vcf.c

/*  vcf.c -- VCF/BCF API functions.

    Copyright (C) 2012, 2013 Broad Institute.
    Copyright (C) 2012-2024 Genome Research Ltd.
    Portions copyright (C) 2014 Intel Corporation.

    Author: Heng Li <lh3@sanger.ac.uk>

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.  */

#define HTS_BUILDING_LIBRARY // Enables HTSLIB_EXPORT, see htslib/hts_defs.h
#include <config.h>

#include <stdio.h>
#include <assert.h>
#include <string.h>
#include <strings.h>
#include <stdlib.h>
#include <limits.h>
#include <stdint.h>
#include <inttypes.h>
#include <errno.h>

#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
#include "fuzz_settings.h"
#endif

#include "htslib/vcf.h"
#include "htslib/bgzf.h"
#include "htslib/tbx.h"
#include "htslib/hfile.h"
#include "hts_internal.h"
#include "htslib/hts_endian.h"
#include "htslib/khash_str2int.h"
#include "htslib/kstring.h"
#include "htslib/sam.h"
#include "htslib/khash.h"

#if 0
// This helps on Intel a bit, often 6-7% faster VCF parsing.
// Conversely sometimes harms AMD Zen4 as ~9% slower.
// Possibly related to IPC differences.  However for now it's just a
// curiousity we ignore and stick with the simpler code.
//
// Left here as a hint for future explorers.
static inline int xstreq(const char *a, const char *b) {
    while (*a && *a == *b)
        a++, b++;
    return *a == *b;
}

#define KHASH_MAP_INIT_XSTR(name, khval_t) \
  KHASH_INIT(name, kh_cstr_t, khval_t, 1, kh_str_hash_func, xstreq)

KHASH_MAP_INIT_XSTR(vdict, bcf_idinfo_t)
#else
KHASH_MAP_INIT_STR(vdict, bcf_idinfo_t)
#endif

typedef khash_t(vdict) vdict_t;

KHASH_MAP_INIT_STR(hdict, bcf_hrec_t*)
typedef khash_t(hdict) hdict_t;


#include "htslib/kseq.h"
HTSLIB_EXPORT
uint32_t bcf_float_missing    = 0x7F800001;

HTSLIB_EXPORT
uint32_t bcf_float_vector_end = 0x7F800002;

HTSLIB_EXPORT
uint8_t bcf_type_shift[] = { 0, 0, 1, 2, 3, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };

static bcf_idinfo_t bcf_idinfo_def = { .info = { 15, 15, 15 }, .hrec = { NULL, NULL, NULL}, .id = -1 };

/*
    Partial support for 64-bit POS and Number=1 INFO tags.
    Notes:
     - the support for 64-bit values is motivated by POS and INFO/END for large genomes
     - the use of 64-bit values does not conform to the specification
     - cannot output 64-bit BCF and if it does, it is not compatible with anything
     - experimental, use at your risk
*/
#ifdef VCF_ALLOW_INT64
    #define BCF_MAX_BT_INT64 (0x7fffffffffffffff)       /* INT64_MAX, for internal use only */
    #define BCF_MIN_BT_INT64 -9223372036854775800LL     /* INT64_MIN + 8, for internal use only */
#endif

#define BCF_IS_64BIT (1<<30)


// Opaque structure with auxilary data which allows to extend bcf_hdr_t without breaking ABI.
// Note that this preserving API and ABI requires that the first element is vdict_t struct
// rather than a pointer, as user programs may (and in some cases do) access the dictionary
// directly as (vdict_t*)hdr->dict.
typedef struct
{
    vdict_t dict;   // bcf_hdr_t.dict[0] vdict_t dictionary which keeps bcf_idinfo_t for BCF_HL_FLT,BCF_HL_INFO,BCF_HL_FMT
    hdict_t *gen;   // hdict_t dictionary which keeps bcf_hrec_t* pointers for generic and structured fields
    size_t *key_len;// length of h->id[BCF_DT_ID] strings
}
bcf_hdr_aux_t;

static inline bcf_hdr_aux_t *get_hdr_aux(const bcf_hdr_t *hdr)
{
    return (bcf_hdr_aux_t *)hdr->dict[0];
}

static char *find_chrom_header_line(char *s)
{
    char *nl;
    if (strncmp(s, "#CHROM\t", 7) == 0) return s;
    else if ((nl = strstr(s, "\n#CHROM\t")) != NULL) return nl+1;
    else return NULL;
}

/*************************
 *** VCF header parser ***
 *************************/

static int bcf_hdr_add_sample_len(bcf_hdr_t *h, const char *s, size_t len)
{
    const char *ss = s;
    while ( *ss && isspace_c(*ss) && ss - s < len) ss++;
    if ( !*ss || ss - s == len)
    {
        hts_log_error("Empty sample name: trailing spaces/tabs in the header line?");
        return -1;
    }

    vdict_t *d = (vdict_t*)h->dict[BCF_DT_SAMPLE];
    int ret;
    char *sdup = malloc(len + 1);
    if (!sdup) return -1;
    memcpy(sdup, s, len);
    sdup[len] = 0;

    // Ensure space is available in h->samples
    size_t n = kh_size(d);
    char **new_samples = realloc(h->samples, sizeof(char*) * (n + 1));
    if (!new_samples) {
        free(sdup);
        return -1;
    }
    h->samples = new_samples;

    int k = kh_put(vdict, d, sdup, &ret);
    if (ret < 0) {
        free(sdup);
        return -1;
    }
    if (ret) { // absent
        kh_val(d, k) = bcf_idinfo_def;
        kh_val(d, k).id = n;
    } else {
        hts_log_error("Duplicated sample name '%s'", sdup);
        free(sdup);
        return -1;
    }
    h->samples[n] = sdup;
    h->dirty = 1;
    return 0;
}

int bcf_hdr_add_sample(bcf_hdr_t *h, const char *s)
{
    if (!s) {
        // Allowed for backwards-compatibility, calling with s == NULL
        // used to trigger bcf_hdr_sync(h);
        return 0;
    }
    return bcf_hdr_add_sample_len(h, s, strlen(s));
}

int HTS_RESULT_USED bcf_hdr_parse_sample_line(bcf_hdr_t *hdr, const char *str)
{
    const char *mandatory = "#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO";
    if ( strncmp(str,mandatory,strlen(mandatory)) )
    {
        hts_log_error("Could not parse the \"#CHROM..\" line, either the fields are incorrect or spaces are present instead of tabs:\n\t%s",str);
        return -1;
    }

    const char *beg = str + strlen(mandatory), *end;
    if ( !*beg || *beg=='\n' ) return 0;
    if ( strncmp(beg,"\tFORMAT\t",8) )
    {
        hts_log_error("Could not parse the \"#CHROM..\" line, either FORMAT is missing or spaces are present instead of tabs:\n\t%s",str);
        return -1;
    }
    beg += 8;

    int ret = 0;
    while ( *beg )
    {
        end = beg;
        while ( *end && *end!='\t' && *end!='\n' ) end++;
        if ( bcf_hdr_add_sample_len(hdr, beg, end-beg) < 0 ) ret = -1;
        if ( !*end || *end=='\n' || ret<0 ) break;
        beg = end + 1;
    }
    return ret;
}

int bcf_hdr_sync(bcf_hdr_t *h)
{
    int i;
    for (i = 0; i < 3; i++)
    {
        vdict_t *d = (vdict_t*)h->dict[i];
        khint_t k;
        if ( h->n[i] < kh_size(d) )
        {
            bcf_idpair_t *new_idpair;
            // this should be true only for i=2, BCF_DT_SAMPLE
            new_idpair = (bcf_idpair_t*) realloc(h->id[i], kh_size(d)*sizeof(bcf_idpair_t));
            if (!new_idpair) return -1;
            h->n[i] = kh_size(d);
            h->id[i] = new_idpair;
        }
        for (k=kh_begin(d); k<kh_end(d); k++)
        {
            if (!kh_exist(d,k)) continue;
            h->id[i][kh_val(d,k).id].key = kh_key(d,k);
            h->id[i][kh_val(d,k).id].val = &kh_val(d,k);
        }
    }

    // Invalidate key length cache
    bcf_hdr_aux_t *aux = get_hdr_aux(h);
    if (aux && aux->key_len) {
        free(aux->key_len);
        aux->key_len = NULL;
    }

    h->dirty = 0;
    return 0;
}

void bcf_hrec_destroy(bcf_hrec_t *hrec)
{
    if (!hrec) return;
    free(hrec->key);
    if ( hrec->value ) free(hrec->value);
    int i;
    for (i=0; i<hrec->nkeys; i++)
    {
        free(hrec->keys[i]);
        free(hrec->vals[i]);
    }
    free(hrec->keys);
    free(hrec->vals);
    free(hrec);
}

// Copies all fields except IDX.
bcf_hrec_t *bcf_hrec_dup(bcf_hrec_t *hrec)
{
    int save_errno;
    bcf_hrec_t *out = (bcf_hrec_t*) calloc(1,sizeof(bcf_hrec_t));
    if (!out) return NULL;

    out->type = hrec->type;
    if ( hrec->key ) {
        out->key = strdup(hrec->key);
        if (!out->key) goto fail;
    }
    if ( hrec->value ) {
        out->value = strdup(hrec->value);
        if (!out->value) goto fail;
    }
    out->nkeys = hrec->nkeys;
    out->keys = (char**) malloc(sizeof(char*)*hrec->nkeys);
    if (!out->keys) goto fail;
    out->vals = (char**) malloc(sizeof(char*)*hrec->nkeys);
    if (!out->vals) goto fail;
    int i, j = 0;
    for (i=0; i<hrec->nkeys; i++)
    {
        if ( hrec->keys[i] && !strcmp("IDX",hrec->keys[i]) ) continue;
        if ( hrec->keys[i] ) {
            out->keys[j] = strdup(hrec->keys[i]);
            if (!out->keys[j]) goto fail;
        }
        if ( hrec->vals[i] ) {
            out->vals[j] = strdup(hrec->vals[i]);
            if (!out->vals[j]) goto fail;
        }
        j++;
    }
    if ( i!=j ) out->nkeys -= i-j;   // IDX was omitted
    return out;

 fail:
    save_errno = errno;
    hts_log_error("%s", strerror(errno));
    bcf_hrec_destroy(out);
    errno = save_errno;
    return NULL;
}

void bcf_hrec_debug(FILE *fp, bcf_hrec_t *hrec)
{
    fprintf(fp, "key=[%s] value=[%s]", hrec->key, hrec->value?hrec->value:"");
    int i;
    for (i=0; i<hrec->nkeys; i++)
        fprintf(fp, "\t[%s]=[%s]", hrec->keys[i],hrec->vals[i]);
    fprintf(fp, "\n");
}

void bcf_header_debug(bcf_hdr_t *hdr)
{
    int i, j;
    for (i=0; i<hdr->nhrec; i++)
    {
        if ( !hdr->hrec[i]->value )
        {
            fprintf(stderr, "##%s=<", hdr->hrec[i]->key);
            fprintf(stderr,"%s=%s", hdr->hrec[i]->keys[0], hdr->hrec[i]->vals[0]);
            for (j=1; j<hdr->hrec[i]->nkeys; j++)
                fprintf(stderr,",%s=%s", hdr->hrec[i]->keys[j], hdr->hrec[i]->vals[j]);
            fprintf(stderr,">\n");
        }
        else
            fprintf(stderr,"##%s=%s\n", hdr->hrec[i]->key,hdr->hrec[i]->value);
    }
}

int bcf_hrec_add_key(bcf_hrec_t *hrec, const char *str, size_t len)
{
    char **tmp;
    size_t n = hrec->nkeys + 1;
    assert(len > 0 && len < SIZE_MAX);
    tmp = realloc(hrec->keys, sizeof(char*)*n);
    if (!tmp) return -1;
    hrec->keys = tmp;
    tmp = realloc(hrec->vals, sizeof(char*)*n);
    if (!tmp) return -1;
    hrec->vals = tmp;

    hrec->keys[hrec->nkeys] = (char*) malloc((len+1)*sizeof(char));
    if (!hrec->keys[hrec->nkeys]) return -1;
    memcpy(hrec->keys[hrec->nkeys],str,len);
    hrec->keys[hrec->nkeys][len] = 0;
    hrec->vals[hrec->nkeys] = NULL;
    hrec->nkeys = n;
    return 0;
}

int bcf_hrec_set_val(bcf_hrec_t *hrec, int i, const char *str, size_t len, int is_quoted)
{
    if ( hrec->vals[i] ) {
        free(hrec->vals[i]);
        hrec->vals[i] = NULL;
    }
    if ( !str ) return 0;
    if ( is_quoted )
    {
        if (len >= SIZE_MAX - 3) {
            errno = ENOMEM;
            return -1;
        }
        hrec->vals[i] = (char*) malloc((len+3)*sizeof(char));
        if (!hrec->vals[i]) return -1;
        hrec->vals[i][0] = '"';
        memcpy(&hrec->vals[i][1],str,len);
        hrec->vals[i][len+1] = '"';
        hrec->vals[i][len+2] = 0;
    }
    else
    {
        if (len == SIZE_MAX) {
            errno = ENOMEM;
            return -1;
        }
        hrec->vals[i] = (char*) malloc((len+1)*sizeof(char));
        if (!hrec->vals[i]) return -1;
        memcpy(hrec->vals[i],str,len);
        hrec->vals[i][len] = 0;
    }
    return 0;
}

int hrec_add_idx(bcf_hrec_t *hrec, int idx)
{
    int n = hrec->nkeys + 1;
    char **tmp = (char**) realloc(hrec->keys, sizeof(char*)*n);
    if (!tmp) return -1;
    hrec->keys = tmp;

    tmp = (char**) realloc(hrec->vals, sizeof(char*)*n);
    if (!tmp) return -1;
    hrec->vals = tmp;

    hrec->keys[hrec->nkeys] = strdup("IDX");
    if (!hrec->keys[hrec->nkeys]) return -1;

    kstring_t str = {0,0,0};
    if (kputw(idx, &str) < 0) {
        free(hrec->keys[hrec->nkeys]);
        return -1;
    }
    hrec->vals[hrec->nkeys] = str.s;
    hrec->nkeys = n;
    return 0;
}

int bcf_hrec_find_key(bcf_hrec_t *hrec, const char *key)
{
    int i;
    for (i=0; i<hrec->nkeys; i++)
        if ( !strcasecmp(key,hrec->keys[i]) ) return i;
    return -1;
}

static void bcf_hrec_set_type(bcf_hrec_t *hrec)
{
    if ( !strcmp(hrec->key, "contig") ) hrec->type = BCF_HL_CTG;
    else if ( !strcmp(hrec->key, "INFO") ) hrec->type = BCF_HL_INFO;
    else if ( !strcmp(hrec->key, "FILTER") ) hrec->type = BCF_HL_FLT;
    else if ( !strcmp(hrec->key, "FORMAT") ) hrec->type = BCF_HL_FMT;
    else if ( hrec->nkeys>0 ) hrec->type = BCF_HL_STR;
    else hrec->type = BCF_HL_GEN;
}


/**
    The arrays were generated with

    valid_ctg:
        perl -le '@v = (split(//,q[!#$%&*+./:;=?@^_|~-]),"a"..."z","A"..."Z","0"..."9"); @a = (0) x 256; foreach $c (@v) { $a[ord($c)] = 1; } print join(", ",@a)' | fold -w 48

    valid_tag:
        perl -le '@v = (split(//,q[_.]),"a"..."z","A"..."Z","0"..."9"); @a = (0) x 256; foreach $c (@v) { $a[ord($c)] = 1; } print join(", ",@a)' | fold -w 48
*/
static const uint8_t valid_ctg[256] =
{
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 1, 0, 1, 1, 1, 1, 0, 0, 0, 1, 1, 0, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1,
    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};
static const uint8_t valid_tag[256] =
{
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,
    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1,
    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};

/**
    bcf_hrec_check() - check the validity of structured header lines

    Returns 0 on success or negative value on error.

    Currently the return status is not checked by the caller
    and only a warning is printed on stderr. This should be improved
    to propagate the error all the way up to the caller and let it
    decide what to do: throw an error or proceed anyway.
 */
static int bcf_hrec_check(bcf_hrec_t *hrec)
{
    int i;
    bcf_hrec_set_type(hrec);

    if ( hrec->type==BCF_HL_CTG )
    {
        i = bcf_hrec_find_key(hrec,"ID");
        if ( i<0 ) goto err_missing_id;
        char *val = hrec->vals[i];
        if ( val[0]=='*' || val[0]=='=' || !valid_ctg[(uint8_t)val[0]] ) goto err_invalid_ctg;
        while ( *(++val) )
            if ( !valid_ctg[(uint8_t)*val] ) goto err_invalid_ctg;
        return 0;
    }
    if ( hrec->type==BCF_HL_INFO )
    {
        i = bcf_hrec_find_key(hrec,"ID");
        if ( i<0 ) goto err_missing_id;
        char *val = hrec->vals[i];
        if ( !strcmp(val,"1000G") ) return 0;
        if ( val[0]=='.' || (val[0]>='0' && val[0]<='9') || !valid_tag[(uint8_t)val[0]] ) goto err_invalid_tag;
        while ( *(++val) )
            if ( !valid_tag[(uint8_t)*val] ) goto err_invalid_tag;
        return 0;
    }
    if ( hrec->type==BCF_HL_FMT )
    {
        i = bcf_hrec_find_key(hrec,"ID");
        if ( i<0 ) goto err_missing_id;
        char *val = hrec->vals[i];
        if ( val[0]=='.' || (val[0]>='0' && val[0]<='9') || !valid_tag[(uint8_t)val[0]] ) goto err_invalid_tag;
        while ( *(++val) )
            if ( !valid_tag[(uint8_t)*val] ) goto err_invalid_tag;
        return 0;
    }
    return 0;

  err_missing_id:
    hts_log_warning("Missing ID attribute in one or more header lines");
    return -1;

  err_invalid_ctg:
    hts_log_warning("Invalid contig name: \"%s\"", hrec->vals[i]);
    return -1;

  err_invalid_tag:
    hts_log_warning("Invalid tag name: \"%s\"", hrec->vals[i]);
    return -1;
}

static inline int is_escaped(const char *min, const char *str)
{
    int n = 0;
    while ( --str>=min && *str=='\\' ) n++;
    return n%2;
}

bcf_hrec_t *bcf_hdr_parse_line(const bcf_hdr_t *h, const char *line, int *len)
{
    bcf_hrec_t *hrec = NULL;
    const char *p = line;
    if (p[0] != '#' || p[1] != '#') { *len = 0; return NULL; }
    p += 2;

    const char *q = p;
    while ( *q && *q!='=' && *q != '\n' ) q++;
    ptrdiff_t n = q-p;
    if ( *q!='=' || !n ) // wrong format
        goto malformed_line;

    hrec = (bcf_hrec_t*) calloc(1,sizeof(bcf_hrec_t));
    if (!hrec) { *len = -1; return NULL; }
    hrec->key = (char*) malloc(sizeof(char)*(n+1));
    if (!hrec->key) goto fail;
    memcpy(hrec->key,p,n);
    hrec->key[n] = 0;
    hrec->type = -1;

    p = ++q;
    if ( *p!='<' ) // generic field, e.g. ##samtoolsVersion=0.1.18-r579
    {
        while ( *q && *q!='\n' ) q++;
        hrec->value = (char*) malloc((q-p+1)*sizeof(char));
        if (!hrec->value) goto fail;
        memcpy(hrec->value, p, q-p);
        hrec->value[q-p] = 0;
        *len = q - line + (*q ? 1 : 0); // Skip \n but not \0
        return hrec;
    }

    // structured line, e.g.
    // ##INFO=<ID=PV1,Number=1,Type=Float,Description="P-value for baseQ bias">
    // ##PEDIGREE=<Name_0=G0-ID,Name_1=G1-ID,Name_3=GN-ID>
    int nopen = 1;
    while ( *q && *q!='\n' && nopen>0 )
    {
        p = ++q;
        while ( *q && *q==' ' ) { p++; q++; }
        // ^[A-Za-z_][0-9A-Za-z_.]*$
        if (p==q && *q && (isalpha_c(*q) || *q=='_'))
        {
            q++;
            while ( *q && (isalnum_c(*q) || *q=='_' || *q=='.') ) q++;
        }
        n = q-p;
        int m = 0;
        while ( *q && *q==' ' ) { q++; m++; }
        if ( *q!='=' || !n )
            goto malformed_line;

        if (bcf_hrec_add_key(hrec, p, q-p-m) < 0) goto fail;
        p = ++q;
        while ( *q && *q==' ' ) { p++; q++; }

        int quoted = 0;
        char ending = '\0';
        switch (*p) {
        case '"':
            quoted = 1;
            ending = '"';
            p++;
            break;
        case '[':
            quoted = 1;
            ending = ']';
            break;
        }
        if ( quoted ) q++;
        while ( *q && *q != '\n' )
        {
            if ( quoted ) { if ( *q==ending && !is_escaped(p,q) ) break; }
            else
            {
                if ( *q=='<' ) nopen++;
                if ( *q=='>' ) nopen--;
                if ( !nopen ) break;
                if ( *q==',' && nopen==1 ) break;
            }
            q++;
        }
        const char *r = q;
        if (quoted && ending == ']') {
            if (*q == ending) {
                r++;
                q++;
                quoted = 0;
            } else {
                char buffer[320];
                hts_log_error("Missing ']' in header line %s",
                              hts_strprint(buffer, sizeof(buffer), '"',
                                           line, q-line));
                goto fail;
            }
        }
        while ( r > p && r[-1] == ' ' ) r--;
        if (bcf_hrec_set_val(hrec, hrec->nkeys-1, p, r-p, quoted) < 0)
            goto fail;
        if ( quoted && *q==ending ) q++;
        if ( *q=='>' )
        {
            if (nopen) nopen--;     // this can happen with nested angle brackets <>
            q++;
        }
    }
    if ( nopen )
        hts_log_warning("Incomplete header line, trying to proceed anyway:\n\t[%s]\n\t[%d]",line,q[0]);

    // Skip to end of line
    int nonspace = 0;
    p = q;
    while ( *q && *q!='\n' ) { nonspace |= !isspace_c(*q); q++; }
    if (nonspace) {
        char buffer[320];
        hts_log_warning("Dropped trailing junk from header line '%s'",
                        hts_strprint(buffer, sizeof(buffer),
                                     '"', line, q - line));
    }

    *len = q - line + (*q ? 1 : 0);
    return hrec;

 fail:
    *len = -1;
    bcf_hrec_destroy(hrec);
    return NULL;

 malformed_line:
    {
        char buffer[320];
        while ( *q && *q!='\n' ) q++;  // Ensure *len includes full line
        hts_log_error("Could not parse the header line: %s",
                      hts_strprint(buffer, sizeof(buffer),
                                   '"', line, q - line));
        *len = q - line + (*q ? 1 : 0);
        bcf_hrec_destroy(hrec);
        return NULL;
    }
}

static int bcf_hdr_set_idx(bcf_hdr_t *hdr, int dict_type, const char *tag, bcf_idinfo_t *idinfo)
{
    size_t new_n;

    // If available, preserve existing IDX
    if ( idinfo->id==-1 )
        idinfo->id = hdr->n[dict_type];
    else if ( idinfo->id < hdr->n[dict_type] && hdr->id[dict_type][idinfo->id].key )
    {
        hts_log_error("Conflicting IDX=%d lines in the header dictionary, the new tag is %s",
            idinfo->id, tag);
        errno = EINVAL;
        return -1;
    }

    new_n = idinfo->id >= hdr->n[dict_type] ? idinfo->id+1 : hdr->n[dict_type];
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
    // hts_resize() can attempt to allocate up to 2 * requested items
    if (new_n > FUZZ_ALLOC_LIMIT/(2 * sizeof(bcf_idpair_t)))
        return -1;
#endif
    if (hts_resize(bcf_idpair_t, new_n, &hdr->m[dict_type],
                   &hdr->id[dict_type], HTS_RESIZE_CLEAR)) {
        return -1;
    }
    hdr->n[dict_type] = new_n;

    // NB: the next kh_put call can invalidate the idinfo pointer, therefore
    // we leave it unassigned here. It must be set explicitly in bcf_hdr_sync.
    hdr->id[dict_type][idinfo->id].key = tag;

    return 0;
}

// returns: 1 when hdr needs to be synced, -1 on error, 0 otherwise
static int bcf_hdr_register_hrec(bcf_hdr_t *hdr, bcf_hrec_t *hrec)
{
    // contig
    int i, ret, replacing = 0;
    khint_t k;
    char *str = NULL;

    bcf_hrec_set_type(hrec);

    if ( hrec->type==BCF_HL_CTG )
    {
        hts_pos_t len = 0;

        // Get the contig ID ($str) and length ($j)
        i = bcf_hrec_find_key(hrec,"length");
        if ( i<0 ) len = 0;
        else {
            char *end = hrec->vals[i];
            len = strtoll(hrec->vals[i], &end, 10);
            if (end == hrec->vals[i] || len < 0) return 0;
        }

        i = bcf_hrec_find_key(hrec,"ID");
        if ( i<0 ) return 0;
        str = strdup(hrec->vals[i]);
        if (!str) return -1;

        // Register in the dictionary
        vdict_t *d = (vdict_t*)hdr->dict[BCF_DT_CTG];
        khint_t k = kh_get(vdict, d, str);
        if ( k != kh_end(d) ) { // already present
            free(str); str=NULL;
            if (kh_val(d, k).hrec[0] != NULL) // and not removed
                return 0;
            replacing = 1;
        } else {
            k = kh_put(vdict, d, str, &ret);
            if (ret < 0) { free(str); return -1; }
        }

        int idx = bcf_hrec_find_key(hrec,"IDX");
        if ( idx!=-1 )
        {
            char *tmp = hrec->vals[idx];
            idx = strtol(hrec->vals[idx], &tmp, 10);
            if ( *tmp || idx < 0 || idx >= INT_MAX - 1)
            {
                if (!replacing) {
                    kh_del(vdict, d, k);
                    free(str);
                }
                hts_log_warning("Error parsing the IDX tag, skipping");
                return 0;
            }
        }

        kh_val(d, k) = bcf_idinfo_def;
        kh_val(d, k).id = idx;
        kh_val(d, k).info[0] = len;
        kh_val(d, k).hrec[0] = hrec;
        if (bcf_hdr_set_idx(hdr, BCF_DT_CTG, kh_key(d,k), &kh_val(d,k)) < 0) {
            if (!replacing) {
                kh_del(vdict, d, k);
                free(str);
            }
            return -1;
        }
        if ( idx==-1 ) {
            if (hrec_add_idx(hrec, kh_val(d,k).id) < 0) {
               return -1;
            }
        }

        return 1;
    }

    if ( hrec->type==BCF_HL_STR ) return 1;
    if ( hrec->type!=BCF_HL_INFO && hrec->type!=BCF_HL_FLT && hrec->type!=BCF_HL_FMT ) return 0;

    // INFO/FILTER/FORMAT
    char *id = NULL;
    uint32_t type = UINT32_MAX, var = UINT32_MAX;
    int num = -1, idx = -1;
    for (i=0; i<hrec->nkeys; i++)
    {
        if ( !strcmp(hrec->keys[i], "ID") ) id = hrec->vals[i];
        else if ( !strcmp(hrec->keys[i], "IDX") )
        {
            char *tmp = hrec->vals[i];
            idx = strtol(hrec->vals[i], &tmp, 10);
            if ( *tmp || idx < 0 || idx >= INT_MAX - 1)
            {
                hts_log_warning("Error parsing the IDX tag, skipping");
                return 0;
            }
        }
        else if ( !strcmp(hrec->keys[i], "Type") )
        {
            if ( !strcmp(hrec->vals[i], "Integer") ) type = BCF_HT_INT;
            else if ( !strcmp(hrec->vals[i], "Float") ) type = BCF_HT_REAL;
            else if ( !strcmp(hrec->vals[i], "String") ) type = BCF_HT_STR;
            else if ( !strcmp(hrec->vals[i], "Character") ) type = BCF_HT_STR;
            else if ( !strcmp(hrec->vals[i], "Flag") ) type = BCF_HT_FLAG;
            else
            {
                hts_log_warning("The type \"%s\" is not supported, assuming \"String\"", hrec->vals[i]);
                type = BCF_HT_STR;
            }
        }
        else if ( !strcmp(hrec->keys[i], "Number") )
        {
            if ( !strcmp(hrec->vals[i],"A") ) var = BCF_VL_A;
            else if ( !strcmp(hrec->vals[i],"R") ) var = BCF_VL_R;
            else if ( !strcmp(hrec->vals[i],"G") ) var = BCF_VL_G;
            else if ( !strcmp(hrec->vals[i],".") ) var = BCF_VL_VAR;
            else
            {
                sscanf(hrec->vals[i],"%d",&num);
                var = BCF_VL_FIXED;
            }
            if (var != BCF_VL_FIXED) num = 0xfffff;
        }
    }
    if (hrec->type == BCF_HL_INFO || hrec->type == BCF_HL_FMT) {
        if (type == -1) {
            hts_log_warning("%s %s field has no Type defined. Assuming String",
                *hrec->key == 'I' ? "An" : "A", hrec->key);
            type = BCF_HT_STR;
        }
        if (var == -1) {
            hts_log_warning("%s %s field has no Number defined. Assuming '.'",
                *hrec->key == 'I' ? "An" : "A", hrec->key);
            var = BCF_VL_VAR;
        }
        if ( type==BCF_HT_FLAG && (var!=BCF_VL_FIXED || num!=0) )
        {
            hts_log_warning("The definition of Flag \"%s/%s\" is invalid, forcing Number=0", hrec->key,id);
            var = BCF_VL_FIXED;
            num = 0;
        }
    }
    uint32_t info = ((((uint32_t)num) & 0xfffff)<<12 |
                     (var & 0xf) << 8 |
                     (type & 0xf) << 4 |
                     (((uint32_t) hrec->type) & 0xf));

    if ( !id ) return 0;
    str = strdup(id);
    if (!str) return -1;

    vdict_t *d = (vdict_t*)hdr->dict[BCF_DT_ID];
    k = kh_get(vdict, d, str);
    if ( k != kh_end(d) )
    {
        // already present
        free(str);
        if ( kh_val(d, k).hrec[info&0xf] ) return 0;
        kh_val(d, k).info[info&0xf] = info;
        kh_val(d, k).hrec[info&0xf] = hrec;
        if ( idx==-1 ) {
            if (hrec_add_idx(hrec, kh_val(d, k).id) < 0) {
                return -1;
            }
        }
        return 1;
    }
    k = kh_put(vdict, d, str, &ret);
    if (ret < 0) {
        free(str);
        return -1;
    }
    kh_val(d, k) = bcf_idinfo_def;
    kh_val(d, k).info[info&0xf] = info;
    kh_val(d, k).hrec[info&0xf] = hrec;
    kh_val(d, k).id = idx;
    if (bcf_hdr_set_idx(hdr, BCF_DT_ID, kh_key(d,k), &kh_val(d,k)) < 0) {
        kh_del(vdict, d, k);
        free(str);
        return -1;
    }
    if ( idx==-1 ) {
        if (hrec_add_idx(hrec, kh_val(d,k).id) < 0) {
            return -1;
        }
    }

    return 1;
}

static void bcf_hdr_unregister_hrec(bcf_hdr_t *hdr, bcf_hrec_t *hrec)
{
    if (hrec->type == BCF_HL_FLT ||
        hrec->type == BCF_HL_INFO ||
        hrec->type == BCF_HL_FMT ||
        hrec->type == BCF_HL_CTG) {
        int id = bcf_hrec_find_key(hrec, "ID");
        if (id < 0 || !hrec->vals[id])
            return;
        vdict_t *dict = (hrec->type == BCF_HL_CTG
                         ? (vdict_t*)hdr->dict[BCF_DT_CTG]
                         : (vdict_t*)hdr->dict[BCF_DT_ID]);
        khint_t k = kh_get(vdict, dict, hrec->vals[id]);
        if (k != kh_end(dict))
            kh_val(dict, k).hrec[hrec->type==BCF_HL_CTG ? 0 : hrec->type] = NULL;
    }
}

static void bcf_hdr_remove_from_hdict(bcf_hdr_t *hdr, bcf_hrec_t *hrec)
{
    kstring_t str = KS_INITIALIZE;
    bcf_hdr_aux_t *aux = get_hdr_aux(hdr);
    khint_t k;
    int id;

    switch (hrec->type) {
    case BCF_HL_GEN:
        if (ksprintf(&str, "##%s=%s", hrec->key,hrec->value) < 0)
            str.l = 0;
        break;
    case BCF_HL_STR:
        id = bcf_hrec_find_key(hrec, "ID");
        if (id < 0)
            return;
        if (!hrec->vals[id] ||
            ksprintf(&str, "##%s=<ID=%s>", hrec->key, hrec->vals[id]) < 0)
            str.l = 0;
        break;
    default:
        return;
    }
    if (str.l) {
        k = kh_get(hdict, aux->gen, str.s);
    } else {
        // Couldn't get a string for some reason, so try the hard way...
        for (k = kh_begin(aux->gen); k < kh_end(aux->gen); k++) {
            if (kh_exist(aux->gen, k) && kh_val(aux->gen, k) == hrec)
                break;
        }
    }
    if (k != kh_end(aux->gen) && kh_val(aux->gen, k) == hrec) {
        kh_val(aux->gen, k) = NULL;
        free((char *) kh_key(aux->gen, k));
        kh_key(aux->gen, k) = NULL;
        kh_del(hdict, aux->gen, k);
    }
    free(str.s);
}

int bcf_hdr_update_hrec(bcf_hdr_t *hdr, bcf_hrec_t *hrec, const bcf_hrec_t *tmp)
{
    // currently only for bcf_hdr_set_version
    assert( hrec->type==BCF_HL_GEN );
    int ret;
    khint_t k;
    bcf_hdr_aux_t *aux = get_hdr_aux(hdr);
    for (k=kh_begin(aux->gen); k<kh_end(aux->gen); k++)
    {
        if ( !kh_exist(aux->gen,k) ) continue;
        if ( hrec!=(bcf_hrec_t*)kh_val(aux->gen,k) ) continue;
        break;
    }
    assert( k<kh_end(aux->gen) );   // something went wrong, should never happen
    free((char*)kh_key(aux->gen,k));
    kh_del(hdict,aux->gen,k);
    kstring_t str = {0,0,0};
    if ( ksprintf(&str, "##%s=%s", tmp->key,tmp->value) < 0 )
    {
        free(str.s);
        return -1;
    }
    k = kh_put(hdict, aux->gen, str.s, &ret);
    if ( ret<0 )
    {
        free(str.s);
        return -1;
    }
    free(hrec->value);
    hrec->value = strdup(tmp->value);
    if ( !hrec->value ) return -1;
    return 0;
}

int bcf_hdr_add_hrec(bcf_hdr_t *hdr, bcf_hrec_t *hrec)
{
    kstring_t str = {0,0,0};
    bcf_hdr_aux_t *aux = get_hdr_aux(hdr);

    int res;
    if ( !hrec ) return 0;

    bcf_hrec_check(hrec);   // todo: check return status and propagate errors up

    res = bcf_hdr_register_hrec(hdr,hrec);
    if (res < 0) return -1;
    if ( !res )
    {
        // If one of the hashed field, then it is already present
        if ( hrec->type != BCF_HL_GEN )
        {
            bcf_hrec_destroy(hrec);
            return 0;
        }

        // Is one of the generic fields and already present?
        if ( ksprintf(&str, "##%s=%s", hrec->key,hrec->value) < 0 )
        {
            free(str.s);
            return -1;
        }
        khint_t k = kh_get(hdict, aux->gen, str.s);
        if ( k != kh_end(aux->gen) )
        {
            // duplicate record
            bcf_hrec_destroy(hrec);
            free(str.s);
            return 0;
        }
    }

    int i;
    if ( hrec->type==BCF_HL_STR && (i=bcf_hrec_find_key(hrec,"ID"))>=0 )
    {
        if ( ksprintf(&str, "##%s=<ID=%s>", hrec->key,hrec->vals[i]) < 0 )
        {
            free(str.s);
            return -1;
        }
        khint_t k = kh_get(hdict, aux->gen, str.s);
        if ( k != kh_end(aux->gen) )
        {
            // duplicate record
            bcf_hrec_destroy(hrec);
            free(str.s);
            return 0;
        }
    }

    // New record, needs to be added
    int n = hdr->nhrec + 1;
    bcf_hrec_t **new_hrec = realloc(hdr->hrec, n*sizeof(bcf_hrec_t*));
    if (!new_hrec) {
        free(str.s);
        bcf_hdr_unregister_hrec(hdr, hrec);
        return -1;
    }
    hdr->hrec = new_hrec;

    if ( str.s )
    {
        khint_t k = kh_put(hdict, aux->gen, str.s, &res);
        if ( res<0 )
        {
            free(str.s);
            return -1;
        }
        kh_val(aux->gen,k) = hrec;
    }

    hdr->hrec[hdr->nhrec] = hrec;
    hdr->dirty = 1;
    hdr->nhrec = n;

    return hrec->type==BCF_HL_GEN ? 0 : 1;
}

bcf_hrec_t *bcf_hdr_get_hrec(const bcf_hdr_t *hdr, int type, const char *key, const char *value, const char *str_class)
{
    int i;
    if ( type==BCF_HL_GEN )
    {
        // e.g. ##fileformat=VCFv4.2
        //      ##source=GenomicsDBImport
        //      ##bcftools_viewVersion=1.16-80-gdfdb0923+htslib-1.16-34-g215d364
        if ( value )
        {
            kstring_t str = {0,0,0};
            ksprintf(&str, "##%s=%s", key,value);
            bcf_hdr_aux_t *aux = get_hdr_aux(hdr);
            khint_t k = kh_get(hdict, aux->gen, str.s);
            free(str.s);
            if ( k == kh_end(aux->gen) ) return NULL;
            return kh_val(aux->gen, k);
        }
        for (i=0; i<hdr->nhrec; i++)
        {
            if ( hdr->hrec[i]->type!=type ) continue;
            if ( strcmp(hdr->hrec[i]->key,key) ) continue;
            return hdr->hrec[i];
        }
        return NULL;
    }
    else if ( type==BCF_HL_STR )
    {
        // e.g. ##GATKCommandLine=<ID=GenomicsDBImport,CommandLine="GenomicsDBImport....">
        //      ##ALT=<ID=NON_REF,Description="Represents any possible alternative allele not already represented at this location by REF and ALT">
        if (!str_class) return NULL;
        if ( !strcmp("ID",key) )
        {
            kstring_t str = {0,0,0};
            ksprintf(&str, "##%s=<%s=%s>",str_class,key,value);
            bcf_hdr_aux_t *aux = get_hdr_aux(hdr);
            khint_t k = kh_get(hdict, aux->gen, str.s);
            free(str.s);
            if ( k == kh_end(aux->gen) ) return NULL;
            return kh_val(aux->gen, k);
        }
        for (i=0; i<hdr->nhrec; i++)
        {
            if ( hdr->hrec[i]->type!=type ) continue;
            if ( strcmp(hdr->hrec[i]->key,str_class) ) continue;
            int j = bcf_hrec_find_key(hdr->hrec[i],key);
            if ( j>=0 && !strcmp(hdr->hrec[i]->vals[j],value) ) return hdr->hrec[i];
        }
        return NULL;
    }
    vdict_t *d = type==BCF_HL_CTG ? (vdict_t*)hdr->dict[BCF_DT_CTG] : (vdict_t*)hdr->dict[BCF_DT_ID];
    khint_t k = kh_get(vdict, d, value);
    if ( k == kh_end(d) ) return NULL;
    return kh_val(d, k).hrec[type==BCF_HL_CTG?0:type];
}

void bcf_hdr_check_sanity(bcf_hdr_t *hdr)
{
    static int PL_warned = 0, GL_warned = 0;

    if ( !PL_warned )
    {
        int id = bcf_hdr_id2int(hdr, BCF_DT_ID, "PL");
        if ( bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,id) && bcf_hdr_id2length(hdr,BCF_HL_FMT,id)!=BCF_VL_G )
        {
            hts_log_warning("PL should be declared as Number=G");
            PL_warned = 1;
        }
    }
    if ( !GL_warned )
    {
        int id = bcf_hdr_id2int(hdr, BCF_DT_ID, "GL");
        if ( bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,id) && bcf_hdr_id2length(hdr,BCF_HL_FMT,id)!=BCF_VL_G )
        {
            hts_log_warning("GL should be declared as Number=G");
            GL_warned = 1;
        }
    }
}

int bcf_hdr_parse(bcf_hdr_t *hdr, char *htxt)
{
    int len, done = 0;
    char *p = htxt;

    // Check sanity: "fileformat" string must come as first
    bcf_hrec_t *hrec = bcf_hdr_parse_line(hdr,p,&len);
    if ( !hrec || !hrec->key || strcasecmp(hrec->key,"fileformat") )
        hts_log_warning("The first line should be ##fileformat; is the VCF/BCF header broken?");
    if (bcf_hdr_add_hrec(hdr, hrec) < 0) {
        bcf_hrec_destroy(hrec);
        return -1;
    }

    // The filter PASS must appear first in the dictionary
    hrec = bcf_hdr_parse_line(hdr,"##FILTER=<ID=PASS,Description=\"All filters passed\">",&len);
    if (!hrec || bcf_hdr_add_hrec(hdr, hrec) < 0) {
        bcf_hrec_destroy(hrec);
        return -1;
    }

    // Parse the whole header
    do {
        while (NULL != (hrec = bcf_hdr_parse_line(hdr, p, &len))) {
            if (bcf_hdr_add_hrec(hdr, hrec) < 0) {
                bcf_hrec_destroy(hrec);
                return -1;
            }
            p += len;
        }
        assert(hrec == NULL);
        if (len < 0) {
            // len < 0 indicates out-of-memory, or similar error
            hts_log_error("Could not parse header line: %s", strerror(errno));
            return -1;
        } else if (len > 0) {
            // Bad header line.  bcf_hdr_parse_line() will have logged it.
            // Skip and try again on the next line (p + len will be the start
            // of the next one).
            p += len;
            continue;
        }

        // Next should be the sample line.  If not, it was a malformed
        // header, in which case print a warning and skip (many VCF
        // operations do not really care about a few malformed lines).
        // In the future we may want to add a strict mode that errors in
        // this case.
        if ( strncmp("#CHROM\t",p,7) && strncmp("#CHROM ",p,7) ) {
            char *eol = strchr(p, '\n');
            if (*p != '\0') {
                char buffer[320];
                hts_log_warning("Could not parse header line: %s",
                                hts_strprint(buffer, sizeof(buffer),
                                               '"', p,
                                               eol ? (eol - p) : SIZE_MAX));
            }
            if (eol) {
                p = eol + 1; // Try from the next line.
            } else {
                done = -1; // No more lines left, give up.
            }
        } else {
            done = 1; // Sample line found
        }
    } while (!done);

    if (done < 0) {
        // No sample line is fatal.
        hts_log_error("Could not parse the header, sample line not found");
        return -1;
    }

    if (bcf_hdr_parse_sample_line(hdr,p) < 0)
        return -1;
    if (bcf_hdr_sync(hdr) < 0)
        return -1;
    bcf_hdr_check_sanity(hdr);
    return 0;
}

int bcf_hdr_append(bcf_hdr_t *hdr, const char *line)
{
    int len;
    bcf_hrec_t *hrec = bcf_hdr_parse_line(hdr, (char*) line, &len);
    if ( !hrec ) return -1;
    if (bcf_hdr_add_hrec(hdr, hrec) < 0)
        return -1;
    return 0;
}

void bcf_hdr_remove(bcf_hdr_t *hdr, int type, const char *key)
{
    int i = 0;
    bcf_hrec_t *hrec;
    if ( !key )
    {
        // no key, remove all entries of this type
        while ( i<hdr->nhrec )
        {
            if ( hdr->hrec[i]->type!=type ) { i++; continue; }
            hrec = hdr->hrec[i];
            bcf_hdr_unregister_hrec(hdr, hrec);
            bcf_hdr_remove_from_hdict(hdr, hrec);
            hdr->dirty = 1;
            hdr->nhrec--;
            if ( i < hdr->nhrec )
                memmove(&hdr->hrec[i],&hdr->hrec[i+1],(hdr->nhrec-i)*sizeof(bcf_hrec_t*));
            bcf_hrec_destroy(hrec);
        }
        return;
    }
    while (1)
    {
        if ( type==BCF_HL_FLT || type==BCF_HL_INFO || type==BCF_HL_FMT || type== BCF_HL_CTG )
        {
            hrec = bcf_hdr_get_hrec(hdr, type, "ID", key, NULL);
            if ( !hrec ) return;

            for (i=0; i<hdr->nhrec; i++)
                if ( hdr->hrec[i]==hrec ) break;
            assert( i<hdr->nhrec );

            vdict_t *d = type==BCF_HL_CTG ? (vdict_t*)hdr->dict[BCF_DT_CTG] : (vdict_t*)hdr->dict[BCF_DT_ID];
            khint_t k = kh_get(vdict, d, key);
            kh_val(d, k).hrec[type==BCF_HL_CTG?0:type] = NULL;
        }
        else
        {
            for (i=0; i<hdr->nhrec; i++)
            {
                if ( hdr->hrec[i]->type!=type ) continue;
                if ( type==BCF_HL_GEN )
                {
                    if ( !strcmp(hdr->hrec[i]->key,key) ) break;
                }
                else
                {
                    // not all structured lines have ID, we could be more sophisticated as in bcf_hdr_get_hrec()
                    int j = bcf_hrec_find_key(hdr->hrec[i], "ID");
                    if ( j>=0 && !strcmp(hdr->hrec[i]->vals[j],key) ) break;
                }
            }
            if ( i==hdr->nhrec ) return;
            hrec = hdr->hrec[i];
            bcf_hdr_remove_from_hdict(hdr, hrec);
        }

        hdr->nhrec--;
        if ( i < hdr->nhrec )
            memmove(&hdr->hrec[i],&hdr->hrec[i+1],(hdr->nhrec-i)*sizeof(bcf_hrec_t*));
        bcf_hrec_destroy(hrec);
        hdr->dirty = 1;
    }
}

int bcf_hdr_printf(bcf_hdr_t *hdr, const char *fmt, ...)
{
    char tmp[256], *line = tmp;
    va_list ap;
    va_start(ap, fmt);
    int n = vsnprintf(line, sizeof(tmp), fmt, ap);
    va_end(ap);

    if (n >= sizeof(tmp)) {
        n++; // For trailing NUL
        line = (char*)malloc(n);
        if (!line)
            return -1;

        va_start(ap, fmt);
        vsnprintf(line, n, fmt, ap);
        va_end(ap);
    }

    int ret = bcf_hdr_append(hdr, line);

    if (line != tmp) free(line);
    return ret;
}


/**********************
 *** BCF header I/O ***
 **********************/

const char *bcf_hdr_get_version(const bcf_hdr_t *hdr)
{
    bcf_hrec_t *hrec = bcf_hdr_get_hrec(hdr, BCF_HL_GEN, "fileformat", NULL, NULL);
    if ( !hrec )
    {
        hts_log_warning("No version string found, assuming VCFv4.2");
        return "VCFv4.2";
    }
    return hrec->value;
}

int bcf_hdr_set_version(bcf_hdr_t *hdr, const char *version)
{
    bcf_hrec_t *hrec = bcf_hdr_get_hrec(hdr, BCF_HL_GEN, "fileformat", NULL, NULL);
    if ( !hrec )
    {
        int len;
        kstring_t str = {0,0,0};
        if ( ksprintf(&str,"##fileformat=%s", version) < 0 ) return -1;
        hrec = bcf_hdr_parse_line(hdr, str.s, &len);
        free(str.s);
    }
    else
    {
        bcf_hrec_t *tmp = bcf_hrec_dup(hrec);
        if ( !tmp ) return -1;
        free(tmp->value);
        tmp->value = strdup(version);
        if ( !tmp->value ) return -1;
        bcf_hdr_update_hrec(hdr, hrec, tmp);
        bcf_hrec_destroy(tmp);
    }
    hdr->dirty = 1;
    return 0; // FIXME: check for errs in this function (return < 0 if so)
}

bcf_hdr_t *bcf_hdr_init(const char *mode)
{
    int i;
    bcf_hdr_t *h;
    h = (bcf_hdr_t*)calloc(1, sizeof(bcf_hdr_t));
    if (!h) return NULL;
    for (i = 0; i < 3; ++i) {
        if ((h->dict[i] = kh_init(vdict)) == NULL) goto fail;
        // Supersize the hash to make collisions very unlikely
        static int dsize[3] = {16384,16384,2048}; // info, contig, format
        if (kh_resize(vdict, h->dict[i], dsize[i]) < 0) goto fail;
    }

    bcf_hdr_aux_t *aux = (bcf_hdr_aux_t*)calloc(1,sizeof(bcf_hdr_aux_t));
    if ( !aux ) goto fail;
    if ( (aux->gen = kh_init(hdict))==NULL ) { free(aux); goto fail; }
    aux->key_len = NULL;
    aux->dict = *((vdict_t*)h->dict[0]);
    free(h->dict[0]);
    h->dict[0] = aux;

    if ( strchr(mode,'w') )
    {
        bcf_hdr_append(h, "##fileformat=VCFv4.2");
        // The filter PASS must appear first in the dictionary
        bcf_hdr_append(h, "##FILTER=<ID=PASS,Description=\"All filters passed\">");
    }
    return h;

 fail:
    for (i = 0; i < 3; ++i)
        kh_destroy(vdict, h->dict[i]);
    free(h);
    return NULL;
}

void bcf_hdr_destroy(bcf_hdr_t *h)
{
    int i;
    khint_t k;
    if (!h) return;
    for (i = 0; i < 3; ++i) {
        vdict_t *d = (vdict_t*)h->dict[i];
        if (d == 0) continue;
        for (k = kh_begin(d); k != kh_end(d); ++k)
            if (kh_exist(d, k)) free((char*)kh_key(d, k));
        if ( i==0 )
        {
            bcf_hdr_aux_t *aux = get_hdr_aux(h);
            for (k=kh_begin(aux->gen); k<kh_end(aux->gen); k++)
                if ( kh_exist(aux->gen,k) ) free((char*)kh_key(aux->gen,k));
            kh_destroy(hdict, aux->gen);
            free(aux->key_len); // may exist for dict[0] only
        }
        kh_destroy(vdict, d);
        free(h->id[i]);
    }
    for (i=0; i<h->nhrec; i++)
        bcf_hrec_destroy(h->hrec[i]);
    if (h->nhrec) free(h->hrec);
    if (h->samples) free(h->samples);
    free(h->keep_samples);
    free(h->transl[0]); free(h->transl[1]);
    free(h->mem.s);
    free(h);
}

bcf_hdr_t *bcf_hdr_read(htsFile *hfp)
{
    if (hfp->format.format == vcf)
        return vcf_hdr_read(hfp);
    if (hfp->format.format != bcf) {
        hts_log_error("Input is not detected as bcf or vcf format");
        return NULL;
    }

    assert(hfp->is_bgzf);

    BGZF *fp = hfp->fp.bgzf;
    uint8_t magic[5];
    bcf_hdr_t *h;
    h = bcf_hdr_init("r");
    if (!h) {
        hts_log_error("Failed to allocate bcf header");
        return NULL;
    }
    if (bgzf_read(fp, magic, 5) != 5)
    {
        hts_log_error("Failed to read the header (reading BCF in text mode?)");
        bcf_hdr_destroy(h);
        return NULL;
    }
    if (strncmp((char*)magic, "BCF\2\2", 5) != 0)
    {
        if (!strncmp((char*)magic, "BCF", 3))
            hts_log_error("Invalid BCF2 magic string: only BCFv2.2 is supported");
        else
            hts_log_error("Invalid BCF2 magic string");
        bcf_hdr_destroy(h);
        return NULL;
    }
    uint8_t buf[4];
    size_t hlen;
    char *htxt = NULL;
    if (bgzf_read(fp, buf, 4) != 4) goto fail;
    hlen = buf[0] | (buf[1] << 8) | (buf[2] << 16) | ((size_t) buf[3] << 24);
    if (hlen >= SIZE_MAX) { errno = ENOMEM; goto fail; }
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
    if (hlen > FUZZ_ALLOC_LIMIT/2) { errno = ENOMEM; goto fail; }
#endif
    htxt = (char*)malloc(hlen + 1);
    if (!htxt) goto fail;
    if (bgzf_read(fp, htxt, hlen) != hlen) goto fail;
    htxt[hlen] = '\0'; // Ensure htxt is terminated
    if ( bcf_hdr_parse(h, htxt) < 0 ) goto fail;
    free(htxt);
    return h;
 fail:
    hts_log_error("Failed to read BCF header");
    free(htxt);
    bcf_hdr_destroy(h);
    return NULL;
}

int bcf_hdr_write(htsFile *hfp, bcf_hdr_t *h)
{
    if (!h) {
        errno = EINVAL;
        return -1;
    }
    if ( h->dirty ) {
        if (bcf_hdr_sync(h) < 0) return -1;
    }
    hfp->format.category = variant_data;
    if (hfp->format.format == vcf || hfp->format.format == text_format) {
        hfp->format.format = vcf;
        return vcf_hdr_write(hfp, h);
    }

    if (hfp->format.format == binary_format)
        hfp->format.format = bcf;

    kstring_t htxt = {0,0,0};
    if (bcf_hdr_format(h, 1, &htxt) < 0) {
        free(htxt.s);
        return -1;
    }
    kputc('\0', &htxt); // include the \0 byte

    BGZF *fp = hfp->fp.bgzf;
    if ( bgzf_write(fp, "BCF\2\2", 5) !=5 ) return -1;
    uint8_t hlen[4];
    u32_to_le(htxt.l, hlen);
    if ( bgzf_write(fp, hlen, 4) !=4 ) return -1;
    if ( bgzf_write(fp, htxt.s, htxt.l) != htxt.l ) return -1;
    if ( bgzf_flush(fp) < 0) return -1;

    free(htxt.s);
    return 0;
}

/********************
 *** BCF site I/O ***
 ********************/

bcf1_t *bcf_init(void)
{
    bcf1_t *v;
    v = (bcf1_t*)calloc(1, sizeof(bcf1_t));
    return v;
}

void bcf_clear(bcf1_t *v)
{
    int i;
    for (i=0; i<v->d.m_info; i++)
    {
        if ( v->d.info[i].vptr_free )
        {
            free(v->d.info[i].vptr - v->d.info[i].vptr_off);
            v->d.info[i].vptr_free = 0;
        }
    }
    for (i=0; i<v->d.m_fmt; i++)
    {
        if ( v->d.fmt[i].p_free )
        {
            free(v->d.fmt[i].p - v->d.fmt[i].p_off);
            v->d.fmt[i].p_free = 0;
        }
    }
    v->rid = v->pos = v->rlen = v->unpacked = 0;
    bcf_float_set_missing(v->qual);
    v->n_info = v->n_allele = v->n_fmt = v->n_sample = 0;
    v->shared.l = v->indiv.l = 0;
    v->d.var_type = -1;
    v->d.shared_dirty = 0;
    v->d.indiv_dirty  = 0;
    v->d.n_flt = 0;
    v->errcode = 0;
    if (v->d.m_als) v->d.als[0] = 0;
    if (v->d.m_id) v->d.id[0] = 0;
}

void bcf_empty(bcf1_t *v)
{
    bcf_clear1(v);
    free(v->d.id);
    free(v->d.als);
    free(v->d.allele); free(v->d.flt); free(v->d.info); free(v->d.fmt);
    if (v->d.var ) free(v->d.var);
    free(v->shared.s); free(v->indiv.s);
    memset(&v->d,0,sizeof(v->d));
    memset(&v->shared,0,sizeof(v->shared));
    memset(&v->indiv,0,sizeof(v->indiv));
}

void bcf_destroy(bcf1_t *v)
{
    if (!v) return;
    bcf_empty1(v);
    free(v);
}

static inline int bcf_read1_core(BGZF *fp, bcf1_t *v)
{
    uint8_t x[32];
    ssize_t ret;
    uint32_t shared_len, indiv_len;
    if ((ret = bgzf_read(fp, x, 32)) != 32) {
        if (ret == 0) return -1;
        return -2;
    }
    bcf_clear1(v);
    shared_len = le_to_u32(x);
    if (shared_len < 24) return -2;
    shared_len -= 24; // to exclude six 32-bit integers
    indiv_len = le_to_u32(x + 4);
#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
    // ks_resize() normally allocates 1.5 * requested size to allow for growth
    if ((uint64_t) shared_len + indiv_len > FUZZ_ALLOC_LIMIT / 3 * 2) return -2;
#endif
    if (ks_resize(&v->shared, shared_len ? shared_len : 1) != 0) return -2;
    if (ks_resize(&v->indiv, indiv_len ? indiv_len : 1) != 0) return -2;
    v->rid  = le_to_i32(x + 8);
    v->pos  = le_to_u32(x + 12);
    if ( v->pos==UINT32_MAX ) v->pos = -1;  // this is for telomere coordinate, e.g. MT:0
    v->rlen = le_to_i32(x + 16);
    v->qual = le_to_float(x + 20);
    v->n_info = le_to_u16(x + 24);
    v->n_allele = le_to_u16(x + 26);
    v->n_sample = le_to_u32(x + 28) & 0xffffff;
    v->n_fmt = x[31];
    v->shared.l = shared_len;
    v->indiv.l = indiv_len;
    // silent fix of broken BCFs produced by earlier versions of bcf_subset, prior to and including bd6ed8b4
    if ( (!v->indiv.l || !v->n_sample) && v->n_fmt ) v->n_fmt = 0;

    if (bgzf_read(fp, v->shared.s, v->shared.l) != v->shared.l) return -2;
    if (bgzf_read(fp, v->indiv.s, v->indiv.l) != v->indiv.l) return -2;
    return 0;
}

#define bit_array_size(n) ((n)/8+1)
#define bit_array_set(a,i)   ((a)[(i)/8] |=   1 << ((i)%8))
#define bit_array_clear(a,i) ((a)[(i)/8] &= ~(1 << ((i)%8)))
#define bit_array_test(a,i)  ((a)[(i)/8] &   (1 << ((i)%8)))

static int bcf_dec_typed_int1_safe(uint8_t *p, uint8_t *end, uint8_t **q,
                                   int32_t *val) {
    uint32_t t;
    if (end - p < 2) return -1;
    t = *p++ & 0xf;
    /* Use if .. else if ... else instead of switch to force order.  Assumption
       is that small integers are more frequent than big ones. */
    if (t == BCF_BT_INT8) {
        *val = *(int8_t *) p++;
    } else {
        if (end - p < (1<<bcf_type_shift[t])) return -1;
        if (t == BCF_BT_INT16) {
            *val = le_to_i16(p);
            p += 2;
        } else if (t == BCF_BT_INT32) {
            *val = le_to_i32(p);
            p += 4;
#ifdef VCF_ALLOW_INT64
        } else if (t == BCF_BT_INT64) {
            // This case should never happen because there should be no
            // 64-bit BCFs at all, definitely not coming from htslib
            *val = le_to_i64(p);
            p += 8;
#endif
        } else {
            return -1;
        }
    }
    *q = p;
    return 0;
}

static int bcf_dec_size_safe(uint8_t *p, uint8_t *end, uint8_t **q,
                             int *num, int *type) {
    int r;
    if (p >= end) return -1;
    *type = *p & 0xf;
    if (*p>>4 != 15) {
        *q = p + 1;
        *num = *p >> 4;
        return 0;
    }
    r = bcf_dec_typed_int1_safe(p + 1, end, q, num);
    if (r) return r;
    return *num >= 0 ? 0 : -1;
}

static const char *get_type_name(int type) {
    const char *types[9] = {
        "null", "int (8-bit)", "int (16 bit)", "int (32 bit)",
        "unknown", "float", "unknown", "char", "unknown"
    };
    int t = (type >= 0 && type < 8) ? type : 8;
    return types[t];
}

static void bcf_record_check_err(const bcf_hdr_t *hdr, bcf1_t *rec,
                                 char *type, uint32_t *reports, int i) {
    if (*reports == 0 || hts_verbose >= HTS_LOG_DEBUG)
        hts_log_warning("Bad BCF record at %s:%"PRIhts_pos
                        ": Invalid FORMAT %s %d",
                        bcf_seqname_safe(hdr,rec), rec->pos+1, type, i);
    (*reports)++;
}

static int bcf_record_check(const bcf_hdr_t *hdr, bcf1_t *rec) {
    uint8_t *ptr, *end;
    size_t bytes;
    uint32_t err = 0;
    int type = 0;
    int num  = 0;
    int reflen = 0;
    uint32_t i, reports;
    const uint32_t is_integer = ((1 << BCF_BT_INT8)  |
                                 (1 << BCF_BT_INT16) |
#ifdef VCF_ALLOW_INT64
                                 (1 << BCF_BT_INT64) |
#endif
                                 (1 << BCF_BT_INT32));
    const uint32_t is_valid_type = (is_integer          |
                                    (1 << BCF_BT_NULL)  |
                                    (1 << BCF_BT_FLOAT) |
                                    (1 << BCF_BT_CHAR));
    int32_t max_id = hdr ? hdr->n[BCF_DT_ID] : 0;

    // Check for valid contig ID
    if (rec->rid < 0
        || (hdr && (rec->rid >= hdr->n[BCF_DT_CTG]
                    || hdr->id[BCF_DT_CTG][rec->rid].key == NULL))) {
        hts_log_warning("Bad BCF record at %"PRIhts_pos": Invalid %s id %d", rec->pos+1, "CONTIG", rec->rid);
        err |= BCF_ERR_CTG_INVALID;
    }

    // Check ID
    ptr = (uint8_t *) rec->shared.s;
    end = ptr + rec->shared.l;
    if (bcf_dec_size_safe(ptr, end, &ptr, &num, &type) != 0) goto bad_shared;
    if (type != BCF_BT_CHAR) {
        hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s type %d (%s)", bcf_seqname_safe(hdr,rec), rec->pos+1, "ID", type, get_type_name(type));
        err |= BCF_ERR_TAG_INVALID;
    }
    bytes = (size_t) num << bcf_type_shift[type];
    if (end - ptr < bytes) goto bad_shared;
    ptr += bytes;

    // Check REF and ALT
    if (rec->n_allele < 1) {
        hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": No REF allele",
                        bcf_seqname_safe(hdr,rec), rec->pos+1);
        err |= BCF_ERR_TAG_UNDEF;
    }

    reports = 0;
    for (i = 0; i < rec->n_allele; i++) {
        if (bcf_dec_size_safe(ptr, end, &ptr, &num, &type) != 0) goto bad_shared;
        if (type != BCF_BT_CHAR) {
            if (!reports++ || hts_verbose >= HTS_LOG_DEBUG)
                hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s type %d (%s)", bcf_seqname_safe(hdr,rec), rec->pos+1, "REF/ALT", type, get_type_name(type));
            err |= BCF_ERR_CHAR;
        }
        if (i == 0) reflen = num;
        bytes = (size_t) num << bcf_type_shift[type];
        if (end - ptr < bytes) goto bad_shared;
        ptr += bytes;
    }

    // Check FILTER
    reports = 0;
    if (bcf_dec_size_safe(ptr, end, &ptr, &num, &type) != 0) goto bad_shared;
    if (num > 0) {
        bytes = (size_t) num << bcf_type_shift[type];
        if (((1 << type) & is_integer) == 0) {
            hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s type %d (%s)", bcf_seqname_safe(hdr,rec), rec->pos+1, "FILTER", type, get_type_name(type));
            err |= BCF_ERR_TAG_INVALID;
            if (end - ptr < bytes) goto bad_shared;
            ptr += bytes;
        } else {
            if (end - ptr < bytes) goto bad_shared;
            for (i = 0; i < num; i++) {
                int32_t key = bcf_dec_int1(ptr, type, &ptr);
                if (key < 0
                    || (hdr && (key >= max_id
                                || hdr->id[BCF_DT_ID][key].key == NULL))) {
                    if (!reports++ || hts_verbose >= HTS_LOG_DEBUG)
                        hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s id %d", bcf_seqname_safe(hdr,rec), rec->pos+1, "FILTER", key);
                    err |= BCF_ERR_TAG_UNDEF;
                }
            }
        }
    }

    // Check INFO
    reports = 0;
    bcf_idpair_t *id_tmp = hdr ? hdr->id[BCF_DT_ID] : NULL;
    for (i = 0; i < rec->n_info; i++) {
        int32_t key = -1;
        if (bcf_dec_typed_int1_safe(ptr, end, &ptr, &key) != 0) goto bad_shared;
        if (key < 0 || (hdr && (key >= max_id
                                || id_tmp[key].key == NULL))) {
            if (!reports++ || hts_verbose >= HTS_LOG_DEBUG)
                hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s id %d", bcf_seqname_safe(hdr,rec), rec->pos+1, "INFO", key);
            err |= BCF_ERR_TAG_UNDEF;
        }
        if (bcf_dec_size_safe(ptr, end, &ptr, &num, &type) != 0) goto bad_shared;
        if (((1 << type) & is_valid_type) == 0
            || (type == BCF_BT_NULL && num > 0)) {
            if (!reports++ || hts_verbose >= HTS_LOG_DEBUG)
                hts_log_warning("Bad BCF record at %s:%"PRIhts_pos": Invalid %s type %d (%s)", bcf_seqname_safe(hdr,rec), rec->pos+1, "INFO", type, get_type_name(type));
            err |= BCF_ERR_TAG_INVALID;
        }
        bytes = (size_t) num << bcf_type_shift[type];
        if (end - ptr < bytes) goto bad_shared;
        ptr += bytes;
    }

    // Check FORMAT and individual information
    ptr = (uint8_t *) rec->indiv.s;
    end = ptr + rec->indiv.l;
    reports = 0;
    for (i = 0; i < rec->n_fmt; i++) {
        int32_t key = -1;
        if (bcf_dec_typed_int1_safe(ptr, end, &ptr, &key) != 0) goto bad_indiv;
        if (key < 0
            || (hdr && (key >= max_id
                        || id_tmp[key].key == NULL))) {
            bcf_record_check_err(hdr, rec, "id", &reports, key);
            err |= BCF_ERR_TAG_UNDEF;
        }
        if (bcf_dec_size_safe(ptr, end, &ptr, &num, &type) != 0) goto bad_indiv;
        if (((1 << type) & is_valid_type) == 0
            || (type == BCF_BT_NULL && num > 0)) {
            bcf_record_check_err(hdr, rec, "type", &reports, type);
            err |= BCF_ERR_TAG_INVALID;
        }
        bytes = ((size_t) num << bcf_type_shift[type]) * rec->n_sample;
        if (end - ptr < bytes) goto bad_indiv;
        ptr += bytes;
    }

    if (!err && rec->rlen < 0) {
        // Treat bad rlen as a warning instead of an error, and try to
        // fix up by using the length of the stored REF allele.
        static int warned = 0;
        if (!warned) {
            hts_log_warning("BCF record at %s:%"PRIhts_pos" has invalid RLEN (%"PRIhts_pos"). "
                            "Only one invalid RLEN will be reported.",
                            bcf_seqname_safe(hdr,rec), rec->pos+1, rec->rlen);
            warned = 1;
        }
        rec->rlen = reflen >= 0 ? reflen : 0;
    }

    rec->errcode |= err;

    return err ? -2 : 0; // Return -2 so bcf_read() reports an error

 bad_shared:
    hts_log_error("Bad BCF record at %s:%"PRIhts_pos" - shared section malformed or too short", bcf_seqname_safe(hdr,rec), rec->pos+1);
    return -2;

 bad_indiv:
    hts_log_error("Bad BCF record at %s:%"PRIhts_pos" - individuals section malformed or too short", bcf_seqname_safe(hdr,rec), rec->pos+1);
    return -2;
}

static inline uint8_t *bcf_unpack_fmt_core1(uint8_t *ptr, int n_sample, bcf_fmt_t *fmt);
int bcf_subset_format(const bcf_hdr_t *hdr, bcf1_t *rec)
{
    if ( !hdr->keep_samples ) return 0;
    if ( !bcf_hdr_nsamples(hdr) )
    {
        rec->indiv.l = rec->n_sample = 0;
        return 0;
    }

    int i, j;
    uint8_t *ptr = (uint8_t*)rec->indiv.s, *dst = NULL, *src;
    bcf_dec_t *dec = &rec->d;
    hts_expand(bcf_fmt_t, rec->n_fmt, dec->m_fmt, dec->fmt);
    for (i=0; i<dec->m_fmt; ++i) dec->fmt[i].p_free = 0;

    for (i=0; i<rec->n_fmt; i++)
    {
        ptr = bcf_unpack_fmt_core1(ptr, rec->n_sample, &dec->fmt[i]);
        src = dec->fmt[i].p - dec->fmt[i].size;
        if ( dst )
        {
            memmove(dec->fmt[i-1].p + dec->fmt[i-1].p_len, dec->fmt[i].p - dec->fmt[i].p_off, dec->fmt[i].p_off);
            dec->fmt[i].p = dec->fmt[i-1].p + dec->fmt[i-1].p_len + dec->fmt[i].p_off;
        }
        dst = dec->fmt[i].p;
        for (j=0; j<hdr->nsamples_ori; j++)
        {
            src += dec->fmt[i].size;
            if ( !bit_array_test(hdr->keep_samples,j) ) continue;
            memmove(dst, src, dec->fmt[i].size);
            dst += dec->fmt[i].size;
        }
        rec->indiv.l -= dec->fmt[i].p_len - (dst - dec->fmt[i].p);
        dec->fmt[i].p_len = dst - dec->fmt[i].p;
    }
    rec->unpacked |= BCF_UN_FMT;

    rec->n_sample = bcf_hdr_nsamples(hdr);
    return 0;
}

int bcf_read(htsFile *fp, const bcf_hdr_t *h, bcf1_t *v)
{
    if (fp->format.format == vcf) return vcf_read(fp,h,v);
    int ret = bcf_read1_core(fp->fp.bgzf, v);
    if (ret == 0) ret = bcf_record_check(h, v);
    if ( ret!=0 || !h->keep_samples ) return ret;
    return bcf_subset_format(h,v);
}

int bcf_readrec(BGZF *fp, void *null, void *vv, int *tid, hts_pos_t *beg, hts_pos_t *end)
{
    bcf1_t *v = (bcf1_t *) vv;
    int ret = bcf_read1_core(fp, v);
    if (ret == 0) ret = bcf_record_check(NULL, v);
    if (ret  >= 0)
        *tid = v->rid, *beg = v->pos, *end = v->pos + v->rlen;
    return ret;
}

static inline int bcf1_sync_id(bcf1_t *line, kstring_t *str)
{
    // single typed string
    if ( line->d.id && strcmp(line->d.id, ".") ) {
        return bcf_enc_vchar(str, strlen(line->d.id), line->d.id);
    } else {
        return bcf_enc_size(str, 0, BCF_BT_CHAR);
    }
}
static inline int bcf1_sync_alleles(bcf1_t *line, kstring_t *str)
{
    // list of typed strings
    int i;
    for (i=0; i<line->n_allele; i++) {
        if (bcf_enc_vchar(str, strlen(line->d.allele[i]), line->d.allele[i]) < 0)
            return -1;
    }
    if ( !line->rlen && line->n_allele ) line->rlen = strlen(line->d.allele[0]);
    return 0;
}
static inline int bcf1_sync_filter(bcf1_t *line, kstring_t *str)
{
    // typed vector of integers
    if ( line->d.n_flt ) {
        return bcf_enc_vint(str, line->d.n_flt, line->d.flt, -1);
    } else {
        return bcf_enc_vint(str, 0, 0, -1);
    }
}

static inline int bcf1_sync_info(bcf1_t *line, kstring_t *str)
{
    // pairs of typed vectors
    int i, irm = -1, e = 0;
    for (i=0; i<line->n_info; i++)
    {
        bcf_info_t *info = &line->d.info[i];
        if ( !info->vptr )
        {
            // marked for removal
            if ( irm < 0 ) irm = i;
            continue;
        }
        e |= kputsn_(info->vptr - info->vptr_off, info->vptr_len + info->vptr_off, str) < 0;
        if ( irm >=0 )
        {
            bcf_info_t tmp = line->d.info[irm]; line->d.info[irm] = line->d.info[i]; line->d.info[i] = tmp;
            while ( irm<=i && line->d.info[irm].vptr ) irm++;
        }
    }
    if ( irm>=0 ) line->n_info = irm;
    return e == 0 ? 0 : -1;
}

static int bcf1_sync(bcf1_t *line)
{
    char *shared_ori = line->shared.s;
    size_t prev_len;

    kstring_t tmp = {0,0,0};
    if ( !line->shared.l )
    {
        // New line created via API, BCF data blocks do not exist. Get it ready for BCF output
        tmp = line->shared;
        bcf1_sync_id(line, &tmp);
        line->unpack_size[0] = tmp.l; prev_len = tmp.l;

        bcf1_sync_alleles(line, &tmp);
        line->unpack_size[1] = tmp.l - prev_len; prev_len = tmp.l;

        bcf1_sync_filter(line, &tmp);
        line->unpack_size[2] = tmp.l - prev_len;

        bcf1_sync_info(line, &tmp);
        line->shared = tmp;
    }
    else if ( line->d.shared_dirty )
    {
        // The line was edited, update the BCF data block.

        if ( !(line->unpacked & BCF_UN_STR) ) bcf_unpack(line,BCF_UN_STR);

        // ptr_ori points to the original unchanged BCF data.
        uint8_t *ptr_ori = (uint8_t *) line->shared.s;

        // ID: single typed string
        if ( line->d.shared_dirty & BCF1_DIRTY_ID )
            bcf1_sync_id(line, &tmp);
        else
            kputsn_(ptr_ori, line->unpack_size[0], &tmp);
        ptr_ori += line->unpack_size[0];
        line->unpack_size[0] = tmp.l; prev_len = tmp.l;

        // REF+ALT: list of typed strings
        if ( line->d.shared_dirty & BCF1_DIRTY_ALS )
            bcf1_sync_alleles(line, &tmp);
        else
        {
            kputsn_(ptr_ori, line->unpack_size[1], &tmp);
            if ( !line->rlen && line->n_allele ) line->rlen = strlen(line->d.allele[0]);
        }
        ptr_ori += line->unpack_size[1];
        line->unpack_size[1] = tmp.l - prev_len; prev_len = tmp.l;

        if ( line->unpacked & BCF_UN_FLT )
        {
            // FILTER: typed vector of integers
            if ( line->d.shared_dirty & BCF1_DIRTY_FLT )
                bcf1_sync_filter(line, &tmp);
            else if ( line->d.n_flt )
                kputsn_(ptr_ori, line->unpack_size[2], &tmp);
            else
                bcf_enc_vint(&tmp, 0, 0, -1);
            ptr_ori += line->unpack_size[2];
            line->unpack_size[2] = tmp.l - prev_len;

            if ( line->unpacked & BCF_UN_INFO )
            {
                // INFO: pairs of typed vectors
                if ( line->d.shared_dirty & BCF1_DIRTY_INF )
                {
                    bcf1_sync_info(line, &tmp);
                    ptr_ori = (uint8_t*)line->shared.s + line->shared.l;
                }
            }
        }

        int size = line->shared.l - (size_t)ptr_ori + (size_t)line->shared.s;
        if ( size ) kputsn_(ptr_ori, size, &tmp);

        free(line->shared.s);
        line->shared = tmp;
    }
    if ( line->shared.s != shared_ori && line->unpacked & BCF_UN_INFO )
    {
        // Reallocated line->shared.s block invalidated line->d.info[].vptr pointers
        size_t off_new = line->unpack_size[0] + line->unpack_size[1] + line->unpack_size[2];
        int i;
        for (i=0; i<line->n_info; i++)
        {
            uint8_t *vptr_free = line->d.info[i].vptr_free ? line->d.info[i].vptr - line->d.info[i].vptr_off : NULL;
            line->d.info[i].vptr = (uint8_t*) line->shared.s + off_new + line->d.info[i].vptr_off;
            off_new += line->d.info[i].vptr_len + line->d.info[i].vptr_off;
            if ( vptr_free )
            {
                free(vptr_free);
                line->d.info[i].vptr_free = 0;
            }
        }
    }

    if ( line->n_sample && line->n_fmt && (!line->indiv.l || line->d.indiv_dirty) )
    {
        // The genotype fields changed or are not present
        tmp.l = tmp.m = 0; tmp.s = NULL;
        int i, irm = -1;
        for (i=0; i<line->n_fmt; i++)
        {
            bcf_fmt_t *fmt = &line->d.fmt[i];
            if ( !fmt->p )
            {
                // marked for removal
                if ( irm < 0 ) irm = i;
                continue;
            }
            kputsn_(fmt->p - fmt->p_off, fmt->p_len + fmt->p_off, &tmp);
            if ( irm >=0 )
            {
                bcf_fmt_t tfmt = line->d.fmt[irm]; line->d.fmt[irm] = line->d.fmt[i]; line->d.fmt[i] = tfmt;
                while ( irm<=i && line->d.fmt[irm].p ) irm++;
            }

        }
        if ( irm>=0 ) line->n_fmt = irm;
        free(line->indiv.s);
        line->indiv = tmp;

        // Reallocated line->indiv.s block invalidated line->d.fmt[].p pointers
        size_t off_new = 0;
        for (i=0; i<line->n_fmt; i++)
        {
            uint8_t *p_free = line->d.fmt[i].p_free ? line->d.fmt[i].p - line->d.fmt[i].p_off : NULL;
            line->d.fmt[i].p = (uint8_t*) line->indiv.s + off_new + line->d.fmt[i].p_off;
            off_new += line->d.fmt[i].p_len + line->d.fmt[i].p_off;
            if ( p_free )
            {
                free(p_free);
                line->d.fmt[i].p_free = 0;
            }
        }
    }
    if ( !line->n_sample ) line->n_fmt = 0;
    line->d.shared_dirty = line->d.indiv_dirty = 0;
    return 0;
}

bcf1_t *bcf_copy(bcf1_t *dst, bcf1_t *src)
{
    bcf1_sync(src);

    bcf_clear(dst);
    dst->rid  = src->rid;
    dst->pos  = src->pos;
    dst->rlen = src->rlen;
    dst->qual = src->qual;
    dst->n_info = src->n_info; dst->n_allele = src->n_allele;
    dst->n_fmt = src->n_fmt; dst->n_sample = src->n_sample;

    if ( dst->shared.m < src->shared.l )
    {
        dst->shared.s = (char*) realloc(dst->shared.s, src->shared.l);
        dst->shared.m = src->shared.l;
    }
    dst->shared.l = src->shared.l;
    memcpy(dst->shared.s,src->shared.s,dst->shared.l);

    if ( dst->indiv.m < src->indiv.l )
    {
        dst->indiv.s = (char*) realloc(dst->indiv.s, src->indiv.l);
        dst->indiv.m = src->indiv.l;
    }
    dst->indiv.l = src->indiv.l;
    memcpy(dst->indiv.s,src->indiv.s,dst->indiv.l);

    return dst;
}
bcf1_t *bcf_dup(bcf1_t *src)
{
    bcf1_t *out = bcf_init1();
    return bcf_copy(out, src);
}

int bcf_write(htsFile *hfp, bcf_hdr_t *h, bcf1_t *v)
{
    if ( h->dirty ) {
        if (bcf_hdr_sync(h) < 0) return -1;
    }
    if ( bcf_hdr_nsamples(h)!=v->n_sample )
    {
        hts_log_error("Broken VCF record, the number of columns at %s:%"PRIhts_pos" does not match the number of samples (%d vs %d)",
            bcf_seqname_safe(h,v), v->pos+1, v->n_sample, bcf_hdr_nsamples(h));
        return -1;
    }

    if ( hfp->format.format == vcf || hfp->format.format == text_format )
        return vcf_write(hfp,h,v);

    if ( v->errcode & ~BCF_ERR_LIMITS ) // todo: unsure about the other BCF_ERR_LIMITS branches in vcf_parse_format_alloc4()
    {
        // vcf_parse1() encountered a new contig or tag, undeclared in the
        // header.  At this point, the header must have been printed,
        // proceeding would lead to a broken BCF file. Errors must be checked
        // and cleared by the caller before we can proceed.
        char errdescription[1024] = "";
        hts_log_error("Unchecked error (%d %s) at %s:%"PRIhts_pos, v->errcode, bcf_strerror(v->errcode, errdescription, sizeof(errdescription)), bcf_seqname_safe(h,v), v->pos+1);
        return -1;
    }
    bcf1_sync(v);   // check if the BCF record was modified

    if ( v->unpacked & BCF_IS_64BIT )
    {
        hts_log_error("Data at %s:%"PRIhts_pos" contains 64-bit values not representable in BCF. Please use VCF instead", bcf_seqname_safe(h,v), v->pos+1);
        return -1;
    }

    BGZF *fp = hfp->fp.bgzf;
    uint8_t x[32];
    u32_to_le(v->shared.l + 24, x); // to include six 32-bit integers
    u32_to_le(v->indiv.l, x + 4);
    i32_to_le(v->rid, x + 8);
    u32_to_le(v->pos, x + 12);
    u32_to_le(v->rlen, x + 16);
    float_to_le(v->qual, x + 20);
    u16_to_le(v->n_info, x + 24);
    u16_to_le(v->n_allele, x + 26);
    u32_to_le((uint32_t)v->n_fmt<<24 | (v->n_sample & 0xffffff), x + 28);
    if ( bgzf_write(fp, x, 32) != 32 ) return -1;
    if ( bgzf_write(fp, v->shared.s, v->shared.l) != v->shared.l ) return -1;
    if ( bgzf_write(fp, v->indiv.s, v->indiv.l) != v->indiv.l ) return -1;

    if (hfp->idx) {
        if (bgzf_idx_push(fp, hfp->idx, v->rid, v->pos, v->pos + v->rlen,
                          bgzf_tell(fp), 1) < 0)
            return -1;
    }

    return 0;
}

/**********************
 *** VCF header I/O ***
 **********************/

static int add_missing_contig_hrec(bcf_hdr_t *h, const char *name) {
    bcf_hrec_t *hrec = calloc(1, sizeof(bcf_hrec_t));
    int save_errno;
    if (!hrec) goto fail;

    hrec->key = strdup("contig");
    if (!hrec->key) goto fail;

    if (bcf_hrec_add_key(hrec, "ID", strlen("ID")) < 0) goto fail;
    if (bcf_hrec_set_val(hrec, hrec->nkeys-1, name, strlen(name), 0) < 0)
        goto fail;
    if (bcf_hdr_add_hrec(h, hrec) < 0)
        goto fail;
    return 0;

 fail:
    save_errno = errno;
    hts_log_error("%s", strerror(errno));
    if (hrec) bcf_hrec_destroy(hrec);
    errno = save_errno;
    return -1;
}

bcf_hdr_t *vcf_hdr_read(htsFile *fp)
{
    kstring_t txt, *s = &fp->line;
    int ret;
    bcf_hdr_t *h;
    tbx_t *idx = NULL;
    const char **names = NULL;
    h = bcf_hdr_init("r");
    if (!h) {
        hts_log_error("Failed to allocate bcf header");
        return NULL;
    }
    txt.l = txt.m = 0; txt.s = 0;
    while ((ret = hts_getline(fp, KS_SEP_LINE, s)) >= 0) {
        int e = 0;
        if (s->l == 0) continue;
        if (s->s[0] != '#') {
            hts_log_error("No sample line");
            goto error;
        }
        if (s->s[1] != '#' && fp->fn_aux) { // insert contigs here
            kstring_t tmp = { 0, 0, NULL };
            hFILE *f = hopen(fp->fn_aux, "r");
            if (f == NULL) {
                hts_log_error("Couldn't open \"%s\"", fp->fn_aux);
                goto error;
            }
            while (tmp.l = 0, kgetline(&tmp, (kgets_func *) hgets, f) >= 0) {
                char *tab = strchr(tmp.s, '\t');
                if (tab == NULL) continue;
                e |= (kputs("##contig=<ID=", &txt) < 0);
                e |= (kputsn(tmp.s, tab - tmp.s, &txt) < 0);
                e |= (kputs(",length=", &txt) < 0);
                e |= (kputl(atol(tab), &txt) < 0);
                e |= (kputsn(">\n", 2, &txt) < 0);
            }
            free(tmp.s);
            if (hclose(f) != 0) {
                hts_log_error("Error on closing %s", fp->fn_aux);
                goto error;
            }
            if (e) goto error;
        }
        if (kputsn(s->s, s->l, &txt) < 0) goto error;
        if (kputc('\n', &txt) < 0) goto error;
        if (s->s[1] != '#') break;
    }
    if ( ret < -1 ) goto error;
    if ( !txt.s )
    {
        hts_log_error("Could not read the header");
        goto error;
    }
    if ( bcf_hdr_parse(h, txt.s) < 0 ) goto error;

    // check tabix index, are all contigs listed in the header? add the missing ones
    idx = tbx_index_load3(fp->fn, NULL, HTS_IDX_SILENT_FAIL);
    if ( idx )
    {
        int i, n, need_sync = 0;
        names = tbx_seqnames(idx, &n);
        if (!names) goto error;
        for (i=0; i<n; i++)
        {
            bcf_hrec_t *hrec = bcf_hdr_get_hrec(h, BCF_HL_CTG, "ID", (char*) names[i], NULL);
            if ( hrec ) continue;
            if (add_missing_contig_hrec(h, names[i]) < 0) goto error;
            need_sync = 1;
        }
        if ( need_sync ) {
            if (bcf_hdr_sync(h) < 0) goto error;
        }
        free(names);
        tbx_destroy(idx);
    }
    free(txt.s);
    return h;

 error:
    if (idx) tbx_destroy(idx);
    free(names);
    free(txt.s);
    if (h) bcf_hdr_destroy(h);
    return NULL;
}

int bcf_hdr_set(bcf_hdr_t *hdr, const char *fname)
{
    int i = 0, n = 0, save_errno;
    char **lines = hts_readlines(fname, &n);
    if ( !lines ) return 1;
    for (i=0; i<n-1; i++)
    {
        int k;
        bcf_hrec_t *hrec = bcf_hdr_parse_line(hdr,lines[i],&k);
        if (!hrec) goto fail;
        if (bcf_hdr_add_hrec(hdr, hrec) < 0) {
            bcf_hrec_destroy(hrec);
            goto fail;
        }
        free(lines[i]);
        lines[i] = NULL;
    }
    if (bcf_hdr_parse_sample_line(hdr, lines[n-1]) < 0) goto fail;
    if (bcf_hdr_sync(hdr) < 0) goto fail;
    free(lines[n-1]);
    free(lines);
    return 0;

 fail:
    save_errno = errno;
    for (; i < n; i++)
        free(lines[i]);
    free(lines);
    errno = save_errno;
    return 1;
}

static int _bcf_hrec_format(const bcf_hrec_t *hrec, int is_bcf, kstring_t *str)
{
    uint32_t e = 0;
    if ( !hrec->value )
    {
        int j, nout = 0;
        e |= ksprintf(str, "##%s=<", hrec->key) < 0;
        for (j=0; j<hrec->nkeys; j++)
        {
            // do not output IDX if output is VCF
            if ( !is_bcf && !strcmp("IDX",hrec->keys[j]) ) continue;
            if ( nout ) e |= kputc(',',str) < 0;
            e |= ksprintf(str,"%s=%s", hrec->keys[j], hrec->vals[j]) < 0;
            nout++;
        }
        e |= ksprintf(str,">\n") < 0;
    }
    else
        e |= ksprintf(str,"##%s=%s\n", hrec->key,hrec->value) < 0;

    return e == 0 ? 0 : -1;
}

int bcf_hrec_format(const bcf_hrec_t *hrec, kstring_t *str)
{
    return _bcf_hrec_format(hrec,0,str);
}

int bcf_hdr_format(const bcf_hdr_t *hdr, int is_bcf, kstring_t *str)
{
    int i, r = 0;
    for (i=0; i<hdr->nhrec; i++)
        r |= _bcf_hrec_format(hdr->hrec[i], is_bcf, str) < 0;

    r |= ksprintf(str, "#CHROM\tPOS\tID\tREF\tALT\tQUAL\tFILTER\tINFO") < 0;
    if ( bcf_hdr_nsamples(hdr) )
    {
        r |= ksprintf(str, "\tFORMAT") < 0;
        for (i=0; i<bcf_hdr_nsamples(hdr); i++)
            r |= ksprintf(str, "\t%s", hdr->samples[i]) < 0;
    }
    r |= ksprintf(str, "\n") < 0;

    return r ? -1 : 0;
}

char *bcf_hdr_fmt_text(const bcf_hdr_t *hdr, int is_bcf, int *len)
{
    kstring_t txt = {0,0,0};
    if (bcf_hdr_format(hdr, is_bcf, &txt) < 0)
        return NULL;
    if ( len ) *len = txt.l;
    return txt.s;
}

const char **bcf_hdr_seqnames(const bcf_hdr_t *h, int *n)
{
    vdict_t *d = (vdict_t*)h->dict[BCF_DT_CTG];
    int i, tid, m = kh_size(d);
    const char **names = (const char**) calloc(m,sizeof(const char*));
    if ( !names )
    {
        hts_log_error("Failed to allocate memory");
        *n = 0;
        return NULL;
    }
    khint_t k;
    for (k=kh_begin(d); k<kh_end(d); k++)
    {
        if ( !kh_exist(d,k) ) continue;
        if ( !kh_val(d, k).hrec[0] ) continue;  // removed via bcf_hdr_remove
        tid = kh_val(d,k).id;
        if ( tid >= m )
        {
            // This can happen after a contig has been removed from BCF header via bcf_hdr_remove()
            if ( hts_resize(const char*, tid + 1, &m, &names, HTS_RESIZE_CLEAR)<0 )
            {
                hts_log_error("Failed to allocate memory");
                *n = 0;
                free(names);
                return NULL;
            }
            m = tid + 1;
        }
        names[tid] = kh_key(d,k);
    }
    // ensure there are no gaps
    for (i=0,tid=0; tid<m; i++,tid++)
    {
        while ( tid<m && !names[tid] ) tid++;
        if ( tid==m ) break;
        if ( i==tid ) continue;
        names[i] = names[tid];
        names[tid] = 0;
    }
    *n = i;
    return names;
}

int vcf_hdr_write(htsFile *fp, const bcf_hdr_t *h)
{
    kstring_t htxt = {0,0,0};
    if (bcf_hdr_format(h, 0, &htxt) < 0) {
        free(htxt.s);
        return -1;
    }
    while (htxt.l && htxt.s[htxt.l-1] == '\0') --htxt.l; // kill trailing zeros
    int ret;
    if ( fp->format.compression!=no_compression ) {
        ret = bgzf_write(fp->fp.bgzf, htxt.s, htxt.l);
        if (bgzf_flush(fp->fp.bgzf) != 0) return -1;
    } else {
        ret = hwrite(fp->fp.hfile, htxt.s, htxt.l);
    }
    free(htxt.s);
    return ret<0 ? -1 : 0;
}

/***********************
 *** Typed value I/O ***
 ***********************/

int bcf_enc_vint(kstring_t *s, int n, int32_t *a, int wsize)
{
    int32_t max = INT32_MIN, min = INT32_MAX;
    int i;
    if (n <= 0) {
        return bcf_enc_size(s, 0, BCF_BT_NULL);
    } else if (n == 1) {
        return bcf_enc_int1(s, a[0]);
    } else {
        if (wsize <= 0) wsize = n;

        // Equivalent to:
        // for (i = 0; i < n; ++i) {
        //     if (a[i] == bcf_int32_missing || a[i] == bcf_int32_vector_end )
        //         continue;
        //     if (max < a[i]) max = a[i];
        //     if (min > a[i]) min = a[i];
        // }
        int max4[4] = {INT32_MIN, INT32_MIN, INT32_MIN, INT32_MIN};
        int min4[4] = {INT32_MAX, INT32_MAX, INT32_MAX, INT32_MAX};
        for (i = 0; i < (n&~3); i+=4) {
            // bcf_int32_missing    == INT32_MIN and
            // bcf_int32_vector_end == INT32_MIN+1.
            // We skip these, but can mostly avoid explicit checking
            if (max4[0] < a[i+0]) max4[0] = a[i+0];
            if (max4[1] < a[i+1]) max4[1] = a[i+1];
            if (max4[2] < a[i+2]) max4[2] = a[i+2];
            if (max4[3] < a[i+3]) max4[3] = a[i+3];
            if (min4[0] > a[i+0] && a[i+0] > INT32_MIN+1) min4[0] = a[i+0];
            if (min4[1] > a[i+1] && a[i+1] > INT32_MIN+1) min4[1] = a[i+1];
            if (min4[2] > a[i+2] && a[i+2] > INT32_MIN+1) min4[2] = a[i+2];
            if (min4[3] > a[i+3] && a[i+3] > INT32_MIN+1) min4[3] = a[i+3];
        }
        min = min4[0];
        if (min > min4[1]) min = min4[1];
        if (min > min4[2]) min = min4[2];
        if (min > min4[3]) min = min4[3];
        max = max4[0];
        if (max < max4[1]) max = max4[1];
        if (max < max4[2]) max = max4[2];
        if (max < max4[3]) max = max4[3];
        for (; i < n; ++i) {
            if (max < a[i]) max = a[i];
            if (min > a[i] && a[i] > INT32_MIN+1) min = a[i];
        }

        if (max <= BCF_MAX_BT_INT8 && min >= BCF_MIN_BT_INT8) {
            if (bcf_enc_size(s, wsize, BCF_BT_INT8) < 0 ||
                ks_resize(s, s->l + n) < 0)
                return -1;
            uint8_t *p = (uint8_t *) s->s + s->l;
            for (i = 0; i < n; ++i, p++) {
                if ( a[i]==bcf_int32_vector_end )   *p = bcf_int8_vector_end;
                else if ( a[i]==bcf_int32_missing ) *p = bcf_int8_missing;
                else *p = a[i];
            }
            s->l += n;
        } else if (max <= BCF_MAX_BT_INT16 && min >= BCF_MIN_BT_INT16) {
            uint8_t *p;
            if (bcf_enc_size(s, wsize, BCF_BT_INT16) < 0 ||
                ks_resize(s, s->l + n * sizeof(int16_t)) < 0)
                return -1;
            p = (uint8_t *) s->s + s->l;
            for (i = 0; i < n; ++i)
            {
                int16_t x;
                if ( a[i]==bcf_int32_vector_end ) x = bcf_int16_vector_end;
                else if ( a[i]==bcf_int32_missing ) x = bcf_int16_missing;
                else x = a[i];
                i16_to_le(x, p);
                p += sizeof(int16_t);
            }
            s->l += n * sizeof(int16_t);
        } else {
            uint8_t *p;
            if (bcf_enc_size(s, wsize, BCF_BT_INT32) < 0 ||
                ks_resize(s, s->l + n * sizeof(int32_t)) < 0)
                return -1;
            p = (uint8_t *) s->s + s->l;
            for (i = 0; i < n; ++i) {
                i32_to_le(a[i], p);
                p += sizeof(int32_t);
            }
            s->l += n * sizeof(int32_t);
        }
    }

    return 0;
}

#ifdef VCF_ALLOW_INT64
static int bcf_enc_long1(kstring_t *s, int64_t x) {
    uint32_t e = 0;
    if (x <= BCF_MAX_BT_INT32 && x >= BCF_MIN_BT_INT32)
        return bcf_enc_int1(s, x);
    if (x == bcf_int64_vector_end) {
        e |= bcf_enc_size(s, 1, BCF_BT_INT8);
        e |= kputc(bcf_int8_vector_end, s) < 0;
    } else if (x == bcf_int64_missing) {
        e |= bcf_enc_size(s, 1, BCF_BT_INT8);
        e |= kputc(bcf_int8_missing, s) < 0;
    } else {
        e |= bcf_enc_size(s, 1, BCF_BT_INT64);
        e |= ks_expand(s, 8);
        if (e == 0) { u64_to_le(x, (uint8_t *) s->s + s->l); s->l += 8; }
    }
    return e == 0 ? 0 : -1;
}
#endif

static inline int serialize_float_array(kstring_t *s, size_t n, const float *a) {
    uint8_t *p;
    size_t i;
    size_t bytes = n * sizeof(float);

    if (bytes / sizeof(float) != n) return -1;
    if (ks_resize(s, s->l + bytes) < 0) return -1;

    p = (uint8_t *) s->s + s->l;
    for (i = 0; i < n; i++) {
        float_to_le(a[i], p);
        p += sizeof(float);
    }
    s->l += bytes;

    return 0;
}

int bcf_enc_vfloat(kstring_t *s, int n, float *a)
{
    assert(n >= 0);
    bcf_enc_size(s, n, BCF_BT_FLOAT);
    serialize_float_array(s, n, a);
    return 0; // FIXME: check for errs in this function
}

int bcf_enc_vchar(kstring_t *s, int l, const char *a)
{
    bcf_enc_size(s, l, BCF_BT_CHAR);
    kputsn(a, l, s);
    return 0; // FIXME: check for errs in this function
}

// Special case of n==1 as it also occurs quite often in FORMAT data.
// This version is also small enough to get inlined.
static inline int bcf_fmt_array1(kstring_t *s, int type, void *data) {
    uint32_t e = 0;
    uint8_t *p = (uint8_t *)data;
    int32_t v;

    // helps gcc more than clang here. In billions of cycles:
    //          bcf_fmt_array1  bcf_fmt_array
    // gcc7:    23.2            24.3
    // gcc13:   21.6            23.0
    // clang13: 27.1            27.8
    switch (type) {
    case BCF_BT_CHAR:
        e |= kputc_(*p == bcf_str_missing ? '.' : *p, s) < 0;
        break;

    case BCF_BT_INT8:
        if (*(int8_t *)p != bcf_int8_vector_end) {
            e |= ((*(int8_t *)p == bcf_int8_missing)
                  ? kputc_('.', s)
                  : kputw(*(int8_t *)p, s)) < 0;
        }
        break;
    case BCF_BT_INT16:
        v = le_to_i16(p);
        if (v != bcf_int16_vector_end) {
            e |= (v == bcf_int16_missing
                  ? kputc_('.', s)
                  : kputw(v, s)) < 0;
        }
        break;

    case BCF_BT_INT32:
        v = le_to_i32(p);
        if (v != bcf_int32_vector_end) {
            e |= (v == bcf_int32_missing
                  ? kputc_('.', s)
                  : kputw(v, s)) < 0;
        }
        break;

    case BCF_BT_FLOAT:
        v = le_to_u32(p);
        if (v != bcf_float_vector_end) {
            e |= (v == bcf_float_missing
                  ? kputc_('.', s)
                  : kputd(le_to_float(p), s)) < 0;
        }
        break;

    default:
        hts_log_error("Unexpected type %d", type);
        return -1;
    }

    return e == 0 ? 0 : -1;
}

int bcf_fmt_array(kstring_t *s, int n, int type, void *data)
{
    int j = 0;
    uint32_t e = 0;
    if (n == 0) {
        return kputc_('.', s) >= 0 ? 0 : -1;
    }

    if (type == BCF_BT_CHAR)
    {
        char *p = (char *)data;

        // Note bcf_str_missing is already accounted for in n==0 above.
        if (n >= 8) {
            char *p_end = memchr(p, 0, n);
            e |= kputsn(p, p_end ? p_end-p : n, s) < 0;
        } else {
            for (j = 0; j < n && *p; ++j, ++p)
               e |= kputc(*p, s) < 0;
        }
    }
    else
    {
        #define BRANCH(type_t, convert, is_missing, is_vector_end, kprint) { \
            uint8_t *p = (uint8_t *) data; \
            for (j=0; j<n; j++, p += sizeof(type_t))    \
            { \
                type_t v = convert(p); \
                if ( is_vector_end ) break; \
                if ( j ) e |= kputc_(',', s) < 0; \
                e |= (is_missing ? kputc('.', s) : kprint) < 0; \
            } \
        }
        switch (type) {
            case BCF_BT_INT8:  BRANCH(int8_t,  le_to_i8, v==bcf_int8_missing,  v==bcf_int8_vector_end,  kputw(v, s)); break;
            case BCF_BT_INT16: BRANCH(int16_t, le_to_i16, v==bcf_int16_missing, v==bcf_int16_vector_end, kputw(v, s)); break;
            case BCF_BT_INT32: BRANCH(int32_t, le_to_i32, v==bcf_int32_missing, v==bcf_int32_vector_end, kputw(v, s)); break;
            case BCF_BT_FLOAT: BRANCH(uint32_t, le_to_u32, v==bcf_float_missing, v==bcf_float_vector_end, kputd(le_to_float(p), s)); break;
            default: hts_log_error("Unexpected type %d", type); exit(1); break;
        }
        #undef BRANCH
    }
    return e == 0 ? 0 : -1;
}

uint8_t *bcf_fmt_sized_array(kstring_t *s, uint8_t *ptr)
{
    int x, type;
    x = bcf_dec_size(ptr, &ptr, &type);
    bcf_fmt_array(s, x, type, ptr);
    return ptr + (x << bcf_type_shift[type]);
}

/********************
 *** VCF site I/O ***
 ********************/

typedef struct {
    int key;            // Key for h->id[BCF_DT_ID][key] vdict
    int max_m;          // number of elements in field array (ie commas)
    int size;           // field size (max_l or max_g*4 if is_gt)
    int offset;         // offset of buf into h->mem
    uint32_t is_gt:1,   // is genotype
             max_g:31;  // maximum number of genotypes
    uint32_t max_l;     // length of field
    uint32_t y;         // h->id[0][fmt[j].key].val->info[BCF_HL_FMT]
    uint8_t *buf;       // Pointer into h->mem
} fmt_aux_t;

// fmt_aux_t field notes:
// max_* are biggest sizes of the various FORMAT fields across all samples.
// We use these after pivoting the data to ensure easy random access
// of a specific sample.
//
// max_m is only used for type BCF_HT_REAL or BCF_HT_INT
// max_g is only used for is_gt == 1 (will be BCF_HT_STR)
// max_l is only used for is_gt == 0 (will be BCF_HT_STR)
//
// These are computed in vcf_parse_format_max3 and used in
// vcf_parse_format_alloc4 to get the size.
//
// size is computed from max_g, max_l, max_m and is_gt.  Once computed
// the max values are never accessed again.
//
// In theory all 4 vars could be coalesced into a single variable, but this
// significantly harms speed (even if done via a union).  It's about 25-30%
// slower.

static inline int align_mem(kstring_t *s)
{
    int e = 0;
    if (s->l&7) {
        uint64_t zero = 0;
        e = kputsn((char*)&zero, 8 - (s->l&7), s) < 0;
    }
    return e == 0 ? 0 : -1;
}

#define MAX_N_FMT 255   /* Limited by size of bcf1_t n_fmt field */

// detect FORMAT "."
static int vcf_parse_format_empty1(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                   const char *p, const char *q) {
    const char *end = s->s + s->l;
    if ( q>=end )
    {
        hts_log_error("FORMAT column with no sample columns starting at %s:%"PRIhts_pos"", bcf_seqname_safe(h,v), v->pos+1);
        v->errcode |= BCF_ERR_NCOLS;
        return -1;
    }

    v->n_fmt = 0;
    if ( p[0]=='.' && p[1]==0 ) // FORMAT field is empty "."
    {
        v->n_sample = bcf_hdr_nsamples(h);
        return 1;
    }

    return 0;
}

// get format information from the dictionary
static int vcf_parse_format_dict2(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                  const char *p, const char *q, fmt_aux_t *fmt) {
    const vdict_t *d = (vdict_t*)h->dict[BCF_DT_ID];
    char *t;
    int j;
    ks_tokaux_t aux1;

    for (j = 0, t = kstrtok(p, ":", &aux1); t; t = kstrtok(0, 0, &aux1), ++j) {
        if (j >= MAX_N_FMT) {
            v->errcode |= BCF_ERR_LIMITS;
            hts_log_error("FORMAT column at %s:%"PRIhts_pos" lists more identifiers than htslib can handle",
                bcf_seqname_safe(h,v), v->pos+1);
            return -1;
        }

        *(char*)aux1.p = 0;
        khint_t k = kh_get(vdict, d, t);
        if (k == kh_end(d) || kh_val(d, k).info[BCF_HL_FMT] == 15) {
            if ( t[0]=='.' && t[1]==0 )
            {
                hts_log_error("Invalid FORMAT tag name '.' at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_TAG_INVALID;
                return -1;
            }
            hts_log_warning("FORMAT '%s' at %s:%"PRIhts_pos" is not defined in the header, assuming Type=String", t, bcf_seqname_safe(h,v), v->pos+1);
            kstring_t tmp = {0,0,0};
            int l;
            ksprintf(&tmp, "##FORMAT=<ID=%s,Number=1,Type=String,Description=\"Dummy\">", t);
            bcf_hrec_t *hrec = bcf_hdr_parse_line(h,tmp.s,&l);
            free(tmp.s);
            int res = hrec ? bcf_hdr_add_hrec((bcf_hdr_t*)h, hrec) : -1;
            if (res < 0) bcf_hrec_destroy(hrec);
            if (res > 0) res = bcf_hdr_sync((bcf_hdr_t*)h);

            k = kh_get(vdict, d, t);
            v->errcode |= BCF_ERR_TAG_UNDEF;
            if (res || k == kh_end(d)) {
                hts_log_error("Could not add dummy header for FORMAT '%s' at %s:%"PRIhts_pos, t, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_TAG_INVALID;
                return -1;
            }
        }
        fmt[j].max_l = fmt[j].max_m = fmt[j].max_g = 0;
        fmt[j].key = kh_val(d, k).id;
        fmt[j].is_gt = (t[0] == 'G' && t[1] == 'T' && !t[2]);
        fmt[j].y = h->id[0][fmt[j].key].val->info[BCF_HL_FMT];
        v->n_fmt++;
    }
    return 0;
}

// compute max
static int vcf_parse_format_max3(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                 char *p, char *q, fmt_aux_t *fmt) {
    int n_sample_ori = -1;
    char *r = q + 1;  // r: position in the format string
    int l = 0, m = 1, g = 1, j;
    v->n_sample = 0;  // m: max vector size, l: max field len, g: max number of alleles
    const char *end = s->s + s->l;

    while ( r<end )
    {
        // can we skip some samples?
        if ( h->keep_samples )
        {
            n_sample_ori++;
            if ( !bit_array_test(h->keep_samples,n_sample_ori) )
            {
                while ( *r!='\t' && r<end ) r++;
                if ( *r=='\t' ) { *r = 0; r++; }
                continue;
            }
        }

        // collect fmt stats: max vector size, length, number of alleles
        j = 0;  // j-th format field
        fmt_aux_t *f = fmt;
        static char meta[256] = {
            // \0 \t , / : |
            1,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,1, 0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
            0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
        };

        char *r_start = r;
        for (;;) {
            // Quickly skip ahead to an appropriate meta-character
            while (!meta[(unsigned char)*r]) r++;

            switch (*r) {
            case ',':
                m++;
                break;

            case '|':
            case '/':
                if (f->is_gt) g++;
                break;

            case '\t':
                *r = 0; // fall through

            default: // valid due to while loop above.
            case '\0':
            case ':':
                l = r - r_start; r_start = r;
                if (f->max_m < m) f->max_m = m;
                if (f->max_l < l) f->max_l = l;
                if (f->is_gt && f->max_g < g) f->max_g = g;
                l = 0, m = g = 1;
                if ( *r==':' ) {
                    j++; f++;
                    if ( j>=v->n_fmt ) {
                        hts_log_error("Incorrect number of FORMAT fields at %s:%"PRIhts_pos"",
                                      h->id[BCF_DT_CTG][v->rid].key, v->pos+1);
                        v->errcode |= BCF_ERR_NCOLS;
                        return -1;
                    }
                } else goto end_for;
                break;
            }
            if ( r>=end ) break;
            r++;
        }
    end_for:
        v->n_sample++;
        if ( v->n_sample == bcf_hdr_nsamples(h) ) break;
        r++;
    }

    return 0;
}

// allocate memory for arrays
static int vcf_parse_format_alloc4(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                   const char *p, const char *q,
                                   fmt_aux_t *fmt) {
    kstring_t *mem = (kstring_t*)&h->mem;

    int j;
    for (j = 0; j < v->n_fmt; ++j) {
        fmt_aux_t *f = &fmt[j];
        if ( !f->max_m ) f->max_m = 1;  // omitted trailing format field

        if ((f->y>>4&0xf) == BCF_HT_STR) {
            f->size = f->is_gt? f->max_g << 2 : f->max_l;
        } else if ((f->y>>4&0xf) == BCF_HT_REAL || (f->y>>4&0xf) == BCF_HT_INT) {
            f->size = f->max_m << 2;
        } else {
            hts_log_error("The format type %d at %s:%"PRIhts_pos" is currently not supported", f->y>>4&0xf, bcf_seqname_safe(h,v), v->pos+1);
            v->errcode |= BCF_ERR_TAG_INVALID;
            return -1;
        }

        if (align_mem(mem) < 0) {
            hts_log_error("Memory allocation failure at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
            v->errcode |= BCF_ERR_LIMITS;
            return -1;
        }

        // Limit the total memory to ~2Gb per VCF row.  This should mean
        // malformed VCF data is less likely to take excessive memory and/or
        // time.
        if ((uint64_t) mem->l + v->n_sample * (uint64_t)f->size > INT_MAX) {
            static int warned = 0;
            if ( !warned ) hts_log_warning("Excessive memory required by FORMAT fields at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
            warned = 1;
            v->errcode |= BCF_ERR_LIMITS;
            f->size = -1;
            f->offset = 0;
            continue;
        }

        f->offset = mem->l;
        if (ks_resize(mem, mem->l + v->n_sample * (size_t)f->size) < 0) {
            hts_log_error("Memory allocation failure at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
            v->errcode |= BCF_ERR_LIMITS;
            return -1;
        }
        mem->l += v->n_sample * f->size;
    }

    {
        int j;
        for (j = 0; j < v->n_fmt; ++j)
            fmt[j].buf = (uint8_t*)mem->s + fmt[j].offset;
    }

    // check for duplicate tags
    int i;
    for (i=1; i<v->n_fmt; i++)
    {
        fmt_aux_t *ifmt = &fmt[i];
        if ( ifmt->size==-1 ) continue; // already marked for removal
        for (j=0; j<i; j++)
        {
            fmt_aux_t *jfmt = &fmt[j];
            if ( jfmt->size==-1 ) continue; // already marked for removal
            if ( ifmt->key!=jfmt->key ) continue;
            static int warned = 0;
            if ( !warned ) hts_log_warning("Duplicate FORMAT tag %s at %s:%"PRIhts_pos, bcf_hdr_int2id(h,BCF_DT_ID,ifmt->key), bcf_seqname_safe(h,v), v->pos+1);
            warned = 1;
            v->errcode |= BCF_ERR_TAG_INVALID;
            ifmt->size = -1;
            ifmt->offset = 0;
            break;
        }
    }
    return 0;
}

// Fill the sample fields
static int vcf_parse_format_fill5(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                  const char *p, const char *q, fmt_aux_t *fmt) {
    static int extreme_val_warned = 0;
    int n_sample_ori = -1;
    // At beginning of the loop t points to the first char of a format
    const char *t = q + 1;
    int m = 0;   // m: sample id
    const int nsamples = bcf_hdr_nsamples(h);

    const char *end = s->s + s->l;
    while ( t<end )
    {
        // can we skip some samples?
        if ( h->keep_samples )
        {
            n_sample_ori++;
            if ( !bit_array_test(h->keep_samples,n_sample_ori) )
            {
                while ( *t && t<end ) t++;
                t++;
                continue;
            }
        }
        if ( m == nsamples ) break;

        int j = 0; // j-th format field, m-th sample
        while ( t < end )
        {
            fmt_aux_t *z = &fmt[j++];
            const int htype = z->y>>4&0xf;
            if (!z->buf) {
                hts_log_error("Memory allocation failure for FORMAT field type %d at %s:%"PRIhts_pos,
                              z->y>>4&0xf, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_LIMITS;
                return -1;
            }

            if ( z->size==-1 )
            {
                // this field is to be ignored, it's either too big or a duplicate
                while ( *t != ':' && *t ) t++;
            }
            else if (htype == BCF_HT_STR) {
                int l;
                if (z->is_gt) {
                    // Genotypes.
                    // <val>([|/]<val>)+... where <val> is [0-9]+ or ".".
                    int32_t is_phased = 0;
                    uint32_t *x = (uint32_t*)(z->buf + z->size * (size_t)m);
                    uint32_t unreadable = 0;
                    uint32_t max = 0;
                    int overflow = 0;
                    for (l = 0;; ++t) {
                        if (*t == '.') {
                            ++t, x[l++] = is_phased;
                        } else {
                            const char *tt = t;
                            uint32_t val;
                            // Or "v->n_allele < 10", but it doesn't
                            // seem to be any faster and this feels safer.
                            if (*t >= '0' && *t <= '9' &&
                                !(t[1] >= '0' && t[1] <= '9')) {
                                val = *t++ - '0';
                            } else {
                                val = hts_str2uint(t, (char **)&t,
                                                   sizeof(val) * CHAR_MAX - 2,
                                                   &overflow);
                                unreadable |= tt == t;
                            }
                            if (max < val) max = val;
                            x[l++] = (val + 1) << 1 | is_phased;
                        }
                        is_phased = (*t == '|');
                        if (*t != '|' && *t != '/') break;
                    }
                    // Possibly check max against v->n_allele instead?
                    if (overflow || max > (INT32_MAX >> 1) - 1) {
                        hts_log_error("Couldn't read GT data: value too large at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
                        return -1;
                    }
                    if (unreadable) {
                        hts_log_error("Couldn't read GT data: value not a number or '.' at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
                        return -1;
                    }
                    if ( !l ) x[l++] = 0;   // An empty field, insert missing value
                    for (; l < z->size>>2; ++l)
                        x[l] = bcf_int32_vector_end;

                } else {
                    // Otherwise arbitrary strings
                    char *x = (char*)z->buf + z->size * (size_t)m;
                    for (l = 0; *t != ':' && *t; ++t)
                        x[l++] = *t;
                    if (z->size > l)
                        memset(&x[l], 0, (z->size-l) * sizeof(*x));
                }

            } else if (htype == BCF_HT_INT) {
                // One or more integers in an array
                int32_t *x = (int32_t*)(z->buf + z->size * (size_t)m);
                int l;
                for (l = 0;; ++t) {
                    if (*t == '.') {
                        x[l++] = bcf_int32_missing, ++t; // ++t to skip "."
                    } else {
                        int overflow = 0;
                        char *te;
                        long int tmp_val = hts_str2int(t, &te, sizeof(tmp_val)*CHAR_BIT, &overflow);
                        if ( te==t || overflow || tmp_val<BCF_MIN_BT_INT32 || tmp_val>BCF_MAX_BT_INT32 )
                        {
                            if ( !extreme_val_warned )
                            {
                                hts_log_warning("Extreme FORMAT/%s value encountered and set to missing at %s:%"PRIhts_pos,
                                                h->id[BCF_DT_ID][fmt[j-1].key].key, bcf_seqname_safe(h,v), v->pos+1);
                                extreme_val_warned = 1;
                            }
                            tmp_val = bcf_int32_missing;
                        }
                        x[l++] = tmp_val;
                        t = te;
                    }
                    if (*t != ',') break;
                }
                if ( !l )
                    x[l++] = bcf_int32_missing;
                for (; l < z->size>>2; ++l)
                    x[l] = bcf_int32_vector_end;

            } else if (htype == BCF_HT_REAL) {
                // One of more floating point values in an array
                float *x = (float*)(z->buf + z->size * (size_t)m);
                int l;
                for (l = 0;; ++t) {
                    if (*t == '.' && !isdigit_c(t[1])) {
                        bcf_float_set_missing(x[l++]), ++t; // ++t to skip "."
                    } else {
                        int overflow = 0;
                        char *te;
                        float tmp_val = hts_str2dbl(t, &te, &overflow);
                        if ( (te==t || overflow) && !extreme_val_warned )
                        {
                            hts_log_warning("Extreme FORMAT/%s value encountered at %s:%"PRIhts_pos, h->id[BCF_DT_ID][fmt[j-1].key].key, bcf_seqname(h,v), v->pos+1);
                            extreme_val_warned = 1;
                        }
                        x[l++] = tmp_val;
                        t = te;
                    }
                    if (*t != ',') break;
                }
                if ( !l )
                    // An empty field, insert missing value
                    bcf_float_set_missing(x[l++]);
                for (; l < z->size>>2; ++l)
                    bcf_float_set_vector_end(x[l]);
            } else {
                hts_log_error("Unknown FORMAT field type %d at %s:%"PRIhts_pos, htype, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_TAG_INVALID;
                return -1;
            }

            if (*t == '\0') {
                break;
            }
            else if (*t == ':') {
                t++;
            }
            else {
                char buffer[8];
                hts_log_error("Invalid character %s in '%s' FORMAT field at %s:%"PRIhts_pos"",
                    hts_strprint(buffer, sizeof buffer, '\'', t, 1),
                    h->id[BCF_DT_ID][z->key].key, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_CHAR;
                return -1;
            }
        }

        // fill end-of-vector values
        for (; j < v->n_fmt; ++j) {
            fmt_aux_t *z = &fmt[j];
            const int htype = z->y>>4&0xf;
            int l;

            if (z->size == -1) // this field is to be ignored
                continue;

            if (htype == BCF_HT_STR) {
                if (z->is_gt) {
                    int32_t *x = (int32_t*)(z->buf + z->size * (size_t)m);
                    if (z->size) x[0] = bcf_int32_missing;
                    for (l = 1; l < z->size>>2; ++l) x[l] = bcf_int32_vector_end;
                } else {
                    char *x = (char*)z->buf + z->size * (size_t)m;
                    if ( z->size ) {
                        x[0] = '.';
                        memset(&x[1], 0, (z->size-1) * sizeof(*x));
                    }
                }
            } else if (htype == BCF_HT_INT) {
                int32_t *x = (int32_t*)(z->buf + z->size * (size_t)m);
                x[0] = bcf_int32_missing;
                for (l = 1; l < z->size>>2; ++l) x[l] = bcf_int32_vector_end;
            } else if (htype == BCF_HT_REAL) {
                float *x = (float*)(z->buf + z->size * (size_t)m);
                bcf_float_set_missing(x[0]);
                for (l = 1; l < z->size>>2; ++l) bcf_float_set_vector_end(x[l]);
            }
        }

        m++; t++;
    }

    return 0;
}

// write individual genotype information
static int vcf_parse_format_gt6(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                                const char *p, const char *q, fmt_aux_t *fmt) {
    kstring_t *str = &v->indiv;
    int i, need_downsize = 0;
    if (v->n_sample > 0) {
        for (i = 0; i < v->n_fmt; ++i) {
            fmt_aux_t *z = &fmt[i];
            if ( z->size==-1 ) {
                need_downsize = 1;
                continue;
            }
            bcf_enc_int1(str, z->key);
            if ((z->y>>4&0xf) == BCF_HT_STR && !z->is_gt) {
                bcf_enc_size(str, z->size, BCF_BT_CHAR);
                kputsn((char*)z->buf, z->size * (size_t)v->n_sample, str);
            } else if ((z->y>>4&0xf) == BCF_HT_INT || z->is_gt) {
                bcf_enc_vint(str, (z->size>>2) * v->n_sample, (int32_t*)z->buf, z->size>>2);
            } else {
                bcf_enc_size(str, z->size>>2, BCF_BT_FLOAT);
                if (serialize_float_array(str, (z->size>>2) * (size_t)v->n_sample,
                                          (float *) z->buf) != 0) {
                    v->errcode |= BCF_ERR_LIMITS;
                    hts_log_error("Out of memory at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
                    return -1;
                }
            }
        }

    }
    if ( need_downsize ) {
        i = 0;
        while ( i < v->n_fmt ) {
            if ( fmt[i].size==-1 )
            {
                v->n_fmt--;
                if ( i < v->n_fmt ) memmove(&fmt[i],&fmt[i+1],sizeof(*fmt)*(v->n_fmt-i));
            }
            else
                i++;
        }
    }
    return 0;
}

// validity checking
static int vcf_parse_format_check7(const bcf_hdr_t *h, bcf1_t *v) {
    if ( v->n_sample!=bcf_hdr_nsamples(h) )
    {
        hts_log_error("Number of columns at %s:%"PRIhts_pos" does not match the number of samples (%d vs %d)",
            bcf_seqname_safe(h,v), v->pos+1, v->n_sample, bcf_hdr_nsamples(h));
        v->errcode |= BCF_ERR_NCOLS;
        return -1;
    }
    if ( v->indiv.l > 0xffffffff )
    {
        hts_log_error("The FORMAT at %s:%"PRIhts_pos" is too long", bcf_seqname_safe(h,v), v->pos+1);
        v->errcode |= BCF_ERR_LIMITS;

        // Error recovery: return -1 if this is a critical error or 0 if we want to ignore the FORMAT and proceed
        v->n_fmt = 0;
        return -1;
    }

    return 0;
}

// p,q is the start and the end of the FORMAT field
static int vcf_parse_format(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v,
                            char *p, char *q)
{
    if ( !bcf_hdr_nsamples(h) ) return 0;
    kstring_t *mem = (kstring_t*)&h->mem;
    mem->l = 0;

    fmt_aux_t fmt[MAX_N_FMT];

    // detect FORMAT "."
    int ret; // +ve = ok, -ve = err
    if ((ret = vcf_parse_format_empty1(s, h, v, p, q)))
        return ret ? 0 : -1;

    // get format information from the dictionary
    if (vcf_parse_format_dict2(s, h, v, p, q, fmt) < 0)
        return -1;

    // FORMAT data is per-sample A:B:C A:B:C A:B:C ... but in memory it is
    // stored as per-type arrays AAA... BBB... CCC...  This is basically
    // a data rotation or pivot.

    // The size of elements in the array grow to their maximum needed,
    // permitting fast random access.  This means however we have to first
    // scan the whole FORMAT line to find the maximum of each type, and
    // then scan it again to find the store the data.
    // We break this down into compute-max, allocate, fill-out-buffers

    // TODO: ?
    // The alternative would be to pivot on the first pass, with fixed
    // size entries for numerics and concatenated strings otherwise, also
    // tracking maximum sizes.  Then on a second pass we reallocate and
    // copy the data again to a uniformly sized array.  Two passes through
    // memory, but without doubling string parsing.

    // compute max
    if (vcf_parse_format_max3(s, h, v, p, q, fmt) < 0)
        return -1;

    // allocate memory for arrays
    if (vcf_parse_format_alloc4(s, h, v, p, q, fmt) < 0)
        return -1;

    // fill the sample fields; at beginning of the loop
    if (vcf_parse_format_fill5(s, h, v, p, q, fmt) < 0)
        return -1;

    // write individual genotype information
    if (vcf_parse_format_gt6(s, h, v, p, q, fmt) < 0)
        return -1;

    // validity checking
    if (vcf_parse_format_check7(h, v) < 0)
        return -1;

    return 0;
}

static khint_t fix_chromosome(const bcf_hdr_t *h, vdict_t *d, const char *p) {
    // Simple error recovery for chromosomes not defined in the header. It will not help when VCF header has
    // been already printed, but will enable tools like vcfcheck to proceed.

    kstring_t tmp = {0,0,0};
    khint_t k;
    int l;
    if (ksprintf(&tmp, "##contig=<ID=%s>", p) < 0)
        return kh_end(d);
    bcf_hrec_t *hrec = bcf_hdr_parse_line(h,tmp.s,&l);
    free(tmp.s);
    int res = hrec ? bcf_hdr_add_hrec((bcf_hdr_t*)h, hrec) : -1;
    if (res < 0) bcf_hrec_destroy(hrec);
    if (res > 0) res = bcf_hdr_sync((bcf_hdr_t*)h);
    k = kh_get(vdict, d, p);

    return k;
}

static int vcf_parse_filter(kstring_t *str, const bcf_hdr_t *h, bcf1_t *v, char *p, char *q) {
    int i, n_flt = 1, max_n_flt = 0;
    char *r, *t;
    int32_t *a_flt = NULL;
    ks_tokaux_t aux1;
    khint_t k;
    vdict_t *d = (vdict_t*)h->dict[BCF_DT_ID];
    // count the number of filters
    if (*(q-1) == ';') *(q-1) = 0;
    for (r = p; *r; ++r)
        if (*r == ';') ++n_flt;
    if (n_flt > max_n_flt) {
        a_flt = malloc(n_flt * sizeof(*a_flt));
        if (!a_flt) {
            hts_log_error("Could not allocate memory at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
            v->errcode |= BCF_ERR_LIMITS; // No appropriate code?
            return -1;
        }
        max_n_flt = n_flt;
    }
    // add filters
    for (t = kstrtok(p, ";", &aux1), i = 0; t; t = kstrtok(0, 0, &aux1)) {
        *(char*)aux1.p = 0;
        k = kh_get(vdict, d, t);
        if (k == kh_end(d))
        {
            // Simple error recovery for FILTERs not defined in the header. It will not help when VCF header has
            // been already printed, but will enable tools like vcfcheck to proceed.
            hts_log_warning("FILTER '%s' is not defined in the header", t);
            kstring_t tmp = {0,0,0};
            int l;
            ksprintf(&tmp, "##FILTER=<ID=%s,Description=\"Dummy\">", t);
            bcf_hrec_t *hrec = bcf_hdr_parse_line(h,tmp.s,&l);
            free(tmp.s);
            int res = hrec ? bcf_hdr_add_hrec((bcf_hdr_t*)h, hrec) : -1;
            if (res < 0) bcf_hrec_destroy(hrec);
            if (res > 0) res = bcf_hdr_sync((bcf_hdr_t*)h);
            k = kh_get(vdict, d, t);
            v->errcode |= BCF_ERR_TAG_UNDEF;
            if (res || k == kh_end(d)) {
                hts_log_error("Could not add dummy header for FILTER '%s' at %s:%"PRIhts_pos, t, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_TAG_INVALID;
                free(a_flt);
                return -1;
            }
        }
        a_flt[i++] = kh_val(d, k).id;
    }

    bcf_enc_vint(str, n_flt, a_flt, -1);
    free(a_flt);

    return 0;
}

static int vcf_parse_info(kstring_t *str, const bcf_hdr_t *h, bcf1_t *v, char *p, char *q) {
    static int extreme_int_warned = 0, negative_rlen_warned = 0;
    int max_n_val = 0, overflow = 0;
    char *r, *key;
    khint_t k;
    vdict_t *d = (vdict_t*)h->dict[BCF_DT_ID];
    int32_t *a_val = NULL;

    v->n_info = 0;
    if (*(q-1) == ';') *(q-1) = 0;
    for (r = key = p;; ++r) {
        int c;
        char *val, *end;
        while (*r > '=' || (*r != ';' && *r != '=' && *r != 0)) r++;
        if (v->n_info == UINT16_MAX) {
            hts_log_error("Too many INFO entries at %s:%"PRIhts_pos,
                          bcf_seqname_safe(h,v), v->pos+1);
            v->errcode |= BCF_ERR_LIMITS;
            goto fail;
        }
        val = end = NULL;
        c = *r; *r = 0;
        if (c == '=') {
            val = r + 1;

            for (end = val; *end != ';' && *end != 0; ++end);
            c = *end; *end = 0;
        } else end = r;
        if ( !*key ) { if (c==0) break; r = end; key = r + 1; continue; }  // faulty VCF, ";;" in the INFO
        k = kh_get(vdict, d, key);
        if (k == kh_end(d) || kh_val(d, k).info[BCF_HL_INFO] == 15)
        {
            hts_log_warning("INFO '%s' is not defined in the header, assuming Type=String", key);
            kstring_t tmp = {0,0,0};
            int l;
            ksprintf(&tmp, "##INFO=<ID=%s,Number=1,Type=String,Description=\"Dummy\">", key);
            bcf_hrec_t *hrec = bcf_hdr_parse_line(h,tmp.s,&l);
            free(tmp.s);
            int res = hrec ? bcf_hdr_add_hrec((bcf_hdr_t*)h, hrec) : -1;
            if (res < 0) bcf_hrec_destroy(hrec);
            if (res > 0) res = bcf_hdr_sync((bcf_hdr_t*)h);
            k = kh_get(vdict, d, key);
            v->errcode |= BCF_ERR_TAG_UNDEF;
            if (res || k == kh_end(d)) {
                hts_log_error("Could not add dummy header for INFO '%s' at %s:%"PRIhts_pos, key, bcf_seqname_safe(h,v), v->pos+1);
                v->errcode |= BCF_ERR_TAG_INVALID;
                goto fail;
            }
        }
        uint32_t y = kh_val(d, k).info[BCF_HL_INFO];
        ++v->n_info;
        bcf_enc_int1(str, kh_val(d, k).id);
        if (val == 0) {
            bcf_enc_size(str, 0, BCF_BT_NULL);
        } else if ((y>>4&0xf) == BCF_HT_FLAG || (y>>4&0xf) == BCF_HT_STR) { // if Flag has a value, treat it as a string
            bcf_enc_vchar(str, end - val, val);
        } else { // int/float value/array
            int i, n_val;
            char *t, *te;
            for (t = val, n_val = 1; *t; ++t) // count the number of values
                if (*t == ',') ++n_val;
            // Check both int and float size in one step for simplicity
            if (n_val > max_n_val) {
                int32_t *a_tmp = (int32_t *)realloc(a_val, n_val * sizeof(*a_val));
                if (!a_tmp) {
                    hts_log_error("Could not allocate memory at %s:%"PRIhts_pos, bcf_seqname_safe(h,v), v->pos+1);
                    v->errcode |= BCF_ERR_LIMITS; // No appropriate code?
                    goto fail;
                }
                a_val = a_tmp;
                max_n_val = n_val;
            }
            if ((y>>4&0xf) == BCF_HT_INT) {
                i = 0, t = val;
                int64_t val1;
                int is_int64 = 0;
#ifdef VCF_ALLOW_INT64
                if ( n_val==1 )
                {
                    overflow = 0;
                    long long int tmp_val = hts_str2int(val, &te, sizeof(tmp_val)*CHAR_BIT, &overflow);
                    if ( te==val ) tmp_val = bcf_int32_missing;
                    else if ( overflow || tmp_val<BCF_MIN_BT_INT64 || tmp_val>BCF_MAX_BT_INT64 )
                    {
                        if ( !extreme_int_warned )
                        {
                            hts_log_warning("Extreme INFO/%s value encountered and set to missing at %s:%"PRIhts_pos,key,bcf_seqname_safe(h,v), v->pos+1);
                            extreme_int_warned = 1;
                        }
                        tmp_val = bcf_int32_missing;
                    }
                    else
                        is_int64 = 1;
                    val1 = tmp_val;
                    t = te;
                    i = 1;  // this is just to avoid adding another nested block...
                }
#endif
                for (; i < n_val; ++i, ++t)
                {
                    overflow = 0;
                    long int tmp_val = hts_str2int(t, &te, sizeof(tmp_val)*CHAR_BIT, &overflow);
                    if ( te==t ) tmp_val = bcf_int32_missing;
                    else if ( overflow || tmp_val<BCF_MIN_BT_INT32 || tmp_val>BCF_MAX_BT_INT32 )
                    {
                        if ( !extreme_int_warned )
                        {
                            hts_log_warning("Extreme INFO/%s value encountered and set to missing at %s:%"PRIhts_pos,key,bcf_seqname_safe(h,v), v->pos+1);
                            extreme_int_warned = 1;
                        }
                        tmp_val = bcf_int32_missing;
                    }
                    a_val[i] = tmp_val;
                    for (t = te; *t && *t != ','; t++);
                }
                if (n_val == 1) {
#ifdef VCF_ALLOW_INT64
                    if ( is_int64 )
                    {
                        v->unpacked |= BCF_IS_64BIT;
                        bcf_enc_long1(str, val1);
                    }
                    else
                        bcf_enc_int1(str, (int32_t)val1);
#else
                    val1 = a_val[0];
                    bcf_enc_int1(str, (int32_t)val1);
#endif
                } else {
                    bcf_enc_vint(str, n_val, a_val, -1);
                }
                if (n_val==1 && (val1!=bcf_int32_missing || is_int64)
                    && memcmp(key, "END", 4) == 0)
                {
                    if ( val1 <= v->pos )
                    {
                        if ( !negative_rlen_warned )
                        {
                            hts_log_warning("INFO/END=%"PRIhts_pos" is smaller than POS at %s:%"PRIhts_pos,val1,bcf_seqname_safe(h,v),v->pos+1);
                            negative_rlen_warned = 1;
                        }
                    }
                    else
                        v->rlen = val1 - v->pos;
                }
            } else if ((y>>4&0xf) == BCF_HT_REAL) {
                float *val_f = (float *)a_val;
                for (i = 0, t = val; i < n_val; ++i, ++t)
                {
                    overflow = 0;
                    val_f[i] = hts_str2dbl(t, &te, &overflow);
                    if ( te==t || overflow ) // conversion failed
                        bcf_float_set_missing(val_f[i]);
                    for (t = te; *t && *t != ','; t++);
                }
                bcf_enc_vfloat(str, n_val, val_f);
            }
        }
        if (c == 0) break;
        r = end;
        key = r + 1;
    }

    free(a_val);
    return 0;

 fail:
    free(a_val);
    return -1;
}

int vcf_parse(kstring_t *s, const bcf_hdr_t *h, bcf1_t *v)
{
    int ret = -2, overflow = 0;
    char *p, *q, *r, *t;
    kstring_t *str;
    khint_t k;
    ks_tokaux_t aux;

//#define NOT_DOT(p) strcmp((p), ".")
//#define NOT_DOT(p) (!(*p == '.' && !p[1]))
//#define NOT_DOT(p) ((*p) != '.' || (p)[1])
//#define NOT_DOT(p) (q-p != 1 || memcmp(p, ".\0", 2))
#define NOT_DOT(p) (memcmp(p, ".\0", 2))

    if (!s || !h || !v || !(s->s))
        return ret;

    // Assumed in lots of places, but we may as well spot this early
    assert(sizeof(float) == sizeof(int32_t));

    // Ensure string we parse has space to permit some over-flow when during
    // parsing.  Eg to do memcmp(key, "END", 4) in vcf_parse_info over
    // the more straight forward looking strcmp, giving a speed advantage.
    if (ks_resize(s, s->l+4) < 0)
        return -1;

    // Force our memory to be initialised so we avoid the technicality of
    // undefined behaviour in using a 4-byte memcmp.  (The reality is this
    // almost certainly is never detected by the compiler so has no impact,
    // but equally so this code has minimal (often beneficial) impact on
    // performance too.)
    s->s[s->l+0] = 0;
    s->s[s->l+1] = 0;
    s->s[s->l+2] = 0;
    s->s[s->l+3] = 0;

    bcf_clear1(v);
    str = &v->shared;
    memset(&aux, 0, sizeof(ks_tokaux_t));

    // CHROM
    if (!(p = kstrtok(s->s, "\t", &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    vdict_t *d = (vdict_t*)h->dict[BCF_DT_CTG];
    k = kh_get(vdict, d, p);
    if (k == kh_end(d)) {
        hts_log_warning("Contig '%s' is not defined in the header. (Quick workaround: index the file with tabix.)", p);
        v->errcode = BCF_ERR_CTG_UNDEF;
        if ((k = fix_chromosome(h, d, p)) == kh_end(d)) {
            hts_log_error("Could not add dummy header for contig '%s'", p);
            v->errcode |= BCF_ERR_CTG_INVALID;
            goto err;
        }
    }
    v->rid = kh_val(d, k).id;

    // POS
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    overflow = 0;
    char *tmp = p;
    v->pos = hts_str2uint(p, &p, 62, &overflow);
    if (overflow) {
        hts_log_error("Position value '%s' is too large", tmp);
        goto err;
    } else if ( *p ) {
        hts_log_error("Could not parse the position '%s'", tmp);
        goto err;
    } else {
        v->pos -= 1;
    }
    if (v->pos >= INT32_MAX)
        v->unpacked |= BCF_IS_64BIT;

    // ID
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    if (NOT_DOT(p)) bcf_enc_vchar(str, q - p, p);
    else bcf_enc_size(str, 0, BCF_BT_CHAR);

    // REF
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    bcf_enc_vchar(str, q - p, p);
    v->n_allele = 1, v->rlen = q - p;

    // ALT
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    if (NOT_DOT(p)) {
        for (r = t = p;; ++r) {
            if (*r == ',' || *r == 0) {
                if (v->n_allele == UINT16_MAX) {
                    hts_log_error("Too many ALT alleles at %s:%"PRIhts_pos,
                                  bcf_seqname_safe(h,v), v->pos+1);
                    v->errcode |= BCF_ERR_LIMITS;
                    goto err;
                }
                bcf_enc_vchar(str, r - t, t);
                t = r + 1;
                ++v->n_allele;
            }
            if (r == q) break;
        }
    }

    // QUAL
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    if (NOT_DOT(p)) v->qual = atof(p);
    else bcf_float_set_missing(v->qual);
    if ( v->max_unpack && !(v->max_unpack>>1) ) goto end; // BCF_UN_STR

    // FILTER
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    if (NOT_DOT(p)) {
        if (vcf_parse_filter(str, h, v, p, q)) {
            goto err;
        }
    } else bcf_enc_vint(str, 0, 0, -1);
    if ( v->max_unpack && !(v->max_unpack>>2) ) goto end; // BCF_UN_FLT

    // INFO
    if (!(p = kstrtok(0, 0, &aux)))
        goto err;
    *(q = (char*)aux.p) = 0;

    if (NOT_DOT(p)) {
        if (vcf_parse_info(str, h, v, p, q)) {
            goto err;
        }
    }
    if ( v->max_unpack && !(v->max_unpack>>3) ) goto end;

    // FORMAT; optional
    p = kstrtok(0, 0, &aux);
    if (p) {
        *(q = (char*)aux.p) = 0;

        return vcf_parse_format(s, h, v, p, q) == 0 ? 0 : -2;
    } else {
        return 0;
    }

 end:
    ret = 0;

 err:
    return ret;
}

int vcf_open_mode(char *mode, const char *fn, const char *format)
{
    if (format == NULL) {
        // Try to pick a format based on the filename extension
        char extension[HTS_MAX_EXT_LEN];
        if (find_file_extension(fn, extension) < 0) return -1;
        return vcf_open_mode(mode, fn, extension);
    }
    else if (strcasecmp(format, "bcf") == 0) strcpy(mode, "b");
    else if (strcasecmp(format, "vcf") == 0) strcpy(mode, "");
    else if (strcasecmp(format, "vcf.gz") == 0 || strcasecmp(format, "vcf.bgz") == 0) strcpy(mode, "z");
    else return -1;

    return 0;
}

int vcf_read(htsFile *fp, const bcf_hdr_t *h, bcf1_t *v)
{
    int ret;
    ret = hts_getline(fp, KS_SEP_LINE, &fp->line);
    if (ret < 0) return ret;
    return vcf_parse1(&fp->line, h, v);
}

static inline uint8_t *bcf_unpack_fmt_core1(uint8_t *ptr, int n_sample, bcf_fmt_t *fmt)
{
    uint8_t *ptr_start = ptr;
    fmt->id = bcf_dec_typed_int1(ptr, &ptr);
    fmt->n = bcf_dec_size(ptr, &ptr, &fmt->type);
    fmt->size = fmt->n << bcf_type_shift[fmt->type];
    fmt->p = ptr;
    fmt->p_off  = ptr - ptr_start;
    fmt->p_free = 0;
    ptr += n_sample * fmt->size;
    fmt->p_len = ptr - fmt->p;
    return ptr;
}

static inline uint8_t *bcf_unpack_info_core1(uint8_t *ptr, bcf_info_t *info)
{
    uint8_t *ptr_start = ptr;
    int64_t len = 0;
    info->key = bcf_dec_typed_int1(ptr, &ptr);
    len = info->len = bcf_dec_size(ptr, &ptr, &info->type);
    info->vptr = ptr;
    info->vptr_off  = ptr - ptr_start;
    info->vptr_free = 0;
    info->v1.i = 0;
    if (info->len == 1) {
        switch(info->type) {
        case BCF_BT_INT8:
        case BCF_BT_CHAR:
            info->v1.i = *(int8_t*)ptr;
            break;
        case BCF_BT_INT16:
            info->v1.i = le_to_i16(ptr);
            len <<= 1;
            break;
        case BCF_BT_INT32:
            info->v1.i = le_to_i32(ptr);
            len <<= 2;
            break;
        case BCF_BT_FLOAT:
            info->v1.f = le_to_float(ptr);
            len <<= 2;
            break;
        case BCF_BT_INT64:
            info->v1.i = le_to_i64(ptr);
            len <<= 3;
            break;
        }
    } else {
        len <<= bcf_type_shift[info->type];
    }
    ptr += len;

    info->vptr_len = ptr - info->vptr;
    return ptr;
}

int bcf_unpack(bcf1_t *b, int which)
{
    if ( !b->shared.l ) return 0; // Building a new BCF record from scratch
    uint8_t *ptr = (uint8_t*)b->shared.s, *ptr_ori;
    int i;
    bcf_dec_t *d = &b->d;
    if (which & BCF_UN_FLT) which |= BCF_UN_STR;
    if (which & BCF_UN_INFO) which |= BCF_UN_SHR;
    if ((which&BCF_UN_STR) && !(b->unpacked&BCF_UN_STR))
    {
        kstring_t tmp;

        // ID
        tmp.l = 0; tmp.s = d->id; tmp.m = d->m_id;
        ptr_ori = ptr;
        ptr = bcf_fmt_sized_array(&tmp, ptr);
        b->unpack_size[0] = ptr - ptr_ori;
        kputc_('\0', &tmp);
        d->id = tmp.s; d->m_id = tmp.m;

        // REF and ALT are in a single block (d->als) and d->alleles are pointers into this block
        hts_expand(char*, b->n_allele, d->m_allele, d->allele); // NM: hts_expand() is a macro
        tmp.l = 0; tmp.s = d->als; tmp.m = d->m_als;
        ptr_ori = ptr;
        for (i = 0; i < b->n_allele; ++i) {
            // Use offset within tmp.s as realloc may change pointer
            d->allele[i] = (char *)(intptr_t)tmp.l;
            ptr = bcf_fmt_sized_array(&tmp, ptr);
            kputc_('\0', &tmp);
        }
        b->unpack_size[1] = ptr - ptr_ori;
        d->als = tmp.s; d->m_als = tmp.m;

        // Convert our offsets within tmp.s back to pointers again
        for (i = 0; i < b->n_allele; ++i)
            d->allele[i] = d->als + (ptrdiff_t)d->allele[i];
        b->unpacked |= BCF_UN_STR;
    }
    if ((which&BCF_UN_FLT) && !(b->unpacked&BCF_UN_FLT)) { // FILTER
        ptr = (uint8_t*)b->shared.s + b->unpack_size[0] + b->unpack_size[1];
        ptr_ori = ptr;
        if (*ptr>>4) {
            int type;
            d->n_flt = bcf_dec_size(ptr, &ptr, &type);
            hts_expand(int, d->n_flt, d->m_flt, d->flt);
            for (i = 0; i < d->n_flt; ++i)
                d->flt[i] = bcf_dec_int1(ptr, type, &ptr);
        } else ++ptr, d->n_flt = 0;
        b->unpack_size[2] = ptr - ptr_ori;
        b->unpacked |= BCF_UN_FLT;
    }
    if ((which&BCF_UN_INFO) && !(b->unpacked&BCF_UN_INFO)) { // INFO
        ptr = (uint8_t*)b->shared.s + b->unpack_size[0] + b->unpack_size[1] + b->unpack_size[2];
        hts_expand(bcf_info_t, b->n_info, d->m_info, d->info);
        for (i = 0; i < d->m_info; ++i) d->info[i].vptr_free = 0;
        for (i = 0; i < b->n_info; ++i)
            ptr = bcf_unpack_info_core1(ptr, &d->info[i]);
        b->unpacked |= BCF_UN_INFO;
    }
    if ((which&BCF_UN_FMT) && b->n_sample && !(b->unpacked&BCF_UN_FMT)) { // FORMAT
        ptr = (uint8_t*)b->indiv.s;
        hts_expand(bcf_fmt_t, b->n_fmt, d->m_fmt, d->fmt);
        for (i = 0; i < d->m_fmt; ++i) d->fmt[i].p_free = 0;
        for (i = 0; i < b->n_fmt; ++i)
            ptr = bcf_unpack_fmt_core1(ptr, b->n_sample, &d->fmt[i]);
        b->unpacked |= BCF_UN_FMT;
    }
    return 0;
}

int vcf_format(const bcf_hdr_t *h, const bcf1_t *v, kstring_t *s)
{
    int i;
    int32_t max_dt_id = h->n[BCF_DT_ID];
    const char *chrom = bcf_seqname(h, v);
    if (!chrom) {
        hts_log_error("Invalid BCF, CONTIG id=%d not present in the header",
                      v->rid);
        errno = EINVAL;
        return -1;
    }

    bcf_unpack((bcf1_t*)v, BCF_UN_ALL & ~(BCF_UN_INFO|BCF_UN_FMT));

    // Cache of key lengths so we don't keep repeatedly using them.
    // This assumes we're not modifying the header between successive calls
    // to vcf_format, but that would lead to many other forms of breakage
    // so it feels like a valid assumption to make.
    //
    // We cannot just do this in bcf_hdr_sync as some code (eg bcftools
    // annotate) manipulates the headers directly without calling sync to
    // refresh the data structures.  So we must do just-in-time length
    // calculation during writes instead.
    bcf_hdr_aux_t *aux = get_hdr_aux(h);
    if (!aux->key_len) {
        if (!(aux->key_len = calloc(h->n[BCF_DT_ID]+1, sizeof(*aux->key_len))))
            return -1;
    }
    size_t *key_len = aux->key_len;

    kputs(chrom, s); // CHROM
    kputc_('\t', s); kputll(v->pos + 1, s); // POS
    kputc_('\t', s); kputs(v->d.id ? v->d.id : ".", s); // ID
    kputc_('\t', s); // REF
    if (v->n_allele > 0) kputs(v->d.allele[0], s);
    else kputc_('.', s);
    kputc_('\t', s); // ALT
    if (v->n_allele > 1) {
        for (i = 1; i < v->n_allele; ++i) {
            if (i > 1) kputc_(',', s);
            kputs(v->d.allele[i], s);
        }
    } else kputc_('.', s);
    kputc_('\t', s); // QUAL
    if ( bcf_float_is_missing(v->qual) ) kputc_('.', s); // QUAL
    else kputd(v->qual, s);
    kputc_('\t', s); // FILTER
    if (v->d.n_flt) {
        for (i = 0; i < v->d.n_flt; ++i) {
            int32_t idx = v->d.flt[i];
            if (idx < 0 || idx >= max_dt_id
                || h->id[BCF_DT_ID][idx].key == NULL) {
                hts_log_error("Invalid BCF, the FILTER tag id=%d at %s:%"PRIhts_pos" not present in the header",
                              idx, bcf_seqname_safe(h, v), v->pos + 1);
                errno = EINVAL;
                return -1;
            }
            if (i) kputc_(';', s);
            if (!key_len[idx])
                key_len[idx] = strlen(h->id[BCF_DT_ID][idx].key);
            kputsn(h->id[BCF_DT_ID][idx].key, key_len[idx], s);
        }
    } else kputc_('.', s);

    kputc_('\t', s); // INFO
    if (v->n_info) {
        uint8_t *ptr = v->shared.s
            ? (uint8_t *)v->shared.s + v->unpack_size[0] +
               v->unpack_size[1] + v->unpack_size[2]
            : NULL;
        int first = 1;
        bcf_info_t *info = v->d.info;

        // Note if we duplicate this code into custom packed and unpacked
        // implementations then we gain a bit more speed, particularly with
        // clang 13 (up to 5%).  Not sure why this is, but code duplication
        // isn't pleasant and it's still faster adding packed support than
        // not so it's a win, just not as good as it should be.
        const int info_packed = !(v->unpacked & BCF_UN_INFO) && v->shared.l;
        for (i = 0; i < v->n_info; ++i) {
            bcf_info_t in, *z;
            if (info_packed) {
                // Use a local bcf_info_t when data is packed
                z = &in;
                z->key  = bcf_dec_typed_int1(ptr, &ptr);
                z->len  = bcf_dec_size(ptr, &ptr, &z->type);
                z->vptr = ptr;
                ptr += z->len << bcf_type_shift[z->type];
            } else {
                // Else previously unpacked INFO struct
                z = &info[i];

                // Also potentially since deleted
                if ( !z->vptr ) continue;
            }

            bcf_idpair_t *id = z->key >= 0 && z->key < max_dt_id
                ? &h->id[BCF_DT_ID][z->key]
                : NULL;

            if (!id || !id->key) {
                hts_log_error("Invalid BCF, the INFO tag id=%d is %s at %s:%"PRIhts_pos,
                              z->key,
                              z->key < 0 ? "negative"
                              : (z->key >= max_dt_id ? "too large" : "not present in the header"),
                              bcf_seqname_safe(h, v), v->pos+1);
                errno = EINVAL;
                return -1;
            }

            // KEY
            if (!key_len[z->key])
                key_len[z->key] = strlen(id->key);
            size_t id_len = key_len[z->key];
            if (ks_resize(s, s->l + 3 + id_len) < 0)
                return -1;
            char *sptr = s->s + s->l;
            if ( !first ) {
                *sptr++ = ';';
                s->l++;
            }
            first = 0;
            memcpy(sptr, id->key, id_len);
            s->l += id_len;

            // VALUE
            if (z->len <= 0) continue;
            sptr[id_len] = '=';
            s->l++;

            if (z->len != 1 || info_packed) {
                bcf_fmt_array(s, z->len, z->type, z->vptr);
            } else {
                // Single length vectors are unpacked into their
                // own info.v1 union and handled separately.
                if (z->type == BCF_BT_FLOAT) {
                    if ( bcf_float_is_missing(z->v1.f) )
                        kputc_('.', s);
                    else
                        kputd(z->v1.f, s);
                } else if (z->type == BCF_BT_CHAR) {
                    kputc_(z->v1.i, s);
                } else if (z->type < BCF_BT_INT64) {
                    int64_t missing[] = {
                        0, // BCF_BT_NULL
                        bcf_int8_missing,
                        bcf_int16_missing,
                        bcf_int32_missing,
                    };
                    if (z->v1.i == missing[z->type])
                        kputc_('.', s);
                    else
                        kputw(z->v1.i, s);
                } else if (z->type == BCF_BT_INT64) {
                    if (z->v1.i == bcf_int64_missing)
                        kputc_('.', s);
                    else
                        kputll(z->v1.i, s);
                } else {
                    hts_log_error("Unexpected type %d at %s:%"PRIhts_pos, z->type, bcf_seqname_safe(h, v), v->pos+1);
                    errno = EINVAL;
                    return -1;
                }
            }
        }
        if ( first ) kputc_('.', s);
    } else kputc_('.', s);

    // FORMAT and individual information
    if (v->n_sample) {
        int i,j;
        if ( v->n_fmt) {
            uint8_t *ptr = (uint8_t *)v->indiv.s;
            int gt_i = -1;
            bcf_fmt_t *fmt = v->d.fmt;
            int first = 1;
            int fmt_packed = !(v->unpacked & BCF_UN_FMT);

            if (fmt_packed) {
                // Local fmt as we have an array of num FORMAT keys,
                // each of which points to N.Sample values.

                // No real gain to be had in handling unpacked data here,
                // but it doesn't cost us much in complexity either and
                // it gives us flexibility.
                fmt = malloc(v->n_fmt * sizeof(*fmt));
                if (!fmt)
                    return -1;
            }

            // KEYS
            for (i = 0; i < (int)v->n_fmt; ++i) {
                bcf_fmt_t *z;
                z = &fmt[i];
                if (fmt_packed) {
                    z->id   = bcf_dec_typed_int1(ptr, &ptr);
                    z->n    = bcf_dec_size(ptr, &ptr, &z->type);
                    z->p    = ptr;
                    z->size = z->n << bcf_type_shift[z->type];
                    ptr += v->n_sample * z->size;
                }
                if ( !z->p ) continue;
                kputc_(!first ? ':' : '\t', s); first = 0;

                bcf_idpair_t *id = z->id >= 0 && z->id < max_dt_id
                    ? &h->id[BCF_DT_ID][z->id]
                    : NULL;

                if (!id || !id->key) {
                    hts_log_error("Invalid BCF, the FORMAT tag id=%d at %s:%"PRIhts_pos" not present in the header", z->id, bcf_seqname_safe(h, v), v->pos+1);
                    errno = EINVAL;
                    return -1;
                }

                if (!key_len[z->id])
                    key_len[z->id] = strlen(id->key);
                size_t id_len = key_len[z->id];
                kputsn(id->key, id_len, s);
                if (id_len == 2 && id->key[0] == 'G' && id->key[1] == 'T')
                    gt_i = i;
            }
            if ( first ) kputsn("\t.", 2, s);

            // VALUES per sample
            for (j = 0; j < v->n_sample; ++j) {
                kputc_('\t', s);
                first = 1;
                bcf_fmt_t *f = fmt;
                for (i = 0; i < (int)v->n_fmt; i++, f++) {
                    if ( !f->p ) continue;
                    if (!first) kputc_(':', s);
                    first = 0;
                    if (gt_i == i) {
                        bcf_format_gt(f,j,s);
                        break;
                    }
                    else if (f->n == 1)
                        bcf_fmt_array1(s, f->type, f->p + j * (size_t)f->size);
                    else
                        bcf_fmt_array(s, f->n, f->type, f->p + j * (size_t)f->size);
                }

                // Simpler loop post GT and at least 1 iteration
                for (i++, f++; i < (int)v->n_fmt; i++, f++) {
                    if ( !f->p ) continue;
                    kputc_(':', s);
                    if (f->n == 1)
                        bcf_fmt_array1(s, f->type, f->p + j * (size_t)f->size);
                    else
                        bcf_fmt_array(s, f->n, f->type, f->p + j * (size_t)f->size);
                }
                if ( first ) kputc_('.', s);
            }
            if (fmt_packed)
                free(fmt);
        }
        else
            for (j=0; j<=v->n_sample; j++)
                kputsn("\t.", 2, s);
    }
    kputc('\n', s);
    return 0;
}

int vcf_write_line(htsFile *fp, kstring_t *line)
{
    int ret;
    if ( line->s[line->l-1]!='\n' ) kputc('\n',line);
    if ( fp->format.compression!=no_compression )
        ret = bgzf_write(fp->fp.bgzf, line->s, line->l);
    else
        ret = hwrite(fp->fp.hfile, line->s, line->l);
    return ret==line->l ? 0 : -1;
}

int vcf_write(htsFile *fp, const bcf_hdr_t *h, bcf1_t *v)
{
    ssize_t ret;
    fp->line.l = 0;
    if (vcf_format1(h, v, &fp->line) != 0)
        return -1;
    if ( fp->format.compression!=no_compression ) {
        if (bgzf_flush_try(fp->fp.bgzf, fp->line.l) < 0)
            return -1;
        if (fp->idx && !fp->fp.bgzf->mt)
            hts_idx_amend_last(fp->idx, bgzf_tell(fp->fp.bgzf));
        ret = bgzf_write(fp->fp.bgzf, fp->line.s, fp->line.l);
    } else {
        ret = hwrite(fp->fp.hfile, fp->line.s, fp->line.l);
    }

    if (fp->idx && fp->format.compression == bgzf) {
        int tid;
        if ((tid = hts_idx_tbi_name(fp->idx, v->rid, bcf_seqname_safe(h, v))) < 0)
            return -1;

        if (bgzf_idx_push(fp->fp.bgzf, fp->idx,
                          tid, v->pos, v->pos + v->rlen,
                          bgzf_tell(fp->fp.bgzf), 1) < 0)
            return -1;
    }

    return ret==fp->line.l ? 0 : -1;
}

/************************
 * Data access routines *
 ************************/

int bcf_hdr_id2int(const bcf_hdr_t *h, int which, const char *id)
{
    khint_t k;
    vdict_t *d = (vdict_t*)h->dict[which];
    k = kh_get(vdict, d, id);
    return k == kh_end(d)? -1 : kh_val(d, k).id;
}


/********************
 *** BCF indexing ***
 ********************/

// Calculate number of index levels given min_shift and the header contig
// list.  Also returns number of contigs in *nids_out.
static int idx_calc_n_lvls_ids(const bcf_hdr_t *h, int min_shift,
                               int starting_n_lvls, int *nids_out)
{
    int n_lvls, i, nids = 0;
    int64_t max_len = 0, s;

    for (i = 0; i < h->n[BCF_DT_CTG]; ++i)
    {
        if ( !h->id[BCF_DT_CTG][i].val ) continue;
        if ( max_len < h->id[BCF_DT_CTG][i].val->info[0] )
            max_len = h->id[BCF_DT_CTG][i].val->info[0];
        nids++;
    }
    if ( !max_len ) max_len = (1LL<<31) - 1;  // In case contig line is broken.
    max_len += 256;
    s = hts_bin_maxpos(min_shift, starting_n_lvls);
    for (n_lvls = starting_n_lvls; max_len > s; ++n_lvls, s <<= 3);

    if (nids_out) *nids_out = nids;
    return n_lvls;
}

hts_idx_t *bcf_index(htsFile *fp, int min_shift)
{
    int n_lvls;
    bcf1_t *b = NULL;
    hts_idx_t *idx = NULL;
    bcf_hdr_t *h;
    int r;
    h = bcf_hdr_read(fp);
    if ( !h ) return NULL;
    int nids = 0;
    n_lvls = idx_calc_n_lvls_ids(h, min_shift, 0, &nids);
    idx = hts_idx_init(nids, HTS_FMT_CSI, bgzf_tell(fp->fp.bgzf), min_shift, n_lvls);
    if (!idx) goto fail;
    b = bcf_init1();
    if (!b) goto fail;
    while ((r = bcf_read1(fp,h, b)) >= 0) {
        int ret;
        ret = hts_idx_push(idx, b->rid, b->pos, b->pos + b->rlen, bgzf_tell(fp->fp.bgzf), 1);
        if (ret < 0) goto fail;
    }
    if (r < -1) goto fail;
    hts_idx_finish(idx, bgzf_tell(fp->fp.bgzf));
    bcf_destroy1(b);
    bcf_hdr_destroy(h);
    return idx;

 fail:
    hts_idx_destroy(idx);
    bcf_destroy1(b);
    bcf_hdr_destroy(h);
    return NULL;
}

hts_idx_t *bcf_index_load2(const char *fn, const char *fnidx)
{
    return fnidx? hts_idx_load2(fn, fnidx) : bcf_index_load(fn);
}

hts_idx_t *bcf_index_load3(const char *fn, const char *fnidx, int flags)
{
    return hts_idx_load3(fn, fnidx, HTS_FMT_CSI, flags);
}

int bcf_index_build3(const char *fn, const char *fnidx, int min_shift, int n_threads)
{
    htsFile *fp;
    hts_idx_t *idx;
    tbx_t *tbx;
    int ret;
    if ((fp = hts_open(fn, "rb")) == 0) return -2;
    if (n_threads)
        hts_set_threads(fp, n_threads);
    if ( fp->format.compression!=bgzf ) { hts_close(fp); return -3; }
    switch (fp->format.format) {
        case bcf:
            if (!min_shift) {
                hts_log_error("TBI indices for BCF files are not supported");
                ret = -1;
            } else {
                idx = bcf_index(fp, min_shift);
                if (idx) {
                    ret = hts_idx_save_as(idx, fn, fnidx, HTS_FMT_CSI);
                    if (ret < 0) ret = -4;
                    hts_idx_destroy(idx);
                }
                else ret = -1;
            }
            break;

        case vcf:
            tbx = tbx_index(hts_get_bgzfp(fp), min_shift, &tbx_conf_vcf);
            if (tbx) {
                ret = hts_idx_save_as(tbx->idx, fn, fnidx, min_shift > 0 ? HTS_FMT_CSI : HTS_FMT_TBI);
                if (ret < 0) ret = -4;
                tbx_destroy(tbx);
            }
            else ret = -1;
            break;

        default:
            ret = -3;
            break;
    }
    hts_close(fp);
    return ret;
}

int bcf_index_build2(const char *fn, const char *fnidx, int min_shift)
{
    return bcf_index_build3(fn, fnidx, min_shift, 0);
}

int bcf_index_build(const char *fn, int min_shift)
{
    return bcf_index_build3(fn, NULL, min_shift, 0);
}

// Initialise fp->idx for the current format type.
// This must be called after the header has been written but no other data.
static int vcf_idx_init(htsFile *fp, bcf_hdr_t *h, int min_shift, const char *fnidx) {
    int n_lvls, fmt;

    if (min_shift == 0) {
        min_shift = 14;
        n_lvls = 5;
        fmt = HTS_FMT_TBI;
    } else {
        // Set initial n_lvls to match tbx_index()
        int starting_n_lvls = (TBX_MAX_SHIFT - min_shift + 2) / 3;
        // Increase if necessary
        n_lvls = idx_calc_n_lvls_ids(h, min_shift, starting_n_lvls, NULL);
        fmt = HTS_FMT_CSI;
    }

    fp->idx = hts_idx_init(0, fmt, bgzf_tell(fp->fp.bgzf), min_shift, n_lvls);
    if (!fp->idx) return -1;

    // Tabix meta data, added even in CSI for VCF
    uint8_t conf[4*7];
    u32_to_le(TBX_VCF, conf+0);  // fmt
    u32_to_le(1,       conf+4);  // name col
    u32_to_le(2,       conf+8);  // beg col
    u32_to_le(0,       conf+12); // end col
    u32_to_le('#',     conf+16); // comment
    u32_to_le(0,       conf+20); // n.skip
    u32_to_le(0,       conf+24); // ref name len
    if (hts_idx_set_meta(fp->idx, sizeof(conf)*sizeof(*conf), (uint8_t *)conf, 1) < 0) {
        hts_idx_destroy(fp->idx);
        fp->idx = NULL;
        return -1;
    }
    fp->fnidx = fnidx;

    return 0;
}

// Initialise fp->idx for the current format type.
// This must be called after the header has been written but no other data.
int bcf_idx_init(htsFile *fp, bcf_hdr_t *h, int min_shift, const char *fnidx) {
    int n_lvls, nids = 0;

    if (fp->format.compression != bgzf) {
        hts_log_error("Indexing is only supported on BGZF-compressed files");
        return -3; // Matches no-compression return for bcf_index_build3()
    }

    if (fp->format.format == vcf)
        return vcf_idx_init(fp, h, min_shift, fnidx);

    if (!min_shift)
        min_shift = 14;

    n_lvls = idx_calc_n_lvls_ids(h, min_shift, 0, &nids);

    fp->idx = hts_idx_init(nids, HTS_FMT_CSI, bgzf_tell(fp->fp.bgzf), min_shift, n_lvls);
    if (!fp->idx) return -1;
    fp->fnidx = fnidx;

    return 0;
}

// Finishes an index. Call after the last record has been written.
// Returns 0 on success, <0 on failure.
//
// NB: same format as SAM/BAM as it uses bgzf.
int bcf_idx_save(htsFile *fp) {
    return sam_idx_save(fp);
}

/*****************
 *** Utilities ***
 *****************/

int bcf_hdr_combine(bcf_hdr_t *dst, const bcf_hdr_t *src)
{
    int i, ndst_ori = dst->nhrec, need_sync = 0, ret = 0, res;
    for (i=0; i<src->nhrec; i++)
    {
        if ( src->hrec[i]->type==BCF_HL_GEN && src->hrec[i]->value )
        {
            int j;
            for (j=0; j<ndst_ori; j++)
            {
                if ( dst->hrec[j]->type!=BCF_HL_GEN ) continue;

                // Checking only the key part of generic lines, otherwise
                // the VCFs are too verbose. Should we perhaps add a flag
                // to bcf_hdr_combine() and make this optional?
                if ( !strcmp(src->hrec[i]->key,dst->hrec[j]->key) ) break;
            }
            if ( j>=ndst_ori ) {
                res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                if (res < 0) return -1;
                need_sync += res;
            }
        }
        else if ( src->hrec[i]->type==BCF_HL_STR )
        {
            // NB: we are ignoring fields without ID
            int j = bcf_hrec_find_key(src->hrec[i],"ID");
            if ( j>=0 )
            {
                bcf_hrec_t *rec = bcf_hdr_get_hrec(dst, src->hrec[i]->type, "ID", src->hrec[i]->vals[j], src->hrec[i]->key);
                if ( !rec ) {
                    res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                    if (res < 0) return -1;
                    need_sync += res;
                }
            }
        }
        else
        {
            int j = bcf_hrec_find_key(src->hrec[i],"ID");
            assert( j>=0 ); // this should always be true for valid VCFs

            bcf_hrec_t *rec = bcf_hdr_get_hrec(dst, src->hrec[i]->type, "ID", src->hrec[i]->vals[j], NULL);
            if ( !rec ) {
                res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                if (res < 0) return -1;
                need_sync += res;
            } else if ( src->hrec[i]->type==BCF_HL_INFO || src->hrec[i]->type==BCF_HL_FMT )
            {
                // Check that both records are of the same type. The bcf_hdr_id2length
                // macro cannot be used here because dst header is not synced yet.
                vdict_t *d_src = (vdict_t*)src->dict[BCF_DT_ID];
                vdict_t *d_dst = (vdict_t*)dst->dict[BCF_DT_ID];
                khint_t k_src  = kh_get(vdict, d_src, src->hrec[i]->vals[0]);
                khint_t k_dst  = kh_get(vdict, d_dst, src->hrec[i]->vals[0]);
                if ( (kh_val(d_src,k_src).info[rec->type]>>8 & 0xf) != (kh_val(d_dst,k_dst).info[rec->type]>>8 & 0xf) )
                {
                    hts_log_warning("Trying to combine \"%s\" tag definitions of different lengths",
                        src->hrec[i]->vals[0]);
                    ret |= 1;
                }
                if ( (kh_val(d_src,k_src).info[rec->type]>>4 & 0xf) != (kh_val(d_dst,k_dst).info[rec->type]>>4 & 0xf) )
                {
                    hts_log_warning("Trying to combine \"%s\" tag definitions of different types",
                        src->hrec[i]->vals[0]);
                    ret |= 1;
                }
            }
        }
    }
    if ( need_sync ) {
        if (bcf_hdr_sync(dst) < 0) return -1;
    }
    return ret;
}

bcf_hdr_t *bcf_hdr_merge(bcf_hdr_t *dst, const bcf_hdr_t *src)
{
    if ( !dst )
    {
        // this will effectively strip existing IDX attributes from src to become dst
        dst = bcf_hdr_init("r");
        kstring_t htxt = {0,0,0};
        if (bcf_hdr_format(src, 0, &htxt) < 0) {
            free(htxt.s);
            return NULL;
        }
        if ( bcf_hdr_parse(dst, htxt.s) < 0 ) {
            bcf_hdr_destroy(dst);
            dst = NULL;
        }
        free(htxt.s);
        return dst;
    }

    int i, ndst_ori = dst->nhrec, need_sync = 0, res;
    for (i=0; i<src->nhrec; i++)
    {
        if ( src->hrec[i]->type==BCF_HL_GEN && src->hrec[i]->value )
        {
            int j;
            for (j=0; j<ndst_ori; j++)
            {
                if ( dst->hrec[j]->type!=BCF_HL_GEN ) continue;

                // Checking only the key part of generic lines, otherwise
                // the VCFs are too verbose. Should we perhaps add a flag
                // to bcf_hdr_combine() and make this optional?
                if ( !strcmp(src->hrec[i]->key,dst->hrec[j]->key) ) break;
            }
            if ( j>=ndst_ori ) {
                res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                if (res < 0) return NULL;
                need_sync += res;
            }
        }
        else if ( src->hrec[i]->type==BCF_HL_STR )
        {
            // NB: we are ignoring fields without ID
            int j = bcf_hrec_find_key(src->hrec[i],"ID");
            if ( j>=0 )
            {
                bcf_hrec_t *rec = bcf_hdr_get_hrec(dst, src->hrec[i]->type, "ID", src->hrec[i]->vals[j], src->hrec[i]->key);
                if ( !rec ) {
                    res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                    if (res < 0) return NULL;
                    need_sync += res;
                }
            }
        }
        else
        {
            int j = bcf_hrec_find_key(src->hrec[i],"ID");
            assert( j>=0 ); // this should always be true for valid VCFs

            bcf_hrec_t *rec = bcf_hdr_get_hrec(dst, src->hrec[i]->type, "ID", src->hrec[i]->vals[j], NULL);
            if ( !rec ) {
                res = bcf_hdr_add_hrec(dst, bcf_hrec_dup(src->hrec[i]));
                if (res < 0) return NULL;
                need_sync += res;
            } else if ( src->hrec[i]->type==BCF_HL_INFO || src->hrec[i]->type==BCF_HL_FMT )
            {
                // Check that both records are of the same type. The bcf_hdr_id2length
                // macro cannot be used here because dst header is not synced yet.
                vdict_t *d_src = (vdict_t*)src->dict[BCF_DT_ID];
                vdict_t *d_dst = (vdict_t*)dst->dict[BCF_DT_ID];
                khint_t k_src  = kh_get(vdict, d_src, src->hrec[i]->vals[0]);
                khint_t k_dst  = kh_get(vdict, d_dst, src->hrec[i]->vals[0]);
                if ( (kh_val(d_src,k_src).info[rec->type]>>8 & 0xf) != (kh_val(d_dst,k_dst).info[rec->type]>>8 & 0xf) )
                {
                    hts_log_warning("Trying to combine \"%s\" tag definitions of different lengths",
                        src->hrec[i]->vals[0]);
                }
                if ( (kh_val(d_src,k_src).info[rec->type]>>4 & 0xf) != (kh_val(d_dst,k_dst).info[rec->type]>>4 & 0xf) )
                {
                    hts_log_warning("Trying to combine \"%s\" tag definitions of different types",
                        src->hrec[i]->vals[0]);
                }
            }
        }
    }
    if ( need_sync ) {
        if (bcf_hdr_sync(dst) < 0) return NULL;
    }
    return dst;
}

int bcf_translate(const bcf_hdr_t *dst_hdr, bcf_hdr_t *src_hdr, bcf1_t *line)
{
    int i;
    if ( line->errcode )
    {
        char errordescription[1024] = "";
        hts_log_error("Unchecked error (%d %s) at %s:%"PRIhts_pos", exiting", line->errcode, bcf_strerror(line->errcode, errordescription, sizeof(errordescription)),  bcf_seqname_safe(src_hdr,line), line->pos+1);
        exit(1);
    }
    if ( src_hdr->ntransl==-1 ) return 0;    // no need to translate, all tags have the same id
    if ( !src_hdr->ntransl )  // called for the first time, see what needs translating
    {
        int dict;
        for (dict=0; dict<2; dict++)    // BCF_DT_ID and BCF_DT_CTG
        {
            src_hdr->transl[dict] = (int*) malloc(src_hdr->n[dict]*sizeof(int));
            for (i=0; i<src_hdr->n[dict]; i++)
            {
                if ( !src_hdr->id[dict][i].key ) // gap left after removed BCF header lines
                {
                    src_hdr->transl[dict][i] = -1;
                    continue;
                }
                src_hdr->transl[dict][i] = bcf_hdr_id2int(dst_hdr,dict,src_hdr->id[dict][i].key);
                if ( src_hdr->transl[dict][i]!=-1 && i!=src_hdr->transl[dict][i] ) src_hdr->ntransl++;
            }
        }
        if ( !src_hdr->ntransl )
        {
            free(src_hdr->transl[0]); src_hdr->transl[0] = NULL;
            free(src_hdr->transl[1]); src_hdr->transl[1] = NULL;
            src_hdr->ntransl = -1;
        }
        if ( src_hdr->ntransl==-1 ) return 0;
    }
    bcf_unpack(line,BCF_UN_ALL);

    // CHROM
    if ( src_hdr->transl[BCF_DT_CTG][line->rid] >=0 ) line->rid = src_hdr->transl[BCF_DT_CTG][line->rid];

    // FILTER
    for (i=0; i<line->d.n_flt; i++)
    {
        int src_id = line->d.flt[i];
        if ( src_hdr->transl[BCF_DT_ID][src_id] >=0 )
            line->d.flt[i] = src_hdr->transl[BCF_DT_ID][src_id];
        line->d.shared_dirty |= BCF1_DIRTY_FLT;
    }

    // INFO
    for (i=0; i<line->n_info; i++)
    {
        int src_id = line->d.info[i].key;
        int dst_id = src_hdr->transl[BCF_DT_ID][src_id];
        if ( dst_id<0 ) continue;
        line->d.info[i].key = dst_id;
        if ( !line->d.info[i].vptr ) continue;  // skip deleted
        int src_size = src_id>>7 ? ( src_id>>15 ? BCF_BT_INT32 : BCF_BT_INT16) : BCF_BT_INT8;
        int dst_size = dst_id>>7 ? ( dst_id>>15 ? BCF_BT_INT32 : BCF_BT_INT16) : BCF_BT_INT8;
        if ( src_size==dst_size )   // can overwrite
        {
            uint8_t *vptr = line->d.info[i].vptr - line->d.info[i].vptr_off;
            if ( dst_size==BCF_BT_INT8 ) { vptr[1] = (uint8_t)dst_id; }
            else if ( dst_size==BCF_BT_INT16 ) { *(uint16_t*)vptr = (uint16_t)dst_id; }
            else { *(uint32_t*)vptr = (uint32_t)dst_id; }
        }
        else    // must realloc
        {
            bcf_info_t *info = &line->d.info[i];
            kstring_t str = {0,0,0};
            bcf_enc_int1(&str, dst_id);
            bcf_enc_size(&str, info->len,info->type);
            uint32_t vptr_off = str.l;
            kputsn((char*)info->vptr, info->vptr_len, &str);
            if( info->vptr_free ) free(info->vptr - info->vptr_off);
            info->vptr_off = vptr_off;
            info->vptr = (uint8_t*)str.s + info->vptr_off;
            info->vptr_free = 1;
            line->d.shared_dirty |= BCF1_DIRTY_INF;
        }
    }

    // FORMAT
    for (i=0; i<line->n_fmt; i++)
    {
        int src_id = line->d.fmt[i].id;
        int dst_id = src_hdr->transl[BCF_DT_ID][src_id];
        if ( dst_id<0 ) continue;
        line->d.fmt[i].id = dst_id;
        if( !line->d.fmt[i].p ) continue;  // skip deleted
        int src_size = src_id>>7 ? ( src_id>>15 ? BCF_BT_INT32 : BCF_BT_INT16) : BCF_BT_INT8;
        int dst_size = dst_id>>7 ? ( dst_id>>15 ? BCF_BT_INT32 : BCF_BT_INT16) : BCF_BT_INT8;
        if ( src_size==dst_size )   // can overwrite
        {
            uint8_t *p = line->d.fmt[i].p - line->d.fmt[i].p_off;    // pointer to the vector size (4bits) and BT type (4bits)
            if ( dst_size==BCF_BT_INT8 ) { p[1] = dst_id; }
            else if ( dst_size==BCF_BT_INT16 ) { i16_to_le(dst_id, p + 1); }
            else { i32_to_le(dst_id, p + 1); }
        }
        else    // must realloc
        {
            bcf_fmt_t *fmt = &line->d.fmt[i];
            kstring_t str = {0,0,0};
            bcf_enc_int1(&str, dst_id);
            bcf_enc_size(&str, fmt->n, fmt->type);
            uint32_t p_off = str.l;
            kputsn((char*)fmt->p, fmt->p_len, &str);
            if( fmt->p_free ) free(fmt->p - fmt->p_off);
            fmt->p_off = p_off;
            fmt->p = (uint8_t*)str.s + fmt->p_off;
            fmt->p_free = 1;
            line->d.indiv_dirty = 1;
        }
    }
    return 0;
}

bcf_hdr_t *bcf_hdr_dup(const bcf_hdr_t *hdr)
{
    bcf_hdr_t *hout = bcf_hdr_init("r");
    if (!hout) {
        hts_log_error("Failed to allocate bcf header");
        return NULL;
    }
    kstring_t htxt = {0,0,0};
    if (bcf_hdr_format(hdr, 1, &htxt) < 0) {
        free(htxt.s);
        return NULL;
    }
    if ( bcf_hdr_parse(hout, htxt.s) < 0 ) {
        bcf_hdr_destroy(hout);
        hout = NULL;
    }
    free(htxt.s);
    return hout;
}

bcf_hdr_t *bcf_hdr_subset(const bcf_hdr_t *h0, int n, char *const* samples, int *imap)
{
    void *names_hash = khash_str2int_init();
    kstring_t htxt = {0,0,0};
    kstring_t str = {0,0,0};
    bcf_hdr_t *h = bcf_hdr_init("w");
    int r = 0;
    if (!h || !names_hash) {
        hts_log_error("Failed to allocate bcf header");
        goto err;
    }
    if (bcf_hdr_format(h0, 1, &htxt) < 0) {
        hts_log_error("Failed to get header text");
        goto err;
    }
    bcf_hdr_set_version(h,bcf_hdr_get_version(h0));
    int j;
    for (j=0; j<n; j++) imap[j] = -1;
    if ( bcf_hdr_nsamples(h0) > 0) {
        char *p = find_chrom_header_line(htxt.s);
        int i = 0, end = n? 8 : 7;
        while ((p = strchr(p, '\t')) != 0 && i < end) ++i, ++p;
        if (i != end) {
            hts_log_error("Wrong number of columns in header #CHROM line");
            goto err;
        }
        r |= kputsn(htxt.s, p - htxt.s, &str) < 0;
        for (i = 0; i < n; ++i) {
            if ( khash_str2int_has_key(names_hash,samples[i]) )
            {
                hts_log_error("Duplicate sample name \"%s\"", samples[i]);
                goto err;
            }
            imap[i] = bcf_hdr_id2int(h0, BCF_DT_SAMPLE, samples[i]);
            if (imap[i] < 0) continue;
            r |= kputc('\t', &str) < 0;
            r |= kputs(samples[i], &str) < 0;
            r |= khash_str2int_inc(names_hash,samples[i]) < 0;
        }
    } else r |= kputsn(htxt.s, htxt.l, &str) < 0;
    while (str.l && (!str.s[str.l-1] || str.s[str.l-1]=='\n') ) str.l--; // kill trailing zeros and newlines
    r |= kputc('\n',&str) < 0;
    if (r) {
        hts_log_error("%s", strerror(errno));
        goto err;
    }
    if ( bcf_hdr_parse(h, str.s) < 0 ) {
        bcf_hdr_destroy(h);
        h = NULL;
    }
    free(str.s);
    free(htxt.s);
    khash_str2int_destroy(names_hash);
    return h;

 err:
    ks_free(&str);
    ks_free(&htxt);
    khash_str2int_destroy(names_hash);
    bcf_hdr_destroy(h);
    return NULL;
}

int bcf_hdr_set_samples(bcf_hdr_t *hdr, const char *samples, int is_file)
{
    if ( samples && !strcmp("-",samples) ) return 0;            // keep all samples

    int i, narr = bit_array_size(bcf_hdr_nsamples(hdr));
    hdr->keep_samples = (uint8_t*) calloc(narr,1);
    if (!hdr->keep_samples) return -1;

    hdr->nsamples_ori = bcf_hdr_nsamples(hdr);
    if ( !samples )
    {
        // exclude all samples
        khint_t k;
        vdict_t *d = (vdict_t*)hdr->dict[BCF_DT_SAMPLE], *new_dict;
        new_dict = kh_init(vdict);
        if (!new_dict) return -1;

        bcf_hdr_nsamples(hdr) = 0;

        for (k = kh_begin(d); k != kh_end(d); ++k)
            if (kh_exist(d, k)) free((char*)kh_key(d, k));
        kh_destroy(vdict, d);
        hdr->dict[BCF_DT_SAMPLE] = new_dict;
        if (bcf_hdr_sync(hdr) < 0) return -1;

        return 0;
    }

    if ( samples[0]=='^' )
        for (i=0; i<bcf_hdr_nsamples(hdr); i++) bit_array_set(hdr->keep_samples,i);

    int idx, n, ret = 0;
    char **smpls = hts_readlist(samples[0]=='^'?samples+1:samples, is_file, &n);
    if ( !smpls ) return -1;
    for (i=0; i<n; i++)
    {
        idx = bcf_hdr_id2int(hdr,BCF_DT_SAMPLE,smpls[i]);
        if ( idx<0 )
        {
            if ( !ret ) ret = i+1;
            continue;
        }
        assert( idx<bcf_hdr_nsamples(hdr) );
        if (  samples[0]=='^' )
            bit_array_clear(hdr->keep_samples, idx);
        else
            bit_array_set(hdr->keep_samples, idx);
    }
    for (i=0; i<n; i++) free(smpls[i]);
    free(smpls);

    bcf_hdr_nsamples(hdr) = 0;
    for (i=0; i<hdr->nsamples_ori; i++)
        if ( bit_array_test(hdr->keep_samples,i) ) bcf_hdr_nsamples(hdr)++;

    if ( !bcf_hdr_nsamples(hdr) ) { free(hdr->keep_samples); hdr->keep_samples=NULL; }
    else
    {
        // Make new list and dictionary with desired samples
        char **samples = (char**) malloc(sizeof(char*)*bcf_hdr_nsamples(hdr));
        vdict_t *new_dict, *d;
        int k, res;
        if (!samples) return -1;

        new_dict = kh_init(vdict);
        if (!new_dict) {
            free(samples);
            return -1;
        }
        idx = 0;
        for (i=0; i<hdr->nsamples_ori; i++) {
            if ( bit_array_test(hdr->keep_samples,i) ) {
                samples[idx] = hdr->samples[i];
                k = kh_put(vdict, new_dict, hdr->samples[i], &res);
                if (res < 0) {
                    free(samples);
                    kh_destroy(vdict, new_dict);
                    return -1;
                }
                kh_val(new_dict, k) = bcf_idinfo_def;
                kh_val(new_dict, k).id = idx;
                idx++;
            }
        }

        // Delete desired samples from old dictionary, so we don't free them
        d = (vdict_t*)hdr->dict[BCF_DT_SAMPLE];
        for (i=0; i < idx; i++) {
            int k = kh_get(vdict, d, samples[i]);
            if (k < kh_end(d)) kh_del(vdict, d, k);
        }

        // Free everything else
        for (k = kh_begin(d); k != kh_end(d); ++k)
            if (kh_exist(d, k)) free((char*)kh_key(d, k));
        kh_destroy(vdict, d);
        hdr->dict[BCF_DT_SAMPLE] = new_dict;

        free(hdr->samples);
        hdr->samples = samples;

        if (bcf_hdr_sync(hdr) < 0)
            return -1;
    }

    return ret;
}

int bcf_subset(const bcf_hdr_t *h, bcf1_t *v, int n, int *imap)
{
    kstring_t ind;
    ind.s = 0; ind.l = ind.m = 0;
    if (n) {
        bcf_fmt_t fmt[MAX_N_FMT];
        int i, j;
        uint8_t *ptr = (uint8_t*)v->indiv.s;
        for (i = 0; i < v->n_fmt; ++i)
            ptr = bcf_unpack_fmt_core1(ptr, v->n_sample, &fmt[i]);
        for (i = 0; i < (int)v->n_fmt; ++i) {
            bcf_fmt_t *f = &fmt[i];
            bcf_enc_int1(&ind, f->id);
            bcf_enc_size(&ind, f->n, f->type);
            for (j = 0; j < n; ++j)
                if (imap[j] >= 0) kputsn((char*)(f->p + imap[j] * f->size), f->size, &ind);
        }
        for (i = j = 0; j < n; ++j) if (imap[j] >= 0) ++i;
        v->n_sample = i;
    } else v->n_sample = 0;
    if ( !v->n_sample ) v->n_fmt = 0;
    free(v->indiv.s);
    v->indiv = ind;
    v->unpacked &= ~BCF_UN_FMT;    // only BCF is ready for output, VCF will need to unpack again
    return 0;
}

int bcf_is_snp(bcf1_t *v)
{
    int i;
    bcf_unpack(v, BCF_UN_STR);
    for (i = 0; i < v->n_allele; ++i)
    {
        if ( v->d.allele[i][1]==0 && v->d.allele[i][0]!='*' ) continue;

        // mpileup's <X> allele, see also below. This is not completely satisfactory,
        // a general library is here narrowly tailored to fit samtools.
        if ( v->d.allele[i][0]=='<' && v->d.allele[i][1]=='X' && v->d.allele[i][2]=='>' ) continue;
        if ( v->d.allele[i][0]=='<' && v->d.allele[i][1]=='*' && v->d.allele[i][2]=='>' ) continue;

        break;
    }
    return i == v->n_allele;
}

static void bcf_set_variant_type(const char *ref, const char *alt, bcf_variant_t *var)
{
    if ( *alt == '*' && !alt[1] ) { var->n = 0; var->type = VCF_OVERLAP; return; }  // overlapping variant

    // The most frequent case
    if ( !ref[1] && !alt[1] )
    {
        if ( *alt == '.' || *ref==*alt ) { var->n = 0; var->type = VCF_REF; return; }
        if ( *alt == 'X' ) { var->n = 0; var->type = VCF_REF; return; }  // mpileup's X allele shouldn't be treated as variant
        var->n = 1; var->type = VCF_SNP; return;
    }
    if ( alt[0]=='<' )
    {
        if ( alt[1]=='X' && alt[2]=='>' ) { var->n = 0; var->type = VCF_REF; return; }  // mpileup's X allele shouldn't be treated as variant
        if ( alt[1]=='*' && alt[2]=='>' ) { var->n = 0; var->type = VCF_REF; return; }
        if ( !strcmp("NON_REF>",alt+1) ) { var->n = 0; var->type = VCF_REF; return; }
        var->type = VCF_OTHER;
        return;
    }

    // Catch "joined before" breakend case
    if ( alt[0]==']' || alt[0] == '[' )
    {
        var->type = VCF_BND; return;
    }

    // Iterate through alt characters that match the reference
    const char *r = ref, *a = alt;
    while (*r && *a && toupper_c(*r)==toupper_c(*a) ) { r++; a++; }     // unfortunately, matching REF,ALT case is not guaranteed

    if ( *a && !*r )
    {
        while ( *a ) a++;
        if ( *(a-1)==']' || *(a-1)=='[' ) { var->type = VCF_BND; return; } // "joined after" breakend
        var->n = (a-alt)-(r-ref); var->type = VCF_INDEL | VCF_INS; return;
    }
    else if ( *r && !*a )
    {
        while ( *r ) r++;
        var->n = (a-alt)-(r-ref); var->type = VCF_INDEL | VCF_DEL; return;
    }
    else if ( !*r && !*a )
    {
        var->n = 0; var->type = VCF_REF; return;
    }

    const char *re = r, *ae = a;
    while ( re[1] ) re++;
    while ( ae[1] ) ae++;
    while ( re>r && ae>a && toupper_c(*re)==toupper_c(*ae) ) { re--; ae--; }
    if ( ae==a )
    {
        if ( re==r ) { var->n = 1; var->type = VCF_SNP; return; }
        var->n = -(re-r);
        if ( toupper_c(*re)==toupper_c(*ae) ) { var->type = VCF_INDEL | VCF_DEL; return; }
        var->type = VCF_OTHER; return;
    }
    else if ( re==r )
    {
        var->n = ae-a;
        if ( toupper_c(*re)==toupper_c(*ae) ) { var->type = VCF_INDEL | VCF_INS; return; }
        var->type = VCF_OTHER; return;
    }

    var->type = ( re-r == ae-a ) ? VCF_MNP : VCF_OTHER;
    var->n = ( re-r > ae-a ) ? -(re-r+1) : ae-a+1;

    // should do also complex events, SVs, etc...
}

static int bcf_set_variant_types(bcf1_t *b)
{
    if ( !(b->unpacked & BCF_UN_STR) ) bcf_unpack(b, BCF_UN_STR);
    bcf_dec_t *d = &b->d;
    if ( d->n_var < b->n_allele )
    {
        bcf_variant_t *new_var = realloc(d->var, sizeof(bcf_variant_t)*b->n_allele);
        if (!new_var)
            return -1;
        d->var = new_var;
        d->n_var = b->n_allele;
    }
    int i;
    b->d.var_type = 0;
    d->var[0].type = VCF_REF;
    d->var[0].n    = 0;
    for (i=1; i<b->n_allele; i++)
    {
        bcf_set_variant_type(d->allele[0],d->allele[i], &d->var[i]);
        b->d.var_type |= d->var[i].type;
        //fprintf(stderr,"[set_variant_type] %d   %s %s -> %d %d .. %d\n", b->pos+1,d->allele[0],d->allele[i],d->var[i].type,d->var[i].n, b->d.var_type);
    }
    return 0;
}

// bcf_get_variant_type/bcf_get_variant_types should only return the following,
// to be compatible with callers that are not expecting newer values
// like VCF_INS, VCF_DEL.  The full set is available from the newer
// vcf_has_variant_type* interfaces.
#define ORIG_VAR_TYPES (VCF_SNP|VCF_MNP|VCF_INDEL|VCF_OTHER|VCF_BND|VCF_OVERLAP)
int bcf_get_variant_types(bcf1_t *rec)
{
    if ( rec->d.var_type==-1 ) {
        if (bcf_set_variant_types(rec) != 0) {
            hts_log_error("Couldn't get variant types: %s", strerror(errno));
            exit(1); // Due to legacy API having no way to report failures
        }
    }
    return rec->d.var_type & ORIG_VAR_TYPES;
}

int bcf_get_variant_type(bcf1_t *rec, int ith_allele)
{
    if ( rec->d.var_type==-1 ) {
        if (bcf_set_variant_types(rec) != 0) {
            hts_log_error("Couldn't get variant types: %s", strerror(errno));
            exit(1); // Due to legacy API having no way to report failures
        }
    }
    if (ith_allele < 0 || ith_allele >= rec->n_allele) {
        hts_log_error("Requested allele outside valid range");
        exit(1);
    }
    return rec->d.var[ith_allele].type & ORIG_VAR_TYPES;
}
#undef ORIG_VAR_TYPES

int bcf_has_variant_type(bcf1_t *rec, int ith_allele, uint32_t bitmask)
{
    if ( rec->d.var_type==-1 ) {
        if (bcf_set_variant_types(rec) != 0) return -1;
    }
    if (ith_allele < 0 || ith_allele >= rec->n_allele) return -1;
    if (bitmask == VCF_REF) {  // VCF_REF is 0, so handled as a special case
        return rec->d.var[ith_allele].type == VCF_REF;
    }
    return bitmask & rec->d.var[ith_allele].type;
}

int bcf_variant_length(bcf1_t *rec, int ith_allele)
{
    if ( rec->d.var_type==-1 ) {
        if (bcf_set_variant_types(rec) != 0) return bcf_int32_missing;
    }
    if (ith_allele < 0 || ith_allele >= rec->n_allele) return bcf_int32_missing;
    return rec->d.var[ith_allele].n;
}

int bcf_has_variant_types(bcf1_t *rec, uint32_t bitmask,
                          enum bcf_variant_match mode)
{
    if ( rec->d.var_type==-1 ) {
        if (bcf_set_variant_types(rec) != 0) return -1;
    }
    uint32_t type = rec->d.var_type;
    if ( mode==bcf_match_overlap ) return bitmask & type;

    // VCF_INDEL is always set with VCF_INS and VCF_DEL by bcf_set_variant_type[s], but the bitmask may
    // ask for say `VCF_INS` or `VCF_INDEL` only
    if ( bitmask&(VCF_INS|VCF_DEL) && !(bitmask&VCF_INDEL) ) type &= ~VCF_INDEL;
    else if ( bitmask&VCF_INDEL && !(bitmask&(VCF_INS|VCF_DEL)) ) type &= ~(VCF_INS|VCF_DEL);

    if ( mode==bcf_match_subset )
    {
        if ( ~bitmask & type ) return 0;
        else return bitmask & type;
    }
    // mode == bcf_match_exact
    if ( bitmask==VCF_REF ) return type==bitmask ? 1 : 0;
    return type==bitmask ? type : 0;
}

int bcf_update_info(const bcf_hdr_t *hdr, bcf1_t *line, const char *key, const void *values, int n, int type)
{
    static int negative_rlen_warned = 0;
    int is_end_tag;

    // Is the field already present?
    int i, inf_id = bcf_hdr_id2int(hdr,BCF_DT_ID,key);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_INFO,inf_id) ) return -1;    // No such INFO field in the header
    if ( !(line->unpacked & BCF_UN_INFO) ) bcf_unpack(line, BCF_UN_INFO);

    is_end_tag = strcmp(key, "END") == 0;

    for (i=0; i<line->n_info; i++)
        if ( inf_id==line->d.info[i].key ) break;
    bcf_info_t *inf = i==line->n_info ? NULL : &line->d.info[i];

    if ( !n || (type==BCF_HT_STR && !values) )
    {
        if ( n==0 && is_end_tag )
            line->rlen = line->n_allele ? strlen(line->d.allele[0]) : 0;
        if ( inf )
        {
            // Mark the tag for removal, free existing memory if necessary
            if ( inf->vptr_free )
            {
                free(inf->vptr - inf->vptr_off);
                inf->vptr_free = 0;
            }
            line->d.shared_dirty |= BCF1_DIRTY_INF;
            inf->vptr = NULL;
            inf->vptr_off = inf->vptr_len = 0;
        }
        return 0;
    }

    if (is_end_tag)
    {
        if (n != 1)
        {
            hts_log_error("END info tag should only have one value at %s:%"PRIhts_pos, bcf_seqname_safe(hdr,line), line->pos+1);
            line->errcode |= BCF_ERR_TAG_INVALID;
            return -1;
        }
        if (type != BCF_HT_INT && type != BCF_HT_LONG)
        {
            hts_log_error("Wrong type (%d) for END info tag at %s:%"PRIhts_pos, type, bcf_seqname_safe(hdr,line), line->pos+1);
            line->errcode |= BCF_ERR_TAG_INVALID;
            return -1;
        }
    }

    // Encode the values and determine the size required to accommodate the values
    kstring_t str = {0,0,0};
    bcf_enc_int1(&str, inf_id);
    if ( type==BCF_HT_INT )
        bcf_enc_vint(&str, n, (int32_t*)values, -1);
    else if ( type==BCF_HT_REAL )
        bcf_enc_vfloat(&str, n, (float*)values);
    else if ( type==BCF_HT_FLAG || type==BCF_HT_STR )
    {
        if ( values==NULL )
            bcf_enc_size(&str, 0, BCF_BT_NULL);
        else
            bcf_enc_vchar(&str, strlen((char*)values), (char*)values);
    }
#ifdef VCF_ALLOW_INT64
    else if ( type==BCF_HT_LONG )
    {
        if (n != 1) {
            hts_log_error("Only storing a single BCF_HT_LONG value is supported at %s:%"PRIhts_pos, bcf_seqname_safe(hdr,line), line->pos+1);
            abort();
        }
        bcf_enc_long1(&str, *(int64_t *) values);
    }
#endif
    else
    {
        hts_log_error("The type %d not implemented yet at %s:%"PRIhts_pos, type, bcf_seqname_safe(hdr,line), line->pos+1);
        abort();
    }

    // Is the INFO tag already present
    if ( inf )
    {
        // Is it big enough to accommodate new block?
        if ( inf->vptr && str.l <= inf->vptr_len + inf->vptr_off )
        {
            if ( str.l != inf->vptr_len + inf->vptr_off ) line->d.shared_dirty |= BCF1_DIRTY_INF;
            uint8_t *ptr = inf->vptr - inf->vptr_off;
            memcpy(ptr, str.s, str.l);
            free(str.s);
            int vptr_free = inf->vptr_free;
            bcf_unpack_info_core1(ptr, inf);
            inf->vptr_free = vptr_free;
        }
        else
        {
            if ( inf->vptr_free )
                free(inf->vptr - inf->vptr_off);
            bcf_unpack_info_core1((uint8_t*)str.s, inf);
            inf->vptr_free = 1;
            line->d.shared_dirty |= BCF1_DIRTY_INF;
        }
    }
    else
    {
        // The tag is not present, create new one
        line->n_info++;
        hts_expand0(bcf_info_t, line->n_info, line->d.m_info , line->d.info);
        inf = &line->d.info[line->n_info-1];
        bcf_unpack_info_core1((uint8_t*)str.s, inf);
        inf->vptr_free = 1;
        line->d.shared_dirty |= BCF1_DIRTY_INF;
    }
    line->unpacked |= BCF_UN_INFO;

   if ( n==1 && is_end_tag) {
        hts_pos_t end = type == BCF_HT_INT ? *(int32_t *) values : *(int64_t *) values;
        if ( (type == BCF_HT_INT && end!=bcf_int32_missing) || (type == BCF_HT_LONG && end!=bcf_int64_missing) )
        {
            if ( end <= line->pos )
            {
                if ( !negative_rlen_warned )
                {
                    hts_log_warning("INFO/END=%"PRIhts_pos" is smaller than POS at %s:%"PRIhts_pos,end,bcf_seqname_safe(hdr,line),line->pos+1);
                    negative_rlen_warned = 1;
                }
                line->rlen = line->n_allele ? strlen(line->d.allele[0]) : 0;
            }
            else
                line->rlen = end - line->pos;
        }
    }
    return 0;
}

int bcf_update_format_string(const bcf_hdr_t *hdr, bcf1_t *line, const char *key, const char **values, int n)
{
    if ( !n )
        return bcf_update_format(hdr,line,key,NULL,0,BCF_HT_STR);

    int i, max_len = 0;
    for (i=0; i<n; i++)
    {
        int len = strlen(values[i]);
        if ( len > max_len ) max_len = len;
    }
    char *out = (char*) malloc(max_len*n);
    if ( !out ) return -2;
    for (i=0; i<n; i++)
    {
        char *dst = out+i*max_len;
        const char *src = values[i];
        int j = 0;
        while ( src[j] ) { dst[j] = src[j]; j++; }
        for (; j<max_len; j++) dst[j] = 0;
    }
    int ret = bcf_update_format(hdr,line,key,out,max_len*n,BCF_HT_STR);
    free(out);
    return ret;
}

int bcf_update_format(const bcf_hdr_t *hdr, bcf1_t *line, const char *key, const void *values, int n, int type)
{
    // Is the field already present?
    int i, fmt_id = bcf_hdr_id2int(hdr,BCF_DT_ID,key);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,fmt_id) )
    {
        if ( !n ) return 0;
        return -1;  // the key not present in the header
    }

    if ( !(line->unpacked & BCF_UN_FMT) ) bcf_unpack(line, BCF_UN_FMT);

    for (i=0; i<line->n_fmt; i++)
        if ( line->d.fmt[i].id==fmt_id ) break;
    bcf_fmt_t *fmt = i==line->n_fmt ? NULL : &line->d.fmt[i];

    if ( !n )
    {
        if ( fmt )
        {
            // Mark the tag for removal, free existing memory if necessary
            if ( fmt->p_free )
            {
                free(fmt->p - fmt->p_off);
                fmt->p_free = 0;
            }
            line->d.indiv_dirty = 1;
            fmt->p = NULL;
        }
        return 0;
    }

    line->n_sample = bcf_hdr_nsamples(hdr);
    int nps = n / line->n_sample;  // number of values per sample
    assert( nps && nps*line->n_sample==n );     // must be divisible by n_sample

    // Encode the values and determine the size required to accommodate the values
    kstring_t str = {0,0,0};
    bcf_enc_int1(&str, fmt_id);
    assert(values != NULL);
    if ( type==BCF_HT_INT )
        bcf_enc_vint(&str, n, (int32_t*)values, nps);
    else if ( type==BCF_HT_REAL )
    {
        bcf_enc_size(&str, nps, BCF_BT_FLOAT);
        serialize_float_array(&str, nps*line->n_sample, (float *) values);
    }
    else if ( type==BCF_HT_STR )
    {
        bcf_enc_size(&str, nps, BCF_BT_CHAR);
        kputsn((char*)values, nps*line->n_sample, &str);
    }
    else
    {
        hts_log_error("The type %d not implemented yet at %s:%"PRIhts_pos, type, bcf_seqname_safe(hdr,line), line->pos+1);
        abort();
    }

    if ( !fmt )
    {
        // Not present, new format field
        line->n_fmt++;
        hts_expand0(bcf_fmt_t, line->n_fmt, line->d.m_fmt, line->d.fmt);

        // Special case: VCF specification requires that GT is always first
        if ( line->n_fmt > 1 && key[0]=='G' && key[1]=='T' && !key[2] )
        {
            for (i=line->n_fmt-1; i>0; i--)
                line->d.fmt[i] = line->d.fmt[i-1];
            fmt = &line->d.fmt[0];
        }
        else
            fmt = &line->d.fmt[line->n_fmt-1];
        bcf_unpack_fmt_core1((uint8_t*)str.s, line->n_sample, fmt);
        line->d.indiv_dirty = 1;
        fmt->p_free = 1;
    }
    else
    {
        // The tag is already present, check if it is big enough to accommodate the new block
        if ( fmt->p && str.l <= fmt->p_len + fmt->p_off )
        {
            // good, the block is big enough
            if ( str.l != fmt->p_len + fmt->p_off ) line->d.indiv_dirty = 1;
            uint8_t *ptr = fmt->p - fmt->p_off;
            memcpy(ptr, str.s, str.l);
            free(str.s);
            int p_free = fmt->p_free;
            bcf_unpack_fmt_core1(ptr, line->n_sample, fmt);
            fmt->p_free = p_free;
        }
        else
        {
            if ( fmt->p_free )
                free(fmt->p - fmt->p_off);
            bcf_unpack_fmt_core1((uint8_t*)str.s, line->n_sample, fmt);
            fmt->p_free = 1;
            line->d.indiv_dirty = 1;
        }
    }
    line->unpacked |= BCF_UN_FMT;
    return 0;
}


int bcf_update_filter(const bcf_hdr_t *hdr, bcf1_t *line, int *flt_ids, int n)
{
    if ( !(line->unpacked & BCF_UN_FLT) ) bcf_unpack(line, BCF_UN_FLT);
    line->d.shared_dirty |= BCF1_DIRTY_FLT;
    line->d.n_flt = n;
    if ( !n ) return 0;
    hts_expand(int, line->d.n_flt, line->d.m_flt, line->d.flt);
    int i;
    for (i=0; i<n; i++)
        line->d.flt[i] = flt_ids[i];
    return 0;
}

int bcf_add_filter(const bcf_hdr_t *hdr, bcf1_t *line, int flt_id)
{
    if ( !(line->unpacked & BCF_UN_FLT) ) bcf_unpack(line, BCF_UN_FLT);
    int i;
    for (i=0; i<line->d.n_flt; i++)
        if ( flt_id==line->d.flt[i] ) break;
    if ( i<line->d.n_flt ) return 0;    // this filter is already set
    line->d.shared_dirty |= BCF1_DIRTY_FLT;
    if ( flt_id==0 )    // set to PASS
        line->d.n_flt = 1;
    else if ( line->d.n_flt==1 && line->d.flt[0]==0 )
        line->d.n_flt = 1;
    else
        line->d.n_flt++;
    hts_expand(int, line->d.n_flt, line->d.m_flt, line->d.flt);
    line->d.flt[line->d.n_flt-1] = flt_id;
    return 1;
}
int bcf_remove_filter(const bcf_hdr_t *hdr, bcf1_t *line, int flt_id, int pass)
{
    if ( !(line->unpacked & BCF_UN_FLT) ) bcf_unpack(line, BCF_UN_FLT);
    int i;
    for (i=0; i<line->d.n_flt; i++)
        if ( flt_id==line->d.flt[i] ) break;
    if ( i==line->d.n_flt ) return 0;   // the filter is not present
    line->d.shared_dirty |= BCF1_DIRTY_FLT;
    if ( i!=line->d.n_flt-1 ) memmove(line->d.flt+i,line->d.flt+i+1,(line->d.n_flt-i-1)*sizeof(*line->d.flt));
    line->d.n_flt--;
    if ( !line->d.n_flt && pass ) bcf_add_filter(hdr,line,0);
    return 0;
}

int bcf_has_filter(const bcf_hdr_t *hdr, bcf1_t *line, char *filter)
{
    if ( filter[0]=='.' && !filter[1] ) filter = "PASS";
    int id = bcf_hdr_id2int(hdr, BCF_DT_ID, filter);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_FLT,id) ) return -1;  // not defined in the header

    if ( !(line->unpacked & BCF_UN_FLT) ) bcf_unpack(line, BCF_UN_FLT);
    if ( id==0 && !line->d.n_flt) return 1; // PASS

    int i;
    for (i=0; i<line->d.n_flt; i++)
        if ( line->d.flt[i]==id ) return 1;
    return 0;
}

static inline int _bcf1_sync_alleles(const bcf_hdr_t *hdr, bcf1_t *line, int nals)
{
    line->d.shared_dirty |= BCF1_DIRTY_ALS;
    line->d.var_type = -1;

    line->n_allele = nals;
    hts_expand(char*, line->n_allele, line->d.m_allele, line->d.allele);

    char *als = line->d.als;
    int n = 0;
    while (n<nals)
    {
        line->d.allele[n] = als;
        while ( *als ) als++;
        als++;
        n++;
    }

    // Update REF length. Note that END is 1-based while line->pos 0-based
    bcf_info_t *end_info = bcf_get_info(hdr,line,"END");
    if ( end_info )
    {
        if ( end_info->type==BCF_HT_INT && end_info->v1.i==bcf_int32_missing ) end_info = NULL;
        else if ( end_info->type==BCF_HT_LONG && end_info->v1.i==bcf_int64_missing ) end_info = NULL;
    }
    if ( end_info && end_info->v1.i > line->pos )
        line->rlen = end_info->v1.i - line->pos;
    else if ( nals > 0 )
        line->rlen = strlen(line->d.allele[0]);
    else
        line->rlen = 0;

    return 0;
}
int bcf_update_alleles(const bcf_hdr_t *hdr, bcf1_t *line, const char **alleles, int nals)
{
    if ( !(line->unpacked & BCF_UN_STR) ) bcf_unpack(line, BCF_UN_STR);
    char *free_old = NULL;
    char buffer[256];
    size_t used = 0;

    // The pointers in alleles may point into the existing line->d.als memory,
    // so care needs to be taken not to clobber them while updating.  Usually
    // they will be short so we can copy through an intermediate buffer.
    // If they're longer, or won't fit in the existing allocation we
    // can allocate a new buffer to write into.  Note that in either case
    // pointers to line->d.als memory in alleles may not be valid when we've
    // finished.
    int i;
    size_t avail = line->d.m_als < sizeof(buffer) ? line->d.m_als : sizeof(buffer);
    for (i=0; i<nals; i++) {
        size_t sz = strlen(alleles[i]) + 1;
        if (avail - used < sz)
            break;
        memcpy(buffer + used, alleles[i], sz);
        used += sz;
    }

    // Did we miss anything?
    if (i < nals) {
        int j;
        size_t needed = used;
        char *new_als;
        for (j = i; j < nals; j++)
            needed += strlen(alleles[j]) + 1;
        if (needed < line->d.m_als) // Don't shrink the buffer
            needed = line->d.m_als;
        if (needed > INT_MAX) {
            hts_log_error("REF + alleles too long to fit in a BCF record");
            return -1;
        }
        new_als = malloc(needed);
        if (!new_als)
            return -1;
        free_old = line->d.als;
        line->d.als = new_als;
        line->d.m_als = needed;
    }

    // Copy from the temp buffer to the destination
    if (used) {
        assert(used <= line->d.m_als);
        memcpy(line->d.als, buffer, used);
    }

    // Add in any remaining entries - if this happens we will always be
    // writing to a newly-allocated buffer.
    for (; i < nals; i++) {
        size_t sz = strlen(alleles[i]) + 1;
        memcpy(line->d.als + used, alleles[i], sz);
        used += sz;
    }

    if (free_old)
        free(free_old);
    return _bcf1_sync_alleles(hdr,line,nals);
}

int bcf_update_alleles_str(const bcf_hdr_t *hdr, bcf1_t *line, const char *alleles_string)
{
    if ( !(line->unpacked & BCF_UN_STR) ) bcf_unpack(line, BCF_UN_STR);
    kstring_t tmp;
    tmp.l = 0; tmp.s = line->d.als; tmp.m = line->d.m_als;
    kputs(alleles_string, &tmp);
    line->d.als = tmp.s; line->d.m_als = tmp.m;

    int nals = 1;
    char *t = line->d.als;
    while (*t)
    {
        if ( *t==',' ) { *t = 0; nals++; }
        t++;
    }
    return _bcf1_sync_alleles(hdr, line, nals);
}

int bcf_update_id(const bcf_hdr_t *hdr, bcf1_t *line, const char *id)
{
    if ( !(line->unpacked & BCF_UN_STR) ) bcf_unpack(line, BCF_UN_STR);
    kstring_t tmp;
    tmp.l = 0; tmp.s = line->d.id; tmp.m = line->d.m_id;
    if ( id )
        kputs(id, &tmp);
    else
        kputs(".", &tmp);
    line->d.id = tmp.s; line->d.m_id = tmp.m;
    line->d.shared_dirty |= BCF1_DIRTY_ID;
    return 0;
}

int bcf_add_id(const bcf_hdr_t *hdr, bcf1_t *line, const char *id)
{
    if ( !id ) return 0;
    if ( !(line->unpacked & BCF_UN_STR) ) bcf_unpack(line, BCF_UN_STR);

    kstring_t tmp;
    tmp.l = 0; tmp.s = line->d.id; tmp.m = line->d.m_id;

    int len = strlen(id);
    char *dst = line->d.id;
    while ( *dst && (dst=strstr(dst,id)) )
    {
        if ( dst[len]!=0 && dst[len]!=';' ) dst++;              // a prefix, not a match
        else if ( dst==line->d.id || dst[-1]==';' ) return 0;   // already present
        dst++;  // a suffix, not a match
    }
    if ( line->d.id && (line->d.id[0]!='.' || line->d.id[1]) )
    {
        tmp.l = strlen(line->d.id);
        kputc(';',&tmp);
    }
    kputs(id,&tmp);

    line->d.id = tmp.s; line->d.m_id = tmp.m;
    line->d.shared_dirty |= BCF1_DIRTY_ID;
    return 0;

}

bcf_fmt_t *bcf_get_fmt(const bcf_hdr_t *hdr, bcf1_t *line, const char *key)
{
    int id = bcf_hdr_id2int(hdr, BCF_DT_ID, key);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,id) ) return NULL;   // no such FMT field in the header
    return bcf_get_fmt_id(line, id);
}

bcf_info_t *bcf_get_info(const bcf_hdr_t *hdr, bcf1_t *line, const char *key)
{
    int id = bcf_hdr_id2int(hdr, BCF_DT_ID, key);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_INFO,id) ) return NULL;   // no such INFO field in the header
    return bcf_get_info_id(line, id);
}

bcf_fmt_t *bcf_get_fmt_id(bcf1_t *line, const int id)
{
    int i;
    if ( !(line->unpacked & BCF_UN_FMT) ) bcf_unpack(line, BCF_UN_FMT);
    for (i=0; i<line->n_fmt; i++)
    {
        if ( line->d.fmt[i].id==id ) return &line->d.fmt[i];
    }
    return NULL;
}

bcf_info_t *bcf_get_info_id(bcf1_t *line, const int id)
{
    int i;
    if ( !(line->unpacked & BCF_UN_INFO) ) bcf_unpack(line, BCF_UN_INFO);
    for (i=0; i<line->n_info; i++)
    {
        if ( line->d.info[i].key==id ) return &line->d.info[i];
    }
    return NULL;
}


int bcf_get_info_values(const bcf_hdr_t *hdr, bcf1_t *line, const char *tag, void **dst, int *ndst, int type)
{
    int i, ret = -4, tag_id = bcf_hdr_id2int(hdr, BCF_DT_ID, tag);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_INFO,tag_id) ) return -1;    // no such INFO field in the header
    if ( bcf_hdr_id2type(hdr,BCF_HL_INFO,tag_id)!=(type & 0xff) ) return -2;     // expected different type

    if ( !(line->unpacked & BCF_UN_INFO) ) bcf_unpack(line, BCF_UN_INFO);

    for (i=0; i<line->n_info; i++)
        if ( line->d.info[i].key==tag_id ) break;
    if ( i==line->n_info ) return ( type==BCF_HT_FLAG ) ? 0 : -3;       // the tag is not present in this record
    if ( type==BCF_HT_FLAG ) return 1;

    bcf_info_t *info = &line->d.info[i];
    if ( !info->vptr ) return -3;           // the tag was marked for removal
    if ( type==BCF_HT_STR )
    {
        if ( *ndst < info->len+1 )
        {
            *ndst = info->len + 1;
            *dst  = realloc(*dst, *ndst);
        }
        memcpy(*dst,info->vptr,info->len);
        ((uint8_t*)*dst)[info->len] = 0;
        return info->len;
    }

    // Make sure the buffer is big enough
    int size1;
    switch (type) {
        case BCF_HT_INT:  size1 = sizeof(int32_t); break;
        case BCF_HT_LONG: size1 = sizeof(int64_t); break;
        case BCF_HT_REAL: size1 = sizeof(float); break;
        default:
            hts_log_error("Unexpected output type %d at %s:%"PRIhts_pos, type, bcf_seqname_safe(hdr,line), line->pos+1);
            return -2;
    }
    if ( *ndst < info->len )
    {
        *ndst = info->len;
        *dst  = realloc(*dst, *ndst * size1);
    }

    #define BRANCH(type_t, convert, is_missing, is_vector_end, set_missing, set_regular, out_type_t) do { \
        out_type_t *tmp = (out_type_t *) *dst; \
        int j; \
        for (j=0; j<info->len; j++) \
        { \
            type_t p = convert(info->vptr + j * sizeof(type_t)); \
            if ( is_vector_end ) break; \
            if ( is_missing ) set_missing; \
            else set_regular; \
            tmp++; \
        } \
        ret = j; \
    } while (0)
    switch (info->type) {
        case BCF_BT_INT8:
            if (type == BCF_HT_LONG) {
                BRANCH(int8_t,  le_to_i8,  p==bcf_int8_missing,  p==bcf_int8_vector_end,  *tmp=bcf_int64_missing, *tmp=p, int64_t);
            } else {
                BRANCH(int8_t,  le_to_i8,  p==bcf_int8_missing,  p==bcf_int8_vector_end,  *tmp=bcf_int32_missing, *tmp=p, int32_t);
            }
            break;
        case BCF_BT_INT16:
            if (type == BCF_HT_LONG) {
                BRANCH(int16_t, le_to_i16, p==bcf_int16_missing, p==bcf_int16_vector_end, *tmp=bcf_int64_missing, *tmp=p, int64_t);
            } else {
                BRANCH(int16_t, le_to_i16, p==bcf_int16_missing, p==bcf_int16_vector_end, *tmp=bcf_int32_missing, *tmp=p, int32_t);
            }
            break;
        case BCF_BT_INT32:
            if (type == BCF_HT_LONG) {
                BRANCH(int32_t, le_to_i32, p==bcf_int32_missing, p==bcf_int32_vector_end, *tmp=bcf_int64_missing, *tmp=p, int64_t); break;
            } else {
                BRANCH(int32_t, le_to_i32, p==bcf_int32_missing, p==bcf_int32_vector_end, *tmp=bcf_int32_missing, *tmp=p, int32_t); break;
            }
        case BCF_BT_FLOAT: BRANCH(uint32_t, le_to_u32, p==bcf_float_missing, p==bcf_float_vector_end, bcf_float_set_missing(*tmp), bcf_float_set(tmp, p), float); break;
        default: hts_log_error("Unexpected type %d at %s:%"PRIhts_pos, info->type, bcf_seqname_safe(hdr,line), line->pos+1); return -2;
    }
    #undef BRANCH
    return ret;  // set by BRANCH
}

int bcf_get_format_string(const bcf_hdr_t *hdr, bcf1_t *line, const char *tag, char ***dst, int *ndst)
{
    int i,tag_id = bcf_hdr_id2int(hdr, BCF_DT_ID, tag);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,tag_id) ) return -1;    // no such FORMAT field in the header
    if ( bcf_hdr_id2type(hdr,BCF_HL_FMT,tag_id)!=BCF_HT_STR ) return -2;     // expected different type

    if ( !(line->unpacked & BCF_UN_FMT) ) bcf_unpack(line, BCF_UN_FMT);

    for (i=0; i<line->n_fmt; i++)
        if ( line->d.fmt[i].id==tag_id ) break;
    if ( i==line->n_fmt ) return -3;                               // the tag is not present in this record
    bcf_fmt_t *fmt = &line->d.fmt[i];
    if ( !fmt->p ) return -3;                                      // the tag was marked for removal

    int nsmpl = bcf_hdr_nsamples(hdr);
    if ( !*dst )
    {
        *dst = (char**) malloc(sizeof(char*)*nsmpl);
        if ( !*dst ) return -4;     // could not alloc
        (*dst)[0] = NULL;
    }
    int n = (fmt->n+1)*nsmpl;
    if ( *ndst < n )
    {
        (*dst)[0] = realloc((*dst)[0], n);
        if ( !(*dst)[0] ) return -4;    // could not alloc
        *ndst = n;
    }
    for (i=0; i<nsmpl; i++)
    {
        uint8_t *src = fmt->p + i*fmt->n;
        uint8_t *tmp = (uint8_t*)(*dst)[0] + i*(fmt->n+1);
        memcpy(tmp,src,fmt->n);
        tmp[fmt->n] = 0;
        (*dst)[i] = (char*) tmp;
    }
    return n;
}

int bcf_get_format_values(const bcf_hdr_t *hdr, bcf1_t *line, const char *tag, void **dst, int *ndst, int type)
{
    int i,j, tag_id = bcf_hdr_id2int(hdr, BCF_DT_ID, tag);
    if ( !bcf_hdr_idinfo_exists(hdr,BCF_HL_FMT,tag_id) ) return -1;    // no such FORMAT field in the header
    if ( tag[0]=='G' && tag[1]=='T' && tag[2]==0 )
    {
        // Ugly: GT field is considered to be a string by the VCF header but BCF represents it as INT.
        if ( bcf_hdr_id2type(hdr,BCF_HL_FMT,tag_id)!=BCF_HT_STR ) return -2;
    }
    else if ( bcf_hdr_id2type(hdr,BCF_HL_FMT,tag_id)!=type ) return -2;     // expected different type

    if ( !(line->unpacked & BCF_UN_FMT) ) bcf_unpack(line, BCF_UN_FMT);

    for (i=0; i<line->n_fmt; i++)
        if ( line->d.fmt[i].id==tag_id ) break;
    if ( i==line->n_fmt ) return -3;                               // the tag is not present in this record
    bcf_fmt_t *fmt = &line->d.fmt[i];
    if ( !fmt->p ) return -3;                                      // the tag was marked for removal

    if ( type==BCF_HT_STR )
    {
        int n = fmt->n*bcf_hdr_nsamples(hdr);
        if ( *ndst < n )
        {
            *dst  = realloc(*dst, n);
            if ( !*dst ) return -4;     // could not alloc
            *ndst = n;
        }
        memcpy(*dst,fmt->p,n);
        return n;
    }

    // Make sure the buffer is big enough
    int nsmpl = bcf_hdr_nsamples(hdr);
    int size1 = type==BCF_HT_INT ? sizeof(int32_t) : sizeof(float);
    if ( *ndst < fmt->n*nsmpl )
    {
        *ndst = fmt->n*nsmpl;
        *dst  = realloc(*dst, *ndst*size1);
        if ( !*dst ) return -4;     // could not alloc
    }

    #define BRANCH(type_t, convert, is_missing, is_vector_end, set_missing, set_vector_end, set_regular, out_type_t) { \
        out_type_t *tmp = (out_type_t *) *dst; \
        uint8_t *fmt_p = fmt->p; \
        for (i=0; i<nsmpl; i++) \
        { \
            for (j=0; j<fmt->n; j++) \
            { \
                type_t p = convert(fmt_p + j * sizeof(type_t)); \
                if ( is_missing ) set_missing; \
                else if ( is_vector_end ) { set_vector_end; break; } \
                else set_regular; \
                tmp++; \
            } \
            for (; j<fmt->n; j++) { set_vector_end; tmp++; } \
            fmt_p += fmt->size; \
        } \
    }
    switch (fmt->type) {
        case BCF_BT_INT8:  BRANCH(int8_t,  le_to_i8, p==bcf_int8_missing,  p==bcf_int8_vector_end,  *tmp=bcf_int32_missing, *tmp=bcf_int32_vector_end, *tmp=p, int32_t); break;
        case BCF_BT_INT16: BRANCH(int16_t, le_to_i16, p==bcf_int16_missing, p==bcf_int16_vector_end, *tmp=bcf_int32_missing, *tmp=bcf_int32_vector_end, *tmp=p, int32_t); break;
        case BCF_BT_INT32: BRANCH(int32_t, le_to_i32, p==bcf_int32_missing, p==bcf_int32_vector_end, *tmp=bcf_int32_missing, *tmp=bcf_int32_vector_end, *tmp=p, int32_t); break;
        case BCF_BT_FLOAT: BRANCH(uint32_t, le_to_u32, p==bcf_float_missing, p==bcf_float_vector_end, bcf_float_set_missing(*tmp), bcf_float_set_vector_end(*tmp), bcf_float_set(tmp, p), float); break;
        default: hts_log_error("Unexpected type %d at %s:%"PRIhts_pos, fmt->type, bcf_seqname_safe(hdr,line), line->pos+1); exit(1);
    }
    #undef BRANCH
    return nsmpl*fmt->n;
}

//error description structure definition
typedef struct err_desc {
    int  errorcode;
    const char *description;
}err_desc;

// error descriptions
static const err_desc errdesc_bcf[] = {
    { BCF_ERR_CTG_UNDEF, "Contig not defined in header"},
    { BCF_ERR_TAG_UNDEF, "Tag not defined in header" },
    { BCF_ERR_NCOLS, "Incorrect number of columns" },
    { BCF_ERR_LIMITS, "Limits reached" },
    { BCF_ERR_CHAR, "Invalid character" },
    { BCF_ERR_CTG_INVALID, "Invalid contig" },
    { BCF_ERR_TAG_INVALID, "Invalid tag" },
};

/// append given description to buffer based on available size and add ... when not enough space
    /** @param buffer       buffer to which description to be appended
        @param offset       offset at which to be appended
        @param maxbuffer    maximum size of the buffer
        @param description  the description to be appended
on failure returns -1 - when buffer is not big enough; returns -1 on invalid params and on too small buffer which are improbable due to validation at caller site
on success returns 0
    */
static int add_desc_to_buffer(char *buffer, size_t *offset, size_t maxbuffer, const char *description) {

    if (!description || !buffer || !offset || (maxbuffer < 4))
        return -1;

    size_t rembuffer = maxbuffer - *offset;
    if (rembuffer > (strlen(description) + (rembuffer == maxbuffer ? 0 : 1))) {    //add description with optionally required ','
        *offset += snprintf(buffer + *offset, rembuffer, "%s%s", (rembuffer == maxbuffer)? "": ",", description);
    } else {    //not enough space for description, put ...
        size_t tmppos = (rembuffer <= 4) ? maxbuffer - 4 : *offset;
        snprintf(buffer + tmppos, 4, "...");    //ignore offset update
        return -1;
    }
    return 0;
}

//get description for given error code. return NULL on error
const char *bcf_strerror(int errorcode, char *buffer, size_t maxbuffer) {
    size_t usedup = 0;
    int ret = 0;
    int idx;

    if (!buffer || maxbuffer < 4)
        return NULL;           //invalid / insufficient buffer

    if (!errorcode) {
        buffer[0] = '\0';      //no error, set null
        return buffer;
    }

    for (idx = 0; idx < sizeof(errdesc_bcf) / sizeof(err_desc); ++idx) {
        if (errorcode & errdesc_bcf[idx].errorcode) {    //error is set, add description
            ret = add_desc_to_buffer(buffer, &usedup, maxbuffer, errdesc_bcf[idx].description);
            if (ret < 0)
                break;         //not enough space, ... added, no need to continue

            errorcode &= ~errdesc_bcf[idx].errorcode;    //reset the error
        }
    }

    if (errorcode && (ret >= 0))  {     //undescribed error is present in error code and had enough buffer, try to add unkonwn error as well§
        add_desc_to_buffer(buffer, &usedup, maxbuffer, "Unknown error");
    }
    return buffer;
}