Release1.1.0

.github/workflows/fuzz-long.yml

@@ -0,0 +1,28 @@
+name: Long Fuzz (reusable)
+
+on:
+  workflow_call: {}
+
+jobs:
+  fuzz:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download goldens artifact
+        uses: actions/download-artifact@v3
+        with:
+          name: goldens
+          path: artifact/golden
+
+      - name: Setup runtime
+        uses: ./.github/actions/ci-setup
+        with:
+          otp-version: '28.1'
+          rebar3-version: '3.25.1'
+          gleam-version: '1.13.0'
+
+      - name: Run long fuzz harness
+        run: |
+          set -eux
+          gleam run -m thrifty/fuzz_cli

.github/workflows/goldens.yml

@@ -0,0 +1,59 @@
+name: Generate and Validate Goldens
+
+on:
+  workflow_dispatch:
+    inputs:
+      regen:
+        description: 'Regenerate goldens'
+        required: false
+        type: boolean
+  push:
+    branches: [ main, master ]
+
+jobs:
+  goldens:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup runtime
+        uses: ./.github/actions/ci-setup
+        with:
+          otp-version: '28.1'
+          rebar3-version: '3.25.1'
+          gleam-version: '1.13.0'
+
+      - name: Cache pip
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cache/pip
+            .venv
+          key: ${{ runner.os }}-pip-${{ hashFiles('tools/golden/generate_compact_vectors.py') }}
+          restore-keys: |
+            ${{ runner.os }}-pip-
+
+      - name: Set up Python
+        run: |
+          set -eux
+          python3 -m venv .venv
+          . .venv/bin/activate
+          pip install --upgrade pip
+          pip install thriftpy2
+
+      - name: Generate goldens
+        run: |
+          set -eux
+          . .venv/bin/activate
+          python3 tools/golden/generate_compact_vectors.py
+
+      - name: Run golden validation (gleam tests)
+        run: |
+          set -eux
+          gleam test
+
+      - name: Upload goldens
+        uses: actions/upload-artifact@v4
+        with:
+          name: goldens
+          path: artifact/golden

CHANGELOG.md

@@ -0,0 +1,108 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
+This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+---
+
+## [1.1.0] — 2026-03-20
+
+### Fixed
+
+#### Wire-format correctness (breaking for interoperability)
+
+- **Field ID in long-form header not zigzag-encoded** (`field.gleam`)
+  The Thrift Compact Protocol specification requires that field IDs in
+  long-form headers (used when the delta from the previous field ID is
+  outside `[1, 15]`) are encoded as a zigzag i16 followed by a varint.
+  The encoder was writing a plain varint and the decoder was reading a
+  plain varint, making payloads with long-form field headers
+  incompatible with every other Thrift implementation.
+  Fix: `encode_field_header` now applies `zigzag.encode_i32` before
+  `varint.encode_varint`; `read_field_header` now applies
+  `zigzag.decode_i32` after reading the varint.
+  Affects any field whose delta from the previous field ID is `> 15`
+  or `<= 0` (e.g. fields with ID > 15, or out-of-order fields).
+
+- **`read_i8` decoded bytes as unsigned instead of signed** (`reader.gleam`)
+  The bit-pattern `<<value:int-size(8)>>` matches unsigned (0–255).
+  Negative i8 values such as `-1` (wire byte `0xFF`) were silently
+  decoded as `255` instead of `-1`.
+  Fix: changed to `<<value:int-signed-size(8)>>`.
+
+- **`message.gleam` sequence ID broken for negative values**
+  `encode_message_header` called `varint.encode_varint` directly on
+  `sequence_id`. Because `encode_varint` requires a non-negative
+  integer, negative sequence IDs produced corrupt output (or, after
+  the `encode_varint` guard was added, a panic).
+  The reference Java implementation encodes the sequence ID as its
+  unsigned 32-bit two's-complement representation.
+  Fix: encoder masks `sequence_id` to uint32 before encoding;
+  decoder sign-extends values `> INT32_MAX` back to negative i32.
+
+#### API correctness
+
+- **`encode_varint` accepted negative integers silently** (`varint.gleam`)
+  Passing a negative integer produced an invalid varint (the MSB
+  continuation bit would be set on the last byte). Now panics with an
+  explicit message including the offending value.
+
+- **`write_i8` silently truncated out-of-range values** (`writer.gleam`)
+  Values outside `[-128, 127]` were accepted and silently truncated by
+  the bit-syntax, e.g. `write_i8(256)` produced `<<0>>`. Now panics
+  with an explicit message, consistent with `write_i32` / `write_i64`.
+
+- **`write_i16` accepted the full i32 range** (`writer.gleam`)
+  `write_i16` delegated to `zigzag.encode_i32`, which allows values up
+  to ±2 147 483 647. Now panics on values outside `[-32 768, 32 767]`.
+
+- **`BoolElementPolicy::AcceptBoth` was identical to `AcceptCanonicalOnly`**
+  (`reader.gleam`, `types.gleam`)
+  Both policies executed the same code, making the distinction
+  meaningless. `AcceptBoth` now also accepts byte value `0` as `False`,
+  providing compatibility with older implementations that used the
+  BinaryProtocol `0 = False / 1 = True` encoding.
+  `AcceptCanonicalOnly` continues to accept only `1` (True) and `2`
+  (False).
+
+#### Documentation
+
+- **Doc comments placed after function bodies** (`varint.gleam`,
+  `zigzag.gleam`, `field.gleam`, `reader.gleam`, `writer.gleam`,
+  `message.gleam`, `writer_highlevel.gleam`)
+  Gleam attaches `///` doc comments to the next item in the file.
+  Comments that appeared after a function body were silently associated
+  with the wrong function in generated documentation. All misplaced
+  comments have been moved to immediately precede the function they
+  describe.
+
+- **Duplicate Apache-2.0 license header** (`varint.gleam`, `field.gleam`)
+  The license block appeared twice at the top of both files. The
+  duplicate has been removed.
+
+#### Dead code
+
+- **`types.Writer` type was never used** (`types.gleam`)
+  `pub type Writer { Writer(buffer: BitArray) }` was defined but never
+  instantiated or referenced anywhere in the library. The actual writer
+  uses `writer.Buffer`. The unused type has been removed.
+
+### Tests added
+
+- `read_i8_negative_roundtrip_test` — roundtrip for i8 values
+  `-128, -42, -1, 0, 1, 127` via `write_i8` / `read_i8`.
+- `i8_write_boundaries_test` — verifies the raw byte written for
+  boundary values `127`, `-128`, `-1` (expected `0xFF`).
+- `i16_write_boundaries_test` — roundtrip for `32 767`, `-32 768`, `-1`
+  via `write_i16` / `read_i16`.
+- `permissive_bool_accepts_zero_as_false_test` — verifies that
+  `AcceptBoth` maps byte `0` to `False`.
+- `message_negative_seqid_roundtrip_test` — roundtrip for sequence IDs
+  `-1`, `-100`, `-2 147 483 648`.
+- `long_form_negative_field_id_test` — verifies that a long-form field
+  header with `field_id = -1` encodes and decodes correctly.
+- `long_form_header_test` updated to use the correct wire bytes:
+  `zigzag(32) = 64`, varint `<<0x40>>` instead of the previous
+  (incorrect) plain varint `<<0x20>>`.