Add unary negation, logical/bitwise not, and bitwise/shift operators#188
Open
WalterGropius wants to merge 5 commits into
Open
Add unary negation, logical/bitwise not, and bitwise/shift operators#188WalterGropius wants to merge 5 commits into
WalterGropius wants to merge 5 commits into
Conversation
Adds lexer tokens for |, ^, ~, <<, >>; introduces EXPR_UNARY and parses unary prefix operators (-, !, ~) above multiplicative; adds binary precedence for bitwise &, |, ^ and shifts << >> between cmp and additive. Checker validates integer-only bitwise/shift operands, signed-only unary negation, integer/bool unary type rules, and rejects shift amounts that exceed the left operand's bit width. Meta evaluator constant-folds the new operators. MIR verifier accepts the new IR opcodes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds AArch64 instruction encodings for ORR, EOR, MVN, NEG (via SUB with XZR), LSLV, LSRV, ASRV, plus AND was reused. The Mach-O backend now lowers IR_VALUE_UNARY (negation, logical not, bitwise not) and the new binary opcodes IR_BIN_BITAND/BITOR/BITXOR/SHL/SHR. Right shift selects LSR for unsigned and ASR for signed operands. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
x86_64 ELF and Windows COFF backends gain encodings for AND/OR/XOR (0x21/0x09/0x31), SHL/SHR (D3 /4-/7), NEG/NOT (F7 /3/2), and Bool not via test+sete+movzx. The AArch64 ELF MVP backend only lowers single-literal returns. To avoid silently dropping the new opcodes, it now walks each function's IR and emits an explicit CGEN004 if it encounters IR_VALUE_UNARY or the new bitwise/shift IR_VALUE_BINARY ops, instead of producing a wrong exit-0 binary. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The lexer no longer folds `>>` into a single token so existing generic syntax like `Foo<Bar<T>>` keeps closing cleanly. The expression parser now recognises two adjacent `>` tokens (no whitespace between them) as the right-shift operator at the new shift precedence. Adds passing fixtures for unary negation, logical not, and bitwise operators (with precedence checks and a UTF-16 surrogate-pair to UTF-8 encode), plus a failing fixture that exercises the shift-amount bit-width check via TYP002. Wires them into the conformance runner. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds an Operators section listing the binary precedence tiers from || down to multiplicative, the unary prefix operators (-, !, ~), and the integer-only constraints for bitwise and shift operators. Notes the special handling of two adjacent `>` tokens as `>>` so `Foo<Bar<T>>` continues to parse. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
@WalterGropius is attempting to deploy a commit to the Vercel Labs Team on Vercel. A member of the Team first needs to authorize it. |
WalterGropius
added a commit
to WalterGropius/zero
that referenced
this pull request
May 22, 2026
This is the B8 narrow slice on top of the existing P0-1 ABI classification
scaffolding: a function declared `fun foo(...) -> f32 { return 1.5 }` (and
the f64 analogue) now builds and runs on the linux-musl-x64 host-runnable
target instead of hitting `CGEN004: direct backend return type is
unsupported actual: f32`.
Per-backend ABI verdict:
- elf64 (System V AMD64): real lowering. Float literals materialise into
xmm0 via `mov eax, imm32 + movd xmm0, eax` (f32) or `movabs rax, imm64 +
movq xmm0, rax` (f64), avoiding the rodata machinery for scalar floats.
Function returns leave xmm0 as the SysV return register. Equality
comparisons (`==` / `!=`) use `ucomis{s,d} xmm1, xmm0` and combine ZF/PF
for IEEE 754 semantics — `eq` is `ordered AND ZF=1`, `ne` is the
complement. Ordered float comparisons (`<` `<=` `>` `>=`) are explicitly
rejected with CGEN004 because the narrow scope does not need them and
unordered handling would otherwise be silently wrong.
- macho64 (Darwin AArch64): keeps the existing CGEN004 path; AAPCS64 s0/d0
lowering is the named follow-up.
- coff (Windows x64): keeps the existing CGEN004 path; the SysV-style
xmm0 lowering would mostly carry over but is not enabled in this PR.
- elf_aarch64 (Linux AArch64): now explicitly rejects f32/f64 returns with
CGEN004 in `a64_reject_unsupported_function`, which is also called from
the literal-only emitter so the silent-drop "MVP subset" trap from
PR vercel-labs#122 / PR vercel-labs#188 can not re-trigger for the new IR float types.
Float params and float locals stay rejected by the IR (`ir_type_is_direct_abi`
unchanged); a new `ir_type_is_direct_abi_return` opens the door only at the
return position. Aggregate (Span/MutSpan/String/Maybe/shape) param+return
marshalling is also still out of scope — that's the next slice and the
classification annotation continues to surface in their CGEN004 messages.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
WalterGropius
added a commit
to WalterGropius/zero
that referenced
this pull request
May 22, 2026
Adds two host-runnable conformance fixtures whose `returnF32() == 1.5` / `returnF64() == 2.25` round-trip exercises the SysV AMD64 xmm0 path on linux-musl-x64. On any host whose runnable direct target is not elf64 (darwin-arm64 macho64, in our case) the `assertCommonRuntimeOrUnsupported` helper accepts the explicit CGEN004 path instead, mirroring the existing runnable-target gating used by other native fixtures. Adds three explicit-rejection assertions — one per still-deferred backend (darwin-arm64 macho, win32-x64.exe coff, linux-arm64 elf) — so the "never silent-drop" invariant from PR vercel-labs#122 / PR vercel-labs#188 is now pinned for the new f32/f64 IR types as well. Updates `docs/articles/target-capabilities.md` with a per-backend f32/f64 table and `zero explain CGEN004` with the corresponding scope note. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds parsing, type-checking, IR lowering, and emitter coverage for the unary
operators
-,!,~and the binary operators|,^,<<,>>(plusbinary
&precedence). Previously these all failed at parse time withPAR100.Source
Addresses B4 + P1-6. Reproduced against main @ 999a46b.
Before this change every form below failed at
PAR100:-1,-x,-1.0!px | y,x ^ y,~xx << 1,x >> 1Multiplicative
*,/,%already worked, and binary&,&&,||alreadyparsed via existing precedence, so no arithmetic work was needed there.
Changes
|,^,~,<<as new tokens.>>is intentionallykept as two
>tokens so generic terminators likeFoo<Bar<T>>stillclose; the expression parser folds two adjacent (no-whitespace)
>intoa
>>shift operator.EXPR_UNARYnode for prefix-,!,~. Binaryprecedence (high to low): unary >
* / %>+ - +% +|><< >>>&>^>|> cmp >&&>||.shifts require an integer left operand and an unsigned shift amount;
constant shift amounts at or beyond the operand bit width are rejected
as
TYP002. Unary-rejects unsigned types;!requiresBool;~requires an integer. The meta evaluator constant-folds all of the new
operators.
IrBinaryOpwithBITAND/BITOR/BITXOR/SHL/SHRand addsIR_VALUE_UNARYwithNEG/NOT/BITNOT. Right-shift selects logical (LSR)for unsigned and arithmetic (ASR/SAR) for signed operands. MIR verifier
enforces operand and result types for the new opcodes.
emit_macho64.c(AArch64 Mach-O): real lowering using AND/ORR/EOR,LSLV/LSRV/ASRV, NEG (SUB with XZR), MVN, and a CBZ-based bool not.
emit_elf64.c(x86_64 ELF) andemit_coff.c(x86_64 COFF): reallowering using
AND/OR/XOR (21/09/31 c8),SHL/SHR/SAR via cl (D3 E0/E8/F8),NEG/NOT (F7 D8/D0), and test+sete+movzx for boolnot.
emit_elf_aarch64.c(AArch64 ELF, MVP literal-return backend):rather than silently dropping new opcodes, walks the IR for each
function and reports an explicit
CGEN004if any new opcode appears.Conformance
Adds three passing fixtures and one failing fixture, wired into
conformance/run.mjs:conformance/native/pass/ops-unary-neg.0– unary negation across types,including nested negation and
-(a + b).conformance/native/pass/ops-not.0–!,!!, and!(...)overBoolexpressions including short-circuit interactions.
conformance/native/pass/ops-bitwise.0–&,|,^,<<,>>,~,precedence checks, and a UTF-16 surrogate-pair (
0xD83D/0xDE00)decoded to its codepoint
U+1F600and re-encoded to the UTF-8 bytesF0 9F 98 80using only the new shift/mask operators.conformance/native/fail/shift-overflow.0–1 << 32against ani32is rejected at check time with
TYP002.Local gates pass on darwin-arm64:
ZERO_NATIVE_TEST_ALLOW_LOCAL=1 node conformance/run.mjs→conformance oknode --experimental-strip-types scripts/snapshot-command-contracts.mts→command contract snapshots okThe
linux-musl-x64cross step innative:test:localcontinues to fail onthis host because no cross compiler / zig is installed; this is
environmental and not a regression introduced here.
Proposed CHANGELOG line
-,!,~and binary|,^,<<,>>for integer expressions; bitwise/shift operands are integer-only,shifts validate the shift amount against the operand bit width, and
>>selects logical or arithmetic shift based on signedness.