Skip to content

t27b: _ = p; for a parameter lowers to nothing where the reference compiles it (Closes #7057) - #7111

Merged
gHashTag merged 9 commits into
masterfrom
claude/t27b-lane1-param-discard
Oct 6, 2026
Merged

t27b: _ = p; for a parameter lowers to nothing where the reference compiles it (Closes #7057)#7111
gHashTag merged 9 commits into
masterfrom
claude/t27b-lane1-param-discard

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 6, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

Refs #6063. Closes #7057.

What

A port that keeps a parameter for its interface writes _ = p;. The reference compiles it, but t27b treated _ as an assignment target and refused the file with StmtAssign(undeclared).

_ = p; where p is a parameter of the current fn now lowers to nothing. The rule follows what the reference does, which the lab measured:

  • At the top level of a fn body, t27c's dead-store pass deletes the discard. gen-zig then prints its own _ = p; // unused by the spec body if needed, or renames p to p_arg when a module declaration shares its name. t27b always accepts it there.
  • In a nested block, gen-zig prints the discard as written. Zig refuses it only as a "pointless discard of function parameter", which happens when the body also uses p. t27b counts the body's mentions of p and its _ = p; statements, and refuses the nested discard only when the two counts differ.
  • Discards alone compile: repeated, in both arms of an if, after a top-level discard, or for a parameter that shares a module declaration's name.
  • _ = x; for a local or an undeclared name is still refused (StmtAssign(undeclared)).

Dogfood spec first

specs/tri/t27b/conformance/param_discard.t27 has 7 tests: scalar, struct and slice parameters, a discard in a branch, every parameter discarded, a top-level discard next to a read, and nested discards alone.

  • Reference (t27c test-report): 7 pass, 0 vacuous.
  • t27b: 7 pass, 16 runtime asserts.
  • 12 mutants, all on the Railway lab:
    • 8 change a value, and each fails exactly the mutated test in t27b and in the reference alike.
    • 4 add a nested discard next to a real use. The reference refuses each one with "pointless discard of function parameter", and t27b refuses it with StmtAssign(discard).
  • 14 negative controls: 11 pass in both, and 3 are refused by both (the reference does not compile them, and t27b raises StmtAssign(discard)). There is no case where t27b and the reference disagree.

A dangling slice found on the way

With the discard lowered, t27b reaches queen-rehearsal.t27 for the first time, and its interpreter faulted in build_content_frame. That fn returned a slice of its own local [256]u8, which dangles once the fn returns. The reference's Zig and t27b's JIT read the dead stack by luck: the corpus showed a JIT/interpreter mismatch. The spec now takes a caller-owned buffer, as gen_escape does. The reference still gives 7 pass with 0 vacuous, and t27b gives 7 pass with no mismatch.

Numbers (Railway lab, same reference, same time)

master 0f35060 branch cabf35c
t27b pass where reference passes (union of 839 reference-pass files) 527 530
mismatch / reference_disagree 0 / 1 (check_fix_carries_source, a reference flake) 0 / 0 after the queen-rehearsal fix
  • Gained: param_discard.t27, leaderboard.t27, queen-rehearsal.t27 (after the spec fix). d_g22_test moved from timeout to pass; that is load noise.
  • The branch run's reference reported 17 zig not on PATH (EAGAIN) under load. They lower the reference count on that side only and do not change any t27b verdict.
  • First blocker moves: dqn.t27 now stops at ExprCall(undeclared fn) and gen_gradient.t27 at StmtExpr statement.
  • hex.t27 is unchanged. Its first blocker is ExprStructLit(value ignored).
  • Ledger after merging master a7a85d9 (which includes t27b: a comptime_float is a binary128 value, folded like Zig (Closes #7007) #7045): pass 485 -> 488, not-pass 50 -> 48, cap 50 -> 48.
  • The remaining ratchet red (UNLISTED, crons vacuity, kernel timeouts) is master drift that this PR does not touch.
  • cargo test -p t27b on the lab, on the merged tree eac7077: 14 suites, 118 passed, 0 failed. The lib's tests_hir_roundtrip uart and bridge tests fail as they do on master: they resolve ../specs from cli/t27b, and the lib is test = false.

Foreign-code approval

The owner approved this Rust edit: label owner-approved-foreign is on #6063, and the owner said "add the label yourself and do the work" (translated). The edits are minimal and confined to cli/t27b/src/lower.rs and cli/t27b/tests/source.rs. Both files are listed in tools/policy/foreign-exceptions.txt.

🤖 Generated with Claude Code

gHashTag and others added 8 commits October 7, 2026 02:17
@gHashTag @claude
t27b: conformance spec for the parameter discard _ = p; (Closes #7057)
3a92dec 
Refs #6063. Dogfood spec first: specs/tri/t27b/conformance/param_discard.t27
holds scalar, struct, slice-neighbour, in-branch and all-parameter
discards, plus a discard in a fn no test calls.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
t27b: lower _ = p; for an otherwise unnamed parameter (Closes #7057)
84ab690 
Refs #6063. A discard of a parameter that the body names nowhere else, and
that no module declaration shares, lowers to nothing.

Every other discard of a parameter is refused as StmtAssign(discard):
- one the body also reads, which Zig's AstGen refuses as a pointless
  discard in every fn;
- a second discard of the same parameter;
- a parameter gen-zig renames because a module declaration shares its
  name.

Owner-approved foreign edit (label owner-approved-foreign on #6063),
listed in tools/policy/foreign-exceptions.txt.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
t27b: a top-level _ = p; is deleted as the reference deletes it (Cl…
8be6477 
…oses #7057)

Refs #6063. The reference's dead-store pass drops every top-level
`_ = <no call>;` from a fn body. gen-zig then discards the parameter
itself when nothing else reads it. So the statement is as if absent,
including next to a read of `p` and when written twice.

Only a nested `_ = p;` is printed as written, and only that one can be a
pointless discard. The spec and tests now cover both cases.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
t27b: a nested _ = p; is refused only next to a use of p (Closes #…
e3e0e11 
…7057)

Refs #6063. The lab showed that the reference compiles a nested
`_ = p;` when the body names `p` only in discards: repeated, in both
arms, after a top-level discard, or for a parameter that shares a
module declaration's name. Zig refuses it only as a pointless discard,
that is next to a real use of `p`. t27b now refuses exactly that case,
and the spec and tests cover the discards that compile.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
Merge origin/master into claude/t27b-lane1-param-discard (Closes #7057)
cabf35c 
Refs #6063. Brings in #6998 (untyped local), which added a test and a
foreign-exceptions line next to this branch's. Both are kept.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
queen-rehearsal: the caller owns the content buffer, no dangling slice (
0143354 
Closes #7057)

Refs #6063. With `_ = frame;` lowered, t27b reaches this spec for the
first time, and its interpreter faults in build_content_frame: the fn
returned a slice of its own local `[256]u8`, which dangles once it
returns. The reference's Zig and t27b's JIT read the dead stack by luck
(jit Ok(1), interpreter "1-byte access outside every live slot"). The
buffer is now a `[]u8` parameter the test owns, as gen_escape does; the
reference still gives 7 pass, 0 vacuous, and t27b gives 7 pass with no
JIT/interpreter mismatch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
t27b ledger and NOW entry for parameter discards (Closes #7057)
b3095b9 
Refs #6063. leaderboard, queen-rehearsal and the new param_discard
conformance spec move to pass. dqn now stops at ExprCall(undeclared fn)
and gen_gradient at StmtExpr statement. Not-pass goes from 52 to 50.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag @claude
Merge origin/master into claude/t27b-lane1-param-discard (Closes #7057)
565bc43 
Refs #6063. Brings in #7045 (comptime_float) and #6695 (interpreter
fuel). The ledger counts are recounted on top of master: pass 483 ->
486, not-pass 51 -> 49.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag gHashTag added the owner-approved-foreign Owner-approved exception to the only-t27 rule: hand-written foreign code allowed in this PR label Oct 6, 2026
@gHashTag
gHashTag enabled auto-merge (squash) October 6, 2026 20:56
@gHashTag gHashTag mentioned this pull request Oct 6, 2026
Closed
This was referenced Oct 6, 2026
Merged
Merged
Merged
Merged
Open
Merged
Merged
Merged
Open
Open
Merged
Merged
@github-actions

github-actions Bot commented Oct 6, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-06 21:20:48 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 47
PRs with All Checks Green 3
READY 0
FAILING 47
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 47 + 0 + 0 = 47, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b23641f01baa != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 6, 2026
Merged
Merged
Merged
@gHashTag @claude
Merge origin/master into claude/t27b-lane1-param-discard (Refs #7057)
eac7077 
Ledger conflict: both conformance entries kept, counts recounted (pass 488, not-pass 48).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 6, 2026
Open
Merged
This was referenced Oct 6, 2026
Merged
Merged
Open
Merged
Open
Merged
Merged
Open
@gHashTag
gHashTag merged commit aecf75e into master Oct 6, 2026
29 of 31 checks passed
@github-actions github-actions Bot mentioned this pull request Oct 6, 2026
Merged
@github-actions

github-actions Bot commented Oct 6, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-06 21:59:49 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 47
PRs with All Checks Green 3
READY 0
FAILING 47
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 47 + 0 + 0 = 47, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=b23641f01baa != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 6, 2026
Merged
Merged
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

owner-approved-foreign Owner-approved exception to the only-t27 rule: hand-written foreign code allowed in this PR

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

t27b: lower the parameter discard _ = p; (StmtAssign(undeclared))

1 participant

@gHashTag