Conversation
`parse_if_stmt` handles two shapes -- `else if` and `else {` -- and has
no third arm. On any other token it consumes `else` and returns, so the
else-body is parsed as the next statement of the ENCLOSING block and
runs unconditionally.
if (a > 10) {
x = 1;
} else
x = 2;
emitted, in every backend, the same thing as if the `else` keyword had
been deleted: `if (a > 10) { x = 1; }` then `x = 2;`. `pick(20)` returned
2 where the spec says 1.
The decisive control is a comparison, not a reading: the braceless output
is BYTE-IDENTICAL to the else-deleted control and DIFFERENT from the
braced one, in Zig, C, Rust and Verilog alike. The token contributes
nothing.
Nothing reported it. `parse-complete`, this project's own
silent-truncation detector, prints "nothing discarded" for a file whose
`else` has vanished.
The then-branch has had this arm all along, twenty lines above:
`// single statement: if (cond) return expr;`. Only the else-branch was
missing it.
4 sites, in `numeric/gf16.t27` and `numeric/tf3.t27`, all of the shape
`if (a_val >= b_val) return a else return b;`. Early return accidentally
preserves their semantics today, so the corruption there is latent and
fires on the first non-return else-body. An earlier audit reported this
same shape with "0 corpus sites".
Two further specs changed output with zero restored `else`
(`igla/coder/pipeline.t27` and its import copy). Sorted content is
identical in both -- only the splice ORDER moved, because the fixpoint
now discovers the else-body's identifiers at a different point. Checked
by sorting and comparing, not assumed.
No measure moved: zig ast-check 228/572, cc 173, iverilog 379, suite
2455 passed 0 failed -- the wrong parse was always valid output.
FROZEN_HASH resealed.
Refs #2880
gHashTag
enabled auto-merge (squash)
August 29, 2026 20:44
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
PR DashboardGenerated at: 2026-08-29 20:51:54 UTC
Summary
Seal Status
|
gHashTag
added a commit
that referenced
this pull request
Aug 29, 2026
… improvements (#2888) 335: an audit ruled out the braceless `else` with "the corpus has 0 such sites -- all 37 `} else <non-brace>` hits are paren-less `else if`". There are four. They do not match that pattern because the `if` has no braces either: `if (a >= b) return a else return b;` is one line, so there is no `}` before the `else` to anchor on. A zero from a pattern that cannot match the real spelling CLOSES the question, which makes the defect harder to find again than if it had never been reported. Before believing a zero, feed the pattern a case you know exists. 336: two ratchets went red in one hour and neither on a regression. The Lean/Rust classifier reported four new disagreements because four specs that could not be PARSED became parseable, so the classifier could finally disagree with a theorem that had always been wrong. The corpus ratchet reported 19 unexpected passes and 1 unexpected failure -- the nineteen are that same unblocking, and the one failure is one of the nineteen, now parsing and therefore now measurable for discard. A down-only ratchet fails on an improvement exactly as on a regression, and that is the design. The work is the same work, and the failure mode is doing it without writing the reason down. Refs #2883, #2887
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.
parse_if_stmthandles two shapes —else ifandelse {— and has no third arm. On any other token it consumeselseand returns, so the else-body is parsed as the next statement of the enclosing block and runs unconditionally.emits, in every backend, the same thing as if the
elsekeyword had been deleted:pick(20)returns 2 where the spec says 1.The decisive control is a comparison, not a reading
Three variants of one spec — braceless else, braced else, and
elsedeleted:The token contributes literally nothing. That answers "is it just a representation I find surprising" — no backend recovers the branch from anywhere else.
Nothing reported it
parse-complete, this project's own silent-truncation detector, printsnothing discardedfor a file whoseelsehas vanished. All five commands exit 0 with zero bytes on stderr.The then-branch has had this arm all along
Twenty lines above, in the same function:
Only the else-branch was missing it.
Scope, stated honestly
4 sites, in
numeric/gf16.t27andnumeric/tf3.t27, all of the shapeif (a_val >= b_val) return a else return b;. Early return accidentally preserves their semantics today, so the corruption there is latent and fires on the first non-return else-body — which the minimal spec above is.An earlier audit reported this same shape with "0 corpus sites". The count is 4.
Two further specs changed output with zero restored
else—igla/coder/pipeline.t27and its import copy. Sorted content is identical in both: only the splice order moved, because the fixpoint now discovers the else-body's identifiers at a different point. Checked by sorting and comparing rather than assumed.No measure moved
zig ast-checkccacceptsiverilogacceptsThe wrong parse was always valid output in all four languages.
bootstrap/stage0/FROZEN_HASHresealed.Refs #2880