Repository navigation
tri misread: the specs the compiler reads wrongly, with a control that refuses - #3228
Merged
Merged
Conversation
…t refuses `tri unparsed` ranks what the parser could not read and its population is defined by "the compiler refused". Anything accepted lies outside it by construction, however wrong the result, so the class where every gate is green had no census. This one reads the GENERATED OUTPUT instead, where being read wrongly becomes visible. Four shapes, each a line that cannot mean anything in its target language: `pub f: ,`, `pub f: 0,`, `Vec<>`, and the C half `0 f;`. The positive control is the design. The command runs a reproducer through the real compiler first and refuses to report on the corpus unless every shape it claims to detect actually fired on a case built to contain it -- because a detector that finds nothing and a detector that is broken print the same thing, which is the defect this command exists for. It earned that on its first run: the reproducer carried only the literal case, two detectors had no case, and it refused instead of printing two comfortable zeros. Corpus: 22 / 1 / 0 / 1. The `Vec<>` zero is a result rather than a silence because the control fired; #3213 removed that defect from the corpus and the detector stays as a regression guard. The run also corrected #3225: the ad-hoc grep behind its "14" required `pub [a-z_]+: ,`, a field-name class with no digits and no capitals, and missed eight specs that name the field Double, Null, Nil, Ready, IgnoreCase, Ascending, RGB and Trace. All 22 are in the rustc-FAIL set. Refs #3225 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
gHashTag
enabled auto-merge (squash)
September 5, 2026 02:04
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
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
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.
tri unparsedranks what the parser could not read, and its population is defined bythe compiler refused. Anything accepted lies outside it by construction, however
wrong the result — so the class where every gate is green had no census. This one
reads the generated output instead, where being read wrongly becomes visible.
Four shapes, each a line that cannot mean anything in its target language:
pub f: ,,pub f: 0,,Vec<>, and the C half0 f;.The positive control is the design, not decoration
The command runs a reproducer through the real compiler first and refuses to
report on the corpus unless every shape it claims to detect actually fired on a case
built to contain it. A detector that finds nothing and a detector that is broken print
the same thing, and that is the defect this command exists for.
It earned that on its first run — the reproducer carried only the literal case, two
detectors had no case, and it refused rather than printing two comfortable zeros:
What it reads now
The
Vec<>row reading 0 is the point: #3213 removed that defect from the corpus,and the control is what makes the zero a result rather than a silence. The detector
stays as a regression guard.
It corrected the issue that motivated it
#3225 said 14. The ad-hoc grep behind that number required
^[[:space:]]+pub [a-z_]+: ,$— a field-name class with no digits and no capitals— and missed eight specs naming the field
Double,Null,Nil,Ready,IgnoreCase,Ascending,RGB,Trace. All enum-variant names, which is the sameunderlying cause. Verified: all 22 are in the rustc-FAIL set, so none is a false
positive.
9 unit tests, including one asserting the reproducer still contains a marker for every
shape — so a "tidied" constant fails at test time rather than making the control pass
vacuously.
Refs #3225