[rasuvaeff]

Fixes #155

A shrink candidate was accepted on the exception class alone, so a body that can fail in more than one place — two assertions, or an assertion beside a call that throws the same type — offered the descent failures it could not tell apart. Shrinking could leave the failure that was found and minimise another one, and the reported minimal counterexample then belonged to a bug the run never started from, with nothing in the output saying so.

Internal\FailureIdentity adds the place to the criterion: the exception class and the line of the property's own file in the failure's stack. It is built once per descent and compared against the ORIGINAL failure, never against the last accepted one.

The two points the issue left open

getFile()/getLine() alone is nearly a no-op for both adapters. PHPUnit raises every constraint failure from Constraint::fail() (Constraint.php:124) and Testo every comparison from assert/Assert.php:55, so the throw site is identical for two assertSame calls in one body — the headline case. Verified on live traces through PhpUnitTrialExecutor and TestoTrialExecutor.

Walking to "the first frame outside the engine" does not work either. Under Testo the body is reached through the framework's interceptor pipeline, so the frames next to the engine belong to Testo, not to the property. The place is instead read from the property's own file, which the class part of the property id names — both adapters derive the id from the test class. The scan is bounded the other way round: it stops at the runner's own file, because beyond it the stack is the engine calling the body, and an adapter that invokes the body from the test file (PHPUnit's check() call) would otherwise report that call as the site of every failure raised outside the body while an adapter that invokes it through reflection reported none — the same property would shrink differently under the two.

Nothing is guessed when the place is unreadable — an id that names no loaded class, a body written in another file (a trait, an included fixture), a failure raised by the engine rather than by the body. There the class stays the whole criterion. A criterion that guessed, or one that compared whole traces or frame counts, would refuse candidates that fail exactly as the original did (a failure reached through recursion arrives on a stack of a different height on every input) and stall the descent on the input it started from — the same silence, at a larger counterexample.

Neither the corpus nor a path replay is affected: a corpus entry stores no failure identity, and a replayed path accepts any falsification by design.

Before / after, through the real adapters

Two assertions in one body, seed: 42, Gen::intBetween(0, 10_000), failing at the first assertion for value >= 100:

before: Shrunk: value=0    Failure: `0` is the same as `-2`     <- the other assertion
after:  Shrunk: value=100  Failure: `100` is the same as `-1`   <- the one that was found

Identical under property-testing-testo and property-testing-phpunit.

Behaviour note for stateful properties

With the rule-based façade the #[Rule] methods live in the test file, so a command sequence whose failure moves from one rule to another is now a different failure and the descent stops before it. No existing test changed, but it is worth knowing.

Verification

• composer build: green (1680 passed, 1 skipped), Psalm 0, cs 0, Rector clean.
• Mutation: gate 97 passes — 3880 killed, 115 escaped (all pre-existing, elsewhere). FailureIdentity on its own: 39/39 mutants killed, no ignores; the two unkillable ones were removed by construction (a redundant unknown-file guard) and pinned by a test (an autoloader registered to assert an id never reaches it).
• Both drift tests fail on master's runner and pass here.
• Adapter suites against this core: testo 224, phpunit 139, names 72 — all green.
• bc-check reports the five pre-existing [BC] SKIPPED entries (enum defaults in promoted readonly properties) and nothing else, which this workflow's gate treats as a pass.

Docs: README.md + README.ru.md, llms.txt, AGENTS.md, and a new "What counts as still failing" section in docs/src/guide/shrinking.md. Changelog entry under ## Unreleased, a minor: the compatibility policy §2 already states that what a descent minimises to is not under SemVer.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 83be7832-2a26-448d-8f4b-33cdba594690

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}