Reproducer for #335 fix #2: parse-asymmetry should fall back to simple hash diff

[tinder-maxwellelliott]

Summary

Adds two @Ignored reproducer tests for #335 fix #2 (the "fall back to computeSimpleImpactedTargets when parse asymmetry is detected" proposal in the issue body).

Today, when the two module-graph JSON payloads passed to get-impacted-targets differ but one of them fails to parse, CalculateImpactedTargetsInteractor.detectChangedModules returns the full set of modules from the parseable side via findChangedModules(emptyMap, fullMap). Every module is then treated as "added". The impacted set explodes in one of two ways:

• With a real BazelQueryService bound, every "added" module fans out into an rdeps query against its canonical repo(s). On the workspace in Upgrading 18.0.x → 18.1.0 with a reused base graph triggers multi-hour queryTargetsDependingOnModules fan-out #335 this produced ~5,000 serial subprocesses and the run took multiple hours.
• With no BazelQueryService bound (or one that errors), the failure-tolerant fallback at CalculateImpactedTargetsInteractor.kt:321-323 returns allTargets.keys — every hashed label is reported as impacted.

Both outcomes are far worse than the per-target hash diff that would have run if no module graph had been supplied. PR #336 made the parser tolerant of stderr-polluted JSON, which covers the historical 18.0.x base graph case, but the asymmetry remains a real failure mode for any genuinely unparseable input (corrupted base graphs pulled from object storage, truncation, a future bazel mod graph serialisation change).

The reproducer

Two @Ignored tests in cli/src/test/kotlin/com/bazel_diff/interactor/CalculateImpactedTargetsInteractorTest.kt cover both code paths that branch on changedModules.isNotEmpty():

• execute_parseAsymmetryFallsBackToSimpleHashDiff_reproducerForIssue335Fix2 — exercises execute() (the default get-impacted-targets path).
• executeWithDistances_parseAsymmetryFallsBackToSimpleHashDiff_reproducerForIssue335Fix2 — exercises executeWithDistances() (the --depsFile path).

Both feed:

• A fromModuleGraphJson with no { at all ("garbage-non-json-payload") so ModuleGraphParser.parseModuleGraph falls through to if (start < 0) return emptyMap().
• A toModuleGraphJson that parses cleanly to one or two modules.
• A startHashes / endHashes pair where only //:changed has actually changed.

Each test asserts that the impacted set is exactly {//:changed}. Today both fail with:

expected to contain only:<["//:changed"]>
 but was:<["//:changed", "//:unchanged_a", "//:unchanged_b"]>

The @Ignored state keeps CI green; the annotations document the issue. Drop @Ignore once an asymmetry-detection fallback lands (e.g. "if parseModuleGraph returns an empty map for one side and a non-empty map for the other, skip queryTargetsDependingOnModules and fall through to computeSimpleImpactedTargets").

Verification

• bazel test //cli:CalculateImpactedTargetsInteractorTest — PASSED (new tests are @Ignored and skipped; existing 19 tests still pass).
• Temporarily dropped the @Ignore annotations and re-ran — FAILED with the assertion above, confirming the reproducer catches the bug.

Test plan

• bazel test //cli:CalculateImpactedTargetsInteractorTest — passes with @Ignored.
• Manually un-@Ignore'd and re-ran — both tests fail with the expected "every target reported" assertion error.
• After the asymmetry fallback lands, drop @Ignore from both tests and re-run — both should pass.

Generated with Claude Code (https://claude.com/claude-code)