Skip to content

[TS] Measure source-replayed branch coverage for unknown-call policies - #414

Open
CaelmBleidd wants to merge 2 commits into
caelmbleidd/shared-ts-adapter-fixesfrom
caelmbleidd/iccq-coverage-v2
Open

CaelmBleidd wants to merge 2 commits into
caelmbleidd/shared-ts-adapter-fixesfrom
caelmbleidd/iccq-coverage-v2

Conversation

@CaelmBleidd

@CaelmBleidd CaelmBleidd commented Oct 1, 2026 •

Copy link
Copy Markdown
Member

Summary

  • Add a coverage search that uses CLOSEST_TO_UNCOVERED_RANDOM without a selected return target.
  • Persist generated inputs as soon as a completed symbolic state adds entry-function CFG coverage.
  • Replay each input in a fresh Node.js process on the original TypeScript and record source statements and exact if branch arms.
  • Replay zero-argument functions through a discarded harness sentinel; 63 of the 264 corpus functions have no inputs.
  • Derive nested 30/60/120-second coverage checkpoints from one 120-second search; replay time is separate.
  • Pin JacoDB to merged revision 873f8e99faf3f1e4736935be96ce1c319985c574.

Validation

  • All three CallsCoverageExperimentTest tests passed with the pinned local JacoDB checkout, including exact replay of a zero-argument function.
  • A corpus pilot replayed two inputs and confirmed both source if arms with no replay errors.
  • The first campaign stopped at a zero-argument replay failure after 60 completed cells. It is preserved separately. A new full paired campaign is running against the fixed commit; results are not included in this PR.

Review focus

  • Candidate selection and coverage checkpoint timing.
  • Exact replay transport and source mapping.
  • Durability and resume behavior of individual cells.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant