Skip to content

Take graphcore c03be3a, which bounds the output of grep_files - #297

Merged
shellygr merged 2 commits into
masterfrom
shelly/bump-graphcore-grep-cap
Nov 5, 2026
Merged

shellygr merged 2 commits into
masterfrom
shelly/bump-graphcore-grep-cap

Conversation

@shellygr

@shellygr shellygr commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

What this is

The graphcore submodule and the pyproject.toml pin move from 9f4e9fc to c03be3a. The range is Certora/graphcore#41 and nothing else. uv.lock does not change, because [tool.uv.sources] resolves graphcore to the local path. The file set is the same as in #227.

What graphcore#41 changes

grep_files returned every match in one result. On a cloud run, one agent searched all files for every function declaration. The result was 2.77 MB: 18,018 lines from 1,122 files, mostly vendored libraries under lib/. The result did not fit in the model's context window, and the next call failed.

Now graphcore cuts a matching line after 500 characters and stops the result at 100,000 characters. When it stops, the last line says how many matches it left out. The line also tells the agent to use a more specific pattern or match_in.

Known limit

The result follows the order of the files on disk, not their relevance. So the part that graphcore leaves out can hold project files. The last line tells the agent that the result is incomplete.

🤖 Generated with Claude Code

Move the graphcore submodule and the pyproject.toml pin from 9f4e9fc
to c03be3a. The range is graphcore#41. uv.lock does not change,
because [tool.uv.sources] resolves graphcore to the local path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@shellygr
shellygr marked this pull request as ready for review November 5, 2026 10:07
@shellygr
shellygr marked this pull request as ready for review November 5, 2026 10:07
@shellygr
shellygr merged commit 327a79c into master Nov 5, 2026
4 checks passed
@shellygr
shellygr deleted the shelly/bump-graphcore-grep-cap branch November 5, 2026 12:56
ericeil added a commit that referenced this pull request Nov 9, 2026
Brings in the evaluate run mode (#282, #292, #296), five property-inference
rounds (#294), autosetup EVM-version fixes (#291, #293), graphcore c03be3a
(#297) and the downstream-notify workflow (#279).

Resolutions:
- core.py: the pinned-run path moves into master's _extract(), so an
  external mode also extracts from a pin; extraction_plugins stays at
  function scope for hoisting.
- ptypes.py: all_failed defers to the external mode's verdict first, then
  keeps staging's rule that a curtailed partial counts as output.
- report schema: staging already used 3.2-3.7, so master's evaluate fields
  (rule_feedback, conf_coverage, coverage_scores, scene, mutation) land as
  3.8. Tests renamed accordingly; the older-report test covers 3.0-3.7.
- report_prover.py: master's fetch_run_verdicts name; staging's job_input
  and make_run_link_fetcher kept on top of it.
- template: staging's hoisted section, then master's widened gaps condition.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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