Skip to content

fix(benchmark): always use current build MCP server, add benchmark skill#210

Merged
jafreck merged 1 commit intomainfrom
fix/benchmark-mcp-server-path
Mar 15, 2026
Merged

fix(benchmark): always use current build MCP server, add benchmark skill#210
jafreck merged 1 commit intomainfrom
fix/benchmark-mcp-server-path

Conversation

@jafreck
Copy link
Owner

@jafreck jafreck commented Mar 15, 2026

Fix findLoreProjectRoot() to always resolve the MCP server from the current checkout, not from a pinned-SHA clone that may lack the realpathSync entry-guard fix.

Changes:

  • findLoreProjectRoot() resolves from import.meta.dirname only; throws if dist/server/server.js is missing
  • Remove redundant --deny-tool for control arm (MCP isn't registered, tools don't exist)
  • Remove token count estimator; use outputTokens from copilot NDJSON directly
  • Add copilot-benchmark skill

…skill (#210)

The lore-enabled benchmark arm was silently getting a dead MCP server
because findLoreProjectRoot() checked the cloned repo first. For
lore-self benchmarks the clone is at a pinned SHA (660be2b) that
predates the realpathSync entry-guard fix (ee708f8), so the old
server.js fails the /var vs /private/var check on macOS and main()
never runs.

Changes:
- findLoreProjectRoot() now resolves exclusively from import.meta.dirname
  (the current checkout), never from the cloned target repo. Throws if
  dist/server/server.js is missing instead of silently falling through.
- Remove redundant --deny-tool for control arm — the MCP isn't registered
  so lore_* tools simply don't exist.
- Remove token count estimator — use actual outputTokens from copilot
  NDJSON stream only.
- Add copilot-benchmark skill (.github/skills/copilot-benchmark/SKILL.md).
@jafreck jafreck merged commit 6141936 into main Mar 15, 2026
1 check passed
@codecov
Copy link

codecov bot commented Mar 15, 2026

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.04%. Comparing base (b3312a8) to head (d7a0af5).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files
@@           Coverage Diff           @@
##             main     #210   +/-   ##
=======================================
  Coverage   88.04%   88.04%           
=======================================
  Files          80       80           
  Lines        9048     9048           
  Branches     2798     2798           
=======================================
  Hits         7966     7966           
  Misses       1082     1082           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

jafreck added a commit that referenced this pull request Mar 15, 2026
…for C/C++ (#211)

* fix(benchmark): always use current build's MCP server, add benchmark skill (#210)

The lore-enabled benchmark arm was silently getting a dead MCP server
because findLoreProjectRoot() checked the cloned repo first. For
lore-self benchmarks the clone is at a pinned SHA (660be2b) that
predates the realpathSync entry-guard fix (ee708f8), so the old
server.js fails the /var vs /private/var check on macOS and main()
never runs.

Changes:
- findLoreProjectRoot() now resolves exclusively from import.meta.dirname
  (the current checkout), never from the cloned target repo. Throws if
  dist/server/server.js is missing instead of silently falling through.
- Remove redundant --deny-tool for control arm — the MCP isn't registered
  so lore_* tools simply don't exist.
- Remove token count estimator — use actual outputTokens from copilot
  NDJSON stream only.
- Add copilot-benchmark skill (.github/skills/copilot-benchmark/SKILL.md).

* feat(scip): auto-install indexers and generate compile_commands.json for C/C++

* fix(ci): exclude installer and compdb from coverage thresholds

* test(scip): add unit tests for installer and compdb with injectable IO mocks

* test(scip): raise installer coverage to 86%, restore statements threshold to 85%
@jafreck jafreck mentioned this pull request Mar 15, 2026
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