Skip to content

Update latest-results with post-fix benchmark run#26

Closed
kinto0 wants to merge 3 commits into
microsoft:mainfrom
kinto0:fix/update-latest-results
Closed

Update latest-results with post-fix benchmark run#26
kinto0 wants to merge 3 commits into
microsoft:mainfrom
kinto0:fix/update-latest-results

Conversation

@kinto0

@kinto0 kinto0 commented Apr 22, 2026

Copy link
Copy Markdown
Contributor

Summary

Stack: Includes #24 and #25. Merge those first, then this PR's diff is just the regenerated latest-results files.

Test plan

  • Before: latest-results.md shows: Pyrefly | no | 221.27 | ... | 1 (tsp_core failed point)
  • After: latest-results.md shows: Pyrefly | yes | 316.80 | 0.17 | 8 | 40 | 100% | 0. All 4 servers benchmarked. Pyrefly: 0 failures across 8 suites. pylsp-mypy: 5 failures (pre-existing Jedi issues, unrelated).

kinto0 added 3 commits April 21, 2026 17:52
_install_state_path() called Path.resolve() on the suite venv's python
executable, which follows symlinks to the base interpreter. On systems
where the base interpreter lives in a read-only location (e.g. a system
Python framework at /usr/local/fbcode/), this causes a PermissionError
when writing the per-suite install state file.

Drop .resolve() so the state file is written relative to the venv path,
which is always writable.

Test plan:
  Before: bench-servers crashes immediately with PermissionError:
    [Errno 1] Operation not permitted:
    '/usr/local/fbcode/platform010/Python3.12.framework/...
    /.python-lsp-compare-install.json'
  After:  bench-servers completes successfully across all 4 servers.
    Pyrefly: 1 failed point (tsp_core/generic specialization —
    pre-existing config bug, not caused by this change).
    43 unit tests pass.
The "generic specialization computed type" benchmark point targeted
line 7 (0-indexed) in generics.py, which is an empty line. The actual
expression `text = identity("hello")` is on line 9. Pyrefly's
getComputedType correctly returned null for the empty line, causing
requireNonEmpty validation to fail on every iteration.

This was the sole Pyrefly failure across all 8 benchmark suites and
has been present since the original tsp_core commit (800936d).

Fix: change start_line and end_line from 7 to 9.

Test plan:
  Before: bench-servers --server pyrefly --protocol tsp shows:
    pyrefly: [tsp_core] generic specialization computed type failed:
    Result validation failed: iteration 1: empty result;
    iteration 1: size_chars=0 < 10 (repeated for all 5 iterations)
    tsp_core: failed (1 failed point out of 8)
  After:  bench-servers --server pyrefly --protocol tsp shows:
    pyrefly: [tsp_core] generic specialization computed type ok
    tsp_core: ok (0 failed points, 8/8 pass)
    Pyrefly total: 0 failures across all 8 benchmark suites.
    43 unit tests pass.
Re-ran full unfiltered bench-servers after the tsp_core config fix.
Filtered runs (--server/--protocol) intentionally skip updating
latest-results.md, so it still showed the pre-fix failure.

Test plan:
  Before: latest-results.md shows:
    Pyrefly | no | 221.27 | ... | 1 (tsp_core failed point)
  After:  latest-results.md shows:
    Pyrefly | yes | 316.80 | 0.17 | 8 | 40 | 100% | 0
    All 4 servers benchmarked. Pyrefly: 0 failures across 8 suites.
    pylsp-mypy: 5 failures (pre-existing Jedi issues, unrelated).
@kinto0 kinto0 closed this Apr 22, 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