Repository navigation
fix(ci): avoid NumPy source builds and duplicate HTTPX coverage - #3573
Merged
Merged
Conversation
Contributor
jbeckwith-oai
approved these changes
Aug 3, 2026
jbeckwith-oai
left a comment
jbeckwith-oai
left a comment
Contributor
There was a problem hiding this comment.
Reviewed the exact head (73f5e4a). The per-interpreter universal NumPy locks, explicit Rye virtualenv cache keys, HTTPX2-only test path, and Release Please uv.lock wiring are consistent and keep the workflow direct. I also regenerated both Rye lockfiles, ran the Python support-policy check and full lint, and verified the exact-head CI matrix is green.
Merged
HAYDEN-OAI
approved these changes
Aug 3, 2026
HAYDEN-OAI
left a comment
HAYDEN-OAI
left a comment
Contributor
There was a problem hiding this comment.
Reviewed exact head 73f5e4a7661f2d4ee448af22e3f0c0336435d0f8. No substantive issues found: interpreter-specific Rye cache keys and invalidation, universal NumPy wheel selections, HTTPX/HTTPX2 and Pydantic v1/v2 coverage, required-check names, fork/release behavior, workflow permissions, and CodeQL coverage are consistent. Confirmed HTTPX2 improved from 10m42s to 2m12s and Python 3.14 from 8m37s to 1m53s. Release automation also propagated the version into uv.lock in #3575.
pull Bot
pushed a commit
to gitupdates/openai-python
that referenced
this pull request
Aug 3, 2026
Automated Release PR --- ## [2.53.0](openai/openai-python@v2.52.1...v2.53.0) (2026-08-03) ### Features * **api:** Add gpt-5.5 and tool name/namespace to Responses types ([openai#3569](openai#3569)) ([dd1202d](openai@dd1202d)) ### Bug Fixes * **ci:** avoid NumPy source builds and duplicate HTTPX coverage ([openai#3573](openai#3573)) ([b58332f](openai@b58332f)) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Justin Beckwith <jbeckwith@openai.com>
yonasb
added a commit
to starslingdev/skills
that referenced
this pull request
Oct 21, 2026
…#137) * ci-speedup: four catalog patterns for slow test jobs (OPT85 to OPT88) Adds four catalog entries, each backed by measured changes OpenAI engineers made in their public repos (every figure re-read from the PR body or post it quotes): - OPT85 Tests Run in Many Processes Instead of One (structural, risk MEDIUM): a subprocess per test case, or a test binary per file. openai/openai-python#3830, openai/codex#2633. A new pattern, not a widening of OPT78: OPT78 is vitest's per-file module registry and its scope note keeps it vitest-only; OPT78 gains a one-line pointer. - OPT86 CI Compiles a Dependency That Ships Prebuilt (data-driven, risk LOW). openai/openai-python#3573 (saving not split between its changes, said so), openai/whisper CPU-only torch index, openai-python UV_NO_BUILD. - OPT87 Tests Wait on a Slow Dependency They Do Not Verify (structural, risk HIGH): stub at the boundary, never mock what the test verifies, keep a real-dependency run. OpenAI Core Experimentation post, openai/evals feed6c75. - OPT88 Test Concurrency Is a Guess, Not a Measurement (data-driven, risk MEDIUM). openai/codex#31352, openai/openai-agents-python#2563 (local measurement, said so). No detector code: all four are catalog-only, like OPT25 and OPT74. The structural pair lands in catalog_structural_patterns_without_detector; the data-driven pair, like every detector: manual entry, is never emitted. Counts 82 -> 86 (75 hygiene + 11 structural) in SKILL.md, ARCHITECTURE.md, evals.json; README 70+ -> 80+. SKILL.md line count unchanged (515). Red-first exemption: no behavior to regress. The change is catalog and doc text plus one scan.py comment; no detector, renderer or verifier code changed. The existing count, breakdown and anchor guards cover the new totals (scoped run: 257 passed). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci-speedup: OPT85 to OPT88 say they are applied by hand, and their fix advice no longer breaks installs Review fixes for the four new catalog entries. - SKILL.md, ARCHITECTURE.md, structural-track.md, MAINTAINERS.md and the scan.py return-key comment now say OPT74, OPT85 and OPT87 have no router and are never emitted, instead of listing them as routed. - Each new entry says no audit phase applies it; a reader applies it by hand. - OPT86: --only-binary / --no-build-package scoped to the package that compiled, with the wide form only once every dependency ships a wheel; the PyTorch CPU index goes in its own install command; detection and the sizing ceiling cover the oversized-download case; UV_NO_BUILD is set "at workflow level", matching the source. - OPT85: the test-list guardrail maps renamed Rust tests; random order is required where the runner can shuffle, with a Rust alternative. - OPT87: recorded stub data must have credentials scrubbed; blog quote apostrophe now matches the source. - codex #2633 is described as a developer-machine time measurement. Text corrections are exempt from red-first (no behavior to regress). The two strengthened guards in test_scan_detectors.py were shown failing on a mutated copy (OPT85 added to structural_detected; SKILL.md split 76/10): AssertionError: OPT85 has no router and must be reported as uncovered AssertionError: SKILL.md claims 10 structural patterns but the catalog has 11 `class: structural` entries and pass on this tree. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci-speedup: OPT85 to OPT88 detection steps match what real before and after CI logs show Wording fixes from checking each entry's detection steps against OpenAI's real before/after runs. - OPT85: the test-category step is a gate only; a helper the parametrized test calls counts as inside the test; cases are counted from the collect-only listing; when the slow run has no slowest-tests report, corroborate against a known-good run and by timing one case with and without the process call. - OPT86: the compile span runs to the closing `Built <pkg>==<version>` line; a `Building <project> @ file://` line is the project itself; only the compile span counts toward the ceiling on a leg slow for other reasons. - OPT87: the post names no repository and links no CI runs; the 2023 evals run history is no longer retrievable. - OPT88: concurrency can be set in the tool's own config files (such as `local_test_jobs` in `.bazelrc`); measure the core count of an undocumented custom runner; say so when nothing forced the setting. Text-only corrections, exempt from red-first (no behavior to regress). Headings and anchors unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This branch was previously deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Root cause
uv.locklacked the Release Please marker needed to keep its SDK version current.Fix
uv.lockfor release updates.Results
Validated with lint, SDK builds, lockfile checks, and both HTTPX/HTTPX 2 suites under Pydantic 1 and 2.
Fixes SDK-221.