Run a smaller subset of targets on pull requests by default by zanieb · Pull Request #1075 · astral-sh/python-build-standalone

zanieb · 2026-04-01T13:05:05Z

Currently, pull requests run all targets by default and labels can be used to select a subset of the matrix. However, CI is quite expensive and it's very rare to need to test the whole matrix, so the default is a bit backwards.

In this change, we update pull requests to run on a subset of targets by default: Python 3.14 on macOS, Linux, and Windows with the most popular architecture. We include both glibc and musl variants of Linux. We include armv7 as an arbitrary cross-compile case — I'd be happy to take suggestions on an alternative there.

This breaks our labeling concept a bit, as our labels currently do pure subsetting of the matrix. The labels will continue to subset, but with some nuances:

Applying a label will generally subset the default selection, e.g., platform:linux will only run the Linux subset of the defaults described above.
Labels that would null the default subset, e.g., python:3.12 or build:debug, will instead change the default set to target that variant.
There are new platform:all, python:all, arch:all, libc:all, and build:all labels that can be used to expand the targets.

zanieb · 2026-04-01T14:50:12Z

ci-matrix.py

+    directives = labels.get("directives", set())
    python_entries = generate_python_build_matrix_entries(
        config,
        runners,
        args.platform,
-        labels,
+        {"directives": directives} if directives else None,


This pattern is really weird?

I don't follow why this line is changing. Passing through labels seems like it would work fine?

I fixed this already

geofft

This seems pretty complicated and I'm not sure that I understand exactly what the logic is and a mental model of which builds will run in response to which targets. To be fair I don't feel like I have a 100% grasp of the current logic, and we could decide that we have few enough existing users that backwards compatibility is not a goal. But it feels like it should be possible to do this task with fewer lines of code and more readably.

I'm okay with deciding that we don't want to put time into that, and we can always have an LLM make further big changes if we don't like the behavior.

ci-matrix.py

geofft · 2026-04-01T17:12:12Z

ci-matrix.py

+    directives = labels.get("directives", set())
    python_entries = generate_python_build_matrix_entries(
        config,
        runners,
        args.platform,
-        labels,
+        {"directives": directives} if directives else None,


I don't follow why this line is changing. Passing through labels seems like it would work fine?

geofft · 2026-04-01T17:24:42Z

ci-matrix.py

-        return False
+def get_all_build_options(ci_config: dict[str, Any], target_triple: str) -> list[str]:
+    """Get all build options (including conditional) for a target from ci-targets.yaml."""
+    for _platform, platform_config in ci_config.items():


Suggested change

for _platform, platform_config in ci_config.items():

for platform_config in ci_config.values():

geofft · 2026-04-01T17:34:55Z

ci-matrix.py

+    if expand_platform or expand_arch or expand_libc:
+        source_triples = {}
+        for platform, platform_config in ci_config.items():
+            for triple, config in platform_config.items():
+                source_triples[triple] = (platform, config)
+    else:
+        source_triples = {}
+        for triple in pr_config["targets"]:
+            platform = find_target_platform(ci_config, triple)
+            source_triples[triple] = (platform, ci_config[platform][triple])


This logic is a little bit complicated for me to make sense of. I think that if we unconditionally do the if-case logic (set source_triples to every possible triple in ci-targets.yaml, without filtering by pr-targets.yaml), and remove the if expand_platform or expand_arch or expand_libc: on line 187 and unconditionally run the matches_default_pattern filter, the outcome will be the same. Is that correct?

geofft · 2026-04-01T17:37:57Z

ci-matrix.py

+                if version in ci_target_config["python_versions"]
+            ]
+        else:
+            python_versions = [pr_default_version]


Seems erroneous for pr_default_version to apply if it's not listed in ci_target_config.

I see there's a validation step later but I think that if you use arch:all or whatever to bring in platforms that aren't in the default set, and they don't support the default Python version, the validation won't notice that.

ci-matrix.py

geofft · 2026-04-01T17:44:07Z

ci-matrix.py

+                for option in sorted(build_filters)
+                if option in all_build_options
+            ]
+            build_options_conditional = []
+        else:
+            build_options = get_default_build_options(ci_config, pr_config, triple)
+            build_options_conditional = []


Same comments here.

ci-matrix.py

ci-defaults.yaml

ci-matrix.py

zanieb · 2026-04-10T14:11:04Z

ci-matrix.py

+        with open(CI_DEFAULTS_YAML) as f:
+            ci_defaults = yaml.safe_load(f)
+
+        pull_request_defaults = ci_defaults.get("pull_request")
+        if pull_request_defaults is None:
+            print(
+                f"error: {CI_DEFAULTS_YAML} is missing a pull_request section",
+                file=sys.stderr,
+            )
+            sys.exit(1)


It seems like this should just be unconditionally loaded and we can do ci_defaults.get(args.event)?

(and not error if there's not a section for the event type, since that just means... use everything)

zanieb added ci:dry-run platform:all Select all platforms, e.g., darwin, linux, windows python:all Select all Python versions arch:all Select all architectures labels Apr 1, 2026

zanieb force-pushed the zb/subset-pr branch from cd45bed to e2a9b19 Compare April 1, 2026 13:34

zanieb added platform:darwin Specific to the macOS platform and removed platform:all Select all platforms, e.g., darwin, linux, windows labels Apr 1, 2026

zanieb force-pushed the zb/subset-pr branch from e2a9b19 to f1ca50f Compare April 1, 2026 13:42

zanieb added build:all platform:linux Specific to the Linux platform and removed platform:darwin Specific to the macOS platform labels Apr 1, 2026

zanieb force-pushed the zb/subset-pr branch 2 times, most recently from 44eec28 to b386b4a Compare April 1, 2026 14:30

zanieb added build:debug and removed platform:linux Specific to the Linux platform python:all Select all Python versions arch:all Select all architectures build:all labels Apr 1, 2026

zanieb force-pushed the zb/subset-pr branch from b386b4a to e019d87 Compare April 1, 2026 14:37

zanieb commented Apr 1, 2026

View reviewed changes

Run a smaller subset of targets on pull requests by default

a86682c

zanieb force-pushed the zb/subset-pr branch from e019d87 to a86682c Compare April 1, 2026 15:04

zanieb marked this pull request as ready for review April 1, 2026 16:16

jjhelmus approved these changes Apr 1, 2026

View reviewed changes

geofft reviewed Apr 1, 2026

View reviewed changes

Review

48494bd

zanieb commented Apr 2, 2026

View reviewed changes

ci-defaults.yaml Outdated Show resolved Hide resolved

zanieb added 2 commits April 2, 2026 17:13

More review

33b6dda

Apply suggestion from @zanieb

940da3c

zanieb commented Apr 10, 2026

View reviewed changes

ci-matrix.py Outdated Show resolved Hide resolved

Apply suggestion from @zanieb

3f7f10c

zanieb commented Apr 10, 2026

View reviewed changes

Simplify default read logic

6cd1bf8

	for _platform, platform_config in ci_config.items():
	for platform_config in ci_config.values():

Uh oh!

Conversation

zanieb commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

geofft left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

zanieb commented Apr 1, 2026 •

edited

Loading