ci: the standards build builds the workspaces a change touches #251

Merged

dcharlot merged 2 commits from ci/build-what-changes into main 2026-10-14 11:10:50 -04:00

Conversation 0 Commits 2 Files changed 3 +319 -14

dcharlot commented 2026-10-14 03:40:27 -04:00

Owner

Copy link

The standards build now builds the workspaces a pull request touches, instead of all 17 for every pull request.

Why

standards-build.yml built and tested every Rust workspace of the tree on every pull request: 9 legs on the one x86_64 runner, 7 on linux-arm64, one on macos, plus the macOS energy job. A change to one standard rebuilt all the others, though none of their inputs had changed. When measured, the x86_64 runner had 90 jobs waiting, most of them these builds.

What changes

• A plan job (on linux-arm64, seconds) runs scripts/standards-touched.py on the pull request's own diff (base...head). It selects each workspace whose inputs the change touches:
  • its own area, the first component of its directory (all of jcp/ for jcp/jcp-rs, with its conformance data and documents);
  • the areas of every crate it depends on by path, followed transitively;
  • any file outside those areas that its Rust sources name with a relative path;
  • anything every leg shares: this workflow, the planning script, the toolchain pin, the cargo configuration. A change to any of these builds everything.
• The build matrix is the entries the plan selects (fromJSON(needs.plan.outputs.include)). The workspace list moves into the plan job, one - { ... } line each, unchanged, with its comments.
• The macOS energy job runs when the change touches a workspace whose energy backend it tests.
• A push to main builds everything, as before. So does anything the plan cannot determine: an unreadable diff or an unknown commit. A wrong "skip" would silently drop a test, while a wrong "build" only costs runner time, so the plan guesses toward building.
• The plan reads the pull request's diff, not the push delta, so merging main into a branch does not rebuild everything.

Evidence

• Last forty merges into main. About half were WAI-only changes, and those would have built none of the 17. The others built between one and ten; a change under jcp/ builds eight, because jcp is a path dependency of most workspaces.
• The forge builds a matrix from a previous job's output. A probe in ci-test showed it creates the legs a plan job lists, and no job at all for a job whose if: is false.
• scripts/standards-touched-selftest.sh runs in the plan job. It builds a scratch repository and checks each case: a file in no area builds nothing; a path-dependency's area builds its dependents; a file a workspace's source includes builds that workspace; the workflow or the script builds everything; a push builds everything; an uncomputable diff builds everything; emitted entries keep their booleans (z3: true) and quoted values.
• This pull request touches the workflow, so its own run builds all 17 through the new plan. That exercises the full path end to end.
• The concurrency check passes.

🤖 Generated with Claude Code

The standards build now builds the workspaces a pull request touches, instead of all 17 for every pull request. ## Why `standards-build.yml` built and tested every Rust workspace of the tree on every pull request: 9 legs on the one x86_64 runner, 7 on `linux-arm64`, one on `macos`, plus the macOS energy job. A change to one standard rebuilt all the others, though none of their inputs had changed. When measured, the x86_64 runner had 90 jobs waiting, most of them these builds. ## What changes - **A `plan` job** (on `linux-arm64`, seconds) runs `scripts/standards-touched.py` on the pull request's own diff (`base...head`). It selects each workspace whose inputs the change touches: - its own area, the first component of its directory (all of `jcp/` for `jcp/jcp-rs`, with its conformance data and documents); - the areas of every crate it depends on by path, followed transitively; - any file outside those areas that its Rust sources name with a relative path; - anything every leg shares: this workflow, the planning script, the toolchain pin, the cargo configuration. A change to any of these builds everything. - **The build matrix** is the entries the plan selects (`fromJSON(needs.plan.outputs.include)`). The workspace list moves into the plan job, one `- { ... }` line each, unchanged, with its comments. - **The macOS energy job** runs when the change touches a workspace whose energy backend it tests. - **A push to main builds everything**, as before. So does anything the plan cannot determine: an unreadable diff or an unknown commit. A wrong "skip" would silently drop a test, while a wrong "build" only costs runner time, so the plan guesses toward building. - The plan reads the pull request's diff, not the push delta, so merging main into a branch does not rebuild everything. ## Evidence - **Last forty merges into main.** About half were WAI-only changes, and those would have built none of the 17. The others built between one and ten; a change under `jcp/` builds eight, because jcp is a path dependency of most workspaces. - **The forge builds a matrix from a previous job's output.** A probe in `ci-test` showed it creates the legs a plan job lists, and no job at all for a job whose `if:` is false. - **`scripts/standards-touched-selftest.sh` runs in the plan job.** It builds a scratch repository and checks each case: a file in no area builds nothing; a path-dependency's area builds its dependents; a file a workspace's source includes builds that workspace; the workflow or the script builds everything; a push builds everything; an uncomputable diff builds everything; emitted entries keep their booleans (`z3: true`) and quoted values. - **This pull request touches the workflow**, so its own run builds all 17 through the new plan. That exercises the full path end to end. - The concurrency check passes. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dcharlot added 1 commit 2026-10-14 03:40:27 -04:00

ci: the standards build builds the workspaces a change touches ... 9edc1ad2aa

Every pull request built and tested all 17 standards workspaces, whatever it
changed, and the one x86_64 runner's queue filled with builds no change could
affect (90 jobs waiting when measured). A plan job on linux-arm64 now reads the
pull request's own diff and selects the workspaces it touches: a workspace's
own area, the areas of everything it depends on by path, any file outside
them its sources name, and everything every leg shares (this workflow, the
planning script, the toolchain pin, the cargo configuration). The build matrix
is the entries it selects; the macOS energy job runs when an energy-metering
workspace is touched. A push to main builds everything, and so does anything
the plan cannot determine.

Over the last forty merges into main, the WAI-only changes, about half,
would have built none of the 17; the rest built between one and ten. A probe
in ci-test confirmed this forge builds a matrix from a previous job's output
and creates no job for an empty list. A self-test runs in the plan job.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot added 1 commit 2026-10-14 04:39:24 -04:00

Merge main into ci/build-what-changes ... 9f7516c539

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot merged commit 9b3275957d into main 2026-10-14 11:10:50 -04:00

dcharlot referenced this pull request from a commit 2026-10-14 11:10:51 -04:00

Merge pull request 'ci: the standards build builds the workspaces a change touches' (#251) from ci/build-what-changes into main

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Transaction-Science/open-standards!251

No description provided.