Use ICF in macOS release builds to reduce binary sizes by zanieb · Pull Request #19615 · astral-sh/uv

zanieb · 2026-05-29T21:30:32Z

Use --icf=safe for optimized macOS release artifacts, via Rust's bundled Mach-O LLD.

Measurements showed useful size wins on macOS; Linux did not show a meaningful artifact-size improvement and are excluded here. See #19615 (comment)

astral-sh-bot · 2026-05-29T21:56:13Z

uv test inventory changes

This PR no longer changes the tests when compared with the latest main baseline.

zanieb · 2026-05-30T05:17:26Z

Hm...

zsol · 2026-05-30T08:21:23Z

Hmm, how does that change if we do more aggressive ICF? FWIW I'm not entirely sure this amount of wins is worth the (slight) risk of confusing crash reports/coredumps. OTOH this decision is easy to reverse

zanieb · 2026-05-30T14:40:50Z

I can't tell if this is actually working as intended — I'm going to poke at it more.

This reverts commit 05e88e7.

zaniebot · 2026-06-08T17:20:00Z

Measurement data

Moved out of the PR body to keep the summary short.

Observed timings

These are GitHub Actions elapsed job times, not isolated linker benchmarks. The no-ICF cohort uses retained ordinary release builds (1, 2, 3); the ICF cohort uses this PR's first run and rerun. Treat the deltas as directional CI observations: the aarch64 baseline range in particular contains a noisy outlier.

Target	No ICF avg (`n=3`)	No ICF range	ICF avg (`n=2`)	ICF range	Delta
`x86_64-apple-darwin`	`7:33`	`7:22-7:49`	`7:18`	`7:16-7:20`	`-3.3%`
`aarch64-apple-darwin`	`8:23`	`7:17-10:30`	`7:05`	`6:56-7:13`	`-15.5%`
`x86_64-unknown-linux-gnu`	`12:03`	`10:59-12:41`	`11:36`	`10:36-12:35`	`-3.7%`

A same-source no-ICF control run toggled only the ICF flags off. Its single elapsed-time sample is slower than the retained cohort, which reinforces that the job-level timings are too noisy to isolate a small linker change:

Target	Same-source no ICF control (`n=1`)	ICF avg (`n=2`)	ICF range
`x86_64-apple-darwin`	`10:20`	`7:18`	`7:16-7:20`
`aarch64-apple-darwin`	`9:38`	`7:05`	`6:56-7:13`
`x86_64-unknown-linux-gnu`	`12:36`	`11:36`	`10:36-12:35`

Artifact sizes

This compares wheel artifacts from the same-source no-ICF control run with this PR's ICF rerun. The control commit changes only the workflow ICF toggle.

Wheel artifact	No ICF bytes	ICF bytes	Delta
`wheels_uv-macos-x86_64`	`23278539`	`22839695`	`-1.89%`
`wheels_uv-aarch64-apple-darwin`	`21760071`	`21518309`	`-1.11%`
`wheels_uv-x86_64-unknown-linux-gnu`	`24857925`	`24854207`	`-0.01%`
`wheels_uv_build-macos-x86_64`	`1473768`	`1467721`	`-0.41%`
`wheels_uv_build-aarch64-apple-darwin`	`1378633`	`1366300`	`-0.89%`
`wheels_uv_build-x86_64-unknown-linux-gnu`	`1608666`	`1608905`	`+0.01%`

The corresponding embedded executable deltas are -2.02% for macOS x86_64 uv, -1.41% for macOS aarch64 uv, -0.67% for macOS x86_64 uv-build, -0.66% for macOS aarch64 uv-build, -0.01% for Linux x86_64 GNU uv, and approximately 0.00% for Linux x86_64 GNU uv-build.

codspeed-hq · 2026-06-08T17:46:53Z

Merging this PR will degrade performance by 5.85%

❌ 1 regressed benchmark
✅ 12 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
❌	WallTime	`unpack_sdist_many_files`	1 s	1.1 s	-5.85%

Tip

Investigate this regression with the CodSpeed MCP and your agent.

_{Comparing zb/icf (c91cff9) with main (89f28ab)}

zsol

Seems a bit counterintuitive that runs with ICF are consistently faster. What should we keep an eye out for in terms of potential negative consequences? Nonsense stack traces in panics?

Released on 2026-06-10. ### Enhancements - Add `--emit-index-url` and `--emit-find-links` to `uv export` ([#18370](#18370)) - Add `--find-links` support for `uv pip list` ([#16103](#16103)) - Group executable install errors during `uv python install` ([#19691](#19691)) - Use ICF in macOS release builds to reduce binary sizes ([#19615](#19615)) ### Preview features - Add initial hidden `uv upgrade` command ([#19678](#19678)) - Reject Git revisions in `uv upgrade` ([#19742](#19742)) ### Configuration - Recognize `UV_NO_INSTALL_PROJECT`, `UV_NO_INSTALL_WORKSPACE`, `UV_NO_INSTALL_LOCAL` ([#19323](#19323)) ### Performance - Speed up discovery of large workspaces ([#18311](#18311)) ### Bug fixes - Allow unknown preview flags with a warning again ([#19669](#19669)) - Apply dependency exclusions to direct requirements ([#19699](#19699)) - Avoid following external symlinks during cache clean ([#19682](#19682)) - Avoid following symlinks during cache prune ([#19543](#19543)) - Fix Git cache keys for worktrees and packed refs ([#19706](#19706)) - Make resolver error handling iterative to avoid stack overflows ([#19695](#19695)) - Pass `VIRTUAL_ENV` through `cygpath` inside `fish` on Windows ([#19703](#19703)) - Rebuild explicit local directory tool installs ([#19591](#19591)) - Validate egg top-level entries as identifiers ([#19679](#19679)) ### Documentation - Document `--find-links` caching behavior ([#19585](#19585)) - Add a small section for malware checks ([#19680](#19680))

This PR contains the following updates: | Package | Type | Update | Change | |---|---|---|---| | [ghcr.io/astral-sh/uv](https://github.com/astral-sh/uv) | stage | patch | `0.11.19` → `0.11.20` | --- ### Release Notes <details> <summary>astral-sh/uv (ghcr.io/astral-sh/uv)</summary> ### [`v0.11.20`](https://github.com/astral-sh/uv/blob/HEAD/CHANGELOG.md#01120) Released on 2026-06-10. ##### Enhancements - Add `--emit-index-url` and `--emit-find-links` to `uv export` ([#18370](astral-sh/uv#18370)) - Add `--find-links` support for `uv pip list` ([#16103](astral-sh/uv#16103)) - Group executable install errors during `uv python install` ([#19691](astral-sh/uv#19691)) - Use ICF in macOS release builds to reduce binary sizes ([#19615](astral-sh/uv#19615)) ##### Preview features - Add initial hidden `uv upgrade` command ([#19678](astral-sh/uv#19678)) - Reject Git revisions in `uv upgrade` ([#19742](astral-sh/uv#19742)) ##### Configuration - Recognize `UV_NO_INSTALL_PROJECT`, `UV_NO_INSTALL_WORKSPACE`, `UV_NO_INSTALL_LOCAL` ([#19323](astral-sh/uv#19323)) ##### Performance - Speed up discovery of large workspaces ([#18311](astral-sh/uv#18311)) ##### Bug fixes - Allow unknown preview flags with a warning again ([#19669](astral-sh/uv#19669)) - Apply dependency exclusions to direct requirements ([#19699](astral-sh/uv#19699)) - Avoid following external symlinks during cache clean ([#19682](astral-sh/uv#19682)) - Avoid following symlinks during cache prune ([#19543](astral-sh/uv#19543)) - Fix Git cache keys for worktrees and packed refs ([#19706](astral-sh/uv#19706)) - Make resolver error handling iterative to avoid stack overflows ([#19695](astral-sh/uv#19695)) - Pass `VIRTUAL_ENV` through `cygpath` inside `fish` on Windows ([#19703](astral-sh/uv#19703)) - Rebuild explicit local directory tool installs ([#19591](astral-sh/uv#19591)) - Validate egg top-level entries as identifiers ([#19679](astral-sh/uv#19679)) ##### Documentation - Document `--find-links` caching behavior ([#19585](astral-sh/uv#19585)) - Add a small section for malware checks ([#19680](astral-sh/uv#19680)) </details> --- ### Configuration 📅 **Schedule**: (UTC) - Branch creation - At any time (no schedule defined) - Automerge - At any time (no schedule defined) 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about this update again. --- - [ ] If you want to rebase/retry this PR, check this box --- This PR has been generated by [Mend Renovate](https://github.com/renovatebot/renovate).  Co-authored-by: Renovate Bot <renovate@bhamm-lab.com> Reviewed-on: https://codeberg.org/blake-hamm/bhamm-lab/pulls/186

zanieb mentioned this pull request May 29, 2026

Enable safe ICF for macOS arm64 release builds #19614

Closed

zanieb commented May 29, 2026

View reviewed changes

Comment thread .github/workflows/build-release-binaries.yml Outdated

zanieb commented May 29, 2026

View reviewed changes

Comment thread .github/workflows/build-release-binaries.yml Outdated

zanieb commented May 29, 2026

View reviewed changes

Comment thread .github/workflows/build-release-binaries.yml Outdated

zanieb mentioned this pull request May 29, 2026

Actually use mold for linking in CI #19613

Closed

zanieb force-pushed the zb/icf branch 12 times, most recently from 63aa9b2 to 0984566 Compare June 1, 2026 20:45

zanieb added 4 commits June 1, 2026 16:25

Use ICF in release builds

7f8abba

Move Linux ICF flags into release matrix

c206ad3

Drop Linux ICF without size improvement

808b2e5

Use bundled LLD for Linux ICF

26a7b49

zanieb force-pushed the zb/icf branch from 0984566 to 26a7b49 Compare June 1, 2026 21:26

zanieb changed the base branch from zb/use-mold-in-release-builds to main June 1, 2026 21:26

zanieb added 3 commits June 1, 2026 16:33

Scope Linux ICF flags to release target

b584604

Temporarily disable ICF for baseline measurement

05e88e7

Revert "Temporarily disable ICF for baseline measurement"

0d38a8e

This reverts commit 05e88e7.

zanieb added 2 commits June 8, 2026 12:20

Scope ICF to macOS release builds

b568ed9

Clarify macOS ICF comments

f22e8e1

zanieb changed the title ~~Use ICF in release builds~~ Use ICF in macOS release builds to reduce binary sizes Jun 8, 2026

zanieb added the releases Related to building and distributing release artifacts of uv label Jun 8, 2026

zanieb marked this pull request as ready for review June 8, 2026 17:24

Refine macOS ICF size comments

c91cff9

zsol self-requested a review June 8, 2026 17:36

charliermarsh approved these changes Jun 8, 2026

View reviewed changes

zsol approved these changes Jun 9, 2026

View reviewed changes

zanieb merged commit 3b8e3ca into main Jun 9, 2026
263 of 274 checks passed

zanieb deleted the zb/icf branch June 9, 2026 14:02

This was referenced Jun 9, 2026

Use ICF for macOS release builds astral-sh/ruff#25780

Draft

Use ICF for macOS release builds astral-sh/ty#3709

Draft

konstin added the enhancement New feature or improvement to existing functionality label Jun 10, 2026

konstin mentioned this pull request Jun 10, 2026

Bump version to 0.11.20 #19765

Merged

BrewTestBot mentioned this pull request Jun 10, 2026

uv 0.11.20 Homebrew/homebrew-core#287331

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use ICF in macOS release builds to reduce binary sizes#19615

Use ICF in macOS release builds to reduce binary sizes#19615
zanieb merged 10 commits into
mainfrom
zb/icf

zanieb commented May 29, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

astral-sh-bot Bot commented May 29, 2026 •

edited

Loading

Uh oh!

zanieb commented May 30, 2026

Uh oh!

zsol commented May 30, 2026

Uh oh!

zanieb commented May 30, 2026

Uh oh!

zaniebot commented Jun 8, 2026

Uh oh!

codspeed-hq Bot commented Jun 8, 2026

Uh oh!

zsol left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

zanieb commented May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

astral-sh-bot Bot commented May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

uv test inventory changes

Uh oh!

zanieb commented May 30, 2026

Uh oh!

zsol commented May 30, 2026

Uh oh!

zanieb commented May 30, 2026

Uh oh!

zaniebot commented Jun 8, 2026

Measurement data

Observed timings

Artifact sizes

Uh oh!

codspeed-hq Bot commented Jun 8, 2026

Merging this PR will degrade performance by 5.85%

Performance Changes

Uh oh!

zsol left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

zanieb commented May 29, 2026 •

edited

Loading

astral-sh-bot Bot commented May 29, 2026 •

edited

Loading