Don't use iOS toolchain on macOS #12578

swolchok · 2025-07-17T00:50:39Z

I don't know why we were doing this, but it seems wrong. Let's see if tests pass.

[ghstack-poisoned]

swolchok · 2025-07-17T00:50:40Z

Stack from ghstack (oldest at bottom):

pytorch-bot · 2025-07-17T00:50:42Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12578

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 11 New Failures, 3 Unrelated Failures

As of commit 4ac87e3 with merge base 986be2a ():

NEW FAILURES - The following jobs have failed:

Apple / build-benchmark-app / macos-job (gh)
RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1
Apple / build-frameworks-ios / macos-job (gh)
RuntimeError: Command bash /Users/runner/work/_temp/exec_script failed with exit code 1
Build Linux Wheels / pytorch/executorch / build-manywheel-py3_10-cpu (gh)
undefined reference to pthread_once'`
Build Linux Wheels / pytorch/executorch / build-manywheel-py3_11-cpu (gh)
undefined reference to pthread_once'`
Build Linux Wheels / pytorch/executorch / build-manywheel-py3_12-cpu (gh)
undefined reference to pthread_once'`
Build Linux Wheels / pytorch/executorch / upload / upload-manywheel-py3_10-cpu (gh)
Unable to download artifact(s): Artifact not found for name: pytorch_executorch__3.10_cpu_x86_64
Build Linux Wheels / pytorch/executorch / upload / upload-manywheel-py3_11-cpu (gh)
Unable to download artifact(s): Artifact not found for name: pytorch_executorch__3.11_cpu_x86_64
Build Linux Wheels / pytorch/executorch / upload / upload-manywheel-py3_12-cpu (gh)
Unable to download artifact(s): Artifact not found for name: pytorch_executorch__3.12_cpu_x86_64
pull / test-eval_llama-mmlu-linux / linux-job (gh)
RuntimeError: Command docker exec -t 8796bbf99a3b33c3ef3f429bda2877bde0302f3f87fbfa341e114b24f5b67026 /exec failed with exit code 1
pull / unittest-arm-backend-with-no-fvp (test_pytest_models) / linux-job (gh)
RuntimeError: Command docker exec -t cb8367700b1b293c4628f9107e7ee751fb116cbe79b416cf1522830d43a61c53 /exec failed with exit code 128
pull / unittest-arm-backend-with-no-fvp (test_pytest_ops) / linux-job (gh)
RuntimeError: Command docker exec -t d72ca0c7a29431775e9e6ad46c26042b963786da850460e473e3bb9c5ecd1a06 /exec failed with exit code 128

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest / linux / linux-job (gh) (trunk failure)
examples/models/llama/tests/test_ring_attention.py::TestRingAttention::test_sliding_window_attention
pull / unittest-editable / linux / linux-job (gh) (trunk failure)
examples/models/llama/tests/test_ring_attention.py::TestRingAttention::test_sliding_window_attention
trunk / unittest-release / linux / linux-job (gh) (trunk failure)
examples/models/llama/tests/test_ring_attention.py::TestRingAttention::test_sliding_window_attention

This comment was automatically generated by Dr. CI and updates every 15 minutes.

I don't know why we were doing this, but it seems wrong. Let's see if tests pass. ghstack-source-id: 26d2882 ghstack-comment-id: 3081988867 Pull-Request: #12578

swolchok · 2025-07-17T00:51:32Z

might want to pick this to release, not sure. comes from #10768

jathu · 2025-07-17T00:55:35Z

I think ios.toolchain.cmake is probably poor naming, but it is a general Apple cmake toolchain to build for all of their platforms (controlled via PLATFORMS): https://github.com/leetal/ios-cmake

swolchok · 2025-07-17T00:57:09Z

I think ios.toolchain.cmake is probably poor naming, but it is a general Apple cmake toolchain to build for all of their platforms (controlled via PLATFORMS): https://github.com/leetal/ios-cmake

using this toolchain breaks sentencepiece's build in the next PR. Why aren't CMake's default settings on macOS sufficient?

jathu · 2025-07-17T01:04:03Z

@shoumikhin do you have context on why we use the toolchain instead of relying the on the default host configs?

shoumikhin · 2025-07-17T01:05:29Z

CMakePresets.json

@@ -12,7 +12,6 @@
      "inherits": ["common"],
      "generator": "Xcode",
      "cacheVariables": {
-        "CMAKE_TOOLCHAIN_FILE": "${sourceDir}/third-party/ios-cmake/ios.toolchain.cmake",


iirc, this is not iOS-only toolchain, we use it to build mac targets too.
It contains a bunch of useful declarations to simplify our live, e.g. the very MAC_ARM64 platform definition.

This is not the normal way that CMake projects expect to build for Mac though. concretely, sentencepiece's build does not like CMAKE_MACOSX_BUNDLE being set on non-iOS platforms, and it wouldn't be set on macOS if not for the use of this toolchain. IMO we should keep everything standard and defaulted unless we have a reason not to. Why do we need to use this to build for Mac?

So we rely on that toolchain platform definitions with the extra cmake flags they add to build frameworks for iOS and macOS. So far they worked well. If you believe there's a better approach, let's explore it too.
I guess we can disabled CMAKE_MACOSX_BUNDLE for sentencepiece specifically when building for mac, that's what we do with the custom script build phase in Xcode when building extension/llm third-party deps.

I guess we can disabled CMAKE_MACOSX_BUNDLE for sentencepiece specifically when building for mac, that's what we do with the custom script build phase in Xcode when building extension/llm third-party deps.

I agree that we can work around this and it's straightforward to do so. My concern is that the setting of this toolchain cost me half an hour of debugging time to track down, because it was surprising that 1) any toolchain was active for macos, since builds work fine without it when not using presets, and 2) "ios.toolchain.cmake" in particular was active. This will presumably continue to be surprising to others in the future.

What if instead we set the CMAKE_TOOLCHAIN_FILE only in build_apple_frameworks.sh? Do we need it anywhere else? (Long-term I think we should investigate using CMake's built-in support for producing Apple frameworks, but I don't see a particular reason to prioritize that.)

I don’t have a strong preference - so long as we can build the frameworks as usual, I’m happy either way.
Just a few quick notes:

Despite its name, this toolchain isn’t limited to iOS.

The last time I tried building the frameworks by manually setting all the macOS-specific flags myself - it ended up costing way more time 😅

Otherwise, this toolchain has worked well so far.

Maybe just disable that particular flag like in #12594

Hmm, maybe #12594 will unblock me. If not I'll see if I can find another way to unblock export since I've found out that this is here for a reason and not just a mistake.

Yep, I rebased and it looks like we don't need this.

swolchok · 2025-07-17T01:24:11Z

build-frameworks-ios error:

2025-07-17T01:17:59.6638640Z Merging libraries:
2025-07-17T01:17:59.6638760Z libexecutorch.a
2025-07-17T01:17:59.6638860Z libexecutorch_core.a
2025-07-17T01:17:59.6638970Z libextension_apple.a
2025-07-17T01:17:59.6639080Z libextension_data_loader.a
2025-07-17T01:17:59.6639210Z libextension_flat_tensor.a
2025-07-17T01:17:59.6639330Z libextension_module.a
2025-07-17T01:17:59.6639440Z libextension_tensor.a
2025-07-17T01:17:59.6639560Z from simulator/Release
2025-07-17T01:17:59.6639880Z into library /Users/runner/work/executorch/executorch/pytorch/executorch/cmake-out/libexecutorch_simulator.a
2025-07-17T01:17:59.6640280Z File simulator/Release/libextension_apple.a does not exist

so I guess this script might need CMAKE_TOOLCHAIN_FILE put back?

Update

4ac87e3

[ghstack-poisoned]

swolchok requested a review from larryliu0820 as a code owner July 17, 2025 00:50

swolchok requested a review from kirklandsign as a code owner July 17, 2025 00:50

meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Jul 17, 2025

swolchok added a commit that referenced this pull request Jul 17, 2025

Don't use iOS toolchain on macOS

a4a6444

I don't know why we were doing this, but it seems wrong. Let's see if tests pass. ghstack-source-id: 26d2882 ghstack-comment-id: 3081988867 Pull-Request: #12578

swolchok requested a review from jathu July 17, 2025 00:51

swolchok added the release notes: build Changes related to build, including dependency upgrades, build flags, optimizations, etc. label Jul 17, 2025

swolchok mentioned this pull request Jul 17, 2025

Use CMake EXPORT #8954

Draft

jathu added ciflow/trunk ciflow/binaries labels Jul 17, 2025

shoumikhin reviewed Jul 17, 2025

View reviewed changes

swolchok mentioned this pull request Jul 17, 2025

Don't build any binaries for 3rd party deps like sentencepiece pytorch-labs/tokenizers#105

Merged

swolchok closed this Jul 17, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Don't use iOS toolchain on macOS #12578

Don't use iOS toolchain on macOS #12578

Uh oh!

swolchok commented Jul 17, 2025

Uh oh!

swolchok commented Jul 17, 2025 •

edited

Loading

Uh oh!

pytorch-bot bot commented Jul 17, 2025 •

edited

Loading

Uh oh!

swolchok commented Jul 17, 2025 •

edited

Loading

Uh oh!

jathu commented Jul 17, 2025

Uh oh!

swolchok commented Jul 17, 2025

Uh oh!

jathu commented Jul 17, 2025

Uh oh!

shoumikhin Jul 17, 2025 •

edited

Loading

Uh oh!

swolchok Jul 17, 2025

Uh oh!

shoumikhin Jul 17, 2025

Uh oh!

swolchok Jul 17, 2025 •

edited

Loading

Uh oh!

shoumikhin Jul 17, 2025

Uh oh!

swolchok Jul 17, 2025

Uh oh!

swolchok Jul 17, 2025

Uh oh!

swolchok commented Jul 17, 2025

Uh oh!

Uh oh!

Don't use iOS toolchain on macOS #12578

Don't use iOS toolchain on macOS #12578

Uh oh!

Conversation

swolchok commented Jul 17, 2025

Uh oh!

swolchok commented Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/12578

❌ 11 New Failures, 3 Unrelated Failures

Uh oh!

swolchok commented Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jathu commented Jul 17, 2025

Uh oh!

swolchok commented Jul 17, 2025

Uh oh!

jathu commented Jul 17, 2025

Uh oh!

shoumikhin Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

swolchok Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

shoumikhin Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

swolchok Jul 17, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shoumikhin Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

swolchok Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

swolchok Jul 17, 2025

Choose a reason for hiding this comment

Uh oh!

swolchok commented Jul 17, 2025

Uh oh!

Uh oh!

swolchok commented Jul 17, 2025 •

edited

Loading

pytorch-bot bot commented Jul 17, 2025 •

edited

Loading

swolchok commented Jul 17, 2025 •

edited

Loading

shoumikhin Jul 17, 2025 •

edited

Loading

swolchok Jul 17, 2025 •

edited

Loading