Unblock the nightly ExecuTorch pin bump - #4728
Open
shoumikhin wants to merge 1 commit into
Open
shoumikhin wants to merge 1 commit into
shoumikhin wants to merge 1 commit into
Conversation
github-actions
Bot
added
component: tests
shoumikhin
force-pushed
the
executorch-pin-extra-fix
branch
from
September 22, 2026 07:53
0dea272 to
7df0609
Compare
The job that moves the ExecuTorch pin forward each night has never landed a bump. It fails while working out which packages to install, so the pin sits still while the rest of the project moves on. Two separate things stop it, and both are fixed here. The first is that the ExecuTorch version is written down twice. The optional install group asks for a range of versions built from the pin, and beside it asks for the companion runtime wheel, which asks for the one exact build it was compiled against. On an ordinary day the two agree. On the day the pin moves they cannot: the range asks for that day's build while every published companion still asks for the one before it, so nothing satisfies both. The companion is now the only place the version is written down. The second is that the list of platforms to solve for is wider than the list of platforms this project supports. It says any Windows, while the required list says Windows on one processor family. The wheels everything here depends on are built for that one family, so solving for any other Windows machine finds nothing to install, and that is reported as a missing wheel rather than as an unsupported platform. The two lists now agree. Nothing gets looser. The companion already turns down a mismatch from its own side, when it is imported, by loading the delegate and checking that it registers itself. Its own notes say its ExecuTorch lands on an older date on purpose. Two checks move with the first change. They used to require every published install group to name the range. They now require them to name the companion instead. Test plan: reproduced the nightly failure on Linux, moving the pin forward and resolving the way the job does, and confirmed both reported causes. With these changes the same step succeeds, records the companion together with the ExecuTorch build it asks for, and leaves the pin file holding the new version. Putting the range back makes the moved checks fail, so they are doing their job.
shoumikhin
force-pushed
the
executorch-pin-extra-fix
branch
from
September 22, 2026 08:01
7df0609 to
31951de
Compare
shoumikhin
changed the title
This branch has not been 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
2 participants
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.
The job that moves the ExecuTorch pin forward each night has never once landed a bump. It fails
while working out which packages to install, so the pin sits still while everything else moves
on. Two separate things stop it. Both are fixed here.
The version is written down twice
The optional install group asks for a range of ExecuTorch versions, built from the pin. Beside
it, the group asks for the companion runtime wheel, and that wheel asks for the one exact
ExecuTorch build it was compiled against.
On an ordinary day the two agree. On the day the pin moves they cannot. The range asks for that
day's build, while every companion published so far still asks for the one before it. Nothing
satisfies both, so the whole thing is rejected, and every failed night adds another day to the
gap.
The companion is now the only place the version is written down. Anyone setting up the project
gets an ExecuTorch and a companion that were built to work together, instead of a pairing that
has never existed.
The platform list is wider than the platforms we support
One list says any Windows. The list below it says Windows on one processor family. The wheels
everything here depends on are built for that one family, so trying to solve for any other
Windows machine finds nothing to install, and that comes out as a missing wheel rather than as
an unsupported platform. The two lists now agree.
Nothing gets looser
The companion already turns down a mismatch from its own side, when it is imported, by loading
the delegate and checking that it registers itself. Its own notes say its ExecuTorch lands on an
older date on purpose, so the pairing was never resting on the range.
Two checks move with the first change. They used to require every published install group to
name the range. They now require them to name the companion instead.
Testing
I reproduced the nightly failure on Linux, moving the pin forward and resolving the way the job
does, and confirmed both reported causes. With these changes the same step succeeds, records the
companion together with the ExecuTorch build it asks for, and leaves the pin file holding the
new version. Putting the range back makes the moved checks fail, so they are doing their job.