Spool native warnings without pipes during EP compilation (#1266) [cherry-pick to v0.3.0] - #1267
Open
KayMKM wants to merge 1 commit into
Open
Spool native warnings without pipes during EP compilation (#1266) [cherry-pick to v0.3.0]#1267KayMKM wants to merge 1 commit into
KayMKM wants to merge 1 commit into
Conversation
## Summary - replace the pipe used by native warning suppression with a file-backed temporary spool - filter and replay preserved native diagnostics only after restoring stderr - retain fail-open behavior, Windows handle restoration, warning filtering, and bounded memory usage ## Root cause The warning filter introduced in #1246 redirected native stderr to a pipe. VitisAI can hang inside `ort.InferenceSession` when its compiler sees that pipe handle. The reader was draining correctly, so this was not the full-buffer deadlock fixed by #1223; changing the reader to defer replay still hung, which isolated the pipe handle itself as the trigger. A temporary file preserves warning filtering without pipe semantics or a finite producer buffer. With an empty VAIP cache and no VitisAI-specific bypass, `facebook/convnext-tiny-224` completed on VitisAI NPU in 144.1 seconds. Disabling warning filtering entirely completed the same workload in 145.1 seconds. ## Validation - `uv run --no-sync pytest tests/unit/utils/test_native_stderr.py tests/unit/commands/test_perf_cli.py -q --basetemp temp/pytest_tmp/native-warning-file-backed-final` (143 passed, 1 platform skip) - `uvx ruff check src/winml/modelkit/utils/native_stderr.py tests/unit/utils/test_native_stderr.py src/winml/modelkit/commands/perf.py tests/unit/commands/test_perf_cli.py` - cold-cache VitisAI NPU perf with warning filtering enabled: PASS in 144.1s
DingmaomaoBJTU
approved these changes
Jul 31, 2026
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.
Cherry-picks #1266 into the v0.3.0 release branch.
This replaces pipe-backed native warning capture with a temporary-file spool, avoiding EP compiler hangs while preserving warning filtering and bounded memory use.
Original commit:
e27a0fafby @KayMKM. Cherry-picked cleanly onto release tipe1d6f874asce11819c.