Read non-finite float values back from an ETDump

[Dev-next-gen]

Summary

I tried to round-trip an ETDump whose debug event holds a -inf double, and deserialize_from_etdump_flatcc failed with json.decoder.JSONDecodeError: Expecting value. The same happens for inf, nan, and for a Float value. A finite value round-trips fine.

The function turns the binary into JSON with flatc --json --strict-json and parses that with json.loads. flatc writes non-finite values as bare tokens: "double_val": -inf, inf, nan, and -nan when the sign bit is set (which is what 0.0 / 0.0 gives on x86). json.loads only understands Infinity, -Infinity and NaN, so a single non-finite value anywhere in the dump makes the whole file unreadable. The runtime logs double and float EValues into debug events as they are, so an attention mask filled with -inf or a NaN intermediate is enough for the Inspector to be unable to load the ETDump, which is the moment you most want to look at it.

The serialize direction already works since #22152, so this only touches the read side: before parsing, the flatc tokens are rewritten into the spellings json.loads accepts. String literals are matched first and copied through, so an event or run named "inf" or "nan" is left untouched. Nothing else calls _deserialize_from_json_to_etdump_flatcc, and JSON without those tokens is unchanged.

Test plan

I added test_serialize_non_finite_floats to devtools/etdump/tests/serialize_test.py. It puts inf in the Float and -inf in the Double of the sample dump and checks the round trip is equal, then checks a nan double comes back as NaN.

python -m unittest executorch.devtools.etdump.tests.serialize_test -v

Without the fix the new test errors with JSONDecodeError: Expecting value: line 81 column 30; with it both tests pass (flatc 24.3.25 built from the pinned third-party/flatbuffers commit, Python 3.12, Linux). Separately I patched a negative NaN (0xFFF8000000000000) into a serialized dump, checked flatc prints it as -nan, and confirmed it now reads back as NaN, and that event names "inf" and "nan" survive the rewrite. I did not run lintrunner locally.

This PR was written with the help of Claude Code.

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22785

• 📄 Preview Python docs built from this PR

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

⚠️ 14 Awaiting Approval

As of commit f35c68b with merge base 903696a ():

AWAITING APPROVAL - The following workflows need approval before CI can run:

• Build documentation (gh)
• Build Presets (gh)
• Lint (gh)
• periodic (gh)
• pull (gh)
• Test ARM Backend (gh)
• Test CoreML Backend (gh)
• Test Cortex-M Backend (gh)
• Test NXP Backend (gh)
• Test QNN Backend (gh)
• Test QNN Windows MSVC build (gh)
• Test Vulkan Backend (gh)
• Test WebGPU Backend (gh)
• Test XNNPACK Backend (gh)

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

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.