1990708 - Crash in [@ mozilla::dom::InferenceSession::Init]

Status: NEW

(Core :: Machine Learning: General, defect)

Description

[BugBot [:suhaib / :marco]]

Crash report: https://crash-stats.mozilla.org/report/index/c87020b5-1058-4947-a737-c92910250920

MOZ_CRASH Reason: MOZ_CRASH(CreateSession error)

Top 10 frames of crashing thread:

0  xul.dll  mozilla::dom::InferenceSession::Init  dom/onnx/InferenceSession.cpp:385
1  xul.dll  mozilla::dom::InferenceSession::Create  dom/onnx/InferenceSession.cpp:285
2  xul.dll  mozilla::dom::InferenceSession_Binding::create  dom/bindings/ONNXBinding.cpp:2314
3  xul.dll  mozilla::dom::StaticMethodPromiseWrapper  dom/bindings/BindingUtils.cpp:3347
4  xul.dll  CallJSNative  js/src/vm/Interpreter.cpp:501
4  xul.dll  js::InternalCallOrConstruct  js/src/vm/Interpreter.cpp:597
4  xul.dll  InternalCall  js/src/vm/Interpreter.cpp:664
4  xul.dll  js::CallFromStack  js/src/vm/Interpreter.cpp:669
4  xul.dll  js::Interpret  js/src/vm/Interpreter.cpp:3287
5  xul.dll  MaybeEnterInterpreterTrampoline  js/src/vm/Interpreter.cpp:395

By querying Nightly crashes reported within the last 2 months, here are some insights about the signature:

• First crash report: 2025-09-20
• Process type: Content
• Is startup crash: No
• Has user comments: No
• Is null crash: No

Comment 1

[BugBot [:suhaib / :marco]]

The severity field is not set for this bug.
:tarek, could you have a look please?

For more information, please visit BugBot documentation.

Comment 2

[BugBot [:suhaib / :marco]]

The bug is linked to a topcrash signature, which matches the following criterion:

• Top 10 desktop browser crashes on nightly

For more information, please visit BugBot documentation.

Comment 3

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Please triage this Nightly crash.

Comment 4

[Donal Meehan [:dmeehan]]

I asked claude to look into why the volume is increasing for Nightly since May, hopefully it's helpful:

Volume grew because the audience for InferenceSession grew, not because the failure rate on any individual call spiked. Between May and now, Firefox Nightly added or expanded:

• Smart Tab Grouping (Bug 2012177 makes it use native)
• Semantic search over history (Bug 2039131 rolls it out via Nimbus, model can be swapped)
• Address autofill classification (Bug 2035232, Nightly-only)
• LLM-based telemetry (Bug 2033537)

Each of those, layered on top of the existing translations and Smart Window consumers, pushes more Nightly users through a code path that MOZ_CRASHes on any CreateSession failure.

The real bug behind the crash

Independent of what's driving volume: MOZ_CRASH("CreateSession error") is the wrong response to a runtime error. If ONNX runtime rejects a model at session-create time, the correct behavior is to reject the JS promise with an appropriate error — not crash the entire content process. Every feature that uses ML would benefit from the fallback code the JS caller already has (best-onnx retries on wasm), but that fallback can't fire because the process is dead first.

Comment 5

[BugBot [:suhaib / :marco]]

Based on the topcrash criteria, the crash signature linked to this bug is not a topcrash signature anymore.

For more information, please visit BugBot documentation.