I am working on tracking down and fixing ubsan errors reported in Mozilla's automated tests. When I use -fsanitize=enum, one of the errors I get is: > [task 2017-09-25T00:47:57.792Z] 00:47:57 INFO - TEST-START | extensions/universalchardet/tests/test_bug426271-utf-8.html > [task 2017-09-25T00:47:57.902Z] 00:47:57 INFO - GECKO(4261) | /builds/worker/workspace/build/src/extensions/universalchardet/src/base/nsCodingStateMachine.h:49:12: runtime error: load of value 5, which is not a valid value for type 'nsSMState' > [task 2017-09-25T00:47:57.903Z] 00:47:57 INFO - GECKO(4261) | /builds/worker/workspace/build/src/extensions/universalchardet/src/base/nsCodingStateMachine.h:39:9: runtime error: load of value 5, which is not a valid value for type 'nsSMState' > [task 2017-09-25T00:47:57.904Z] 00:47:57 INFO - GECKO(4261) | /builds/worker/workspace/build/src/extensions/universalchardet/src/base/nsCodingStateMachine.h:46:30: runtime error: load of value 5, which is not a valid value for type 'nsSMState' > [task 2017-09-25T00:47:57.906Z] 00:47:57 INFO - GECKO(4261) | /builds/worker/workspace/build/src/extensions/universalchardet/src/base/nsCodingStateMachine.h:46:30: runtime error: load of value 5, which is not a valid value for type 'nsSMState' the same error was reported here: https://bugs.freedesktop.org/show_bug.cgi?id=101032 where they replaced the nsSMState by an unsigned int. Here's the patch: https://cgit.freedesktop.org/uchardet/uchardet/commit/?id=53f7ad0e0b0894f11c948ee75f74595ab1b61405 Should we do the same thing?

Yes, nsSMState shouldn't use enum. It uses non-enum value too.

Thanks. Here's a version of the patch for mozilla-central.

Thank you for the review. Try server results: https://treeherder.mozilla.org/#/jobs?repo=try&revision=be58c7cf95e9

Pushed by ryanvm@gmail.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/2e390724da6f Fix ubsan runtime error for nsSMState enum. r=m_kato