I'm working on fixing ubsan enum errors in the codebase. In running the full Linux test suite with -fsanitize=enum, I observe the following errors: > [task 2017-09-25T00:46:49.602Z] 00:46:49 INFO - TEST-START | GeckoProfiler.FeaturesAndParams > [task 2017-09-25T00:46:49.603Z] 00:46:49 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1399:38: runtime error: load of value 4294967195, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:46:49.603Z] 00:46:49 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1404:39: runtime error: load of value 4294967195, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:46:49.603Z] 00:46:49 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1411:40: runtime error: load of value 4294967195, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:46:49.604Z] 00:46:49 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1612:52: runtime error: load of value 4294967195, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:46:53.619Z] 00:46:53 INFO - TEST-PASS | GeckoProfiler.FeaturesAndParams | test completed (time: 4047ms) and > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - TEST-START | LulDwarfEHFrame.SimpleFDE > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1381:38: runtime error: load of value 4294967226, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1497:35: runtime error: load of value 4294967226, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1504:36: runtime error: load of value 4294967226, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - /builds/worker/workspace/build/src/tools/profiler/lul/LulDwarf.cpp:1620:58: runtime error: load of value 4294967226, which is not a valid value for type 'DwarfPointerEncoding' > [task 2017-09-25T00:47:00.109Z] 00:47:00 INFO - TEST-PASS | LulDwarfEHFrame.SimpleFDE | test completed (time: 0ms) So I looked at the code near tools/profiler/lul/LulDwarf.cpp:1399 and here it is: > case DW_Z_has_personality_routine: > // The CIE's augmentation data holds the personality routine > // pointer's encoding, followed by the pointer itself. > cie->has_z_personality = true; > // Fetch the personality routine pointer's encoding from the > // augmentation data. > if (data >= data_end) return ReportIncomplete(cie); > cie->personality_encoding = DwarfPointerEncoding(*data++); > if (!reader_->ValidEncoding(cie->personality_encoding)) { > reporter_->InvalidPointerEncoding(cie->offset, > cie->personality_encoding); > return false; > } The problem is, `data` is defined as a (const signed char *). In the first four errors, `*data` is equal to 0xFF (DW_EH_PE_omit), and it is cast to a DwarfPointerEncoding, so we get int(-1) instead of 0xFF. In the last four errors, for some reason `*data` is equal to 0xBA. That doesn't correspond to any allowed enum value, and I haven't tracked down why yet. But probably we should be checking for a correct value *before* casting to a DwarfPointerEncoding type. In any case, I noticed there is a newer version of the same code in the crashreporter: https://dxr.mozilla.org/mozilla-central/rev/76a26ef7c493311c170ae83eb0c1d6592a21396d/toolkit/crashreporter/google-breakpad/src/common/dwarf/dwarf2reader.cc#2348 and in that version, `data` is a (const uint8_t*) instead. So that would fix the first four runtime errors ast least. So I can think of three options: 1. Refactor to use the crashreporter copy of dwarf2reader.cc in crashreporter 2. Update LulDwarf.cpp to the latest breakpad code 3. Apply patch https://chromium.googlesource.com/breakpad/breakpad/+/bc44efdc274aaf5b3b575d66f7e245754c0fa1e1%5E%21/ which switched from unsigned char to uint8_t.

Hi Nicholas -- any opinions as to how to proceed? Thanks.

I will defer to jseward, who is the expert on LUL.