This is what I get if I add a MOZ_CRASH early in XREMain in a debug build: - on Windows x86-64: Hit MOZ_CRASH(foobar) at z:/build/build/src/toolkit/xre/nsAppRunner.cpp:3453 #01: XRE_main (z:\build\build\src\toolkit\xre\nsAppRunner.cpp:4841) #02: NS_internal_main (z:\build\build\src\browser\app\nsBrowserApp.cpp:293) #03: wmain (z:\build\build\src\toolkit\xre\nsWindowsWMain.cpp:129) #04: __scrt_common_main_seh (f:\dd\vctools\crt\vcstartup\src\startup\exe_common.inl:288) #05: BaseThreadInitThunk[C:\WINDOWS\System32\KERNEL32.DLL +0x13034] #06: RtlUserThreadStart[C:\WINDOWS\SYSTEM32\ntdll.dll +0x73691] Removing all frame skipping, I get those additional frames: #01: MozStackWalkThread[C:\Users\glandium\Desktop\foo\firefox\mozglue.dll +0x333f3] #02: MozStackWalk[C:\Users\glandium\Desktop\foo\firefox\mozglue.dll +0x33821] #03: MOZ_ReportCrash (z:\build\build\src\obj-firefox\dist\include\mozilla\Assertions.h:184) #04: XREMain::XRE_mainInit (z:\build\build\src\toolkit\xre\nsAppRunner.cpp:3453) - on Windows x86: Hit MOZ_CRASH(foobar) at z:/build/build/src/toolkit/xre/nsAppRunner.cpp:3453 #01: XREMain::XRE_main (z:\build\build\src\toolkit\xre\nsAppRunner.cpp:4729) #02: XRE_main (z:\build\build\src\toolkit\xre\nsAppRunner.cpp:4841) #03: mozilla::BootstrapImpl::XRE_main (z:\build\build\src\toolkit\xre\Bootstrap.cpp:39) #04: NS_internal_main (z:\build\build\src\browser\app\nsBrowserApp.cpp:293) #05: wmain (z:\build\build\src\toolkit\xre\nsWindowsWMain.cpp:129) #06: __scrt_common_main_seh (f:\dd\vctools\crt\vcstartup\src\startup\exe_common.inl:288) #07: BaseThreadInitThunk[C:\WINDOWS\System32\KERNEL32.DLL +0x18484] #08: RtlAreBitsSet[C:\WINDOWS\SYSTEM32\ntdll.dll +0x63ab8] #09: RtlAreBitsSet[C:\WINDOWS\SYSTEM32\ntdll.dll +0x63a88] Without any skipping: #01: MozStackWalkThread[C:\Users\glandium\Desktop\bar\firefox\mozglue.dll +0x2ab6b] #02: MozStackWalk[C:\Users\glandium\Desktop\bar\firefox\mozglue.dll +0x2af28] #03: nsTraceRefcnt::WalkTheStack (z:\build\build\src\xpcom\base\nsTraceRefcnt.cpp:773) #04: MOZ_ReportCrash (z:\build\build\src\obj-firefox\dist\include\mozilla\Assertions.h:182) #05: XREMain::XRE_mainInit (z:\build\build\src\toolkit\xre\nsAppRunner.cpp:3453) At the very least, the XREMain::XRE_mainInit frame should not be skipped. But it also seems that WalkTheStack shouldn't be assuming that its caller should be skipped, too. Also, this is all fragile in the face of inlining changing frame counts. The weird thing is that the number of skipped frame is different, although it's supposed to be the same, per: https://dxr.mozilla.org/mozilla-central/rev/c2593a3058afdfeaac5c990e18794ee8257afe99/xpcom/base/nsTraceRefcnt.cpp#839 https://dxr.mozilla.org/mozilla-central/rev/c2593a3058afdfeaac5c990e18794ee8257afe99/mozglue/misc/StackWalk.cpp#254

The XREMain::XRE_mainInit frame is also skipped over on Linux. > Also, this is all fragile in the face of inlining changing frame counts. ... or tail call optimizations or whatever... as a result, the same crash yields, on aarch64 windows, the following frames: #01: MozStackWalkThread #02: MOZ_ReportCrash #03: XREMain::XRE_mainInit #04: XREMain::XRE_main

The XREMain::XRE_mainInit frame is also skipped over on OSX.

Pushed by mh@glandium.org: https://hg.mozilla.org/integration/autoland/rev/f7b0cdc3aeb0 Make MozStackWalk/MozWalkTheStack frame skipping more reliable. r=gerald,nika,bobowen,jld

Pushed by mh@glandium.org: https://hg.mozilla.org/integration/autoland/rev/5c6b15fcea71 Make MozStackWalk/MozWalkTheStack frame skipping more reliable. r=gerald,nika,bobowen,jld

Pushed by mh@glandium.org: https://hg.mozilla.org/integration/autoland/rev/a00cdb29df1f Make MozStackWalk/MozWalkTheStack frame skipping more reliable. r=gerald,nika,bobowen,jld