Created attachment 364007 [details] useless backtrace from the webprocess the webprocess crashes on SIGTRAP when running javascript from http://httpbin.org/ This behaviour has been observed in webkitgtk in 2.23.3, 2.23.90, 2.23.91, 2.23.92, in 64bit builds. Attaching gdb to the webprocess and trying to get a backtrace has proven to be futile: One such backtrace is attached. I would appreciate any info on how to get more info from webkit on this failure
Hi, thanks for the bug report! I believe this to be fixed in the 2.24.0 release — I have just tried poking around httpbin.org for a while with a fresh x86_64 build and I am not able to get any crash. Could you please check with version 2.24.0 and confirm whether this issue is gone or whether the report is still relevant?
We're talking on IRC. You need to find some way to build with -g in order to get a backtrace. There's nothing actionable here otherwise.
Please bear with me as I'm a bit slow: I was able to build webkit-2.24.0 with -ggdb (with some difficulty with just 4GB RAM) but as I expected, I still get the same sort of useless stacktrace on the SIGTRAP with no function names. I have verified that the correct libraries (1.8G libwebkit2gtk-4.0.so) are being used. No additional information is reported by gdb on the webprocess over my previous report. As I am only seeing this on native 64bit gentoo builds, I suspect this problem may be peculiar to my setup, but I am unable to figure out what the problem may be or how to go about figuring why this particular javascript crashes and why it crashes in this way - without providing any debuggable clues. I am still looking for suggestions,
Created attachment 365505 [details] backtrace with -ggdb
Actually There is additional information in the stacktrace which I am attaching: The relevant line is #94 0x00007fe75cb078fd in JSC::operationCompareStrictEq (exec=0x7ffe4bf113b0, encodedOp1=140629277965936, encodedOp2=-281474976710656) at /var/tmp/portage/net-libs/webkit-gtk-2.24.0/work/webkit-gtk-2.24.0/Source/JavaScriptCore/jit/JITOperations.cpp:1245
"WebKitWebProces" received signal SIGTRAP - this probably means that there is WTFCrash involved, probably ASSERT Also, you are using portage to build WebKitGTK, it might be that backtrace is obscured by some custom CFLAGS which you've passed via make.conf. Could you build it stand-alone from release tarball, as described in https://trac.webkit.org/wiki/BuildingGtk ?
BTW, if you are low on ram, using lower number of parallel compiler jobs should help
Thank you, it tindeed urns out to be a case of compiler flags. The problem only occurs whencompiling with gcc-8.2.0 with -Os -O2 (-ggdb is not relevant) Apparently there is jsc code which gcc miscompiles when it is supplied with -Os -O2 and the crash smashes the stack so that there is no information in gdb. I was alerted to the problem by a similar unexpected crash in other software in a specific case, and possibly the gcc bug can be tracked down. I didn't bother to recompile without -Os on gcc-8.2.0 but am trying out clang instead
If you compile with -Os -O2 result should be the same as just -O2, because only last of -O options takes effect.
If -O2 were all that's required for breakage, we would be hearing a lot of complaints from distros.
Yes, so there should be some other flag, affecting optimizer, e.g. specific -march, or compiler itself might be miscompiled