test-domain-with-abort-on-uncaught-exception.js fails on PPC platforms

[mhdawson]

I've seen failures on this test starting yesterday, looks like it was added as part of:

#3035

and

#3036

not ok 196 test-domain-with-abort-on-uncaught-exception.js
#
#assert.js:89
#  throw new assert.AssertionError({
#  ^
#AssertionError: false == true
#    at ChildProcess.onChildExited (/jenkins-workspace/slave-j9nodelevm1/workspace/convergance-test-aix/arch/ppcle64/build/build/gcc48/gcc48/os/linux/node/test/parallel/test-domain-with-abort-on-uncaught-exception.js:168:22)
#    at emitTwo (events.js:87:13)
#    at ChildProcess.emit (events.js:172:7)
#    at Process.ChildProcess._handle.onexit (internal/child_process.js:200:12)

Have kicked off runs in community CI to confirm we see it there as well:

https://ci.nodejs.org/job/iojs+pr+ppc/20/

fyi @misterdjules

[misterdjules]

@mhdawson Can I get access to a shell on this machine?

[mhdawson]

Yes, was just checking how we'd do that. Can you send me your ssh public key and I'll add temporarily

[misterdjules]

@mhdawson Sent by email.

Also, why are ppc Jenkins agents not running https://ci.nodejs.org/job/node-test-pull-request/?If we want to avoid regressions for these platforms, we'll need some way for collaborators to check for regressions before changes get merged.

[dnakamura]

Ive looked into it at it appears that v8::base::OS::abort() is generating a SIGTRAP rather than a SIGILL or SIGABRT as expected

Core was generated by `/home/devinn/node/out/Release/node --abort_on_uncaught_exception /home/devinn/node/test/parallel/test-domain-with-abort-on-uncaught-exception.js child throwInDomainErrHandler'.
Program terminated with signal 5, Trace/breakpoint trap.
#0  0x0000000010eaf874 in v8::base::OS::Abort()
Missing separate debuginfos, use: debuginfo-install glibc-2.17-78.el7.ppc64 libgcc-4.8.3-9.el7.ppc64 libstdc++-4.8.3-9.el7.ppc64
(gdb) bt
#0  0x0000000010eaf874 in v8::base::OS::Abort()
#1  0x000000001090ded8 in v8::internal::Isolate::Throw(v8::internal::Object*, v8::internal::MessageLocation*) ()

#2  0x0000000010a694dc in v8::internal::Runtime_Throw(int, v8::internal::Object**, v8::internal::Isolate*) ()
#3  0x00000204b7120128 in ?? ()
(gdb)

[misterdjules]

@dnakamura Thank you! This is what I expected. @mhdawson @dnakamura Can you try this patch: https://gist.github.com/misterdjules/d660a9a9acb1ddd9436c and let me know if that fixes the problem on ppc/AIX?

[dnakamura]

@misterdjules It seems to fix the issue on Linux PPC, AIX still seems broken. Looking into the issue now

[misterdjules]

@dnakamura Are you on IRC in #io.js (Freenode server) by any chance? That would make realtime communication a lot easier :)

[dnakamura]

@misterdjules I can be. Also, as it turns out I was using an old build :P looks like it works on AIX too.

[misterdjules]

@dnakamura OK, then no need to chat on IRC for that issue, but it would definitely help for future ones :)

Thank you for your help!

@mhdawson What's the best way for me to run the tests suite on all ppc platforms?

[mhdawson]

You can run for a given commit with the following job: https://ci.nodejs.org/job/node-test-commit-plinux/

[kapouer]

Hello,
it's also failing on arm64, mips, mipsel (see https://buildd.debian.org/status/package.php?p=nodejs&suite=unstable). I'm trying @misterdjules patch on those now.

UPDATE: on arm64 at least the test pass with that patch.