1303212 - OS X Startup crash Hit MOZ_CRASH(CLOCK_MONOTONIC is absent!) at gecko-dev/mozglue/misc/TimeStamp_posix.cpp:181 after Xcode 8 Update

Status: RESOLVED FIXED

(Firefox Build System :: General, defect, P1)

Description

[Mason Chang [Inactive] [:mchang]]

Xcode decided to update itself today

clang --version
Apple LLVM version 8.0.0 (clang-800.0.38)
Target: x86_64-apple-darwin15.6.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

After building Firefox today, I'm getting a start up crash. This only seems to happen with Xcode 8 and El Capitan. The Xcode 8 update and the macOS Sierra GM I'm running on another machine don't have this problem.

Comment 1

[Mason Chang [Inactive] [:mchang]]

Attachment: Temp Fix

It looks like CONFIG['HAVE_CLOCK_MONOTONIC'] is true now for some reason, still trying to find out where this is defined, but this is a temporary fix.

Comment 2

[Mason Chang [Inactive] [:mchang]]

Attachment: Do a runtime check for clock_Gettime

Not sure you're the right person, but it looks like in Xcode 8, clock_gettime was added to the SDK. The current configure just tries to link it, but not run it. However, at runtime, clock_gettime fails with "No file or directory found". Doing a test clock_gettime outside of gecko, I get a correct link and compile but a runtime error of:

"dyld: lazy symbol binding failed: Symbol not found: _clock_gettime
  Referenced from: /Users/masonchang/Projects/temp/./a.out (which was built for Mac OS X 10.12)
  Expected in: /usr/lib/libSystem.B.dylib
"

On macOS sierra, everything works as expected and clock_gettime is implemented, but not on El Capitan. I suspect before Xcode 8, clock_gettime didn't even link.

This patch changes the check in our configure script to do a runtime check.

Just an FYI, I'm going on PTO for two weeks, so if this is properly reviewed, someone else can please try to land this. Thanks!

Comment 3

[Mike Hommey [:glandium]]

Comment on attachment 8792081 [details] [diff] [review]
Do a runtime check for clock_Gettime

Review of attachment 8792081 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/old-configure.in
@@ +1223,4 @@
>                 [for libs in "" -lrt; do
>                      _SAVE_LIBS="$LIBS"
>                      LIBS="$LIBS $libs"
> +                    AC_TRY_RUN([

AC_TRY_RUN needs to be avoided at all costs. It breaks cross-compilations (think Android).
Considering we're targeting versions of OSX that don't have the function, at least for now, we can simplify the problem and just not check for clock_gettime on that platform (assuming the test does return no currently)

Comment 4

[Mike de Boer [:mikedeboer]]

(In reply to Mike Hommey [:glandium] from comment #3)
> AC_TRY_RUN needs to be avoided at all costs. It breaks cross-compilations
> (think Android).
> Considering we're targeting versions of OSX that don't have the function, at
> least for now, we can simplify the problem and just not check for
> clock_gettime on that platform (assuming the test does return no currently)

Or we can branch out the AC_TRY_RUN using $OS_TARGET = Darwin and therewith save cross-compilation, or dumb idea?

Comment 5

[Andrew McCreight [:mccr8]]

(In reply to Mike Hommey [:glandium] from comment #3)
> AC_TRY_RUN needs to be avoided at all costs. It breaks cross-compilations
> (think Android).

I happened to do an all-platform push with this patch in the queue, and I can confirm that it breaks compilation for Android. :)
https://treeherder.mozilla.org/#/jobs?repo=try&revision=1c6e526473c2

Comment 6

[u408661]

Attachment: Don't check for CLOCK_MONOTONIC on darwin

Comment 7

[u408661]

The patch above is building just fine on 10.12 for me, and here's a try run to make sure it doesn't mess things up on our builders https://treeherder.mozilla.org/#/jobs?repo=try&revision=02c5273d0ae3

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/e568feccbb14
Don't check for CLOCK_MONOTONIC on darwin r=glandium

Comment 11

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/e568feccbb14

Comment 12

[Mike de Boer [:mikedeboer]]

Thanks so much, Mason, Nicholas and Mike!

Comment 13

[Gerry Chang [:gchang]]

Mark 51 as fix-optional. If it's worth uplifting to 51, feel free to nominate it.

Comment 14

[Mike de Boer [:mikedeboer]]

Well, it might be worth uplifting, but it also won't help much: currently my local Fx 51 build (tried both debug and opt) segfaults at startup. Even though it's wonderful to be able to build it, running is kind of another requirement.