Closed Bug 1106174 Opened 5 years ago Closed 5 years ago

Frequent Win PGO smimever.obj : fatal error LNK1270: invalid file; timestamp does not match file linked with /LTCG:PGINSTRUMENT (or decode_core.obj)

Categories

(Firefox Build System :: General, defect)

x86
Windows XP
defect
Not set

Tracking

(firefox35 unaffected, firefox36 fixed, firefox37 fixed, firefox-esr31 unaffected)

RESOLVED FIXED
mozilla37
Tracking Status
firefox35 --- unaffected
firefox36 --- fixed
firefox37 --- fixed
firefox-esr31 --- unaffected

People

(Reporter: philor, Assigned: dmajor)

References

(Depends on 1 open bug)

Details

(Keywords: intermittent-failure)

Because the first two I saw were on the push that upgraded us to NSS 3.18, https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&revision=925290052100, and the first five I saw were in smimever.obj, and the failures did not admit to being clobbers while the green retriggers (after which we just blew off the failure) did claim to be clobbers, my first theory was that the NSS upgrade was needs-clobber.

Then post-clobber we got the same failure, but in decode_core.obj.
There was a very recent patch that landed to remove PGO workarounds for MSVC2010. There was a comment in that bug about a similar error IIRC. Don't have the bug number handy, though.
Before landing bug 1095103 I encountered several such issues locally, and they were all pretty easy for glandium to fix (and were good for build system health too).

I bet this one is easy too, I just need the time to debug it. If it causes too much pain in the meantime, feel free to back out bug 1095103.
Blocks: 1095103
I was able to reproduce the smimever.obj failure by running the incremental build from comment 0 on inbound:

> mach clobber
> hg up 9d533f7027b7
> mach build
> hg up 925290052100
> mach build

In the second build, we rebuild smimever.obj (and smime.lib), but we don't re-PGINSTRUMENT nss3.dll. We go straight to the re-PGOPTIMIZE of nss3.dll. Glandium recalls seeing a bug about NSS rebuilds; I'm going to guess that it's 959928.

The one-off issues -- decode_core, LTP_scale_ctrl_FLP, and scale_vector_FLP -- are all libopus. We may have a similar issue there as well. I haven't tried to reproduce that one yet.
Depends on: 959928
> The one-off issues -- decode_core, LTP_scale_ctrl_FLP, and scale_vector_FLP
> -- are all libopus. We may have a similar issue there as well. I haven't
> tried to reproduce that one yet.

Yikes, comment 13 was on an auto-clobber build. Didn't reproduce for me locally (no surprise).
Bug 1095103 backed out from trunk now too.
Assignee: nobody → dmajor
Status: NEW → RESOLVED
Closed: 5 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla37
Product: Core → Firefox Build System
You need to log in before you can comment on or make changes to this bug.