Closed Bug 1020804 Opened 10 years ago Closed 10 years ago

Most Win64 Windows 8 Debug tests are failing -- mozjs.dll issue?

Categories

(Infrastructure & Operations Graveyard :: CIDuty, task)

x86_64
Windows 8
task
Not set
normal

Tracking

(Not tracked)

RESOLVED FIXED

People

(Reporter: away, Assigned: catlee)

References

Details

Attachments

(1 file)

Almost all tests are orange for Windows 8 Debug at https://tbpl.mozilla.org/?tree=Date&rev=cd4ed657d827. The few passing cppunittests are the standalone ones that don't have dependencies on Firefox libraries. When I try to run the server's build locally with gflags set to "Show loader snaps", I get this debugger message: LdrpProcessStaticImports - ERROR: Probing for the manifest of DLL "C:\firefox-win64debug\mozjs.dll" failed with status 0xc000007b That's the same error code (in hex) that the xpcshell tests and failing cppunittests are logging. Substituting my locally-built mozjs.dll fixes the problem. I see no obvious differences with depends.exe. Given the above, it is strange that the JIT tests are passing.
> Given the above, it is strange that the JIT tests are passing. I guess it's because the JIT tests use js.exe and not mozjs.dll.
Comparisons of dumpbin.exe /headers: Shipping mozjs.dll: 0 [ 0] RVA [size] of Resource Directory 3A7400 [ 2070] RVA [size] of Certificates Directory Date branch opt mozjs.dll: 0 [ 0] RVA [size] of Resource Directory 0 [ 0] RVA [size] of Certificates Directory Date branch debug mozjs.dll: B07400 [ 4F0] RVA [size] of Resource Directory 0 [ 0] RVA [size] of Certificates Directory My debug mozjs.dll: 0 [ 0] RVA [size] of Resource Directory 0 [ 0] RVA [size] of Certificates Directory At RVA B07400 of the Date debug mozjs.dll, I see some similar bytes as at RVA 3A7400 of the shipping mozjs.dll, except with strings like "Mozilla Fake CA". So it looks like: we are fake-signing the debug DLL, but calling it a Resource rather than a Certificate (?!)
So apparently we do this fake-signing-resource thing to all of the DLLs. Why is mozjs.dll the only one failing? The other DLLs have length 0x4E0 for that directory. Mozjs.dll is the only one with length 0x4F0. Why is it special? The only difference I know of is that mozjs lacks version information (bug 520849 / bug 634282). Could that be confusing the signing servers?
Ben or Chris, can you help David figuring out what's wrong with binary signing on automation?
Flags: needinfo?(catlee)
Flags: needinfo?(bhearsum)
The resource-versus-certificate issue may be a red herring. Some good DLLs are showing the same confusion; maybe it's just something that dumpbin isn't interpreting right. So for now let's just try to figure out why mozjs.dll has a signature of a different length from all the other DLLs.
The only difference I can see in the link lines for mozjs vs. other DLLs is the lack of a resource file in the mozjs link.
(In reply to Mike Hommey [:glandium] from comment #4) > Ben or Chris, can you help David figuring out what's wrong with binary > signing on automation? https://bugzilla.mozilla.org/show_bug.cgi?id=711210
Flags: needinfo?(bhearsum)
signing for win64 is completely disabled ATM, due to the bug ben linked to.
Flags: needinfo?(catlee)
(In reply to Chris AtLee [:catlee] from comment #8) > signing for win64 is completely disabled ATM, due to the bug ben linked to. I see that signing is disabled for opt builds, but for debug builds we are still fake-signing. Is that intentional? Can we disable it?
Flags: needinfo?(catlee)
(In reply to David Major [:dmajor] (UTC+12) from comment #9) > (In reply to Chris AtLee [:catlee] from comment #8) > > signing for win64 is completely disabled ATM, due to the bug ben linked to. > > I see that signing is disabled for opt builds, but for debug builds we are > still fake-signing. Is that intentional? Can we disable it? It is not intentional! I'll disable it shortly.
Flags: needinfo?(catlee)
Attachment #8435909 - Flags: review?(rail)
Attachment #8435909 - Flags: review?(rail) → review+
Attachment #8435909 - Flags: checkin+
Is the change supposed to take effect right away? Does it need to be merged around or something? I still see the same signing happening on the Date branch.
In production.
The debug tests are loading: https://tbpl.mozilla.org/?tree=Date&rev=f33b631a2cbd Thanks Chris!
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Assignee: nobody → catlee
Component: General → Platform Support
Flags: checkin+
Product: Core → Release Engineering
QA Contact: coop
Version: 32 Branch → unspecified
Component: Platform Support → Buildduty
Product: Release Engineering → Infrastructure & Operations
Product: Infrastructure & Operations → Infrastructure & Operations Graveyard
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: