600373 - Android not starting on Tracemonkey due to ENABLE_SHARED_JS=1

Status: RESOLVED FIXED

(Firefox Build System :: General, defect)

Description

[Michael Wu [:mwu]]

Attachment: Only ENABLE_SHARED_JS on Windows

Comment 1

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Comment on attachment 479205 [details] [diff] [review]
Only ENABLE_SHARED_JS on Windows

:-P

Comment 2

[Michael Wu [:mwu]]

Attachment: Only ENABLE_SHARED_JS on Windows, v2

Comment 4

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

See https://bugzilla.mozilla.org/show_bug.cgi?id=600311#c1 before we check this in.

Comment 5

[Robert Sayre]

checked this into to TM for a test.

http://hg.mozilla.org/tracemonkey/rev/49890751e255

Comment 6

[(not currently active) Ted Mielczarek]

RE: reporting the PGO crash to Microsoft. When Luke's code was triggering that crash in the optimizer, I generated a linkrepro and attached it to a bug I filed at Microsoft Connect:
http://support.microsoft.com/kb/974229

It makes the linker copy all the input files and command line options to a directory that you can zip up and attach to a bug report. We should definitely try to do that if the linker crash is 100% reproducible. We should also probably try reproducing with Visual C++ 2010, because if it's already fixed there they're likely to just resolve our bug as DONTCARE.

Comment 7

[Robert Sayre]

(In reply to comment #5)
> checked this into to TM for a test.
> 
> http://hg.mozilla.org/tracemonkey/rev/49890751e255

Doing this appears to have fixed both the Android issue and the OS 10.5.8 issue.

Comment 8

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/49890751e255

Comment 9

[Mike Shaver (:shaver emeritus)]

(In reply to comment #6)
> RE: reporting the PGO crash to Microsoft. When Luke's code was triggering that
> crash in the optimizer, I generated a linkrepro and attached it to a bug I
> filed at Microsoft Connect:
> http://support.microsoft.com/kb/974229
> 
> It makes the linker copy all the input files and command line options to a
> directory that you can zip up and attach to a bug report. We should definitely
> try to do that if the linker crash is 100% reproducible. We should also
> probably try reproducing with Visual C++ 2010, because if it's already fixed
> there they're likely to just resolve our bug as DONTCARE.

Ted: can you file another bug and do this?  Repro should be as simple as turning on static JS for windows and pushing to try, plus whatever you have to do to the build system generate the linkrepro.  Would really like to get libjs back into libxul, it's making Taras sad. :-( :-( :-(