Bug 1152 - Can't build webkit-gtk
Summary: Can't build webkit-gtk
Status: RESOLVED FIXED
Alias: None
Product: OE-Core
Classification: Build System, Metadata & Runtime
Component: connectivity (show other bugs)
Version: 1.0
Hardware: x86 Multiple
: Medium normal
Target Milestone: 1.1
Assignee: Edwin Zhai
QA Contact:
URL:
Whiteboard: should be fixed now after last upgrad...
Depends on:
Blocks:
 
Reported: 2011-06-09 14:38 UTC by Gary Thomas
Modified: 2011-08-11 05:02 UTC (History)
9 users (show)

See Also:
OS type for building Yocto: ---
Type of Regression: ---
Verified:
Documentation change: ---


Attachments
Build log from webkit-gtk-1.3.7+svnr72836-r72836-r1 (28.91 KB, text/plain)
2011-06-29 11:34 UTC, Gary Thomas
no flags Details
Build log from webkit-gtk-1.5.1+svnr90727-r0 (98.63 KB, application/octet-stream)
2011-08-08 01:48 UTC, Mei Lei
no flags Details

Note You need to log in before you can comment on or make changes to this bug.
Description Gary Thomas 2011-06-09 14:38:47 UTC
When configured to use GCC-4.6.0, webkit-gtk cannot be built

MACHINE=beagleboard
REVISION=7a086abc257ed433fda7a8a21c23a23199dc7cd5

Compile log attached.
Comment 1 Dexuan Cui 2011-06-09 21:54:02 UTC
(In reply to comment #0)
> When configured to use GCC-4.6.0, webkit-gtk cannot be built
Gary, do you imply with gcc-4.5 we can build fine?

> MACHINE=beagleboard
> REVISION=7a086abc257ed433fda7a8a21c23a23199dc7cd5
This is just before the perl-native change that begins from commit 5565b30.
So there must be other reason. 

> Compile log attached.
log not attached yet? :-)
Comment 2 Gary Thomas 2011-06-10 04:56:31 UTC
Log file was too large to attach. Find it at http://www.mlbassoc.com/poky/webkit-gtk-1.3.7+svnr72836-r72836.do_compile

>> MACHINE=beagleboard
>> REVISION=7a086abc257ed433fda7a8a21c23a23199dc7cd5
>This is just before the perl-native change that begins from commit 5565b30.
>So there must be other reason. 

This does not imply that revision causes the problem, rather that this
was the tip of the master branch when I tested it.

I'm checking now to make sure that it builds with GCC-4.5.1.  I know that
this package did build just a few weeks ago.
Comment 3 Dongxiao Xu 2011-06-15 22:08:21 UTC
Change the assignee to Edwin since webkit-gtk recipe is under his maintanence.

Thanks,
Dongxiao
Comment 4 Edwin Zhai 2011-06-22 22:21:46 UTC
Gary,
I can't reproduce this bug on latest master . 
MACHINE=beagleboard
REVISION=b914de55a45029658116f729ffda3abead654c90

Default gcc-4.6.0 compiles webkit-gtk successfully.

Can you double check it?
Comment 5 Gary Thomas 2011-06-29 11:30:00 UTC
Still fails for me - today's master:
OE Build Configuration:
BB_VERSION        = "1.13.2"
METADATA_BRANCH   = "local_master"
METADATA_REVISION = "a4f3e006e1f2fd93f156012af2a05adccf41d065"
TARGET_ARCH       = "arm"
TARGET_OS         = "linux-gnueabi"
MACHINE           = "beagleboard"
DISTRO            = "poky"
DISTRO_VERSION    = "1.0+snapshot-20110629"
TARGET_FPU        = "soft"
BB_NUMBER_THREADS = "4"
PARALLEL_MAKE = "-j 4"
Comment 6 Gary Thomas 2011-06-29 11:34:58 UTC
Created attachment 179 [details]
Build log from webkit-gtk-1.3.7+svnr72836-r72836-r1

Build straight out of the box with latest master from scratch:
  % bitbake web-webkit
Comment 7 Dexuan Cui 2011-06-29 17:32:26 UTC
(In reply to comment #6)
> Created attachment 179 [details]
> Build log from webkit-gtk-1.3.7+svnr72836-r72836-r1
> Build straight out of the box with latest master from scratch:
>   % bitbake web-webkit

I didn't try to reproduce the issue.
I only found some suspicious things in the building log:

cc1plus: warning: include location "/usr/include" is unsafe for cross-compilation [-Wpoison-system-directories]
...
/usr/include/bits/mathinline.h:534:1: error: impossible constraint in 'asm'

Looks ${STAGING_INCDIR}/bits/mathinline.h doesn't exist somehow so the host's header is used???
Comment 8 Gary Thomas 2011-06-30 02:17:31 UTC
Note that this failure only happens for ARM targets.  A build for qemux86 using the same repository succeeds.  

The failure does not seem to be related to GCC (on ARM) - both 2.5.1 and 2.6.0 fail with the same error.
Comment 9 Gary Thomas 2011-07-21 16:04:05 UTC
(In reply to comment #8)
> Note that this failure only happens for ARM targets.  A build for qemux86 using
> the same repository succeeds.  
> 
> The failure does not seem to be related to GCC (on ARM) - both 2.5.1 and 2.6.0
> fail with the same error.

This package has been updated and now builds on ARM with GCC 4.6.1
  webkit-gtk-1.5.1+svnr90727-r0
so I think the bug can be closed.
Comment 10 Edwin Zhai 2011-08-07 23:09:56 UTC
I have tried it on latest master (5fbac72eb) and compiled webkit-gtk successfully.
But Lei reported an error when install.

Lei,
Could you pls. give details of your failure, such as changeset/platform, to help reproducing it?
Comment 11 Mei Lei 2011-08-08 01:48:59 UTC
Created attachment 202 [details]
Build log from webkit-gtk-1.5.1+svnr90727-r0

This is webkit-gtk install failure log.
environment is below:
OE Build Configuration:
BB_VERSION        = "1.13.3"
TARGET_ARCH       = "x86_64"
TARGET_OS         = "linux"
MACHINE           = "qemux86-64"
DISTRO            = "poky"
DISTRO_VERSION    = "1.0+snapshot-20110808"
TUNE_FEATURES     = "m64"
TARGET_FPU        = ""
meta              
meta-yocto        = "master:5fbac72eb8c3cddf10876cf007aff54180c7264e"
Comment 12 Edwin Zhai 2011-08-10 00:28:13 UTC
Error comes from 
"x86_64-poky-linux-g++: error: ./.libs/libWebCore.a: Value too large for defined data type", where libWebCore.a is ~ 2.5G.
Then x86_64-poky-linux-g++ complains about the big file when link the final libwebkitgtk-1.0.so...

This bug only occurs on Lei's machine, where we cross-compile x86-64 target on i686.  On my x86-64 host, everything runs well.

Another tricky thing is: running script run.do_install from source directory can work, but bitbake webkit-gtk always fail when install. I have checked both use same command line, and share same ulimit.

Does anybody know any input file limitation for g++?
Comment 13 Edwin Zhai 2011-08-11 01:40:14 UTC
It's caused by memory limitation on i686 host. Build chromium(also linked with webkit) has same issue. Community suggestion is using x86-64 host: 

"This typically happens with 32bit linkers when building with debugging symbols, they'll exhaust their address space when linking chrome. Switching to a 64bit system or doing a release build will likely fix the issue."

Since it's a different bug, I'd like close this one and open a new "Wont' fix" bug for it.
Comment 14 Gary Thomas 2011-08-11 03:59:06 UTC
Note that these last failures (on x86) are not related to the originally reported bug which was only on ARM (see comment #8).  That problem has been fixed, probably as a side effect of code updates.
Comment 15 Edwin Zhai 2011-08-11 05:02:13 UTC
Agree with Gary.

Created new one for it: Bug 1363