[OE-core] Reproducible build problem with BB_NUMBER_THREADS=8

[Eric Bénard]

Hi,

I often meet a problem related to C++ headers which are not found (despite 
existing)n often when bitbake starts compiling mysql & directfb.
I tested on 2 build hosts (both with Core i7, one i686 ubuntu, one x64_64 
Fedora) and manage to reproduce the problem easily with BB_NUMBER_THREADS=8 in 
both cases.

The configuration is quite simple :
angstrom setup script + beagleboard machine + bitbake qt4e-demo-image
with
PARALLEL_MAKE     = "-j6"
BB_NUMBER_THREADS = "8"

Please find attached the 2 logs :
- log1.txt (with one custom overlay containing a custom image) :
mysql fails with
| ../include/my_global.h:1516:15: fatal error: new: No such file or directory
the file exists :
build/tmp-angstrom_2010_x-eglibc/sysroots/beagleboard/usr/include/c++/new
it seems that include/c++ is not in the header search path
reducing BB_NUMBER_THREADS to 2 leads to the same kind of problem in libcgroup :
| libcg_ba.cpp:18:18: fatal error: string: No such file or directory
the file exists :
build/tmp-angstrom_2010_x-eglibc/sysroots/beagleboard/usr/include/c++/debug/string
=> the only solution was to remove tmp and start again with a reduced 
BB_NUMBER_THREADS

- log2.txt (plain oe-core + angstrom setup to remove the doubt of my overlay 
being the source of the problem) :
directfb fails with
mkdgifft.cpp:64:16: fatal error: list: No such file or directory
the file exists :
build/tmp-angstrom_2010_x-eglibc/sysroots/beagleboard/usr/include/c++/list
here again, it seems that include/c++ is not in the header search path
=> I reduced BB_NUMBER_THREADS to 1 and launched again bitbake (build actually 
in progress).

The workaround is to reduce BB_NUMBER_THREADS to <=4 which seems to never 
bring the problem (at least for a dozen of builds).

Does anyone meet the same problem ?
Any idea of what could be wrong here ?

Thanks,
Eric
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: log1.txt
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20111115/615ceeb8/attachment-0004.txt>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: log2.txt
URL: <http://lists.openembedded.org/pipermail/openembedded-core/attachments/20111115/615ceeb8/attachment-0005.txt>