[BUILDBOT] build breaker of linux32 and win7

[Andre Fischer <af...@a-w-f.de>]

The linux32 and win7, but not linux64, did break when trying to apply 
patches in the python module.  The reason for this seems to be that the 
patches are applied to already patched files.  This is a bit strange 
because there are marker files in $INPATH/misc/build that usually 
prevent this from happening.  But there seem to be dependencies of these 
marker files on the patch files: update a patch file and the marker file 
is deleted and the patch is applied again.
(a lot of "seems", have to look at this more closely to be sure)

To make a long story short.  Removing the output trees in the python 
module before the next build should do the trick.
@Andrew: could you that?

Thanks,
Andre

Re: [BUILDBOT] build breaker of linux32 and win7

[Andre Fischer <af...@a-w-f.de>]

More technical details:

There really is a dependency of the unpack flag file on the patch files.
So, why does it not work?  Because the rule 
$(PACKAGE_DIR)/$(UNTAR_FLAG_FILE) in tg_ext.mk uses the wrong test to 
detect the existence of the unpacked python source code: -f instead of -d
After fixing this (SVN revision 1348771) the directory is removed. 
Therefore unpacking it again succeeds and the files can be patched 
without errors.

The python modules on the buildbots still have to be cleaned (even 
though unpacking fails (silently), its flag file is still written and 
therefore will not be done again.)

-Andre

Re: [BUILDBOT] build breaker of linux32 and win7

[Andrew Rist <an...@oracle.com>]

On 6/11/2012 1:40 AM, Andre Fischer wrote:
> The linux32 and win7, but not linux64, did break when trying to apply 
> patches in the python module.  The reason for this seems to be that 
> the patches are applied to already patched files.  This is a bit 
> strange because there are marker files in $INPATH/misc/build that 
> usually prevent this from happening.  But there seem to be 
> dependencies of these marker files on the patch files: update a patch 
> file and the marker file is deleted and the patch is applied again.
> (a lot of "seems", have to look at this more closely to be sure)
>
> To make a long story short.  Removing the output trees in the python 
> module before the next build should do the trick.
> @Andrew: could you that?
sure - will do
A.
>
> Thanks,
> Andre