Re: work-shared not being correctly shared

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Mon, 2012-10-01 at 12:50 +0100, Phil Blundell wrote:
> On Mon, 2012-09-24 at 21:28 +0200, Martin Jansa wrote:
> > On Mon, Sep 24, 2012 at 05:22:57PM +0100, Phil Blundell wrote:
> > > On Mon, 2012-09-24 at 08:42 -0700, Saul Wold wrote:
> > > > Yes, I am seeing something similar with do_patch happening twice on the 
> > > > work-shared area, I talked with Richard a bit ago about it, and I am 
> > > > digging into the siginfo files, you can use bitbake-sigdiff to see what 
> > > > sigs are changed causing it unpack twice (or in my case patch twice).
> > > 
> > > Ah, right, it seems to be rm_old_work that's causing the problem.  I'll
> > > disable that for now and figure out how to fix it later.
> > 
> > That was already discussed few weeks ago, you should be able to find it
> > in archive.
> 
> I found this thread:
> 
> http://lists.linuxtogo.org/pipermail/openembedded-core/2012-July/026185.html
> 
> but there didn't appear to be any clear conclusion about what the right
> fix for the problem was.  Does anybody have any further insight into
> that?

The issue is that rm_old_work references PN in its task definition.
gcc's shared work directory assumes it can get the same sstate checksums
for fetch/unpack/patch which is broken by:

addtask rm_old_works before do_unpack

The trick would seem to be to change do_unpack for do_configure which is
how others worked around it.

Not a perfect fix but equally not annoying enough to lose lots of sleep
over.

Cheers,

Richard