[OE-core] [PATCH 0/8] Extensible SDK improvements

[Paul Eggleton]

On Sat, 30 Jul 2016 22:01:07 Paul Eggleton wrote:
> On Fri, 29 Jul 2016 08:26:13 Richard Purdie wrote:
> > On Mon, 2016-07-25 at 19:42 +1200, Paul Eggleton wrote:
> > > On Sat, 23 Jul 2016 13:11:39 Paul Eggleton wrote:
> > > > On Fri, 22 Jul 2016 23:31:38 Burton, Ross wrote:
> > > > > On 22 July 2016 at 13:38, Paul Eggleton <
> > > > > paul.eggleton at linux.intel.com>
> > > > > 
> > > > > wrote:
> > > > > > Some fixes and enhancements centered around how the toolchain
> > > > > > is
> > > > > > incorporated in the eSDK, as well as a couple of tweaks for
> > > > > > buildhistory
> > > > > > that I found we needed during development.
> > > > > 
> > > > > This is happening in mut, could your branch be responsible:
> > > > > 
> > > > > http://errors.yoctoproject.org/Errors/Details/73022/
> > > > > 
> > > > > Ross (who will look further when its not 23:30 on Friday!)
> > > > 
> > > > Hmm, it will be, but I can't tell immediately from the error
> > > > message what
> > > > could be wrong...
> > > 
> > > Ah, I believe this is bash vs. dash. Fix incoming, just need to test
> > > it.
> > 
> > Thanks for that fix. I think but am not 100% sure that this patchset
> > may have also caused:
> > 
> > https://autobuilder.yoctoproject.org/main/builders/nightly-oecore/build
> > s/865
> > 
> > which happens only on OE-Core testing :/.
> 
> I'm a 100% sure it's *exposed* the issue, but I can't say what would be
> causing the error itself. I'll investigate on Monday.

OK, so I have a fix for this - I was attempting to run the build system before
UNINATIVE_CHECKSUM was set. If you're using poky or something based on it then
that is already true, hence it worked there and thus the issue wasn't apparent.

However this still won't work, for two reasons:

1) The change to support gcc < 5 on the host in uninative.bbclass [1] causes
the signatures for all native targets to change. That shouldn't be a serious
problem, except...

2) As far as I can tell, locked signatures aren't working. You get
"taskhash mismatch" errors, but worse than that the tasks inexplicably still
execute instead of being restored from shared state, except the hash used for
the stamp is the locked value.

Clearly we need to examine the tests at least for #2 as well because they seem
to be either incomplete or broken.

Cheers,
Paul

[1] http://cgit.openembedded.org/openembedded-core/commit/?id=1925ead3828dcd50ef96212c2d1ea9c35bc9f13c

-- 

Paul Eggleton
Intel Open Source Technology Centre