[OE-core] Wrong sstate signatures for cross toolchain Was: [yocto] Building gdb-7.10 for arm with master

[Richard Purdie]

On Wed, 2016-03-09 at 23:50 +0100, Martin Jansa wrote:
> On Sat, Mar 05, 2016 at 09:40:48PM +0000, Burton, Ross wrote:
> > On 5 March 2016 at 21:19, Rudolf J Streif <rudolf.streif at gmail.com>
> > wrote:
> > 
> > > compile fails with the following error message:
> > > 
> > > > make[2]: *** No rule to make target
> > > > '../readline/libreadline.a', needed
> > > by 'gdb'.  Stop.
> > > > make[2]: *** Waiting for unfinished jobs....
> > > 
> > > The reason seems to be a broken path to the built-in readline in
> > > the
> > > makefile as the default configuration is
> > > --without-system-readline. Enabling
> > > 
> > > PACKAGECONFIG_append_pn-dbg = " readline"
> 
> See http://patchwork.openembedded.org/patch/116843/ discussion, the
> patch from Ross mentioned bellow fixes it better and is already
> merged
> in master.
> 
> > The default is with system readline...
> > 
> > gdb-common.inc:PACKAGECONFIG ??= "readline"
> 
> BTW: this is causing gdb-cross-* to depend on target readline, which
> may
> have different signatures for different MACHINEs with the same arch.
> 
> E.g. after adding qemuarm7 MACHINE, you can easily get gdb-cross-arm
> being rebuilt/extracted-from-sstate after each MACHINE change.
> 
> I'm not sure why selftest didn't catch this, but it can be seen with:
> openembedded-core/scripts/sstate-diff-machines.sh --machines="qemuarm
> qemuarm7" --targets=gdb-cross-arm --tmpdir=tmp-glibc

Something has regressed. Do you know when this last passed? Looks like
we need to add some extra tests to selftest and get this fixed.

Sadly I'm drowning in problems to fix at the moment :(

Cheers,

Richard