[OE-core] race between aclocal and unstaging of .m4 files

[Richard Purdie]

On Mon, 2012-10-29 at 14:42 +0000, Phil Blundell wrote:
> If do_configure() from one recipe runs in parallel with
> sysroot_cleansstate() from another then aclocal may fail because it
> doesn't react very well to .m4 files disappearing underneath it.  This
> manifests as slightly obscure failures such as:
> 
> | aclocal: error: aclocal: file '.../tmp-eglibc/sysroots/x86_64-linux/share/aclocal/alsa.m4' does not exist
> 
> where the .m4 file in question is not one that the recipe being built
> would actually want to use.  (The alsa.m4 error above actually occurred
> during a build of attr.)
> 
> There seem to be two obvious ways of fixing this:
> 
> a) Add more locking so that these things can't happen in parallel
> 
> b) Patch aclocal to make this situation non-fatal
> 
> My inclination would be to do (b) since it's less intrusive.  Any
> objections or better suggestions?

Does current master not resolve this by creating a copy of the aclocal
directory using hardlinks and using that?

We've gone around in circles on this but I think its fixed once and for
all now.

Cheers,

Richard