[poky] Is sstate broken

[Richard Purdie]

On Wed, 2011-03-09 at 11:42 -0700, Gary Thomas wrote:
> I updated to master (fab742bd4693ed3092690a20dd32d53fe27c3d4c) and tried again.
> I don't see any difference - the run using the sstate cache as a mirror
> seems to do all the same work as without.  Here's how I tested it.
> 
> * Build original tree
>    % POKYCONF=meta-cobra3530p60/conf . /tmp/poky-amltd/poky-init-build-env /local/p60_step1
>    ... adjust conf/local.conf
>    % bitbake amltd-console-image
> 
> * Rebuild, using previous result for SSTATE_MIRRORS
>    % POKYCONF=meta-cobra3530p60/conf . /tmp/poky-amltd/poky-init-build-env /local/p60_step2
>    ... adjust conf/local.conf
>    % bitbake amltd-console-image
> 
> The only difference between the two runs is enabling SSTATE_MIRRORS in local.conf
>    %  diff -u /local/p60_step?/conf/local.conf
>     --- /local/p60_step1/conf/local.conf    2011-03-09 08:28:18.266933061 -0700
>     +++ /local/p60_step2/conf/local.conf    2011-03-09 09:57:51.365932951 -0700
>     @@ -53,4 +53,7 @@
>      IMAGE_LINGUAS ?= "en-us"
> 
>      # Minimize feature set
>      DISTRO_FEATURES ?= "alsa"
>     +SSTATE_MIRRORS ?= "\
>     +file://.* file:///local/p60_step1/sstate-cache/"
> 
> The results seem to have gone through all the same steps (or nearly so).  The output
> from the runs is at
>    http://www.mlbassoc.com/poky/build.step1
>    http://www.mlbassoc.com/poky/build.step2
> 
> Comparing the two build trees:
>    % ls /local/p60_step1/tmp/work/*/*/temp/log.do_compile | wc
>      144     144   12521
>    % ls /local/p60_step2/tmp/work/*/*/temp/log.do_compile | wc
>      143     143   12427
>    % du -s /local/p60_step?
>    15229296        /local/p60_step1
>    15162760        /local/p60_step2
> 
> I know this procedure used to work (or at least close).  Am I doing
> something wrong?

You're not doing anything wrong and this is the same scenario I've been
testing with. After I'd fixed the origin problem it created a problem
with file urls containing globing. I managed to break the original patch
with the globing fix. The good news is the problem is simple and I've
pushed a fix.

It should work *much* better than you've seen above, not needing to
rerun many tasks so if its working it will be very obvious.

Cheers,

Richard