[Buildroot] [PATCH][autobuild] Use last 4 lines of log instead of 3 when getting failure reason

[Vlad Zakharov]

Hi Thomas.

On Wed, 2016-10-26 at 13:52 +0200, Thomas Petazzoni wrote:
> Hello,
> 
> On Wed, 26 Oct 2016 13:45:17 +0300, Vlad Zakharov wrote:
> > 
> > Latest versions of make produce a bit different output than older
> > versions do. They also print number of line in Makefile where the
> > failure occurred.
> > 
> > E.g. latest make produces the following output:
> > ---------------------------------->8------------------------------------  
> > GNUmakefile:277: recipe for target 'cgiedit.o' failed
> > make[2]: *** [cgiedit.o] Error 1
> > package/pkg-generic.mk:201: recipe for target '/home/buildroot/autobuild/run/instance-2/output/build/privoxy-
> > 3.0.26/.stamp_built' failed
> > make[1]: *** [/home/buildroot/autobuild/run/instance-2/output/build/privoxy-3.0.26/.stamp_built] Error 2
> > Makefile:88: recipe for target '_all' failed
> > make: *** [_all] Error 2
> > make: Leaving directory '/home/buildroot/autobuild/run/instance-2/buildroot'
> > ---------------------------------->8------------------------------------  
> > 
> > When with old version we get something like:
> > ---------------------------------->8------------------------------------  
> > make[2]: *** [cgiedit.o] Error 1
> > make[1]: *** [/home/buildroot/autobuild/run/instance-1/output/build/privoxy-3.0.26/.stamp_built] Error 2
> > make: *** [_all] Error 2
> > make: Leaving directory `/home/buildroot/autobuild/run/instance-1/buildroot'
> > ---------------------------------->8------------------------------------  
> > 
> > So latest make produce much more "noise" and we have to analyze
> > at least 4 last lines of log to be sure that they contain a line
> > with failure reason.
> > 
> > Signed-off-by: Vlad Zakharov <vzakhar at synopsys.com>
> 
> I'm a little bit worried about this change, because we did a fairly
> similar change in commit 585358277b80dfe18b17b13035675bfe7db48b20,
> which I quickly reverted in commit
> a616a8c592caf2a9c9ce82ac1cac9681775452d3 because it was breaking
> everything. Following 585358277b80dfe18b17b13035675bfe7db48b20, for a
> large number of builds, the detected build failure was "all" or
> "all-recursive".
> 
> So we really want to be extra careful, by testing the logic change on
> existing build results first.

But I haven't updated the regex, only lines count printed by tail.
The regex suggested by 585358277b80dfe18b17b13035675bfe7db48b20 was able to recognize both
"make:" and "make[n]:". Note that "all" and "all-recursive" targets appears with the second variant.
I guess that my changes would not affect failure reason parsing from output of older make versions.

I can offer another solution:
We analyze last 4 lines but use the last matched string. As I understand from my experience of analyzing build logs the
string with a real build failure as a rule comes after "all" and "all-recursive".

> 
> Also, another thing that is annoying that we have two places where we
> calculate the failure reason: in autobuild-run (on the autobuild slave)
> and in import.inc.php (which runs on autobuild.b.o when an autobuild
> result is submitted).
> 
> Historically, the failure reason was only extracted in import.inc.php,
> from the submitted log. But later on, some logic was added in
> autobuild-run to extract the reason, which was used to make sure we
> keep in build-end.log not just the 500 last lines of the build, but
> the entire build log from the beginning of the failing package (which
> requires knowing the package that failed building, and therefore
> calculating the reason).
> 
> It would be great to get rid of this duplication, and have
> autobuild-run calculate the reason, store it in a "reason" file,
> submitted to autobuild.b.o as part of the tarball, and import.inc.php
> simply reads that file. This way we have the reason calculation logic
> in just one place.

I can write a separate patch that will exclude duplication and store failure reason in a "reason" file. 

> 
> Thanks,
> 
> Thomas
-- 
Best regards,
Vlad Zakharov <vzakhar at synopsys.com>