Re: Do not terminate default test runs on test failure

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Tue, 11 Mar 2014, Roland McGrath wrote:

> > Apart from needing make -k being unnecessarily different from normal 
> > conventions for how testsuites work, it also doesn't work for getting the 
> > test summaries: if any tests (dependencies of the tests target) fail then 
> > the makefile rule for the tests target (which in a subdirectory generates 
> > subdir-tests.sum, and at toplevel also generates tests.sum) doesn't get 
> > run.  The point of this patch is to generate a normal summary file of test 
> > results at toplevel, regardless of whether individual tests passed or 
> > failed - so you don't need to use -k and postprocess make output to 
> > identify failures.
> 
> Yeah, that's sensible enough.  Just so long as the top-level make exits
> nonzero when there were failures.

Whether to exit with error status if there were FAILs, or only if there 
were ERRORs, was one of my questions on the first version of this patch, 
where the only previous comment was from Brooks preferring FAILs not to 
cause an error status 
<https://sourceware.org/ml/libc-alpha/2014-02/msg00301.html>.  I'm happy 
either way; error exit for any FAILs is just two extra grep lines like 
those ensuring error exit for any ERRORs.

-- 
Joseph S. Myers
joseph@codesourcery.com