Do not terminate default test runs on test failure

[Joseph S. Myers]

On Tue, 1 Apr 2014, Andreas Schwab wrote:

> "Joseph S. Myers" <joseph@codesourcery.com> writes:
> 
> > That means that some dependency of the toplevel tests target failed 
> > (possibly because a test failed to compile), in which case the commands 
> > for that target don't get run.  Generation of summaries relies on 
> > dependencies not failing
> 
> This is a serious regression, please fix ASAP.

I see no regression from my patch.

Suppose there is a build failure for some test (and you need to make clear 
exactly what errors you see, on what architecture, before we can start to 
work out the correct way to address them).

If you don't use -k, then, both before and after my patch, the test run 
stops with error status when that build failure occurs, without printing a 
summary of results.  No regression.

If you do use -k, then, both before and after my patch, the test run 
continues past the failure to run as many tests as possible, and once the 
possible tests have run it finishes with error status, without printing a 
summary of results.  No regression.

Maybe some other patch - say, the -Wundef patch - caused a regression, but 
you haven't given sufficient information about the actual failures you see 
and the configuration in which you see them to determine the right fix.

-- 
Joseph S. Myers
joseph@codesourcery.com