Re: PATCH: Run V3 tests on installed compiler

[Benjamin Kosnik <bkoz at redhat dot com>]

> > In practice, how do you pass a list of just files in say
> > testsuite_27/basic_istream? Or testsuite/tr1? Or testsuite/ext? 
> 
> You can use:
> 
> runtest --tool libstdc++ normal.exp=`find $srcdir/17_intro -name *.cc`
> 
> or:
> 
> runtest --tool libstdc++ normal.exp=`cat testsuite_files`
> 
> (I just tried this with a "testsuite_files" that contained just a few 
> tests, and it worked fine.)

Hmm. I must be doing something wrong: neither works for me. I've attached my output.

Help? I tried doing this in the gcc build directory and in the libstdc++
build directory, neither worked. Same with adjusting paths. 

This is probably something I'm doing wrong.

> This the same procedure that you can use elsewhere in the compiler; it's 
> the normal DejaGNU way of doing things.  I think that if we can make the 
> V3 testsuite work more like the other testsuites, without making it 
> harder for you to do development, then that's a win.

Yep.

> > http://gcc.gnu.org/onlinedocs/libstdc++/test.html
> > 
> > Please bother to read it before complete changing things underneath our
> > feet, without warning. If you're changing current practice, please
> > update this doc at the same time.
> 
> OK.  If you do not ask me to revert my patch, I will update that 
> document to use the example above to show how to feed the contents 
> testsuite_files into DejaGNU.

Please do the doc updates.

> I'm also surprised in that there was no negative reaction to Geoff's 
> patch, other than the technical problem that it didn't work with 
> multilibs.  I didn't see any indication that the V3 maintainers were 
> unhappy with the concept.

Geoff's patch was quickly reverted, thus I figured I'd have time to comment at the next iteration.

It seems I was mistaken.
 
> > A way to repair the damage would be to keep the testsuite_files makefile
> > rules, not have it run by default, and *if this files exists* (ie
> > maintainers have run the make rule) use it for the files, if not
> > generate dynamically. That would allow library maintainers to keep this
> > more efficient workflow, and allow you to test install as well.
> 
> To be clear, the rules are still there, and are in fact still run by 
> default.  I wonder if the fact that my initial message was unclear about 
> what was actually checked in vis a vis what I'd like to do in future has 
> muddled the situation so badly that you're not able to discern what's 
> actually happened?

I'm running with your patch applied sir. I'm pretty clear on what the
current state is vs. the future plans.

> Your plan would work too, but rather than having DejaGNU look for the 
> file, I think it would be technically better to have the user use the 
> normal DejaGNU mechanisms for selecting which tests to run.  Perhaps, we 
> could put the logic you're looking for into the Makefile, so that "make 
> check-maintainer" automatically passed in testsuite_files, if typing the 
> runtest command manually is noxious.  I don't feel very strongly about 
> this; if you think its vital that DejaGNU look for the file, I'll make 
> that change ASAP.

No make check-maintainer rule please. 

I was just offering a solution: it looks like you've thought of another that is superior.

-benjamin

Attachment: mark_testsuite.txt
Description: Text document