Cross-testing: check-target-libstdc++-v3 with dejagnu in build tree.

[Tom Tromey]

>>>>> ">" == Hans-Peter Nilsson <hp@bitrange.com> writes:

>> The patch is no good as is, since Makefile.in in
>> libstdc++-v3/testsuite is generated from Makefile.am, and this
>> check-DEJAGNU rule is an automake-generated thingy.  I think it
>> exposes a problem with automake, which seems to by default assume
>> that libraries are only tested as host libraries, in which case the
>> library dir is one .. closer to the host expect build dir.

Yes.  This code was written a long time ago -- I think it was 1996 --
and as far as I know hasn't changed substantially since then.  Back
then I knew nothing about target libraries or multilibbing, so I
ignored those features.

>> IIRC the automake documented remedy is to enter the check-DEJAGNU
>> rule manually in Makefile.am, but I think the automake-generated
>> rule needs fixing too.

Actually, doing it by hand is a hack.  This isn't an exported name, so
it could change from release to release.  Someday it will (when we
start moving internal targets to their own namespace).

>> I don't really know how or if the test should be tweaked to deal
>> with the case where libstdc++-v3 is a multilib.

That's the issue.  Someone has to just sit down and figure out how we
can get information about the "real" top-of-tree, and then put this
into dejagnu.am.

Could you submit an automake PR for this problem?  That way we can at
least track it.  You can find Gnats by following links from the home
page:

    http://sources.redhat.com/automake/

Thanks.

Tom