Re: A few more test case cleanups for -fPIC and SCO

[Kean Johnston <jkj at sco dot com>]

> I'm not finished looking through your changes, but have a couple of
> comments.  First of all, there's a check in gcc.dg/tls/tls.exp for TLS
> support.  If that check isn't detecting that your target does not
> support TLS, then the appropriate thing is to fix that check so the
> whole directory is skipped.
Ok that test is wrong then. The code compiles, so the test passes
and it tries the tests. But the platform doesn't support the notion
of TLS. So one of two things is wrong (and I dont know which):
either the platform port is wrong in that it compiles the code when
it should be producing an error, or the test in tls.exp needs to
be trained to ignore the platform.

Jakub mentioned that there was a patch for a target specific
check for TLS, so we could use something like
     { dg-require-effective-target tls-runtime }
which seems an equally appropriate way to go.

> Your new checks are for *-*-sco3.2v5; does that cover all of the
> targets it should, or is the version too specific?
No, its just right. sco3.2v5* is the only set in the 3.2 series
that is supported by gcc, and it doesnt, and never will, support
threads or TLS. The later versions all support threads just fine.

> Test gcc.dg/ucnid-3.c changes from XFAIL to skip for other targets;
> is that on purpose?
Yes indeed. The xfail causes a false-negative. The test compiles
but fails to execute, so it gets a FAIL. By changing it to skip,
it doesn't even try to compile it.

Kean