Re: Testsuite patch for Solaris 10

[Janis Johnson <janis187 at us dot ibm dot com>]

On Fri, Nov 12, 2004 at 05:10:12PM -0500, Daniel Jacobowitz wrote:
> On Fri, Nov 12, 2004 at 10:04:17PM +0000, Joseph S. Myers wrote:
> > On Fri, 12 Nov 2004, Janis Johnson wrote:
> > 
> > >   { dg-require [expr [istarget i?86-*-*] && [expr ![is-option "-m64"]]] }
> > > 
> > > What do you think, besides the daunting task of updating all those
> > > tests?
> > 
> > It's not clear to me that embedding pieces of TCL code syntax in 
> > individual tests - rather than dg-whatever-keyword which have defined 
> > simple syntax independent of the particular test harness - is an 
> > improvement.  It seems desirable for the test format to be seen as a 
> > specification independent of implementation language - which might, for 
> > example, be used with QMTest's DejaGnu emulation, not just with the 
> > particular current harness.
> 
> Amen.  Janis, while I think your idea is an excellent one, how about
> keeping all of the TCL bits in predicates in some separate .exp file?

I experimented with lots of ways of using conditional expressions in
dg-target-list but had problems with them not being evaluated early
enough.  Joseph, can you hold off for a few days so I can try to find
a way to support conditional expressions within the test framework
other than via TCL?

I'm worried about the proliferation of dg- commands used in tests and
unexpected interactions between them.  For example, as it's written
dg-forbid-option doesn't clear xfail flags that might have been set in
an earlier line in the testsuite, so those flags will apply to the next
test that is executed; it should use skip_test_and_clear_xfail which
was added a day or two ago.

My goal is to provide flexibility for marking tests as UNSUPPORTED or
XFAIL so we can have clean test reports, with tests skipped for targets
or options for which they are not relevant and xfailed when there is an
open PR.  The gcc.dg/vect tests are a good example of the challenges.

Janis