Re: dejagnu advice needed for g77

[mrs at wrs dot com (Mike Stump)]

> From: Dave Love <d.love@dl.ac.uk>
> Date: 27 Apr 1998 19:08:37 +0100

> Having dropped a test in g77.f-torture/compile at Craig's behest, I
> now realize it's wrong because the code (compile/980419-1.f) elicits
> a compiler error now it doesn't actually abort.

> I'm ignorant of DejaGnu and any egcs specifics so don't know what to
> do about such cases.  Is RTFMing something specific in order?
> Otherwise could someone advise on setting up such tests (in a
> `noncompile' directory?) or even do it?

> Sorry to be asking without much research, but I'm meant to be finding
> time for g77 alpha-test stuff...

Well, you have a few choices.  You can retroactively `fix' the test
case to be correct, as long as the fixed test case still dumps the old
compiler (if you want to retain the regressions test case), or you can
remove it, and add support for a noncompile directory to the fortran
driver, and then readd the testcase to the noncompile directory, or
you can just remove the testcase altogether, or you can write a .x
file to run it specially.

If you want to setup a noncompile directory, the style used in g++ is
a better style than the one in c-torture.noncompile.  I used the `new'
style, while the `old' style support was being written.  It just so
happens that that style was copied out from g++ for gcc in the
transition period.  :-( The `new' style can be identified by words
like `prebase' and `postbase', the old style by `old-deja'.  The term
`new' and `old' in this context are misleading, I should have used
`hack', and `good' to differentiate between the two.  :-)