Re: crash in 1.0.3a fortran compilation

[Craig Burley <burley at gnu dot org>]

[These subthread seems more appropriate for egcs than egcs-bugs, IMO,
so I've switched it.]

>Here is test.f. It is a routine from LAPACK. I'm actually amazed that
>you don't compile standard libraries like this as part of the
>regression test.

That's a good point.  OT1H, there's apparently plenty of freely
redistributable Fortran code out there that could be snarfed into
the g77.f-torture suite.  OTOH, that would make the `make check'
phase of typical builds take *much* longer, and the distributions
larger as well, and it's not clear (to me anyway) why that'd be okay
while, at the same time, we don't snarf major C packages (like GNU
Emacs, lynx, and so on ;-) into gcc.c-torture.

Ah, if only it were a SMOP to create a `make full-check' target that
would be implemented by simply including canonical names and
specific version numbers of packages people might already have
on their systems (as source), each combination with URLs to the
source and mirror distribution points and the recipe(s) (or URLs
to source/mirrors of same) for how to fit the package into the
test-suite stuff.

In the meantime, as long as we find compiler bugs exposed by chunks
of code in packages out there, and that code is distributable under
the terms of egcs (the test suite anyway), and isn't too large or
inherently unportable to put into the current test structure, surely
it'd be welcome to add?

        tq vm, (burley)