Re: quasi-patch for 2.95 g77/dyn linker bug.

[craig at jcb-sc dot com]

>With Jeff's suggestions, I was able to find the offending file that was
>making the OSR5 dynamic linker drop core.  The bad news (for the poor
>SCO staffer that's going to hear about this while I rant and rave) is
>that it's undebatably a dynamic linker bug.  The good news (for EGCS
>2.95) is that it has a relatively painless workaround.
>
>Without this patch, it will fail on 24 tests.
>
>With this patch applied, a 
>   make check-g77 RUNTESTFLAGS='--target_board unix{-fPIC} execute.exp'
>will pass 470 out of 470 tests.

Hmm, well, that's pretty persuasive...

>If this patch is in the general ballpark of correctness (Dave, Craig?)
>let me know and I'll work up a less rude patch.  If you prefer to
>quarrantine this kludge to OpenServer I can make it do this for only
>OSR5 by sticking it inside a test for #if _SCO_DS.   

...I'd personally lean away from applying more g77-(libg2c-)specific
patches to libf2c, as we already have too many, which has led to our
creating bugs by simply integrating new versions of (official patches to)
libf2c into libg2c.  (I'm probably 90% or more responsible for the
existing differences.)

>In the interim, I'm going to try to reduce it to something to give to
>our dudes so they can fix the real problem.

That would be the best thing.

I'd prefer, for now, to document the need for this interim patch in
the install notes for the release (and include the patch therein, since
it's so small), so only those people who need to build g77 for the
pertinent target system would bother applying it.  (It'd be wise to
"enhance" the patch to also modify Version.c in libI77 so it gives a
different date for `g77 -v'.  Makes responding to proper bug reports
just a tad easier.)

How does that sound?

        tq vm, (burley)