Re: egcs-1.1 branch, required for bootstrap on alpha-*-openbsd

[Marc Espie <espie at quatramaran dot ens dot fr>]

In article <28630.920170209@hurl.cygnus.com> Jeff writes:
>  > Sat Feb 27 14:55:51 CET 1999	Marc Espie <espie@openbsd.org>
>  > 	* config/t-openbsd: avoid problems with the name mkstemps 
>  > 	  on machines without dynamic library support.
>I've added this patch.  However, I'm still not convinced that adding OpenBSD
>support to this release was a good idea.  The patch itself is rather
>disgusting.

Well, what can I say ? Do you believe I enjoy asking for a butt-ugly patch
to be included in the 1.1.2 release ?

For my defense:

1/ 1.1.1 misconfigures i386-*-openbsd as i386-*-netbsd, which is the reason
I pushed for inclusion in the first place.

2/ the current mainline fixes the problem very cleanly. libiberty does 
autocheck for mkstemps and never compiles the problematic file under OpenBSD.

3/ not having dynamic libraries is not the normal state of affairs on OpenBSD
platforms. There are various ugly issues that led to that problem on alpha.
First part is that we can't use netbsd code, as there are copyright issues
that OpenBSD must avoid for some reasons. Second part is that we still don't
have any alpha killer machines, as DEC and other vendors don't give a damn
about us, and we have better things to do than spend all the project limited
time begging.

4/ this patch only alters things on openbsd platforms, and we have a working
mkstemps in libc.

5/ we do have some users who will try to compile egcs on alpha and not 
understand what's going on when it fails... linked to 1/. Those same
folks did not even notice that OpenBSD was not really supported in 1.1.

So yep, it's a hack. It solves the problem. 1/ was compelling for me in
asking for OpenBSD support. 2/ is the right way to solve the problem from
the egcs point of view, but I see why you would not want to bring all those
changes over to 1.1. 3/ is the correct way to solve the problem from
the OpenBSD point of view (given unlimited resources and time, we would have
upgraded to binutils 2.9.1, hacked at it some more to get all the support
we need, fixed the mkstemps problem, added dynamic libraries to the alpha port,
and ported OpenBSD to the Commodore 64 already...)

Let's drop that, shall we ? or at least take this out of the egcs 
mailing-lists.