libtool
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Problems with AIX4 Compiler


From: Erik Lindahl
Subject: Re: Problems with AIX4 Compiler
Date: Sat, 21 Jul 2001 11:00:24 +0200

Tim Mooney wrote:

>libtool (and many GNU projects) support based on "popularity", meaning
>that the "stuff" (features, OSes, etc. -- in this case, compilers and
>build environments) that is most widely used gets support first, and
support
>for new "stuff" comes when someone needs it and helps get it in place.
>Since most users on AIX systems are using either gcc or IBM's cc,
that's
>what's supported the best.  This also means that libtool works for the
>vast majority of people on AIX.  It misses the "edge cases", and here's

>a good one you've just pointed out.

Hi Tim,

Sure I'm aware of the GNU development, I wouldn't ever dream of
complaining
about it, on the contrary, I'm writing patches :-)

>I believe it should be very do-able to add support for an additional
compiler
>on AIX.  Since you have access to mpcc, perhaps you would want to
consider
>helping with this?  You might want to take a look at the stanza for
Linux
>and C++ or C -- since there are multiple options for C or C++ compiler
there,
>it might make a good template for how to add support for mpcc on AIX.
Once
>you have that in place there, other people will benefit too, since MPI
is
>also popular on SGI's Origin's, Beowulf clusters, and elsewhere.
>

It's actually much easier; I've never seen a system where mpcc is a
separate
compiler. It's essentially always a scripts that replaces

mpcc prog.c

with

cc -I<path to mpi include> -L<path to mpi libs> prog.c -lmpi

and possible some special hardware libs too... This is a good thing,
since it
means all GNU tools work out of the box.

The only thing that breaks this on sp2/aix is that the -blibpath option
removes the hardcoding of <path to mpi libs>. The executables need
this, since they use dynamic linking per default (an environment
variable is used to decide which communication libraries should be
linked at runtime).

So, I have a patch for this, but it would mean removing the -blibpath
option and relying on hardcoding of the -L flags instead, like on aix
version 3.
The question is, would you find that acceptable from a security point
of view?

>If you've got a patch that you think is close (and it's not huge)
submit
>it to this list for comments and review.  Since I don't have access to
an
>SP2 or other IBM system with MPI, I can't test, but there might be
others
>that could, and the peer review might turn up some issues.

It's a very small patch - the only issue is the libpath policy :-)
If  you can live with the same setup as aix3 uses I'll submit the patch
by monday...

Best wishes,

Erik






reply via email to

[Prev in Thread] Current Thread [Next in Thread]