[petsc-dev] Experimental GNU make build system

[Jed Brown]

Satish Balay <balay at mcs.anl.gov> writes:

> On Tue, 28 May 2013, Jed Brown wrote:
>
>> Satish Balay <balay at mcs.anl.gov> writes:
>> 
>> > Pushed a check to 'balay/gnumake-cygwin'
>> >
>> >>>>>
>> > commit 9044defc4215d73a805105aae748d53a5be06605
>> > Author: Satish Balay <balay at mcs.anl.gov>
>> > Date:   Tue May 28 14:19:40 2013 -0500
>> >
>> >     gmake: check cygwin version - and its 1.7.18 or older, use V=1
>> 
>> I cherry-picked this onto 'jed/gnumake' because I didn't want the
>> unrelated patch before it.  Its commit is now:
>> 
>>   b10d012affee97fea384b801cb0aeb042614a900
>> 
>> and I have deleted 'balay/gnumake-cygwin'.  Thanks.
>
> Ah - I haven't noticed this before [primarily because most of my
> testing was with --with-fc=0 - so the link command was reduced by
> half]
>
> But now I get this error with fortran enabled.
>
>>>>>>>>>
> balay at msnehalem2 ~/petsc.clone
> $ make -f gmakefile PETSC_ARCH=arch-cmplx V=1 > x
> /usr/bin/sh: /home/balay/petsc.clone/bin/win32fe/win32fe: Argument list too long
> make: *** [/home/balay/petsc.clone/arch-cmplx/lib/libpetsc.lib] Error 126
>
> balay at msnehalem2 ~/petsc.clone
> $ grep win32fe x | wc
>       1     964   53867
> <<<<<<<<<
>
> Looks like windows programs have '32k' limit on the 'command-line
> argument buffer' [and win32fe, lib etc are windows binaries]
>
> google gives one reference to this problem.
> http://cygwin.com/ml/cygwin/2004-04/msg00285.html
>
> Perhaps we should split up AR command into multiple invocations? [for windows?]

We discussed this earlier in this thread [1]:

http://mail-archive.com/search?l=mid&q=87bo81sjsr.fsf@mcs.anl.gov

I prefer response files to incremental AR because response files also
work for shared libraries.

[1] Trying the mail-archive link because I can't link to our archives
without brute-force search through the dozens of emails in the thread.