[petsc-dev] Configure problem on Titan

[Barry Smith]

> On Mar 11, 2015, at 11:43 PM, Satish Balay <balay at mcs.anl.gov> wrote:
> 
> On Wed, 11 Mar 2015, Jed Brown wrote:
> 
>> Barry Smith <bsmith at mcs.anl.gov> writes:
>>>  If one fails then configure moves on an tries with the other. Is
>>>  this a problem? We do it all over the place on configure.
>> 
>> The problem is that both may "work" (link without errors), but exhibit
>> subtle errors at run-time due to ABI mismatch.
>> 
> 
> The current usage mode here is:
> 
> --with-clib-autodetect=0 --with-cxxlib-autodetect=0 --with-fortranlib-autodetect=0 LIBS=-lstdc++
> 
> In this mode - I don't think configure should be gussing - and let the
> user provide the appropriate compatibility library list.
> 
> If any fixes - they should go for usage:
> --with-clib-autodetect=1 --with-cxxlib-autodetect=1 --with-fortranlib-autodetect=1

  Oh yes. But then I've always hated the --with-clib-autodetect=0 --with-cxxlib-autodetect=0 --with-fortranlib-autodetect=0 since it requires too much knowledge of the user.

   Why not have --with-clib-autodetect=1 --with-cxxlib-autodetect=1 --with-fortranlib-autodetect=1  

first check if Cray system and try just -lstdc++ ?  Then you don't get the huge mess of libraries of libraries to link against but don't require  users to know machine specific things.

  Barry




> 
> Satish