On 2012-03-17, at 12:06 PM, Tyler Luchko wrote:
>
> On 2012-03-17, at 12:26 AM, Ross Walker wrote:
>
>> Hi Tyler,
>>
>>> I think we should have a consistent policy as to what happens when some
>>> functionality is not available with a particular compiler or whatever.
>>> IMO, I think configure should _always_ die with an error message and
>>> explicit instructions on how to proceed. My previous error message
>>> would return the user's original configure command modified such that
>>> RISM would not be built by adding '-norism' and removing any instances
>>> of '-rism'. This way they could copy and paste to quickly get to the
>>> next step but also knew exactly what functionality was missing.
>>
>> The problem is that ./configure -norism gnu no longer works. It still tries
>> to build rism because of the dependency on rism that has been introduced in
>> the building of libsff. So right now it makes no difference if -norism is
>> specific or not - the build process still tries to build FFTW3 and that's
>> where it crashes.
>
>
> I've pushed out a working version that still automatically turns on -norism for older compilers.
>
> I apologize for the problems as this my fault. I tried to clean up the build process a bit and did test the -norism flag (it didn't produced any RISM executables) but neglected to test it with GNU 4.1.2.
>
> I will now re-implement the original error message for configure unless there are any objections.
>
I've changed configure to throw an error and give an explicit error message with instructions on how to proceed. It has been tested with CentOS 5.7 (GNU 4.1.2 [serial and MPI], Intel 10.1 018 [serial and MPI] and PGI 9.0-1 and 9.0-4), Ubuntu 11.10 (GNU 4.6.2) and Mac OS X 10.6.8 (4.6.3).
There are many places in configure that throw similar errors (or maybe they should, e.g. automatically turning off MTK++ for certain flags) and would benefit from providing the user with a modified configure command. If there are no objections, I can update these error messages to have the same uniform output and information.
Tyler
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Mar 17 2012 - 14:00:02 PDT
