[petsc-dev] worthless information

[Vijay S. Mahadevan]

> Some people, when confronted with a problem, think "I know, I'll use regular
> expressions."

I've been using them for several years and have found them to be
powerful to extract expressions I care about. Most often, they are
handy in the configuration process.

> FWIW, we care at least as much about the path of the compiler as its
> version. Usually the first word of "mpicc -show" has this (provided spaces
> in paths are escaped appropriately).

I do not think mpicc -show gives the full path either. It does provide
the name of the compiler but version, I think is all the more
important. I understand that getting this info via autoconf or some
other regular build system wasn't the cause for the discussion but
think waf is extensible in this context, based on my experiences. The
snippet above is generic enough to be added to any python based
configure system though.

> My first complaint about waf is it's lame handling of configuration for
> batch systems. I also don't see a good waf-based library for figuring out
> how to build against dependent libraries, which mostly makes it a make
> replacement (the easy part) instead of a configuration system (the hard
> part).

I've personally not used waf much on batch systems yet but have used
to configure and build against several dependencies
(petsc/slepc/libmesh/deal + other C++ libraries). I had to write a few
utilities of my own on top of the original waf system to get this
right though for a complex system like PETSc, there might be some
additional needs.

On Tue, Aug 30, 2011 at 3:08 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> On Tue, Aug 30, 2011 at 14:47, Vijay S. Mahadevan <vijay.m at gmail.com> wrote:
>>
>>        vnum = re.findall(r'^.*?\d+\.\d+(?:\.\d+|(?:\.\d+\.\d+))?',s)
>
> Some people, when confronted with a problem, think "I know, I'll use regular
> expressions."
> Now they have two problems.
>
> FWIW, we care at least as much about the path of the compiler as its
> version. Usually the first word of "mpicc -show" has this (provided spaces
> in paths are escaped appropriately). Finding this information is not
> difficult, we're just discussing what information is most important in the
> summary and taking it as an opportunity to complain about existing systems.
> My first complaint about waf is it's lame handling of configuration for
> batch systems. I also don't see a good waf-based library for figuring out
> how to build against dependent libraries, which mostly makes it a make
> replacement (the easy part) instead of a configuration system (the hard
> part).