[Libmesh-devel] configure should fail on invalid PETSC_DIR or PETSC_ARCH

[John P. <pet...@cf...>]

Mike Taylor writes:
 > I think it would be less confusing for most folks if configure failed when 
 > PETSC_DIR or PETSC_ARCH are missing or invalid.  Print a nice message 
 > explaining the situation and what to do about it.  One of the "fixes" 
 > suggested to the user could be --disable-petsc.
 > 
 > At one point, I had PETSC_DIR set correctly but had failed to set 
 > PETSC_ARCH.  This caused the build system to put together a string like:
 > /usr/lib/petscdir/2.1.6/bmake//packages 
 > , try to use it, and fail with a less than informative message.
 > 
 > I think it would be less confusing and more consistent to communicate 
 > petsc configuration data though configure switches rather than environment 
 > variables.  However, you guys have been doing this for a while and may 
 > have good reasons for continuing to configure petsc through environment 
 > variables.  If so, a few lines about these variables in configure --help 
 > would be nice.

Hi Mike,

You bring up a good point.  The reason we have used environment variables
in the past is that Petsc (when you built it yourself) made you do it.
I guess now they are moving to a more GNU-like system so they might be
phasing it out?  Regardless, if you have Petsc as an RPM, you might not
have the environment variables even though you of course have Petsc.

For now I think we will stick with the environment variables, but if one
is missing we'll set --disable-petsc during configure.  The other developers
may want to comment as well...

-John