eliminating a cause of my recent rant and argument with Matt

Satish Balay balay at mcs.anl.gov
Tue Dec 15 15:30:18 CST 2009


Wrt batch mode, the relavent PETSc configure flags are:

  '--with-batch=1',
  '--with-mpi-shared=0',
  '--with-memcmp-ok',
  '--sizeof_char=1',
  '--sizeof_void_p=4',
  '--sizeof_short=2',
  '--sizeof_int=4',
  '--sizeof_long=4',
  '--sizeof_size_t=4',
  '--sizeof_long_long=8',
  '--sizeof_float=4',
  '--sizeof_double=8',
  '--bits_per_byte=8',
  '--sizeof_MPI_Comm=4',
  '--sizeof_MPI_Fint=4',
  '--have-mpi-long-double=1',


wrt autoconf - I think these [sizeof stuff] are just shell
variables. The last time I tried doing something equivalnet to
--with-batch - I just set these variables on the command line.

./configure ac_cv_sizeof_long_int=8 ac_cv_sizeof_int=4

etc.. So perhaps the current sizeof_char type options already mimic
autoconf. But if we are using --known-sizeofint=8 type flags - thats
also fine.

The other odd flags are --with-memcmp-ok and --have-mpi-long-double=1

Satish

On Tue, 15 Dec 2009, Barry Smith wrote:

> 
>   I think a root of the miscommunication between Matt and I with regard to
> --with-mpi-shared is that --with is used for TWO different reasons
> 
> 1) --with-mpi=0 --with-x=1 etc are used to indicate we want to turn off/on a
> package. In other words they are COMMANDS to BuildSystem
> 
> 2) --with-mpi-shared=1, --with-sizeofint=8 tells BuildSystems FACTS that it
> would normally be figured out by BuildSystem. Usually they are provided on the
> command line because ./configure cannot run compiled programs (using
> --with-batch for example).
> 
>   I suggest we can the syntax for 2) and consistently use --known-mpi-shared
> --known-sizeofint=8 etc everywhere in help and BuildSystem. It makes things
> much clearer and can prevent loads of confusion.
> 
>   This is likely not compatible with how autoconf does things. Do they have a
> similar split and different names? Is there anything we can copy from their
> approach? Do we care?
> 
> 
>   What do others think?
> 
>   Barry
> 




More information about the petsc-dev mailing list