[petsc-dev] make ./configure not a symlink

Lisandro Dalcin dalcinl at gmail.com
Sat Oct 2 13:51:47 CDT 2010

On 2 October 2010 12:52, Barry Smith <bsmith at mcs.anl.gov> wrote:
> On Oct 2, 2010, at 12:05 AM, Lisandro Dalcin wrote:
>> On 1 October 2010 22:05, Barry Smith <bsmith at mcs.anl.gov> wrote:
>>> On Oct 1, 2010, at 6:48 PM, Lisandro Dalcin wrote:
>>>> Barry, could we make ./configure a proper file, using
>>>> execfile("config/configure.py") (actually, I'll make it a bit more
>>>> robust than that).  The reason is that symlinks sometimes cause
>>>> trouble, some versions of pip fail to install PETSc because of the
>>>> tarball containing a symlink. I've already had a report of such
>>>> failures.
>>>   Stupid PIP!
>> Well, I think it is not actually a pip problem, but something deep
>> inside the tarfile module in Python stdlib. Fortunately, very recent
>> pip versions seems to somewhat ignore or workaround the error. Anyway,
>> symlinks are definitely not portable. On Windows, ./configure is
>> basically broken, if you hg clone the repo, ./configure is actually a
>> text file with a single line saying "config/configure.py", i.e.
>> totally useless. Stupid HG! ;-)
>>>   I tried not using a link before and always had trouble passing all the arguments with any shell escape stuff they have into the inner python script.  So if it is 100% robust I have no problem changing it, but can you make it completely robust?
>> Can you show and example of what could be problematic? As I'm thinking
>> about using execfile, the inner script will see basically the same
>> sys.argv as the outer one, I fail to see how this could break.
>  Like I said if you can get it to work it is fine with me. I've always had trouble with scripts that are suppose to call another script with the exact same arguments and never had trouble with links, so I am just coming from a different perspective.


Lisandro Dalcin
Predio CONICET-Santa Fe
Colectora RN 168 Km 472, Paraje El Pozo
Tel: +54-342-4511594 (ext 1011)
Tel/Fax: +54-342-4511169

More information about the petsc-dev mailing list