[SciPy-dev] build error on OSX 10.3

[David Cournapeau]

John Hunter wrote:
> On 10/14/07, David Cournapeau <david at ar.media.kyoto-u.ac.jp> wrote:
>
>    > Complex numbers support may be broken (some problems with
> passing/returning
>    > complex values on 64-bit targets, and not checked against the
> requirements of the
>    > C99 standard). However, the _Complex keyword now works.
>
> Are any 10.3 platforms 64 bit?  If not, there may be little danger in
> simply adding it, or conditionally adding it for non-64 bit platforms.
>   
Cpu-wise, I think G5 is 64 bits capable ? 64 bits platform is a bit 
ambiguous, or at last I am not sure what it really means. Does it mean 
that the platform can have more than 4 Gb of Ram ? Does it mean that 
each process can have more than 32 bits address space ? (Panther 
supports the former but not the later). In our case, the problem seems 
more related to an API problem (difference between 32 bits and 64 bits).
> The other question is: what has recently changed vis-a-vis veclib that
> is now causing the compile on 10.3 to break.  I compiled scipy from
> svn several months ago, though I don't have the  revision number off
> the top of head.
>   
I submitted the above fix because otherwise, you have a problem when 
using complex values returned from fortran (which was causing the 
check_dot errors mentionned sometimes here: veclib fortran interface 
place complex return values in memory, but gfortran expects it to be in 
a register, or something like this). But because the information on 
veclib is so scarse, I did not find a lot of information: veclib gives 
~25 000 hits only on google. In particular, I did not find any mention 
of differences between panther and tiger on this issue.

I am a bit reluctant to add the complex.h before I can understand the 
issue (specially since panther uses gcc 3.3, which has some problems 
with complex support).

cheers,

David