COMPLEX functions and g77 compatibility (was: Re: Can gfortran compile fftw3?)

[Steven G. Johnson]

Tobias Schlüter wrote:
> I thought passing / returning COMPLEX is handled differently, but I was sure
> someone would report a bug if this was truly the case, so maybe it works the
> same after all.  Or maybe it's just not important.

Hi Tobias,

I just checked this.  First, argument passing of COMPLEX is apparently 
the same, and calling FFTW's transform routines from gfortran produces 
the correct results (aside from an I/O glitch, described in another 
message).

However, *returning* COMPLEX from a *function* is apparently different. 
  This doesn't affect FFTW (it only presents a subroutine interface to 
Fortran, since cross-language function returns are problematic), but I 
think it is a relatively important problem.

For example, this means that you cannot have a single optimized BLAS 
library on your machine that is usable from both g77 and gfortran 
(thanks to functions like zdotc).

I've filed a bug report with a test case: bug #20178