ASM allowed in Fortran? and OpenDir?

[Walter Spector]

Janne Blomqvist wrote:
> 
> On 8/8/07, FX Coudert <fxcoudert@gmail.com> wrote:
> > >     integer :: dirid
> > >     integer :: jdirent
> > >     integer :: pxferr
> > >
> > >     call pxfopendir (mydir, len (mydir), dirid, pxferr)
> >
> > This assumes that the directory id can fit in an integer, which might
> > be an issue for 64bit platforms, isn't it?

It depends on the size of your integer.  PXF specifies the default-sized
integer because that is all the Fortran-77 (and -90/95/2003 for that matter)
Standard requires.  And yes, these days that usually means 32 bits.

> Not just 64-bit platforms, any systems with LFS, which includes pretty
> much everything released within the last 5 years... Of course, when you do
> hit that bug it'll be one of those infuriating "works on system A and
> not on system B, although they have identical software"...

A quality implementation would detect that the integer size has
been exceeded, and return ETRUNC in the IERROR argument.  This would
at least allow the caller to know that there is a problem.

> ... I suppose supporting even a withdrawn
> standard would be better than nothing, but the contortions PXF* goes
> through to access C from F77 looks rather painful. Some F90+-iy
> interface making use of derived types would surely be much nicer (I
> suppose PXF90 doesn't exist?).

It really isn't that painful.  The steps are identical to what one
would use in C.

That said, I agree that it would have been nice to have a 'PXF90'.
There are several badly needed extensions - which I made in my private
version.  Two are pxfint8get - which returns 8 byte fields (and is compatible
with the Intel and IRIX extension of the same name), and pxfstrerror - which
returns a string describing a given errno.  I also support symbolic links,
a pxfsystem, and a few other things.

Interestingly, I have not felt an extreme need for derived types.
The problem, again, is that if one has a module with derived
types defined in it, there is no way to automagically extract the proper
layout from the master definitions in the C headers.  You end up with a lot
of conditional code, multiple versions, porting problems, etc.  The only
way I know to avoid this is to write a pxfstructcreate-like routine in C.
But if one is going to go that far, why not just write pxfstructcreate and
be done with it?

W.