Re: F2003's IOMSG feature and Janne's FLUSH patch

[Steve Kargl <sgk at troutmask dot apl dot washington dot edu>]

On Sat, Jul 30, 2005 at 09:36:30PM +0100, Paul Brook wrote:
> On Saturday 30 July 2005 20:47, Steve Kargl wrote:
> > I've finally found some time to review Janne's implementation
> > of the F2003 flush command.  In expanding Janne's testcase to
> > cover ERR=, I came across IOMSG= in the draft of the F2003 standard.
> > gfortran does not  support IOMSG.  So I have 2 questions.
> >
> > (1) Should we implement IOMSG before committing Janne's patch?
> >
> > (2) Can I commit Janne's patch with an appropriate TODO comment?
> 
> I don't mind incomplete implementations, provided:
> - They as commented as such

I'll add comments.  I'll also document the behavior in 
gfortran.info.

> - The missing bits can be implemented without having to redo the work of the 
> original patch.

I think this is the case.  AFAICT, IOMSG would be a clone of IOSTAT
with the integer codes replaced by a message catalog.  If I have
time I'll look into this.  Note, IOMSG is needed for all data transfer
statements in Fortran 2003.

> - We never generate incorrect code because of the missing features. Refusing 
> to compile anything is better than getting some cases wrong.
> 
> My preference would be that the compiler call sorry() or gfc_todo() if it 
> encounter things that aren't supported.

At the moment we get

 In file flush_1.f90:29

   flush(unit=10, iomsg=msg)
                1
Error: Syntax error in FLUSH statement at (1)

Is this sufficient?

-- 
Steve