Re: [Plplot-devel] F95 changes

[Andrew R. <and...@us...>]

On Wed, Jan 25, 2012 at 09:45:25AM +0000, Andrew Ross wrote:
> On Wed, Jan 25, 2012 at 09:35:23AM +0100, Arjen Markus wrote:
> > Hi Alan,
> > 
> > On 2012-01-24 20:06, Alan W. Irwin wrote:
> > > 
> > > Actually, trailing blanks in character strings do have a significant 
> > > effect in Fortran.  In particular, if a Fortran
> > > user wants to put a trailing blank into one of the character
> > > string arguments of PLplot, she should be allowed to do so.
> > > 
> > > In fact our Fortran bindings did preserve Fortran trailing blanks up
> > > to when a regression was introduced by the recent changes in
> > > bindings/f95/strutil.f90.  I have now fixed that regression with a
> > > one-line change (revision 12152).  The result is example 28 (and
> > > presumably example 20 which I didn't have time to test) implemented
> > > with C and Fortran now give consistent results again.
> > > 
> > > Arjen, please review revision 12152 to make sure it is consistent with
> > > the spirit of your recent changes.
> > > 
> > 
> > I misread the intention of the original code, I guess. I will have a
> > look and see if the new code is indeed equivalent.
> > 
> > Regards,
> > 
> > Arjen
> 
> Alan's changes also don't work. The len instrinsic returns the length 
> of the fortran character array, not the actual length of the string. 
> This means that, for example, passing a character array of len 8 which
> you have set to 'test ' will actually pass 'test    ' to the C interface.
> This causes example 14 to fail when setting the driver since the padding
> spaces makes the name invalid. It also causes postscript regression on 
> a large number of other examples. 

To follow up, I've looked back at the old code for converting fortran 
strings to C strings and it basically fudged it.

If the string contained characters, then just the characters were passed 
(with trailing spaces removed). If a null string was passed (i.e. of 
length 0), then the returned string was a C NULL string. If the fortran 
string was of length > 0, but was all spaces, then a single space was 
passed to C. This seems entirely arbitrary. It made the examples work, 
but wouldn't definitely do what the user expected.

I suggest we continue as we are following Alan's changes, i.e. pass 
EXACTLY the fortran string to C. The onus is then on the user to ensure 
the string is the right length or to strip any unnecessary spaces. We 
can't do this manually since we don't know which trailing spaces are 
necessary and which are not. 

This will be an API change for the f95 bindings, and will require 
a number of examples to be updated. Are people happy with this?

Andrew