fprintf for GMP objects?

[Ronald Bruck]

I've just started using GMP, and am delighted at finally being able 
to get good multi-precision results.  But I have a couple of 
questions:

1. Is there a C++ wrapper for it yet?  I'm mostly interested in the 
floating-point stuff.

2. Failing that, has anyone written a relacement for fscanf and 
fprintf which will read/write GMP reals instead of doubles?

I have ported Brian Borchers' csdp (a semidefinite-programming 
optimizer) using GMP rationals and floats instead of doubles, and 
life would have been SO much easier with (1).  I wrote my own kludgy 
version of fprintf, just to print single floats (with preceding and 
following text) using %w.de notation, but I hate reinventing the 
wheel.  Surely someone has written a full fprintf?!  If nothing else, 
using K&R's example of how to use a variable argument list, 
intercepting the "%e" call.

I also wrote a routine to read an object from a file, deciding 
on-the-fly whether it's an integer, a rational, or a float; and 
storing the value in a GMP integer, rational, and float, as they make 
sense.  (For example, if it reads an object it decides is an integer, 
it stores it in the GMP integer, GMP rational, and GMP float.) 
Again, this is surely reinventing the wheel.

And there are implementation questions I've sloughed off, because I 
have no one to ask.  For example, if my fprintf finds a format 
specification of %e, it diverges from the standard fprintf in that it 
will use an mpf_out_str with a precision of 0, i.e. it will print 
FULL PRECISION.  That means it won't be compatible with the standard 
fscanf.  For my purposes that's not important, but if it were to be 
public code, this would have to be thought about.

So what sort of i/o is available?  What sort of wrapper?

I'll make csdp publicly available when I've tested it more 
thoroughly.  Right now it's kind of rough.  But if you want to peek 
at it, you can get it by anonmyous ftp from:  imperator.usc.edu.



--Ron Bruck