Re: [bug-gawk] missing function powm and friends

[Michael Plugge]

Hi Arnold,
thank you for your friendly mai. It was a little laziness to ask for 
inclusion of powm() and next_prime() to the basic functions of gawk :-) 
- you are right, the gawk function set is compact and limited, and 
should remain so. Bundling the functions into an extension is the better 
way. Invocation and performance are the same, but the basic function set 
is clearly defined and separated from more specialized function sets.

Am 27.11.2014 um 21:27 schrieb Aharon Robbins:
> Hi Michael.
>
> Thank you for your offer of the new functions. I appreciate your
> willingness to contribute.
>
> > Date: Thu, 27 Nov 2014 14:45:24 -0500
> > From: "Andrew J. Schorr" <address@hidden>
> > To: Michael Plugge <address@hidden>
> > Cc: address@hidden
> > Subject: Re: [bug-gawk] missing function powm and friends
> >
> > On Thu, Nov 27, 2014 at 03:49:37PM +0100, Michael Plugge wrote:
> > > I wrote a small patch to use the powm() function (and also gcd() and
> > > next_prime()) of gmp; this works even for very large parameters and
> > > exponents. The appended patch file includes also an integer version of the
> > > routines. It would be really great if it could be integrated into gawk :-).
> > I wish we could simply package this as a shared library extension, but the
> > problem is that the extension mechanism passes numeric values as double
> > precision numbers and does not expose the MPFR/GMP numbers used with -M.  The
> > gawk-mpfr extension converts numbers to strings in order to pass them between
> > gawk and the library, so that is one approach to dealing with this.
> I would recommend this as the way to make a first attempt at
> packaging up these functions as an extension.
>
> > My sense is that there's a limited appetite for adding new functions to
> > gawk, and a preference for adding new functionality through extensions.
> That is exactly right. I think these functions would be useful if
> available as an extension.  Michael, can you work with Andy to package
> them into gawkextlib, either as part of the MPFR extension or as a small
> standalone extension?
yes. First I'll read the extension code, mpfr.c and gawkextlib. Maybe 
that will need a while...

> Another option might be to write them entirely in awk using basic
> mathematical operations and assuming the availability of -M.
That could be  a first quick way to implement some functions. powm(), 
gcd() and friends are easy, I'll write a little awk file for them. But 
the final goal should be to include the gmp functions because they are 
much more efficient (and there are many nice-to-have functions in gmp... 
:-) ).

> > But it is challenging when considering extensions that would benefit
> > from the arbitrary-precision representation.  Maybe it's worth giving
> > some thought to how to enhance the shared library interface to support
> > alternate numeric representations.  That may be a big job.
> This is worth some thought, but since it's not likely to happen very
> soon, I suggest that you and Michael work to get his code into gawkextlib
> now. It can be revised later once the gawk team figures out if and how
> to pass MPFR values to/from an extension.
Would it be possible (some day) to pass a node* to an extension, perhaps 
as read-only variable?

Kind regards, and thanks again

Michael


> Thanks!
>
> Arnold