Status update: mini-gmp

[Torbjorn Granlund]

nisse at lysator.liu.se (Niels Möller) writes:

  A status update. It's almost usable by guile (may need to implement
  mp_set_memory_functions, and there are some patches needed to guile to
  avoid using mpq functions for conversions between fractions and
  doubles).

A nice milestone!

  As for GMP's own uses, I can compile gmp it with dumbmp.c cut down to
  the attached version, and then with copies of mini-gmp.[ch] in the top
  directory.

Good.  I suppose it doesn't matter much if some dumbmp.c functions
remain, but I suggest that we at least rename the remainder to something
more apt.  After all, it isn't that dumb, that left stuff.

  Maybe it's getting time to move mini-gmp into the main gmp tree? As for
  the organization, it would be simplest to put the mini-gmp.[ch] files in
  the top directory, but then there's also the mini-gmp README file and
  the testsuite, which I'm not sure how they should be handled.

It would be nicer to have it in a separate directory, both for GMP
directory cleanness and for easy extraction by users.
  
  Maybe one should put everything in a subdirectory, and have gmp make
  check set up testing of it (passing any needed configuration as make
  variables on the make command line, since the mini-gmp testsuite is a
  plain Makefile with no autoconf). The mini-gmp testsuite still depends
  on GMP, which might be an undesirable circular dependency.
  
Does such test dependency matter?  Which are the dependencies?

>From GMP's perspective, I don't think mini-gmp unit testing should be
necessary.  We didn't perform unit testing of dumbmp, but instead relied
on testing of the generated data.  (Assuming generated data testing is
complete, testing mini-gmp as part of a GMP make check has a slight
drawback; it might stumble on compiler bugs that are irrelevent to GMP,
making "make check" cry wolf about the compiled GMP.)

-- 
Torbjörn