[Gm2] cvs checkins

[Gaius Mulley]

Hi,

here is a summary of the commit log since the beginning of the month:

gm2
  - fixed bug in FpuIO.mod which failed to issue a '-' when emitting
    real numbers.
  - added many new tests to the regression directory (autotest).

  - fixed bugs in the bnf descriptions. Compiler is now using automatically
    generated parser for P1, P2 and P3. Just PH remains hand built.

  - Introduced a number of small regression tests following Olivier's
    extensive bug report on (m2f). Fixed a number of bugs reported,
    see gm2/m2/comp/BUGS for details.

  - Fixed bug in M2Lexical.mod causing mismatch between source line numbers and
    gdb information.

  - Fixed + - / of constant sets.

  - removed tests for the interim (they exist under bitset3.mod.breaks and
    string.mod.breaks respectively)

  - modifications to stop overflow on max CARDINAL when converting numbers into
    strings (NumberIO.mod)

m2f related changes:
  - added ability to link against C libraries via the -lxyz option (m2f.mod).
  - fixed two bugs in mod and div arithmetic with negative numbers.

the state of play is that:

   m2f will build itself reliably and can run the verify tests correctly.

   gm2 will build but will _fail_ on `make paranoid' (with gcc-20001204)
       thus the some of changes above have broken gm2. Probably due
       to the new P3 parser..

the tar'ed gzipped archive of gcc-20001204 + gm2 passes the make
paranoid (the gm2 version being older than these modifications of
course).

Gaius