Re: [Gm2] About LONGINT fixes

[Gaius Mulley]

Hi,

I've just uploaded the latest cvs to floppsie.comp.glam.ac.uk. In
summary I believe the constant integers (and REALs) to be fixed. GNU
Modula-2 now uses the internal Z type for all constants and then
converts then to the correct type when they are used with variables.

Importantly only 24 regression tests fail out of 2148 on the Opteron
LP64 [9 fail within ulm libraries (Procedure/Module scope nesting) and
15 fail with ISO import/export of enumeration types]. The same happens
in the i386 32 bit implementation - regained some stability :-)

I suspect there may still be some bugs in the constant handling, as
the changes to include the Z type permeated through into the core of
the compiler.  However I think those bugs reported (LONGINT/INTEGER
category) have had regression tests written for them and they are all
fixed.

I'm also uploading gm2-harness-0.3 which fixes a few bugs and
tests for the presence of dejagnu before running the regression tests.

enjoy (and please test :-), have a good holiday,

Gaius


The changes include the following:

2004-12-22      Gaius Mulley <address@hidden>

        * more fixes to enable to the Z type to work.
        * introduced GetM2ZRealType which maps onto the largest
          real type tree.
        * 24 regression test failures out of 2148 on the Opteron LP64
          9 fail within ulm libraries (Procedure/Module scope nesting).
          15 fail with ISO import/export of enumeration types.
        * exactly the same regression test failures on 32 bit i386
          implementation.
        * fixed regression test path, all tests are build and run
          using gm2-harness-0.3
        
2004-11-29      Gaius Mulley <address@hidden>

        * internal changes to ordinal type which now use the Z
          type (long_long_int) for internal calculations before
          being converted into the actual ordinal type at assignment
          to a variable or procedure parameter or return value.
        * works on a 32 bit i386 but fails on the Opteron LP64 model.
        
2004-11-22      Gaius Mulley <address@hidden>

        * compiler now checks for overflow at compile time
          when constructing constant expressions.