Re: [Gm2] Build Problems on Mac

[Gaius Mulley]

Michael Lambert <address@hidden> writes:

> Hi Gaius,
> 
> On my Intel box 'make paranoid' passes, but, as likely expected, there
> are problems in the regression tests.  I did apply a patch
> 
> --- i386.c.original     2006-05-16 21:11:59.000000000 -0400
> +++ i386.c      2007-02-07 21:26:30.000000000 -0500
> @@ -16637,7 +16637,7 @@
>      {
>        fprintf (file, "\tcall LPC$%d\nLPC$%d:\tpopl %%eax\n", label,
> label); fprintf (file, "\tmovl %s-LPC$%d(%%eax),%%edx\n",
> lazy_ptr_name, label);
> -      fprintf (file, "\tjmp %%edx\n");
> +      fprintf (file, "\tjmp *%%edx\n");
>      }
>    else
>      fprintf (file, "\tjmp *%s\n", lazy_ptr_name);
> 
> in the directory gcc-4.1.1/gcc/config/i386.  This is a workaround for a
> change in Apple's assembler (fixed in either gcc-4.1-branch or
> gcc-4.2-branch).  Without it, I think the resulting messages are only
> cosmetic, but they confuse the regression tests.
> 
> On my PowerPC box I just updated the CVS and did a make--it compiled,
> but I didn't try the tests.
> 
> Michael
> 

Hi Michael,

many thanks for running the build.  At least we know the recent fix
works on the compiler.  I'm having some success with the optimisation
switch - although I can still find failures on small code fragments
(StrLen currently breaks).  But SymbolTable.mod compiles - whether it
has compiled correctly has yet to be seen - but certainly the
optimised assembler for gm2/gm2-compiler/FifoQueue.mod looked correct.
GCC does a pretty awesome job during `unit_at_a_time' optimisation :-)

regards,
Gaius