fix for df.c (was: Partial merge from new-regalloc-branch to HEAD)

[Michael Matz]

Hi,

On 22 Jun 2002, Daniel Egger wrote:

> > It seems so, doesn't it? ;-)  But it isn't.  Look at
> > df_insn_table_realloc().
>
> What about a more clearer
>
> while (uid >= df->insn_size)
>   df_insn_table_realloc (df, 0);

This would have fixed the potential problem.  But I believe the patch I
attached last time is better.  At the time of the calls we know exactly
how large the table must be (at least).  It's unlogical to then only
submit the difference to the called function instead of the absolute
value.  (in the meanwhile it bootstrapped on i686-linuxm and didn't cause
any regressions)


Ciao,
Michael.