On 3 December 2013 10:08, <lucio@proxima.alt.za> wrote:

> I seem to remember
> Charles suggesting that it is difficult to do.  And as it caught a
> rather embarrassing case of sloppy programming
>

It's not sloppy programming, but an underlying assumption about RISC
machines in the (then) future
having a respectable number of (truly) general-purpose registers, and the
x86
does not. The S-U numbers guide register allocation, but there is no
spilling code,
except for nested function calls, and a few special cases that don't use
the S-U numbers.
Given that the out of fixed registers diagnostic occurs rarely in the
normal source,
and the S-U numbers might or might not be strictly accurate, the potential
disadvantage
of introducing inefficiency in existing compiled code by spill code
misguided by inaccurate
estimates made it seem not worthwhile at the time.