Re: [PATCH 0/3] yacc: compute the best type for the state number

[Kaz Kylheku]

On 2019-10-01 06:53, Hans Åberg wrote:
> One should note that the unsigned types are required to be 2’s
> complement C/C++, unlike the signed ones, cf.

That is untrue by definition, since two's complement is strictly
a mechanism for representing negative values, which the unsigned
types do not do. They follow a "pure binary enumeration"
(I think that's the wording).

The unsigned types can *simulate* some aspects of two's complement
quite well. Unsigned and two's complement addition basically look
the same at the bit level; just the interpretation of the upper
two bits is different for determining sign and overflow. If you're
writing a software emulator for a two's complement CPU, you can use
unsigned for the basic ALU operations.

Code which uses unsigned numbers in a thoroughly disciplined way
to simulate two's complement math does not represent the typical
uses of the unsigned types in C programs.

When unsigned types are used to simulate signed, comparisons must
be done very carefully. The value UINT_MAX is representing -1
and so it has to compare less than zero. You need two write yourself
a a twocomp_less(x, y) function and consistently use it to get this
right. Most code that you will see in the wild that works with
unsigned numbers still uses the regular < and > operators, though.
(That's one way the bugs creep in.)

Unsigned multiplication and division also require careful handling
in order to correctly simulate two's complement.