Re: [PATCH v2 1/8] minmax: Put all the clamp() definitions together

From: Linus Torvalds
Date: Sun Jul 28 2024 - 13:24:58 EST


On Sun, 28 Jul 2024 at 07:18, David Laight <David.Laight@xxxxxxxxxx> wrote:
>
> +#define min_t(type, x, y) __careful_cmp(min, (type)(x), (type)(y))
> +#define max_t(type, x, y) __careful_cmp(max, (type)(x), (type)(y))

This is unrelated to your patch, but since it moves things around and
touches these, I reacted to it..

We should *not* use __careful_cmp() here.

Why? Because part of __careful_cmp() is the "only use arguments once".

But *another* part of __careful_cmp() is "be careful about the types"
in __cmp_once().

And being careful about the types is what causes horrendous expansion,
and is pointless when we just forced things to be the same type.

So we should split __careful_cmp() into one that does just the "do
once" and one that then also does the type checking.

But I think even if we don't do that, I wonder if we can just do this:

#define __cmp_once(op, x, y, unique_x, unique_y) ({ \
typeof(x) unique_x = (x); \
typeof(y) unique_y = (y); \
static_assert(__types_ok(x, y), \
...

and change it to

#define __cmp_once(op, x, y, unique_x, unique_y) ({ \
__auto_type unique_x = (x); \
__auto_type unique_y = (y); \
static_assert(__types_ok(unique_x, unique_y), \
...

because while that may screw up the "constant integer" case (because
it now goes through that "unique_XY" variable, maybe it doesn't? At
least gcc has been known to deal with things like arguments to inline
functions well enough (ie a constant argument means that the arguments
shows as __builtin_constant_p(), and we already depend on that).

That single change would cut down on duplication of 'x' and 'y'
_enormously_. No?

(You already did the __auto_type part elsewhere)

Note that this would require the more relaxed "__is_noneg_int()" that
I suggested that allows for any expression, not just C constant
expressions)

Linus