bug#65491: [PATCH] Improve performance allocating vectors

[Paul Eggert]

On 2023-09-16 22:18, Eli Zaretskii wrote:

> It seems to me that in a 32-bit build with wide
> ints just
>
>    #define XUNTAG(a, type, ctype) ((ctype *) XLP (a))
>
> should be enough, since XLP yields a 'void *', no?

That would complicate the code unnecessarily. Let's not go there. Patch 
P works as-is, and it's simpler. Let's do that instead.


> what I
> meant was to have a separate definition of XUNTAG for 32-bit builds
> with wide ints (which could still remove the undefined behavior),

Yes, that's exactly what I suggest not to do. Why complicate the source 
code unnecessarily? And if we complicate it here, why not complicate it 
in similar ways in dozens of other places?

I went through a lot of this when adding support for --with-wide-int in 
the first place, years ago. When doing so, I strove to avoid having 
multiple copies of the code whenever I could. And I pretty much 
succeeded: there are only two WITH_WIDE_INT conditionals in lisp.h (and 
only three in other source files, all introduced relatively recently by 
others to work around compiler bugs, and all which should be rewritten 
without the #if).

It's an obvious win to have just one copy of the code instead of two, 
when one copy works and is just as efficient.  As much as possible, 
--with-wide-int should not be a special case.  We should not have 
"#ifdef WITH_WIDE_INT" scattered all over the place.  Keep it simple.