Re: master b0ba0d42b0f: * src/lisp.h (EQ): Improve generated code.

[Andrea Corallo <acorallo@gnu.org>]

Stefan Kangas <stefankangas@gmail.com> writes:

> Eli Zaretskii <eliz@gnu.org> writes:
>
>>> From: Mattias Engdegård <mattias.engdegard@gmail.com>
>>> Date: Thu, 28 Nov 2024 18:53:46 +0100
>>> Cc: Andrea Corallo <acorallo@gnu.org>,
>>>  Emacs Devel <emacs-devel@gnu.org>
>>>
>>> We should probably define some variant of likely/unlikely because
>>> those cover most needs of __builtin_expect, rather than using it
>>> directly. Then portability wouldn't be a problem.
>>
>> Portability is already not a problem, so I see no need for inventing
>> new macros for this purpose.
>
> Right, since this is already done for us by Gnulib. But the other
> reason to having such macros is readability.
>
> To be clear, the proposed macros would look something like this (copied
> here from the Linux kernel):
>
> #define likely(x)      __builtin_expect(!!(x), 1)
> #define unlikely(x)    __builtin_expect(!!(x), 0)
>
> Andrea's code would then read like so:
>
> -  return BASE_EQ ((__builtin_expect (symbols_with_pos_enabled, false)
> +  return BASE_EQ ((unlikely (symbols_with_pos_enabled)

Hi Stefan,

I think I'm with Eli in suggesting we should just use __builtin_expect.
Yes it reads a little longer but:

1- We define a layer of macros less

2- It's to me still very readable

3- In my mind introducing ad-hoc macros might encourage developers using
them, while I think developers should refrain using them unless they
really know what they are doing.  IOW annotating branchs should be the
rare exception rather than the rule, and for these rare cases
__builtin_expect is IMO fine.

  Andrea