Re: [PATCH v2 0/2] Lock and Pointer guards

From: Valentin Schneider
Date: Tue May 30 2023 - 09:59:18 EST

On 30/05/23 11:23, Peter Zijlstra wrote:
> On Sat, May 27, 2023 at 12:18:04PM -0700, Linus Torvalds wrote:
>> And it turns out that the above two trivial macros are actually quite
>> useful in themselves. You want to do an auto-cleanup version of
>> 'struct fd'? It's trivial:
>> /* Trivial "getfd()" wrapper */
>> static inline void release_fd(struct fd *fd)
>> { fdput(*fd); }
>> #define auto_getfd(name, n) \
>> auto_release_name(struct fd, name, fdget(n), release_fd)
>> - I think the above is simpler and objectively better in every way
>> from the explicitly scoped thing
> Well, I think having that as a option would still be very nice.

IMO the explicit scoping can help with readability. It gives a clear visual
indication of where critical sections are, and you can break it up with a
scope + guards as in migrate_swap_stop() to stay at sane indentation
levels (with Python context managers, this would all be one scope).

I'd argue that for these, the scope/indentation is beneficial and not just
a byproduct. Even for longer functions like try_to_wake_up(), this works
out alright.

This obviously falls apart when dealing with too many guards
(e.g. copy_process()) or if the resulting indentation is nuts, but I concur
that keeping the explicit scope as an option would be nice.