[prev in list] [next in list] [prev in thread] [next in thread] 

List:       git
Subject:    Re: [PATCH v3 0/3]
From:       Ævar Arnfjörð Bjarmason <avarab () gmail ! com>
Date:       2021-06-30 18:00:05
Message-ID: 87y2ardygj.fsf () evledraar ! gmail ! com
[Download RAW message or body]


On Wed, Jun 30 2021, Jeff King wrote:

> On Wed, Jun 30, 2021 at 01:34:07PM -0400, Jeff King wrote:
>
>> As an aside, having to have a separate bundle_header_init() and
>> BUNDLE_HEADER_INIT is annoying (because they both must be kept up to
>> date with each other), but quite common in our code base. I wonder if
>> writing:
>> 
>>   void bundle_header_init(struct bundle_header *header)
>>   {
>> 	struct bundle_header blank = BUNDLE_HEADER_INIT;
>> 	memcpy(header, &blank, sizeof(*header));
>>   }
>> 
>> would let a smart enough compiler just init "header" in place without
>> the extra copy (the performance of a single bundle_header almost
>> certainly doesn't matter, but it might for other types).
>> 
>> Just musing. ;)
>
> For my own curiosity, the answer is yes: https://godbolt.org/z/s54dc6ss9
>
> With "gcc -O2" the memcpy goes away and we init "header" directly.
>
> If we want to start using this technique widely, I don't think it should
> be part of your series, though. This probably applies to quite a few
> data structures, so it would make more sense to have a series which
> converts several.

That's cool, yeah that would make quite a lot of code better. Thanks!
[prev in list] [next in list] [prev in thread] [next in thread]