Re: [PATCH] hash: reduce size of algo member of object ID

From: Eric Wong <e@80x24.org>
Date: Oct 4, 2021, 07:31 UTC
Message-ID: <20211004073142.GA25608@dcvr>
In-Reply-To: <8da7bad2-b5a8-5aef-284b-dfa4e78556a9@web.de>

René Scharfe <l.s.r@web.de> wrote:

Show 6 quoted lines

> cf0983213c (hash: add an algo member to struct object_id, 2021-04-26)
> introduced the algo member as an int.  This increased the size of struct
> object_id by 4 bytes (12.5%) and imposed a 4-byte alignment.  Currently
> we only need to stored the values 0, 1 and 2 in it.  Let's use an
> unsigned char instead to reduce the size overhead and lift the alignment
> requirement.

I like it.  Btw, would a bitfield enum be portable enough for us?

	enum git_hash_algo algo:8;  /* or algo:2 */

I've used those in other projects, but AFAIK they were for gcc||clang-only.

> Not sure how to measure the performance impact of this change.  The perf
> tests mentioned by cf0983213c don't show much of a difference with
> GIT_PERF_REPEAT_COUNT=10 for me:

Thanks for that info. I'm not familiar enough with most git internals to know if it ought to make a difference as-is.

In my experience, changes like this open the door to further struct-packing opportunities (possibly combined with conditional use of __attribute__((packed)) ).