From: Jeff King <peff@peff.net>
Date: Fri, 20 Mar 2026 06:18:52 GMT
Subject: Re: [RFC] cocci: .buf in a strbuf object can never be NULL
Message-ID: <20260320061852.GA35538@coredump.intra.peff.net>
In-Reply-To: <20260320055709.GA35291@coredump.intra.peff.net>

On Fri, Mar 20, 2026 at 01:57:09AM -0400, Jeff King wrote:

> I haven't measured to see what the exact cost is, but I know that
> looping over a strbuf (with a reset in the loop, or the implied reset
> from a getline call) is a common optimization trick that does have a
> measurable improvement for some cases.

Try something like this:

  # input file is all objects in linux.git
  cd linux.git
  git repack -ad
  git show-index <.git/objects/pack/pack-*.idx | cut -d' ' -f2 >input

  # now do something that primarily reads a bunch of lines
  git cat-file --buffer --batch-check='%(objectname)" <input

That's a somewhat silly command, though it does do something useful (it
checks that each object exists). Here is master ("git.old") versus
applying your patch ("git.new"):

  Benchmark 1: ./git.old cat-file --buffer --batch-check="%(objectname)" <input
    Time (mean ± σ):      3.152 s ±  0.057 s    [User: 3.067 s, System: 0.085 s]
    Range (min … max):    3.048 s …  3.225 s    10 runs
  
  Benchmark 2: ./git.new cat-file --buffer --batch-check="%(objectname)" <input
    Time (mean ± σ):      3.377 s ±  0.065 s    [User: 3.295 s, System: 0.082 s]
    Range (min … max):    3.279 s …  3.480 s    10 runs
  
  Summary
    ./git.old cat-file --buffer --batch-check="%(objectname)" <input ran
      1.07 ± 0.03 times faster than ./git.new cat-file --buffer --batch-check="%(objectname)" <input

That's a fairly extreme example, but I think shows that the extra
allocations do have measurable overhead. If you ask it do more work
(asking for %(objecttype) or something) the relative change becomes
smaller, but the absolute slowdown (a few hundred ms) remains.

-Peff