Re: [PATCH] last-modified: implement faster algorithm

From: Toon Claes <toon@iotcl.com>
Date: Oct 21, 2025, 13:00 UTC
Message-ID: <878qh4tnop.fsf@iotcl.com>
In-Reply-To: <87jz0tu3yh.fsf@iotcl.com>

> Taylor Blau <me@ttaylorr.com> writes:

Show 5 quoted lines

>> I'd suggest including those benchmarks as well, and potentially running
>> them on linux.git, or another comparably larger open-source repository.
>> git.git is large enough to show some interesting behavior, but I always
>> have found it useful to compare the results against a larger repository
>> as well.

I've submitted v2 of my patch. In that version I included the benchmark for a repo without and with Bloom filters. But I didn't include benchmarks on linux.git. So for the record I'm sharing it here:

    Benchmark 1: master   no bloom
      Time (mean ± σ):     12.290 s ±  0.105 s    [User: 11.893 s, System: 0.370 s]
      Range (min … max):   12.089 s … 12.461 s    10 runs

    Benchmark 2: master with bloom
      Time (mean ± σ):      1.536 s ±  0.072 s    [User: 1.418 s, System: 0.114 s]
      Range (min … max):    1.500 s …  1.740 s    10 runs

    Benchmark 3: HEAD     no bloom
      Time (mean ± σ):     868.7 ms ±   3.0 ms    [User: 810.7 ms, System: 56.1 ms]
      Range (min … max):   863.1 ms … 873.5 ms    10 runs

    Benchmark 4: HEAD   with bloom
      Time (mean ± σ):     110.0 ms ±   1.8 ms    [User: 89.3 ms, System: 20.3 ms]
      Range (min … max):   106.6 ms … 114.2 ms    27 runs

    Summary
      HEAD   with bloom ran
        7.90 ± 0.13 times faster than HEAD     no bloom
       13.96 ± 0.69 times faster than master with bloom
      111.73 ± 2.02 times faster than master   no bloom

-- 
Cheers,
Toon