]> git.ipfire.org Git - thirdparty/git.git/commit
repack: add --path-walk option
authorDerrick Stolee <stolee@gmail.com>
Fri, 16 May 2025 18:11:57 +0000 (18:11 +0000)
committerJunio C Hamano <gitster@pobox.com>
Fri, 16 May 2025 19:15:39 +0000 (12:15 -0700)
commit5f711504d9540ef5c8ca4c9d43e6f5a1d0259d3c
treef1271c8225a00f782250a6c4bab5e434030f3a15
parent6e95bf80b5066abca85277bd8c142bf36c5fbe0d
repack: add --path-walk option

Since 'git pack-objects' supports a --path-walk option, allow passing it
through in 'git repack'. This presents interesting testing opportunities for
comparing the different repacking strategies against each other.

Add the --path-walk option to the performance tests in p5313.

For the microsoft/fluentui repo [1] checked out at a specific commit [2],
the --path-walk tests in p5313 look like this:

Test                                                     this tree
-------------------------------------------------------------------------
5313.18: thin pack with --path-walk                      0.08(0.06+0.02)
5313.19: thin pack size with --path-walk                           18.4K
5313.20: big pack with --path-walk                       2.10(7.80+0.26)
5313.21: big pack size with --path-walk                            19.8M
5313.22: shallow fetch pack with --path-walk             1.62(3.38+0.17)
5313.23: shallow pack size with --path-walk                        33.6M
5313.24: repack with --path-walk                         81.29(96.08+0.71)
5313.25: repack size with --path-walk                             142.5M

[1] https://github.com/microsoft/fluentui
[2] e70848ebac1cd720875bccaa3026f4a9ed700e08

Along with the earlier tests in p5313, I'll instead reformat the
comparison as follows:

Repack Method    Pack Size       Time
---------------------------------------
Hash v1             439.4M      87.24s
Hash v2             161.7M      21.51s
Path Walk           142.5M      81.29s

There are a few things to notice here:

 1. The benefits of --name-hash-version=2 over --name-hash-version=1 are
    significant, but --path-walk still compresses better than that
    option.

 2. The --path-walk command is still using --name-hash-version=1 for the
    second pass of delta computation, using the increased name hash
    collisions as a potential method for opportunistic compression on
    top of the path-focused compression.

 3. The --path-walk algorithm is currently sequential and does not use
    multiple threads for delta compression. Threading will be
    implemented in a future change so the computation time will improve
    to better compete in this metric.

There are small benefits in size for my copy of the Git repository:

Repack Method    Pack Size       Time
---------------------------------------
Hash v1             248.8M      30.44s
Hash v2             249.0M      30.15s
Path Walk           213.2M     142.50s

As well as in the nodejs/node repository [3]:

Repack Method    Pack Size       Time
---------------------------------------
Hash v1             739.9M      71.18s
Hash v2             764.6M      67.82s
Path Walk           698.1M     208.10s

[3] https://github.com/nodejs/node

This benefit also repeats in my copy of the Linux kernel repository:

Repack Method    Pack Size       Time
---------------------------------------
Hash v1               2.5G     554.41s
Hash v2               2.5G     549.62s
Path Walk             2.2G    1562.36s

It is important to see that even when the repository shape does not have
many name-hash collisions, there is a slight space boost to be found
using this method.

As this repacking strategy was released in Git for Windows 2.47.0, some
users have reported cases where the --path-walk compression is slightly
worse than the --name-hash-version=2 option. In those cases, it may be
beneficial to combine the two options. However, there has not been a
released version of Git that has both options and I don't have access to
these repos for testing.

Signed-off-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Documentation/git-repack.adoc
builtin/repack.c
t/perf/p5313-pack-objects.sh