areas for performance improvements

[rbuckland]

Is your feature request related to a problem? Please describe.

We notice that as our monorepo gets bigger, the time to generate a version, for some of our packages, is getting longer, or, at times is longer.

Today, generating a bump version took approx 40s.
I am unsure if this is related to the distance between the "last" tag, or monorepo and needing to filter out paths, or other.

Describe the solution you'd like
assist with investigation and provide feedback on ways to improve the performance.

Additional context

Our monorepo "produces" 7+ packages. Each package is prefixed on its tag (simple directory is tag)

tags/lib/mylib/v7.6.1 --> for lib/mylib
tags/apps/app123/v2.8.0 --> for apps/app123

Running convco looks like

❯ time convco version --prefix lib/mylib/v --paths lib/mylib --bump  
7.5.0
convco version --prefix lib/mylib/v --paths lib/mylib --bump  42.50s user 0.65s system 99% cpu 43.153 total

which takes approx 42s

Attached is analysis using flamegraph, (https://github.com/flamegraph-rs/flamegraph)
from the same run above.

• The last version tag was in december 15 2023.
• 575 commits in total between that tag and "today",
• of which 318 were directly related to this library (lib/mylib)

> git log --oneline lib/mylib/v7.4.0..HEAD | wc -l
575
> git log --oneline lib/mylib/v7.4.0..HEAD -- lib/mylib | wc -l
318

[hdevalke]

https://github.com/convco/convco/blob/master/src/git.rs#L167 could be faster if the for each stops at the first match.

What is the expected time? For example how long does a git log previous..HEAD -- path/ takes?

[hdevalke]

I was wrong, the code already stopped the for_each at the first path found.
I made a fix that will probably improve performance, but i am not sure about it.

Would you be able to test #195?
Or do you know a similar public repo I could test this?

I thought there is a second improvement possible by only doing 1 diff for version, but this will not be possible as we need to make 1 diff per commit to know if the path is included.

We know make a diff to before parsing the commit to a conventional commit. It might be useful to parse the commit first and filter on commits that are breaking or increment the version. This would skip useless diff calculations.

[hdevalke]

I tried to do some benchmarks, but i could not find a difference.