[Feature Request] Using git library to extract git related information

[filzrev]

Is your feature request related to a problem? Please describe.
Current docfx v2 extracts git related information by invoking git command using CommandUtility.cs.
It takes about 30-60 milliseconds every calls to get file information.
and when it need to get additional information (like committer-date) it require extra command invoke costs.

Describe the solution you'd like
It seems docfx v3 using LibGit2 to get git information.
I hope these features can backport to v2. If it run faster than current implementation.

Describe alternatives you've considered

• Using libgit2sharp

[KalleOlaviNiemitalo]

FWIW, nbgv 3.6.133 ships with these unmanaged libraries:

Uncompressed size	Path
1125832	tools/net6.0/any/runtimes/linux-arm/native/libgit2-b7bad55.so
1447504	tools/net6.0/any/runtimes/linux-arm64/native/libgit2-b7bad55.so
1520212	tools/net6.0/any/runtimes/linux-musl-arm/native/libgit2-b7bad55.so
1619160	tools/net6.0/any/runtimes/linux-musl-arm64/native/libgit2-b7bad55.so
1663736	tools/net6.0/any/runtimes/linux-musl-x64/native/libgit2-b7bad55.so
1658528	tools/net6.0/any/runtimes/linux-x64/native/libgit2-b7bad55.so
1109816	tools/net6.0/any/runtimes/osx-arm64/native/libgit2-b7bad55.dylib
1183856	tools/net6.0/any/runtimes/osx-x64/native/libgit2-b7bad55.dylib
1429504	tools/net6.0/any/runtimes/win-arm64/native/git2-b7bad55.dll
1703424	tools/net6.0/any/runtimes/win-x64/native/git2-b7bad55.dll
1346048	tools/net6.0/any/runtimes/win-x86/native/git2-b7bad55.dll

The compressed sizes are approximately half of those. In the docfx tool package, I assume it would be possible to deduplicate the libraries across net6.0, net7.0, and net8.0.

Would there have to be a fallback implementation that executes git, in case someone runs DocFX on another architecture like RISC-V (dotnet/runtime#36748 in progress)?

Does the linking exception of the libgit2 license apply when libgit2 is distributed as a shared library rather than "linked into a combined executable"?

[filzrev]

It seems better to using fully managed version git reader library.

For example:

• https://www.nuget.org/packages/GitReader
• https://github.com/dotnet/Nerdbank.GitVersioning/tree/main/src/NerdBank.GitVersioning/ManagedGit

I thought following functionalities required to replace git command to library.

1. GetRepoRootCommand

Git Command: git rev-parse --show-toplevel

2. GetLocalBranchCommand

Git Command: git rev-parse --abbrev-ref HEAD

3. GetLocalBranchCommitIdCommand

Git Command:: git rev-parse HEAD

4. GetRemoteBranchCommand

GitCommand: git rev-parse --abbrev-ref @{u}

5. GetGitCommiterDate (Currently not included)

Git Command: git log -1 --format=%cI --no-show-signature {filename}

[yufeih]

It seems better to using fully managed version git reader library.

Fantastic!

[filzrev]

@yufeih
Is it able to re-open this issue?
I thought following problems are resolved at latest GitReader 1.6.0 release.

• Wrong remote URL is returned by GitReader
• Can't get information managed by git submodules.

Remaining Issues

• Mixed output path separators (/,\) chars problems ([Bug] Invalid Edit-this-page URL after update to docfx v2.72.0 #9337)
• Cache GitReader's PrimitiveRepository instance. (As far as I've confirmed it's safe to be called by multiple thread)

[yufeih]

GitReader has been replaced with a different implementation due to bugs in the library. We should fix the path problem.

[filzrev]

Can't get information managed by git submodules.

It seems current docfx implementation don't support git submodule repository. (Currently, it returns root repository URL)

It needs to implement extra gitdir: path lookup logics when directory contains git file.
I'll try to create PR that fix this issue later.

[filzrev]

Git submodule supports issue is resolved by PR(#9540).

My remaining issue is to get committer date (Expected to be used for Sitemap/RSS feed).

5. GetGitCommiterDate (Currently not included)

Git Command: git log -1 --format=%cI --no-show-signature {filename}

It seems complex to get this information from .git repository structure without library.
So I'm expected to get this information by using git.exe external executable call.