Support git-lfs (large file storage)

[cgarling]

This adds the path to the git-lfs binary provided by Git_LFS_jll.jl to the PATH in the environment of the command returned by Git.git(). This enables working with repositories that use the Git LFS protocol.

I wrote a test case (which does work) but the test repo it clones is a bit big (~200 MB) so if anyone knows of / can create a smaller test repo with LFS files we can test, feel free to suggest a different one.

With respect to the tests, I think all github runners have git-lfs installed by default and so I don't know if we should be testing something to make sure we are using the correct git-lfs. Anecdotally I have a mac OS laptop that does not have git-lfs installed in its global path and I can confirm that this works on that machine.

Thanks to @giordano for providing advice on how to get started with this implementation.

[cgarling]

The tests I added are in the "Git.jl" testset, failures seem to occur in the OpenSSH tests that only run on CI. Looks like a permissions issue -- LMK if it requires any changes on my end.

[DilumAluthge]

Likely because CI_READONLY_DEPLOYKEY_FOR_CI_TESTSUITE_PRIVATEKE is not available on fork PRs.

@giordano We should set up GitHub Merge Queue on this repo.

[giordano]

Sounds like that test should be run only when the env var exists/is non-empty.

[cgarling]

I tried to fix the OpenSSH tests but I'm not having luck. It seems like it's detecting the env var and starting the OpenSSH tests despite my new check but failing when it actually tries to use the key, I think in this line

Git.jl/test/runtests.jl

Line 112 in bf81c75

proc = run(`$(git()) clone --depth=1 git@github.com:JuliaVersionControl/Git.jl.git`)

I'd like to get this merged and don't have much experience with these permissions issues. Could we merge this regardless of the CI issue? I can revert my attempts to fix it if you'd like.

[giordano]

Sounds like that test should be run only when the env var exists/IS NON-EMPTY.

☝️ Emphasis is mine

[giordano]

Also, doing this in a separate PR would be a good idea, since it's logically unrelated to this PR.

[giordano]

This is logically the same as b98ca06, not a great surprise this doesn't work either. My suggestion was

ssh_privkey = get(ENV, "CI_READONLY_DEPLOYKEY_FOR_CI_TESTSUITE_PRIVATEKEY", "")
if !isempty(ssh_privkey

Also, the other checks are useless at this point.

[codecov]

Codecov Report

❌ Patch coverage is 85.71429% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 95.00%. Comparing base (1201aed) to head (9610718).

Files with missing lines	Patch %	Lines
src/git_function.jl	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #64      +/-   ##
==========================================
- Coverage   96.96%   95.00%   -1.97%     
==========================================
  Files           1        1              
  Lines          33       40       +7     
==========================================
+ Hits           32       38       +6     
- Misses          1        2       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[cgarling]

Thanks for the tip! I implemented your suggestion and most of the tests are passing now. I'm not sure what's causing the "Julia min" tests to fail on windows, they seem to have passed the last time full CI was run (see here).

[mortenpi]

I think this might be incorrect? On Windows, this means that git will now use the system ENV, rather than the one set by the JLL, right? So it's missing the relevant JLL paths? Which might also be the reason for the CI failure?

I was just implementing my own version of this PR (as I hadn't notice this one 😅), and I ended up parsing git_cmd.env to ensure that I push to the correct path always. It's a bit noisy and ugly, but it was looking like this:

function _push_git_lfs_to_path!(git_cmd::Cmd)
    GitLFSExt = Base.get_extension(@__MODULE__, :GitLFSExt)
    if isnothing(GitLFSExt)
        throw(GitLFSNotLoadedError())
    end

    idx = findfirst(startswith("PATH="), git_cmd.env)
    path = if isnothing(idx)
        ""
    else
        # dropping the `PATH=` part
        git_cmd.env[idx][6:end]
    end
    path = split(get(ENV, LIBPATH_env, ""), path)
    pushfirst!(path, GitLFSExt.git_lfs_path())

    return addenv(git_cmd, "PATH" => join(path, pathsep))
end

A simpler approach might be to pull something like

path = vcat(dirname(Git_jll.git_path), split(get(ENV, "PATH", ""), pathsep))
libpath = vcat(Git_jll.LIBPATH_list, split(get(ENV, LIBPATH_env, ""), pathsep))

out of the if and conditionally only add the OpenSSH paths in the if? But I wasn't 100% sure that's correct either, so I opted for the more convoluted approach where I know I am only minimally changing the PATH variable.

[cgarling]

Ah yeah, I pulled path = split(get(ENV, "PATH", ""), pathsep) out of a block that only runs if the system is not windows so it would make sense I could have missed something like this. The part that confused me was why it passed on the Windows runs with more recent Julia versions and failed on the Julia 1.6 run.

I'll try to give this another try tonight or tomorrow. If you see an easy solution, you're welcome to submit a PR against my branch here if you'd like.

[cgarling]

Hey @mortenpi, thanks for the suggestion! I think I have something working based on your solution that I'm about to push to see if it fixes the CI errors, but I'm not sure what your line

path = split(get(ENV, LIBPATH_env, ""), path)

is supposed to do and it was causing errors so I removed it. LMK of any other changes you would suggest.

Edit: new commit did not fix CI errors -.-

[giordano]

Also, doing this in a separate PR would be a good idea, since it's logically unrelated to this PR.

☝️

[cgarling]

Ok, let's fix the CI for external PRs first and then we can revisit this. See #66.

[mortenpi]

I was able to repro this (on a Windows machine) locally with Julia 1.6.0, but the tests pass with 1.6.7. I'll try to figure out what actually is failing (and if it's a bug that got fixed), but in the meanwhile, any thoughts on bumping minimum Julia to 1.6.X or even 1.10.0?

[giordano]

Sounds OK to me, @DilumAluthge?

[mortenpi]

This looks like and MWE (for 1.6.0). This works on 1.6.7. So no fault of this PR as far as I can tell.

julia> addenv(`git status`, "PATH" => ENV["PATH"]) |> run;
On branch git-lfs
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
        modified:   Project.toml
        modified:   src/git_function.jl

no changes added to commit (use "git add" and/or "git commit -a")

julia> addenv(addenv(`git status`, "PATH" => ENV["PATH"]), "PATH" => ENV["PATH"]) |> run;
ERROR: IOError: could not spawn setenv(`git status`,["PATH=C:\\Windows\\System32;C:\\Windows;C:\\Windows\\System32\\wbem;C:\\Windows\\System32\\WindowsPowerShell\\v1.0\\;C:\\Windows\\System32\\OpenSSH\\;C:\\Program Files\\dotnet\\;C:\\Program Files (x86)\\IDEMIA\\AWP\\DLLs;C:\\Program Files\\IDEMIA\\AWP\\DLLs;C:\\Program Files (x86)\\NVIDIA Corporation\\PhysX\\Common;C:\\Program Files (x86)\\GnuPG\\bin;C:\\Program Files\\Git\\cmd;C:\\Program Files\\WireGuard\\;C:\\Program Files\\NVIDIA Corporation\\NVIDIA app\\NvDLISR;C:\\Program Files\\GitHub CLI\\;C:\\Users\\morte\\AppData\\Local\\Microsoft\\WindowsApps;;C:\\Users\\morte\\AppData\\Local\\Programs\\Microsoft VS Code\\bin;C:\\texlive\\2024\\bin\\windows;C:\\Users\\morte\\AppData\\Local\\Programs\\cursor\\resources\\app\\bin;c:\\Users\\morte\\AppData\\Roaming\\Code\\User\\globalStorage\\github.copilot-chat\\debugCommand", "USERDOMAIN_ROAMINGPROFILE=RAZA", "HOMEPATH=\\Users\\morte", "VSCODE_INJECTION=1", "PATHEXT=.COM;.EXE;.BAT;.CMD;.VBS;.VBE;.JS;.JSE;.WSF;.WSH;.MSC;.CPL", "SESSIONNAME=Console", "VSCODE_NONCE=4a49bbe4-b463-4de9-b912-0560883f4031", "SYSTEMROOT=C:\\WINDOWS", "APPDATA=C:\\Users\\morte\\AppData\\Roaming", "PSMODULEPATH=C:\\Users\\morte\\Documents\\WindowsPowerShell\\Modules;C:\\Program Files\\WindowsPowerShell\\Modules;C:\\WINDOWS\\system32\\WindowsPowerShell\\v1.0\\Modules", "COMMONPROGRAMW6432=C:\\Program Files\\Common Files", "PROGRAMDATA=C:\\ProgramData", "PUBLIC=C:\\Users\\Public", "USERDOMAIN=RAZA", "OS=Windows_NT", "PROCESSOR_REVISION=a502", "TMP=C:\\Users\\morte\\AppData\\Local\\Temp", "COMSPEC=C:\\WINDOWS\\system32\\cmd.exe", "ALLUSERSPROFILE=C:\\ProgramData", "COMPUTERNAME=RAZA", "USERNAME=mortenpi", "VSCODE_GIT_ASKPASS_NODE=C:\\Users\\morte\\AppData\\Local\\Programs\\Microsoft VS Code\\Code.exe", "VSCODE_GIT_ASKPASS_EXTRA_ARGS=", "VSCODE_GIT_IPC_HANDLE=\\\\.\\pipe\\vscode-git-1f4de4c6bb-sock", "ONEDRIVECONSUMER=C:\\Users\\morte\\OneDrive", "USERPROFILE=C:\\Users\\morte", "PROCESSOR_LEVEL=6", "PROGRAMW6432=C:\\Program Files", "TEMP=C:\\Users\\morte\\AppData\\Local\\Temp", "HOMEDRIVE=C:", "COLORTERM=truecolor", "WINDIR=C:\\WINDOWS", "LOCALAPPDATA=C:\\Users\\morte\\AppData\\Local", "PROCESSOR_IDENTIFIER=Intel64 Family 6 Model 165 Stepping 2, GenuineIntel", "NUMBER_OF_PROCESSORS=16", "GIT_ASKPASS=c:\\Users\\morte\\AppData\\Local\\Programs\\Microsoft VS Code\\resources\\app\\extensions\\git\\dist\\askpass.sh", "EFC_8040_1592913036=1", "LANG=en_US.UTF-8", "COMMONPROGRAMFILES(X86)=C:\\Program Files (x86)\\Common Files", "TERM_PROGRAM_VERSION=1.102.1", "COMMONPROGRAMFILES=C:\\Program Files\\Common Files", "ONEDRIVE=C:\\Users\\morte\\OneDrive", "TERM_PROGRAM=vscode", "PROGRAMFILES(X86)=C:\\Program Files (x86)", "VSCODE_GIT_ASKPASS_MAIN=c:\\Users\\morte\\AppData\\Local\\Programs\\Microsoft VS Code\\resources\\app\\extensions\\git\\dist\\askpass-main.js", "OPENBLAS_NUM_THREADS=8", "PROGRAMFILES=C:\\Program Files", "CHROME_CRASHPAD_PIPE_NAME=\\\\.\\pipe\\crashpad_5848_YSKAVRUJGEMCUVJU", "LOGONSERVER=\\\\RAZA", "DRIVERDATA=C:\\Windows\\System32\\Drivers\\DriverData", "ORIGINAL_XDG_CURRENT_DESKTOP=undefined", "VSCODE_STABLE=1", "SYSTEMDRIVE=C:", "=C:", "PROCESSOR_ARCHITECTURE=AMD64", "OPENBLAS_MAIN_FREE=1"]): invalid argument (EINVAL)
Stacktrace:
 [1] _spawn_primitive(file::String, cmd::Cmd, stdio::Vector{Any})
   @ Base .\process.jl:99
 [2] #637
   @ .\process.jl:112 [inlined]
 [3] setup_stdios(f::Base.var"#637#638"{Cmd}, stdios::Vector{Any})
   @ Base .\process.jl:196
 [4] _spawn
   @ .\process.jl:111 [inlined]
 [5] run(::Cmd; wait::Bool)
   @ Base .\process.jl:439
 [6] run
   @ .\process.jl:438 [inlined]
 [7] |>(x::Cmd, f::typeof(run))
   @ Base .\operators.jl:858
 [8] top-level scope
   @ REPL[17]:1

[cgarling]

Following from #67 (thanks for the help @mortenpi), julia = "~1.6.6, 1.7.3" is working on this PR but it looks like no CI is run against 1.7.3. Should we add something to the CI matrix to make sure we test against both 1.6.6 and 1.7.3?

[DilumAluthge]

Should we add something to the CI matrix to make sure we test against both 1.6.6 and 1.7.3?

Yes, that sounds good.

[cgarling]

Okay I added a single job on windows-latest with Julia 1.7.3 which passes (see here). It doesn't seem worth it to add 1.7.3 to the full CI matrix since we only ever had issues on Windows, but LMK if you'd like to have it in the matrix so we run Julia 1.7.3 against all the other OSes as well.