Preserve against garbage collection in LibGit2

Internals & Design
1

Maybe I am wrong but I have a concern about the implementation of LibGit2 regarding garbage collection. To ilustrate this, please have a look at the code of a simple function such as:

function Base.length(blob::GitBlob)
    ensure_initialized()
    return ccall((:git_blob_rawsize, libgit2), Int64, (Ptr{Cvoid},), blob.ptr)
end

in stdlib/LibGit2/src/blob.jl. My understanding is that there is nothing above to prevent blob to be garbage collected. This could be done by:

function Base.length(blob::GitBlob)
    ensure_initialized()
    GC.@preserve blob begin
        return ccall((:git_blob_rawsize, libgit2), Int64, (Ptr{Cvoid},), blob.ptr)
    end
end

Thanks to the specialization of the Base.unsafe_convert method for Git objects in stdlib/LibGit2/src/types.jl, another (better) possibility would be to just replace blob.ptr by blob:

Base.unsafe
function Base.length(blob::GitBlob)
    ensure_initialized()
    return ccall((:git_blob_rawsize, libgit2), Int64, (Ptr{Cvoid},), blob)
end

The same remark applies for other types of Git objects and other methods. If I am right, there are a number of ccalls in stdlib/LibGit2/src that could benefit from this (small) change.

1 Like
2

Yeah, the libgit2 wrappers weren’t written particularly carefully. Most of these sorts of issues that caused practical problems in CI’d usage were fixed, but there’s others remaining. PRs welcome.

3

OK I’d be happy to make a PR but do you confirm my analysis and the proposed fix? I do not want to introduce other errors.

4

You are correct.

5

The PR is here: Preserve Git objects from being garbage collected by emmt · Pull Request #55142 · JuliaLang/julia · GitHub

There are many changes and I have a few doubts (indicated by FIXME comments) but all tests in LibGit2 were successul.

9 Likes