Support Git sparse-index sdir required index extension

Inizia dal punto di ingresso git_repository_index e riproduci il fallimento usando i comandi shell di sparse-index e il consumer C forniti. Traccia il modo in cui il lettore dell’indice gestisce l’estensione sdir richiesta; il lavoro è completato quando il consumer apre correttamente l’indice, preservando il comportamento corretto dell’indice per le voci Sparse-Directory.

Version tested

Repro'd against libgit2 main:

• Repository: https://github.com/libgit2/libgit2
• git describe --tags --always --dirty: d23f18f
• Commit: d23f18fd881b8c5bb5e7a7abfcbc71ddf0c9a5ea
• Build environment: Windows, Visual Studio 2022 compiler, CMake + Ninja from Visual Studio

Summary

libgit2 fails to open a Git index that contains Git's sparse-index sdir required extension.

Observed error from git_repository_index on main:

git_repository_index failed: unsupported mandatory extension: 'sdir'

The same repository works with the Git CLI. The failure occurs when libgit2 reads the repository index.

This matters for large monorepos that rely on cone-mode sparse checkout plus sparse-index. Temporarily disabling sparse-index works around the failure, but it can be expensive because the index expands to include all paths at HEAD instead of scaling with the sparse-checkout cone.

Minimal repro steps

The following creates a small repository whose index contains sparse-directory entries and the sdir extension:

git init sdir-repro
cd sdir-repro
git config user.name 'Sparse Index Repro'
git config user.email 'sparse-index-repro@example.invalid'

printf 'initial\n' > initial.txt
git add initial.txt
git commit -m 'initial'

for i in $(seq 1 200); do
  mkdir -p "src/dir$i"
  printf 'base %s\n' "$i" > "src/dir$i/file$i.txt"
done

git add src
git commit -m 'seed sparse base tree'

git checkout -b feature/sparse-index
printf 'base 1\nfeature line\n' > src/dir1/file1.txt
git commit -am 'feature dir1'

git sparse-checkout init --cone --sparse-index
git sparse-checkout set src/dir1

printf 'base 1\nfeature line\nreview fix\n' > src/dir1/file1.txt
git add src/dir1/file1.txt

Confirm that the index contains sdir:

python - <<'PY'
from pathlib import Path
print(b'sdir' in Path('.git/index').read_bytes())
PY

This prints:

True

Minimal libgit2 reproducer

From inside the repro repository above, a minimal libgit2 consumer only needs to open/read the index:

#include <git2.h>
#include <stdio.h>

int main(void)
{
    git_repository *repo = NULL;
    git_index *index = NULL;
    const git_error *err;
    int rc;

    git_libgit2_init();

    rc = git_repository_open_ext(&repo, ".", 0, NULL);
    if (rc < 0) {
        err = git_error_last();
        fprintf(stderr, "git_repository_open_ext failed: %s\n",
            err ? err->message : "unknown error");
        git_libgit2_shutdown();
        return 1;
    }

    rc = git_repository_index(&index, repo);
    if (rc < 0) {
        err = git_error_last();
        fprintf(stderr, "git_repository_index failed: %s\n",
            err ? err->message : "unknown error");
        git_repository_free(repo);
        git_libgit2_shutdown();
        return 1;
    }

    printf("index entries: %u\n", (unsigned)git_index_entrycount(index));
    git_index_free(index);
    git_repository_free(repo);
    git_libgit2_shutdown();
    return 0;
}

Actual result on main

The repro repository contains sdir, the C reproducer compiles, and running it against libgit2 main fails:

indexContainsSdir=True
compileExit=0
git_repository_index failed: unsupported mandatory extension: 'sdir'
reproExit=1

Expected behavior

libgit2 should support reading sparse-indexes containing the required sdir extension, either by safely expanding sparse-directory entries to a full in-memory index where necessary or by making the relevant index APIs sparse-index aware.

Current workaround

Temporarily disable sparse-index, run the libgit2-backed tool, then restore sparse-index:

git sparse-checkout reapply --no-sparse-index
# run libgit2-backed tool
git sparse-checkout reapply --sparse-index

This preserves sparse-checkout patterns, but it can be costly in very large monorepos.

Related references

• Git sparse-index documentation: https://git-scm.com/docs/sparse-index
• GitHub sparse-index background: https://github.blog/open-source/git/make-your-monorepo-feel-small-with-gits-sparse-index/
• Existing broad sparse-checkout issue: https://github.com/libgit2/libgit2/issues/2263