fix(vcs): select the root directory a clone command was given

[HarshMN2345]

The bug

generateCloneCommand() writes the root directory straight into .git/info/sparse-checkout, where git matches it gitignore-style. A ./ prefix looks for a directory literally named ., so the clone checks out nothing.

Reproduced against real git — bare repo, core.sparseCheckout, depth-1 pull:

pattern docs     -> checked out: [docs]
pattern ./docs   -> checked out: []
pattern .        -> checked out: []
pattern ./       -> checked out: []

. and ./ are the conventional spellings of the repository root, and a caller reading a path from a directory picker or config file routinely holds the ./ form. The failure is silent at clone time and surfaces later as a missing entrypoint — npm ENOENT ... package.json and the like.

Why it slipped through

Each adapter tested for the root itself, in two different shapes, and neither handled . or ./:

adapter	test
GitHub.php:1192, Gitea.php:1060, Origin.php:894	$rootDirectory === '' || $rootDirectory === '0'
GitLab.php:981, Bitbucket.php:1508	\in_array($rootDirectory, ['', '0', '/'], true)

Meanwhile the package already answers this exact question correctly one file over: Git::normalizeRepositoryPath() has resolved these sentinels for the contents APIs since c2c60ae ("Resolve path sentinels the same way on every provider … a caller passing '.', './' or 'src//' gets the same answer everywhere"). It was applied to the 11 contents call sites and not to the one clone call site.

The change

Each adapter resolves the root directory through normalizeRepositoryPath() first, and its root test becomes === ''. That test is explicit rather than ?:, because '0' is a real directory name and PHP treats it as falsy.

'0' goes with it. That was an empty() leftover — a repository whose top-level directory is named 0 was treated as root and cloned whole. It now checks out that directory. Behaviour change, deliberate, called out in case anyone disagrees.

'/' still resolves to * on every adapter, so GitLab and Bitbucket keep their existing handling and the other three gain it.

Tests

generateCloneCommand needs no provider, so the shared unit base covers it for every adapter — 60 cases, 6 adapters, over '', ., ./, /, docs, docs/, ./docs, ./docs/, ./astro/starter, and 0.

Building a clone command in the shared base also means the Gitea-family adapters must carry an endpoint, as every other operation on them already assumes — Gitea, Gogs and Forgejo test factories now call setEndpoint().

phpunit --testsuite unit: 145 passing. pint: clean. phpstan: no errors. bin/monorepo validate: all packages valid.

Context

Found while tracing a root-directory bug in Appwrite (appwrite/appwrite#13574). Appwrite does not hit this today only because its builds worker canonicalizes one line before the call — other consumers are exposed.

[greptile-apps]

Confidence Score: 5/5

The production change appears safe to merge, though the prior non-blocking request for an observable clone test remains unresolved.

Findings

1.  Test Mirrors Command Text ▶

Fix with agent prompt

### Issue 1
packages/vcs/tests/Unit/Base.php:undefined-232
This test checks only that the shell-escaped pattern occurs somewhere in the generated command. It can therefore pass if the pattern is written to the wrong file, duplicated, or later overridden, and it never verifies the checkout tree that this regression actually broke. This violates the repository directive to test observable behavior instead of mirroring source or configuration. The requirement must be satisfied before merging by exercising the command against a local Git fixture and asserting the resulting files.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

• Treats ., ./, /, and the empty path as repository-root selections.
• Preserves 0 as a valid directory name rather than a root sentinel.
• Configures Gitea-family test adapters with the endpoint required to generate clone commands.
• The previous implementation-coupled test finding remains outstanding because the test still checks command text rather than observable checkout behavior.

_{Reviews (2) · Last reviewed commit: "refactor(vcs): resolve the clone root di..."}

[greptile-apps]

Test Mirrors Command Text

This test checks only that the shell-escaped pattern occurs somewhere in the generated command. It can therefore pass if the pattern is written to the wrong file, duplicated, or later overridden, and it never verifies the checkout tree that this regression actually broke. This violates the repository directive to test observable behavior instead of mirroring source or configuration. The requirement must be satisfied before merging by exercising the command against a local Git fixture and asserting the resulting files.

Context Used: Call out and harshly judge implementation-coupled ... (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: packages/vcs/tests/Unit/Base.php
Line: 232

Comment:
**Test Mirrors Command Text**

This test checks only that the shell-escaped pattern occurs somewhere in the generated command. It can therefore pass if the pattern is written to the wrong file, duplicated, or later overridden, and it never verifies the checkout tree that this regression actually broke. This violates the repository directive to test observable behavior instead of mirroring source or configuration. The requirement must be satisfied before merging by exercising the command against a local Git fixture and asserting the resulting files.

**Context Used:** Call out and harshly judge implementation-coupled ... ([source](https://app.greptile.com/review/custom-context?memory=instruction-0))

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!