feat: improved performances when checking for conflicts on pull requests (#7727)
Some checks are pending
/ release (push) Waiting to run
testing / backend-checks (push) Waiting to run
testing / frontend-checks (push) Waiting to run
testing / test-unit (push) Blocked by required conditions
testing / test-e2e (push) Blocked by required conditions
testing / test-remote-cacher (redis) (push) Blocked by required conditions
testing / test-remote-cacher (valkey) (push) Blocked by required conditions
testing / test-remote-cacher (garnet) (push) Blocked by required conditions
testing / test-remote-cacher (redict) (push) Blocked by required conditions
testing / test-mysql (push) Blocked by required conditions
testing / test-pgsql (push) Blocked by required conditions
testing / test-sqlite (push) Blocked by required conditions
testing / security-check (push) Blocked by required conditions

- `testPatch` is a function that is called to test a pull request and determine the state of the pull request. Checking for merge conflicts, check if the diff is empty and if the pull request modifies any protected files.
- The checking for merge conflict and if the diff is empty used git commands that relied on a working tree to correctly functions. Forgejo store repositories in a bare format which do not contain a working tree. This means that a temporary copy was created every time a pull request had to be re-checked and for large repositories involving quite some I/O interaction.
- This patch adjusts those codepaths to instead use newer Git plumbing commands that work without requiring a work tree and can thus be used directly on the bare repository. The merge conflict is now done via [`git-merge-tree(1)`](https://git-scm.com/docs/git-merge-tree/) and checking if the diff is empty is done via [`git-diff-tree(1)`](https://git-scm.com/docs/git-diff-tree).
- If the function is called to test a patch where the head and base repository are not the same, then [Git alternate](https://git-scm.com/docs/gitglossary#Documentation/gitglossary.txt-aiddefalternateobjectdatabaseaalternateobjectdatabase) is used to make the head commit available in the base repository, this done on a per git command basis via the `GIT_ALTERNATE_OBJECT_DIRECTORIES` environment.
- As far as I can understand the documentation and the existing code, there's no edge case that the new code cannot handle. It also results in a cleaner codepath, as the existing code did a lot of checking and merging in a more traditional approach that required a lot of (parsing) code, while the new code offloads this to git and has a trivial parser of the output.
- Resolves forgejo/forgejo#7701
- Added exhaustive integration testing.

Reviewed-on: https://codeberg.org/forgejo/forgejo/pulls/7727
Reviewed-by: Earl Warren <earl-warren@noreply.codeberg.org>
Reviewed-by: Otto <otto@codeberg.org>
Co-authored-by: Gusted <postmaster@gusted.xyz>
Co-committed-by: Gusted <postmaster@gusted.xyz>
This commit is contained in:
Gusted 2025-05-16 12:40:38 +00:00 committed by Gusted
parent 2ab5b585f6
commit 7c150be23d
13 changed files with 645 additions and 156 deletions

View file

@ -37,6 +37,7 @@ var (
SupportHashSha256 bool // >= 2.42, SHA-256 repositories no longer an experimental curiosity
InvertedGitFlushEnv bool // 2.43.1
SupportCheckAttrOnBare bool // >= 2.40
SupportGitMergeTree bool // >= 2.38
HasSSHExecutable bool
@ -189,6 +190,7 @@ func InitFull(ctx context.Context) (err error) {
}
InvertedGitFlushEnv = CheckGitVersionEqual("2.43.1") == nil
SupportGitMergeTree = CheckGitVersionAtLeast("2.38") == nil
if setting.LFS.StartServer {
if CheckGitVersionAtLeast("2.1.2") != nil {