mirror of
https://github.com/go-gitea/gitea.git
synced 2026-08-08 08:57:28 +00:00
7733f1953f
## Issue Gitea fails to dump LFS (and other object-storage) files when Azure Blob Storage is configured as the storage backend. The dump reports: Failed to dump LFS objects: /file/path: copying contents: file does not exist This happens with any non-empty base path (the default for LFS storage), which is why the user could only work around it by using `--skip-*` flags. The root cause is in `AzureBlobStorage.IterateObjects()`: Azure's list API already returns each blob's name including the configured base path, but the code was building the read client by running that name through the base-path-prepending helper a second time. This doubled the base path (e.g. `gitea-lfs/gitea-lfs/aa/bb/hash`), pointing at a blob that doesn't exist. `Stat()` still succeeded because it doesn't touch the network, so the failure only surfaced when the dumper actually tried to read the object's contents. ## Solution Add `getBlobClientByFullName()`, which builds a blob client from a name that is already fully qualified, without re-applying `buildAzureBlobPath()`. `IterateObjects()` now uses it for names obtained from Azure's list API. `getBlobClient()` (used by `Open`, `Stat`, `Delete`, `ServeDirectURL`, which take relative paths) is unchanged in behavior. Also add `TestAzureBlobStorageDumpArchive`, a regression test that drives the real dump path (`IterateObjects` → `Stat` → `dump.Dumper.AddFileByReader` → `mholt/archives` zip writer) against a **non-empty** `BasePath`, and verifies the produced archive contains the object with the correct content. The existing Azure tests use an empty `BasePath` and never read object content via `IterateObjects`, which is why they didn't catch this. Fixes #35476 --------- Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>