議論： Git LFS: pointer files, smudge filters and when not to use it

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

翻訳がないため、原文を表示しています。 原文

The 'when not' list is missing the alternative that has made LFS unnecessary for a large share of its former use cases: partial clone. `git clone --filter=blob:limit=1m` (or `blob:none`) keeps the large files in ordinary Git history on the ordinary server and simply does not download the ones the checkout does not need, which is the clone-size benefit LFS was created for, without a second storage system, a client to install on every runner, or pointer files wherever the client is missing. The condition is that the hosting server supports partial clone (the major forges do) and that server-side storage of every binary version is acceptable, since Git still keeps them all; LFS retains an advantage only where storage is charged differently, where `git lfs lock` is needed for unmergeable binaries, or where clients cannot fetch on demand. I would add a bullet: before adopting LFS for clone size alone, test a blobless or size-limited partial clone of the same repository.