# Bug report: AM 3-way patching fails with partial cloning 2 messages from 2023-07-03 to 2023-07-04. Participants: Marius Schiffer, Junio C Hamano. Thread: https://gitlist.dev/t/59947 ## Marius Schiffer, 2023-07-03 16:46 Subject: Bug report: AM 3-way patching fails with partial cloning Message-ID: <1e79047e-944c-0112-ffd0-ca52e31226c3@mschiffer.de> URL: https://gitlist.dev/e/1e79047e-944c-0112-ffd0-ca52e31226c3%40mschiffer.de ``` When applying patches with git am on a repository cloned with filter=blob:none, the patching sometimes fails when it can't find the required SHA1 information. Applying the same thing on a normal clone works. I'm not sure yet what exactly i s missing that is not fetched, but I guess git should be able to fetch the missing blobs. I would also be very happy to hear workarounds, as I really love git partial cloning. Using git v2.41.0. To reproduce: 1. Download this patch: https://raw.githubusercontent.com/projectceladon/vendor-intel-utils/master/bsp_diff/common/kernel/linux-intel-lts2021/0007-9p-freezing-aborts-when-use-wait_event_killable.patch 2. Clone this repository: git clone --filter=blob:none https://github.com/projectceladon/linux-intel-lts2021 3. Apply the patch: git am -3 ../0007-9p-freezing-aborts-when-use-wait_event_killable.patch -- Marius ``` ## Junio C Hamano, 2023-07-04 22:02 Subject: Re: Bug report: AM 3-way patching fails with partial cloning Message-ID: URL: https://gitlist.dev/e/xmqqedln49nz.fsf%40gitster.g In-Reply-To: <1e79047e-944c-0112-ffd0-ca52e31226c3@mschiffer.de> ``` Marius Schiffer writes: > When applying patches with git am on a repository cloned with > filter=blob:none, the patching sometimes fails when it can't find the > required SHA1 information. > Applying the same thing on a normal clone works. This is very much expected, especially if your patches are not generated with --full-index option, because partial clones cannot on-demand request objects without knowing their full object names to the promissor remote. I do not offhand know if "am -3" codepath, if the full object name is available, is arranged to trigger lazy fetching of the objects, but nobody sane are sending patches with the --full-index option to begin with, so that is academic, I would say. ```