Re: [PATCH v3] packfile: freshen the mtime of packfile by configuration

Taylor Blau [off-list ref] writes:

Hmm. I'm still quite unconvinced that we should be taking this direction
without better motivation. We talked about your assumption that NFS
seems to be invalidating the block cache when updating the inodes that
point at those blocks, but I don't recall seeing further evidence.

Me neither.  Not touching the pack and not updating the "most
recently used" time of individual objects smells like a recipe
for repository corruption.

My opinion is that it is not, and that the bizarre caching behavior you
are seeing is out of Git's control.

2021年7月20日 08:07，Junio C Hamano [off-list ref] 写道：

Taylor Blau [off-list ref] writes:

quoted

Hmm. I'm still quite unconvinced that we should be taking this direction
without better motivation. We talked about your assumption that NFS
seems to be invalidating the block cache when updating the inodes that
point at those blocks, but I don't recall seeing further evidence.

Me neither.  Not touching the pack and not updating the "most
recently used" time of individual objects smells like a recipe
for repository corruption.

quoted

My opinion is that it is not, and that the bizarre caching behavior you
are seeing is out of Git's control.

Thanks Junio, I will try to get a more detial reports of the NFS caches and
share it if it is valuable. Not touching the mtime of packfiles really has
potencial problems just as Taylor said.