On Wed, Dec 16, 2020 at 11:46:54AM -0500, Josef Bacik wrote: > My recent set of patches to reduce lock contention on the extent root by > running delayed refs resulted in a regression in generic/371.
As you write that often in patches, I want to point out that references to 'my recent patches doing something' are too vague and don't stand the test of time. Reading the patch in a year won't give me much clue what patches are that. Also as patches don't get merged in the order they've sent, 'recent' could be referring to something that would be in fact committed after this patch, making it more confusing and hard to find. If the patches are yet to be merged, referencing by subject should be sufficient, Filipe has been doing that and I find it really convenient to get more background to understand the changes or put them to context in case there's some dependency. > This test > fallocate()'s the fs until it's full, deletes all the files, and then > tries to fallocate() until full again. > > Before my delayed refs patches we would run all of the delayed refs And another one. If there's a hard patchset-to-patchset dependency and they get merged in that order, that's fine to refer to it like that and we've been doing that. The difference is that the patches are serialized in git and going back in git history will end up in the referred patches eventually. Unlike when it's still in the mailinglist, but still a more specific reference is better.
