On Mon, May 20, 2024 at 01:23:26PM +1000, Stephen Rothwell wrote: > On Tue, 14 May 2024 23:57:36 +0100 Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote: > > > > At LSFMM we're talking about the need to do more integrated testing with > > the various fs trees, the fs infrastructure and the vfs. We'd like to > > avoid that testing be blocked by a bad patch in, say, a graphics driver. > > > > A solution we're kicking around would be for linux-next to include a > > 'fs-next' branch which contains the trees which have opted into this > > new branch. Would this be tremendously disruptive to your workflow or > > would this be an easy addition? > > How would this be different from what happens at the moment with all > the separate file system trees and the various "vfs" trees? I can > include any tree. What we were hoping for was that you would merge together the vfs, iomap, and various fs-specific trees (e.g., bcachefs, btrfs, ext4, f2fs, xfs, etc.) together, and then publish it as "fs-next". You could then use fs-next as something that would be merged into linux-next instead of the component fs trees, so hopefully it wouldn't be a significant amount of extra work for you. As Willy stated, the advantages of having an official daily "fs-next" tree is that multiple file system developers would be able to test the same branch and compare notes when regressions are found. And the advantage of fs-next versus the full linux-next is that it reduces the chances of tests getting blocked by non-fs-relevant changes. Cheers, - Ted
