On Tue, Jul 2, 2013 at 7:07 AM, Greg Stein <gst...@gmail.com> wrote: > On Sun, Jun 30, 2013 at 6:48 AM, Stefan Fuhrmann > <stefan.fuhrm...@wandisco.com> wrote: > > Hi all, > > > > This is based on what was discussed in Berlin already. > > > > The goal is to get FSFS improvements reviewed & integrated > > into /trunk a.s.a.p. and to bring the code for the new backend > > to /trunk as well and continue development there. So, the > > plan is: > > > > * On the fsfs-format7 branch, duplicate the fsfs-f7 code and > > turn it into a new experimental fs backend. I will name it FSX, > > with "X" standing for "experimental". It pronounce it "fisiks" > > which underlines its design goals. > > Today, you ripped out all FSFS support from the FSX backend. That > implies you have an entirely new backend, rather than an upgrade path > for existing FSFS user. >
There are no upgrade paths even within FSFS itself. We simply said "from now on, you may try feature X, if that should not be prevented by some earlier feature". 1.8 for the first time, tried to actually touch existing data during an upgrade and ran into (resolvable) trouble. On the implementation side, those franken-repos that have seen various format upgrades are very troublesome. As noted on IRC earlier, we just deprecated BDB so that we wouldn't > have to continue supporting multiple backends. But it seems you have > just created a third/new backend. > Well, we still support BDB. My understand is that we simply don't add new features to it and that we reserve the right to drop support entirely at any future release. FSX is a new backend, true. But initial support for it will be less than for BDB: We should run regression tests but explicitly mark it as "unfit for production usage" (see my other post for more on that). By the time FSX may become a stable and fully supported backend, it will likely have an FS2 interface. It might also be next-to-impossible to implement FS2 on a 1.0 FSFS repository or even any fsfs format. In that scenario, FSX would be the only backend and everything else is legacy. > > * Rip out the f7 code from the fsfs backend. > > > > * Open a "fsfs-improvements" integration branch. Merge all fsfs > > relevant changes in there in a hopefully review-friendly way. > > Is the idea that people can test FSX independently? And then changes > will go into FSFS on this branch? And then it will get to trunk as > part of FSFS? And that FSX will never land on trunk? > > If that is true, then why rip out the f7 support from FSFS on the branch? > See my other post. Haven't been entirely clear upon the whole strategy until I actually removed backwards compat stuff to see how much that helps. > > * Let people review (give them 2 weeks) & merge the integration > > branch to /trunk. > > > > * Continue work on fsx during that period and merge it directly > > to /trunk once the fsfs-improvements branch got closed. > > See. This part confuses me. It sounds like we're moving to multiple > back ends again. > > I thought we wanted to avoid that? > See above. For now, FSX is an experimental test-bed for everything nice-to-have on the backend side. The things we learn from that will help us make much more informed decisions about FS2 as well as the future of FSFS and FSX alike. -- Stefan^2.
