From: Philip Oakley Date: Tue, 04 Feb 2014 21:03:41 GMT Subject: Re: [RFC/PATCH] howto/maintain-git.txt: new version numbering scheme Message-ID: <43FB93CCA3984B8890CA448E9E75E44E@PhilipOakley> In-Reply-To: From: "Junio C Hamano" > "Philip Oakley" writes: > >> If we are progressing from V1.9 to V2.0 quickly (one cycle?), which I >> understand is the plan, then mixing the minor development items >> (patch >> series which progress to master) with the maintenance fixes over the >> next few months, thus only having 1.9.x releases, sounds reasonable. >> >> If there is going to be separate maintenance fixes from the patch >> series >> developments then keeping to the previous 1.9.x.y for maintenance >> would >> be better. >> >> Will the new rapid counting continue after V2.0, such that we get to >> V2.9 -> V3.0 rather more quickly than V1.0 -> V2.0 ? >> >> The key discriminator would be to say when V2.0 will be out for >> deciding >> the V1.9 sequence. > > I do not quite follow. The time distance between v1.9 and v2.0 > should not affect anything. If it is a long road, there may be > v1.10, v1.11, v1.12, ... I wasn't sure if you were considering going past either 1.9.9 to 1.9.10, and going from 1.9 to 1.10 was something that hadn't occurred to me (somewhat of a Doh! moment maybe). > before we have v2.0. If not, v2.0 may > immediately follow v1.9 as a new feature release. There may be > maintenance releases based on v1.9 that does not add any new > features. > > Right now, if you count the maintenance releases, there are > potentially four kinds of version gaps: > > - Between v1.8.5 and v1.8.5.1, there are fixes but no new features; > > - Between v1.8.5 and v1.8.6, there are new features but no > compatibility worries; > > - Between v1.8.6 and v1.9.0, there are new features, no > compatibility worries, but somehow the jump is larger than the > one between v1.8.5 and v1.8.6; and > > - Between v1.9.0 and v2.0.0, there are new features and also > compatibility concerns. > > Switching to 2-digit scheme and calling the upcoming one v1.9 (and > the next major one v2.0) was meant to make the naming more flat, OK I'd buy that flattening approach. > as > the third item in the above list "somehow the jump is larger" does > not seem to add much value to the end users. So the logical > numbering becomes more like this: > > - Between v1.9 and v1.9.1, there are fixes but no new features; > > - Between v1.9.x and v1.10, there are new features but no > compatibility worries; > > - Between v1.9.x and v2.0, there are new features and also > compatibility concerns. > > With a twist, though. There seem to be many places where at least > three digits are assumed to exist in our version numbers, so in > order to make life easier, the updated document says vX.Y (a feature > release) will identify itself as vX.Y.0 Yes. I'd be happy to support that third 'digit' for the maint releases, with zero as the initial release. Git Gui has a version string checking routine but its regex only needs two parts X.Y (we looked into the version string back in $gmane/217189 > > Thanks for the clarifications. Philip.