That all sounds correct. The one edge case is that when the .0 release hasn't been cut yet but RCs exist, it's important to include in fix versions any releases that would need to be included if the current RC passed.
e.g. once 1.4.0 RC0 comes out, we need to include 1.5.0 because if RC0 passes, such an issue will actually be in 1.4.1 and not 1.4.0. I'm not sure if we should set 1.4.0 or 1.4.1 in such a case. When prepping for 1.2.0 I tried to account for folks who had picked either 1.2.0 or 1.2.1 when generating the next RC, by correcting fix versions as needed. On Thu, Nov 9, 2017 at 1:28 PM, Dave Latham <[email protected]> wrote: > Cool. Don't mean to suggest this is a change or new, just thinking through > and writing down what I think I've observed and folks seem to be doing, > which makes sense. > I don't have the bandwidth at the moment to figure out the doc format and > go through the process to create a patch, but if any of the above is > helpful, would be happy for anyone inclined to get it into the doc. > > On Thu, Nov 9, 2017 at 11:24 AM, Andrew Purtell <[email protected]> wrote: > >> > Yes, those are the rules I applied. >> >> To clarify: Those are the rules I applied after going back and adding back >> fixversions such that the set {1.3.1, ...} becomes {1.4.0, 1.3.1, ...} as >> Sean suggested. >> >> Now, 1.4.0 was dropped only if there is a fixversion 1.3.0 in the set. And, >> 1.5.0 was dropped wherever we have 1.4.0 in the set, and that is currently >> everywhere because at this moment branch-1.4 == branch-1, but will begin to >> diverge as soon as 1.4.0 is released. >> >> >> >> On Thu, Nov 9, 2017 at 11:21 AM, Andrew Purtell <[email protected]> >> wrote: >> >> > Yes, those are the rules I applied. >> > >> > While I was in there adjusting fixversions I noticed that generally we >> > follow exactly that approach for numbering. I think we inherit it from >> > Hadoop, because our old timers came from that community. >> > >> > Dave - Please consider submitting a patch for our online book for this, >> if >> > the text doesn't already exist in there somewhere. >> > >> > On Thu, Nov 9, 2017 at 11:17 AM, Dave Latham <[email protected]> >> wrote: >> > >> >> +1 for making it as simple as possible to determine if a given fix is >> in a >> >> given release purely from the release numbers, without having to consult >> >> the dates of when branches were made or release candidates were built. >> >> >> >> I think the rules should be >> >> >> >> Fix version of X.Y.Z => fixed in all releases X.Y.Z' where Z' >= Z >> >> Fix version of X.Y.0 => fixed in all releases X.Y'.* where Y' >= Y >> >> Fix version of X.0.0 => fixed in all releases X'.*.* where X' >= X >> >> >> >> By this policy, fix version of 1.3.0 implies 1.4.0, but 1.3.2 does not >> >> imply 1.4.0, as we could not tell purely from the numbers which release >> >> came first. >> >> >> >> To track a fix then, I think that means that there should usually be a >> fix >> >> version added for each branch that a commit was pushed to, with the >> >> exception of master if there is a branch for the next major release that >> >> has not happened yet. >> >> >> >> I think this is probably just repeating what Sean was saying, but it was >> >> helpful for me to write out and perhaps helpful for others to think >> about. >> >> >> >> On Thu, Nov 9, 2017 at 10:37 AM, Andrew Purtell <[email protected]> >> >> wrote: >> >> >> >> > Related, on HBASE-18996 Peter said: >> >> > >> >> > the fix version is not correct. It should include 1.5 too. Did you >> >> remove >> >> > that on purpose? >> >> > >> >> > and my response: >> >> > >> >> > Yes, I removed 1.5.0 for anything that is going out in 1.4.0. Fix >> >> versions >> >> > in HBase have historically meant the same thing as in Hadoop, which is >> >> "in >> >> > this release and any later". That said, I'm only touching fix versions >> >> for >> >> > branch-1. I won't presume to mess with fix version accounting for >> >> branch-2. >> >> > We have another RM tending to that. >> >> > At this point branch-1 (1.5.0) == branch-1.4 (1.4.0) so having both >> in >> >> > fixversions would be fine but redundant, and a future RM would have >> some >> >> > work to do. As things stood, they were horribly inconsistent, with >> many >> >> > 1.5.0 fixversions missing. >> >> > >> >> > >> >> > On Thu, Nov 9, 2017 at 10:27 AM, Andrew Purtell <[email protected]> >> >> > wrote: >> >> > >> >> > > You mean put back the 1.4.0 fixversion for anything released in >> >> 1.3.1? I >> >> > > can do that. I don't have a strong opinion either way. Let me make a >> >> pass >> >> > > now. >> >> > > >> >> > > Any other suggestions? >> >> > > >> >> > > >> >> > > On Thu, Nov 9, 2017 at 6:15 AM, Sean Busbey <[email protected]> >> >> wrote: >> >> > > >> >> > >> I'd like to try convincing you to just drop issues that were in >> >> 1.3.0. >> >> > >> >> >> > >> I assume that 1.4 will become our new stable line. Suppose that I >> am >> >> > >> running on our stable 1.2 release line. When considering an upgrade >> >> to >> >> > >> 1.4.z, which CHANGES files do I have to read to get a sense of >> what I >> >> > >> have to look out for in changes? >> >> > >> >> >> > >> Presuming the 1.3.0 CHANGES files contains everything that changed >> >> > >> since 1.2.0, I could just read the 1.3.0 CHANGES and then the >> CHANGE >> >> > >> for the 1.4.z release I am aiming for (since it will have 1.4.0 - >> >> > >> 1.4.z in it). >> >> > >> >> >> > >> If you use a later 1.3.z as the basis, then I have to find what the >> >> > >> last 1.3.z that had its RC created before 1.4.0. (a later one will >> >> > >> cover changes that are not included in 1.4.0 and might not be in >> any >> >> > >> 1.4.z yet) >> >> > >> >> >> > >> On Thu, Nov 9, 2017 at 12:05 AM, Stack <[email protected]> wrote: >> >> > >> > On Wed, Nov 8, 2017 at 4:31 PM, Andrew Purtell < >> >> [email protected]> >> >> > >> wrote: >> >> > >> > >> >> > >> >> You may notice I'm dropping '1.4.0' from fix versions wherever >> we >> >> > have >> >> > >> >> something that went in to any 1.3.x. This is because I am basing >> >> the >> >> > >> >> CHANGES.txt changelog for 1.4.0 on the latest from branch-1.3, >> so >> >> > >> anything >> >> > >> >> that went in on branch-1.3 will be included in the changelog at >> >> the >> >> > >> correct >> >> > >> >> point in history. Anything in the 1.4.0 section of the changelog >> >> for >> >> > >> >> branch-1.4 will be for changes in branch-1.4 not in branch-1.3. >> >> > >> >> >> >> > >> >> >> >> > >> > If 1.4.0 goes out before 1.3.2, does that mean, there will be >> fixes >> >> > that >> >> > >> > are in 1.4.0 (because they were committed post 1.3.1) not >> >> mentioned in >> >> > >> > 1.4.0 CHANGES.txt? >> >> > >> > >> >> > >> > Seems fine. Just asking. >> >> > >> > >> >> > >> > St.Ack >> >> > >> > >> >> > >> > >> >> > >> > >> >> > >> >> -- >> >> > >> >> Best regards, >> >> > >> >> Andrew >> >> > >> >> >> >> > >> >> Words like orphans lost among the crosstalk, meaning torn from >> >> > truth's >> >> > >> >> decrepit hands >> >> > >> >> - A23, Crosstalk >> >> > >> >> >> >> > >> >> >> > > >> >> > > >> >> > > >> >> > > -- >> >> > > Best regards, >> >> > > Andrew >> >> > > >> >> > > Words like orphans lost among the crosstalk, meaning torn from >> truth's >> >> > > decrepit hands >> >> > > - A23, Crosstalk >> >> > > >> >> > >> >> > >> >> > >> >> > -- >> >> > Best regards, >> >> > Andrew >> >> > >> >> > Words like orphans lost among the crosstalk, meaning torn from truth's >> >> > decrepit hands >> >> > - A23, Crosstalk >> >> > >> >> >> > >> > >> > >> > -- >> > Best regards, >> > Andrew >> > >> > Words like orphans lost among the crosstalk, meaning torn from truth's >> > decrepit hands >> > - A23, Crosstalk >> > >> >> >> >> -- >> Best regards, >> Andrew >> >> Words like orphans lost among the crosstalk, meaning torn from truth's >> decrepit hands >> - A23, Crosstalk >>
