On 6/4/13 2:34 PM, Christopher wrote:
Personally, I'm fine with not having a super-long running release
branch, but I like them for a few reasons:
1. It doesn't force rapid releases that are time-consuming and require
testing, but we still have that option if we want.
2. We can batch a few bugfixes and work on a bugfix release schedule
for nice-to-have, but not urgent bugfixes (patch Tuesdays? :)
3. We're going to create this branch anyway, so we can stabilize
things for testing prior to release... so why not leave it around for
bugfixes?
4. Pretending we won't have bugfixes for a prior release is possibly
delusional, and it seems like it doesn't really avoid anything
regarding merging over multiple branches... when we patch, we'll just
have to create these branches from their respective tags before we do
this process. It doesn't seem to save us from this merging process.
In short, I don't think leaving these branches around diverges from
the proposed branching model... it just keeps them around until we
decide that branch is EOL. If we leave a branch around until it
reaches EOL, and we never make a bugfix in it, then no big deal, no
harm done... just delete the branch at EOL.
I just want to keep make note that this will affect things much more
than previously in SVN as a merge is more than a guess at metadata.
If you have long-lived branches, this necessitates that there are very
many viable options in which a bug-fix can be applied (any branch at
all). Not keeping these around will force the developer to think about
where their fix is supposed to go (earliest, non-EOL'ed version).
Again, this is mostly a stylistic argument, and where I tend to disagree
with you. :)
I think branches should be hierarchical for subprojects and contribs,
as in: contrib/InstamoOrWhatever (unless there's a better way to
handle contribs...?)
I believe we'll need to make separate repos for subprojects and
contribs. This is one of the things I need to investigate how it's
currently handled. New repos are certainly the easiest to manage.
I think ticket/feature branches should be deleted after they've merged
back to develop, and should also be hierarchical, as in:
feature/ACCUMULO-9001
Works for me.
As for tags, I'm not comfortable tagging anything we haven't voted on
as an official release, and I think the time-consuming nature of that
should be factored in, when considering these long-running bugfix
branches for a release line.
I personally like being tag-heavy; however, there is a fine line here as
I already stated. There's no reason to not make a tag for something
you're working on in a feature branch that you want to share with
someone else (while you continue work on that feature). They can
checkout that tag which is assumed stable to an extent (the reason you
wanted someone else to pull it in the first place) and then make a
branch if they want to change something.
Point being, I do not want to preclude developers from making tags
unless it's an ASF release as I believe this gimps Git quite a bit. IMO,
for tags, the line should be drawn between "internal" and "external".
Tags advertised or intended for public use should (probably) be voted
on, while internal-only intended (doesn't preclude someone from finding
and downloading themselves) are encouraged.
--
Christopher L Tubbs II
http://gravatar.com/ctubbsii
On Tue, Jun 4, 2013 at 9:35 AM, Keith Turner <[email protected]> wrote:
On Tue, Jun 4, 2013 at 9:07 AM, Josh Elser <[email protected]> wrote:
Yay, Git. Wait...
I was going to wait to respond until I collected all of the info, but
since I still haven't gotten that done yet, I figured I should just say
"sure".
The one thing I do want to get hammered out is the general workflow we
plan to use. I believe that one "unstable" or "development" branch will
satisfy most use cases as we typically don't have much active development
against previous major releases.
The thing I don't care for (putting it mildly) is long-running
minor-release branches. I'm curious of suggestions that people might have
for how to work around this. One
Why? What problems are you thinking of w/ long-running minor release
branches?
possibility would be to be git-tag heavy while being more lax on official
apache releases.
Merits:
- Less merging through 2-3 branches which a bug-fix might apply
(1.4->1.5->1.6)
- Less clutter in the branch space (could be moot if we impose some sort
of "hierarchy" in branch names, e.g. bugfixes/ACCUMULO-XXXX,
minor/ACCUMULO-XXXX)
- Quicker availability of fixes for consumers (after a fix, a new tag is
made)
Downsides:
- Could create more work for us as we would be noting new minor releases.
Does Christopher's release work mitigate some/most of this?
- Creating git-tags without making an official release _might_ skirt a
line on ASF releases. Some artifact that is intended for public consumption
is meant to follow the release process.
It seems like you have a specific workflow in mind, but its not clear to me
exactly what you are thinking. Are you planning on elaborating on this
tonight? Is this workflow written up somewhere? If its not written up, a
few quick example scenarios would probably help me get on the same page.
Personally, I'd consider making the bold assumption that, over time, we
will create the infrastructure for ourselves to make better and better
releases which will also mitigate this. I'm curious what everyone else
thinks.
I'll try to make time tonight to get info on all of the necessary below.
On 6/1/13 4:28 PM, Christopher wrote:
Reviewing this thread, it seems everyone is pretty much in favor of
this. I propose we proceed by electing a "git transition advisor"...
someone who:
1) is sufficiently knowledgeable with git to identify roadblocks
during the transition and is willing to make it a point to propose and
follow through with solutions,
2) can serve as the main point of contact with INFRA tickets related
to the transition,
3) can advise other developers on best practices for things like when
to branch, where to put contribs, when to delete a branch (if ever),
etc. for things related to the shared repo.
Item 3 is really something we can all do to the extent that our
expertise allows, but it'd be nice to have somebody willing to do item
2, in the context of their experience in item 1.
I would like to nominate Josh Elser, because I think he's qualified,
and because his exact response to this inquiry was "Yay, git."
--
Christopher L Tubbs II
http://gravatar.com/ctubbsii
On Tue, May 21, 2013 at 5:44 PM, Christopher <[email protected]> wrote:
I'm sure this has been entertained before, I'm starting to think that
we should seriously consider switching to git, maybe sooner (during
1.6 development cycle?).
--
Christopher L Tubbs II
http://gravatar.com/ctubbsii
