Greetings, all.
I'm working on putting AST into CVS, and part of that involves making a
branch where I can safely autoconf AST without getting in David's way.
It therefore seemed useful to work through the Branching patterns document
I've pointed to before, and make notes. I've put these on the wiki at
<http://wiki.starlink.ac.uk/twiki/bin/view/Starlink/BranchingPolicy>.
Before anyone gets excited, I'm doing this because I think I need to
branch right now. I don't imagine we'll start branching frantically
-- our requirements are too simple for that.
Part of branching is to have consistent tag names. There's some
discussion of that on the page where Mark summarised a discussion here:
<http://wiki.starlink.ac.uk/twiki/bin/view/Starlink/CvsTagging>
See:
Wiki notes
http://wiki.starlink.ac.uk/twiki/bin/view/Starlink/BranchingPolicy
ACME branching patterns -- copious
http://acme.bradapp.net/branching/ Download the lot at
http://acme.bradapp.net/branching/printing.html
Terser best-practice document
http://www.magic-cauldron.com/cm/cvs-bestpractices/
FAQ-o-matic
Page `How do I create and checkout a branch?', currently at
<http://www.loria.fr/~molli/fom-serve/cache/6.html> though the URL
might theoretically change.
Here is some suggested elaboration of these, partly as a strawman.
Mark suggests tagging releases as `rMajor-minor'. Fairy Nuff. Less
is more.
Miscellaneous other tags, specific to a package: Mark suggests prefixing
these with the package name. I imagine that the only person adding
such tags is likely to be the `codeline owner' in ACME terminology,
so there's probably little need to disambiguate this further.
Branch names for releases (release x.y)
* tag the branch point as bp-release-x-y (the FAQ-o-matic article
above suggests that it's useful to always make a branch starting
from a tag rather than an un-named random snapshot of the code,
so that it's easier to work out what diffs have happened since
the fork).
* name the branch release-x-y and work on it to add any last-minute
changes for the release (this assumes the `S2.1 Parallel
Releasing/Development Lines' pattern which I mention in the wiki
page -- this seems eminently sensible).
* finally tag the release rx-y (as above). Any minor bugfixing
happens on this branch and is tagged and merged to the development
code on the trunk as appropriate.
The only problem with this is that the string `release-x-y' looks a
bit like a tag rather than a branch. We could have bp-release-x-y
and branch-release-x-y instead but that's more verbose. It depends
a bit on how often anyone would be typing or looking at these things:
my guess is not much, when it came down to it, so it doesn't actually
matter that much.
Branch names for experiments (such as autoconfing AST) or largeish
feature additions (such as adding TimeFrame to AST or perhaps XmlChan):
* branch point bp-dev-nxg-20031121
* branch name dev-nxg-20031121
Perhaps something like bp-dev-nxg-20031121-autoconfing would be both
unique and have handy mnemonic value. So it's long; but you don't
have to type it often. That's the one I think I prefer.
There are other possible reasons for having a branch
<http://acme.bradapp.net/branching/line-elems.html>, but I don't think
any of them are useful to us at this stage -- we want to keep things
simple.
In the branch names, the `dev' tells you that this is some sort of
development branch, and that it's expected to be merged back on to the
trunk (in our case) at some point. Similarly the `release' in the release
branches tells you that this is not expected to be finally merged back
onto the trunk, but may have merges (of bugfixes) as required. In either
merge case, there should be a tag added with <branch-name>_MERGED (added
with -F if there's already one) at the point where the merge is made:
this indicates both that the merge has happened (in case you forget) and
means that if you have to do more than one merge from this codeline you
know which to merge _from_ (CVS unfortunately doesn't stop you re-merging
stuff you've already merged, and such a tag can help you avoid this).
The point of this is that you can tell from a tag roughly what's happening
on it, and what is expected to happen to it later. Thus it is an implicit
<http://acme.bradapp.net/branching/branch-policy.html#CodelinePolicy>,
which works because the only place things will be remerged in our case
is back onto the trunk (so we don't have to spell that out) and because
we probably won't be creating too many branches, so as long as the tag
indicates who made the branch, it's likely they'll still be able to
remember why they did it and whether it's still live (if that were in
doubt, and important).
As regards `codeline owners': should we perhaps note these down
somewhere, either centrally, or within the packages themselves (a
file called `Maintenance' at the top of the package, perhaps?). This
is part of the `Codeline Policy' mentioned in the last paragraph.
This is sounding bureaucratic, but is probably handy (`I've had to fix
something in package X, but I've forgotten whom I should tell, so
they get a chance to veto or correct it'), and could also act as a
store of informal notes about whether a line is frozen, or abandoned,
or whatever. I'm thinking here of the useful practicalities of the
<http://acme.bradapp.net/branching/branch-policy.html#CodelineOwnership>
pattern, without the potentially severe bureaucracy it could imply.
Voila one strawman. Beat, berate or burn to taste.
See you,
Norman
--
---------------------------------------------------------------------------
Norman Gray http://www.astro.gla.ac.uk/users/norman/
Physics and Astronomy, University of Glasgow, UK [log in to unmask]
|