On Tue, Jul 14, 2009 at 03:29:42PM +0100, James Rowe wrote:
> * W. Trevor King (wking@drexel.edu) wrote:
> > One problem is that we don't actually have "releases". People just
> > clone a branch, install, and go.
>
> This is actually the main reason I've manually mirrored the tree in
> the past, so that users of our projects can get BE. If tarballs were
> available I probably wouldn't even bother, but bzr really isn't a nice
> dependency for just submitting/commenting on bugs.
Fair enough. It will be good when we get a commit-able web interface
and/or email interface going.
> Isn't there a bzr web interface that at least supports creating
> tarballs/zips? It is pretty standard functionality for most other VCS'
> web interfaces so I'm guessing there must be, but loggerhead seems not
> to support it.
Unfortunately, people would still need bzr to install the versioned source:
libbe/_version.py:
bzr version-info --format python > $@
So you'll need a "release" target in the makefile to build a bzr-less
install. While you're at it, you should probably compile the manpage
too to remove the docbook-to-man dependency.
Do people want a HEAD tarball too? There must be a bzr equivalent of
.git/hooks/post-update
but I don't know what it is.
> > If you're worried about stability, just clone from a more stable branch
> > (i.e., Chris' trunk). I think > this is good for distributed development,
> > but maybe makes it hard to package into a conventional release-based system.
> > With the bzr patch number in setup.py as the patch release number, I would be
> > releasing my 0.1.363 while Chris releases his 0.1.314, even though they're at
> > about the same point. I would rather be releasing my
> > 0.1.20090714121347
> > while Chris releases his
> > 0.1.20090713154540
> > Since then the similarity is clearer.
>
> Both approaches seem pretty odd to me, as a user you would have no
> idea if 0.1.200910302359 has the fixes you required in a release you
> were using that was numbered 0.1.200907141554. Surely you'd at least be
> {pre,suf}fixing a branch name to the version.
"be --version" currently gives you the revision id:
wking@drexel.edu-20090714121347-c6rloikst1t3m5yl
which tells you exactly which commit your installed version is based on.
If we want stick with revision numbers, how about:
major.minor.revno-branch_nick
But then we'd have to pick "unique" branch nicknames...
I'd sliced out the timestamp portion of the revision id so that the
"patch-number" would be an integer and the branch name wasn't
references, so that "upgrading" from one branch to another could be
transparent to the users (who just see an increading timestamp), but
still available to the developers (who know when commits were made to
the branches they track, and the likelyhood of to-the-second commit
collisions in official packages is small).
On Wed, Jul 15, 2009 at 12:54:05AM +1000, Ben Finney wrote:
> "W. Trevor King" <wking@drexel.edu> writes:
>
> > On Tue, Jul 14, 2009 at 10:36:26PM +1000, Ben Finney wrote:
> > > Please, no. Timestamps aren't version strings, that's conflating two
> > > pieces of information with very different meanings. Correlating the
> > > two is the job of a changelog.
> >
> > Which we don't bother keeping (also NEWS), since "bzr log" works so
> > nicely.
>
> That's not a changelog, that's a commit log of every source-level commit
> made. Far too much detail for a changelog of *user-visible* changes
> associated with a release.
I need a user around to help me determine "user-visable" changes ;).
My labmates loose interest after be init/new/comment :p. None of
which has ever changed, other than set-root -> init ;).
> > The timestamp should at least replace the patch release number, which
> > you agree is-desirable-to increase motonically ;).
>
> I still disagree that a timestamp is the right thing to use there. If
> you want a monotonically-increasing indicator of which revision we're up
> to, that's immediately available with the revision number from VCS on
> the main branch. That also has the advantage of producing consecutive
> numbers for each revision, by definition.
But not during branch-switches, while my method skips large regions,
but probably increases during any reasonable branch-switch. For
example, when I upgraded to rich root to pull Ben's patch, I'm not
sure if Chris upgraded the trunk and merged my branch, or just ditched
the trunk and cloned my branch. Using actual bzr revision numbers
would make switching branches that either wrong (in the case of
rev-id decreases) or confusing (in the case of a single
non-consecutive increase).
On Tue, Jul 14, 2009 at 11:11:31AM -0400, Chris Ball wrote:
> > I agree that's a problem. I think the solution is to start making
> > releases, with specific version strings, as source tarballs.
>
> I'm happy to do this if people think it would be useful, and I don't
> yet have a strong opinion on whether the releases should come with
> version numbers or timestamps.
I imagine the news from 2006 to now will be somewhat abridged ;).
--
This email may be signed or encrypted with GPG (http://www.gnupg.org).
The GPG signature (if present) will be attached as 'signature.asc'.
For more information, see http://en.wikipedia.org/wiki/Pretty_Good_Privacy
My public key is at http://www.physics.drexel.edu/~wking/pubkey.txt
