Re: Criticism of "version numbering"-document
| From: | Stefan Neufeind | Date: | Mon, 08 Mar 2004 14:51:07 +0000 |
| Subject: | Re: Criticism of "version numbering"-document | ||
| References: | 1 | Groups: | php.pear.qa |
| Request: | Send a blank email to pear-qa+get-867@lists.php.net to get a copy of this message | ||
On 8 Mar 2004 at 15:14, Lukas Smith wrote:
> Stefan Neufeind wrote:
>
> > writing about:
> > http://pear.php.net/group/docs/20040226-vn.php
>
> > But I agree with Pierre that adding the package state as a "postfix"
> > is imho not a good idea since this information is already present in
> > the state-field. What I personally dislike the most is adding a
> > version after the state, so we might result in 0.2.0alpha2 instead
> > of simply 0.1.1alpha or something like that.
>
> > "The state should always be added as a postfix unless the state is
> > stable. [...] The postfix consists of the state followed by a number
> > which is incremented with every subsequent release with the same
> > state."
>
> The reason why this was added (and note there are a fair number of
> people in the group which didnt/dont agree anymore or are atleast not
> 100% happy with it either) was:
Nice to hear that. So I'm not alone :-))
> 1. the state is missing in alot of places where it was felt that it
> was missing to better transport the state of the package, where only
> the version name was available
Well, then these could maybe have been reworked?
> 2. we dont have a state "RC" (yet)
Agreed.
> 3. during earlier discussion (specifically while talking about the
> BC breaking releases regulation) on this it was decided that new
> major version continue the version counting where the previous
> package left off .. which is why the first release of MDB2 was
> 2.0.0alpha1, and the next 2.0.0beta1 with the final stable being
> 2.0.0
Well I see what you mean. But is it realistic to assume that between
alphaX, betaY and the stable no new features will be added or changed
or anything? If that occured already a new sub-release (x.x.Y) would
have been required. Maybe people like seeing a 2.0.0 being released
as stable. But why would a 2.0.2, state stable have been a problem?
The initial 2.0-kickoff would have been "new API, but currently in
the works". imho no problem with that. And then 2.0.1 as beta and
2.0.2 as stable could have followed.
> The solution used in the version naming regulation takes a
> straighforward way of making things work.
>
> There are other solutions:
> 1) add the state whereever we currently only show the version name
Would be fine with me, yeah. Or if, for obvious reasons, we don't
want to do it that way then I'd suggest to at least make it
"2.0.0alpha" instead of adding another versionnumber after that.
Why not a version-numbering like that:
2.0.0alpha
2.0.1alpha
2.0.2beta
2.0.3
From the first alpha there were a few issues fixed, then it went
straight forward. Pierre already pointed out that the amount of
numbers is unlimited :-)) If if we already extend major and minor
numbers by another sub-version (x.x.Y), which is a good thing imho,
then we should avoid confusing people too much.
> 2) add a state "RC"
Right between beta and stable you mean? That's also fine with me,
yeah. Then we should have a hierarchy of
cvs -> alpha -> beta -> rc -> stable
Whereas stable should *not* be included in the name (if we make the
state part of the filename as it is now), and cvs/rc are completely
optional to use ( surely :-) ).
3) dont continue counting on new major versions, so
> my first MDB2 release could have been 0.1.0 with state alpha
Well, I think this is no option since it has already been discussed
quite long. At that time some people pointed out that end-users (not
devs) might be confused with a "MDB2 1.0" and "MDB 2.0" or something.
A clean solution would have been to use "MDB 2.x"-naming, but since
PHP does not support namespaces we at least need to rename the class
MDB to MDB2 (and the sub-directory also) in that step. So we agreed
that maybe having a MDB2 (or MDBv2 or MDB_v2, in that order, as
appropriate) would be the best solution.
Again, I'm not saying that the group has made a bad, not well-thought
about decision. And it *might* make sense to demand to append the
state to the name since not in all places the state-field might be
yet implemented (visible). But imho there are only two "good"
options:
a) Going for a "2.0.1alpha"-name instead of "2.0.0alpha2".
b) Implement the missing indicators for the state where possible and
avoid having the state in the package version completely. (imho
*really* the best and cleanest solution).
Please Let's discuss this, since I believe this is quite an essential
decision!
Stefan