On 10/07/2010 03:18 PM, Vincent Massol wrote:
On Oct 7, 2010, at 2:56 PM, Alex Busenius wrote:
Hi Vincent,
On 10/07/2010 09:38 AM, Vincent Massol wrote:
Hi Caleb and all,
Here's my take: 1) I agree it's currently hard to know the compatibility of a given extension 2) Putting the XE version in the extension version is limiting IMO. Consider the following typical use case: -- we release XE version N and the Acme Extension version N -- we make changes to Acme Extension -- we release XE version N+1 and thus Acme Extension N+1 -- I'm a user with XE version N installed. I check to see if there's a new version of the Acme Extension I use. I see Acme Extension N+1 but N+1 doesn't match my XE version so I assume it's not going to work with XE version. -- The problem is that this assumption is false. We would need to re-release every extension N times whenever there are changes to it in order just to name them with the past released XE versions...
It depends on how we define compatibility, if we say that all x.y.* versions are compatible among each other, then the choice will be easy. In the future, Extension Manager will handle versions transparently for the users (independent from versioning scheme), so it will become even easier. The overhead for releasing unchanged extensions is not that big IMO because there are no exceptions and only one version to think about.
3) We're developing an Extension Manager in charge of finding all available extensions for you directly from your wiki and to install/upgrade them easily 4) The way the Extension Manager works is by looking at the dependency version in the Maven POM for the extension. Thus the best practice is to use version ranges to show the compatibility matrix of a given extension with XE core. Thus I propose several things: -- that we start using version ranges in our extension POMs -- that indeed we have a template on xwiki.org for users to contribute information on whether a given extension works on a given XE release, in order to distribute the workload of creating the compatibility matrix -- when a user reports it work, we update the POM for the extension accordingly (same when a user reports that it doesn't work)
Indeed, the Extension Manager will be a great improvement, but it will solve only a *half* of the problem - it will make it much easier for the users, but not for the developers.
Currently, Java code is handled nicely in that respect - we *always* have a branch for the next version, we have a development practices for introducing of incompatible changes, bugs are fixed in both trunk and last branch, new features go to trunk etc.
But why does this scheme not apply to the applications,
It does apply to everything, including applications.
Branches are only created on a need basis since there's no point in creating branches if nothing will go in them and it would be useless to create branches for 100 extensions when you don't know at all if any changes will be made in them.
The problem with on-demand branches is that nobody creates them and all changes just go to trunk, which works most of the time, but not always. Consider the following a use case (happened with 2.4.1): - both XE version 2.4.0 (stable) and 2.5 are using App version X - as usual, we have trunk for XE 2.5 and branch for XE 2.4.0 - App was not changed yet, so there is no branch - the first change makes App dependent on 2.5 API, the change (correctly) goes to trunk (future X+1), no branch is created since it is still not needed - a few weeks later, some other changes to App are committed to trunk as usual, they do not introduce incompatibilities, so nobody thinks about branching (again) - in a few more weeks we release 2.4.1, all changed apps are released as usual, among them App X+1, which breaks XE 2.4.1 The problem is that (as far as I understand the "on-demand" branching) the _release manager_ needs to know which apps are compatible with the stable branch and makes branches for them if needed. Actually, _anyone_ who wants to fix a bug in App needs to check all changes made to it so far, understand which ones are compatible with the stable XE, branch App-X and merge all important compatible changes from trunk + the actual bugfix to the new branch. With synchronized versions, there would be no need to think about this, new features would just go to trunk and bugfixes to both trunk and stable branch (like everywhere else).
which also contain code and have exactly the same problems? It seems to me that the only reason is the huge overhead created by the fact that *each* application has its own version and release cycle. We ship a wild mix of applications with different core versions, and everyone who wants to make changes to them is supposed to take care of compatibility.
Indeed, that's how it's working right now.
IMO the way applications are currently handled is a mess, which will only be *hidden* from the users by the Extension Manager, and the only way to make it easy for developers is to synchronize versions and apply the strategy we have for core to all the applications at the same time.
Can you please answer to my use case above to see how you'd handle it, because that's the main issue I see. IMO your solution is very costly or it forces users to always upgrade to the next XE version even when all they need is an application upgrade.
See above for the answer. Users don't have to upgrade if they don't want to, they can always try a newer version of the app and hope for the best (like they do it now). Thanks, Alex
Thanks -Vincent
Of course this strategy cannot be applied to 3rd party applications, but we will only use them, so here the Extension Manager will take care of them.
Thanks, Alex
In conclusion I don't think we need to change the current versioning scheme for extensions and in any case we can't control all the extensions out there since the idea is to promote the creation of XWiki extensions outside of the xwiki dev team and thus we need a better mechanism than one based on the name.
Note: if you wanted to put the XE version in the extension name you'd need to allow version range syntax in the name which is a pain since you'd need to re-release everytime you learn that your extensions works or doesn't work with a given XE release (barring the issue with special chars for some filesystems...).
Thanks -Vincent
On Oct 6, 2010, at 2:28 PM, Caleb James DeLisle wrote:
I have been observing problems with the versioning scheme which we are using.
Because applications are not branched along with core, when a bugfix version of a stable branch is released, new versions of applications are typically pulled in. This means that `experimental' code is being introduced into a `stable' branch in a bugfix version. This is not the path I would choose but more importantly we can't honestly say that our code goes through a milestone/release candidate verification process if some of the code is allowed to bypass it. This situation has caused me to make a mistake which I was able to correct during the release without major issue, I think the same issue is behind the release of 2 bogus versions (2.4.1 and 2.4.2)
There is another issue, users who want to mix and match applications to build their own wiki are faced with a set of version numbers and no way to know what is compatible with what. A user who I spoke with last night had this very problem. We could publish a compatibility matrix but if we were to show all the versions a given application is compatible with, that would require testing each application version against each core version and I think we need to concentrate on testing what gets released in XE.
Both of these problems would be fixed if version numbers were synchronized and everything was branched for a release. Relevant questions which come to mind are "do we need the capability to release applications at separate times?" and "is there no way to do that with synchronized version numbers?"
Am I missing any other reasons? Should this not become a proposal?
Caleb
_______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs