On Tue, May 22, 2012 at 9:08 AM, Thomas Mortagne <[email protected]> wrote: > On Mon, May 21, 2012 at 7:12 PM, Jerome Velociter <[email protected]> > wrote: >> Hi devs, >> >> Following a thread on github, here's a mail to start a discussion >> about the way we determine what is API or not. >> >> Our current rule is : >> 1) "Non user-public code must be located in an internal package just >> after the module name." c.f. [1] (implied that what is not in internal >> is public) >> 2) What is public has to go through the deprecation strategy described at >> [2]). >> >> I think the rule is good but there is a problem in its enforcement >> right now, mainly because : >> * There is some legacy code where 1) does not make much sense because >> it "arrived too late at the party" (for example in oldcore) >> * There is some "new code" where some classes/interfaces hasn't been >> made internal when they likely should have been. For example I've been >> playing around with the WYSIWYG recently and in the client module, >> there are *only* user-public classes [3]. I'm sure we can find a lot >> of examples of such practice and I'm OK to work and list some if >> needed. >> >> The problem I see coming is that the cost of refactoring will increase >> for a lot of modules/classes/etc., while at the same time those >> modules should not have been API to begin with, and are probably not >> even being used as such by anybody. The risk is our progress being >> bogged down for no good reason. >> So what can we do to enforce a solid backward-compatibility policy for >> API code, while keeping flexibility in XWiki internals ? >> >> We could : >> A) Not do anything :) Maybe it's just me that sees this as a potential >> problem. >> B) Do the work to move everything that ought to be in an internal >> package to an internal package. But I don't believe we're reading to >> make that effort, so that's not going to happen IMHO >> C) Do the work to move what ought to be in an internal package "on the >> fly", when refactoring code. That would be on a case-by-case case, >> probably requiring a mail to announce it ; or more coercive, a VOTE. >> C) We change the rule. We could decide that instead of having >> everything be an API and enforce the "internal" status in a special >> package, we take it the other way around and Day everything is >> internal, and APIs needs to be in a special package (or be annotated >> with a special annotation). This could also be the opportunity to >> introduce another rule that says that such APIs should be referenced >> in their own module documentation on extensions.xwiki.org. >> >> What do you think ? > > Big -1 to force putting public APIs in a specific useless package. > > The main issue is mostly with oldcore which is supposed to be > refactored at some point peace by peace anyway so that's not a big > issue IMO. It's just another argument to get rid of it ASAP. > > For "new" modules, if some code is is public while it should not then > it should be fixed ASAP and indicated in the release note (since it's > a breakage even if it should never have existed in the first place) or > made @Deprecated if it's too old and jugged probably already used to > be removed like that, in both cases it would require a VOTE IMO. Note > sure if that's B or C. > > The status of new code is not as bad as you think IMO
Besides the WYSIWYG, there is the REST module that also has (wrongly) only public code. For other modules it looks better, though I think there are quite some interfaces that are offered as public for the sheer fact they are interfaces ; which is a mistake IMO. Also, there is the case of the DocumentAccessBridge, which by essence is meant to grow, thus be broken (but maybe it is "complete" now, though if you think about it, our present rule would not have permit us to elaborate it in the time-frame we did - which is a couple of years) Jerome > and I don't see > the point in changing a good rule because it's not applicable to old > code that's supposed to be replaced anyway at some point. Whatever the > new rule you will never be able to apply it to oldcore anyway... For > example C does not work since we have a lot of things we do want > public for java code which are not in the "api" package and we are not > going to deprecated and move them there since it does not worth the > effort. > >> >> Jerome. >> >> [1] http://dev.xwiki.org/xwiki/bin/view/Community/JavaCodeStyle#HPackagenames >> [2] >> http://dev.xwiki.org/xwiki/bin/view/Community/DevelopmentPractices#HBackwardCompatibility) >> [3] >> https://github.com/xwiki/xwiki-platform/tree/master/xwiki-platform-core/xwiki-platform-wysiwyg/xwiki-platform-wysiwyg-client/src/main/java/org/xwiki/gwt/wysiwyg/client/ >> >> -- >> Jérôme Velociter >> Winesquare >> http://www.winesquare.net/ >> _______________________________________________ >> devs mailing list >> [email protected] >> http://lists.xwiki.org/mailman/listinfo/devs > > > > -- > Thomas Mortagne > _______________________________________________ > devs mailing list > [email protected] > http://lists.xwiki.org/mailman/listinfo/devs -- Jérôme Velociter Winesquare http://www.winesquare.net/ _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
