Hi Vincent, I have the exact same reading then Thomas. The @since tag should not be use for anything else than what it is made for: notifying the user of any package/class/method at which point it has been introduced. A package/class/method is define by its canonical name, and any change in it is therefore another package/class/method, even if it has the exact same signature. While I understand that this could have not been properly followed in the past, this is clearly a mistake, and it is mostly due to the lack of automation/check on these annotations. The removal of the @Unstable annotation has nothing to do with @since, and should not be mixed up. I found very annoying to see the need to know for sure the introduction time of the @Unstable annotation, simple to respect a rules we have only voted to avoid those @Unstable to be left for ever. Normally, the "owner" of the newly introduced API, should take care to remove those @Unstable as earlier as possible, and the maximum should only be reach in very rare cases. So, my feeling is that those rare case could be solved using Git history/blame if ever needed. So I am not in favor of any of your "new" options. I would simply clarify the rules for @since (if really needed for those not understanding the obvious). I also remind all committers that it is your responsibility to remove the @Unstable you have introduce in due time. If you really want to add new annotations, I would be far more interested to distinguish API and SPI, and provide them a different stability, allowing us more flexibility :) Thanks, On Tue, Feb 11, 2014 at 3:22 PM, Thomas Mortagne <[email protected]>wrote:
On Tue, Feb 11, 2014 at 3:19 PM, Thomas Mortagne <[email protected]> wrote:
On Tue, Feb 11, 2014 at 3:15 PM, [email protected] <[email protected]> wrote:
See below.
On 11 Feb 2014 at 13:17:03, Marius Dumitru Florea (
[email protected](mailto:[email protected])) wrote:
I agree with Thomas and Denis, but I must admit that I haven't updated the @since version when I did refactorings in the past. I'll pay attention to this next time.
Thanks, Marius
On Mon, Feb 10, 2014 at 2:51 PM, Denis Gervalle wrote:
On Sun, Feb 9, 2014 at 11:16 PM, Thomas Mortagne wrote:
On Sun, Feb 9, 2014 at 6:10 PM, [email protected] wrote: > Hi devs, > > I always ask myself this question so I think we need a common
agreement.
> > So here's the question: > * I have added some code in version N and this I have a "@since N" in the code > * In version M (M > N), I move the class/interface to a new package > > Question: Do I change the @since annotation to "@since M" or not? > > 2 possibilities: > * Reasoning 1: it's a new class/interface since the FQN of the class/interface has changed and thus we should use "@since M" > * Reasoning 2: even though the FQN has changed it's still the same code that was moved and from a user POV, it was still introduced in version N and thus we should keep "@since N" > > WDYT? > > I'm hesitating. The most technically correct answer is Reasoning 1 IMO but the most useful one is probably Reasoning 2 since the question we wish to answer is probably: "when was this code first introduced?". > > Thus reasoning 2 seems slightly better to me.
Big -1 for 2 which is totally out of context, @since indicate that you can use that class or method since that version in you code and indicate you which version you are going to be compatible with. If you change the class or method your can't keep the same @since. If you want to know since when the feature exist look at xwiki.org...
I completely agree with Thomas, a -1 for 2) I would add that if you want to know from where the code come from, Git is your best friend.
<playing devil's advocate for the sake of the discussion>
I don't fully agree with this.
The point of the @since tag is exactly to NOT have to check in Git to see when some code was introduced! And with your logic, the @since tag is never needed at all since we can always check in Git, and it's as easy to check in Git for Reasoning 1 than it is for Reasoning 2.
The point of @since is to indicate since when a signature exist which means that 2 is completely wrong. I never talked about git, I don't care how you know since when a feature exist but please don't use @since which has a different meaning for that.
If you start changing the @since then it makes it impossible to
properly remove the @Unstable annotations later on since for each @Unstable annotation you'll need to do some deep Git archeology to reconstruct the first time the API was introduced.
Also, I can tell you that a lot of devs (the majority, if not 80%) have
been doing Reasoning 2 since the beginning of our usage of @since, since it's the simplest thing to do and it's what you get by default if you don't do anything... I know I did it, I know Marius did too and I'm pretty sure others too.
So to recap, my points are: * If you need to find out when some class was moved in another package
you can always check Git and you don't need the @since that for this
* Reasoning 1 makes it almost impossible in practice to remove @Unstable annotations * Reasoning 2 is complex to implement (the proof being that for most of our code it wasn't done)
</playing devil's advocate for the sake of the discussion>
Note that this discussion is important since we never formalized how to use the @since annotation (it's not documented on http://dev.xwiki.org/xwiki/bin/view/Community/DevelopmentPractices).
Also note that http://www.oracle.com/technetwork/java/javase/documentation/index-137868.html#@sincedoesn't explain what to do when a class/interface is moved elsewhere. I googled and couldn't find how other projects handle the since tag.
It does no say it because it's obvious, if you change the package you change the name, the class short name does not have any value in Java.
So taking everything into account we have the following options:
A) Update the @since value when we move a class/interface to another
package. But force the @Unstable annotation to have a text specified (ATM it's optional) with the rule of specifying when the API is introduced. For example @Unstable("Introduced in 5.4M1").
B) Keep what we've been doing implicitly, which is to not change the
@since value when a class/interface is moved to another package and consider that this @since tag corresponds to when the code was first introduced independently of its class/interface location. In this case no need to use a text for the @Unstable annotation.
C) Use some other annotation like for example @Introduced("5.2") or
@Introduced in 5.2 (javadoc).
As for automating the addition of the since tags, I couldn't find
anything good for us to use. FTR I found:
- http://stackoverflow.com/questions/3417243/automatic-since-javadoc-tag-for-r... the maven plugin doesn't do magic. - http://help.eclipse.org/indigo/index.jsp?topic=%2Forg.eclipse.pde.doc.user%2...
While I prefer A) which I find more technically correct I think it's also a lot more work to enforce (in lots of places it means duplicating information between @since and @unstable) so I'm hesitating, especially since we've been doing B implicitly.
Any idea/preference?
Thanks -Vincent
I take the occasion to also mention that it would be nice to have a better way to maintain those @since. At least a check of presence, or even better a check of correctness, in the build would nice to have. The must being to have those @since added automagically :)
> > Thanks > -Vincent
_______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Thomas Mortagne
-- Thomas Mortagne _______________________________________________ devs mailing list [email protected] http://lists.xwiki.org/mailman/listinfo/devs
-- Denis Gervalle SOFTEC sa - CEO