Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [cross-project-issues-dev] JFace Generics

Hi Mikael,

For me it's very hard to value the benefits of a generified Viewer framework. Personally I'd definitely not need that but I'm willing to accept that other have other opinions on this question.

If this fundamental change (in the light of its impact to existing code) is welcomed by a majority then I'm willing to adjust my working code accordingly. For sure this will be a huge effort (with no value for me as I said), so I'd expect at least a public announcement through the appropriate channels and that the change is well-thought, correct and complete.

I totally disagree with your opinion on warnings in the code. They're absolutely unacceptable (even by rule in my project). I'd better disable a compiler check (or findbugs rule, for that matter) than risk that any new potential problem gets overlooked in a sea of warnings that nobody ever wants to address. Once you've got more than 5 warnings in the problems view, you'll never ever look at that view again.

I'd really appreciate if the projects that I need to consume think carefully about the warnings that they introduce. If they're necessary and solveable it's certainly okay, but please don't do it "peu à peu" or back and forth.

Cheers
/Eike

----
http://www.esc-net.de
http://thegordian.blogspot.com
http://twitter.com/eikestepper



Am 29.08.2013 15:25, schrieb Mickael Istria:
On 08/29/2013 02:36 PM, Ed Merks wrote:
Not to rain on anyone's parade, but respectfully suggest the platform team should reconsider the gain verses the
client impact for generifying the JFace APIs.   Designing generic containers isn't easy, and it's all too easy to make
mistakes that will be difficult to correct in the future
[...]
In the end, many of the things being changed are effectively SPI. One implements the APIs for providers and passes
them to a generic container that does the right things with it.  There's little to be gained from adding generics, to
justify the cost to the ecosystem, and getting it right is much harder than it appears at first glance...
Although I trust people saying that adding generics is difficult and error-prone (<T> or <? extends T> ?), I believe
that adding generics in JFace has a lot of value, simply because of the number of cast necessary when dealing with a
ContentProvider/LabelProvider couple. Dealing with current APIs forces us to remind the actual types for the JFace
viewer, and it is a tricky exercise that is error prone: I've often seen/written some ClassCastException that takes more
time to notice and debug that if I could see immediate feedback in IDE thanks to generic.
Overall, adding generics to JFace will really boost productivity of JFace users. That's IMO a valid reason to introduce
warnings on existing code.

What is the current issue? Is it that existing code using JFace will now show warnings? Is this really an issue, does it
prevent code to work? I don't think so. Those warnings just reveals that current usages of JFace don't leverage this new
smart mechanism to prevent from ClassCastException. It's not a big deal to see them.
IMO, just asking to roll back everything is just like disabling a FindBugs rule: it doesn't solve anything, it just
hides a place for improvement. In the end, it reduce code quality.

So, IMHO, the balance is more in favor of adding generics that in many cases will provide a better productivity and quality.
And I think it is totally fine to start merging such change so early in the release train to get maximal feedback.

I see (in Gerrit; thanks Matthias) that you've committed changes for TreeViewer and here I think the whole approach is
completely questionable.  When is it the case that a tree view has uniformly the same elements throughout?   I think
that's so rarely the case that it's worse than useless to make such an assumption; I would argue it's mostly just
noise that will never solve real problems.

I've seen many TreeViewers using a super type other than Object, especially in the modeling world, where almost
everything is an EObject (and often a specialized super-type shared across the whole metamodel), or in some JDT or PDE
code where there are some TreeNodeAdapter (or whatever pattern which create a dedicated type to put in
TreeContentProvider). So it seems to make sense for TreeViewers as well. And for other cases, it's not a big deal to
write "TreeContentProvider<? extends Object>".
--
Mickael Istria
Eclipse developer at JBoss, by Red Hat <http://www.jboss.org/tools>
My blog <http://mickaelistria.wordpress.com> - My Tweets <http://twitter.com/mickaelistria>


_______________________________________________
cross-project-issues-dev mailing list
cross-project-issues-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cross-project-issues-dev




Back to the top