Lambdafication (was Re: Default methods in JFX-8)
Stephen F Northover
steve.x.northover at oracle.com
Mon Oct 7 07:45:23 PDT 2013
It is still interesting to see the results and compare with the
non-lambda version.
Steve
On 2013-10-07 10:21 AM, Tom Schindl wrote:
> ... I can't see any real improvement for using lambdas beside that you
> killed off warnings which could be fixed otherwise as well.
>
> So I really don't get what it would help us beside making e.g. RoboVM
> our sole chance to get FX on iOS harder to maintain.
>
> Tom
>
> On 07.10.13 16:16, Sven Reimers wrote:
>> Ok. So I will file a P4 saying Lambdafication for Controls
>> and send the diff to Richard/Kevin/Jonathan to be attached..
>>
>> should I base the change on b110 (master)?
>>
>> I could create a public bitbucket branch based on master and add my changes
>> there - better idea?
>>
>> What approach is most simple for review?
>>
>> Should I split test and library code changes?
>>
>> -Sven
>>
>> P.S. Shall I try to get this done as well for other modules? Which would be
>> preferred? (Just in case I have some more time to spend)
>>
>>
>>
>> On Mon, Oct 7, 2013 at 1:26 PM, Kevin Rushforth
>> <kevin.rushforth at oracle.com>wrote:
>>
>>> **
>>>
>>> 5. Should we enable more -Xlint warnings in OpenJFX build?
>>>
>>> 6. Any chances anything of this can still go in 8 (e.g. get rid of warnings
>>>
>>>
>>> We have 2 weeks where we can still accept P4-P5 bugs into FX 8, and
>>> getting rid of warnings would be a P4 bug. I guess it depends on how
>>> intrusive the changes are and whether someone has time to review it in the
>>> next two weeks.
>>>
>>> -- Kevin
>>>
>>>
>>>
>>>
>>> Sven Reimers wrote:
>>>
>>> Ok. So here are the results of trying to add lambda and diamond to the
>>> controls module:
>>>
>>> 1. A lot of generics and typing to be fixed (esp. in tests). Seems you can
>>> get some anonymous inner classes type checked by the compiler, but not the
>>> lambda equivalent.. very interesting.
>>>
>>> 2. 279 Files modified (including tests)
>>>
>>> 3. A lot of the automatic replacements could probably be nicer e.g.
>>>
>>> ft.setOnFinished(new EventHandler<ActionEvent>() {
>>> @Override public void handle(ActionEvent
>>> actionEvent) {
>>> getChildren().remove(tm.textNode);
>>> }
>>> });
>>>
>>> was replaced to:
>>>
>>> ft.setOnFinished((ActionEvent actionEvent) -> {
>>> getChildren().remove(tm.textNode);
>>> });
>>>
>>> most unobtrusive code probably:
>>>
>>> ft.setOnFinished((actionEvent) -> getChildren().remove(tm.textNode));
>>>
>>> 4. A lot of illegal forward reference errors - these were result of missing
>>> this in the automatic transformation from anonymous inner to lambdas (seems
>>> the rules are not identical - you have to add "this." as prefix if using
>>> lambdas - not sure this is the expected way it should work)
>>>
>>> 5. Should we enable more -Xlint warnings in OpenJFX build?
>>>
>>> 6. Any chances anything of this can still go in 8 (e.g. get rid of warnings
>>>
>>> 7. Probably more things I just can't think of at the moment...
>>>
>>> How to take this forward? If there is interest in the change I could make
>>> available...
>>>
>>> Comments? Ideas?
>>>
>>> -Sven
>>>
>>>
>>>
>>> On Fri, Oct 4, 2013 at 2:19 PM, Sven Reimers <sven.reimers at gmail.com> <sven.reimers at gmail.com> wrote:
>>>
>>>
>>>
>>> Oh and btw - would you go for lambda with or without additional type info
>>> before parameter name?
>>>
>>> -Sven
>>>
>>>
>>> On Fri, Oct 4, 2013 at 2:05 PM, Sven Reimers <sven.reimers at gmail.com> <sven.reimers at gmail.com>wrote:
>>>
>>>
>>>
>>> Ok. Here you go...
>>>
>>> I just did an inspection run for the controls module and my IDE came up
>>> with (drum roll) 888 possible lambda conversions..
>>>
>>> Looking through them I discovered that usage of <> (aka diamond syntax)
>>> is not used (or at least not used a lot) in at least the controls
>>> modules. My IDE showed me 1171 occurrences.
>>>
>>> Is there a good reason not to use diamonds?
>>>
>>> Will now try to apply all those changes and figure out if this still
>>> builds... up next: go through the other modules...
>>>
>>> -Sven
>>>
>>>
>>> On Fri, Oct 4, 2013 at 1:35 AM, Richard Bair <richard.bair at oracle.com> <richard.bair at oracle.com>wrote:
>>>
>>>
>>>
>>> Brian was telling me at J1 that whether parallel gets you performance or
>>> not depends on the size of the collection and the complexity of the work to
>>> perform. There is definitely a point at which parallel helps -- and a point
>>> at which it hurts.
>>>
>>> Richard
>>>
>>> On Oct 3, 2013, at 3:33 PM, Hervé Girod <herve.girod at gmail.com> <herve.girod at gmail.com> wrote:
>>>
>>>
>>>
>>> Here is a nice example, taking advantage of the ease of going
>>>
>>>
>>> parallel. Apparently the performance without parallel will also further
>>> improve.http://blog.hersen.name/blog/2013/10/01/project-lambda-it-was-worth-the-wait/
>>>
>>> Hervé
>>>
>>> Sent from my iPad
>>>
>>>
>>>
>>> On 4 oct. 2013, at 00:20, David Grieve <david.grieve at oracle.com> <david.grieve at oracle.com>
>>>
>>> wrote:
>>>
>>>
>>> And what about Stream? I like the declarative code that comes from
>>>
>>>
>>> using Stream and I can see places in the code where Stream could be used,
>>> but I wonder about its performance relative to iterators and/or enhanced
>>> for loops.
>>>
>>>
>>> On Oct 3, 2013, at 4:45 PM, Richard Bair <richard.bair at oracle.com> <richard.bair at oracle.com>
>>>
>>> wrote:
>>>
>>>
>>> Hello, OpenJFX Community.
>>>
>>> There's a question about using Java 8 features in FX.
>>>
>>> I've been working on the support for InputMethods in JFXPanel which
>>>
>>>
>>> is an important feature for many users who speak hieroglyphic languages.
>>>
>>>
>>> The issue is tracked under:
>>>
>>>
>>> https://javafx-jira.kenai.com/browse/RT-13248
>>>
>>> In order to have a high-quality support we need to change
>>>
>>>
>>> javafx.scene.input.InputMethodRequests interface and introduce 3 new
>>> methods. This is not needed for pure FX applications right now, but
>>> absolutely required for InputMethods in the JFXPanel. However, the
>>> interface is public and it was present since FX2.0, so changing it would
>>> become a breaking change. So the only way to avoid the problem is using the
>>> default methods. Those would return some stub values, the JDK is OK with
>>> that, as it would not crash or throw exceptions, but text composition would
>>> not work correctly.
>>>
>>>
>>> I know that we want to avoid using the Java 8 features in the
>>>
>>>
>>> JFX-8, so I wanted to ask - is it OK to use the default methods here?
>>>
>>>
>>> If you are staying away from JDK8 features for the JFX78 backport,
>>>
>>>
>>> don't worry. There are more issues with new JDK8 APIs than with the new
>>> language features.
>>>
>>>
>>> For example there were default methods put into some collections
>>>
>>>
>>> classes that we solved by pushing them down to the first implements. But
>>> the Date and Time picker depends on the new time package. The threeten
>>> backport won't be updated until after 8 ships, so that has been removed so
>>> far.
>>>
>>>
>>> I'de be interested to know what a wholesale lamdaization would
>>>
>>>
>>> result in speed wise and code size wise (both source and compiled). From
>>> what I can tell the IDEs can lambda and de-lambda fairly easily, so it jsut
>>> makes the backport more of a busy work proposition. If there were
>>> performance gains it would also make a great front page story in the next
>>> java magazine or a case study..
>>>
>>>
>>> After having used Lambda's for JavaOne, I'd love to make the
>>>
>>>
>>> conversion, even if in the end the performance was the same, because the
>>> savings in noise in the Java files is so big. At one time I just took the
>>> concurrent classes and lambda-ized them to measure the impact on those
>>> classes. You could maybe pick a package and just lambda-ize that one
>>> package and see what happens in terms of size reduction. We might see:
>>>
>>>
>>> + A reduction in the overall class size (not pack-200'd)
>>> - An increase in startup time (have to spin up synthetic classes
>>>
>>>
>>> created at usage time)
>>>
>>>
>>> +/- And increase or decrease in performance
>>> + A decrease in source code
>>>
>>> It would be interesting to get some data for these points and see
>>>
>>>
>>> what effect lambda's have. Especially if an IDE can just do it in bulk…
>>>
>>>
>>> Richard
>>>
>>>
>>> --
>>> Sven Reimers
>>>
>>> * Senior Expert Software Architect
>>> * NetBeans Dream Team Member: http://dreamteam.netbeans.org
>>> * Community Leader NetBeans: http://community.java.net/netbeans
>>> Desktop Java:http://community.java.net/javadesktop
>>> * Duke's Choice Award Winner 2009
>>> * Blog: http://nbguru.blogspot.com
>>>
>>> * XING: https://www.xing.com/profile/Sven_Reimers8
>>> * LinkedIn: http://www.linkedin.com/in/svenreimers
>>>
>>> Join the NetBeans Groups:
>>> * XING: http://www.xing.com/group-20148.82db20
>>> * NUGM: http://haug-server.dyndns.org/display/NUGM/Home
>>> * LinkedIn: http://www.linkedin.com/groups?gid=1860468
>>> http://www.linkedin.com/groups?gid=107402
>>> http://www.linkedin.com/groups?gid=1684717
>>> * Oracle: https://mix.oracle.com/groups/18497
>>>
>>>
>>> --
>>> Sven Reimers
>>>
>>> * Senior Expert Software Architect
>>> * NetBeans Dream Team Member: http://dreamteam.netbeans.org
>>> * Community Leader NetBeans: http://community.java.net/netbeans
>>> Desktop Java:http://community.java.net/javadesktop
>>> * Duke's Choice Award Winner 2009
>>> * Blog: http://nbguru.blogspot.com
>>>
>>> * XING: https://www.xing.com/profile/Sven_Reimers8
>>> * LinkedIn: http://www.linkedin.com/in/svenreimers
>>>
>>> Join the NetBeans Groups:
>>> * XING: http://www.xing.com/group-20148.82db20
>>> * NUGM: http://haug-server.dyndns.org/display/NUGM/Home
>>> * LinkedIn: http://www.linkedin.com/groups?gid=1860468
>>> http://www.linkedin.com/groups?gid=107402
>>> http://www.linkedin.com/groups?gid=1684717
>>> * Oracle: https://mix.oracle.com/groups/18497
>>>
>>>
>>>
>>>
>>
More information about the openjfx-dev
mailing list