[External] : Re: Dependencies on java.desktop
Kevin Rushforth
kevin.rushforth at oracle.com
Wed May 19 19:07:48 UTC 2021
That's an interesting idea, although splitting it up to the degree you
listed below would be unwieldy at best and likely unworkable (due to
split package issues and interdependencies). It also would be wrong to
create any implementation modules unless they are 100% hidden from the
application developer. If any of them would (or even could) ever show up
in a developer's .pom, .xml, or .gradle file, then it would be a
non-starter since that would amount to leaking implementation details.
Irrespective of how the graphics module might be broken up, I suspect
that splitting any of our existing modules might run into compatibility
problems in the maven.central deployment case, unless gradle and maven
have completely solved the module dependencies properly such that
application can specify, for example, javafx.controls and have it
automatically pull in all of the other javafx dependencies.
Leaving that last point aside, I can see how splitting it up into just
two modules *might* be workable with a bit of refactoring:
javafx.graphics --> depends on javafx.graphics.* (requires transitive)
javafx.graphics.core (includes all of the public API except printing,
and all of the implementation except printing + j2d pipeline)
javafx.graphics.printing --> depends on javafx.graphics.core,
java.desktop (includes printing api, printing impl, j2d pipeline)
An application that doesn't need printing could depend on
javafx.graphics.core. Of course, that leaves out UI controls-based
applications, since javafx.controls "requires transitive"
javafx.graphics. So you'd have to split that one too, and possibly
javafx.fxml, in order to really get any benefit from this.
Anyway, I'm not really in favor of this approach, just pointing out a
less radical idea that we could at least discuss further as a possibility.
I still think making printing optional might be a feasible alternative
to reimplementing it natively (which isn't as odd an idea as it might
seem initially, since embedded / mobile systems don't typically have a
printer). If so, a better approach to throwing an exception would
probably be to just to report no printers if the Java2D library (the
java.desktop module) isn't available. Need to think about this one some
more.
> Out of curiosity I tried that locally and it just works ;-)
I'm sure it depends on your definition of "just works". If you mean
taking the class files after the fact and cobbling together some modules
from the built class files and running something, then maybe, although
even then I don't see how it could work seamlessly.
-- Kevin
On 5/19/2021 10:20 AM, Tom Schindl wrote:
> Hi,
>
> Well I looked a bit closer now and the situation and you are right. I
> think need to do that one by one.
>
> I think getting rid of HostServices::showDocument is quite easy as the
> code in Java-AWT is just 1 JNI-Method so copying that to OpenJFX
> should be fairly easy.
>
> Now on the printing story which is the real problem a fairly radical
> approach would be to split up javafx.graphics into multiple modules
> and make javafx.graphics a META-Module who provides the backward
> compat using "requires transitive".
>
> We would end up with:
>
> * javafx.graphics
> * javafx.graphics.base
> * javafx.graphics.print
> * javafx.graphics.prism.base
> * javafx.graphics.prism.j2d
> * javafx.graphics.prism.es2
> * javafx.graphics.prism.d3d
> * javafx.graphics.glass.base
> * javafx.graphics.glass.win32
> * javafx.graphics.glass.mac
> * ...
>
> I know that sounds radical but from a pure architectual point of view
> this would be better than optional features in javafx.graphics and
> would not require us to reimplement printing to get rid of java.desktop.
>
> Out of curiosity I tried that locally and it just works ;-)
>
> Tom
>
> Am 18.05.21 um 23:40 schrieb Kevin Rushforth:
>> As noted in the thread you quoted below, removing the dependency on
>> java.desktop from javafx.base isn't a particularly hard problem, and
>> is tracked by JDK-8240844 [1]. And even though it will require a spec
>> change (meaning a CSR), it doesn't result in any loss of
>> functionality, since in order to usefully do anything with the
>> JavaBeanXxxxx classes requires java.desktop anyway.
>>
>> Removing the dependency on java.desktop from javafx.graphics is a
>> larger issue because of printing. There is also an implementation
>> dependency on the Desktop class to implement
>> HostServices::showDocument that would need to be redone. And of
>> course it depends on the above (eliminating the dependency from
>> javafx.base) being done first.
>>
>> For printing we would need to do one of two things:
>>
>> 1. Eliminate the implementation dependency on the Java2D printing
>> code. This is a large effort, but would preserve existing functionality.
>>
>> 2. Make the JavaFX Printing function "optional" (i.e., make it a
>> "ConditionalFeature"), such that if java.desktop is not present,
>> printing doesn't work (all of the printing APIs would throw an
>> UnsupportedOperationException if java.desktop is not available). An
>> application that wants to do printing would need to include
>> java.desktop.
>>
>> Option 1 would be my preferred approach, but as mentioned above it's
>> a lot of work. Option 2 would need a spec change, and I'm not
>> convinced we want to do it. If there are enough other developers who
>> would want this, we could open it up for discussion for some future
>> version (not JavaFX 17).
>>
>> Phil might have some thoughts on this.
>>
>> -- Kevin
>>
>> [1] https://bugs.openjdk.java.net/browse/JDK-8240844
>>
>>
>> On 5/18/2021 10:45 AM, Tom Schindl wrote:
>>> Uff - I'd like to revisit this topic. As I did some jlink stuff for
>>> our applications adding java.desktop once more bugged me.
>>>
>>> I guess the first thing to do is to file a JIRA-Ticket but it really
>>> starts to bug me to include java.desktop although I don't plan to
>>> use printing (and I guess > 90% of the JavaFX don't use the printing
>>> API either but produce PDFs or whatever) and Java-Beans.
>>>
>>> Tom
>>>
>>> Am 27.03.18 um 14:26 schrieb Kevin Rushforth:
>>>> Hi Tom,
>>>>
>>>> Yes, this is an unfortunate dependency. It is "only" an
>>>> implementation dependency, meaning that nothing in the public API
>>>> depends on java.desktop (which is why we don't "requires transient
>>>> java.desktop"), so it should be possible to remove this dependency
>>>> in the future. As noted, it is only there because Java Beans is
>>>> part of the java.desktop module.
>>>>
>>>> In the interim, your suggestion of "requires static java.base"
>>>> could be the way to go. It would need a spec change to the JavaFX
>>>> beans adapter classes documenting that they would throw an
>>>> UnsupportedOperationException if java.desktop was not present at
>>>> runtime, along with a recommendation that applications needing that
>>>> functionality should add "requires java.desktop" to their own
>>>> module-info.java.
>>>>
>>>> Note that this would only help non-graphical JavaFX applications
>>>> that use javafx.base for its collections, properties, and bindings,
>>>> since javafx.graphics requires java.desktop in a way that currently
>>>> cannot easily be made optional (not without reimplementing printing
>>>> support anyway).
>>>>
>>>> -- Kevin
>>>>
>>>>
>>>> Tom Schindl wrote:
>>>>> Hi,
>>>>>
>>>>> Anyone else has an opinion on that? Is require static the way to go?
>>>>>
>>>>> Tom
>>>>>
>>>>> On 21.03.18 23:23, Tom Schindl wrote:
>>>>>> Hi,
>>>>>>
>>>>>> I always thought the JavaFX-Codebase should be able to run with
>>>>>> just the
>>>>>> java.base module but I was browsing the codebase a bit and was
>>>>>> suprised
>>>>>> (or rather shocked) that even the base-module requires java.desktop.
>>>>>>
>>>>>> If I get it correct this because of the java.beans (provided by the
>>>>>> adapters) stuff is found in there. Why hasn't the requires there not
>>>>>> defined as:
>>>>>>
>>>>>> requires static java.desktop;
>>>>>>
>>>>>> Tom
>>>>>>
>>
More information about the openjfx-dev
mailing list