A WIP for JDK-8238080 for review/discussion (Re: "Internal review ID 9063426: "FXMLLoader: if script engines implement javax.script.Compilabel compile scripts"

Tue Apr 14 17:52:44 UTC 2020

Yes, I agree that enough time has gone by. Go ahead with your proposal. 
I would wait a bit to create the CSR until the review is far enough 
along to know which direction we intend to go.

Unless there is a real concern about possible regressions if scripts are 
compiled by default, I think "enabled by default" is the way to go. Your 
argument that such script engines are broken seems reasonable, since 
this only applies to script engines that implement 
javax.script.Compilable in the first place. We still might want to add 
way to turn compilation off for individual scripts. One other thing to 
consider is that if compilation fails, it might make sense to log a 
warning and fall back to the existing interpreted mode.

Does anyone else have any concerns with this?

-- Kevin

On 4/14/2020 9:48 AM, Rony G. Flatscher wrote:
> Hi there,
>
> as there was probably enough time that has passed by I would intend to create a CSR in the next days
> with the PR as per Kevin's suggestion.
>
> (For the case that this feature should not be active by default, the CSR will suggest to define a
> new "compile" PI in the form <?compile [true|false] ?> (default, if no PI data given: true), which
> is independent of the existence of a language PI (this way it becomes also possible to allow
> compilation of external scripts denoted with the script-element, which do not need a page language
> to be set as the file's extension allows the appropriate script engine to be loaded and used for
> execution). A compile-PI would allow for turning compilation of scripts on by just adding the PI
> <?compile?> or <?compile true?>  to FXML files (and <?compile false?> to turn off), which seems to
> be simple and self-documentary. In general employing such compile PIs allows for setting compilation
> of scripts on and off throughout an FXML file.)
>
> ---rony
>
>
> On 04.04.2020 18:03, Rony G. Flatscher wrote:
>
>> Hi Kevin,
>>
>> On 03.04.2020 01:21, Kevin Rushforth wrote:
>>> I see that you updated the PR and sent it for review.
>>>
>>> Before we formally review it in the PR, let's finish the discussion as to whether this is a useful
>>> feature, and if so, what form this feature should take.
>>>
>>>  From my point of view, this does seem like a useful feature. Would other users of FXML benefit
>>> from it?
>> Script code should be executed faster after compilation, so any FXML page that hosts script code may
>> benefit.
>>
>> The benefits depend on the type of script (and maybe its size and its complexity) and also on the
>> types of event handlers the scripts serve, e.g. move or drag event handlers may benefit
>> significantly. This is because repeated invocation of compiled script event handlers do not cause
>> the reparsing of that script's source and interpreting it on each invocation, which may be expensive
>> depending on the script engine, but rather allows the immediate evaluation/execution of the compiled
>> script by the script engine.
>>
>>> I'm not certain whether we want it to be implicit, compiling the script if the script engine in
>>> question implements Compilable, or via a new keyword or tag. What are the pros / cons?
>> In principle there are three possibilities:
>>
>>      1) If a script engine implements javax.script.Compilable, compile the script and execute the
>>      compiled version. In the case of event handlers compile and buffer the compiled script and
>>      execute the compiled script each time its registered event fires.
>>
>>        o Pro: immediately benefits all existing FXML pages that host scripts
>>        o Con: it is theoretically possible (albeit quite unlikely) that there are scripts that fail
>>          compiling but have been employed successfully in interpreted mode
>>
>>      2) Introduce some form of an optional attribute (e.g. "compile={true|false}") to the FXML
>>      language PI that switches on compilation of scripts hosted in FXML definitions if the script
>>      engine implements the javax.script.Compilable interface. If missing it would default to "false".
>>      (Alternatively, add a "compile" PI, that if present causes the compilation of scripts, if the
>>      script engine supports it. It would be an error if the "compile" PI was present, but the
>>      "language" PI was not.)
>>
>>        o Pro: compilation of FXML hosted scripts is done only, if the FXML definition of the language
>>          PI gets changed
>>        o Con: benefit not made available automatically to existing FXML pages that host scripts
>>
>>      3) Another possibility would be to define a boolean attribute/property "compile" for script and
>>      node elements and only compile the scripts, if the property is set to true
>>
>>        o Pro: compilation of FXML hosted scripts is done only, if the FXML definition gets changed
>>          accordingly
>>        o Con: potential benefit not made available automatically to existing FXML pages that host scripts
>>
>> 2 and 3 could be combined, where 2 would define the default compilation behavior that then could be
>> overruled individually by 3.
>>
>> The question would be whether 2 or/and 3 are really necessary as it can be expected that compilation
>> of scripts by the script engines would find the same errors as while interpreting the very same
>> scripts (if not, the script engine is badly broken and it could be argued that then the
>> interpretation part of the script engine might be expected to be broken as well which would be quite
>> dangerous from an integrity POV; the same consideration applies as well if the execution of the
>> compiled script would behave differently to interpreting the very same script by the same script
>> engine).
>>
>> The current WIP implements 1 above and includes an appropriate test unit.
>>
>>> What do others think?
>>
>>> In either case, we would need to make sure that this doesn't present any security concerns in the
>>> presence of a security manager. As long as a user-provided class is on the stack, it will be fine,
>>> but we would need to ensure that.
>> The compilation of scripts needs to be done by the Java script engines (implementing both,
>> javax.script.Engine and javax.script.Compilable) as well as controlling the execution of compiled
>> scripts ([javax.script.CompiledScript]
>> (https://docs.oracle.com/en/java/javase/14/docs/api/java.scripting/javax/script/CompiledScript.html)).
>>
>> ---rony
>>
>>
>>
>>> On 4/2/2020 10:41 AM, Rony G. Flatscher wrote:
>>>> After merging master, applying some fixes and changing the title to reflect the change from WIP to a
>>>> pull request I would kindly request a review of this pull request!
>>>>
>>>> Here the URL: <https://github.com/openjdk/jfx/pull/129>, title changed to: "8238080: FXMLLoader: if
>>>> script engines implement javax.script.Compilable compile scripts".
>>>>
>>>> ---rony
>>>>
>>>>
>>>> On 28.02.2020 19:22, Rony G. Flatscher wrote:
>>>>> Here is a WIP [1] implementation of [2]. The WIP is based on [3], which is currently in review, and
>>>>> has an appropriate test unit going with it as well, please review.
>>>>>
>>>>> There should be no compatibility issue with this implementation.
>>>>>
>>>>> Discussion: another solution could be to not compile by default. Rather compile, if some new
>>>>> information is present with the FXML file which cannot possibly be present in existing FXML files.
>>>>> In this scenario one possible and probably simple solution would be to only compile scripts if the
>>>>> language process instruction (e.g. <?language rexx?>) contains an appropriate attribute with a
>>>>> value
>>>>> indicating that compilation should be carried out (e.g.: compile="true"). This way only FXML with
>>>>> PIs having this attribute set to true would be affected. If desired I could try to come up with a
>>>>> respective solution as well.
>>>>>
>>>>> ---rony
>>>>>
>>>>> [1] "Implementation and test unit": <https://github.com/openjdk/jfx/pull/129>
>>>>>
>>>>> [2] "JDK-8238080 : FXMLLoader: if script engines implement javax.script.Compilable compile
>>>>> scripts":
>>>>> <https://bugs.java.com/bugdatabase/view_bug.do?bug_id=JDK-8238080>
>>>>>
>>>>> [3] "8234959: FXMLLoader does not populate ENGINE_SCOPE Bindings with FILENAME and ARGV":
>>>>> <https://github.com/openjdk/jfx/pull/122/commits>
>>>>>
>>>>>
>>>>> On 24.01.2020 16:26, Kevin Rushforth wrote:
>>>>>
>>>>>> Thank you for filing this enhancement request. As an enhancement it should be discussed on this
>>>>>> list before proceeding with a pull request (although a "WIP" or Draft PR can be used to illustrate
>>>>>> the concept).
>>>>>>
>>>>>> For my part, this seems like a reasonable enhancement, as long as there are no compatibility
>>>>>> issues, but it would be good to hear from application developers who heavily use FXML.
>>>>>>
>>>>>> -- Kevin
>>>>>>
>>>>>>
>>>>>> On 1/24/2020 7:21 AM, Rony G. Flatscher wrote:
>>>>>>> Just filed a RFE with the following information:
>>>>>>>
>>>>>>>      * Component:
>>>>>>>          o JavaFX
>>>>>>>
>>>>>>>      * Subcomponent:
>>>>>>>          o fxml: JavaFX FXML
>>>>>>>
>>>>>>>      * Synopsis:
>>>>>>>          o "FXMLLoader: if script engines implement javax.script.Compilabel compile scripts"
>>>>>>>
>>>>>>>      * Descriptions:
>>>>>>>          o "FXMLLoader is able to execute scripts in Java script languages
>>>>>>> (javax.script.ScriptEngine
>>>>>>>            implementations) if such a Java script language gets defined as the controller
>>>>>>> language in
>>>>>>>            the FXML file.
>>>>>>>
>>>>>>>            If a script engine implements the javax.script.Compilable interface, then such scripts
>>>>>>> could
>>>>>>>            be compiled and the resulting javax.script.CompiledScript could be executed instead
>>>>>>> using
>>>>>>>            its eval() methods.
>>>>>>>
>>>>>>>            Evaluating the CompiledScript objects may help speed up the execution of script
>>>>>>> invocations,
>>>>>>>            especially for scripts defined for event attributes in FXML elements (e.g. like
>>>>>>> onMouseMove)
>>>>>>>            which may be repetitevly invoked and evaluated."
>>>>>>>
>>>>>>>      * System /OS/Java Runtime Information:
>>>>>>>          o "All systems."
>>>>>>>
>>>>>>> Received the internal review ID: "9063426"
>>>>>>>
>>>>>>> ---rony