mmm, makes sense. That's worth a try!
2015-09-22 14:04 GMT+02:00 Jamie Echlin <jamie.ech...@gmail.com>:
> > The question is what you would use that information for
>
> So, basically, users provide a script in the UI, the value that it
> returned is indexed (in Lucene) by one of several various indexers that is
> configured elsewhere. You use JIRA, so you know what I'm talking about...
> it's JIRA.
>
> Only certain return types make sense depending on the configured indexer.
>
> Let's say I'm using the indexer that requires a Double. It occurs to me
> (actually my colleague) that the script could implicitly extend some other
> class than Script, eg:
>
> class DoubleScript {
> abstract Double run()
> }
>
> This could be set on the compilerconfiguration.
>
> I think if we do that the type-checker would point out the errors in the
> following script:
>
> if (xyz()) {
>
> 111D
>
> }
>
> else {
>
> "sss"
>
> }
>
>
> which would solve the problem. Sounds sensible?
> cheers, jamie
>
>
>
> On Tue, Sep 22, 2015 at 10:39 AM, Cédric Champeau <
> cedric.champ...@gmail.com> wrote:
>
>> Not necessarily. The question is what you would use that information for.
>>
>> 2015-09-22 11:36 GMT+02:00 Jamie Echlin <jamie.ech...@gmail.com>:
>>
>>> No, I guess so. So it sounds like you're saying I should forget it and
>>> move on to the next thing? I.e. it's not feasible to get at this in a
>>> static analysis context?
>>>
>>> On Tue, Sep 22, 2015 at 10:24 AM, Cédric Champeau <
>>> cedric.champ...@gmail.com> wrote:
>>>
>>>> Just for context and history, early versions of the type checker had
>>>> return type inference for methods too. However, we removed it for a good
>>>> reason: overrides. A method "foo" in class A can have an inferred type T,
>>>> but "foo" from B extends A could very well have a different inferred type
>>>> T2. So you can make no guarantee, at least for non private methods.
>>>>
>>>>
>>>> 2015-09-22 11:19 GMT+02:00 Jamie Echlin <jamie.ech...@gmail.com>:
>>>>
>>>>> Hi Shil,
>>>>>
>>>>> I did that actually, but it tells me the return type of the run method
>>>>> of Script is Object. Which is expected, if it's just looking at the actual
>>>>> method signature.
>>>>>
>>>>> But what I was hoping for is that given a script:
>>>>> "new Double(2)"
>>>>>
>>>>> it would be able to infer the actual return type is a Double. Now, I
>>>>> can imagine scripts where you randomly pick from any class on the
>>>>> classpath, invoke a no-args constructor, and return that... in that case I
>>>>> don't expect magic. But most scripts have one or two possible return
>>>>> statements (maybe implicit), I'm wondering if there is already code that
>>>>> can statically analyse a method/script and say that there are only one,
>>>>> two
>>>>> or whatever possible return types.
>>>>>
>>>>> I'll take a look at ReturnAdder, thanks for that.
>>>>>
>>>>> cheers, jamie
>>>>>
>>>>>
>>>>> On Mon, Sep 21, 2015 at 11:55 PM, Shil Sinha <shil.si...@gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Actually, the second option can be made safe by initializing
>>>>>> ReturnAdder with a listener, which will allow you to 'visit' all implicit
>>>>>> returns without actually adding return statements.
>>>>>>
>>>>>> On Mon, Sep 21, 2015 at 5:30 PM, Shil Sinha <shil.si...@gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> You could call 'getInferredReturnType' on the 'run' method node of
>>>>>>> your script class, but inferred return types only appear to be stored
>>>>>>> for
>>>>>>> closures and certain binary expressions (based on the usages of the
>>>>>>> INFERRED_RETURN_TYPE marker and the storeInferredReturnType method).
>>>>>>>
>>>>>>> You could also add all implicit return statements to the 'run'
>>>>>>> method via ReturnAdder, and then infer a return type by visiting all of
>>>>>>> the
>>>>>>> return statements, but I'm not sure if that kind of AST modification in
>>>>>>> a
>>>>>>> type checking extension is safe.
>>>>>>>
>>>>>>> Shil
>>>>>>>
>>>>>>> On Mon, Sep 21, 2015 at 4:34 PM, Jamie Echlin <
>>>>>>> jamie.ech...@gmail.com> wrote:
>>>>>>>
>>>>>>>> Hi...
>>>>>>>>
>>>>>>>> Is it possible to infer the return type, or types of a script
>>>>>>>> during static compilation? Presumably there is always a finite list of
>>>>>>>> branches where the script could end.
>>>>>>>>
>>>>>>>> I'm looking
>>>>>>>> at
>>>>>>>> org.codehaus.groovy.transform.stc.StaticTypeCheckingVisitor#getInferredReturnType,
>>>>>>>> but I'm not really sure how to call it from a type checking script, nor
>>>>>>>> what adds the inferred return type to the metadata which is used in
>>>>>>>> that
>>>>>>>> method.
>>>>>>>>
>>>>>>>> cheers, jamie
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>
