[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Newsgroup Home]
[news.eclipse.modeling.mdt.ocl] Re: custom operators and functions using model information

Hi, Elio,

The only unary operator currently recognized by OCL is "-" (possibly "+", but I rather doubt that).  For that, the getJavaMethodFor(...) method will indicate the translation.

There is no facility for defining arbitrary infix operators in OCL.  The grammar fixes them quite strictly.

Cheers,

Christian


On Fri, 2009-09-04 at 21:59 +0000, Elio Damaggio wrote:
Hi Christian,

I was able to have the operator work :)

How is the situation for unary operators?
Is there a way to define those for custom datatypes?
Thanks,

Elio


Christian W. Damus wrote:

> Hi, Ed, Elio,

> Yes, OCL is trained to look for Java names "plus", "minus", "lessThan",
> "greaterThan", etc. to implement these symbolic operators, but it does
> expect to find the symbolic names (e.g., "+", "<") in the model's
> EClass.  The "<" / "lessThan" operation is particularly handy because it
> allows your type to be used in "orderedBy" iterators.

> One of the callOperation() or some such methods in the
> EvaluationEnvironment type hierarchy has the mapping from symbolic to
> Java names, in looking up the Method objects to invoke reflectively.  I
> should, of course, have documented it in the SDK's developer guide,
> too  :-(

> Cheers,

> Christian


> On Sat, 2009-08-29 at 06:52 +0100, Ed Willink wrote:

>> Hi Elio, Ed
>> 
>> [Ed: you may be able to provide a simple solution for b).]
>> 
>> a) parser modification
>> 
>> The parser is intended to be extended rather than modified. Be very wary 
>> of modifying; you will have te re-engineer each time MDT-OCL improves. 
>> That said we are in the process of migrating from LPG 1 to LPG 2,
>> https://bugs.eclipse.org/bugs/show_bug.cgi?id=242153, which will require 
>> about four trivial edit procedures to existing extensions.
>> 
>> b) operators
>> 
>> In your Ecore model of custom data types (EClass not EDataType) you can 
>> add operations such as "+", which you can then use them naturally in 
>> OCL. Unfortunately these give a validation error in EMF so you cannot 
>> genModel them.
>> 
>> I'm not sure if the situation is any better for a UML model.
>> 
>> This is a spelling problem I've been considering already: For OCL it 
>> must be "+" for Java it must be perhaps "_plus". (OMG) OCL needs to 
>> specify an encoding scheme for awkward characters and to specify a 
>> concrete syntax access scheme. MDT-OCL supports access via double quotes 
>> e.g:
>> 
>> def: "+"(d : Date) ...
>> 
>> I'm afraid I don't know of any solution to making Date + Date work today.
>> 
>> The XSD support for EMF provides a form of renaming. I've copied this to 
>> Ed Merks who may be able to explain how it could be exploited.
>> 
>> c) call path
>> 
>> I'm not very familiar with the EvaluationEnvironment, so nearly all of 
>> what I'm about to suggest may already be the current MDT-OCL functionality.
>> 
>> You can introduce a derived LOTREnvironment which is created at each 
>> stage of evaluation, with a reference to its outer context. At any point 
>> these references give you the call path. You can then add custom methods 
>> to evaluate any function of the call path that you like. Perhaps you 
>> encapsulate these all in a richer Environment object and implement just 
>> a single .oclEnvironment() method to access it. If you get this working 
>> it might make for a useful contribution to MDT-OCL and a submission for 
>> a future OCL specification.
>> 
>> 	Regards
>> 
>> 		Ed Willink
>> 
>> Elio Damaggio wrote:
>> > Hi all,
>> > 
>> > I am modifying the OCL parser/evaluator in order to (besides other 
things):
>> > 
>> > * include a series of time related datatypes (Date, Time, Duration, 
>> > Timestamp). These datatypes should have standard infix operators (+, -, 
>> > ..) like integers and so on... I think I have to modify the 
>> > EvaluationVisitor. Is this the best course of action?
>> > 
>> > * include predefined methods that, if applied on an attribute, have 
>> > access to the model information of that attribute (i.e. which class they 
>> > belong, in addition to the actual instance).
>> > For instance, lets assume book have an attribute called author. And the 
>> > "LOTR" book instance has "Tolkien" as its author attribute. Then the 
>> > predefined method LOTR.author.info() should be able to know that it is 
>> > invoked on the "author" attribute of a "book" class.
>> > Do you have any idea on how to implement these methods?
>> > 
>> > Thank you all,
>> > 
>> > Elio
>> >