[news.eclipse.technology.dltk] Re: type hierarchy and code completion

[Andrei Sobolev <haiodo@xxxxxxxxx>]

Hi Chuck,

AST tree and Structure model are two different models.
AST is used to represent language internal strucures. Also AST are used to do completion, selection etc.

Structure model are used to represent only top level elements.
This model represent projects, folders, files, classes, methods, field.

Structure model are fixed and could not be extended by DLTK users.
Language implementor could map some langauge elements to structure model elements.
Using of structure model gives some free features like Script Explorer, outline, basic search.

Some time ago we thinked about a way to show custom content fromd  all generic DLTK views.
Then will be possibly to display any kind of information.

>I noticed that IType declares two codeComplete methods.
>Those methods are implemented as dummy/no-op methods.
This is JDT lagacy code, which is not realy used from anythere in DLTK.

How to get AST from ICompletionEngine.complete()
Original idea is to parse content with some advanced completion parser and build some completion model.
For example in Tcl we use Tcl AST tree but with some additional processing, required for completion only.

Two ISourceModule interfaces are also JDT legacy.
JDT guys use similar class then they need completion for some code parts, then no real ISourceModule are exists.
In DLTK we also could use this classes in same context, but not do it yet.
For example completion from wizards, breakpoint configuration view, etc.

>b)    How does one make use of declarations from another module, i.e. Java import?
You could obtain ISourceModule from search, or using some other way. For example for python it will be eacy to get 
current ISourceModule
parent and check for imported module name. etc.

For  examples of how to use search please look to SearchTests class from org.eclipse.dltk.tcl.core.tests project,
and AbstractDLTKSearchTests class from org.eclipse.dltk.core.tests project.

Mixins:
Some of scripting lanuages could have declarations from more then one file.
For example Tcl namespace could be declared from any set of files.
Same thing is for Ruby classes, modules.

Example:
file1.rb:
class MyClass
	def myMethod()
	end
end

file2.rb:
class MyClass
	def myMethod2()
	end
end
In runtime Ruby will see both methods.

Mixin model is intended to handle search in such langauges.
Mixin model in two words:
1) From Script Builder for each type, method, field we report special keys.
Key represent path to element decoded as string, and could contain subkeys.
Example of keys for previous Ruby code:

file1.rb:
MyClass
MyClass{myMethod

file2.rb:
MyClass{
MyClass{myMethod2

In this example MyClass{myMethod2 is subkey for MyClass

2) MixinModel user could ask for elements by key. Mixin will search for all subkeys, and build tree of elements.
For example if we ask for MyClass we wil recieve easy way to iterate all it's children.
Also language mixin parser could report some object instances, and we could do correct completion, selection, without 
additional searchs.

Best regards,
Andrei Sobolev.