Jump to: navigation, search

DLTK Core Architecture

Revision as of 01:28, 29 August 2007 by Fourdman.xored.com (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Build paths

Similar to java class paths, DLTK has a concept of build paths.

Build paths is a set of source folders, library containers and references to another projects. This paths are used as for model building, so for launching.

Paths is stored in .buildpath file from project's root folder. DLTK automatically parses it when required. You can get project build paths via IScriptProject.getRawBuildpath() method. To set buildpath to project you should use setRawBuildpath() method. Elements of IBuildpathEntry may be created with DLTKCore.new*Entry(...) methods.

Model

Model is a key thing in DLTK you should know about. In fact, as you probably know, DLTK model had been cloned from JDT model. So, if you are familiar with JDT model, you will understand everything here quickly. As JDT, DLTK uses an in-memory object model to represent the workspace structure from projects level to source file internals. This structure is derived from the project's build path. The model is hierarchical. The following table summarizes the different kinds of model elements. All elements classes support the IModelElement interface.

Element JDT-analog Description
IScriptModel IJavaModel Represents the root model element, corresponding to the

workspace. The parent of all projects with the script natures. It also

gives you access to the projects without the script nature.
IScriptProject IJavaProject Represents a script project in the workspace. (Child of IScriptModel)
IProjectFragment IPackageFragmentRoot Represents a project fragment, and maps the contents to an

underlying resource which is either a folder, JAR, or ZIP file. (Child

of IScriptProject)
IScriptFolder IPackageFragment Represents a folder containing script files inside. (Child of IProjectFragment )
ISourceModule ICompilationUnit Represents a source file. (Child of IScriptFolder)
IPackageDeclaration IPackageDeclaration Represents a package declaration in a source module. (Child of ISourceModule )
IType IType Represents either a class/module/namespace inside a source file.
IField IField Represents a field inside a type. (Child of IType)
IMethod IMethod Represents a method or constructor inside a type. (Child of IType)

You can use DLTKCore.create(...) methods to access model. These methods allow to create appropriate model element from file, resource or project.

Model creation

DLTK Model elements from workspace level down to source modules level are built automatically. All things, that you should do for that:

  • Contribute IDLTKLanguageToolkit interface implementation via org.eclipse.dltk.core.language extension point.
  • Set correct nature as in extension point attribute, so in getNatureId() method.
  • Methods validateSourceModule() and validateSourcePackage() should return OK if source module/package is a real source module, or anything else if it's a regular file/folder.

After that, projects, that have given nature will be considered as script project and model for them will be built accordingly to internal structure, validate...() methods and build paths.

For building source module's internal model there is another mechanism called source element parsers. They are contributed via org.eclipse.dltk.core.sourceElementParsers ext. point and should implement ISourceElementParser.

Main task of source element parser is to parse source file and report internal model elements info to given ISourceElementRequestor.

Search engine

Indexes

Indexes is a main facility which search engine uses. Index is a set of documents and keys associated with them. There are possible several different indexes (for type names, for methods, ...).

DLTK automatically index all source files in a separate thread. It uses standard source element parser with a requester set to SourceIndexerRequestor object. Source element parser doesn't know anything about search and just reports model elements info. Task of SourceIndexerRequestor is to report appropriate index keys to given SourceIndexer. User may extend SourceIndexerRequestor if it's required.

Search

Before using the search engine user should prepare a special object: search pattern. It's a SearchPattern class object. It may be a TypeDeclarationPattern or may be MethodPattern. Static method SearchPattern.createPattern() may be used for that purpose. After that user should specify a search scope(project, workspace,...). It may be created with a SearchEngine.createSearchScope() method. And final item is a search requestor. It's a SearchRequestor object, that will receive all successfull search matches. After that SearchEngine::search() may be called. Here is an example:

		SearchRequestor requestor = new SearchRequestor() {

			public void acceptSearchMatch(SearchMatch match)
					throws CoreException {
				// process match
			}

		};
		SearchPattern pattern = SearchPattern.createPattern(namePattern,
				IDLTKSearchConstants.METHOD, IDLTKSearchConstants.DECLARATIONS,
				SearchPattern.R_PATTERN_MATCH | SearchPattern.R_EXACT_MATCH);
		IDLTKSearchScope scope = SearchEngine.createWorkspaceScope(RubyLanguageToolkit
					.getDefault());
		try {
			SearchEngine engine = new SearchEngine();
			engine.search(pattern, new SearchParticipant[] { SearchEngine
					.getDefaultSearchParticipant() }, scope, requestor, null);
		} catch (CoreException e) {
			if (DLTKCore.DEBUG)
				e.printStackTrace();
		}

How search works?

Extending a search engine requires understanding of how search engine works. After you had called search() method, a special PatternSearchJob are being created. Inside it all indexes are being enumerated. As result, list of documents, containing a matching key will be received. After that, using a MatchLocator class each document will be reparsed and appropriate SearchMatch objects reported. User given MatchLocatorParser are being used for reparsing. It knows a MatchLocator and PossibleMatch objects and while parsing should call match(...) method from the locator.

Runtime model ("mixin" model)

DLTK has a very simple, but really powerful structure for managing runtime source files model. If model elements may be constructed from several source files(that's why model is called "mixin") or modified during execution, this approach can help.

In fact, mixin-model is built on top of the standard indexes facility. Mixin-parser(IMixinParser implementation, ext. point org.eclipse.dltk.core.mixin) reports keys(String objects) while parsing. One key may be reported many times from many places. To each key some Object may be attached. And... that's all. DLTK worries about everything else.

Lets consider the following example for Ruby.

file1.rb

class Foo # key "Foo" reported, IType object attached
end

class Foo # key "Foo" reported, IType object attached
	def doo # key "Foo{doo" reported, IMethod object attached
	end
end

file2.rb

class Foo # key "Foo" reported, IType object attached
	def doo2 # key "Foo{doo2" reported, IMethod object attached
	end
end

Now, if we'll ask model for Foo key(MixinModel.get() method), we'll fetch IMixinElement with information, that this key has been reported from file1.rb and file2.rb and we'll be able to get every IType object attached. More. "{" is a standard delimeter, so get can call getChilds() and fetch information about "Foo{doo" and "Foo{doo2" keys.

Type inference

DLTK has a language independent engine for building type inference systems. It uses demand-driven analysis with subgoal pruning algorithm. Key abstractions here:

  • Goal. It may be type of some ASTNode in source file, incoming/outcoming data flow, whatever. Goals are unique. So, two object of IGoal having a common ASTNode object is equal.
  • Goal evaluator. Strategy, that knows how to evaluate goal. While evaluating it may produce another helper goals (sub-goals), may wait for their results and only after that produce it's own result.
  • Goal evaluator factory. This factory constructs a GoalEvaluator objects for given IGoals.
  • Pruner. Pruner is a straregy that are able to cancel some evaluators as non-important. Also using pruner is possible to implement time limits for evaluations.

// TODO

Launching

Interpreter management

Each interpreter installation in system is stored inside a IInterpreterInstall object. Such object knows interpreter name, executable path, arguments, library paths and also installation type and IInterpreterRunner object. Installation type is represented via IInterpreterInstallType object. Installation type knows about all installations with such type, knows how to fetch default library locations and able to validate installations.

For each language, DLTK stores a separate set of IInterpreterInstalls. One of them should be marked as "default".

The key class for accessing installation is a ScriptRuntime.

However, ScriptRuntime allows only to fetch installs, not to modify them. In fact, there is no beautiful way for that. So here we'll describe relatively low-level way to do that.

All information about interpreter installs are stored in plugin preferences in XML format. XML data can be read/created using an InterpreterDefinitionsContainer. This class represent a set of interpreter installs and allows to read them from XML or store to XML.

For setting set of interpreters in preferences, there are exists an IntepreterUpdater class. It takes an InterpreterDefinitionsContainer object and stores in plugin preferences. After that ScriptRuntime will be able to read new values.

Launching a script

DLTK launching engine is fully based on standard Eclipse launching, so there are not a lot of new things are here.

Cause each InterpreterInstall has an IInterpreterRunner object, launch process looks like following:

  • get selected interpreter install info from launch configuration (it should be put there before)
  • get runner
  • launch it

In fact, all this stuff is already implemented. First, there is a AbstractScriptLaunchConfigurationDelegate class. It requires only two methods from user: getLanguageId() and createInterpreterConfig(). InterpreterConfig is a simple structure containing information for launch in a low-level form.

Also, for IInterpreterRunner exists a AbstractInterpreterRunner class, that requires from user only a launching plugin id and created process type.

For launching scripts programmatically there is a ScriptLaunchUtil class. It contains lots of methods for launching scripts.

DLTK AST

ASTNote Superclass of all the ast nodes.
ASTListNode Represents list of nodes.
ModuleDeclaration Top-level node for a source file.
TypeDeclaration Declaration of class/module/namespace.
MethodDeclaration Declaration of procedure or method.


Other classes you can find in org.eclipse.dltk.ast.* packages. Usage of DLTK AST is not mandatory, but some DLTK features like folding may rely on it and greatly simplify implementation.