The goal of this topic is to describe the architecture of IntelliJ IDEA from a plugin developer's point of view. It will be organized in a task-based manner: rather than listing all the things that you can do with each object and describing how they are all implemented, it will try to answer questions "what can I do with this object", "how do I get to this object" and so on.
This topic assumes that the reader is familiar with the basic concepts of IntelliJ IDEA plugin development. If you don't know anything at all about plugin development, you should start with the live demo and tutorials at http://www.jetbrains.com/idea/plugins/index.html, and then return to this document.
This topic covers the following subjects:
In general, the data structures of IntelliJ IDEA are covered by a single "multiple readers / single writer" lock. Reading data is allowed from any thread. Reading data from the UI thread does not require any special effort, however, read operations performed from any other thread need to be wrapped in a read action (ApplicationManager.getApplication().runReadAction()). Writing the data is only allowed from the UI thread, and write operations always need to be wrapped in a write action (ApplicationManager.getApplication().runWriteAction()).
To pass control from a background thread to the event dispatch thread, instead of the standard SwingUtilities.invokeLater(), plugins should use ApplicationManager.getApplication().invokeLater(). The latter API allows to specify the modality state for the call - the stack of modal dialogs under which the call is allowed to execute. Passing ModalityState.NON_MODAL means that the operation will be executed after all modal dialogs are closed, and passing ModalityState.stateForComponent() means that the operation may be executed while the specified component (part of a dialog) is still visible.
A virtual file (com.intellij.openapi.vfs.VirtualFile ) is IntelliJ IDEA's representation of a file in a file system (VFS). The most common case of a virtual file is a file in your local file system. However, IDEA supports multiple pluggable file system implementations, so virtual files can also represent classes in a JAR file, old revisions of files loaded from the CVS repository and so on.
The VFS level deals only with binary content. You can get or set the contents of a VirtualFile as a stream of bytes, but concepts like encodings and line separators are handled by higher levels of the system.
- From an action: e.getData(PlatformDataKeys.VIRTUAL_FILE). If you are interested in multiple selection, you can also use e.getData(PlatformDataKeys.VIRTUAL_FILE_ARRAY).
- From a path in the local file system: LocalFileSystem.getInstance().findFileByIoFile()
- From a PSI file: psiFile.getVirtualFile() (may return null if the PSI file exists only in memory)
- From a document: FileDocumentManager.getInstance().getFile()
Traverse the file system, get file contents, rename, move, delete - typical file operations.
Recursive iteration should be performed using VfsUtilCore.iterateChildrenRecursively to prevent endless loops caused by recursive symlinks.
The VFS is built incrementally, by scanning the file system up and down starting from the project root. New files appearing in the file system are detected by VFS refreshes. A refresh operation can be initiated programmatically using (VirtualFileManager.getInstance().refresh() or VirtualFile.refresh()). VFS refreshes are also caused by the file system change notifications received by file system watchers (availble for the Windows and Mac operating systems).
As a plugin developer, you may need to invoke a VFS refresh if you need to access a file that has just been created by an external tool through IDEA's APIs.
A particular file on disk is represented by the equal VirtualFile instances for the entire lifetime of the IDEA process. There might be several instances corresponding to the same file, and they can be garbage-collected. The file is a UserDataHolder, and the user data is shared between those equal instances. If a file is deleted, its corresponding VirtualFile instance becomes invalid ( the isValid() method returns false and operations cause exceptions).
Usually you don't: as a rule, files are created either through the PSI API or through the regular java.io.File API. If you do need to create a file through VFS, you can use the VirtualFile.createChildData() method to create a VirtualFile instance and the VirtualFile.setBinaryContent() method to write some data to the file.
The VirtualFileManager.addVirtualFileListener() method allows you to receive notifications about all changes in the VFS.
To provide an alternative file system implementation (for example, an FTP file system), implement the VirtualFileSystem class (most likely you'll also need to implement VirtualFile), and register your implementation as an application component . To hook into operations performed in the local file system (for example, if you deal with development of a version control system integration that needs custom rename/move handling), implement the LocalFileOperationsHandler interface and register it through the LocalFileSystem.registerAuxiliaryFileOperationsHandler method.
See IntelliJ IDEA Virtual File System for a detailed description of the VFS architecture and usage rules.
Sample plugin projects that illustrate how to work with virtual files are available in the <%IDEA project directory%>/community/samples/vfs and <%IDEA project directory%>/community/samples/textEditor directories.
A document is an editable sequence of Unicode characters, which typically corresponds to the text contents of a virtual file. Line breaks in a document are always normalized to \n. IntelliJ IDEA handles encoding and line break conversions when loading and saving documents transparently.
- From an action: e.getData(PlatformDataKeys.EDITOR).getDocument()
- From a virtual file: FileDocumentManager.getDocument(). This call forces the document content to be loaded from disk if it wasn't loaded previously; if you're only interested in open documents or documents which may have been modified, use FileDocumentManager.getCachedDocument() instead.
- From a PSI file: PsiDocumentManager.getInstance().getDocument() or PsiDocumentManager.getInstance().getCachedDocument()
Any operations which access or modify the file contents on "plain text" level (as a sequence of characters, not as a tree of Java elements).
Document instances are created when some operation needs to access the text contents of a file (in particular, this is needed to build the PSI for a file). Also, document instances not linked to any virtual files can be created temporarily, for example, to represent the contents of a text editor field in a dialog.
Document instances are weakly referenced from VirtualFile instances. Thus, an unmodified Document instance can be garbage-collected if it isn't referenced by anyone, and a new instance will be created if the document contents is accessed again later. Storing Document references in long-term data structures of your plugin will cause memory leaks.
If you need to create a new file on disk, you don't create a document: you create a PSI file and then get its document. If you need to create a document instance which isn't bound to anything, you can use EditorFactory.createDocument.
- Document.addDocumentListener allows you to receive notifications about changes in a particular Document instance.
- EditorFactory.getEventMulticaster().addDocumentListener allows you to receive notifications about changes in all open documents.
- FileDocumentManager.addFileDocumentManagerListener allows you to receive notifications when any document is saved or reloaded from disk.
The general read/write action rules are in effect. In addition to that, any operations which modify the contents of the document must be wrapped in a command (CommandProcessor.getInstance().executeCommand()). executeCommand() calls can be nested, and the outermost executeCommand call is added to the undo stack. If multiple documents are modified within a command, undoing this command will by default show a confirmation dialog to the user.
If the file corresponding to a document is read-only (for example, not checked out from the version control system), document modifications will fail. Thus, before modifying the document, it is necessary to call ReadonlyStatusHandler.getInstance(project).ensureFilesWritable() to check out the file if necessary.
All text strings passed to document modification methods (setText, insertString, replaceString) must use only \n as line separators.
A sample plugin project that illustrates how to work with document files is available in the <%IDEA project directory%>/community/samples/textEditor directory. For more information, see Sample Text File Editor.
A PSI (Program Structure Interface) file is the root of a structure representing the contents of a file as a hierarchy of elements in a particular programming language. The PsiFile class is the common base class for all PSI files, while files in a specific language are usually represented by its subclasses. For example, the PsiJavaFile class represents a Java file, and the XmlFile class represents an XML file.
Unlike VirtualFile and Document, which have application scope (even if multiple projects are open, each file is represented by the same VirtualFile instance), PSI has project scope (the same file is represented by multiple PsiFile instances if the file belongs to multiple projects open at the same time).
- From an action: e.getData(LangDataKeys.PSI_FILE).
- From a VirtualFile: PsiManager.getInstance(project).findFile()
- From a Document: PsiDocumentManager.getInstance(project).getPsiFile()
- From an element inside the file: psiElement.getContainingFile()
- To find files with a specific name anywhere in the project, use FilenameIndex.getFilesByName(project, name, scope)
Most interesting modification operations are performed on the level of individual PSI elements, not files as a whole.
To iterate over the elements in a file, use psiFile.accept(new PsiRecursiveElementWalkingVisitor()...);
As the PSI is language-dependent, PSI files are created through the Language object, using the LanguageParserDefinitions.INSTANCE.forLanguage(language).createFile(fileViewProvider) method.
Like documents, PSI files are created on demand when the PSI is accessed for a particular file.
Again, like documents, PSI files are weakly referenced from the corresponding VirtualFile instances and can be garbage collected if not referenced by anyone.
PsiManager.getInstance(project).addPsiTreeChangeListener() allows you to receive notifications about all changes to the PSI tree of a project.
The PSI can be extended to support additional languages through custom language plugins. Developing such plugins is extensively documented in another article.
Any changes done to the content of PSI files are reflected in documents, so all rules for working with documents (read/write actions, commands, read-only status handling) are in effect.
A file view provider (see the FileViewProvider class) is a new concept in IntelliJ IDEA beginning with version 6.0. Its main purpose is managing access to multiple PSI trees within a single file. For example, a JSPX page has a separate PSI tree for the Java code in it (PsiJavaFile), a separate tree for the XML code (XmlFile), and a separate tree for JSP as a whole (JspFile ). Each of the PSI trees covers the entire contents of the file, and contains special "outer language elements" in the places where contents in a different language can be found.
A FileViewProvider instance corresponds to a single VirtualFile, a single Document, and can be used to retrieve multiple PsiFile instances.
- From a VirtualFile: PsiManager.getInstance(project).findViewProvider()
- From a PSI file: psiFile.getViewProvider()
- To get the list of all languages for which PSI trees exist in a file: fileViewProvider.getLanguages()
- To get the PSI tree for a particular language: fileViewProvider.getPsi(language), where the language parameter can take values of the Language type defined in StdLanguages class. For example, to get the PSI tree for XML, use fileViewProvider.getPsi(StdLanguages.XML).
- To find an element of a particular language at the specified offset in the file: fileViewProvider.findElementAt(offset,language)
To create a file type that has multiple interspersing trees for different languages, your plugin must contain an extension to the fileType.fileViewProviderFactory extension point available in the IntelliJ IDEA core. This extension point is declared using the FileTypeExtensionPoint bean class. To access this extension point, create a Java class that implements the FileViewProviderFactory interface, and in this class, override the createFileViewProvider method.
To declare the extension to the fileType.fileViewProviderFactory extension point, to the <extensions> section of the plugin.xml file, add the following syntax:
where the %file type% refers to the type of the file to create (for example, to "JFS"), and the %class name% refers to the name of your Java class that implements the FileViewProviderFactory interface.
A PSI (Program Structure Interface) file represents a hierarchy of PSI elements (so called PSI trees). A single PSI file may include several PSI trees in a particular programming language. A PSI element, in its turn, can have child PSI elements.
PSI elements and operations on the level of individual PSI elements are used to explore the internal structure of source code as it is interpreted by IntelliJ IDEA. For example, you can use PSI elements to perform the code analysis, such as code inspections or intention actions .
The PsiElement class is the common base class for PSI elements.
- From an action: e.getData(LangDataKeys.PSI_ELEMENT). Note: if an editor is currently open and the element under caret is a reference, this will return the result of resolving the reference. This may or not be what you need.
- From a file by offset: PsiFile.findElementAt(). Note: this returns the lowest level element at the specified offset, which is normally a lexer token. You most likely need to use PsiTreeUtil.getParentOfType() to find the element you really need.
- By iterating through a PSI file: using a PsiRecursiveElementWalkingVisitor.
- By resolving a reference: PsiReference.resolve()
See also PSI Cookbook
See PSI Cookbook
to be continued