|
org.netbeans.spi.editor.lib2.highlighting
The Highlighting SPI is a new way of influencing how text in an editor component
is rendered. The editor framework in Netbeans is an extension of the Swing Text SPI
framework and as such it uses things like Elements and Views
to render a Document on a screen.
Since the editor framework is primarily designed to support various different
types of files in the IDE it has to give module a chance to participate in documents
rendering. Modules providing support for different languages usually need to influence
colors and fonts of different parts of a source file depending on what code it
contains (i.e. syntax coloring) or what other information the module needs presenting
to a user (e.g. text annotations, hyperlinking, etc.). This all and more can
be achieved by using the Highlighting SPI.
Key parts of the SPI
The very basic idea behind the SPI is to render a document as a sandwich of
independent layers, which will say what colors and font should be used for rendering
particular parts of the document. These parts of the document together with their
rendering attributes (i.e. colors or font) are called highlighted areas
or highlights. Each layer can provide as many non-overlapping highlights
as it likes and each module can provide as many layers as it needs. The implementation
behind the SPI will collect all layers registered for a particular document type
(i.e. mime type), ask each of them for its highlights, merge those highlights
together and finally send them to the draw engine, which will render the document.
The whole SPI is organized around the
HighlightsLayer
class, which is the ultimate thing that modules need to implement in order to provide
a list of highlights for a document. The HighlightsLayers are created
by HighlightsLayerFactory,
which should be registered in MimeLookup under the mime-type of a document
that the layer should be used for. All layers registered for one type of a document
are ordedred according to the
ZOrder
they provide. Besides of ZOrder the layers provide additional information
about nature of highlights they maintain.
The HighlightsLayer class implements
HighlightsContainer
interface, which is the fundamental part of the SPI.
The HighlightsContainer interface allows to get a list of highlighs
and to listen on changes in highlights that it contains.
Besides of HighlightsLayer there are two other implementations of
this interface and they are
OffsetsBag
and
PositionsBag.
Both OffsetsBag and PositionsBag classes allow adding
and removing highlights dynamically. The highlights can be added either one-by-one
or in chunks; each change is reported to listeners.
HighlightsLayer registration
The registration of HighlightsLayers has to be done through an
instance of the HighlightsLayerFactory class. The factory should
be registered in MimeLookup under the mime-type of documents, which
the HighlightsLayer should be used for. For example, if a module
wants to provide HighlightsLayer for text/x-something documents
it should implement its own HighlightsLayerFactory (e.g.
org.some.module.HLFactory class) and register it in MimeLookup
using its XML layer as it is shown on the example below.
<folder name="Editors">
<folder name="text">
<folder name="x-something">
<file name="org-some-module-HLFactory.instance" />
</folder>
</folder>
</folder>
The HLFactory class will simply return a new instance of
the module's implementation of the HighlightsLayer class from its
createLayers
method. The parameter of the createLayers method provides access to
a JTextComponent and its Document, which the layer
is being created for. The method can create and return multiple HighlightsLayers.
HighlightsLayer lifecycle
The lifecycle of HighlightsLayers is tied to the lifecycle of
Document. The infrastructure creates new instances of layers by calling
registered HighlightsLayerFactory objects every time it needs to
visualize a new Document. The layers created for one Document
are not cached or resused in any way. This means that the layers themselvs do not have to take
care about a potential change of a Document instance in
JTextComponent. The infrastructre will always create a new set of
layers if the Document instance changes. Therefore the layers can
simply hold their instance of JTextComponent and/or Document
and treat them as invariants.
Locking and Document changes
The basics of the locking and events model of Swing documents is
described in the javadoc of the
javax.swing.text.AbstractDocument
class. Netbeans documents use the same patterns and so does the Highlighting SPI,
because of its tight relation to documents. The fundamentals of the Swing documents
locking model are that any changes to a document are done under the
document's write lock, the document's listeners are notified synchronously on the
mutating thread and have full read access to the document, but can't modify it.
The main functionality of the Highlighting SPI is to maintain highlights
of certain areas of a document. These highlights are specified as a triple of
starting offset, ending offset and a set of attributes. The offsets are usually passed
in and out accross the SPI boundaries in form of ints and even though
some implementations (e.g. PositionsBag) use
Positions
the esential rule is that any calls in and out from the SPI have to be made
under the document's read lock. Let's have a look on a few examples demonstrating
what this means.
-
Any calls from the infrastructure to
HighlightsLayer.getHighlights()
have to be carried on under the read lock of the document, which the HighlightsLayer
was created for.
-
When
HighlightsLayer needs to notify its listeneres that some
of its highlights have changed all the events have to be fired under the
layer's document's read lock. Obviously, the listeners are not allowed to
modify the document from the event notification methods.
-
Any calls to
OffsetsBag or PositionsBag that modify
their content have to be done under the read lock of the document, which the bag
was created for. Both bags makes all the
changes synchronously on the caller's thread and this even includes firing
notification events. Therefore all events fired from OffsetsBag
or PositionsBag
will be fired under the document's read lock if the mutating thread holds the
read lock.
The Highlighting SPI does not use any special threads and any processing it
does is always done on the caller's thread. This means that the above described
constraints hardly cause any limitation for practical use. The majority of
things happening around a document are done from within
DocumentListeners,
which hold the document's read lock anyway.
The Highlighting SPI is generally thread-safe meaning that any implementation
behind the SPI can be used simultaneously from multiple threads if not stated
otherwise. This doesn't change in any way the rule of acquiring a read lock
before calling the SPI. Swing documents generally allow access for multiple
readers that can run concurrently.
Z-order
Since there can be multiple layers suplying highlights for one document and the
highlights can generally overlap it is important to sort the layers according
to their Z-order. For this purpose each layer has to supply an appropriate
ZOrder.
ZOrder maintains a position of a layer
relatively to other layers as a simple integer number. The higher the number
the higher (more visible) the layer is in the z-order hierarchy.
Instances of the ZOrder class are immutable making it
impossible to dynamically change a position of a layer in the z-order stack
created for a document.
The ZOrder class contains several predefined constants,
which can be used as well-known positions. These constants are called z-order
racks and are meant to be used as a starting point for positioning a layer. An
exact z-order can then be specified by choosing an integer position of the layer
within a rack. The racks are listed below in their respective z-order.
TOP_RACK - the top most rack-
SHOW_OFF_RACK - layers providing short-lived highlights that
can temporarily override highlights from other layers. An example can be
text selection or text search layers.
DEFAULT_RACK - the rack for general layersCARET_RACK - layers that highlight a caret-
SYNTAX_RACK - layers providing syntax or semantic
highlighting of text
BOTOM_RACK - the bottom rack
Using AttributeSet
The Highlighting SPI uses javax.swing.text.AttributeSet to define
attributes for particular highlights. These attributes can be anything,
which the editor's drawing engine understands and can render. Usually the
attribute names are constants from javax.swing.StyleConstants or
org.netbeans.api.editor.settings.EditorStyleConstants. The values
depend on the meaning of each particular attribute, but they usually are instances
of java.awt.Color, java.lang.Integer, Boolean.TRUE
or Boolean.FALSE and similar.
Since there can be more highlighting layers participating on one document and
they can provide highlights that overlap the infrastructure will merge attributes
from all AttributeSets provided for areas with overlapping highlights.
The merging is done in the order defined by ZOrders of the participating
layers, which means that if two layers provide an attribute with the same name
then the merged AttributeSet will contain the attribute from the
layer, which is placed higher in the z-order hierarchy.
There are two important rules for using AttributeSets, which should
be carefully followed by all highlighting layer implementations. Violating these
rules may potentialy break the rendering of a document or may cause performance
problems.
- Immutability - All instances of
AttributeSets should
always be treated as immutable objects. Once you create an AttibuteSet
and use it for a highlight you should not modify it. Your modification is
most likely to be ignored or can have unpredictable results.
- One instance only - There can be thousands of highlights supplied
by layers for one document and a lot of them will be the same (e.g. all
keywords in a java document are highlighted with the same color and font).
It would be unreasonable to create a new
AttributeSet for
each of those highlights when all of them would in fact be the same. Instead
you should always create one instance of AttributeSet and share
it among all highlights that render the same text category (e.g. token or
token category, etc.).
The AttributeSets used for highlighting are often created by calling
FontColorSettings
and it is a responsibility of this class to prevent excesive creation of
AttributeSets it provides. However, if your highlighting layer creates
its own AttributeSets they should always be cached and reused. You can
use methods from the
AttributesUtilities
class for creating immutable AttributeSets.
Use case 1. - Caret selection
The Netbeans editor as any other modern editor allows selecting blocks of text
and highlighting them to a user for easier identification. We call this functionality
caret selection services and it includes things as simple as marking a block of text
that the user selected for copy/paste operation or highlighting a line where
the caret is placed to more complex ones such as highlighting occurences of a text
that the user search for using the 'Find dialog', etc.
This functionality usually only needs to create one highlight and update it
depending on the caret movements/selection notified from JTextComponent.
The more complex cases may need to create several highlights (e.g. to show the
text being searched for). Generally, the highlights are created independently on the
text changes in the document itself (e.g. the caret move or searching for a text).
However, they have to survive editing the document (e.g. the highlighted occurences
of the searched text have to remain highlighted when other parts of the document
are edited).
The caret selection highlights are generally short-lived and have higher importance
than other highlights (e.g. syntax or semantic coloring). They usually change
the background color to highlight the selection, but also retain as much of
a visual appearance of the highlighted text as possible.
Use case 2. - Syntax highlighting
This type of a document coloring shows 'words' or characters in different colors
to indicate their meaning in the structure of the text document. This is very
popular with highly structured documents such as source code files, scripts,
SGML-like documents, etc. It's usually not used for plain text documents containing
text in a human language.
Syntax highlighting in Netbeans editor is based on a lexical analysis done by
lexer plug-ins registered for various types of documents. The lexers are written
using the APIs provided by the Lexer module.
During the lexical analysis text gets split into tokens of different types and
categories. Each token type or category can have defined its own coloring information
such as font and foreground and background colors, etc. Tokens know their position
in text (i.e. offset and length), which information can then be used for creating
highlights.
Decoupling the lexers
and making them pluggable lets the syntax highlighting be very flexible. A single
layer based on the Lexer API can colorify all sorts of documents providing that
there is a lexer registered for each type of a document.
Generally a syntax analysis is very fast and syntax highlighting immediately
reflects changes done in text. The syntax highlighting layer is usually at
the bottom of the hierarchy of highlighting layers.
Use case 3. - Semantic highlighting
In fact semantic coloring regardless of the language it is provided for is
very similar to syntax coloring. Words or groups of characters are highlighted
depending on their meaning in the text. The difference is in the amount of
information that is needed to make this type of coloring meaningful. While with
syntax coloring all the information needed is in the text itself in semantic
coloring parts of text can be colored depending on information found in a
completely different document (e.g. in another source file, library, project, etc.).
Semantic highlighting is highly dependent on the type of a document and therefore
is usually provided on case-by-case basis and only for the most important types
of documents (i.e. those most frequenty used such as java files in Netbeans). Also,
semantic coloring is generally not very fast, because of the amount of information
that is sometimes needed to gather before a document can be colored. Therefore,
while all the effort is made to make semantic coloring reflect text changes as
soon as possible, it is generally done asynchronously outside of the documents
event model and highlights are created as soon as they are available. The tokens
created during the semantic analysis always contain token's position within
the text in some form (i.e. either offset or Position). If Positions
are available they should be accepted and re-used by the Highlighting SPI.
Use case 4. - Embedded languages
An embedded language is a language of a part of a document that is different
than the main language of the document. An example can be a java scriplet in
a JSP page or JavaScript in an HTML document. The main language of a JSP page
is 'text/x-jsp' and the emebedded language in the case of a java scriplet is
'text/x-java'. For the HTML document the main language is 'text/html' and if
a JavaScript part is included in the document the 'text/x-javascript' is the
embedded language.
The language embedding is supported by Lexer API and therefore there is no
problem with supporting it for syntax coloring. For semantic coloring all the
work lies on the highlighting layers providing semantic coloring support for
a particular language. These layers have to be prepared to provide highlights
for parts of a document, which does not contain text in the language they support,
but which contains some embedded parts in that language. The Highlighting
infrastructure will scan the document for all languages it contains and then
it will create appropriate highlighting layers. The layers can be added dynamically
as user inserts parts of text in a new language. The layers, however, may not be
removed immediately when the last part of text in a language they suppor is removed.
Therefore the layers should be prepared to provide no highlights if there is
no text they recognize.
In certain situations JTextComponent or JEditorPane
are used for other purposes than editing. For example debugger may want to show
JEditorPane for adding a new watch, where a user could write a piece
of java code that should evaluated. This pane should use basically the same layers
so that the entered code looks like properly colored and formatted java code.
However, it is not desirable to use exactly the same layers as for an ordinary
java editor, because some highlightings have a little value in this context or
could even be disturbing. There is no point in highlighting the row with the caret,
because watches are essentially one-line expressions. There is also a little point
in showing text-search related highlights, because hardly anybody will use text
search in these simple expressions anyway. On the other hand it makes sense to
highlight selected text if user selects some.
There can be a whole range of usecases where modules need to show an editor
pane, but do not want to use a particular set of highlighting layers, which are
registered for the mime type of text that the module is trying to display and
which would normally be used for an ordinary editor pane. These usecases are
very specific for each module and its way of implementing some features.
The editor insfrastructure supports this usecase through allowing modules to
set special properties on the editor pane that they want to use for displaying text.
The properties are called HighlightsLayerIncludes and HighlightsLayerExcludes.
The value of those properties can be String or String []
of regular expressions that will be used for finding the matching layers by evaluating
each regular expression against the layer's type id. The exact interpretation of those
two properties is described below.
-
HighlightsLayerIncludes - Defines the set of layers that
will be used for rendering text in an editor pane that defines this property.
Every layer, which type Id matches at least one of the regular expressions
defined by this property, will be included for rendering. The default value
is null, which means that all registered layers will be used.
-
HighlightsLayerExcludes - Defines the set of layers that
will not be used for rendering. Every layer, which type Id matches
at least one of the regular expressions defined by this property, will be
excluded from rendering. The default value is null,
which means that no layer will be excluded.
The filters defined by those two properties are used in the same order as they
were listed above. That is the includes are used first and whatever layers they
includ are then filtered by the excludes filter. The result is then used for
rendering text in an editor component, which defined those properties.
The example below shows how to disable the caret row highlighting layer
on JEditorPane.
JEditorPane pane = new JEditorPane();
pane.putClientProperty(
"HighlightsLayerExcludes",
"^org\\.netbeans\\.modules\\.editor\\.lib2\\.highlighting\\.CaretRowHighlighting$"
);
Other usecases
The main usecases described above are certainly not the only usecases of the
Highlighting SPI. In general the SPI can be used for binding any type of
information to parts of text in a document. While this information should have
limited size to keep a good performance of Netbeans editor it can be pretty much
anything. Information provided in highlights is currently used only by the editor's
drawing engine, which provides a limited set of features useful mostly for
rendering text. Some other uses could be for example text annotations, hyperlinking,
showing icons in text, etc.
|