Basic OGNL expressions are very simple. The language has become quite rich with features, but you don't generally need to worry about the more complicated parts of the language: the simple cases have remained that way. For example, to get at the name property of an object, the OGNL expression is simply name. To get at the text property of the object returned by the headline property, the OGNL expression is headline.text.
What is a property? Roughly, an OGNL property is the same as a bean property, which means that a pair of get/set methods, or alternatively a field, defines a property (the full story is a bit more complicated, since properties differ for different kinds of objects; see below for a full explanation).
The fundamental unit of an OGNL expression is the navigation chain, usually just called "chain." The simplest chains consist of the following parts:
| Expression Element Part | Example |
|---|---|
| Property names | like the name and headline.text examples above |
| Method Calls | hashCode() to return the current object's hash code |
| Array Indices | listeners[0] to return the first of the current object's list of listeners |
All OGNL expressions are evaluated in the context of a current object, and a chain simply uses the result of the previous link in the chain as the current object for the next one. You can extend a chain as long as you like. For example, this chain:
name.toCharArray()[0].numericValue.toString()
This expression follows these steps to evaluate:
Note that this example can only be used to get a value from an object, not to set a value. Passing the above expression to the Ognl.setValue() method would cause an InappropriateExpressionException to be thrown, because the last link in the chain is neither a property name nor an array index.
This is enough syntax to do the vast majority of what you ever need to do.
This section outlines the details the elements of OGNL's expressions.
OGNL has the following kinds of constants:
OGNL treats different kinds of objects differently in its handling of property references. Maps treat all property references as element lookups or storage, with the property name as the key. Lists and arrays treat numeric properties similarly, with the property name as the index, but string properties the same way ordinary objects do. Ordinary objects (that is, all other kinds) only can handle string properties and do so by using "get" and "set" methods (or "is" and "set"), if the object has them, or a field with the given name otherwise.
Note the new terminology here. Property "names" can be of any type, not just Strings. But to refer to non-String properties, you must use what we have been calling the "index" notation. For example, to get the length of an array, you can use this expression:
array.length
But to get at element 0 of the array, you must use an expression like this:
array[0]
Note that Java collections have some special properties associated with them.
As discussed above, the "indexing" notation is actually just property reference, though a computed form of property reference rather than a constant one.
For example, OGNL internally treats the "array.length" expression exactly the same as this expression:
array["length"]
And this expression would have the same result (though not the same internal form):
array["len" + "gth"]
Array and List Indexing
For Java arrays and Lists indexing is fairly simple, just like in Java. An integer index is given and that element is the referrent. If the index is out of bounds of the array or List and IndexOutOfBoundsException is thrown, just as in Java.
JavaBeans Indexed Properties
JavaBeans supports the concept of Indexed properties. Specifically this means that an object has a set of methods that follow the following pattern:
OGNL can interpret this and provide seamless access to the property through the indexing notation. References such as
someProperty[2]
are automatically routed through the correct indexed property accessor (in the above case through getSomeProperty(2) or setSomeProperty(2, value)). If there is no indexed property accessor a property is found with the name someProperty and the index is applied to that.
OGNL Object Indexed Properties
OGNL extends the concept of indexed properties to include indexing with arbitrary objects, not just integers as with JavaBeans Indexed Properties. When finding properties as candidates for object indexing, OGNL looks for patterns of methods with the following signature:
The PropertyType and IndexType must match each other in the corresponding set and get methods. An actual example of using Object Indexed Properties is with the Servlet API: the Session object has two methods for getting and setting arbitrary attributes:
public Object getAttribute(String name) public void setAttribute(String name, Object value)
An OGNL expression that can both get and set one of these attributes is:
session.attribute["foo"]
OGNL calls methods a little differently from the way Java does, because OGNL is interpreted and must choose the right method at run time, with no extra type information aside from the actual arguments supplied. OGNL always chooses the most specific method it can find whose types match the supplied arguments; if there are two or more methods that are equally specific and match the given arguments, one of them will be chosen arbitrarily.
In particular, a null argument matches all non-primitive types, and so is most likely to result in an unexpected method being called.
Note that the arguments to a method are separated by commas, and so the comma operator cannot be used unless it is enclosed in parentheses. For example,
method( ensureLoaded(), name )
is a call to a 2-argument method, while
method( (ensureLoaded(), name) )
is a call to a 1-argument method.
OGNL has a simple variable scheme, which lets you store intermediate results and use them again, or just name things to make an expression easier to understand. All variables in OGNL are global to the entire expression. You refer to a variable using a number sign in front of its name, like this:
#var
OGNL also stores the current object at every point in the evaluation of an expression in the this variable, where it can be referred to like any other variable. For example, the following expression operates on the number of listeners, returning twice the number if it is more than 100, or 20 more than the number otherwise:
listeners.size().(#this > 100? 2*#this : 20+#this)
OGNL can be invoked with a map that defines initial values for variables. The standard way of invoking OGNL defines the variables root (which holds the initial, or root, object), and context (which holds the Map of variables itself).
To assign a value to a variable explicitly, simply write an assignment statement with a variable reference on the left-hand side:
#var = 99
As you would expect, an expression enclosed in parentheses is evaluated as a unit, separately from any surrounding operators. This can be used to force an evaluation order different from the one that would be implied by OGNL operator precedences. It is also the only way to use the comma operator in a method argument.
If you use a parenthetical expression after a dot, the object that is current at the dot is used as the current object throughout the parenthetical expression. For example,
headline.parent.(ensureLoaded(), name)
traverses through the headline and parent properties, ensures that the parent is loaded and then returns (or sets) the parent's name.
Top-level expressions can also be chained in this way. The result of the expression is the right-most expression element.
ensureLoaded(), name
This will call ensureLoaded() on the root object, then get the name property of the root object as the result of the expression.
To create a list of objects, enclose a list of expressions in curly braces. As with method arguments, these expressions cannot use the comma operator unless it is enclosed in parentheses. Here is an example:
name in { null,"Untitled" }This tests whether the name property is null or equal to "Untitled".
The syntax described above will create a instanceof the List interface. The exact subclass is not defined.
Sometimes you want to create Java native arrays, such as int[] or Integer[]. OGNL supports the creation of these similarly to the way that constructors are normally called, but allows initialization of the native array from either an existing list or a given size of the array.
new int[] { 1, 2, 3 }This creates a new int array consisting of three integers 1, 2 and 3.
To create an array with all null or 0 elements, use the alternative size constructor
new int[5]
This creates an int array with 5 slots, all initialized to zero.
Maps can also be created using a special syntax.
#{ "foo" : "foo value", "bar" : "bar value" }This creates a Map initialized with mappings for "foo" and "bar".
Advanced users who wish to select the specific Map class can specify that class before the opening curly brace
#@java.util.LinkedHashMap@{ "foo" : "foo value", "bar" : "bar value" }The above example will create an instance of the JDK 1.4 class LinkedHashMap, ensuring the the insertion order of the elements is preserved.
OGNL provides a simple way to call the same method or extract the same property from each element in a collection and store the results in a new collection. We call this "projection," from the database term for choosing a subset of columns from a table. For example, this expression:
listeners.{delegate}returns a list of all the listeners' delegates. See the coercion section for how OGNL treats various kinds of objects as collections.
During a projection the #this variable refers to the current element of the iteration.
objects.{ #this instanceof String ? #this : #this.toString()}The above would produce a new list of elements from the objects list as string values.
OGNL provides a simple way to use an expression to choose some elements from a collection and save the results in a new collection. We call this "selection," from the database term for choosing a subset of rows from a table. For example, this expression:
listeners.{? #this instanceof ActionListener}returns a list of all those listeners that are instances of the ActionListener class.
In order to get the first match from a list of matches, you could use indexing such as listeners.{? true }[0]. However, this is cumbersome because if the match does not return any results (or if the result list is empty) you will get an ArrayIndexOutOfBoundsException.
The selection syntax is also available to select only the first match and return it as a list. If the match does not succeed for any elements an empty list is the result.
objects.{^ #this instanceof String }Will return the first element contained in objects that is an instance of the String class.
You can create new objects as in Java, with the new operator. One difference is that you must specify the fully qualified class name for classes other than those in the java.lang package.
This is only true with the default ClassResolver in place. With a custom class resolver packages can be mapped in such a way that more Java-like references to classes can be made. Refer to the OGNL Developer's Guide for details on using ClassResolver class (for example, new java.util.ArrayList(), rather than simply new ArrayList()).
OGNL chooses the right constructor to call using the same procedure it uses for overloaded method calls.
You can call a static method using the syntax @class@method(args). If you leave out class, it defaults to java.lang.Math, to make it easier to call min and max methods. If you specify the class, you must give the fully qualified name.
If you have an instance of a class whose static method you wish to call, you can call the method through the object as if it was an instance method.
If the method name is overloaded, OGNL chooses the right static method to call using the same procedure it uses for overloaded instance methods.
You can refer to a static field using the syntax @class@field. The class must be fully qualified.
If you follow an OGNL expression with a parenthesized expression, without a dot in front of the parentheses, OGNL will try to treat the result of the first expression as another expression to evaluate, and will use the result of the parenthesized expression as the root object for that evaluation. The result of the first expression may be any object; if it is an AST, OGNL assumes it is the parsed form of an expression and simply interprets it; otherwise, OGNL takes the string value of the object and parses that string to get the AST to interpret.
For example, this expression
#fact(30H)
looks up the fact variable, and interprets the value of that variable as an OGNL expression using the BigInteger representation of 30 as the root object. See below for an example of setting the fact variable with an expression that returns the factorial of its argument. Note that there is an ambiguity in OGNL's syntax between this double evaluation operator and a method call. OGNL resolves this ambiguity by calling anything that looks like a method call, a method call. For example, if the current object had a fact property that held an OGNL factorial expression, you could not use this approach to call it
fact(30H)
because OGNL would interpret this as a call to the fact method. You could force the interpretation you want by surrounding the property reference by parentheses:
(fact)(30H)
OGNL has a simplified lambda-expression syntax, which lets you write simple functions. It is not a full-blown lambda calculus, because there are no closures---all variables in OGNL have global scope and extent.
For example, here is an OGNL expression that declares a recursive factorial function, and then calls it:
#fact = :[#this<=1? 1 : #this*#fact(#this-1)], #fact(30H)
The lambda expression is everything inside the brackets. The #this variable holds the argument to the expression, which is initially 30H, and is then one less for each successive call to the expression.
OGNL treats lambda expressions as constants. The value of a lambda expression is the AST that OGNL uses as the parsed form of the contained expression.
There are some special properties of collections that OGNL makes available. The reason for this is that the collections do not follow JavaBeans patterns for method naming; therefore the size(), length(), etc. methods must be called instead of more intuitively referring to these as properties. OGNL corrects this by exposing certain pseudo-properties as if they were built-in.
| Collection | Special Properties |
|---|---|
| Collection (inherited by Map, List & Set) |
|
| List |
|
| Map |
|
| Set |
|
| Iterator |
|
| Enumeration |
|
For the most part, OGNL's operators are borrowed from Java and work similarly to Java's operators. See the OGNL Reference for a complete discussion. Here we describe OGNL operators that are not in Java, or that are different from Java.
ensureLoaded(), name
{ null, true, false }name in {null,"Untitled"} || nameAs stated before, some values that are gettable are not also settable because of the nature of the expression. For example,
names[0].location
is a settable expression - the final component of the expression resolves to a settable property.
However, some expressions, such as
names[0].length + 1
are not settable because they do not resolve to a settable property in an object. It is simply a computed value. If you try to evaluate this expression using any of the Ognl.setValue() methods it will fail with an InappropriateExpressionException.
It is also possible to set variables using get expressions that include the '=' operator. This is useful when a get expression needs to set a variable as a side effect of execution.
Here we describe how OGNL interprets objects as various types. See below for how OGNL coerces objects to booleans, numbers, integers, and collections.
Any object can be used where a boolean is required. OGNL interprets objects as booleans like this:
Numerical operators try to treat their arguments as numbers. The basic primitive-type wrapper classes (Integer, Double, and so on, including Character and Boolean, which are treated as integers), and the "big" numeric classes from the java.math package (BigInteger and BigDecimal), are recognized as special numeric types. Given an object of some other class, OGNL tries to parse the object's string value as a number.
Numerical operators that take two arguments use the following algorithm to decide what type the result should be. The type of the actual result may be wider, if the result does not fit in the given type.
Operators that work only on integers, like the bit-shifting operators, treat their arguments as numbers, except that BigDecimals and BigIntegers are operated on as BigIntegers and all other kinds of numbers are operated on as Longs. For the BigInteger case, the result of these operators remains a BigInteger; for the Long case, the result is expressed as the same type of the arguments, if it fits, or as a Long otherwise.
The projection and selection operators (e1.{e2} and e1.{?e2}), and the in operator, all treat one of their arguments as a collection and walk it. This is done differently depending on the class of the argument:
This section has a fairly detailed treatment of OGNL's syntax and implementation. See below for a complete table of OGNL's operators, a section on how OGNL coerces objects to various types, and a detailed description of OGNL's basic expressions.
OGNL borrows most of Java's operators, and adds a few new ones. For the most part, OGNL's treatment of a given operator is the same as Java's, with the important caveat that OGNL is essentially a typeless language. What that means is that every value in OGNL is a Java object, and OGNL attempts to coerce from each object a meaning appropriate to the situation it is used in (see the section on coercion).
The following table lists OGNL operators in reverse precedence order. When more than one operator is listed in the same box, these operators have the same precedence and are evaluated in left-to-right order.
| Operator | getValue() Notes | setValue() Notes |
|---|---|---|
|
Both e1 and e2 are evaluated with the same source object, and the result of e2 is returned. | getValue is called on e1, and then setValue is called on e2. |
|
getValue is called on e2, and then setValue is called on e1 with the result of e2 as the target object. | Cannot be the top-level expression for setValue. |
|
getValue is called on e1 and the result is interpreted as a boolean. getValue is then called on either e2 or e3, depending on whether the result of e1 was true or false respectively, and the result is returned. | getValue is called on e1, and then setValue is called on either e2 or e3. |
|
getValue is called on e1 and the result is interpreted as a boolean. If true, that result is returned; if false, getValue is called on e2 and its value is returned. | getValue is called on e1; if false, setValue is called on e2. Note that e1 being true prevents any further setting from taking place. |
|
getValue is called on e1 and the result is interpreted as a boolean. If false, that result is returned; if true, getValue is called on e2 and its value is returned. | getValue is called on e1; if true, setValue is called on e2. Note that e1 being false prevents any further setting from taking place. |
|
e1 and e2 are interpreted as integers and the result is an integer. | Cannot be the top-level expression passed to setValue. |
|
e1 and e2 are interpreted as integers and the result is an integer. | Cannot be the top-level expression passed to setValue. |
|
e1 and e2 are interpreted as integers and the result is an integer. | Cannot be the top-level expression passed to setValue. |
|
Equality is tested for as follows. If either value is null, they are equal if and only if both are null. If they are the same object or the equals() method says they are equal, they are equal. If they are both Numbers, they are equal if their values as double-precision floating point numbers are equal. Otherwise, they are not equal. These rules make numbers compare equal more readily than they would normally, if just using the equals method. | Cannot be the top-level expression passed to setValue. |
|
The ordering operators compare with compareTo() if their arguments are non-numeric and implement Comparable; otherwise, the arguments are interpreted as numbers and compared numerically. The in operator is not from Java; it tests for inclusion of e1 in e2, where e2 is interpreted as a collection. This test is not efficient: it iterates the collection. However, it uses the standard OGNL equality test. | Cannot be the top-level expression passed to setValue. |
|
e1 and e2 are interpreted as integers and the result is an integer. | Cannot be the top-level expression passed to setValue. |
|
The plus operator concatenates strings if its arguments are non-numeric; otherwise it interprets its arguments as numbers and adds them. The minus operator always works on numbers. | Cannot be the top-level expression passed to setValue. |
|
Multiplication, division, which interpret their arguments as numbers, and remainder, which interprets its arguments as integers. | Cannot be the top-level expression passed to setValue. |
|
Unary plus is a no-op, it simply returns the value of its argument. Unary minus interprets its argument as a number. Logical not interprets its argument as a boolean. Bitwise not interprets its argument as an integer. The class argument to instanceof is the fully qualified name of a Java class. | Cannot be the top-level expression passed to setValue. |
|
Generally speaking, navigation chains are evaluated by evaluating the first expression, then evaluating the second one with the result of the first as the source object. | Some of these forms can be passed as top-level expressions to setValue and others cannot. Only those chains that end in property references (e.property), indexes (e1[e2]), and subexpressions (e1.(e2)) can be; and expression evaluations can be as well. For the chains, getValue is called on the left-hand expression (e or e1), and then setValue is called on the rest with the result as the target object. |
|
Basic expressions | Only property references (property), indexes ([e]), and variable references (#variable) can be passed as top-level expressions to setValue. For indexes, getValue is called on e, and then the result is used as the property "name" (which might be a String or any other kind of object) to set in the current target object. Variable and property references are set more directly. |