Seldom used and expert built-ins

These are the built-ins that normally you should not use, but in exceptional situations (debugging, advanced macros) they can be useful. If you need to use these in your normal page templates, you may revisit the data-model so you don't need to use these.

absolute_template_name

Converts a template name to an absolute name, which can be safely passed to <#include name> or <#import name as ns> or .get_optional_template(name) and such in another template, as it won't be misinterpreted to be relative to the directory of the template that contains the include, import, etc. For example, if you are in template "dir/here.ftl", then "target.ftl" is converted to "/dir/target.ftl" (note the initial /). If now you pass this value to a template in "other-dir/there.ftl", where it's passed to the include directive, then it won't be misinterpreted as "other-dir/target.ftl", like "target.ftl" would have been.

Optionally, you can specify a root based name (a name that's either relative to the template root directory, or is absolute) that will be used instead of the name of the current template, like pathToConver?absolute_template_name(otherTemplateName).

Example of an application (also uses .caller_template_name and .get_optional_template):

<#--
  <@smileyInclude name /> behaves like <#include name>, but prints a "(:" before the
  template, or prints "):" instead if the template is missing.

  Note that just like with #include, if name is relative, it's resolved based on the
  directory of the caller template, not of the template that defines this macro. As
  .get_optional_template resolves relative names based on the current template, we
  had to convert the name to an absolute name based on the caller template before
  passing it to it.
-->
<#macro smileyInclude name>
  <#local t = .get_optional_template(
      name?absolute_template_name(.caller_template_name))>
  <#if t.exists>
    (:
    <@t.include />
  <#else>
    ):
  </#if>
</#macro>

api, has_api

value?api provides access to the API (usually, the Java API) of value, like value?api.someJavaMethod() or value?api.someBeanProperty, if the value itself supports exposing its API. This meant to be used rarely, when you need to call a Java method of an object, but the by-design simplistic view of the value that FreeMarker exposes to the templates hides that, and there's no equivalent built-in either. For example, when you put a Map into the data-model (and you are using the default object wrapper), myMap.myMethod() in a template basically translates to ((Method) myMap.get("myMethod")).invoke(...) in Java, thus you can't call myMethod. If, however, you write myMap?api.myMethod() instead, that means myMap.myMethod() in Java. Similarly, myMap?api.myProperty translates to myMap.getMyProperty() in Java, instead of to myMap.get("myProperty").

You should avoid using api, and rely on the capabilities of the FTL types and the related built-ins as far as possible. For example, don't use users?api.size(), but users?size. The variation that uses ?api is more verbose, slower, more easily breaks when FreeMarker configuration settings are changed, and most importantly, more prone to break as the technical details of the data-model change. For example, if users is changed from a List to an array, users?size will keep working, while users?api.size() will break.

Avoid calling methods that modify an object (especially Map-s and Collection-s) or that aren't thread safe from other reasons. Templates usually aren't expected to modify the objects exposed to them, just to display them. Thus the application may passes some objects to multiple (possibly concurrent) template processings.

The api built-in is not everywhere available, some requirements has to be met:

• The api_builtin_enabled configuration setting must be set to true. Its default is false (at least as of 2.3.22) for not lowering the security of existing applications.

• The value itself has to support it. We are talking about the value as the template sees it, which is created from the original object (that's coming from the data-model or from a Java method return value) value via object wrapping. Hence, this depends on the object_wrapper FreeMarker configuration setting, and on the class of the wrapped (the original) object:

  • When the object wrapper is a DefaultObjectWrapper with its incompatibleImprovements set to 2.3.22 or higher (see how to set it here), FTL values made from Map-s and List-s support ?api. (Actually, what matters is that its useAdaptersForContainer property is set to true, but that's the default with said incompatibleImprovements.) Other java.util.Collections (such as Set-s) only support ?api if DefaultObjectWrapper's forceLegacyNonListCollections property is set to false (the default is true for better out-of-the-box backward compatibility).

  • When wrapped with pure BeansWrapper, all values support ?api.

  • Custom TemplateModel-s can support ?api by implementing the freemarker.template.TemplateModelWithAPISupport interface.

Using ?api when it's not allowed in the configuration or when the value doesn't support it will abort template processing with error.

Whether a value supports ?api can be checked like value?has_api, which returns a boolean value. Note that the result of ?has_api isn't influenced by the api_builtin_enabled setting.

byte, double, float, int, long, short

Returns a SimpleNumber which contains the same value as the original variable, but uses java.lang.Type for the internal representation of the value. This is useful if a method is overloaded, or if a TemplateModel unwrapper has problem with automatically choosing the suitable java.lang.* type. Note that since version 2.3.9 the unwrapper has been improved substantially, so you will hardly ever need to use these built-ins to convert between numerical types, except for resolving ambiguity in overloaded method invocation.

The long built-in can also be used with date, time and date-time values to get the value as java.util.Date.getTime() would return. This is useful if you have to call a Java methods that expect a timestamp as a long.

eval

This built-in evaluates a string as an FTL expression. For example "1+2"?eval returns the number 3. (To render a template that's stored in a string, use the interpret built-in instead.)

The evaluated expression sees the same variables (such as locals) that are visible at the place of the invocation of eval. That is, it behaves similarly as if in place of s?eval you had the value of s there. Except, it can't use loop variable built-ins that refer to a loop variable that was created outside s.

Regarding the configuration settings that affect the parsing (like syntax) and evaluation the rules are the same as with the interpret built-in.

has_content

It is true if the variable exists (and isn't Java null) and is not "empty", otherwise it is false. The meaning of "empty" depends on the concrete case. This follows intuitive common-sense ideas. The following are empty: a string with 0 length, a markup output value with 0 length markup, a sequence or hash with no sub variables, a collection which has passed the last element. If the value is not of any of these types, then it counts as non-empty if it's a number or a date or a boolean (e.g. 0 and false are not empty), otherwise it counts as empty. Note that when your data-model implements multiple template model interfaces you may get unexpected results. However, when in doubt you can use always use expr!?size > 0 or expr!?length > 0 instead of expr?has_content.

This buit-in is exceptional in that you can use the parentheses trick like with the default value operator. That is, you can write both product.color?has_content and (product.color)?has_content. The first doesn't handle the case when product is missing, the last does.

interpret

This built-in parses a string as an FTL template, and returns an user-defined directive that executes that template, just as if a template with that content were include-d at that point. Example:

<#assign x=["a", "b", "c"]>
<#assign templateSource = r"<#list x as y>${y}</#list>">
<#-- Note: That r was needed so that the ${y} is not interpreted above -->
<#assign inlineTemplate = templateSource?interpret>
<@inlineTemplate />

The output:

abc

As you can see, inlineTemplate is a user-defined directive that, when executed, runs the template whose content is the value of templateSource.

The name of the template created by interpret is the name of the template that calls interpret, plus "->anonymous_interpreted". For example, if the template that calls the built-in is "foo/bar.ftl", then the name of the resulting template is "foo/bar.ftl->anonymous_interpreted". Thus, relative paths inside the interpreted template are relative to this path (i.e., the base directory will be "foo"), and errors inside the interpreted template will point to this generated template name.

For more helpful error messages, you can override the template name part after the "->". For example, let's say mailTemplateSource comes from the mail_template database table, and in the case of error, you want the error log to contain the database ID of the failing template:

<#assign inlineTemplate = [mailTemplateSource, "mail_templates id=${mailTemplateId}"]?interpret>

As you can see, interpret can be applied on a sequence of two items, in which case the first item is the FTL string to interpret, and the second items is the template name used after the "->".

The configuration settings that affect the interpreted template are the same as of the surrounding template, except that parser settings specified in the ftl directive or was established via tag syntax or naming convention auto-detection are instead coming from the Configuration object (or naturally, from the TemplateConfiguration, if there's any). Thus the tag syntax, naming convention, whitespace handling, etc. of the interpreted template is independent of that established inside the surrounding template. An important exception from this rule is that the output format and auto-escaping policy is inherited from the lexical context where interpret is called from. For example in a template that has <#ftl output_format="XML"> header (or if you are inside a <#output_format "XML">...</#output_format> block), interpret calls in it will produce directives with XML output format.

is_...

These built-ins check the type of a variable, and returns true or false depending on the type. The list of is_... built-ins:

markup_string

Returns the markup stored inside a markup output value as string. This is useful if the value has to be passed to a Java method for a String parameter, or if we want to manipulate the markup directly in the template. Note that the resulting string can be converted back to markup output value with ?no_esc.

namespace

This built-in returns the namespace (i.e. the "gate" hash to the namespace) associated with a macro or function variable. You can use it with macros and functions only.

new

This is to create a variable of a certain TemplateModel implementation.

On the left side of ? you specify a string, the full-qualified class name of a TemplateModel implementation. The result is a method variable that calls the constructor, and returns the new variable.

Example:

<#-- Creates an user-defined directive be calling the parameterless constructor of the class -->
<#assign word_wrapp = "com.acmee.freemarker.WordWrapperDirective"?new()>
<#-- Creates an user-defined directive be calling the constructor with one numerical argument -->
<#assign word_wrapp_narrow = "com.acmee.freemarker.WordWrapperDirective"?new(40)>

For more information about how the constructor parameters are unwrapped and how overloaded constructor is chosen, read: Programmer's Guide/Miscellaneous/Bean wrapper

This built-in can be a security concern because the template author can create arbitrary Java objects and then use them, as far as they implement TemplateModel. Also the template author can trigger static initialization for classes that don't even implement TemplateModel. You can (since 2.3.17) restrict the classes accessible with this built-in using Configuration.setNewBuiltinClassResolver(TemplateClassResolver) or the new_builtin_class_resolver setting. See the Java API docs for more information. If you are allowing not-so-much-trusted users to upload templates then you should definitely look into this topic.

number_to_date, number_to_time, number_to_datetime

These are used to convert a number (usually a Java long) to a date, time or date-time, respectively. This does them same as new java.util.Date(long) in Java, that is, the number is interpreted as the milliseconds passed since the epoch. The number can be anything and of any type as far as its value fits into a long. If the number isn't a whole number, it will be rounded to whole with half-up rule.

Example:

${1305575275540?number_to_datetime}
${1305575275540?number_to_date}
${1305575275540?number_to_time}

The output will be something like this (depending on the current locale and time zone):

May 16, 2011 3:47:55 PM
May 16, 2011
3:47:55 PM

sequence

This built-in is used to convert a listable value (one that you can iterate through with the list directive) to a more capable sequence value. Sequences support operations like xs[index] and xs?size. Also, the resulting value is listable for multiple times, even if the original value was backed by a java.util.Iterator (which gives error when you try to list it for the 2nd time). This built-in is typically used to work around data-model problems, in case you can't fix the data-model itself. If you can, always fix the data-model instead (give a java.util.List or array to the template instead of a more restricted object, like a non-List java.util.Collection, or a java.util.Iterator).

If the value is already a sequence, then this built-in just returns that as is. If the value is not something that the list directive could list, then template processing will be aborted with error. Otherwise, it usually fetches all the values, and stores them into a sequence. Be careful if you can have a huge number of items, as all of them will be held in memory on the same time. However, in some special cases fetching and/or storing all elements is avoided; see about the optimizations later.

You should convert a value with sequence only once. If you need the resulting sequence at multiple places, always assign the result to a variable, because if the value you convert is only listable once, converting it for the second time will result in error or an empty sequence. Also the conversion is somewhat costly for big collections, so it's better to do it only once.

Example: Let's say you find that users is only listable once (because it's a java.util.Iterator), but you need to list it for multiple times in the template, and you can't fix the data-model. Then you could do this:

<#-- Collect all the users into a sequence: -->
<#assign usersSeq = users?sequence>

<#list usersSeq as user>...</#list>
Again:
<#list usersSeq as user>...</#list>

Optimizations

Since version 2.3.29, if the result of the sequence built-in is directly the input of to the [index] or [range] operator, or of ?size, or of ?first, or a chain of these operations, then the elements will not be collected into the memory, and only as many elements as strictly necessary will be fetched. For example anIterator?sequence[1] will just fetch the first 2 items (instead of building a sequence that contains all the elements, and then getting the 2nd element from that). Or, if you write anIterator?sequence?size, it will just skip through all elements to count them, but won't store them in memory.

The optimizations will only work within the same chain of built-in calls, so for example in <#assign seq = anIterator?sequence>${seq[1]} the ?sequence step will collect all the elements into the memory, as anIterator?sequence and seq[1] are separated. On the other hand, the optimizations will work in anIterator?sequence[10..]?size, as both [range] and ?size supports it, and they are directly chained together.

with_args

The goal of this built-in is to add parameters dynamically to the call of a directive (like a macro), function or method. Dynamically means that parameters are added based on a hash value (like {'a': 1, 'b': 2, 'c': 3} or a Java Map), or a sequence value (like [1, 2, 3] or a Java List), whose actual content is might only known at the moment when the call happens.

For example, we have this macro m:

<#macro m a b c>a=${a}, b=${b}, c=${c}</#macro>

Normally you call it like:

<@m a=1 b=2 c=3 />

Below call does the same, assuming dynArgs is the hash {'a': 1, 'b': 2, 'c': 3}:

<@m?with_args(dynArgs) />

a=1, b=1, c=1

Below call also does the same, but combines dynamic arguments from dynArgsAB, assumed to be {'a': 1, 'b': 2}, and argument c specified directly:

<@m?with_args(dynArgsAB) c=3 />

a=1, b=1, c=1

To understand why this works, you need to realize that macros, custom directives, functions, and methods in FreeMarker are just values, just like numbers, strings, etc. <#macro m ...> just creates a value that's a macro (as opposed to a number, or string, etc.), and then assigns it to variable m. Thus, m in itself is a valid expression, which evaluates to the macro (but it doesn't call the macro). <@m ... /> evaluates the expression m (and you can use arbitrarily complex expressions there too, like m?with_args(...)), and then calls the resulting macro. m?with_args(dynArgs) returns a macro that's very similar to the original macro (that's stored in m), but its arguments default to the values specified in dynArgs. So the result of m?with_args({'b': 22, 'c': 33}) is similar to a modified macro that was created as <#macro unspefiedName a b=22 c=33>. With an example:

<#assign mWithDefs = m?with_args({'b': 22, 'c': 33})>
<@myWithDefs a=1 c='overridden'/>

a=1, b=22, c=overridden

Above we have created a new macro based on the value of m, stored it in variable mWithDefs, and then later we called it with <@myWithDefs ... />.

with_args can also be applied on functions (crated with <#function ...>) and Java methods (usually get from the data-model, like myObject.myMethod). But because functions and methods can only be called with positional arguments (like f(1, 2, 3), and not as f(a=1, b=2, c=3)), the argument to with_args must be a sequence instead of a hash. Other than that, the same tricks work as with macros:

<#function f(a, b, c)><#return "a=${a}, b=${b}, c=${c}"></#function>
<#assign dynArgs=[1, 2, 3]>

${f(1, 2, 3)}
Same as:
${f?with_args(dynArgs)()}
or as:
${f?with_args([1, 2])(3)}
or as:
${f?with_args([1])(2, 3)}

<#assign fWithOneAsFirstArg = f?with_args([1])>
${fWithOneAsFirstArg(2, 3)} <#-- same as f(1, 2, 3) -->

Note the double application of (...) above, like in f?with_args(dynArgs)(). That's because f?with_args(dynArgs) just returns a new function (which is just a value), but doesn't call it. So if you want to call that new function immediately (as opposed to assigning it to a variable for example), you need the second ().

Because macro calls support both named and positional arguments, the with_args argument can be a sequence for macros as well (though using a hash is usually a better practice):

<#macro m a b c>a=${a}, b=${b}, c=${c}</#macro>

<#-- Called with named parameters: -->
<@m a=1 b=2 c=3 />
Same as:
<#-- Called with positional parameters: -->
<@m 1 2 3 />
Same as:
<@m?with_args([1, 2, 3]) />
Same as:
<#-- Sequence with_args with positional c parameter: -->
<@m?with_args([1, 2]) 3 />
Same as:
<#-- Sequence with_args with named c parameter: -->
<@m?with_args([1, 2]) c=3 />

To summarize, depending on the type of the value with_args is applied on, the type of argument to with_args can be:

• Function or method: sequence. Note that WRONG f?with_args(1, 2) is WRONG, the correct form is f?with_args([1, 2]).

• Macro: hash or sequence

• Directive (user defined): hash

The return type of with_args is the same as the type of value it was applied on, like if it's applied on a method (like myObj.myMethod?with_args(dynArgs)), then it returns a method.

Note that it's not possible to apply with_args on built-in directives, like <#if ...>, <#list ...>, etc., because they aren't available as values.

This built-in is often used together with the .args special variable. For example:

<#macro m1 a b c>
  m1 does things with ${a}, ${b}, ${c}
</#macro>

<#macro m2 a b c>
  m2 does things with ${a}, ${b}, ${c}
  Delegate to m1:
  <@m1?with_args(.args) />
</#macro>

<@m2 a=1 b=2 c=3 />

  m2 does things with 1, 2, 3
  Delegate to m1:
  m1 does things with 1, 2, 3

FreeMarker syntax allows using the name before the ?with_args(...) in the end-tag, just as if the ?with_args(...) wasn't there:

<@myMacro?with_args({'a': 1})>...</@myMacro>

Note that as far as the order of arguments is concerned, arguments coming from with_args(...) are added before the arguments specified in the call to the returned directive/function/method. In some use cases it's more desirable to add them at the end instead, in which case use the with_args_last built-in.

with_args_last

Same as with_args, but if the order of the arguments in resulting final argument list may differs (but not the values in it). This only matters if you pass parameters by position (typically, when calling functions or methods), or when there's catch-all argument.

A typical example with positional arguments is when you want to add the dynamic argument to the end of the parameter list:

<#function f a b c d>
  <#return "a=${a}, b=${b}, c=${c}, d=${d}">
</#function>

<#assign dynamicArgs=[3, 4]>

with_args:
${f?with_args(dynamicArgs)(1, 2)}

with_args_last:
${f?with_args_last(dynamicArgs)(1, 2)}

with_args:
a=3, b=4, c=1, d=2

with_args_last:
a=1, b=2, c=3, d=4

In the case of name arguments, while the primary mean of identifying an argument is the its name, catch-all arguments (others... below) still have an order:

<#macro m a b others...>
  a=${a}
  b=${b}
  others:
  <#list others as k, v>
    ${k} = ${v}
  </#list>
</#macro>

<#assign dynamicArgs={'e': 5, 'f': 6}>

with_args:
<@m?with_args(dynamicArgs) a=1 b=2 c=3 d=4 />

with_args_last:
<@m?with_args_last(dynamicArgs) a=1 b=2 c=3 d=4 />

with_args:
  a=1
  b=2
  others:
    e = 5
    f = 6
    c = 3
    d = 4

with_args_last:
  a=1
  b=2
  others:
    c = 3
    d = 4
    e = 5
    f = 6

If you specify a named parameter that are not catch-all, so they are declared in the macro tag (as a and b below), then with_args and with_args_last are no different, since the argument order is specified by the macro definition, not the macro call:

<#macro m a=0 b=0>
  <#-- We use .args to demonstrate the ordering of a and b: -->
  <#list .args as k, v>
    ${k} = ${v}
  </#list>
</#macro>

<#assign dynamicArgs={'b': 1}>

with_args:
<@m?with_args(dynamicArgs) a=1 />

with_args_last:
<@m?with_args_last(dynamicArgs) a=1 />

with_args:
    a = 1
    b = 1

with_args_last:
    a = 1
    b = 1

If both the macro or directive call, and the with_args_last argument specifies named catch-all argument with the same name (like b below), then the placement of those parameters is decide by the macro/directive call:

<#macro m others...>
  <#list others as k, v>
    ${k} = ${v}
  </#list>
</#macro>

<#assign dynamicArgs={'b': 0, 'd': 4}>

with_args:
<@m?with_args(dynamicArgs) a=1 b=2 c=3 />

with_args_last:
<@m?with_args_last(dynamicArgs) a=1 b=2 c=3 />

with_args:
    b = 2
    d = 4
    a = 1
    c = 3

with_args_last:
    a = 1
    b = 2
    c = 3
    d = 4