Template engines
1. Introduction
Groovy supports multiple ways to generate text dynamically including GStrings
, printf
and MarkupBuilder just to name a few.
In addition to these, there is a dedicated template framework which is well-suited to applications where the text to be generated follows the form of a static template.
2. Template framework
The template framework in Groovy consists of a TemplateEngine
abstract base class that engines must implement
and a Template
interface that the resulting templates they generate must implement.
Included with Groovy are several template engines:
-
SimpleTemplateEngine
- for basic templates -
StreamingTemplateEngine
- functionally equivalent toSimpleTemplateEngine
, but can handle strings larger than 64k -
GStringTemplateEngine
- stores the template as writeable closures (useful for streaming scenarios) -
XmlTemplateEngine
- works well when the template and output are valid XML -
MarkupTemplateEngine
- a very complete, optimized, template engine
3. SimpleTemplateEngine
Shown here is the SimpleTemplateEngine
that allows you to use JSP-like scriptlets (see example below), script, and EL expressions
in your template in order to generate parametrized text. Here is an example of using the system:
def text = 'Dear "$firstname $lastname",\nSo nice to meet you in <% print city %>.\nSee you in ${month},\n${signed}'
def binding = ["firstname":"Sam", "lastname":"Pullara", "city":"San Francisco", "month":"December", "signed":"Groovy-Dev"]
def engine = new groovy.text.SimpleTemplateEngine()
def template = engine.createTemplate(text).make(binding)
def result = 'Dear "Sam Pullara",\nSo nice to meet you in San Francisco.\nSee you in December,\nGroovy-Dev'
assert result == template.toString()
While it is generally not deemed good practice to mix processing logic in your template (or view), sometimes very simple logic can be useful. E.g. in the example above, we could change this:
$firstname
to this (assuming we have set up a static import for capitalize inside the template):
${firstname.capitalize()}
or this:
<% print city %>
to this:
<% print city == "New York" ? "The Big Apple" : city %>
3.1. Advanced Usage Note
If you happen to be embedding your template directly in your script (as we did above) you have to be careful about backslash escaping. Because the template string itself will be parsed by Groovy before it is passed to the templating framework, you have to escape any backslashes inside GString expressions or scriptlet 'code' that are entered as part of a Groovy program. E.g. if we wanted quotes around The Big Apple above, we would use:
<% print city == "New York" ? "\\"The Big Apple\\"" : city %>
Similarly, if we wanted a newline, we would use:
\\n
in any GString expression or scriptlet 'code' that appears inside a Groovy script. A normal “\n” is fine within the static template text itself or if the entire template itself is in an external template file. Similarly, to represent an actual backslash in your text you would need
\\\\
in an external file or
\\\\
in any GString expression or scriptlet 'code'. (Note: the necessity to have this extra slash may go away in a future version of Groovy if we can find an easy way to support such a change.)
4. StreamingTemplateEngine
The StreamingTemplateEngine
engine is functionally equivalent to the SimpleTemplateEngine
,
but creates the template using writable closures making it more scalable for large templates.
Specifically this template engine can handle strings larger than 64k.
It uses JSP style <% %> script and <%= %> expression syntax or GString style expressions. The variable 'out' is bound to the writer that the template is being written to.
Frequently, the template source will be a file but here we show a simple example providing the template as a string:
def text = '''\
Dear <% out.print firstname %> ${lastname},
We <% if (accepted) out.print 'are pleased' else out.print 'regret' %> \
to inform you that your paper entitled
'$title' was ${ accepted ? 'accepted' : 'rejected' }.
The conference committee.'''
def template = new groovy.text.StreamingTemplateEngine().createTemplate(text)
def binding = [
firstname : "Grace",
lastname : "Hopper",
accepted : true,
title : 'Groovy for COBOL programmers'
]
String response = template.make(binding)
assert response == '''Dear Grace Hopper,
We are pleased to inform you that your paper entitled
'Groovy for COBOL programmers' was accepted.
The conference committee.'''
5. GStringTemplateEngine
As an example of using the GStringTemplateEngine
, here is the example above done again (with a few changes to show some other options).
First we will store the template in a file this time:
Dear "$firstname $lastname",
So nice to meet you in <% out << (city == "New York" ? "\\"The Big Apple\\"" : city) %>.
See you in ${month},
${signed}
Note that we used out
instead of print
to support the streaming nature of GStringTemplateEngine
.
Because we have the template in a separate file, there is no need to escape the backslashes. Here is how we call it:
def f = new File('test.template')
def engine = new groovy.text.GStringTemplateEngine()
def template = engine.createTemplate(f).make(binding)
println template.toString()
and here is the output:
Dear "Sam Pullara", So nice to meet you in "The Big Apple". See you in December, Groovy-Dev
6. XmlTemplateEngine
XmlTemplateEngine
for use in templating scenarios where both the template source and the expected output are intended to be XML.
Templates may use the normal ${expression}
and $variable
notations to insert an arbitrary expression into the template.
In addition, support is also provided for special tags: <gsp:scriptlet>
(for inserting code fragments) and <gsp:expression>
(for code fragments which produce output).
Comments and processing instructions will be removed as part of processing and special XML characters such
as <
, >
, "
and '
will be escaped using the respective XML notation.
The output will also be indented using standard XML pretty printing.
The xmlns namespace definition for gsp: tags will be removed but other namespace definitions will be preserved (but may change to an equivalent position within the XML tree).
Normally, the template source will be in a file but here is a simple example providing the XML template as a string:
def binding = [firstname: 'Jochen', lastname: 'Theodorou', nickname: 'blackdrag', salutation: 'Dear']
def engine = new groovy.text.XmlTemplateEngine()
def text = '''\
<document xmlns:gsp='http://groovy.codehaus.org/2005/gsp' xmlns:foo='baz' type='letter'>
<gsp:scriptlet>def greeting = "${salutation}est"</gsp:scriptlet>
<gsp:expression>greeting</gsp:expression>
<foo:to>$firstname "$nickname" $lastname</foo:to>
How are you today?
</document>
'''
def template = engine.createTemplate(text).make(binding)
println template.toString()
This example will produce this output:
<document type='letter'>
Dearest
<foo:to xmlns:foo='baz'>
Jochen "blackdrag" Theodorou
</foo:to>
How are you today?
</document>
7. The MarkupTemplateEngine
This template engine is a template engine primarily aimed at generating XML-like markup (XML, XHTML, HTML5, …), but that can be used to generate any text based content. Unlike traditional template engines, this one relies on a DSL that uses the builder syntax. Here is a sample template:
xmlDeclaration()
cars {
cars.each {
car(make: it.make, model: it.model)
}
}
If you feed it with the following model:
model = [cars: [new Car(make: 'Peugeot', model: '508'), new Car(make: 'Toyota', model: 'Prius')]]
It would be rendered as:
<?xml version='1.0'?>
<cars><car make='Peugeot' model='508'/><car make='Toyota' model='Prius'/></cars>
The key features of this template engine are:
-
a markup builder like syntax
-
templates are compiled into bytecode
-
fast rendering
-
optional type checking of the model
-
includes
-
internationalization support
-
fragments/layouts
7.1. The template format
7.1.1. Basics
Templates consist of Groovy code. Let’s explore the first example more thoroughly:
xmlDeclaration() (1)
cars { (2)
cars.each { (3)
car(make: it.make, model: it.model) (4)
} (5)
}
1 | renders the XML declaration string. |
2 | opens a cars tag |
3 | cars is a variable found in the template model, which is a list of Car instances |
4 | for each item, we create a car tag with the attributes from the Car instance |
5 | closes the cars tag |
As you can see, regular Groovy code can be used in the template. Here, we are calling each
on a list (retrieved from the model), allowing us to
render one car
tag per entry.
In a similar fashion, rendering HTML code is as simple as this:
yieldUnescaped '<!DOCTYPE html>' (1)
html(lang:'en') { (2)
head { (3)
meta('http-equiv':'"Content-Type" content="text/html; charset=utf-8"') (4)
title('My page') (5)
} (6)
body { (7)
p('This is an example of HTML contents') (8)
} (9)
} (10)
1 | renders the HTML doctype special tag |
2 | opens the html tag with an attribute |
3 | opens the head tag |
4 | renders a meta tag with one http-equiv attribute |
5 | renders the title tag |
6 | closes the head tag |
7 | opens the body tag |
8 | renders a p tag |
9 | closes the body tag |
10 | closes the html tag |
The output is straightforward:
<!DOCTYPE html><html lang='en'><head><meta http-equiv='"Content-Type" content="text/html; charset=utf-8"'/><title>My page</title></head><body><p>This is an example of HTML contents</p></body></html>
With some configuration, you can have the output pretty printed, with newlines and indent automatically added. |
7.1.2. Support methods
In the previous example, the doctype declaration was rendered using the yieldUnescaped
method. We have also seen the xmlDeclaration
method.
The template engine provides several support methods that will help you render contents appropriately:
Method | Description | Example |
---|---|---|
yield |
Renders contents, but escapes it before rendering |
Template:
Output:
|
yieldUnescaped |
Renders raw contents. The argument is rendered as is, without escaping. |
Template:
Output:
|
xmlDeclaration |
Renders an XML declaration String. If the encoding is specified in the configuration, it is written in the declaration. |
Template:
Output:
If Output:
|
comment |
Renders raw contents inside an XML comment |
Template:
Output:
|
newLine |
Renders a new line. See also |
Template:
Output:
|
pi |
Renders an XML processing instruction. |
Template:
Output:
|
tryEscape |
Returns an escaped string for an object, if it is a |
Template:
Output:
|
7.1.3. Includes
The MarkupTemplateEngine
supports inclusion of contents from another file. Included contents may be:
-
another template
-
raw contents
-
contents to be escaped
Including another template can be done using:
include template: 'other_template.tpl'
Including a file as raw contents, without escaping it, can be done like this:
include unescaped: 'raw.txt'
Eventually, inclusion of text that should be escaped before rendering can be done using:
include escaped: 'to_be_escaped.txt'
Alternatively, you can use the following helper methods instead:
-
includeGroovy(<name>)
to include another template -
includeEscaped(<name>)
to include another file with escaping -
includeUnescaped(<name>)
to include another file without escaping
Calling those methods instead of the include xxx:
syntax can be useful if the name of the file to be included is dynamic (stored in a variable for example).
Files to be included (independently of their type, template or text) are found on classpath. This is one of the reasons why the MarkupTemplateEngine
takes
an optional ClassLoader
as constructor argument (the other reason being that you can include code referencing other classes in a template).
If you don’t want your templates to be on classpath, the MarkupTemplateEngine
accepts a convenient constructor that lets you define the directory where
templates are to be found.
7.1.4. Fragments
Fragments are nested templates. They can be used to provide improved composition in a single template. A fragment consists of a string, the inner template, and a model, used to render this template. Consider the following template:
ul {
pages.each {
fragment "li(line)", line:it
}
}
The fragment
element creates a nested template, and renders it with a model which is specific to this template. Here,
we have the li(line)
fragment, where line
is bound to it
. Since it
corresponds to the iteration of pages
,
we will generate a single li
element for each page in our model:
<ul><li>Page 1</li><li>Page 2</li></ul>
Fragments are interesting to factorize template elements. They come at the price of the compilation of a fragment per template, and they cannot be externalized.
7.1.5. Layouts
Layouts, unlike fragments, refer to other templates. They can be used to compose templates and share common structures. This is often interesting if you have, for example, a common HTML page setup, and that you only want to replace the body. This can be done easily with a layout. First of all, you need to create a layout template:
html {
head {
title(title) (1)
}
body {
bodyContents() (2)
}
}
1 | the title variable (inside the title tag) is a layout variable |
2 | the bodyContents call will render the body |
Then what you need is a template that includes the layout:
layout 'layout-main.tpl', (1)
title: 'Layout example', (2)
bodyContents: contents { p('This is the body') } (3)
1 | use the main-layout.tpl layout file |
2 | set the title variable |
3 | set the bodyContents |
As you can see, bodyContents
will be rendered inside the layout, thanks to the bodyContents()
call in the layout file. As
a result, the template will be rendered as this:
<html><head><title>Layout example</title></head><body><p>This is the body</p></body></html>
The call to the contents
method is used to tell the template engine that the block of code is in fact a specification of a
template, instead of a helper function to be rendered directly. If you don’t add contents
before your specification, then
the contents would be rendered, but you would also see a random string generated, corresponding to the result value of the block.
Layouts are a powerful way to share common elements across multiple templates, without having to rewrite everything or use includes.
Layouts use, by default, a model which is independent from the model of the page where they are used. It is however possible to make them inherit from the parent model. Imagine that the model is defined like this:
model = new HashMap<String,Object>();
model.put('title','Title from main model');
and the following template:
layout 'layout-main.tpl', true, (1)
bodyContents: contents { p('This is the body') }
1 | note the use of true to enable model inheritance |
then it is not necessary to pass the title
value to the layout as in the previous example. The result will be:
<html><head><title>Title from main model</title></head><body><p>This is the body</p></body></html>
But it is also possible to override a value from the parent model:
layout 'layout-main.tpl', true, (1)
title: 'overridden title', (2)
bodyContents: contents { p('This is the body') }
1 | true means inherit from the parent model |
2 | but title is overridden |
then the output will be:
<html><head><title>overridden title</title></head><body><p>This is the body</p></body></html>
7.2. Rendering contents
7.2.1. Creation of a template engine
On the server side, rendering templates require an instance of groovy.text.markup.MarkupTemplateEngine
and a
groovy.text.markup.TemplateConfiguration
:
TemplateConfiguration config = new TemplateConfiguration(); (1)
MarkupTemplateEngine engine = new MarkupTemplateEngine(config); (2)
Template template = engine.createTemplate("p('test template')"); (3)
Map<String, Object> model = new HashMap<>(); (4)
Writable output = template.make(model); (5)
output.writeTo(writer); (6)
1 | creates a template configuration |
2 | creates a template engine with this configuration |
3 | creates a template instance from a String |
4 | creates a model to be used in the template |
5 | bind the model to the template instance |
6 | render output |
There are several possible options to parse templates:
-
from a
String
, usingcreateTemplate(String)
-
from a
Reader
, usingcreateTemplate(Reader)
-
from a
URL
, usingcreateTemplate(URL)
-
given a template name, using
createTemplateByPath(String)
The last version should in general be preferred:
Template template = engine.createTemplateByPath("main.tpl");
Writable output = template.make(model);
output.writeTo(writer);
7.2.2. Configuration options
The behavior of the engine can be tweaked with several configuration options accessible through the TemplateConfiguration
class:
Option | Default value | Description | Example |
---|---|---|---|
declarationEncoding |
null |
Determines the value of the encoding to be written when |
Template:
Output:
If Output:
|
expandEmptyElements |
false |
If true, empty tags are rendered in their expanded form. |
Template:
Output:
If Output:
|
useDoubleQuotes |
false |
If true, use double quotes for attributes instead of simple quotes |
Template:
Output:
If Output:
|
newLineString |
System default (system property |
Allows to choose what string is used when a new line is rendered |
Template:
If Output:
|
autoEscape |
false |
If true, variables from models are automatically escaped before rendering. |
|
autoIndent |
false |
If true, performs automatic indentation after new lines |
|
autoIndentString |
four (4) spaces |
The string to be used as indent. |
|
autoNewLine |
false |
If true, performs automatic insertion of new lines |
|
baseTemplateClass |
|
Sets the super class of compiled templates. This can be used to provide application specific templates. |
|
locale |
Default locale |
Sets the default locale for templates. |
Once the template engine has been created, it is unsafe to change the configuration. |
7.2.3. Automatic formatting
By default, the template engine will render output without any specific formatting. Some configuration options can improve the situation:
-
autoIndent
is responsible for auto-indenting after a new line is inserted -
autoNewLine
is responsible for automatically inserting new lines based on the original formatting of the template source
In general, it is recommended to set both autoIndent
and autoNewLine
to true if you want human-readable, pretty printed, output:
config.setAutoNewLine(true);
config.setAutoIndent(true);
Using the following template:
html {
head {
title('Title')
}
}
The output will now be:
<html>
<head>
<title>Title</title>
</head>
</html>
We can slightly change the template so that the title
instruction is found on the same line as the head
one:
html {
head { title('Title')
}
}
And the output will reflect that:
<html>
<head><title>Title</title>
</head>
</html>
New lines are only inserted where curly braces for tags are found, and the insertion corresponds to where the nested content is found. This means that tags in the body of another tag will not trigger new lines unless they use curly braces themselves:
html {
head {
meta(attr:'value') (1)
title('Title') (2)
newLine() (3)
meta(attr:'value2') (4)
}
}
1 | a new line is inserted because meta is not on the same line as head |
2 | no new line is inserted, because we’re on the same depth as the previous tag |
3 | we can force rendering of a new line by explicitly calling newLine |
4 | and this tag will be rendered on a separate line |
This time, the output will be:
<html>
<head>
<meta attr='value'/><title>Title</title>
<meta attr='value2'/>
</head>
</html>
By default, the renderer uses four(4) spaces as indent, but you can change it by setting the TemplateConfiguration#autoIndentString
property.
7.2.4. Automatic escaping
By default, contents which is read from the model is rendered as is. If this contents comes from user input, it can be sensible, and you might
want to escape it by default, for example to avoid XSS injection. For that, the template configuration provides an option which will automatically
escape objects from the model, as long as they inherit from CharSequence
(typically, `String`s).
Let’s imagine the following setup:
config.setAutoEscape(false);
model = new HashMap<String,Object>();
model.put("unsafeContents", "I am an <html> hacker.");
and the following template:
html {
body {
div(unsafeContents)
}
}
Then you wouldn’t want the HTML from unsafeContents
to be rendered as is, because of potential security issues:
<html><body><div>I am an <html> hacker.</div></body></html>
Automatic escaping will fix this:
config.setAutoEscape(true);
And now the output is properly escaped:
<html><body><div>I am an <html> hacker.</div></body></html>
Note that using automatic escaping doesn’t prevent you from including unescaped contents from the model. To do this, your template should then explicitly
mention that a model variable should not be escaped by prefixing it with unescaped.
, like in this example:
html {
body {
div(unescaped.unsafeContents)
}
}
7.2.5. Common gotchas
Strings containing markup
Say that you want to generate a <p>
tag which contains a string containing markup:
p {
yield "This is a "
a(href:'target.html', "link")
yield " to another page"
}
and generates:
<p>This is a <a href='target.html'>link</a> to another page</p>
Can’t this be written shorter? A naive alternative would be:
p {
yield "This is a ${a(href:'target.html', "link")} to another page"
}
but the result will not look as expected:
<p><a href='target.html'>link</a>This is a to another page</p>
The reason is that the markup template engine is a streaming engine. In the original version, the first yield
call
generates a string which is streamed to the output, then the a
link is generated and streamed, and then the last yield
call is streamed, leading in an execution in order. But with the string version above, the order of execution is different:
-
the
yield
call requires an argument, a string -
that arguments need to be evaluated before the yield call is generated
so evaluating the string leads to an execution of the a(href:…)
call before yield
is itself called. This is not
what you want to do. Instead, you want to generate a string which contains markup, which is then passed to the yield
call. This can be done this way:
p("This is a ${stringOf {a(href:'target.html', "link")}} to another page")
Note the stringOf
call, which basically tells the markup template engine that the underlying markup needs to be rendered
separately and exported as a string. Note that for simple expressions, stringOf
can be replaced by an alternate tag
notation that starts with a dollar sign:
p("This is a ${$a(href:'target.html', "link")} to another page")
It is worth noting that using stringOf or the special $tag notation triggers the creation of a distinct string writer
which is then used to render the markup. It is slower than using the version with calls to yield which perform direct
streaming of the markup instead.
|
7.2.6. Internationalization
The template engine has native support for internationalization. For that, when you create the TemplateConfiguration
, you can provide
a Locale
which is the default locale to be used for templates. Each template may have different versions, one for each locale. The
name of the template makes the difference:
-
file.tpl
: default template file -
file_fr_FR.tpl
: french version of the template -
file_en_US.tpl
: american english version of the template -
…
When a template is rendered or included, then:
-
if the template name or include name explicitly sets a locale, the specific version is included, or the default version if not found
-
if the template name doesn’t include a locale, the version for the
TemplateConfiguration
locale is used, or the default version if not found
For example, imagine the default locale is set to Locale.ENGLISH
and that the main template includes:
include template: 'locale_include_fr_FR.tpl'
then the template is rendered using the specific template:
Texte en français
Using an include without specifying a locale will make the template engine look for a template with the configured locale, and if not, fallback to the default, like here:
include template: 'locale_include.tpl'
Default text
However, changing the default locale of the template engine to Locale.FRANCE
will change the output, because the template engine will now look for a file
with the fr_FR
locale:
Texte en français
This strategy lets you translate your templates one by one, by relying on default templates, for which no locale is set in the file name.
7.2.7. Custom template classes
By default, templates created inherit the groovy.text.markup.BaseTemplate
class. It may be interesting for an application to provide a different
template class, for example to provide additional helper methods which are aware of the application, or customized rendering primitives (for HTML,
for example).
The template engine provides this ability by setting an alternative baseTemplateClass
in the TemplateConfiguration
:
config.setBaseTemplateClass(MyTemplate.class);
The custom base class has to extend BaseClass
like in this example:
public abstract class MyTemplate extends BaseTemplate {
private List<Module> modules
public MyTemplate(
final MarkupTemplateEngine templateEngine,
final Map model,
final Map<String, String> modelTypes,
final TemplateConfiguration configuration) {
super(templateEngine, model, modelTypes, configuration)
}
List<Module> getModules() {
return modules
}
void setModules(final List<Module> modules) {
this.modules = modules
}
boolean hasModule(String name) {
modules?.any { it.name == name }
}
}
This example shows a class which provides an additional method named hasModule
, which can then be used directly in the template:
if (hasModule('foo')) {
p 'Found module [foo]'
} else {
p 'Module [foo] not found'
}
7.3. Type checked templates
7.3.1. Optional type checking
Even if templates are not type checked, they are statically compiled. This means that once the templates are compiled, performance should be very good. For some applications, it might be good to make sure that templates are valid before they are actually rendered. This means failing template compilation, for example, if a method on a model variable doesn’t exist.
The MarkupTemplateEngine
provides such a facility. Templates can be optionally type checked. For that, the developer must provide additional information at
template creation time, which is the types of the variables found in the model. Imagine a model exposing a list of pages, where a page is defined as:
public class Page {
Long id
String title
String body
}
Then a list of pages can be exposed in the model, like this:
Page p = new Page();
p.setTitle("Sample page");
p.setBody("Page body");
List<Page> pages = new LinkedList<>();
pages.add(p);
model = new HashMap<String,Object>();
model.put("pages", pages);
A template can use it easily:
pages.each { page -> (1)
p("Page title: $page.title") (2)
p(page.text) (3)
}
1 | iterate on pages from the model |
2 | page.title is valid |
3 | page.text is not (should be page.body ) |
Without type checking, the compilation of the template succeeds, because the template engine doesn’t know about the model until a page is actually rendered. This means that the problem would only surface at runtime, once the page is rendered:
No such property: text
In some situations, this can be complicated to sort out or even notice. By declaring the type of the pages
to the template engine, we’re now capable of failing at compile time:
modelTypes = new HashMap<String,String>(); (1)
modelTypes.put("pages", "List<Page>"); (2)
Template template = engine.createTypeCheckedModelTemplate("main.tpl", modelTypes) (3)
1 | create a map which will hold the model types |
2 | declare the type of the pages variables (note the use of a string for the type) |
3 | use createTypeCheckedModelTemplate instead of createTemplate |
This time, when the template is compiled at the last line, an error occurs:
[Static type checking] - No such property: text for class: Page
This means that you don’t need to wait for the page to be rendered to see an error. The use of createTypeCheckedModelTemplate
is mandatory.
7.3.2. Alternative declaration of types
Alternatively, if the developer is also the one who writes the templates, it is possible to declare the types of the expected variables
directly in the template. In this case, even if you call createTemplate
, it will be type checked:
modelTypes = { (1)
List<Page> pages (2)
}
pages.each { page ->
p("Page title: $page.title")
p(page.text)
}
1 | types need to be declared in the modelTypes header |
2 | declare one variable per object in the model |
7.3.3. Performance of type checked templates
An additional interest of using type checked models is that performance should improve. By telling the type checker what are the expected types, you also let the compiler generate optimized code for that, so if you are looking for the best performance, consider using type checked templates.
8. Other solutions
Also, there are other templating solutions that can be used along with Groovy, such as FreeMarker, Velocity, StringTemplate and others.