Filtering and Buffering¶

Expression Filtering¶

As described in the chapter Syntax, the “|” operator can be applied to a “${}” expression to apply escape filters to the output:

${"this is some text" | u}

The above expression applies URL escaping to the expression, and produces this+is+some+text.

The built-in escape flags are:

u : URL escaping, provided by urllib.quote_plus(string.encode('utf-8'))
h : HTML escaping, provided by markupsafe.escape(string)

New in version 0.3.4: Prior versions use cgi.escape(string, True).
x : XML escaping
trim : whitespace trimming, provided by string.strip()
entity : produces HTML entity references for applicable strings, derived from htmlentitydefs
unicode (str on Python 3): produces a Python unicode string (this function is applied by default)
decode.<some encoding>: decode input into a Python unicode with the specified encoding
n : disable all default filtering; only filters specified in the local expression tag will be applied.

To apply more than one filter, separate them by a comma:

${" <tag>some value</tag> " | h,trim}

The above produces <tag>some value</tag>, with no leading or trailing whitespace. The HTML escaping function is applied first, the “trim” function second.

Naturally, you can make your own filters too. A filter is just a Python function that accepts a single string argument, and returns the filtered result. The expressions after the | operator draw upon the local namespace of the template in which they appear, meaning you can define escaping functions locally:

<%!
    def myescape(text):
        return "<TAG>" + text + "</TAG>"
%>

Here's some tagged text: ${"text" | myescape}

Or from any Python module:

<%!
    import myfilters
%>

Here's some tagged text: ${"text" | myfilters.tagfilter}

A page can apply a default set of filters to all expression tags using the expression_filter argument to the %page tag:

<%page expression_filter="h"/>

Escaped text:  ${"<html>some html</html>"}

Result:

Escaped text: &lt;html&gt;some html&lt;/html&gt;

The `default_filters` Argument¶

In addition to the expression_filter argument, the default_filters argument to both Template and TemplateLookup can specify filtering for all expression tags at the programmatic level. This array-based argument, when given its default argument of None, will be internally set to ["unicode"] (or ["str"] on Python 3), except when disable_unicode=True is set in which case it defaults to ["str"]:

t = TemplateLookup(directories=['/tmp'], default_filters=['unicode'])

To replace the usual unicode/str function with a specific encoding, the decode filter can be substituted:

t = TemplateLookup(directories=['/tmp'], default_filters=['decode.utf8'])

To disable default_filters entirely, set it to an empty list:

t = TemplateLookup(directories=['/tmp'], default_filters=[])

Any string name can be added to default_filters where it will be added to all expressions as a filter. The filters are applied from left to right, meaning the leftmost filter is applied first.

t = Template(templatetext, default_filters=['unicode', 'myfilter'])

To ease the usage of default_filters with custom filters, you can also add imports (or other code) to all templates using the imports argument:

t = TemplateLookup(directories=['/tmp'],
                   default_filters=['unicode', 'myfilter'],
                   imports=['from mypackage import myfilter'])

The above will generate templates something like this:

# ....
from mypackage import myfilter

def render_body(context):
    context.write(myfilter(unicode("some text")))

Turning off Filtering with the `n` Filter¶

In all cases the special n filter, used locally within an expression, will disable all filters declared in the <%page> tag as well as in default_filters. Such as:

${'myexpression' | n}

will render myexpression with no filtering of any kind, and:

${'myexpression' | n,trim}

will render myexpression using the trim filter only.

Including the n filter in a <%page> tag will only disable default_filters. In effect this makes the filters from the tag replace default filters instead of adding to them. For example:

<%page expression_filter="n, json.dumps"/>
data = {a: ${123}, b: ${"123"}};

will suppress turning the values into strings using the default filter, so that json.dumps (which requires imports=["import json"] or something equivalent) can take the value type into account, formatting numbers as numeric literals and strings as string literals.

New in version 1.0.14: The n filter can now be used in the <%page> tag.

Filtering Defs and Blocks¶

The %def and %block tags have an argument called filter which will apply the given list of filter functions to the output of the %def:

<%def name="foo()" filter="h, trim">
    <b>this is bold</b>
</%def>

When the filter attribute is applied to a def as above, the def is automatically buffered as well. This is described next.

Buffering¶

One of Mako’s central design goals is speed. To this end, all of the textual content within a template and its various callables is by default piped directly to the single buffer that is stored within the Context object. While this normally is easy to miss, it has certain side effects. The main one is that when you call a def using the normal expression syntax, i.e. ${somedef()}, it may appear that the return value of the function is the content it produced, which is then delivered to your template just like any other expression substitution, except that normally, this is not the case; the return value of ${somedef()} is simply the empty string ''. By the time you receive this empty string, the output of somedef() has been sent to the underlying buffer.

You may not want this effect, if for example you are doing something like this:

${" results " + somedef() + " more results "}

If the somedef() function produced the content “somedef's results”, the above template would produce this output:

somedef's results results more results

This is because somedef() fully executes before the expression returns the results of its concatenation; the concatenation in turn receives just the empty string as its middle expression.

Mako provides two ways to work around this. One is by applying buffering to the %def itself:

<%def name="somedef()" buffered="True">
    somedef's results
</%def>

The above definition will generate code similar to this:

def somedef():
    context.push_buffer()
    try:
        context.write("somedef's results")
    finally:
        buf = context.pop_buffer()
    return buf.getvalue()

So that the content of somedef() is sent to a second buffer, which is then popped off the stack and its value returned. The speed hit inherent in buffering the output of a def is also apparent.

Note that the filter argument on %def also causes the def to be buffered. This is so that the final content of the %def can be delivered to the escaping function in one batch, which reduces method calls and also produces more deterministic behavior for the filtering function itself, which can possibly be useful for a filtering function that wishes to apply a transformation to the text as a whole.

The other way to buffer the output of a def or any Mako callable is by using the built-in capture function. This function performs an operation similar to the above buffering operation except it is specified by the caller.

${" results " + capture(somedef) + " more results "}

Note that the first argument to the capture function is the function itself, not the result of calling it. This is because the capture function takes over the job of actually calling the target function, after setting up a buffered environment. To send arguments to the function, just send them to capture instead:

${capture(somedef, 17, 'hi', use_paging=True)}

The above call is equivalent to the unbuffered call:

${somedef(17, 'hi', use_paging=True)}

Decorating¶

New in version 0.2.5.

Somewhat like a filter for a %def but more flexible, the decorator argument to %def allows the creation of a function that will work in a similar manner to a Python decorator. The function can control whether or not the function executes. The original intent of this function is to allow the creation of custom cache logic, but there may be other uses as well.

decorator is intended to be used with a regular Python function, such as one defined in a library module. Here we’ll illustrate the python function defined in the template for simplicities’ sake:

<%!
    def bar(fn):
        def decorate(context, *args, **kw):
            context.write("BAR")
            fn(*args, **kw)
            context.write("BAR")
            return ''
        return decorate
%>

<%def name="foo()" decorator="bar">
    this is foo
</%def>

${foo()}

The above template will return, with more whitespace than this, "BAR this is foo BAR". The function is the render callable itself (or possibly a wrapper around it), and by default will write to the context. To capture its output, use the capture() callable in the mako.runtime module (available in templates as just runtime):

<%!
    def bar(fn):
        def decorate(context, *args, **kw):
            return "BAR" + runtime.capture(context, fn, *args, **kw) + "BAR"
        return decorate
%>

<%def name="foo()" decorator="bar">
    this is foo
</%def>

${foo()}

The decorator can be used with top-level defs as well as nested defs, and blocks too. Note that when calling a top-level def from the Template API, i.e. template.get_def('somedef').render(), the decorator has to write the output to the context, i.e. as in the first example. The return value gets discarded.

Mako 1.1.6 Documentation