minijinja 3.0.3

MiniJinja for Python: a powerful template engine for Rust and Python

minijinja-py is an experimental binding of MiniJinja to Python. It has somewhat limited functionality compared to the Rust version. These bindings use maturin and pyo3.

You might want to use MiniJinja instead of Jinja2 when the full feature set of Jinja2 is not required and you want to have the same rendering experience of a data set between Rust and Python.

With these bindings MiniJinja can render some Python objects and values that are passed to templates, but there are clear limitations with regards to what can be done.

To install MiniJinja for Python you can fetch the package from PyPI:

$ pip install minijinja

Basic API

The basic API is hidden behind the Environment object. It behaves almost entirely like in minijinja with some Python specific changes. For instance instead of env.set_debug(True) you use env.debug = True. Additionally instead of using add_template or attaching a source you either pass a dictionary of templates directly to the environment or a loader function.

from minijinja import Environment

env = Environment(templates={
    "template_name": "Template source"
})

To render a template you can use the render_template method:

result = env.render_template('template_name', var1="value 1", var2="value 2")
print(result)

Purpose

MiniJinja attempts a reasonably high level of compatibility with Jinja2, but it does not try to achieve this at all costs. As a result you will notice that quite a few templates will refuse to render with MiniJinja despite the fact that they probably look quite innocent. It is however possible to write templates that render to the same results for both Jinja2 and MiniJinja. This raises the question why you might want to use MiniJinja.

The main benefit would be to achieve the exact same results in both Rust and Python. Additionally MiniJinja has a stronger sandbox than Jinja2 and might perform ever so slightly better in some situations. However you should be aware that due to the marshalling that needs to happen in either direction there is a certain amount of loss of information.

Dynamic Template Loading

MiniJinja's Python bindings inherit the underlying behavior of how MiniJinja loads templates. Templates are loaded on first use and then cached. The templates are loaded via a loader. To trigger a reload you can call env.reload() or alternatively set env.reload_before_render to True.

def my_loader(name):
    segments = []
    for segment in name.split("/"):
        if "\\" in segment or segment in (".", ".."):
            return None
        segments.append(segment)
    try:
        with open(os.path.join(TEMPLATES, *segments)) as f:
            return f.read()
    except (IOError, OSError):
        pass

env = Environment(loader=my_loader)
env.reload_before_render = True
print(env.render_template("index.html"))

Alternatively templates can manually be loaded and unloaded with env.add_template and env.remove_template.

Auto Escaping

The default behavior is to use auto escaping file files ending in .html. You can customize this behavior by overriding the auto_escape_callback:

env = Environment(auto_escape_callback=lambda x: x.endswith((".html", ".foo")))

MiniJinja uses markupsafe if it's available on the Python side. It will honor __html__.

Finalizers

Instead of custom formatters like in MiniJinja, you can define a finalizer instead which is similar to how it works in Jinja2. It's passed a value (or optional also the state as first argument when pass_state is used) and can return a new value. If the special NotImplemented value is returned, the original value is rendered without any modification:

from minijinja import Environment

def finalizer(value):
    if value is None:
	return ""
    return NotImplemented

env = Environment(finalizer=finalizer)
assert env.render_str("{{ none }}") == ""

State Access

Functions passed to the environment such as filters or global functions can optionally have the template state passed by using the pass_state parameter. This is similar to pass_context in Jinja2. It can be used to look at the name of the template or to look up variables in the context.

from minijinja import pass_state

@pass_state
def my_filter(state, value):
    return state.lookup("a_variable") + value

env.add_filter("add_a_variable", my_filter)

Runtime Behavior

MiniJinja uses it's own runtime model which is not matching the Python runtime model. As a result there are gaps in behavior between the two but some limited effort is made to bridge them. For instance you will be able to call some methods of types, but for instance builtins such as dicts and lists do not expose their methods on the MiniJinja side in all cases. A natively generated MiniJinja map (such as with the dict global function) will not have an .items() method, whereas a Python dict passed to MiniJinja will.

Here is what this means for some basic types:

• Python dictionaries and lists (as well as other objects that behave as sequences) appear in the MiniJinja side very similar to how they do in Python.
• Python tuples appear as tuples in MiniJinja, use Python-compatible tuple rendering, and remain tuples when passed back to Python. Lists remain distinct from tuples.
• Python objects are represented in MiniJinja similarly to dicts, but they retain all their meaningful Python APIs. This means they stringify via __str__ and they allow the MiniJinja code to call their non-underscored methods. Note that there is no extra security layer in use at the moment so take care of what you pass there.
• MiniJinja's python binding understand what __html__ is when it exists on a string subclass. This means that a markupsafe.Markup object will appear as safe string in MiniJinja. This information can also flow back to Python again.
• Stringification of objects uses __str__ which is why mixed Python and MiniJinja objects can be a bit confusing at times.
• Where in Jinja2 there is a difference between foo["bar"] and foo.bar which can be used to disambiugate properties and keys, in MiniJinja there is no such difference. However methods are disambiugated so foo.items() works and will correctly call the method in all cases.
• Operator overloading is not supported. This in particular means that operators like + will not invoke the __add__ method of most objects. This is an intentional limitation of the engine.
• When MiniJinja objects are exposed to the Python side they lose most of their functionality. For instance plain objects such as functions are currently just represented as strings, maps are dictionaries etc. This also means you cannot call methods on them from the Python side.

Threading

MiniJinja's Python binding is thread-safe. Multiple threads can render templates from the same environment at the same time. Changes to an environment (for instance adding filters or globals or changing settings) are only picked up by renders that start after the change; renders that are already in progress continue to use the configuration they started with.

Sponsor

If you like the project and find it useful you can become a sponsor.

License and Links

• Documentation
• Examples
• Issue Tracker
• MiniJinja Playground
• License: Apache-2.0