Python Liquid

A Python implementation of Liquid, the safe, customer-facing template language for flexible web apps.

from liquid import Template

template = Template("Hello, {{ you }}!")
print(template.render(you="World"))  # "Hello, World!"
print(template.render(you="Liquid"))  # "Hello, Liquid!"

Installing

Install Python Liquid using Pipenv:

$ pipenv install python-liquid

Or pip:

$ python -m pip install -U python-liquid

Links

• Documentation: https://jg-rp.github.io/liquid/
• Change Log: https://github.com/jg-rp/liquid/blob/main/CHANGES.rst
• PyPi: https://pypi.org/project/python-liquid/
• Source Code: https://github.com/jg-rp/liquid
• Issue Tracker: https://github.com/jg-rp/liquid/issues

Related Projects

• django-liquid: A Django template backend for Liquid. Render Liquid templates in your Django apps.
• Flask-Liquid: A Flask extension for Liquid. Render Liquid templates in your Flask applications.
• python-liquid-extra: A growing collection of extra tags and filters for Python Liquid. Highlights of which are:
  • an if tag that supports not and grouping with parentheses.
  • macro and call tags for defining and calling parameterized Liquid snippets.
  • inline if/else expressions. For example {{ 'active' if link.active else '' }} or {% assign selected = true if product.selected_variant else false %}.
  • a JSON encoding filter.
• golden-liquid: A test suite for Liquid. See how various Liquid template engines compare to the reference implementation.

Compatibility

We strive to be 100% compatible with the reference implementation of Liquid, written in Ruby. That is, given an equivalent render context, a template rendered with Python Liquid should produce the same output as when rendered with Ruby Liquid.

Known Issues

Please help by raising an issue if you notice an incompatibility.

• Error handling. Python Liquid might not handle syntax or type errors in the same way as the reference implementation. We might fail earlier or later, and will almost certainly produce a different error message.
• The built-in date filter uses dateutil for parsing strings to datetimes, and strftime for formatting. There are likely to be some inconsistencies between this and the reference implementation's equivalent parsing and formatting of dates and times.
• In Ruby Liquid, the built-in increment and decrement tags can, in some cases, mutate "global" context and keep named counters alive between renders. Although not difficult to implement, I can't quite bring myself to do it.
• If a range literal uses a float literal as its start or stop value, the float literal must have something after the decimal point. This is OK (1.0..3). This is not (1...3). Ruby Liquid will accept either, resulting in a sequence of [1,2,3].

Benchmark

You can run the benchmark using make benchmark (or python -O performance.py if you don't have make) from the root of the source tree. On my ropey desktop computer with a Ryzen 5 1500X, we get the following results.

Best of 5 rounds with 100 iterations per round and 60 ops per iteration (6000 ops per round).

lex template (not expressions): 1.3s (4727.35 ops/s, 78.79 i/s)
                 lex and parse: 6.4s (942.15 ops/s, 15.70 i/s)
                        render: 1.7s (3443.62 ops/s, 57.39 i/s)
         lex, parse and render: 8.2s (733.30 ops/s, 12.22 i/s)

And PyPy3.7 gives us a decent increase in performance.

Best of 5 rounds with 100 iterations per round and 60 ops per iteration (6000 ops per round).

lex template (not expressions): 0.58s (10421.14 ops/s, 173.69 i/s)
                 lex and parse: 2.9s (2036.33 ops/s, 33.94 i/s)
                        render: 1.1s (5644.80 ops/s, 94.08 i/s)
         lex, parse and render: 4.2s (1439.43 ops/s, 23.99 i/s)

On the same machine, running rake benchmark:run from the root of the reference implementation source tree gives us these results.

/usr/bin/ruby ./performance/benchmark.rb lax

Running benchmark for 10 seconds (with 5 seconds warmup).

Warming up --------------------------------------
             parse:     3.000  i/100ms
            render:     8.000  i/100ms
    parse & render:     2.000  i/100ms
Calculating -------------------------------------
             parse:     39.072  (± 0.0%) i/s -    393.000  in  10.058789s
            render:     86.995  (± 1.1%) i/s -    872.000  in  10.024951s
    parse & render:     26.139  (± 0.0%) i/s -    262.000  in  10.023365s

I've tried to match the benchmark workload to that of the reference implementation, so that we might compare results directly. The workload is meant to be representative of Shopify's use case, although I wouldn't be surprised if their usage has changed subtly since the benchmark fixture was designed.

Contributing

• Install development dependencies with Pipenv
• Python Liquid uses type hints and static type checking. Run mypy or tox -e typing to check for typing issues.
• Format code using black.
• Write tests using unittest.TestCase.
• Run tests with make test or python -m unittest or pytest.
• Check test coverage with make coverage and open htmlcov/index.html in your browser.
• Check your changes have not adversely affected performance with make benchmark.