how can I correctly plug in my own Jinja Environment?

[gagan-bajwa]

Hi team,

I’m re-skinning Meridian reports to match our company’s design system.
I copied the original *.html.jinja files into src/branding/templates/ and load them via a custom Jinja environment:

src/branding/env.py
from jinja2 import Environment, FileSystemLoader
from pathlib import Path
``
_TEMPLATES = Path(__file__).parent / "templates"

def get_brand_env() -> Environment:
    return Environment(
        loader=FileSystemLoader([_TEMPLATES]),
        autoescape=False,        # want raw HTML, not escaped
        trim_blocks=True,
        lstrip_blocks=True,
    )

In my API entry-point / training script I patch Meridian like this:

import meridian.analysis.formatter as _fmt
import meridian.analysis.summarizer as _summ
from src.branding.env import get_brand_env

_brand_env = get_brand_env()
_fmt.create_template_env = get_brand_env          
_summ._TEMPLATE_ENV      = _brand_env             

The new templates and CSS load fine (header, chips, colours, logo).
Problem: every pre-rendered fragment (card, chart, stats, table) still arrives HTML-escaped, so the browser prints

<card id="model-fit" …>
&lt;chart class="…"&gt; … &lt;/chart&gt;

instead of rendering the grid.

What I’ve tried
Adding |safe whenever I print card, item, etc. — helps only in the outer template; the inner markup is still encoded.

Verified that my own templates are actually used (added dummy text, saw it appear).

If I monkey-patch jinja2.escape = lambda x: x, the tags render (so it really is an escaping issue).

Questions
Is there an official extension point to supply a custom Environment (ideally with autoescape=False) instead of patching _TEMPLATE_ENV?

Where in the pipeline do the fragments (CardSpec, ChartSpec, …) get rendered? Is there a recommended flag / setting to mark them “safe” so the outer template can embed them without HTML-encoding?

If disabling auto-escape globally is a no-go, what’s the proper way to inject already-rendered HTML snippets into summary.html.jinja?

Any guidance or code pointers would be greatly appreciated.happy to PR docs once I get this working.

Thanks!

[santoso-wijaya]

Is there an official extension point to supply a custom Environment (ideally with autoescape=False) instead of patching _TEMPLATE_ENV?

There isn't at the moment.

Where in the pipeline do the fragments (CardSpec, ChartSpec, …) get rendered?

I can demonstrate this with the following snippets:

1. The patching.

from meridian.analysis import formatter
import jinja2
import os

# stand-in for a "custom" loader; this would just load the same built-in templates, though.
_template_loader = jinja2.FileSystemLoader(os.path.abspath(os.path.dirname(formatter.__file__)) + '/templates')


def _custom_create_template_env() -> jinja2.Environment:
  return jinja2.Environment(
      loader=_template_loader,
      autoescape=False,
      trim_blocks=True,
      lstrip_blocks=True,
  )

formatter.create_template_env = _custom_create_template_env

2. Create cards HTMLs piecewise

from meridian.analysis import summarizer

sum = summarizer.Summarizer(mmm)

# Setting up the (custom) template env
_template_env = formatter.create_template_env()
_template_env.globals["start_date"] = "Jan 1, 2021"
_template_env.globals["end_date"] = "Jan 22, 2024"

# Using the (now custom) template env, load invidivial card each as its own <card> html DOM
cards_htmls = sum._create_cards_htmls(
    _template_env,
    selected_times=None,  # == (start_date, end_date) above
)

3. Inspect each card HTML

print(cards_htmls[0])   # no escapes

4. Produce the summary report page

# Load the top-level page container (summary html template)
html_template = _template_env.get_template('summary.html.jinja')

report = html_template.render(title="My Title", cards=cards_htmls)
print(report)   # no escapes, either