I've seen a large scale Mako codebase and the amount of business logic that had sneaked into the templates was horrifying - really made me double-down on my preference for limited template languages that help people avoid that temptation.

I helped design Django's template language and I now think we took keeping it restrictive a bit too far though - these days I prefer working with Jinja because it has better support for complex expressions and lets you call functions within your templates. That does mean it requires a bit more discipline to use responsibly though, but being able to control which functions are exposed to the template context helps a lot there.

Having never used Mako the appeal to me over Jinja is that it allows you to use real control structures. Jinja is a bit of a pain to debug and develop because it does some things slightly differently than you'd do in native Python so even though I can setup breakpoints and evaluate the application state I can't easily convert the code changes I make in the debugger back to the template. I get that there are reasons why Jinja and other templating systems guard against this, but it does have a cost.

I hate jinja. I really don’t get it. So here’s a complaining reply, so you hopefully won’t feel as alone.

When using jinja2 for website builders, it just feels like a poor version of old school php.

I’ve also seen it used for generating json/yaml config files, which to me is absolutely bunkers compared to just building a dictionary in Python and exporting it.