No error messages when accessing undefined variables in config templating

[mikn]

So, poking around in your settings system I noticed that the templates, both on the input from the user-data.toml and to the output in the templates, you use kebab-case, but in rust you use snake_case (few programming languages allow you to use kebab-case, I guess).

This took me an embarrassingly long time to figure out (I just assumed that since it was decoded into rust structs with snake_case, that the templating would use snake_case also).

Ultimately, what was not ergonomic in this was that I got no real error when trying to access template variables that does not exist and is not set. You do take great care to validate it in Rust, etc - perhaps there is a way to make sure that when accessing variables in the templating stage, that you check that the variable exists? Would have saved me a lot of time!

[cbgbt]

Hey @mikn, thanks for the report.

The template renderer is supposed to throw an error if we attempt to use a variable which isn't defined. We enable strict mode when constructing the template registry (relevant docs).

I wonder if we are rendering templates via a codepath that doesn't use these registry settings, or if these errors are not surfaced in a way that makes problems clear.

[cbgbt]

Yep, we don't use that codepath when we render template files in Bottlerocket.

Here's where we render templates, and that calls this function which does not use the registry code I linked above.

Phooey, I almost certainly introduced this regression when I last worked on schnauzer.

We can re-enable strict mode for these renderings, but we'll need to be very cautious in doing so -- it's a backwards incompatible change for core-kit. Existing templates which successfully render today could very well stop.