Hacker News Reader

I also think Jinja in YAML is an antipattern, but I think that due to them using the default block and expression characters of "{%" and "{{" both of which are yaml characters and thus every single such occurrence needs to be quoted, versus the very sane "${{" used by GHA or even "<%" and "<<" (running the risk that, yes, "<<:" is a yamlism but is not legal jinja)

If you meant that having any executable inside yaml (e.g. not just picking on jinja, but "a yaml document is not code") then I fear that ship has sailed because people realized that inverting the literal parts and only sometimes having executable parts makes for a great way to ASP/JSP/PHP up some content

If I wanted start a sibling flame war within this thread, we'd talk about HCL and for_each :rolling_eyes: (but, for clarity, please don't talk about it, or at least not here, submit your own noHCL.com link)

> 1. Write a configuration library for a real programming language. Have this library generate a JSON configuration file that is treated purely as an inspectable artifact, not something to edit by hand. The user will have their configuration as tooling-enabled code checked into version control (hopefully). That it will be harder to make an emergency fix directly on the server has its downsides and upsides.

this is the route Amazon chose with CDK. it kinda works... but it really feels like you're building a rube goldberg machine to do anything non-trivial. how much of it is the CDK and how much is cloudformation plainly sucking I don't know.

I agree. I find working with templated YAML so cumbersome that I ended up creating a tool (Cels - https://github.com/pacha/cels) because of it. I like Jsonnet and Starlark but in practice I don’t usually need a new programming language for most use cases. Most of the time I just want to create a base document and apply patches to do modifications. That simplifies everything a lot. The experience of writing pure YAML is actually not that bad (a couple of things about the format are really questionable, but it is workable). I find that problems appear from the complexity of the solutions you have to add to adapt documents to different environments.

> I think Jinja-in-YAML is very much an antipattern.

The thought of working with Ansible again is my nightmare due to this. Apparently they had an actual Python DSL you could use instead of YAML for a while, but discontinued it? So now loops and if-thens are these awful drawn-out messes of YAML that interpret jinja in completely unpredictable ways.

Do you mean the Ansible Runner interface? It is apparently still available. It was featured more prominently in early documentation (http://web.archive.org/web/20120314070643/http://ansible.git...). The discontinued OpsMop by the original author of Ansible (https://github.com/opsmop/opsmop) used a Python DSL.

I would also like to avoid writing Ansible playbooks when I need to manage servers again. Something I want to try out is Pyinfra (https://pyinfra.com/). It is like Ansible in that it is agentless and works over SSH, but with configuration in Python instead of YAML and Jinja. There is also itamae (Ruby, works over SSH or on the local machine, https://github.com/itamae-kitchen/itamae) and mitamae (mruby, local-only, https://github.com/itamae-kitchen/mitamae).

Is there a standalone Starlark tool? Similar to `jsonnet`, something that I can use to run Starlark code to generate JSON or YAML?

I once tried to find a simple standalone interpreter to play with and couldn't. There is a list of projects that use Starlark at https://github.com/bazelbuild/starlark#users. It lists https://github.com/carvel-dev/ytt for YAML templating and https://github.com/jathu/starfig for generating JSON and YAML configs. (I have not used either.)