related to #129

thanks for sharing that approach. after a first review, here are some thoughts:

• i would seperate file- and service-validation, the latter would be done in process_container_options
• there are some checks in other parts of the code, that
  • can be moved to the validation
  • are ineffective after validation and therefore should be moved to the validation or be removed
• the DOCKER_CONFIG_KEYS should be derived programatically from the validation schema
• the validation could be far more specific regarding the values, e.g.:
  • expose-keys must contain only digits
  • paths are expected differently on different OS
    • their existence in the fs could even be checked, but that seems too far off
• i also found these two packages which seem to be capable to do the job, are more pythonic and extendable:
  • Volutuous
  • Cerberus
    • after reading the docs, i'd prefer that one

@aanand @bfirsh @dnephin @draghuram i'd appreciate any feedback before i get on it.

Thanks for both your PRs - I'm going to take a proper look at them soon.

This looks like a good start. I don't feel especially strongly about this one vs @funkyfuture's so if you've not got the time, that one will probably win by default.

I know @dnephin has advocated for using JSON schema in the past and I can see a hypothetical argument for it (should the Compose format ever become something that's serializable to JSON) but I don't think it's a particularly strong one.

@aanand Thanks for the comments. I agree that we should go with @funkyfuture's PR considering that he is willing to see the feature through.

Handled by #1808