CLI does not accept a schema without instance any longer

[krassowski]

Hi, thanks for maintaining jsonschema! We were using jsonschema to validate the integrity of the schema itself (i.e. is it a valid schema) as an early step in CI in https://github.com/python-lsp/python-lsp-server/ (if schema is not valid, we want to short-circuit early). When upgrading CI from Python 3.6 (which was pulling jsonschema 3.2.0) to a newer Python version (resolving jsonschema to 4.2.1) on python-lsp/python-lsp-server#130 the CI started failing:

jsonschema pylsp/config/schema.json
Failed to parse <stdin>: Expecting value: line 1 column 1 (char 0)

I did not find a changelog entry warning about to this change, but comparing jsonschema -h between versions I see that -i option is no longer really optional (i.e. it is now implicitly required):

3.2.0:

usage: jsonschema [-h] [-i INSTANCES] [-F ERROR_FORMAT] [-V VALIDATOR] [--version] schema

JSON Schema Validation CLI

positional arguments:
  schema                the JSON Schema to validate with (i.e. schema.json)

optional arguments:
  -h, --help            show this help message and exit
  -i INSTANCES, --instance INSTANCES
                        a path to a JSON instance (i.e. filename.json) to validate (may be specified multiple times)

4.2.1:

usage: jsonschema [-h] [-i INSTANCES] [-F ERROR_FORMAT] [-o {plain,pretty}] [-V VALIDATOR] [--base-uri BASE_URI] [--version] schema

JSON Schema Validation CLI

positional arguments:
  schema                the path to a JSON Schema to validate with (i.e. schema.json)

optional arguments:
  -h, --help            show this help message and exit
  -i INSTANCES, --instance INSTANCES
                        a path to a JSON instance (i.e. filename.json) to validate (may be specified multiple times).
                        If no instances are provided via this option, one will be expected on standard input.

the addition of If no instances are provided via this option, one will be expected on standard input. forces users to always pass a valid instance via standard input if they want to validate the schema itself.

One solution could be to add --no-instances optional argument (in 4.x). Another would be to just pass if no instance is given. What do you think?

For posterity coming here from search engines, I will cross-reference #888 (although not asking about CLI, this issue could answer the question from the title of the other one).

[Julian]

Hi! Thanks for the kind words.

This is probably an oversight (or, well, something that a test was missing for, so yeah an unintentional oversight in reworking the CLI). #643 is I think the relevant PR.

I'm not so keen on a --no-instances argument initially -- so I'm not sure what a fix looks like yet, but I'm certainly willing to fix this somehow. Another option is forcing someone to pass - to indicate stdin, but that'd now be another breaking change... Not sure. Will have to think, but feel free to share any additional thoughts.

[sirosen]

In this case, the schema is being validated under a relevant metaschema.
So rather than --no-instances ("don't do X"), is there a positive phrasing ("do Y") like --check-metaschema?

jsonschema pylsp/config/schema.json --check-metaschema

[Julian]

Any flag I think will cause implementation annoyance -- namely it will introduce the need for mutually exclusive options (--check-metaschema and -i), or it will make --check-metaschema's behavior confusing if it's normally implied by -i.

I'm still somewhat personally leaning towards reverting the read from stdin such that one can go back to providing no instances, but supporting reading from stdin by passing - as the instance as many CLI tools tend to. I suspect a PR implementing this would be acceptable.

More ideas are good though as well.

[sirosen]

I've implemented this as a --check-metaschema flag in check-jsonschema. Because I already need to deal with mutex options there, it wasn't too much of an additional burden.

It is a bit of a mouthful to write check-jsonschema --check-metaschema foo.json, but at least it's clear and explicit.

I am the author of check-jsonschema and we are currently discussing making it the official CLI for jsonschema. For details, you can follow #910. If check-jsonschema has a behavior which you believe should be changed to handle your use-case, please open an issue!

[sirosen]

We're going to move forward and begin calling check-jsonschema the official CLI. I'm therefore closing this as a wontfix.

Definitely let me know if you'd like a hand updating python-lsp-server to use check-jsonschema where you're using jsonschema's built-in CLI. I'm happy to stop by and see how I can help.