Set LabApp to fail when config files have errors.

[Zsailer]

If jupyterlab config file contains an error, the jupyterlab app throws a warning, ignores the config file, and continues running the application.

Would it make more sense to, instead, fail fast when errors are found? This is a simple change. Just need to add raise_config_file_errors = True as an attribute in the LabApp.

This was raised in issue #5614. Opening this issue to discuss, as this will change the default behavior of jupyterlab.

[jasongrout]

Would it make more sense to, instead, fail fast when errors are found?

Yes! Several of us agree.

[Zsailer]

@jasongrout
raise_config_file_errors=False by default in traitlets (docs here). To fail fast, a traitlets application just needs to set this to True.

My question is, how deep+wide in the inheritance tree should we go? Meaning, do we only want the LabApp to fail on config file errors? or all JupyterLab subcommands too?

Further, all JupyterLab apps inherit from JupyterApp at some point. We could set this trait at the JupyterApp level so all jupyter applications behave the same.

[vidartf]

Meaning, do we only want the LabApp to fail on config file errors? or all JupyterLab subcommands too?

I would say that only having the LabApp fail would be a good start. I think having it on subcommands could become counterproductive. Are there any examples of lab subcommands that will do the wrong thing if the config is incorrect?

[bollwyvl]

A misconfigured app_dir in 'build' or 'labextension' could definitely give you something you don't want or mysteriously can't use... But here "misconfigured" probably just means "not what I expect" and not "a number". Further, build may actually bring it into existence, so even if there was a trait for PathWhatExists, it wouldn't work here.

…

[Madhu94]

@bollwyvl, for the example you mentioned - jupyter lab buildand jupyter labextension don't look up the config file for app_dir, either JUPYTERLAB_DIR has to be set or the --app-dir flag needs to be passed in.

Do any of the other subcommands look into the config file?

We could set this trait at the JupyterApp level so all jupyter applications behave the same.

So the change would be in jupyter_core instead ?

[goanpeca]

Wondering what is the current state of this issues :)

Was this implemented @jupyterlab/committers ?

[vidartf]

I think this can be closed now? After Ian's fix in jupyter-core, I think we can now just set JupyterApp.raise_config_file_errors=True, and it should fail as expected (though I haven't actually confirmed this).