Errors on command load should throw errors instead of invalidating the command about core HOT 6 CLOSED

@mdonnalley Thanks for the help! This isn't a super-high priority for us either, but it's good to have this issue for reference for when it inevitably comes up again.

from core.

@nferch This is where we're handling the MODULE_NOT_FOUND error: https://github.com/oclif/core/blob/3.16.0/src/config/plugin.ts#L136

Can you help me better understand what you'd like to see changed? Is it that you want the option to always throw on that error? If so, we could make that configurable either through the oclif section in the package.json or through an environment variable

from core.

@mdonnalley yeah, I came across that when debugging and thought that might be the place. Out of curiosity, under what condition is it useful to swallow that error? Is this the same codepath when a command doesn't actually exist?

I've been seeing the behavior when there's some bad code introduced or some other build error. What has been happening for us is that all commands fail to import so it appears that our CLI has no commands at all. Which is a confusing failure mode for both us and our users, as well as a bit awkward to detect in smoketests.

So I'd prefer to see it handled as a fatal error rather than a command not existing. Having an option to throw the error would probably be fine, but I also wonder if there'd be some way to distinguish the error being thrown on import rather than the import missing altogether. Does that make sense?

from core.

Out of curiosity, under what condition is it useful to swallow that error?

Off the top of my head, I think the idea was that you don't want one bad actor to bring down the entire CLI. So imagine a scenario where someone has installed a plugin that has a faulty init hook. In that scenario every single command in your CLI will fail because of that one plugin.

Is this the same codepath when a command doesn't actually exist?

That code path is here.

I've been seeing the behavior when there's some bad code introduced or some other build error. What has been happening for us is that all commands fail to import so it appears that our CLI has no commands at all.

Are you able to share a repo that replicates this? Thanks!

from core.

Out of curiosity, under what condition is it useful to swallow that error?

Off the top of my head, I think the idea was that you don't want one bad actor to bring down the entire CLI. So imagine a scenario where someone has installed a plugin that has a faulty init hook. In that scenario every single command in your CLI will fail because of that one plugin.

That makes sense. In our case these are all commands in our own CLI, so we'd prefer for the CLI to fail fatally.

Is this the same codepath when a command doesn't actually exist?

That code path is here.

I've been seeing the behavior when there's some bad code introduced or some other build error. What has been happening for us is that all commands fail to import so it appears that our CLI has no commands at all.

Are you able to share a repo that replicates this? Thanks!

The CLI in question is located at https://github.com/dxos/dxos/tree/main/packages/devtools/cli . It's currently in a non-broken state, but you can simulate the failure by adding an invalid import to any of the commands.

from core.

@nferch Sounds like we might just want to make your desired behavior configurable. Is this something you only need for your tests and build processes? If that's the case, then I'd say that it should be an environment variable. But if it's something that you want in production, then it should be a setting in the package.json

We don't have the bandwidth to prioritize this at the moment so I'd recommend making the PR yourself if it's something you need soon. If you do decide to do that, I'm happy to help with any questions you might have.

from core.