Improve watch mode

[mrlubos]

• add timeout option to handle servers that stay down for tens of seconds while reloading
• handle servers that do not expose HEAD method for spec
• improve console message on reload (clear)

[Matsuuu]

This would be highly appriciated feature as I run this in watch mode while developing and having the CLI crash out constantly isn't quite the DX you want.

Just having the json parsing in a try-catch block could help to not crash the software

[mrlubos]

Hi @Matsuuu, please try the latest release v0.64.1 and let me know if it works as expected!

[Matsuuu]

I am still getting crashes on the latest package version.

/node_modules/@hey-api/json-schema-ref-parser/dist/lib/resolvers/url.js:44

ResolverError: Error requesting http://localhost:5220/openapi/v1.json
fetch failed

I can try and make a minimal repro repository at some point if needed

[mrlubos]

@Matsuuu Are you able to find out why is it failing? What kind of request is it sending vs expected?

[Matsuuu]

So I'm running a background service, which needs to be taken down, re-built and then started again, due to it not supporting a watch mode.

When I take the API down, the watch mode breaks as it's not able to query the endpoint.

Any setup I tried that supports actual watch modes worked flawlessly with this, but a compiled API that needs a "full restart" with a few seconds of downtime will break the watch mode.

The stacktrace points at /node_modules/@hey-api/json-schema-ref-parser/dist/lib/resolvers/url.js:44:15

I understand that the API being down should result in an warning / error but when using watch mode, I think a good dev experience would be that it just reports that "I'm not able to find anything at pointed url", instead of straight up throwing and error and stopping the whole process.

This can be easily simulated here by following the steps: Starting the api, starting the app, shutting down the api.

I would like that either by default, or via a flag these situations wouldn't crash the watcher.

---

I think there's 2 approaches to this.

1. In watch mode, we report that the URL doesn't provide a output, with some info about it

2. We only make it throw the error if it's the "Initial" run of a watch mode and not on concecutive ones.

I'd be happy to discuss further as I'm starting to adopt this tool in plenty of my projects and maybe even contributing if there's spots to contribute to.

[Matsuuu]

Provided a small PR that fixes it. It ain't the most elegant thing but it does the job. I'd be happy to discuss if there's a better way #1748

[mrlubos]

Yep, will merge, thank you!