Open dcblack-google opened 2 months ago
I'm not all that clear about this one; I think @dglazkov will have a view on it, though.
This should totally "just work", as of three weeks ago or so. I wonder if googleplex visual editor is running an older version?
Now that I try again from scratch I'm getting an internal exception, both on googleplex and on the external site:
Internal Exception: TypeError: Failed to fetch at .invoke (https://breadboard-ai.googleplex.com/index-DO-MixRG.js:1567:21) at .invoke (https://breadboard-ai.googleplex.com/define-ciGEp0Q7.js:308:24) at https://breadboard-ai.googleplex.com/path-registry-C9U4iGRp.js:15:5 at new Promise (
very simple example board: testboard.json
Yup, can reproduce. The error is hidden away in DevTools:
index-BQSQ4KEq.js:1567 Mixed Content: The page at 'https://breadboard-ai.web.app/?board=idb%3A%2F%2Fdefault%2Fblank-board-2.bgl.json' was loaded over HTTPS, but requested an insecure resource 'http://motherfuckingwebsite.com/'. This request has been blocked; the content must be served over HTTPS.
If I change the URL protocol to https, I get another error:
Access to fetch at 'https://motherfuckingwebsite.com/' from origin 'https://breadboard-ai.web.app' has been blocked by CORS policy: No 'Access-Control-Allow-Origin' header is present on the requested resource. If an opaque response serves your needs, set the request's mode to 'no-cors' to fetch the resource with CORS disabled.
Good bug! We need to add a bit more graceful error handling to fetch
, so that the user doesn't have to dig into DevTools to find the actual error.
... it's respecting CORS? Gosh. Well, good to know - guess I need to come up with a completely different plan for how to fetch web content.
Check out this one as a possible strategy: https://breadboard-ai.web.app/?board=https%3A%2F%2Fbreadboard.live%2Fboards%2F%40dimitri%2Ftool-page-as-markdown.bgl.json
That was the very first thing I tried, and then swiftly abandoned when it immediately started spitting out quota exceeded errors. (I was only calling it a couple times so it wasn't me.)
Currently have an internal solution half-built, hope to get it up and running next week, but getting arbitrary web content feels like a base capability that should be better supported by the platform.
The fetch node doesn't seem to respect the "raw" checkbox and always tries to parse what it gets as JSON. I'm trying to use it to just grab a vanilla webpage's html and it's spitting out "Internal Exception: SyntaxError: Unexpected token '<', " <hea"... is not valid JSON }" even with raw checked.