Fixes IMAGES-2308 and (the ancient) #4385

fetch(url, { cf: { image: { ... } } }) allows workers to take advantage of the longstanding transform via Workers feature to transform images . Prior to this change, this feature is not supported in local wrangler dev - the cf object is dropped and the source image bytes are returned.

We already use Sharp to mock some of the more basic capabilities of the images binding. We are repeating the trick here and using it to resize our images we fetch. After we ship, we'll have much closer parity with production.

Changes

• Outbound fetches now route through an outbound interceptor worker (a renaming + restructuring of the existing strip-cf-connecting-ip worker). It does the existing CF-Connecting-IP header rewrite and the new cf.image transform, then continues down the chain.
• When a request features cf.image options, the interceptor fetches the origin image (stamping Via: image-resizing for loop-prevention parity) and hands the bytes to a new Node-side Sharp fetcher (cfImageLocalFetcher) over the existing loopback bridge.
• Attempts to follow production semantics. We support various fit modes, the format: "json" output shape, the cf-resized header. Only resize, rotate and format are honoured, unsupported options are ignored.

Reviewer considerations

• The outbound interceptor worker is in the path for all worker outbound fetches. Non-image requests are passed through. I don't see any major problems with this, but worth being aware.
• stripCfConnectingIp: false has been changed for a toggle on a worker to instead toggling the header strip functionality in our outbound inteceptor worker.
• Local failures fail open (returning the source image). We don't attempt to replicate any of the "Free Transforms Exceeded" logic that may befall customers running in production.

Testing

New spec (test/plugins/core/cf-image.spec.ts). Sanity-checked end-to-end against a demo worker under a local wrangler dev build vs published wrangler.

• Tests
  • Tests included/updated
  • Automated tests not possible - manual testing has been completed as follows:
  • Additional testing not necessary because:
• Public documentation
  • Cloudflare docs PR(s):
  • Documentation not necessary because: Current behaviour isn't documented.

A picture of a cute animal (not mandatory, but encouraged)