🐛 BUG: warning always received about multiple bind addresses

[Cherry]

Which Cloudflare product(s) does this pertain to?

Wrangler core

What version(s) of the tool(s) are you using?

3.8.0

What version of Node are you using?

18.17.1

What operating system are you using?

Windows 10

Describe the Bug

Every time that you start wrangler dev, you get a large red error like so (from workerd it would seem):

kj/async-io-win32.c++:982: warning: Bind address resolved to multiple addresses.  Only the first address will be used.  If this is incorrect, specify the address numerically.  This may be fixed in the future.; addrs[0].toString() = 127.0.0.1:59208

It's possible that this only occurs in Windows, or when a machine has multiple network adapters. Hitting b still works fine, and opening up http://127.0.0.1:8787/ is fine too, so perhaps the error should just be hidden?

Reproduction steps

• npm create cloudflare@latest
• use all defaults
• cd into worker dir and run npm start

[edge-rps]

I was installing PartyKit.io that uses cf and got the same error.

Windows 10

[goktugcy]

Same error

Windows 11

[william0bra]

Same here.

Windows 10

Multiple ips provided, but no one seems to be working.

[EDIT]
When I tried Astro's guide, the error appeared, and no URL was working.
https://docs.astro.build/en/guides/integrations-guide/cloudflare

Starting it with 'npm create cloudflare@latest' made that work despite the errors, but not loading env vars to Astro.locals.runtime.
https://developers.cloudflare.com/workers/get-started/guide/

[Cherry]

This also appears every single time the server reloads too.

[genesem]

Found a quick fix of the above for win10.

Use wrangler with parameter: --ip=127.0.0.1 to specify exact ip to bind.

[ghost]

Found a quick fix of the above for win10.

Use wrangler with parameter: --ip=127.0.0.1 to specify exact ip to bind.

I still get the error, but its acting like there is only one address, so rebuilds work now. Thanks 👍

[kentonv]

I suspect this is because localhost maps to two addresses -- one ipv4 (127.0.0.1) and one ipv6 (::1). workerd is telling you that it doesn't know how to listen on both addresses at once, so it chose to listen only to the ipv4 address.

This isn't a big deal in practice because everything trying to connect to "localhost" will almost certainly use the first address first, which is the one workerd bound to, so things will just work fine.

That said, it turns out I actually fixed this on the Unix side a while ago: capnproto/capnproto@f812158

But I didn't update async-io-win32.c++. If someone makes the same change to async-io-win32.c++ as was made to async-io-unix.c++ in that commit, it should allow workerd to bind to both IPs at once.

[1000hz]

This should be fixed in the next release.