`DOCKER_HOST` without `unix://` prefix prints a confusing error: `Cannot connect to the Docker daemon at tcp://localhost:2375/foo.sock.`

[AkihiroSuda]

Description

$ DOCKER_HOST=/invalid.sock docker version
Client:
 Version:           27.5.1
 API version:       1.47
 Go version:        go1.22.11
 Git commit:        9f9e405
 Built:             Wed Jan 22 13:39:08 2025
 OS/Arch:           linux/arm64
 Context:           default
Cannot connect to the Docker daemon at tcp://localhost:2375/invalid.sock. Is the docker daemon running?

This error message is quite confusing, as the original invalid DOCKER_HOST value does not contain tcp://localhost:2375. Probably it should return an error like invalid $DOCKER_HOST value "/invalid.sock": must start with "unix://" or "tcp://"

[thaJeztah]

Thx! Yeah, I think there's some dubious (and inconsistent) handling for these in some areas.

I suspect I may have introduced this when I replaced some code with url.Parse (adding a scheme if missing), but it's possible it was elsewhere as I found multiple places where logic for this was implemented, but ever-so-slightly different.

[thaJeztah]

Oh, and if we're gonna add that to the error message, ssh:// should probably also be there.

I suspect part of this may be due to logic where hostname (without scheme) would be considered tcp://hostname, and we may be using tcp as default if not set, which in this case, gets seen as <host missing>/invalid.sock -> <default host and scheme>/invalid.sock.

But perhaps in this case, we should consider checking if the value starts with something that indicates a path (/, ./), as in that case the user very likely meant "path to a socket".

[ZayaanRahman]

Hi! I'm interested in contributing to the Docker cli, and was wondering if this was a good issue I could get started on.