Change https://golang.org/cl/363374 mentions this issue: net: add missing AddrPort Listen and Dial functions for UDP and TCP

I am generally in favor of API symmetry between related types, but I am not sure that 5 additional constructors are justified for dial/listen operations since they are not used in allocation hot paths like the aforementioned UDP read/write methods.

It's a one liner to convert netip.AddrPorts to the appropriate type for each of the existing Dial/Listen constructors (as you've done in the CL), so is it necessary to expand the API surface of net for these minor conveniences?

I assume the goal would be to eventually mark everything having to do with UDP/TCPAddr and net.IP as deprecated, and refactor those APIs to be converters to native netip.Addr/AddrPort functions. So adding the correct API is a necessary step in getting rid of the old problematic types (or, rather, relegating them to a deprecated wrapper file for Go 1).

as deprecated

Speaking of deprecated: anything without a context can be considered deprecated, so I wouldn't add anything new without context support.

We were discussing that the only API we should probably add are new methods on Dialer (ala #49097 (comment)) and ListenConfig.

Speaking of deprecated: anything without a context can be considered deprecated,

Is this documented anywhere? If the doc comments marked these as deprecated, it might motivate a lot of folks to move to context, as IDEs tend to mark deprecated function invocations with a strike or italics or some other marker.

And that's exactly why we haven't deprecated them officially. There's been talk (#48665, etc) about a lighter deprecation annotation that's like, "There are better ways to do this, but this way's fine I guess."

Ah. I was thinking the strike/italics/etc as a good way to enact change without breaking the Go 1 promise, rather than a prohibitive one. But I suppose if it's still "fine", seems like moving those over to netip.Addr[Port] would be a good idea?

I was thinking the strike/italics/etc as a good way

That's #16666

I don't think we should deprecate net.DialTCP. Deprecate means IDEs and such will warn about usage, and that's unnecessary.

Let's redirect this conversation to #49097. Closing as duplicate of that.

Before you close this issue, there's still https://go-review.googlesource.com/c/go/+/363374/ to consider, right?

This proposal has been added to the active column of the proposals project
and will now be reviewed at the weekly proposal review meetings.
— rsc for the proposal review group

@zx2c4, clearly I forgot to close this issue last week. CL 363374 implements this proposal, but I think we want to instead add the new functions in Dialer only. Using the typed forms is pretty rare for non-UDP things. I will leave this open to let this discussion about the non-Dialer functions continue toward a consensus, but it seems like we should do just the Dialer changes and not do CL 363374.

Does anyone object to only adding new Dialer methods, and not adding any top-level non-Dialer functions for AddrPort?

I'm probably a lone voice out there, so feel free to disregard that, but my feeling is that if we're moving to a new type, we should have API parity, and I sort of like the convenience of the top level functions for one-off little scripts and stuff. I don't really feel so strongly about it, though, and I recognize I'm likely in the minority, so 🤷 .

These functions are so little used that the API parity may not be worthwhile.
It sounds like there aren't too many objections to only adding new Dialer methods, so we should probably start with that.
We can always come back and add the others later if the need becomes clearer.

This will become a likely decline, and the Dialer methods are #49097.

Based on the discussion above, this proposal seems like a likely decline.
— rsc for the proposal review group

No change in consensus, so declined.
— rsc for the proposal review group