And if we moved the remaining functions to a new package, io/fileio, we could deprecate io/ioutil entirely.

But should this happen as part of a stdlib v2? It is backwards compatible, but at the cost of a fair amount of duplication.

The fileio suggestion is #19660 for what it's worth.

For what it's worth, I think we can correct these three symbols without waiting for a more comprehensive stdlib v2.
They clearly belong in io, by almost any criteria. (And we're not likely to start chopping up io or anything like that.)

The remaining symbols I think we can deal with at another time. It's possible they just belong in os, instead of isolated in a side package all by themselves, except that there's already an os.TempDir.

The reactions I see above (one comment, bunch of emoji) are all positive. Does anyone object to accepting this proposal?

No objection from me. I can never remember what's in ioutil vs io.

Based on the discussion above, this seems like a likely accept.

(Part of the reason to do this is that then you never have to use io/ioutil when working with a non-operating-system fs.FS. io/ioutil becomes OS-only.)

Sorry to be the naysayer here, but how should we deal with the break in backwards compatibility this move causes? The go statement in the go.mod file and go fix or go fmt or such? It seems a lot of work for what is in essence a cosmetic change.

Once we have generics, we will have to rethink the whole standard library anyway. I prefer all breaking changes like this one to be made at once for the V2 standard library, then I only will have to upgrade once and not a dozen times.

My assumption was that we would move the base symbols to io, then type alias them into io/ioutil. We can deprecated the aliases too if that is not excessive.

package io

var Discard io.Writer = devNull(0) // maybe we could make this a const?
func NopCloser(r io.Reader) io.ReadCloser  { /*...*/ }
func ReadAll(r io.Reader) ([]byte, error) { /*...*/ }

package ioutil

var Discard io.Writer = io.Discard
func NopCloser(r io.Reader) io.ReadCloser  { return io.NopCloser(r) }
func ReadAll(r io.Reader) ([]byte, error) { return io.ReadAll(r) }

From the proposal:

Obviously we cannot delete these three from io/ioutil. But we can move the implementations to io and leave wrappers behind.

I've clarified in the text above that nothing will break.

Other than that comment, seems like there is no change in consensus. Accepted.

Change https://golang.org/cl/245657 mentions this issue: io: move Discard, NopCloser from ioutil

Change https://golang.org/cl/245658 mentions this issue: io: reimplement and move ReadAll from ioutil

Change https://golang.org/cl/263141 mentions this issue: io: adopt Discard, NopCloser, ReadAll from io/ioutil

Change https://golang.org/cl/266364 mentions this issue: os: add ReadFile, WriteFile, CreateTemp (was TempFile), MkdirTemp (was TempDir) from io/ioutil

Change https://golang.org/cl/285377 mentions this issue: doc/go1.16: mention deprecation of io/ioutil