proposal: crypto, hash: add WriteString support

[josharian]

Add WriteString to crypto/sha1, crypto/sha256, hash/crc64, etc. Similar to #11805. Probably easy pickings.

[josharian]

And see CL 20524 for a place this would get used.

[sctb]

Initial CL submitted. I expect it's not quite there yet; glad to make any changes necessary.

[gopherbot]

CL https://golang.org/cl/22422 mentions this issue.

[minux]

My thought on this issue: If you want high performance hashing, then using strings is very probable already a mistake (the string must come from somewhere else, so it gets converted from []byte in the first place.) But if the string does not come from outside (I/O), then it's still better to use []byte directly. Could you please given a real scenarios where you want WriteString support on a crypto.Hash while still care about the utmost performance? While I agree that WriteString support on io.MultiWriter is fine, adding WriteString support to all io.Writer implementations doesn't look necessary esp. if doesn't bring enough performance improvement.

[josharian]

I think you're probably right. Before abandoning this, I'd like to do a bit of research in my Go corpus, but that might take a while.

[Scratch-net]

Having WriteString might indirectly suggest using it to hash passwords, which is a very, very bad idea.

[rsc]

I am very skeptical of an unsafe implementation of WriteString here.
Even the one in os.File is not unsafe - it does Write([]byte(s)).
I don't believe crypto should set a precedent that "unsafe is OK if it's for performance",
for others to follow.

[rsc]

Based on the discussion above (and the general lack of interest over 4 years), this seems like a likely decline.

Leaving open for a week for final comments.

[rsc]

No change in consensus, so declined.

[rsc]

Over on #38776 we realized that WriteString can be added without requiring use of unsafe, so maybe we should do that. WriteString is now included in #38776.