Add StringReader, StringWriter, and StringStream

[emberian]

The docs for std::io::mem mention that there should be some sort of implementation for strings.

The Reader is straightforward.

For Writer, this is tricky, since one cannot just write arbitrary data to a string, it must be UTF-8. Where should the boundary be drawn? Should individual writes be checked for UTF-8ness? Should the writer instead check validity of the underlying string when it is destructed? Should these operate on a &str (and thus not be growable), or should they make their own ~str?

I'm leaning towards a StringWriter that writes to an internal ~[u8], with an explicit fn get_string(self) -> Option<~str> that destroys the writer and returns the string if it is valid, None otherwise. Or, maybe it should be Result<~str, ~[u8]>, so the buffer could still be handled if desired.

[huonw]

Assuming we're not checking each byte (which seems reasonable, one could feasibly be streaming 1000 bytes at a time from something and so get split across a codepoint), couldn't the StringWriter be implemented as the get_string method on MemWriter? (Although, on naming, I'd say into_str() -> ....)

(Similarly, isn't StringReader just BufReader::new(s.as_bytes()), i.e. it could be left as that, or there could be from_utf8(s: &str) creators.)

[emberian]

Yeah, adding a method to MemWriter sounds reasonable! The BufReader makes sense too.

[pongad]

I'm on it. Since it hasn't been updated in 4 months, if this is no longer relevant please let me know :D

[sfackler]

I'm not sure this makes a ton of sense to add. Doing the conversion manually is pretty painless: str::from_utf8_owned(w.unwrap()).

[pongad]

@sfackler I see your point. If there is a consensus, we can remove this issue.

[alexcrichton]

A very similar rfc was postponed, so I'm going to close this for now.

These operations are all possible with MemWriter and friends with the utf8 -> str conversion functions.