address review comments

Author: novalis

text/0000-read-all.md

@@ -26,7 +26,7 @@ read into, and will attempt to fill that entire slice with data.
 It will loop, calling read() once per iteration and attempting to read
 the remaining amount of data.  If read returns EINTR, the loop will
 retry.  If there are no more bytes to read (as signalled by a return
-of Ok(0) from read()), a new error type, ErrorKind::ReadZero, will be
+of Ok(0) from read()), a new error type, ErrorKind::ShortRead, will be
 returned.  In the event of another error, that error will be
 returned. After a read call returns having successfully read some
 bytes, the total number of bytes read will be updated.  If that
@@ -46,5 +46,14 @@ One alternative design would return some new kind of Result which
 could report the number of bytes sucessfully read before an error.
 This would be inconsistent with write_all, but arguably more correct.
 
+Another would be that ErrorKind::ShortRead would be parameterized by
+the number of bytes read before EOF.  The downside of this is that it
+bloats the size of io::Error.
+
+Finally, in the event of a short read, we could return Ok(number of
+bytes read before EOF) instead of an error.  But then every user would
+have to check for this case.  And it would be inconsistent with
+write_all.
+
 Or we could leave this out, and let every Rust user write their own
 read_all function -- like savages.