Re: svn commit: r1242397 - in /subversion/trunk/subversion/libsvn_subr: skel.c stream.c

I (Julian Foad) wrote:

> I suggest we follow up by doing both of:
>
>   * Change the implementation to always add an extra byte to that requested, to
> help buggy callers.  This is valid because the number passed in is already only
> a lower bound so callers can't assume that we don't increase it.
>
>   * Rev the API to 'svn_stringbuf_ensure2()' and document that the
> caller doesn't have to allow for the trailing NUL.
>
> Of course the implementation of _ensure() and _ensure2() would then be
> identical.  The point of adding one in the old _ensure() implementation is to
> help callers in third-party code, given that we've demonstrated a high
> chance of the existing API being misused.  The point of defining the _ensure2()
> API is so we can cleanly change callers to not add 1, without them contradicting
> the old API.  (It would add confusion for people reading the code if we removed
> the caller's "+1" and didn't rev the name.)

I'm following up with a patch in a new thread "[PATCH] String creation -- ensuring space for the NUL", <http://svn.haxx.se/dev/archive-2012-02/0415.shtml>.

- Julian
Received on 2012-02-14 13:31:54 CET