svn commit: samba r7759 - in branches/SAMBA_4_0/source/lib/ldb/common: .

[Andrew Tridgell]

Derrell,

 > That's kinda yucky.  (No, that's really yucky.)  Is there some reason that you
 > wouldn't want the caller to just free the error string, or better yet, for the
 > error string to always just be allocated on the ldb context so that it gets
 > freed automagically?  With everything so clean in ldb, this seems like a real
 > oversight.

yes, there are very good reasons for doing it the way it is done!

1) callers mostly don't look for the error string, so if we allocate
   at the time the error happens and the caller doesn't call
   ldb_errstring() then we have leaked. So the backend has to free it
   in that case anyway.

2) when the backend gets an allocation error, the last thing you want
   to do to report the error is another allocation!

3) always allocating on the ldb context would give us a massive memory
   leak for long lived connections.

Notice that ldb_errstring() retruns a "const char *". The "const" is a
hint that the caller does not free (note that talloc_free() takes a
non-const, just like free()).

Having an error function that returns a string that the caller does
not free is the normal approach to this problem in lots and lots of
APIs. Have a look at calls like ldap_err2string() for an example.

Cheers, Tridge