dovecot - Mar 2010 - Fix for the consume_results leak

I'm not really comfortable with making non-trivial changes to this
code but here is at least a fix for the PGresult leak (verified to
both apply to 1.2.11 and 2.0.beta4).

-----
--- dovecot/src/lib-sql/driver-pgsql.c	15 Mar 2010 18:18:14 -0000	1.1.1.2
+++ dovecot/src/lib-sql/driver-pgsql.c	26 Mar 2010 14:13:18 -0000	1.1.1.2.4.1
@@ -247,6 +247,17 @@
 	return 0;
 }
 
+static inline int more_results(PGconn *pg)
+{
+	PGresult *pgres;
+
+	pgres = PQgetResult(pg);
+	if (!pgres) return 0;
+
+	PQclear(pgres);
+	return 1;
+}
+
 static void consume_results(struct pgsql_db *db)
 {
 	do {
@@ -255,7 +266,7 @@
 
 		if (PQisBusy(db->pg))
 			return;
-	} while (PQgetResult(db->pg) != NULL);
+	} while (more_results(db->pg));
 
 	if (PQstatus(db->pg) == CONNECTION_BAD)
 		io_remove_closed(&db->io);

On Fri, 2010-03-26 at 15:50 +0100, Rainer Weikusat
wrote:
> I'm not really comfortable with making non-trivial changes to this
> code but here is at least a fix for the PGresult leak (verified to
> both apply to 1.2.11 and 2.0.beta4).
Thanks. I did a slightly different change though:
http://hg.dovecot.org/dovecot-2.0/rev/5a6aaf88f15c

(the extra result->pgres = NULL probably doesn't fix anything, but
avoids potential problems and it wasn't worth a separate commit :)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
URL:
<http://dovecot.org/pipermail/dovecot/attachments/20100327/ad60e7a6/attachment-0002.bin>

Timo Sirainen <tss at iki.fi> writes:
> On Fri, 2010-03-26 at 15:50 +0100, Rainer Weikusat wrote:
>> I'm not really comfortable with making non-trivial changes to this
>> code but here is at least a fix for the PGresult leak (verified to
>> both apply to 1.2.11 and 2.0.beta4).
>
> Thanks. I did a slightly different change though:
> http://hg.dovecot.org/dovecot-2.0/rev/5a6aaf88f15c
That was the first idea I had, but I thought that the original layout
of the loop (loop until all results have been dealt with) made more
sense. Since there didn't seemed to be a sensible way to do this
without goto (some people strongly object to :->), I then had the idea
that this should really be a problem for the compiler and wrote the
inline routine which ended up creating exactly the 'loop while
PQconsumeInput didn't return an error'-loop in machine code.