Add a comment explaining one of the ways that pgbench fails to fully
authorTom Lane <tgl@sss.pgh.pa.us>
Wed, 12 Mar 2008 02:18:33 +0000 (02:18 +0000)
committerTom Lane <tgl@sss.pgh.pa.us>
Wed, 12 Mar 2008 02:18:33 +0000 (02:18 +0000)
comply with TPC-B.  Per Itagaki Takahiro and discussion of bug#3681.

contrib/pgbench/pgbench.c

index 70daa3ec071a1d56d4e70d5ac22ec4f5e58b4cca..e5b04e95a2afbc25968bdbfce5eb7043c15c8f4e 100644 (file)
@@ -811,6 +811,16 @@ init(void)
 {
        PGconn     *con;
        PGresult   *res;
+       /*
+        * Note: TPC-B requires at least 100 bytes per row, and the "filler"
+        * fields in these table declarations were intended to comply with that.
+        * But because they default to NULLs, they don't actually take any
+        * space.  We could fix that by giving them non-null default values.
+        * However, that would completely break comparability of pgbench
+        * results with prior versions.  Since pgbench has never pretended
+        * to be fully TPC-B compliant anyway, we stick with the historical
+        * behavior.
+        */
        static char *DDLs[] = {
                "drop table if exists branches",
                "create table branches(bid int not null,bbalance int,filler char(88)) with (fillfactor=%d)",