[mdb-dev] [PATCH] read all table rows

[David M. <mdb...@dm...>]

Hi Brian, list,

I've been using an in-house heavily patched version of 0.5 for a long 
time now, and finally decided to try to update to the latest CVS, 
because you guys have made a lot of great progress.  Many things are 
greatly improved over the old 0.5 version.

I found a small bug that actually traces back to a mis-merge of some 
patches I sent you a year ago!

The issue is the the structure for a table, num_rows contains a total 
count of rows in the table (actually, only counts non-deleted rows). 
However, the method mdb_read_row takes a row number relative to the 
block only.

The bug occurs when you have a table with a small number of rows, all in 
one block, but with a bunch of deleted rows.

In my case there are 33 total non-deleted rows (table->num_rows == 33). 
  There are 44 rows in the first block, and 3 in the second.  Doing the 
math, there are 14 deleted rows.

However, the check in mdb_read_row:

         if (table->num_rows <= row)
                 return 0;

in mdb_read_row causes the routine to skip row indexes 33 -> 44 in the 
first data block.

Looking back at my source (0.5 plus patches) the code reads:

	if (table->num_rows == 0)
		return 0;

which is what it had been for years before that, and I think is still 
valid as an optimization.

Changing the code back makes it work for me again.

This may be white-space mangled due to cut and paste.

===================================================================
RCS file: /cvsroot/mdbtools/mdbtools/src/libmdb/data.c,v
retrieving revision 1.87
diff -u -r1.87 data.c
--- src/libmdb/data.c   15 Jan 2005 05:02:08 -0000      1.87
+++ src/libmdb/data.c   6 Feb 2005 20:16:21 -0000
@@ -229,7 +229,7 @@
         MdbField fields[256];
         int num_fields;

-       if (table->num_rows <= row)
+       if (table->num_rows == 0)
                 return 0;

         row_start = mdb_pg_get_int16(mdb, (fmt->row_count_offset + 2) + 
(row*2));