BUG#46961 - archive engine loses rows during self joining select!

SELECT with join (not only self-join) from archive table may
return incomplete result set, when result set size exceeds
join buffer size.

The problem was that archive row counter was initialzed too
early, when ha_archive::info() method was called. Later,
when optimizer exceeds join buffer, it attempts to reuse
handler without calling ha_archive::info() again (which is
correct).

Fixed by moving row counter initialization from
ha_archive::info() to ha_archive::rnd_init().

mysql-test/r/archive.result:
  A test case for BUG#46961.
mysql-test/t/archive.test:
  A test case for BUG#46961.
storage/archive/ha_archive.cc:
  Since a cursor may get reused without a call to ::info(),
  move assignment of scan_rows to a proper place, that is
  ::rnd_init().

1 files changed, 1 insertions, 1 deletions

diff --git a/storage/archive/ha_archive.cc b/storage/archive/ha_archive.cc
index 1146b2eb73a..f1b9c26dfd6 100644
--- a/storage/archive/ha_archive.cc
+++ b/storage/archive/ha_archive.cc
@@ -993,6 +993,7 @@ int ha_archive::rnd_init(bool scan)
   /* We rewind the file so that we can read from the beginning if scan */
   if (scan)
   {
+    scan_rows= stats.records;
     DBUG_PRINT("info", ("archive will retrieve %llu rows", 
                         (unsigned long long) scan_rows));
 
@@ -1461,7 +1462,6 @@ int ha_archive::info(uint flag)
   stats.records= share->rows_recorded;
   pthread_mutex_unlock(&share->mutex);
 
-  scan_rows= stats.records;
   stats.deleted= 0;
 
   DBUG_PRINT("ha_archive", ("Stats rows is %d\n", (int)stats.records));