Description of problem: bdb test fails on s390x
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. build RPM with "make test" enabled
The test results show bogus values for the "rows" column of EXPLAIN
output. The values are a factor of 2^32 more than they should be,
which looks a whole lot like a 32-vs-64-bit programming error, but
there is no sign of similar problems on our other 64-bit arches.
Perhaps it is a compiler bug?
Even more interesting: on RHEL3, mysql-3.23.58 also fails self-test on s390x and no other
architecture. But it's a different failure. Maybe there is some kind of generic s390x issue in mysql?
Ah-hah, it's a bogus assumption that atomic_add can be used on 8-byte
quantities. Fixed in our 4.1.11 release, and reported upstream as
The observed failure is relatively cosmetic: wrong row count estimates output by
EXPLAIN. I am not sure if a wrong rows value could lead to any more-serious
problems... but we will need to back-patch 4.1.11 into RHEL4 anyway, so we might
as well include this too.
Further thought: maybe the RHEL3 issue is also from trying to use atomic_add on
the wrong data width, even though it's not BDB specifically? Should try
backporting the no-atomic patch.
Back-patched into RHEL4 for QU1 ... still need to investigate whether a similar
change would improve matters for RHEL3.
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.