Bug#26536 func_time failure on vm-win2003-64-b, occurs every time

 - Avoid overflow in sec_since_epoch by shifting the time
   back 2 days for times close to the maximum range of my_time_t
 - Improve comment about why we need my_time_t
 - Patch will also fix timezone2.test


include/my_time.h:
  Improve comment for my_time_t, especially that we require
  it to be at least a 32bit unsigned type
sql/tztime.cc:
  - Add assert in sec_since_epoch to guard it against broken down time
  values that would cause overflow in my_time_t on systems with 32 bit
  my_time_t
  - Avoid overflow in sec_since_epoch by "shifting" the broken down
  time back 2 days for times that are near the max value.c

1 files changed, 7 insertions, 1 deletions

diff --git a/include/my_time.h b/include/my_time.h
index 14726d3a01d..3c8f4274409 100644
--- a/include/my_time.h
+++ b/include/my_time.h
@@ -30,7 +30,13 @@ extern uchar days_in_month[];
 
 /*
   Portable time_t replacement.
-  Should be signed and hold seconds for 1902-2038 range.
+  Should be signed and hold seconds for 1902 -- 2038-01-19 range
+  i.e at least a 32bit variable
+
+  Using the system built in time_t is not an option as
+  we rely on the above requirements in the time functions
+
+  For example QNX has an unsigned time_t type
 */
 typedef long my_time_t;