Fix Bug #9191 "TIMESTAMP/from_unixtime() no longer accepts 2^31-1"

(4.1 version, with post-review fixes)
  
  The fix for another Bug (6439) limited FROM_UNIXTIME() to
  TIMESTAMP_MAX_VALUE which is 2145916799 or 2037-12-01 23:59:59 GMT,
  however unix timestamp in general is not considered to be limited 
  by this value. All dates up to power(2,31)-1 are valid.
  
  This patch extends allowed TIMESTAMP range so, that max
  TIMESTAMP value is power(2,31)-1. It also corrects
  FROM_UNIXTIME() and UNIX_TIMESTAMP() functions, so that
  max allowed UNIX_TIMESTAMP() is power(2,31)-1. FROM_UNIXTIME()
  is fixed accordingly to allow conversion of dates up to
  2038-01-19 03:14:07 UTC. The patch also fixes CONVERT_TZ()
  function to allow extended range of dates.
  
  The main problem solved in the patch is possible overflows
  of variables, used in broken-time representation to time_t
  conversion (required for UNIX_TIMESTAMP).


acinclude.m4:
  Add new macro to check time_t range
configure.in:
  Call the macro to check time_t range
include/my_time.h:
  Move time-related defines to proper place.
  Add a function to perform a rough check if
  a TIMESTAMP value fits into the boundaries.
  Note: it is defined as "static inline", as
  otherwise libmysql won't compile (due to the
  way how gcc handles "inline" directive).
mysql-test/r/func_time.result:
  Update test result
mysql-test/r/timezone.result:
  Update test result
mysql-test/r/timezone2.result:
  Update test result
mysql-test/t/func_time.test:
  Add test for Bug#9191 and update test to be consistent
  with new TIMESTAMP boundaries
mysql-test/t/timezone.test:
  Update old tests to be consistent
  with new TIMESTAMP boundaries
mysql-test/t/timezone2.test:
  Update tests for convert_tz to be consistent with new
  TIMESTAMP boundaries
sql/item_timefunc.cc:
  Fix convert_tz to allow dates from the new (extended)
  TIMESTAMP range
sql/mysql_priv.h:
  Move time handling defaults to my_time.h
sql-common/my_time.c:
  Because of increased TIMESTAMP_MAX_VALUE overflows in my_system_gmt_sec()
  became possible. Here we make it safe against the overflows by stepping
  back from the boundary dates which are likely to trigger them.
sql/time.cc:
  Update TIME_to_timestamp to allow conversion of
  extended date range
sql/tztime.cc:
  Fix new (4.1) implementation of broken-down time representation
  to time_t conversion routine to avoid overflows during conversion
  of boundary dates
mysql-test/r/timezone4.result:
  New BitKeeper file ``mysql-test/r/timezone4.result''
mysql-test/t/timezone4-master.opt:
  New BitKeeper file ``mysql-test/t/timezone4-master.opt''
mysql-test/t/timezone4.test:
  New BitKeeper file ``mysql-test/t/timezone4.test''

1 files changed, 24 insertions, 0 deletions

diff --git a/acinclude.m4 b/acinclude.m4
index 102f869e0d4..0337b9de0cd 100644
--- a/acinclude.m4
+++ b/acinclude.m4
@@ -1835,6 +1835,30 @@ dnl END OF MYSQL_CHECK_NDBCLUSTER SECTION
 dnl ---------------------------------------------------------------------------
 
 
+dnl
+dnl  Macro to check time_t range: according to C standard
+dnl  array index myst be greater then 0 => if time_t is signed
+dnl  the code in the macros below won't compile.
+dnl
+
+AC_DEFUN([MYSQL_CHECK_TIME_T],[
+    AC_MSG_CHECKING(if time_t is unsigned)
+    AC_COMPILE_IFELSE([AC_LANG_PROGRAM(
+        [[
+#include <time.h>
+        ]],
+        [[
+        int array[(((time_t)-1) > 0) ? 1 : -1];
+        ]] )
+    ], [
+    AC_DEFINE([TIME_T_UNSIGNED], 1, [Define to 1 if time_t is unsigned])
+    AC_MSG_RESULT(yes)
+    ],
+    [AC_MSG_RESULT(no)]
+    )
+])
+
+
 dnl By default, many hosts won't let programs access large files;
 dnl one must use special compiler options to get large-file access to work.
 dnl For more details about this brain damage please see: