diff options
| author | unknown <konstantin@mysql.com> | 2005-03-31 03:44:51 +0400 |
|---|---|---|
| committer | unknown <konstantin@mysql.com> | 2005-03-31 03:44:51 +0400 |
| commit | f485db429a26666a8468c1ec1227d6bd64a45682 (patch) | |
| tree | d891b9f81b26c11be96ca19e7eca03abf5ddfbfd /sql/item_cmpfunc.cc | |
| parent | 30c85129548f7d8267fab809a723672161c6fce9 (diff) | |
| download | mariadb-git-f485db429a26666a8468c1ec1227d6bd64a45682.tar.gz | |
A fix for Bug#9443 "mysql_client_test fails on linux and some solaris
platforms": yet another issue with floating pointer comparisons.
The fix uses the workaround with volatiles.
sql/item_cmpfunc.cc:
A fix for the failing mysql_client_test on some Intel platforms
when compiled with optimization.
We don't use -ffloat-store compileation as it may slow all
floating point operations.
Diffstat (limited to 'sql/item_cmpfunc.cc')
| -rw-r--r-- | sql/item_cmpfunc.cc | 10 |
1 files changed, 8 insertions, 2 deletions
diff --git a/sql/item_cmpfunc.cc b/sql/item_cmpfunc.cc index 690da1be18c..8498ab4800e 100644 --- a/sql/item_cmpfunc.cc +++ b/sql/item_cmpfunc.cc @@ -393,10 +393,16 @@ int Arg_comparator::compare_e_binary_string() int Arg_comparator::compare_real() { - double val1= (*a)->val(); + /* + Fix yet another manifestation of Bug#2338. 'Volatile' will instruct + gcc to flush double values out of 80-bit Intel FPU registers before + performing the comparison. + */ + volatile double val1, val2; + val1= (*a)->val(); if (!(*a)->null_value) { - double val2= (*b)->val(); + val2= (*b)->val(); if (!(*b)->null_value) { owner->null_value= 0; |