Re: bk commit into 6.0 tree (malff:1.2644) BUG#36457

[Marc Alff]

Hi Davi


Davi Arnaut wrote:
> Hi Marc,
>
>
> marc.alff@stripped wrote:
>
> [..]
>
>   
>> diff -Nrup a/sql/my_decimal.cc b/sql/my_decimal.cc
>> --- a/sql/my_decimal.cc	2007-10-11 11:29:06 -06:00
>> +++ b/sql/my_decimal.cc	2008-05-13 17:42:38 -06:00
>> @@ -41,7 +41,7 @@ int decimal_operation_results(int result
>>  			"", (long)-1);
>>      break;
>>    case E_DEC_OVERFLOW:
>> -    push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
>> +    push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_WARN,
>>                          ER_TRUNCATED_WRONG_VALUE,
>>                          ER(ER_TRUNCATED_WRONG_VALUE),
>>  			"DECIMAL", "");
>>     
>
> How this will be elevated to a error when sql_mode is set to
> strict_all_tables (traditional)? or, shouldn't we also enforce the
> traditional mode on selects too?
>
>   
Please note that this bug report, and the patch, is not about SQL_MODE,
and that no logic involving SQL_MODE is changed here.

To answer your question, when the 'strict_all_tables' mode is in effect:
- the code in my_decimal.c calls push_warning_printf(WARN_LEVEL_WARN)
- push_warning_printf() calls push_warning()
- in push_warning, thd->really_abort_on_warning() will return TRUE,
so that the push_warning() implementation internally change the level to 
error and calls my_error().

This is how the strict mode is implemented today, I am explaining it, 
not defending it.

THD::really_abort_on_warning() depends on:
(a) - THD::abort_on_warning
(b) - THD::transaction.stmt.modified_non_trans_table
(c) - THD::variable.sql_mode & MODE_STRICT_ALL_TABLES

(a) is set only for some -- not all -- statements
(b) relates to MODE_STRICT_TRANS_TABLES
(c) is MODE_STRICT_ALL_TABLES

Should the traditional mode be enforced on selects also ?
That depends on the semantic of this mode:

'strict' modes are *not* used to protect the execution of any SQL statement,
so the semantic is not comparable to, for example, "gcc -Wall -Werr".

' strict' modes are used to protect the *data* in a table, and are 
enforced only when data is written to a table.
This is why THD::abort_on_warning is set only in a few places, in 
sql_insert.cc, sql_update.cc, etc.

... so no, strict modes should not be enforced during a SELECT, given 
the current definition of this mode.

This topic is the subject of long debates every 2 years or so, and 
traces back to very old history in mysql.

For example, see bug#5929, "to be fixed later", which was to be fixed in 
5.1, then in 5.2 ...

Best Regards,
Marc.