>>>>> "Bonis," == Bonis, Wim <bonis@stripped> writes:
>> The bug should be in this functions: (sql/field.cc):
>>
>> void Field_double::store(double nr)
>> {
>> if (decimals != 31)
>> nr=floor(nr*log_10[decimals]+0.5)/log_10[decimals]; // To
>> fixed point
>> doublestore(ptr,nr);
>> }
>>
>> double Field_double::val_real(void)
>> {
>> double j; doubleget(j,ptr);
>> return j;
>> }
>>
>> The doublestore / doubleget macros are defined in include/global.h
Bonis,> I tried it on a mips(qube2) but the functions you quoted may not have been
Bonis,> executed. I have added printf´s into the two functions but the following
Bonis,> session did not trigger them:
mysql> select 1.01+1;
Bonis,> +---------------------------------------------------------------------------
Bonis,> ----------------------------------------------------------------------------
Bonis,> ----------------+
Bonis,> | 1.01+1
Bonis,> |
Bonis,> +---------------------------------------------------------------------------
Bonis,> ----------------------------------------------------------------------------
Bonis,> ----------------+
Bonis,> |
Bonis,> -416149464319656471792043282545976376852126312593339196866087348513668409107
Bonis,> 6452136392930344036015456104570933577105313965196989532097205520492824844667
Bonis,> 5369066496.00 |
Bonis,> +---------------------------------------------------------------------------
Bonis,> ----------------------------------------------------------------------------
Bonis,> ----------------+
The above will not trigger the Field_double code. The code is
triggered if you insert/retrieve something to/from a column declared
as 'double'.
Bonis,> here is what the developers of the QUBE say to that Problem:
Bonis,> Maybe you could help them (and us, the users of the qube) to fix the
Bonis,> Problem.
<cut>
Bonis,> egcs-1.0.2-9 that i use does not handle alignment issues related to double
Bonis,> arguments consistently. nor does the latest port available for mipsel. a
Bonis,> patch
Bonis,> has recently been released for i386 (feb 99) and there is quite some
Bonis,> dicussion
Bonis,> on it (search for "egcs double alignment" on the web).
Bonis,> more details:
Bonis,> mysql uses an Item_real C++ object that has a constructor which takes
Bonis,> arguments
Bonis,> (const char *, double, uint, uint). though the caller aligns the double at
Bonis,> a
Bonis,> 64-bit boundary, the code generated for the constructor accesses the double
Bonis,> from the 32-bit pad after the const char *. i could be wrong as the
Bonis,> fprintf i use could have the same problems too. otoh, the fprintf that mysql
Bonis,> uses to print out the result could be hosed. getting gdb to break execution
Bonis,> when the constructor is invoked has been abortive.
<cut>
If you have problems debugging mysqld because its threaded, the
--one-thread switch to mysqld may help!
Does the above aligment problem also exists with egcs 1.1.2 ?
Can you try to verify if the bug is really in the call to the
constructor or in the double -> char conversion.
A breakpoint at String::set(double num,uint decimals) should answer
this one...
Regards,
Monty
