Failure to detect error in allocating transcation_prealloc_size

Bug when reported in MySQL http://lists.mysql.com/commits/24309. Looks like we have this fix in our source base

probably a % in format somewhere.

or another lovely artifact of the horrible variable system....

Test case has been commented out for now in variables-big test, please make sure this is uncommented after the bug is fixed

this looks to be happening in set_var.cc in the cast to (uint32_t) for values greater then 4294967295 in the test case. The result of this cast is then passed through to throw_bounds_warning() and fails with a value of 0 for setting a transaction_prealloc_size to 4294967296. Other larger values dont throw warnings although their values are bogus from the cast they are still within range.

static bool get_unsigned32(Session *, set_var *var)
{
  if (var->value->unsigned_flag)
    var->save_result.uint32_t_value= (uint32_t) var->value->val_int();
  else
  {
    int64_t v= var->value->val_int();
    var->save_result.uint32_t_value= (uint32_t) ((v < 0) ? 0 : v); <----- possibly bad cast
  }
  return 0;
}

Seems like a pretty straightforward fix:

Either change the transaction_prealloc_size to a uint64_t or fix the throw_bounds_warning() function to not truncate/change the actual value returned by val_int() (which is of type int64_t...

Joe, care to give it a try?

-jay

Sure, I'll give it a try!

Thanks

   -Diego

On Wed, Aug 26, 2009 at 12:56 PM, Jay Pipes<email address hidden> wrote:
> Seems like a pretty straightforward fix:
>
> Either change the transaction_prealloc_size to a uint64_t or fix the
> throw_bounds_warning() function to not truncate/change the actual value
> returned by val_int() (which is of type int64_t...
>
> Joe, care to give it a try?
>
> -jay
>
> --
> Failure to detect error in allocating transcation_prealloc_size
> https://bugs.launchpad.net/bugs/377826
> You received this bug notification because you are a member of Drizzle-
> developers, which is subscribed to Drizzle.
>
> Status in A Lightweight SQL Database for Cloud and Web: Confirmed
>
> Bug description:
> Two problems with the variables-big.test in the big test suite.
>
> 1. Error is not detected when running the command:
>
> set session transaction_prealloc_size=1024*1024*1024*5;
>
> 2. Incorrect error message is being produced when running the command:
>
> set session transaction_prealloc_size=1024*1024*1024*4; The error message says:
>
> Error   1292    Truncated incorrect transaction_prealloc_size value: '0'
>
> and it should say:
>
> Error   1292    Truncated incorrect transaction_prealloc_size value: '4294967296'
>
> The big test suite is being run from on Hudson on drizzle-os (OpenSolaris x86) machine, see http://gorf.tangent.org/hudson/job/drizzle-drizzle-os-test-big/
>
> % diff suite/big/r/variables-big.re*
> 15c15
> < Error 1292    Truncated incorrect transaction_prealloc_size value: '0'
> ---
>> Warning       1292    Truncated incorrect transaction_prealloc_size value: '4294967296'
> 19a20,21
>> Warnings:
>> Warning       1292    Truncated incorrect transaction_prealloc_size value: '5368709120'
>

--
Diego Medina
Web Developer
http://www.fmpwizard.com

Will be pushing the fix soon

Migrating trans_prealloc_size to uint64_t causes an incredible slowdown on solaris sparc... so I made it back into a uint32_t and then added a bounds check in get_unsigned32.