Re: LuaJIT optimization bug?

[David Kastrup <dak@...>]

Pierre Chapuis <catwell@archlinux.us> writes:

> On Wed, 17 Aug 2011 16:09:25 +0200, Pierre Chapuis wrote:
>
>> To give you an idea of the butterfly effect I face:
>>
>>   -- incorrect result
>>   bit_set_1(_bset,_byte,_bit)
>>
>>   -- incorrect result
>>   bit_set_1(_bset,_byte,_bit)
>>   assert(true)
>>
>>   -- incorrect result
>>   bit_set_1(_bset,_byte,_bit)
>>   assert(true)
>>   assert(true)
>>
>>   -- correct result (!)
>>   bit_set_1(_bset,_byte,_bit)
>>   assert(true)
>>   assert(true)
>>   assert(true)
>>
>>   -- incorrect result (!!!)
>>   bit_set_1(_bset,_byte,_bit)
>>   assert(true)
>>   assert(true)
>>   assert(true)
>>   assert(true)
>
> It's actually worse than I thought... It works with 3, 6 or
> 19 successive assert(true) statements. It doesn't work for any
> other number of assert(true) statements between 0 and 20.
>
> Chaos? (http://www.wolframscience.com/nksonline/page-27)

It will probably work with 22 again.  Sounds like a question of
alignment.

-- 
David Kastrup