rubylang.10.webmeister / spamgourmet.com wrote:
> <blockquote>
> Can you try any other code which uses 64-bit math?
> </blockquote>
> 
> Can you recommend any test code? Those compilers build the whole Arch Linux ARM repository, so if it were a common problem, it should have been noticed by now.

Maybe some crypto or audio/video codecs.  It could also be compiler
options for math worth trying.

> <blockquote>
> 64-bits is needed to avoid overflow on VM state changes.
> Otherwise our caches could give false hits and crash.
> </blockquote>

> I see. But isn't it dangerous then to fall back to `unsigned long`,
> which only guarantees at least 32 bits? Or is the probability for a
> crash with only 32 bits still very low, and 64 bits just provide some
> extra safety?

unsigned long is 64-bits on 64-bit systems.  The probability of a 32-bit
crash is low and only affects codebases with frequent dynamic
method/module/class/const creation.  Maybe few use the 32-bit unsigned
long case and don't hit the problem.