From: Markus <markus@...>
Date: 2004-09-01T00:13:13+09:00
Subject: Re: (Revised) [PATCH] Subtle bug in bignum.c



On Tue, 2004-08-31 at 07:59, ts wrote:
> >>>>> "M" == Markus  <markus@reality.com> writes:
> 
> M> I've always thought (but remember, I don't really know C) that volatile
> M> was a hint to the compiler that code that _used_ something should _not_
> M> hold it in a register, because something else might modify it at any
> M> time (e.g. the system clock).  So I am not sure why it would help.
> 
>  Yes, you are right.
> 
>  Now `x' and `y' will not be in a register, and ruby will find it on the
>  stack if the GC is called in bignew() *even* if a new Bignum is created by
>  rb_big_clone() and stored at this address.
> 
>  The initial problem is that the new Bignum (created by rb_big_clone()) was
>  stored in a register (because `x' was in a register), then the content of
>  the register was replaced with another variable (RBIGNUM(y)->digits) when
>  the gc was called and this is why ruby was unable to mark the Bignum
> 
> 
> Guy Decoux

Ah yes.  Thank you.  I had not realized (but should have guessed) that
the gc walked the stack to find things that were reachable.  I will work
up & test a third (and I hope final) patch.

The only remaining question is: do you think there may be analogous bugs
lurking elsewhere in the code?  I suppose it might be possible to hunt
for them in a semi-automated way, but before I embark on that is there
any reason to suppose that this situation is more or less common?

-- Markus