Re: Lua 5.1 work 0 bug reports?

[Mark Hamburg <mhamburg@...>]

Write barriers are an incremental GC issue. They basically come up when you
change a field in an object that's already been marked so that it points to
an unmarked object. A writer barrier keeps one from losing the second
object.

Mark

on 7/6/04 7:49 AM, Virgil Smith at Virgil@Nomadics.com wrote:

>> I believe that lua_setfenv needs a write barrier.
>> Adding one made another bug I was hunting down go away.
> 
> OK, I've written device drivers for multiple platforms (write barriers are
> nice for performing hardware access and fast/specific thread
> synchronization), however, I feel that I for one am missing the point of
> something.
> 
> Why would the Lua code need write barriers?
> 
> 
> --
> Either I've yet to learn something with respect to Lua's design and
> implementation constraints to make it multithreading friendly, or I'm
> missing something with regards to C sequencing guarantees and how they
> relate to <all> software, not just segments accessing hardware (either for
> direct hardware control, or for multi-threading/interrupt handling).
> And maybe others could benefit from the explanation so I'll be brave enough
> to admit possible shortcomings in the open :-)
> 
>