Re: [Patch] Finalization of function objects

["Thomas Harning Jr." <harningt@...>]

On Fri, Jul 31, 2009 at 3:07 PM, Thomas Harning Jr.<harningt@gmail.com> wrote:
> Any further work in the finalization patch?
>
> I tried rebuilding to see if perhaps it was my environment at the
> time, but I get the same error:
>
> x86_64
> http://pastebin.com/m69d835c0
> x86
> http://pastebin.com/m43f079ac
>
> I'll try to dig into this more tonight...
I've narrowed this down to the GC'ed coroutine. Ah hah!

do
	local g
	function collect() g = true end
	local function foo()
		finally(collect)
		coroutine.yield()
	end

	local co = coroutine.create(foo)
	coroutine.resume(co)
	-- force GC
	co = nil
	collectgarbage("collect")
	assert(g, "Not finalized")
	print"OK"
end

Works

do
	local g
	
	local function foo()
		finally(function() g = true end)
		coroutine.yield()
	end

	local co = coroutine.create(foo)
	coroutine.resume(co)
	-- force GC
	co = nil
	collectgarbage("collect")
	assert(g, "Not finalized")
	print"OK"
end

Does not safely work!

My guess is that the locally defined function somehow got cleaned up
during the coroutine finalization.... then the op got called and it
passes due to the fact that the memory hadn't yet been written over.

I don't know how valgrind misses it on your system (from way back)...
but I can reproducibly cause valgrind errors and 'fix' them.  Now to
see how in the world to fix this in the patch.

-- 
Thomas Harning Jr.