Re: tolua++ porting to 5.2

[Jay Carlson <nop@...>]

On Jan 14, 2013, at 1:30 PM, Tim Mensch wrote:

> On 1/14/2013 3:25 AM, Joop Boonen wrote:
>> I wonder if tolua++ could be a part of lua?
>> It's not supported any more and it's still quite widely used.
> 
> I can't recommend tolua++ to anyone; the core architecture they use for binding objects is profoundly broken, as I detailed in a previous message [1].

Quick summary: tolua++ is suitable if you can guarantee the lifespan of each C-side object is always longer than the Lua-side lifespan. Otherwise you get heisenbugs. And this is a hard guarantee to make.

"Never free anything until exit" has this property, and so does "Lua lifespan determines C lifespan"--but that is difficult to arrange for unless your objects do not reference other objects. 

Disciplined programmers who "know" when a reference should not be held can cover for this in the short run. The LuaFLTK binding I did back in the 4.0 era depended on smart programmers who never screwed up, but I had not realized the extent until Tim pointed it out.

The fix for this would involve C-side deallocation turning Lua-side references into tombstones, and I don't see an obvious way to implement tombstones other than something that works out to be Tim's serial numbers.

I have no idea how other general binding systems deal with the lifecycle mismatch problem, especially in the presence of reachability through member variables--somebody can squirrel away a reference to the Point in your Circle. 

LuaJIT ffi's philosophy is "we suck just as much as C"--in other words, it does not attempt to solve this for you, rather than get it wrong.

Jay

> [1] http://lua-users.org/lists/lua-l/2012-12/msg00601.html