Re: Why GCTM stop gc step before calling userdata's finalizer?

["Soni L." <fakedme@...>]

On 09/01/16 11:54 PM, 书呆彭, Peng Yi wrote:
> 在 2016/1/9 20:35, 张伟智 写道:
> > Lua version is 5.2.4.
> > In lgc.c line 820, function GCTM,
> > g->gcrunning is set to 0 to avoid GC steps,
> > I think if g->gcrunning remains unchanged,
> > line 824 luaD_pcall will trigger luaC_step->luaC_forcestep->GCTM 
> > during executing __gc finalizer.
> > But what is the problem if recursive enter GCTM ?
>
> In my understanding, the Lua VM is reentrant, but the finalizers are 
> meant to be called
> in sequence. according to Lua's semantic, the finalizers are called in 
> the reverse order
> that the objects were marked for for finalization. so the finalizer 
> for the next object
> must be called _after_ the current object's finished its execution.
What if it were to add the finalizers to a queue instead?

Also, setmetatable({}, {__gc=function(t) while true do 
table.insert(t,{}) end end}) -- (instant DoS. try it with a debug 
hook-based sandbox, it didn't work in Lua 5.1 because you needed either 
newproxy() or an unprotected userdata metatable to do it. this is also 
an issue when debugging finalizers with e.g. ZeroBrane Studio.)

--
Disclaimer: these emails may be made public at any given time, with or without reason. If you don't agree with this, DO NOT REPLY.