Re: omission in lua_yield docs

[John Belmonte <john@...>]

Of course-- thanks.

I think it would still be helpful to clarify this in the yield docs,
as is done e.g. for lua_error.


On Sat, May 26, 2012 at 11:18 PM, Peng Zhicheng
<pengzhicheng1986@gmail.com> wrote:
> 于 2012-5-27 10:40, John Belmonte 写道:
>
>> I noticed that the lua_yield / yieldk docs in 5.1 and 5.2 don't define
>> the return value of the function.  (If I understand correctly it's the
>> number of arguments placed on the stack by lua_resume.)
>>
> No.
> in plain Lua (not LuaJIT, I mean), lua_yield actually doesn't return at the
> C API side.
> lua_yield would do a `longjmp' to perform the yield, which would unwind all
> the C stack
> till the nearest recover point, i.e. the nearest setjmp call.
>
> this is the reason why you can't use pcall in a yieldable coroutine in Lua
> 5.1, in which case
> the yield would be caught by the recover point set by pcall, not the one by
> lua_resume.
> Lua 5.2 fixed this restriction by not setting a recover point in lua_pcall
> if one is already set by lua_resume.
>
> the yield/resume semantic is all about the Lua side. lua_resume simply puts
> the values onto
> the stack of the thread and calls into the Lua VM to execute the thread.
> for the Lua side, this has the effect that the the last call to
> `coroutine.yield' has just returned.
> but the C side lua_yield would never return at all.
>
>
>