Hi,
I’m trying to build a kind of pre-emptive
scheduler around Lua VM 5.2.
The idea is to execute different lua_State
in “parallel’.
To give the illusion of parallelism, the
idea is to run a thread for a given number of Lua VM
instructions and then run the next thread in a list for this
number of instructions and so on…
From the LUA regular API, when starting the
execution of a thread, it is not possible to control the
number of Lua VM instructions to execute.
But from the LUA debug API, it is possible
to define a ‘hook’ to execute every COUNT of instructions.
The first idea was to install such hook and
perform the round-robin scheduling from the hook. The
lua_State of active thread seems to be saved before running
such hook.
But I didn’t found any means to overwrite
the thread to be resumed from the hook.
Then the idea was to ‘yield’ from the hook,
back to the round robin scheduler.
This seems to work in the simplest cases.
But I’ve noticed some strange “intereferences” when thread use
coroutines on its own.
The Lua resume API has a ‘from’ parameter:
int lua_resume (lua_State *L, lua_State *from, int nargs);
But from the description, I didn’t catch
the how this param has to be used.
I’ve written a small “C” main program to
illustrate the concept and the “interferences”…
In the program here below, the first
iteration of the loop works as expected.
Then the L[0] and its coroutine are
resumed, but the hook is reached faster than for the first
iteration and the L[1] is rescheduled.
After L[0] will be never rescheduled, only
L[1].
Thanks in advance, Dominique Torette.
