[MIT-Scheme-devel] callbacks

[Matt Birkholz]

> From: Matt Birkholz <address@hidden>
> Date: Sat, 19 Mar 2005 01:13:29 +0000
> 
> [...]
> Given what I have learned this past week, the latter trick looks
> doable.  I already have a couple threads (a console REPL and Edwin)
> running, time-sliced, in an idle function of Gtk's main loop, with a
> GTk main window displaying the slice count.  A GTk-based select
> registry should complete the picture.  Wish me luck?

To integrate Scheme and the toolkit's main loop, I am running Scheme
(the Interpret function) inside a GSource dispatch function.  When
Scheme yields to the toolkit, it returns from Interpret.  When the
toolkit yields to Scheme, it calls Interpret(1) -- with the
pop_return_p arg #t.

Getting into and out of Scheme this way seems to be working well, but
I do not really understand why PRIMITIVE_CANONICALIZE_CONTEXT is
necessary -- how/why compiled code leaves the stack in a difficult
condition for the interpreter.  SO I would like some re-assurance from
the list that my technique is sound.

My technique is to introduce two new "aborts" -- dispatch codes for
abort_to_interpreter.  Here is how I have changed interp.c:

*** interp.c.~9.100.~   Thu Mar  6 05:41:19 2003
--- interp.c    Tue Mar 22 17:39:29 2005
***************
*** 437,442 ****
--- 437,452 ----
   repeat_dispatch:
    switch (dispatch_code)
      {
+ 
+     case PRIM_CONTINUE_AFTER_C:
+       PROCEED_AFTER_PRIMITIVE ();
+       return;
+ 
+     case PRIM_RETRY_AFTER_C:
+       BACK_OUT_AFTER_PRIMITIVE ();
+       Save_Cont ();
+       return;
+ 
      case PRIM_APPLY:
        PROCEED_AFTER_PRIMITIVE ();
      case CODE_MAP (PRIM_APPLY):

In my modified TEST-SELECT-* primitives (when Scheme yields to the
toolkit for a poll(2)), I do this:

  PRIMITIVE_CANONICALIZE_CONTEXT ();
  <setup/check GSource>
  PRIMITIVE_ABORT (PRIM_RETRY_AFTER_C);

Did my foot get shot?