New by-need model turned an assertion wrong

[Raphael Collet]

Raphael Collet wrote:
> 
> The assertion was that a space cannot wake up a suspension outside it.
> This assertion is incorrect with the new by-need model.  A space must be
> able to make a global variable needed, otherwise you simply get stuck
> when loading the FD module, for instance.
> 
> The fix was simply to remove the assertion.

Well, that fix was a bug :P
This Ozcar bug has now be fixed properly.

Technically, the debug stream in AM has been replaced by a port.  This
allows a debug message to be sent from inside a subspace, provided the
message does not contain space-sited data.  The message never contains
sited data, because only toplevel threads are traced by Ozcar.  You can
test the fix with the following example:

declare X
thread {WaitNeeded X} X=1 end
S = {Space.new fun {$} X+1 end}
{Wait {Space.ask S}}
{Show {Space.merge S}}

Launch Ozcar, then feed the lines above.  The following scenario shows
that Ozcar correctly handles interactions between the toplevel space and
a subspace:
 - Let the thread at line 2 block on X before creating the space.
 - After the space is created, you see the thread at line 2 become
runnable as expected.
 - Continue the main thread until it blocks on Space.ask.
 - Let the "WaitNeeded" thread terminate.  This unblocks the thread
inside the space, which makes the space stable.  You can see the main
thread unblock.
 - Let the main thread terminate.


Cheers,

raph