Smalltalk › Gnu

[PATCH] Process>>#isSuspendedInCCall

_‹ Previous Topic Next Topic _›

Classic

List

Threaded

4 messages Options

Gwenaël Casaccio

[PATCH] Process>>#isSuspendedInCCall

Hi,

I've added Process>>#isSuspendedInCCall useful when you try
to make the difference between a suspended process and a process
to make a c-call which is reentrant.

Cheers,
Gwen

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

0002-Add-Process-isSuspendedInCCall.patch (6K) Download Attachment

Gwenaël Casaccio

Re: [PATCH] Process>>#isSuspendedInCCall

On 12/04/2014 11:15, Gwenaël Casaccio wrote:
> Hi,
>
> I've added Process>>#isSuspendedInCCall useful when you try
> to make the difference between a suspended process and a process
> to make a c-call which is reentrant.
>
> Cheers,
> Gwen
>

Don't apply it I've a better implementation not yet finished but here
are the big lines:
In the primitive I get the context (if it's a block I get the method
context in the outer context)
and I set or unset a bit in the flag.

Now I'm not sure the it's not the current context that should be tagged
and not the outer context
but that requires more changes in the vm.

Cheers,
Gwen

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

Gwenaël Casaccio

Re: [PATCH] Process>>#isSuspendedInCCall

On 14/04/2014 20:17, Gwenaël Casaccio wrote:

> On 12/04/2014 11:15, Gwenaël Casaccio wrote:
>> Hi,
>>
>> I've added Process>>#isSuspendedInCCall useful when you try
>> to make the difference between a suspended process and a process
>> to make a c-call which is reentrant.
>>
>> Cheers,
>> Gwen
>>
>
> Don't apply it I've a better implementation not yet finished but here
> are the big lines:
> In the primitive I get the context (if it's a block I get the method
> context in the outer context)
> and I set or unset a bit in the flag.
>
> Now I'm not sure the it's not the current context that should be
> tagged and not the outer context
> but that requires more changes in the vm.
>
> Cheers,
> Gwen
>

Here is the patch: I've add one slot to ContextPart (ctxtFlags) and
thisContext is tagged while
being reentrant. And few primitives (compile, ccall) are extended to tag
the context.

Gwen

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk

0002-Add-Process-isSuspendedInCCall.patch (15K) Download Attachment

Holger Freyther

Re: [PATCH] Process>>#isSuspendedInCCall

On Tue, Apr 15, 2014 at 05:47:49PM +0200, Gwenaël Casaccio wrote:

Hi,

sorry for the delay in reviewing.

> Here is the patch: I've add one slot to ContextPart (ctxtFlags) and
> thisContext is tagged while
> being reentrant. And few primitives (compile, ccall) are extended to tag the
> context.

The C-Call/primitive will only be invoked from a method context
in normal operation. What about just adding the ctxFlags to the
methodContext then?

> asyncCall [
> + <category: 'calling'>
> +
> "Perform the call-out for the function represented by the receiver.
> The arguments (and the receiver if one of the arguments has type
> #self or #selfSmalltalk) are taken from the parent context.
> @@ -140,7 +142,7 @@ to perform the actual call-out to C routines.'>
> ^self isValid
> ifFalse:
> [SystemExceptions.CInterfaceError signal: 'Invalid C call-out ' , self name]
> - ifTrue: [self asyncCallNoRetryFrom: thisContext parentContext]
> + ifTrue: [self asyncCallNoRetryFrom: thisContext parentContext ]
> ]

unrelated change just creating distraction. :)

> @@ -522,6 +525,7 @@ _gst_interpret (OOP processOOP)
>
> monitor_byte_codes:
> SET_EXCEPT_FLAG (false);
> + finish_debugging_step = false;
>
> /* First, deal with any async signals. */
> if (async_queue_enabled)
> @@ -540,6 +544,7 @@ monitor_byte_codes:
> {
> _gst_async_signal (single_step_semaphore);
> single_step_semaphore = NULL;
> + finish_debugging_step = true;
> }

are you mixing things here? The 'tagging' of a process being
debugged might be good in a separate patch? We will also need
to add this to the JIT.

> +/* Anwser whether execution is going to be reentrant. */
> +#define MCF_IS_REENTRANT 2

What does this mean?

> + this_context = (gst_method_context) OOP_TO_OBJ (_gst_this_context_oop);
> + this_context->ctxtFlags |= MCF_IS_REENTRANT;
> +
> oop2 = POP_OOP ();
> oop1 = POP_OOP ();
> if (IS_CLASS (oop2, _gst_string_class))
> @@ -5231,6 +5235,8 @@ primitive VMpr_Behavior_primCompile [succeed]
>
> if (interrupted)
> stop_execution ();
> + else
> + this_context->ctxtFlags &= ~MCF_IS_REENTRANT;

this and most of the errors look unbalanced. So if the function
is re-entered the first one to leave will unset this flag? Is
this the intention? What does it buy us?

_______________________________________________
help-smalltalk mailing list
[hidden email]
https://lists.gnu.org/mailman/listinfo/help-smalltalk