Tracer tool

> I'm working on a tool that trace the excecution of the block (I need to know
> all the objects and methods touched by the block and the methodes inside these
> methods). With something like this :
>
> [ blablabla ] trace.
>
> So I want to mark the context with a flag to know if the current context is
> traced or not and in the method send_message I simply look at the flag in the
> current context and store the result or not.
>
> I've looked at the VM code and it seems that the contexts are allocated with
> the method  alloc_stack_context. So in this method I initialise the trace
> variable with 0.

> > I'm working on a tool that trace the excecution of the block (I need to
> > know all the objects and methods touched by the block and the methodes
> > inside these methods). With something like this :
> >
> > [ blablabla ] trace.
> >
> > So I want to mark the context with a flag to know if the current context
> > is traced or not and in the method send_message I simply look at the flag
> > in the current context and store the result or not.
> >
> > I've looked at the VM code and it seems that the contexts are allocated
> > with the method  alloc_stack_context. So in this method I initialise the
> > trace variable with 0.
>
> Have you checked that using DebugTools is too slow?
>
> Initialization of the new context is done in activate_new_context,
> that's probably a better place.
>
> Also, alternatively you can use the checks that are already in place for
> the profiler, so that you don't introduce more ifs on a very hpt path.
>
> Alternatively, what about instrumenting the methods and replacing them
> dynamically?  That is, rewrite something like:
>
>    self abc.
>    self def: 3.
> to
>
>    Tracer current on: self perform: #abc.
>    Tracer current on: self perform: #def: with: 3.
>
> Inside the tracer object, you keep a mapping between untraced methods
> and traced methods, and do something like:
>
>    on: receiver perform: aSymbol
>
>        | method |
>
>        "Make this a primitive if it is too slow."
>        method := receiver class lookupSelector: aSymbol.
>        method isNil ifTrue: [ ... handle doesNotUnderstand ... ].
>        self on: receiver trace: method.
>        ^receiver perform: (self tracingMethodAt: method)
>
> Right now there is no instruction stream class (it's all in STCompiler
> in the Parser package), but maybe you have already written one? :-)

> > Thanks for your answser :)
> >
> > This is strange in the activate_new_context method I've done this :
> >
> >   newContext = alloc_stack_context (size);
> >
> >   newContext->traced = 0;
>
> You definitely want FROM_INT (0) or _gst_nil_oop here, or expect crashes
> at the next GC.
>
> >   fprintf (stderr, "traced value : %d\n", newContext->traced);
> >   fprintf (stderr, "parent value : %d\n", ((gst_method_context)
> > OOP_TO_OBJ (_gst_this_context_oop))->traced);
> >
> > And if the traced value for the parent context is not nil any idea of the
> > problem ?
>
> nil is not zero, it is _gst_nil_oop (usually an hex value ending with
> 0x...800).
>
> Look at all callers of alloc_stack_context, in particular
> _gst_prepare_execution_environment.  But, adding new instance variables
> to contexts is a no-no. :-)
>
> By the way, with your GSOC application having been accepted "with very
> high probability", I would like you to publish your code somewhere as
> soon as possible.  I don't care in what state it is, don't be afraid of
> that.
>

> > Thanks for your answser :)
> >
> > This is strange in the activate_new_context method I've done this :
> >
> >   newContext = alloc_stack_context (size);
> >
> >   newContext->traced = 0;
>
> You definitely want FROM_INT (0) or _gst_nil_oop here, or expect crashes
> at the next GC.
>
> >   fprintf (stderr, "traced value : %d\n", newContext->traced);
> >   fprintf (stderr, "parent value : %d\n", ((gst_method_context)
> > OOP_TO_OBJ (_gst_this_context_oop))->traced);
> >
> > And if the traced value for the parent context is not nil any idea of the
> > problem ?
>
> nil is not zero, it is _gst_nil_oop (usually an hex value ending with
> 0x...800).
>
> Look at all callers of alloc_stack_context, in particular
> _gst_prepare_execution_environment.  But, adding new instance variables
> to contexts is a no-no. :-)
>
> By the way, with your GSOC application having been accepted "with very
> high probability", I would like you to publish your code somewhere as
> soon as possible.  I don't care in what state it is, don't be afraid of
> that.
>
> Paolo

> On Monday 20 April 2009 12:33:20 you wrote:
> > > Thanks for your answser :)
> > >
> > > This is strange in the activate_new_context method I've done this :
> > >
> > >   newContext = alloc_stack_context (size);
> > >
> > >   newContext->traced = 0;
> >
> > You definitely want FROM_INT (0) or _gst_nil_oop here, or expect crashes
> > at the next GC.
> >
> > >   fprintf (stderr, "traced value : %d\n", newContext->traced);
> > >   fprintf (stderr, "parent value : %d\n", ((gst_method_context)
> > > OOP_TO_OBJ (_gst_this_context_oop))->traced);
> > >
> > > And if the traced value for the parent context is not nil any idea of
> > > the problem ?
> >
> > nil is not zero, it is _gst_nil_oop (usually an hex value ending with
> > 0x...800).
> >
> > Look at all callers of alloc_stack_context, in particular
> > _gst_prepare_execution_environment.  But, adding new instance variables
> > to contexts is a no-no. :-)
> >
> > By the way, with your GSOC application having been accepted "with very
> > high probability", I would like you to publish your code somewhere as
> > soon as possible.  I don't care in what state it is, don't be afraid of
> > that.
> >
> > Paolo
>
> Is it possible to send a message to an object inside the
> methodsend_block_value ?