Davisp python-spidermonkey

Comments and changes to this ticket

• You flagged this item as spam.
  

  qvx 3000 November 9th, 2009 @ 10:00 AM

  I see that you are using Context->root as this scope object. How about adding enter_scope/leave_scope which manipulate the root scope object.

  Enter would create new object, set root as its parent and then replace the root with itself (all in one step):
  

    Context_enter_scope(){
  
  
  
  
  
  
  

  self->root = JS_NewObject(self->cx, &js_global_class, NULL, self->root);
  

  
  
  
  
    }
  

  Leave would do the opposite:
  

    Context_leave_scope(){
  
  
  
  
  
  
  

  self->root = JS_GetParent(self->cx, self->root);
  

  
  
  
  
    }
  

  I don't know if anything has to be destroyed/unallocated here.
• 

  Paul J. Davis November 9th, 2009 @ 01:21 PM

  I tried this at one point but I can't find the code for it. The bottom line though is that even though it looks like this should work, it's not quite 100% correct. There are weird interactions with how variable declarations work and which object they get attached to.

  You might look into the V8 bindings. The V8 documentation I read seemed to target this specific use case quite pointedly. I haven't the slightest if that functionality is exposed in the bindings or not though.

• You flagged this item as spam.
  

  qvx 3000 November 9th, 2009 @ 04:26 PM

  I'v done as I proposed above and the simple tests are working correctly. What did you have in mind when you said that it is not 100% percent correct?

  tests:

  @t.cx()
  def test_scope(cx):
      t.eq(cx.execute("var g=1; var x = 4; x * x;"), 16)
      t.eq(cx.execute("g"), 1)
      t.eq(cx.execute("x"), 4)
      t.eq(cx.enter_scope(), None)
      t.eq(cx.execute("var x = 5; x * x;"), 25)
      t.eq(cx.execute("g"), 1)
      t.eq(cx.execute("x"), 5)
      t.eq(cx.leave_scope(), None)
      t.eq(cx.execute("g"), 1)
  t.eq(cx.execute("x"), 4)
  t.raises(Exception, cx.leave_scope)
  

  P.S. comment preview would be a nice feature to have.

• You flagged this item as spam.
  

  qvx 3000 November 9th, 2009 @ 04:33 PM

  And this is the implementation:

  PyObject*
  Context_enter_scope(Context* self, PyObject* args, PyObject* kwargs)
  {
      PyObject* ret = NULL;
      JSObject* root = NULL;
  
      JS_BeginRequest(self->cx);
  
      root = JS_NewObject(self->cx, &js_global_class, NULL, self->root);
      if(root == NULL)
      {
          PyErr_SetString(PyExc_RuntimeError, "Error creating root object while entering scope.");
          goto error;
      }
      self->root = root;
  
      JS_EndRequest(self->cx);
      ret = Py_None;
      Py_INCREF(ret);
      goto success;
  
  error:
      JS_EndRequest(self->cx);
  success:
      return ret;
  }
  
  PyObject*
  Context_leave_scope(Context* self, PyObject* args, PyObject* kwargs)
  {
      PyObject* ret = NULL;
      JSObject* root = NULL;
  
      JS_BeginRequest(self->cx);
  
      root = JS_GetParent(self->cx, self->root);
      if(root == NULL)
      {
          PyErr_SetString(PyExc_RuntimeError, "Cannot leave scope if it was not entered.");
          goto error;
      }
      self->root = root;
  
      JS_EndRequest(self->cx);
      ret = Py_None;
      Py_INCREF(ret);
      goto success;
  
  error:
      JS_EndRequest(self->cx);
  success:
      return ret;
  }
  

• 

  Paul J. Davis November 9th, 2009 @ 04:55 PM

  Oh, I think the reason is that you're creating a blank scope where as I was doing a leaky thing.

  What happens if you do something like:

  @t.cx()
  def test_scope(cx):
      t.eq(cx.execute("var x = 4; x;"), 4)
      t.eq(cx.enter_scope(), None)
      t.eq(cx.execute("x;"), 4)
  

  I swear I had a ticket around here that had some code that another guy wrote to walk the parent chain to get the different bits.

  The 100% correct thing is that with the other patch that I had working, variable assignments were leaking outside of the nested scopes. It was finicky on how you used var instantiation and which global got written to etc etc. I ended up dropping it because of that weirdness.

• 

  Paul J. Davis November 9th, 2009 @ 04:58 PM

  Oh found it, it's an email thread I had with some guy. If you email me at paul.joseph.davis@gmail.com I'll forward you the conversation and what my thinking was that time so you can comment.

• You flagged this item as spam.
  

  qvx 3000 November 9th, 2009 @ 05:17 PM

  I see! You are using custom "global class" implementation to track global objects. It is an error to create another global when entering a new scope. To fix this I have to say:

  root = JS_NewObject(self->cx, NULL, NULL, self->root);
  

  There are other changes, like introducing root0 to Context struct to keep track of original root (the one that is used for global object tracking). More details in scope.patch file.

  New test:

  @t.rt()
  def test_scope(rt):
      glbl = {}
      cx = rt.new_context(glbl=glbl)
      t.eq(cx.execute("var g=1; var x = 4; x * x;"), 16)
      t.eq(cx.add_global('a', 1), None)
      t.eq(cx.execute("g"), 1)
      t.eq(cx.execute("x"), 4)
      t.eq(glbl, {'g':1, 'a':1, 'x':4})
      t.eq(cx.enter_scope(), None)
      t.eq(cx.execute("var dummy=0; var x = 5; x * x;"), 25)
      t.eq(cx.execute("g"), 1)
      t.eq(cx.execute("x"), 5)
      t.eq(cx.add_global('A', 1), None)
      t.eq(glbl, {'g':1, 'a':1, 'A':1, 'x':4})
      t.eq(cx.leave_scope(), None)
      t.eq(cx.execute("a"), 1)
      t.eq(cx.execute("A"), 1)
      t.eq(cx.execute("g"), 1)
      t.eq(cx.execute("x"), 4)
      t.eq(glbl, {'g':1, 'a':1, 'A':1, 'x':4})
      t.raises(Exception, cx.leave_scope)
  

  • scopepatch.txt 4 KB

• You flagged this item as spam.
  

  qvx 3000 November 9th, 2009 @ 11:14 PM

  More recent versions.

  • scope3patch.txt 8.4 KB
  • test-scope.py 2.1 KB