Re: Giving Globals the CV treatment [WAS: Runtime JIT Proposals]

> At first I don't understand why you are trying to deallocate variable. > > + if (CG(active_op_array)->T == (parent->u.var - 1)) { > + CG(active_op_array)->T--; > + } > > Isn't the same variable reused as result of ZEND_FETCH_DIM? > Yeah, I hastily recanted that about three minutes after my post... It was accidentally left in from a prior attempt. You're absolutely right that it doesn't belong there. Just my sloppiness in preparing the patch for posting. > The rest of patch seems proper, but I am not sure about place. > Why we optimize @$a['b'] but not @$a->b and @$a->b()? > > I would prefer to find more general solution. > Well, we could do it in the end_variable_parse, but at that point we actually *do* have to erase an op and a temp var. Doing it at this stage prevents the extra op/tempvar from being allocated at all. Looking at do_fetch_property, I notice that we're already doing this (after a fashion) for turning $this->prop into an immediate FETCH_OBJ (setting op1 to UNUSED to indicate $this). This approach really just builds on that existing solution. I've updated the patch (adding FETCH_OBJ support and applying some concepts from do_fetch_property's implementation for $this optimization to the FETCH_DIM/OBJ CV implementations). I suppose the one alternate approach we could take would be to make a macro/inline func out of this rewriting step so that the DIM/OBJ parsers can use the same code-base. I might even suggest we look at supporting op1->IS_UNUSED for $this when used with array access on objects (but that would require an update so that the opcode handler supports it as well). -Sara