Re: ExecEvalVar does not have appropriate ExprState?

On 28.03.2011 10:02, Vaibhav Kaushal wrote:
> Hi,
>
> I see the ExecInitExpr says something like this:
> ------------
> case T_Var:
> state = (ExprState *) makeNode(ExprState);
> state->evalfunc = ExecEvalVar;
> ---------------
>
>
> But the ExecEvalVar function definition says:
>
> ------------------------
> Var *variable = (Var *) exprstate->expr;
> TupleTableSlot *slot;
> AttrNumber attnum;
>
> if (isDone)
> *isDone = ExprSingleResult;
>
> /* Get the input slot and attribute number we want */
> switch (variable->varno)
> {
> case INNER: /* get the tuple from the inner node */
> -------------------------
>
>
> Since ExprState -> expr in its final form would finally contain only:
>
> NodeTag type;
>
> I think that the pointer being cast in the ExecEvalVar is actually
> already a form of Var which is passed around as Expr for sake of
> function call.

Right, exprstate->expr is a Var in ExecEvalVar.

> So, was the node in the Expr tree for the corresponding ExprState node
> of the ExprState tree actually a 'Var'?

Yes.

> Also, I have tried, but failed to fidn the exact place where the Expr
> tree is created. Just pointing me to the file / function which does this
> would be of great help.

A raw expression tree is created in the grammar,
src/backend/parser/gram.y. It is then transformed in parse analysis
phase to the form the planner accepts, in transformExpr(). The planner
can do some further transformations, like replacing immutable function
calls with Consts.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com