Re: pl/pgsql feature request: shorthand for argument and local variable references

st 17. 3. 2021 v 9:20 odesílatel Michael Paquier <michael(at)paquier(dot)xyz>
napsal:

> On Wed, Mar 17, 2021 at 02:06:57PM +0800, Julien Rouhaud wrote:
> > I also think that there should be a single usable top label, otherwise
> it will
> > lead to confusing code and it can be a source of bug.
>
> Okay, that's fine by me. Could it be possible to come up with an
> approach that does not hijack the namespace list contruction and the
> lookup logic as much as it does now? I get it that the patch is done
> this way because of the ordering of operations done for the initial ns
> list creation and the grammar parsing that adds the routine label on
> top of the root one, but I'd like to believe that there are more solid
> ways to achieve that, like for example something that decouples the
> root label and its alias, or something that associates an alias
> directly with its PLpgSQL_nsitem entry?
>

I am checking it now, and I don't see any easy solution. The namespace is a
one direction tree - only backward iteration from leafs to roots is
supported. At the moment, when I try to replace the name of root ns_item,
this root ns_item is referenced by the function's arguments (NS_TYPE_VAR
type). So anytime I need some helper ns_item node, that can be a
persistent target for these nodes. In this case is a memory overhead of
just one ns_item.

orig_label <- argument1 <- argument2

The patch has to save the original orig_label because it can be referenced
from argument1 or by "FOUND" variable. New graphs looks like

new_label <- invalidated orig_label <- argument1 <- ...

This tree has a different direction than is usual, and then replacing the
root node is not simple.

Second solution (significantly more simple) is an additional pointer in
ns_item structure. In almost all cases this pointer will not be used.
Because ns_item uses a flexible array, then union cannot be used. I
implemented this in a patch marked as "alias-implementation".

What do you think about it?

Pavel

> --
> Michael
>