Re: Assignment statement does not evaluate all its expressions first

[Francisco Olarte <folarte@...>]

Xmilia Hermit:

On Sun, Sep 12, 2021 at 11:58 AM Xmilia Hermit <xmilia.hermit@gmail.com> wrote:

...
> node.next, node = new, new
> could also not evaluate the node from node.next before node is assigned
> to new, however, this case is explicitly checked in
> https://github.com/lua/lua/blob/9db4bfed6bb9d5828c99c0f24749eedf54d70cc2/lparser.c#L1330.
> Why even do this check in the first place if the prefix-expression parts
> do not need to be evaluated before the assignments?
> Further, from https://www.lua.org/manual/5.4/manual.html#3.4 I would
> assume that prefixexp are also expressions.

Yeah, it may weel be that, I'm not really that familiar.

I think the docs try to point you can use multiple asignementes to do
temp-less swap/rotate tricks in things like a,b=b,a or a,b,c=b,c,a,
but this levels of detail is justo too much for me.

> Thus the code
> i = 3
> i, a[i] = i+1, 20
> sets a[3] to 20, without affecting a[4] because the i in a[i] is
> evaluated (to 3) before it is assigned 4.

I suppose it does, but haven't dug too much on that. I mean, both your
examples seem to be using multi-assignement where it is not needed (
node.next=new, node=new and a[i]=20, i=i+1 seem clearer for me than
that ) and I have not found much room for confussions in real use
cases, may be because I only use multi-assign where I need it.

> It would be nice to know how this is meant to work as the my example is
> a bit contrived, however, this could happen in a more complex __newindex
> metamethod which for example calls to other hook functions.

It could be, but I think any  _newindex which makes things like
node.x, node.y=1,2 work "weirdly", as your example, should be
thoroughly documented and kept for well-trained users.

Anyway, I agree the documentation could use some work stating that
side effects may alter things. I just do not know how to suggest one.

Francisco Olarte.