Re: [Notion-devel] Object inheritance

[Etan R. <de...@un...>]

I think renaming to "super" is a pointless idea. Either you know what type
the parent is, in which case typing its name instead of super isn't a big
deal; or you don't, in which case you can't use super anyway since you
can't know what to cast it to.

It also doesn't tell you anything useful, since the immediate super
relationship isn't particularly helpful whereas the type of the member
actually is (especially if you are simply passing it to another function
which expects a certain type).

Consider:
    WScreen *scr=current_screen();
    mplex_switch_next(&(scr->mplex));
versus
    WScreen *scr=current_screen();
    mplex_switch_next(&(scr->super));

Which one tells you immediately that you've gotten the argument type to
mplex_switch_next right and which one leaves you having to remember and or
guess?

Also consider your given example. In what way is
> 'rw->scr.super.super.win=None;'
superior to
> 'rw->scr.mplex.win.win=None;'
?

The super version reads less well (to me), gives me less information, and
at that point I would really want a cast (or cast macro) or something to
hide it (you probably want that in either case really).

What is the current inconsistency exactly? I think there are actually
valid reasons to use both the specific cast and the &(obj->mem)
indirection.

To my mind the cast is more appropriate for assignment to local variables
for continued use (it says "I know this X is a Y and I intend to use it as
such from now on") whereas the member indirection is useful for examples
like the above (it says "I have an X, which I know is a Y, and you need
just that bit of it").

I'm not certain we need both, mind you. I'd be fine standardizing on the
cast (I wouldn't standardize the other way though). I'd also be fine
adding typecast macros (the way glib does), like what ebik said (though
I'm not sure we can reasonably get runtime tests the way glib sometimes
does).

    -Etan