On 06.10.2017 00:06, Jesper Steen Møller wrote:
Hi List
On 5 Oct 2017, at 20.01, [email protected]
<mailto:[email protected]> wrote:
[...]
`a??.b.c.` is just `a?.b?.c`
Are there any cases where a?.b.c would every make sense -- if a is null,
then the expression would throw anyway, wouldn't it? Or is there some
dynamic mixin case I'm not taking into consideration?
In other words, IMHO a?.b.c should implicitly mean a?.b?.c or at least
give a warning that the construct doesn't make sense. In either case ??.
wouldn't make much sense.
I personally think we made a mistake with ?. Instead of safe navigation,
we should stop evaluation
x = a?.b?.c?.d
is translated as:
var tmp
tmp = a==null?null: a.b
tmp = tmp==null?null: tmp.c
tmp = tmp==null?null: tmp.d
x = tmp
and x = a?.b.c.d is
var tmp
tmp = a==null?null: a.b
x = tmp.c.d
while x = a??.b.c.d is
var tmp
tmp = a==null?null: a.b.c.d
x = tmp
As you can see the constructs a??.b.c.d and a?.b?.c?.d can have
different implementations.
I would favor replacing ?. with what ??. does. I doubt it would really
break any programs.
As for the other constructs, I'm also not a big fan of introducing lots
of new syntax just because we can.
There is a collateral cost to pay, e.g. in the type checker and static
compiler (within the Groovy project's control), too, and in every AST
transformer which needs to know about the new stuff (out of the
Groovyproject's control!) .
well, that is where the current version shines... They are parse time
constructs, no new AST required, no new compilation required (only for
optimization), no new checks required.
But I would go further... just because the collateral cost is low, does
not make the construct more appealing. For syntax, semantics and
use-case have to decide.
bye Jochen
