This makes sense. What I really meant was: Let's not try to do this
checking during the type check itself, as we initially did, but rather
as a later lint step. This also allows you to disable it if you know
what you're doing and for some reason the code is cleaner as you wrote it.
Niko
Graydon Hoare wrote:
On 12-12-20 05:59 AM, Niko Matsakis wrote:
Lindsey Kuper wrote:
Our thinking at the time we implemented suffix inference was that if
you went so far as to write the optional type annotation, then you
should get the type you asked for, no questions asked, overflows and
all. After all, sometimes people do want intentional overflow. But
if it's causing nasty surprises, then, yeah, it should be revisited.
I think the best thing would be a simple lint check that visits each
literal and checks whether it can fit into the type assigned to it. It
will catch many simple cases like this one.
It is certainly the simplest thing for this case, but I'd also like not
to paper it over, but to address the bug its general form. The question
of exactly what constant-folding to do in the front/middle ends, and
what sorts of conditions during that count as errors, is worth nailing
down; it will resurface (already has, I've seen several "surprise"
constant / non-constant behaviors, and heard others stubbing toes on same).
It's already partly-implemented, and I think there's a plan coming into
focus, just needs more care / thought.
-Graydon
_______________________________________________
Rust-dev mailing list
Rust-dev@mozilla.org
https://mail.mozilla.org/listinfo/rust-dev
