inferred literals and integer-const expressions aren't overflow checked

[graydon]

The most egregious example is: let x:u8 = 10_000; which compiles without complaint. But more generally, if we're going to make a promise to constant-fold the integer-constant grammar, we might well consider overflow checks on all those arithmetic nodes. Opinions?

[graydon]

cc @lkuper

[nikomatsakis]

My suggestion for how to deal with this:

Leave type check as it is. Instead, insert a lint check that examines all literals (or perhaps constant expressions) and checks whether they fit into the type that was inferred for them.

[sophiebits]

(Mostly a dupe of #3084.)

[x--]

I have written this patch x--@e9aab10 that basically add checks for assign and assign_op expressions.
@graydon recommended that these checks would be more useful in middle::const_eval so, gonna try to put the checks there.

[nikomatsakis]

What I had envisioned was similar to your patch @x-- except that I thought we'd just visit integer literals and check the type that was assigned to them, rather than putting the checks on assignments. However, @graydon's suggestion sounds like a superset of this check.

[nikomatsakis]

Not critical for 0.6; de-milestoning

[catamorphism]

Nominating for milestone 2 (backwards-compatible).

[brson]

cc #5551

[graydon]

sub-bug of #5551

[graydon]

accepted for well-defined milestone