Avoid translation of unreachable arms in lazy binops

[dotdash]

No description provided.

[rust-highfive]

r? @brson

(rust_highfive has picked a reviewer for you, use r? to override)

[brson]

Can you add a one-line comment explaining what this early return is doing?

[brson]

I presume the idea here is to generate less IR since LLVM will figure this out on its own. Do you have a sense of how often this occurs in real code and what kind of savings it creates?

[dotdash]

I only noticed this when looking at the unoptimized IR for libsyntax, looking for br i1 (false|true). A few dozen of those were from with_cond calls (fixed in an earlier PR), and a few hundred from from lazy binops. I don't know how representative that is or if it saves much in terms of compile time, but the places I looked at were rather hard to read due to the extra branches and basic blocks. The fact that some of them were inside matches didn't exactly help ;-)

[dotdash]

Looking through the IR for libsyntax again, most of these seem to originate from derived PartialEq implementations. Maybe "fixing" those is preferable?

[dotdash]

Compile times for libsyntax are unchanged. I'd be fine with dropping this and maybe fixing the deriving code at some point. In the meantime, I can easily locally apply this patch when I feel the need.

[dotdash]

Not worth the added complexity