Make the static arrays of style-affecting attributes thread-local.

[pcwalton]

... instead of regenerating and dropping them on the stack over and over.

Reduces I-cache footprint and memory barrier traffic on ARM.

r? @SimonSapin

[SimonSapin]

Could they be completely static, with https://crates.io/crates/lazy_static , rather than thread-local?

[Ms2ger]

I've never seen any proof that this code is actually necessary.

[pcwalton]

@SimonSapin I looked into that, but lazy_static causes a full memory barrier (dmb ish) on ARM each time the static is accessed, which is too expensive.

[Manishearth]

Can't we make atom! use a constfn?

[pcwalton]

Do we have const fn in Servo's Rust?

[Manishearth]

Not yet. I can rustup though. There's some major plugins renaming but not much more as far as I can tell.

(Having done the feature audit I sort of can gauge rustup difficulty now :) )

[michaelwu]

I thought we got it in the last rustup. At the very least, the last rustup had support for const functions in libsyntax asts.

[Manishearth]

Oh, right. We should have it then.

[Ms2ger]

@pcwalton ping

[metajack]

pinging @pcwalton again

[nox]

atom!() is const now, so we can just use static.

[SimonSapin]

Hasn’t atom!() always been const, expanding to an struct literal with an integer literal ?

[SimonSapin]

Ah, here is why they’re not const:

src/matching.rs:539:15: 539:30 error: statics are not allowed to have destructors [E0493]
src/matching.rs:539         atom: atom!("hidden"),
                                  ^~~~~~~~~~~~~~~
src/matching.rs:539:15: 539:30 note: in this expansion of atom! (defined in <string_cache macros>)

… which is not gonna change as long as we use reference counting.

[SimonSapin]

… unless the language-level restriction is lifted: rust-lang/rfcs#1111