Special case quote!/quote_spanned! for 1 and 2 tts.

[nnethercote]

These make crates that use quote compile faster.

I also tried specializing for 3 tts, but the rules are more complex and
the additional perf wins are much smaller than they are for 1 and 2 tts.

[nnethercote]

Some instruction counts results for non-incremental cargo check builds on some crates that use quote heavily.

Benchmark	Profile	Scenario	% Change	Significance Factor?
ctor-0.1.21	check	full	-12.39%	61.93x
num-derive-0.3.3	check	full	-11.05%	55.26x
pest_generator-2.1.3	check	full	-11.02%	55.11x
enum-as-inner-0.4.0	check	full	-9.73%	48.67x
mockall_derive-0.11.0	check	full	-6.63%	33.17x
futures-macro-0.3.19	check	full	-5.82%	29.12x
wasm-bindgen-backend-0.2.79	check	full	-4.44%	22.20x
wayland-scanner-0.29.4	check	full	-3.75%	18.75x
pyo3-macros-backend-0.16.3	check	full	-2.37%	11.83x
tonic-build-0.7.0	check	full	-2.34%	11.68x
clap_derive-3.1.7	check	full	-2.16%	10.78x
prost-derive-0.10.0	check	full	-2.16%	10.78x
ref-cast-impl-1.0.6	check	full	-1.95%	9.74x
diesel_derives-1.4.1	check	full	-1.84%	9.20x
stdweb-derive-0.5.3	check	full	-1.58%	7.88x
vsdb_derive-0.21.1	check	full	-1.01%	5.05x
enumflags2_derive-0.7.4	check	full	-0.08%	0.40x

[dtolnay]

Thanks, this looks promising.

The failing ui test should be fixed if you rebase past 31c3be4. Rustc's diagnostics sometimes change.

Edit: nope, that isn't it. This PR is causing the diagnostic change. You can run TRYBUILD=overwrite cargo +nightly test --test compiletest to update the test case.

[nnethercote]

CI gets one test error, but I get two locally -- the second one in tests/ui/not-repeatable.stderr. Not sure why. I have pushed a new version that updates the expected output just for tests/ui/not-quotable.rs. Hopefully that'll be enough to satisfy CI.

[nnethercote]

Whoops, no, I pushed a version with both test changes. This time it should just be the one change.

[dtolnay]

Thanks!