Add clap-3.1.6 as a new benchmark.

[nnethercote]

This is a little faster to compile than the old clap-rs benchmark.

        clap-rs   clap-3.1.6
check   3.17s     2.78s
debug   4.44s     4.00s
opt     5.79s     5.16s

clap-rs will be removed soon.

[rylev]

The derive feature is heavily used (having superseded the structopt crate. I wonder if including that would be useful though that might skew the benchmark heavily towards proc macro benchmarking.

[nnethercote]

The derive feature is heavily used (having superseded the structopt crate. I wonder if including that would be useful though that might skew the benchmark heavily towards proc macro benchmarking.

Interesting idea. I tried enabling the derive feature by default like so:

diff --git a/collector/benchmarks/clap-3.1.6/Cargo.toml b/collector/benchmarks/clap-3.1.6/Cargo.toml
index 6fdaa1a5..24d4d1df 100644
--- a/collector/benchmarks/clap-3.1.6/Cargo.toml
+++ b/collector/benchmarks/clap-3.1.6/Cargo.toml
@@ -413,7 +413,7 @@ default-features = false
 cargo = ["lazy_static"]
 color = ["atty", "termcolor"]
 debug = ["clap_derive/debug", "backtrace"]
-default = ["std", "color", "suggestions"]
+default = ["std", "color", "suggestions", "derive"]
 derive = ["clap_derive", "lazy_static"]
 env = []
 std = ["indexmap/std"]

It certainly affected overall compile time, due to the proc-macro2 and syn dependencies being added:

      default  w/derive
check  2.7s     7.1s
debug  4.0s     8.4s
opt    5.1s     8.9s

But for the final crate itself the change in compile times was well under 1%. I think you would see the effect of the derive feature in crates that depend on clap, rather than clap itself. So I'll stick with the original code.