Add rudimentary timing of parsing and formatting phases

[marcusklaas]

I was interested to see how much time is spent parsing vs formatting. This adds a timer to Summary to measure these things.

When run on rustfmt itself, using $ cargo run --release --bin rustfmt -- --verbose src/lib.rs, I get the following results on my machine:

Spent PT0.109887983S in the parsing phase, and PT0.414788861S in the formatting phase

In this case, formatting takes about 4x as long as parsing. This suggests that it should be worthwhile to look into trimming the fat from rustfmt.

cc #338

[nrc]

Looks good, thanks! I'd prefer if we could use std rather than add a dep though.

Interesting stats too. I'd be interested to see if that varies with the crate or whether it is fairly consistent.

So, it looks like parsing time includes IO time reading files from disk, but the formatting time does not include writing to disk. That would make the ratio look even worse for rustfmt (I would naturally expect formatting to take longer than parsing, but not by that much). It would be really interesting to record IO vs computation time, but I guess that needs a profiler, rather than just timing code.

[nrc]

Could you use std::time instead of the time crate? I think it has everything you need without the extra dep.

[marcusklaas]

Updated!

[nrc]

This still has the dependency info in the Cargo.lock, I think you need to build before committing in order to update the lock file.

[davidalber]

I have a rebased (onto current master) version of @marcusklaas's branch here that also updates Cargo.lock, in case you want to move this PR forward. You are welcome to use that, just "manually" update the original branch, or I'd be happy to open a new PR with that branch as the source.

[nrc]

rebased and merged. Thanks!