google / benchmark / 2201

92%

master: 92%

Pull #861

travis-ci

Perform IWYU on library sources

Was looking at time distribution clang spends on my projects
that builds benchmark in-tree, and to my surprise
benchmark_register.cc + benchmark.cc + benchmark_runner.cc +
statistics.cc + benchmark.h were near the top.

There are two (well three) solutions here - to include less
stuff in benchmark's sources, to further partition benchmark's
sources (and to not build it in-tree :S).

This performs the first solution, since it's most straight-forward
and least controversial. The diff looks scary, and indeed i will be
quite surprized if this doesn't break build on some platform.
But I don't foresee any way to avoid that risk.

Repeating the measurements show that this either helps
(- ~2s clang front-end time), or the initial measurement was noisy.

2 of 2 new or added lines in 1 file covered. (100.0%)

1780 of 1940 relevant lines covered (91.75%)

5287531.32 hits per line