Call for experimentation: branch zack/no-inline-string-opts

[Zack Weinberg]

For a long time people have been arguing over whether the inline
string optimizations in glibc's headers (bits/string.h,
bits/string2.h) are still valuable.  It's hard to test, because
removing even just one has unpredictable knock-on effects, up to and
including breaking the build.

I've put together a branch (available right now:
zack/no-inline-string-opts) in which both bits/string.h and
bits/string2.h are completely removed.  I am not proposing that we
should just land this branch right now; I did it to make it easier for
people to benchmark real applications with and without the string
optimizations, which should help us pin down which of the
optimizations are actually valuable.

(I don't plan on doing any such benchmarking myself.  I have just
about exhausted my hacking-not-related-to-my-actual-job budget for the
quarter already.)

zw