CA-421084: xstringext: Improve performance of escaped

[cplaursen]

Updated implementation to use map_unlikely, which is optimised for this sort of thing.

[edwintorok]

Assuming that the map_unlikely function works correctly, this looks good.
If we want to be sure could add a quickcheck test with a copy of the old implementation and new implementations and have it try to find differences in behaviour.

There are examples of QCheck2 usage in XAPI already

[psafont]

Are there any mini benchmarks to see the improvement?

I have a wip branch with more cleanup around strings, especially consolidating escaping functions, which might be a worthwile thing to revise while you're on the area. Unfortunately I don't have much time to test the changes, which is why I haven't opened a PR.

master...psafont:xen-api:dev/pau/strings

The series removes map_unlikely, and changes escaped. That needs to be changed to work well with your change, of course

[cplaursen]

I'm happy to produce some benchmarks with the tests. I'll also take a look at the branch you've linked, thank you, though I'll focus on getting this change through for now.

[cplaursen]

Benchmarking against a naive implementation gives:

String size 100:
  Optimized: 1789.445 μs
  Reference: 2478.163 μs
  Improvement: 27.8% faster

String size 500:
  Optimized: 8766.232 μs
  Reference: 12855.552 μs
  Improvement: 31.8% faster

String size 1000:
  Optimized: 17661.768 μs
  Reference: 26854.626 μs
  Improvement: 34.2% faster

[cplaursen]

It would be good to have a unified benchmark markup library within xapi, rather than reproducing bechamel_simple_cli.ml whenever a benchmark is needed

[lindig]

I would expect an improvement that is not visible in the benchmark: this creates much less memory garbage and that improvement will not show up in the benchmark, I would assume.