Avoid extra allocations when checking for namespace separators

[ghost]

Description

Hi, I am trying to optimize YARD a bit to speed up documentation runs on a large codebase.

The codebase is private, so I am using https://github.com/watir/watir/ as a smaller stand-in to report benchmarking numbers from https://github.com/SamSaffron/memory_profiler.

Test set-up

• Ruby 2.7.6 (installed via rvm on Linux)
• Watir version: 293bed2b57c
• MemoryProfiler version: 1.0.0
• YARD baseline: 0a550939f9b
• Test command:
  • ruby-memory-profiler --no-color --retained-strings=500 --allocated-strings=500 --max=300 -o prof.log ./run-yard.rb stats
  • run-yard.rb is a smaller version of /bin/yard (MemoryProfiler was not able to run /bin/yard)
  • I am using the stats command to avoid memory exhaustion

Performance data

• Before
  • Total allocated: 187901758 bytes (1987572 objects)
  • Total retained: 4052113 bytes (53463 objects)
• After
  • Total allocated: 180313268 bytes (1942822 objects)
  • Total retained: 4052153 bytes (53464 objects)
• Savings
  • Total allocated: 7588490 bytes (44750 objects)
  • Total retained: -40 bytes (-1 object)

Completed Tasks

• [x] I have read the Contributing Guide.
• [x] The pull request is complete (implemented / written).
• [x] Git commits have been cleaned up (squash WIP / revert commits).
• [x] I wrote tests and ran bundle exec rake locally (if code is attached to PR).

[ghost]

This feels like a brittle change. NSEP is not guaranteed to be 2 characters, and thus, any changes away from a Regex would require a more generalized string comparison.

Even though this might seem like the :: separation is fairly baked into the Ruby language, YARD does allow for extensions to modify the source types being parsed (yard-js is one example), and this would break those plugins.

It's a pretty minor change (suggestions already provided) but I'm wondering if this significantly affects the performance improvements?

Thanks for taking a look. I can re-test this week, but I just wanted to note that existing code seems to assume that NSEP is two characters:

https://github.com/lsegal/yard/blob/8e82a8ab27f2e711bf555bb6ba0c037a1eb3ccfb/lib/yard/code_objects/proxy.rb#L37-L40

https://github.com/lsegal/yard/blob/15c8b888c52a9bcad498c4281004969fda6f87eb/lib/yard/code_objects/base.rb#L199-L202

Do these spots need to use NSEP.size as well?

[ghost]

I will not have time to work on this branch in the short term