optimized copyuntil

[stevengj]

Speeds up copyuntil (and hence readuntil, readline, etcetera) in Julia 1.11 (via JuliaLang/julia#48273).

(I also verified locally that the new readlines tests pass with Julia master, and that the test indeed exercises the new copyuntil method.)

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.40% 🎉

Comparison is base (9602735) 87.32% compared to head (66c7cf2) 87.72%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #76      +/-   ##
==========================================
+ Coverage   87.32%   87.72%   +0.40%     
==========================================
  Files           5        5              
  Lines         426      440      +14     
==========================================
+ Hits          372      386      +14     
  Misses         54       54              

Files Changed	Coverage Δ
src/bufferedinputstream.jl	97.68% <100.00%> (+0.13%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[KristofferC]

For posterity, what benchmark did you use and what were the results?

[stevengj]

Using the benchmark from JuliaLang/julia#48273 (comment) (with readuntil! updated to copyuntil), I got

PRIOR TO THIS PR:

• sum(length, eachline(seekstart(bio), keep=true)) (allocating a String per line): 3.363 ms (31055 allocations: 2.44 MiB)
• doit!((seekstart(bio); bio), buf, '\n') (re-using a StringView): 2.874 ms (0 allocations: 0 bytes)

AFTER THIS PR:

• sum(length, eachline(seekstart(bio), keep=true)) (allocating a String per line): 2.048 ms (31052 allocations: 2.44 MiB)
• doit!((seekstart(bio); bio), buf, '\n') (re-using a StringView): 1.666 ms (0 allocations: 0 bytes)

which is a nice improvement but not overwhelming since the fallback method of character-by-character reading of a BufferedInputStream is not too terrible.

[KristofferC]

Side note, it doesn't feel like the tests of this package should have to take this long.