Define getindex(::InlineString, range) to return InlineString

[nickrobinson251]

This is another case where we could be keeping things as InlineString rather than returning a SubString{InlineString} (like chop, chomp, strip etc) (xref #5)

...but i don't love the (worse than linear) slowdown i'm seeing for larger InlineString types

I guess it makes sense since we're doing more work to construct a new InlineString (two nextind calls and then several bit-shifts) than the work involved in creating a SubString (which is basically a single nextind call)?

Micro-benchmark

      for S in (
           InlineString3, InlineString7, InlineString15, InlineString31,
           InlineString63, InlineString127, InlineString255,
       )
           str = S(repeat("x", sizeof(S)-1))
           ix = (firstindex(str) + 1):(lastindex(str) - 1)
           @btime $str[$ix]
       end

On main (returning SubString)

  14.194 ns (0 allocations: 0 bytes)
  11.720 ns (0 allocations: 0 bytes)
  16.658 ns (0 allocations: 0 bytes)
  18.681 ns (0 allocations: 0 bytes)
  29.732 ns (0 allocations: 0 bytes)
  57.562 ns (0 allocations: 0 bytes)
  109.481 ns (0 allocations: 0 bytes)

On this branch (returning InlineString)

  10.218 ns (0 allocations: 0 bytes)
  12.930 ns (0 allocations: 0 bytes)
  14.570 ns (0 allocations: 0 bytes)
  29.313 ns (0 allocations: 0 bytes)
  98.969 ns (0 allocations: 0 bytes)
  328.744 ns (0 allocations: 0 bytes)
  1.283 μs (0 allocations: 0 bytes)

---

There's also getindex(str, vec) that allows selecting non-contiguous elements... so maybe if we add this we should add that too? (currently returns SubString)

[quinnj]

This looks great; thanks @nickrobinson251! The test failure is just Julia 1.3 being really old and having some other ambiguous range error. I'm happy to bump up to Julia 1.6 and we can do a new minor release.

[nickrobinson251]

oooft, looks like v1.6 has the same ambiguity error as 1.3 😞 I think i'd rather us support 1.6 if we can, so i'll probably just oush a fix