Performance on strings

[simsurace]

Somewhere along the way performance on strings must have taken a hit:

julia> data = rand(UInt8, 10^6);

julia> str = String(data);

julia> @time md5(data);
  0.000047 seconds (6 allocations: 288 bytes)

julia> @time md5(str);
  2.112903 seconds (6 allocations: 288 bytes)

julia> versioninfo()
Julia Version 1.11.6
Commit 9615af0f269 (2025-07-09 12:58 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin24.0.0)
  CPU: 12 × Apple M3 Pro
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, apple-m2)
Threads: 4 default, 0 interactive, 2 GC (on 6 virtual cores)
Environment:
  JULIA_CONDAPKG_BACKEND = Null
  JULIA_EDITOR = code
  JULIA_VSCODE_REPL = 1
  JULIA_NUM_THREADS = 4

Is there an easy fix? Even md5(collect(codeunits(str))) would be way better than this:

julia> @time md5(collect(codeunits(str)));
  0.002464 seconds (10 allocations: 992.359 KiB)

[oxinabox]

Wild.
Must be something that was changed in codeunits
Or perhaps that's always been slow?

the code is nothing more: https://github.com/JuliaCrypto/MD5.jl/blob/master/src/MD5.jl#L25C47-L26C1

[simsurace]

I suspect that the reason for the slow-down might be something further down in the call chain that makes codeunits behave very differently that a plain Vector{UInt8} even though CodeUnits <: AbstractVector{UInt8}.

In my application I can afford to put the string into an IOBuffer since I don't need it anymore afterwards. That's also fast, but not a general solution.