vector-sorting-benchmarks VS text

I recently tried to implement quicksort on vectors in Haskell aiming to make it fast enough to be comparable to C++. If you're curious you can check out https://github.com/sergv/vector-sorting-benchmarks. After some time and research on what C++ actually does I reimplemented it in Haskell and the result managed to stay within reasonable % of C++. E.g. sorting 1000 arrays of 20000 8-byte integers on my machine takes 523ms in C++ (with ffi overhead) and 545ms in fastest Haskell algorithm with closest others taking around 650ms.

String is a linked list of UTF-32 codepoints, which is just as bad as it sounds. Haskell programmers usually use Text from the text package instead.

This is the first release that was executed via the new GitHub CI. Everything went smoothly, except for an unexpected packaging bug on windows due to text-2.0 linking against libstdc++ by default.

In addition to what's mentioned, the "default" libraries people use are often not the best-performing ones. E.g. Data.Vector.Hashtables is often much faster than Data.HashMap.Strict (which again is typically faster than Data.Map). And we find performance papercuts in common libraries that may simply be due to not enough people optimising for speed.

Used by nearly every Haskell project: https://hackage.haskell.org/package/text 60% of the modules are ".Internal".

text

Here is a representative patch for text: https://github.com/haskell/text/pull/365/commits/37a2157245457a287d638bbceb472fe93b71f224 Hopefully it clarifies why achieving C performance is problematic even if you have enough primitive operations.

I want to run an app in a Docker container, but detached. This causes getLine to throw an EOF exception, since the terminal detaches. That is, if I understand this thread correctly https://github.com/haskell/text/issues/258.