Potential bug in u64_array_remove_duplicates

[frwdrik]

I think there's a bug in u64_array_remove_duplicates. When input is an array of size N > 0 consisting of a single value x, repeated N times, it doesn't return what I expect: Instead of returning an array { x } of count one, it returns an empty array.

Small reproducible test case:

int main() {
  Arena *arena = arena_alloc();

  U64Array arr = {};
  arr.count = 2;
  arr.v = push_array_no_zero(arena, U64, arr.count);
  arr.v[0] = 1;
  arr.v[1] = 1;
  
  U64Array uniq = u64_array_remove_duplicates(arena, arr);
  AssertAlways(uniq.count == 1); // This (surprisingly) traps

  return 0;
}

Compiling with GCC 14.2.1 and running on Linux 6.12.4, this causes a trap from the AssertAlways. The expected behaviour would be that the assert passes, like in this modified version:

int main() {
  Arena *arena = arena_alloc();

  U64Array arr = {};
  arr.count = 3;
  arr.v = push_array_no_zero(arena, U64, arr.count);
  arr.v[0] = 1;
  arr.v[1] = 1;
  arr.v[2] = 2;
  
  U64Array uniq = u64_array_remove_duplicates(arena, arr);
  AssertAlways(uniq.count == 2); // This doesn't trap
  return 0;
}

Note that by adding a different element 2, the assert now passes. In this case the different element was inserted at the end of the array, but that is (as one would expect) not significant.

I came across this while studying the various functions in the base layer of this project. The overall style and design of its building blocks are a great inspiration. Thanks for making this project's source public.

If you confirm this is a bug, I can have a try at a PR to fix it.

[Mallchad]

This API is specific to the linker it seems. and it very clearly removes contiguous duplicates, not making a single instance in the array.
B32 is_unique = in.v[i - 1] != in.v[i];
It's just not obvious this is unintended, but hey, it would be nice if it had a short docstring.