https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127690
--- Comment #7 from Oleksandr Kulkov <adamant.pwn at gmail dot com> --- I now went over the whole of pb_ds with AI assistance, comparing its capabilities and performance to the standard library. Ultimately, I think the only part whose removal would really hurt is dynamic order statistics: efficient insertion/erasure together with order_of_key() and find_by_order(). If pb_ds is to go away, could we consider a substitute for this capability specifically? I’m thinking of a focused ordered container with one balanced-tree implementation, fixed subtree-size metadata, and proper support for move semantics and stateful allocators. I’d keep the initial scope to ordinary ordered-container operations plus rank/select. I’d be happy to work on a modern C++ implementation if there’s any chance to eventually accepting it as a libstdc++ extension. Would something along those lines be worth pursuing?
