I don't have any problem with ADTs or abstraction ... In fact I'd say the strength of C++, and why I use it, is precisely that you can create your own abstractions, more so than C.
I'm just saying that using a canned set with "many useful" operations is suboptimal, design by committee is suboptimal, etc.
As opposed to analyzing the ops your app needs and creating custom data structures. i.e. there are many different ADTs for "hash table" or "set"
That might seem obvious, but it's not how software development is being done today. There's a lot of code reuse that leads to suboptimal software; it would actually be better to copy and paste and refine more.
I still think you're mostly conflating "the C++ standard library provides bad implementations of this stuff" which is famously true, with "generic solutions are just bad" which is at least non-obvious and I'd argue generally false.
> copy and paste and refine more.
That doesn't get you better algorithms, which is what you need. If you start with the C++ standard library's unordered_set implementation in a source file and you "refine" that you won't get from there to absl::flat_hash_set except in the same sense you could start with the footage from "Grease" and end up making "Bugsy Malone". Start over with a different design.
Picking from a handful of options like absl::flat_hash_set rather than always hand-rolling is a clear win for productivity. As with anything else in performance, measure first and only solve problems you actually have applies to the idea that maybe the choice doesn't work - even for the abysmal std::unordered_set if you don't have measurements showing it's a problem then it probably just isn't a problem.
I'm just saying that using a canned set with "many useful" operations is suboptimal, design by committee is suboptimal, etc.
As opposed to analyzing the ops your app needs and creating custom data structures. i.e. there are many different ADTs for "hash table" or "set"
That might seem obvious, but it's not how software development is being done today. There's a lot of code reuse that leads to suboptimal software; it would actually be better to copy and paste and refine more.