Comment by cb321

I tried to defend you against a hard to support "most people didn't want this" with a workload (either static or dynamic) assertion. You come back saying you don't need such charitability while simultaneously clarifying in terms of a static workload! Seemingly, "most written C++" (again charitably moving from "people writing" to "written code" because the latter is probably more relevant and surely more auditable/studyable - "source code files easier than people").

I think you should be more careful about claiming you know what people want or know. Many people I know don't seem to know what they want.. it's some melange of desiderata with various eternal tensions. As just one example, I would doubt most people even want performance above all else. I really doubt most people understand well how their various static and dynamic workloads relate to their performance. I doubt I'm alone in having known many C++ programmers who, from spoken conversation, revealed they thought STL map was a hash table, for example.

I only said "one of the main" and almost added "with pointer stability being the bigger", but it had already been observed by rurban. Delete-while-iterating has close ties with cell motion (though convoluted workarounds may exist, they may well cost more than you personally would like, it sounds). How to best explain all these things is always questionable, but I was trying to supplement. It's generally not easy without visual aids and definitely not if people are being combative. Python has ever & always since the late 80s used open addressing but only more recently tried to `raise` on dict edits during iterations.

It seems to be flip-flopping that you are now calling separately chained hash tables a "weird container type" when 3 months ago it seemed to be news to you that they were NOT the main strategy for decades before open addressing on some Nim -ish comment on a Carbon thread: https://news.ycombinator.com/item?id=44754821

Anything else I would say, such as Stepanov not even choosing a hash table in the first place, gpderetta said very well in a sibling. For the record, I still very much disagree with the early (& later) design of the STL and am no fan of WG21. Have a good weekend.