Bitmapped vector tries vs. arrays

[deadalnix]

Le 18/12/2011 09:22, Andrei Alexandrescu a écrit :
> This is bound to turn interesting:
>
> http://www.reddit.com/r/programming/comments/nddlj/extreme_cleverness_functional_data_structures_in/
>
>
>
> Andrei

Interesting speach, but thoses datastructures are kind of trying dealing 
with a nail using a screwdriver. Problems solved here are not problem 
that nicely solve in functionnal, so why use functionnal ?

I did always think it is stupid to enforce a paradigm because it is 
supposed to be supperior, blablabla . . .

Java did it with OOP, and we end up with many stupid thing, static main 
within a class. OOP is working for some stuff, and not for others. Just 
as functionnal.

Secondly, I think this conf is going out of sanity during the butmap 
vector trie presentation (about 40 mins). Many claims not backed with 
facts, and messy explaination (the cache stuff doesn't exlain in any way 
how this datastruct can be faster than ArrayList, and I bet it isn't 
expect on a flawed benchmark). No idea why polymorphism is revealant 
here either - but this is most likely an error and not the word the 
speaker wanted to use.

Some argument are not convincing. OK, log32(n) is close to O(1) in many 
cases but it isn't O(1). Matter of fact.

It look like the author try to convince us that oranges can taste like 
banana, and show us many techniques to go in that direction, but this is 
useless as logn as we have banana available.

Thoses datastruct are interesting anyway on a research perspective.