Re: [boost] Determining interest in container with efficient random access insert and erase

Subject: Re: [boost] Determining interest in container with efficient random access insert and erase
From: Ion Gaztañaga (igaztanaga_at_[hidden])
Date: 2015-09-12 06:46:23

On 12/09/2015 1:10, Glen Fernandes wrote:
>> Useful links:
>>   - Source code: https://github.com/det/segmented_tree_seq
>>   - Documentation: http://det.github.io/segmented_tree_seq/
>>   - Benchmarks: https://github.com/det/segmented_tree_seq/tree/benchmarks
>>
>> Any feedback is greatly appreciated.
>
> The benchmarks are impressive. I hope to take a deeper look this
> weekend, but from my cursory glance: the implementation looks very
> clean too.
>
> Ion: Is this something you would be interested in reviewing?

I'd like to separate the proposal to put that data structure into Boost 
and the proposal to put it into Boost.Container.

The implementation looks good although I have no experience with B+Trees 
or augmented data structures. I think we should have more innovative 
container in Boost and there have been more proposals to similar data 
structures that were not.

Regarding Boost.Container, one of the biggest problems with adding new 
stuff into Boost.Container is that it should be consistent with the rest 
of the library. The library supports C++03, supports Interprocess 
requirements, recursive types, SCARY iterators, etc... it must be 
maintained and documented as the rest of the library. I wouldn't like to 
maintain data structures I really don't fully understand or just left 
them unmaintained. We added stable_vector and static_vector but those 
were easier data structures.

Just like we have Boost.CircularBuffer or Boost.Unordered I'd like to 
propose adding adding new containers in its own library. Maybe an 
(optionally augmented) B+Tree library would be nice, and it could also 
contain a sequence container based on that data structure.

Just my 2 cents, does it make sense?

Ion