How to use this most efficiently?

[markg85]

How to use this most efficiently?

[infusion]

Hi
currently you can't control that behavior, since the library is optimized for speed, not memory. The 10 elements are a reasonable pre-allocation at the beginning to save re-allocations in most basic usecases < 320 bit. Especially when you narrow it down the way you're asking for you can't handle big jumps properly, such as .set(2), .set(1000). In your desired way you would increase the size linearly, instead of pre-allocating larger ranges.
Do you have a use case where 3*4 byte really make such a difference?

[markg85]

Hi,

Personally i'd argue that using a bitset most often implies one is going for memory efficiency above performance.

I do have a usecase where this really matters though it's very much niche.
I'm just playing with a site to show crypto prices. I'm using the crypto.com ticker API for that which gives me that number of 214 (that many different cryptos). In my specific case i get enough data from the ticker to send an update to the browser every .05 seconds. I'm already limiting the data i send a lot to the bare minimum but that's still a list of 214 entries with the difference between the previous message.

This brings down data usage per 0.5 seconds to ~300 bytes.

Now if i have a bitset that tells me which entries changed then i only need to send:

• that bitset (ideally 216 bits, 27 bytes)
• each changed value (that's 20 at most in this interval thus at most 80 bytes where i assume each entry is a 4 byte float)
• new likely max packet size ~100 bytes per 0.5 seconds

So yeah, this very niche case, does in fact save quite a bit of data if it were possible with BitSet.

[markg85]

So, a very quick experiment with https://github.com/Callidon/bloom-filters/blob/master/src/bloom/bit-set.ts (from the bloom-filters package, the bitset is just for internal purposes).

That is memory efficient and it does in fact turn out that using such a structure gives me 0.5 second data blobs between 40-100 bytes. The API of that class is weird though... Oh well, works for my needs.

Could you perhaps consider adding a settable flag to your class to choose between memory efficiency and performance?