doc: make it easier to work out size of bloom filter #19130

I've manually traced through the bloom filter code to work out how big a filter ends up being a few times now; this just adds some python code as a comment so it's easy to do that by cutting and pasting into an interpreter instead.

Looks right, but perhaps it should go in the .h file (which has an easier approximate formula already).

An alternative would be to run the code in massif and show the size of the object. :grimacing:

To have an idea of the approximation, the formula "4.328 bits per element and per bit of fprate" is at most 0.115% off for better than 8-bit fprate (fprate < 1/256). For 10-bit, at most 0.074%. For 12-bit, at most 0.052%. For 20-bit, 0.019%.

Looks right, but perhaps it should go in the .h file (which has an easier approximate formula already).

I agree. In general, this kind of documentation would be better in the header file. This also gets included in doxygen etc.

@ajtowns did you want to follow up here and move this into the header?

This is really helpful. I'd love to see it merged into the header file at some point.

#19968 did that, i think ?

#19968 did that, i think ?

Oh! I remembered there was a python snippet to calculate this and couldn't find it in master. I see now that #19968 included the formula without the script. That's good.