bloomfilter-blocked: update the README

Author: jorisdral

bloomfilter-blocked/README.md

@@ -1,29 +1,71 @@
-# A fast, space efficient Bloom filter implementation
+# bloomfilter-blocked
 
-Copyright 2008, 2009, 2010, 2011 Bryan O'Sullivan <[email protected]>.
+`bloomfilter-blocked` is a Haskell library providing multiple fast and efficient
+implementations of [bloom filters][bloom-filter:wiki]. It is a full rewrite of
+the [`bloomfilter`][bloomfilter:hackage] package, originally authored by Bryan
+O'Sullivan <[email protected]>.
 
-This package provides both mutable and immutable Bloom filter data
-types, along with a family of hash function and an easy-to-use
-interface.
+A bloom filter is a space-efficient data structure representing a set that can
+be probablistically queried for set membership. The set membership query returns
+no false negatives, but it might return false positives. That is, if an element
+was added to a bloom filter, then a subsequent query definitely returns `True`.
+If an element was *not* added to a filter, then a subsequent query may still
+return `True` if `False` would be the correct answer. The probabiliy of false
+positives -- the false positive rate (FPR) -- is configurable, as we will
+describe later.
 
-To build:
+The library includes two implementations of bloom filters: classic, and blocked.
 
-    cabal install bloomfilter
+* **Classic** bloom filters, found in the `Data.BloomFilter.Classic` module: a
+  default implementation that is faithful to the canonical description of a
+  bloom filter data structure.
 
-For examples of usage, see the Haddock documentation and the files in
-the examples directory.
+* **Blocked** floom filters, found in the `Data.BloomFilter.Blocked` module: an
+  implementation that optimises the memory layout of a classic bloom filter for
+  speed (cheaper CPU cache reads), at the cost of a slightly higher FPR for the
+  same amount of assigned memory.
 
+The FPR scales inversely with how much memory is assigned to the filter. It also
+scales inversely with how many elements are added to the set. The user can
+configure how much memory is asisgned to a filter, and the user also controls
+how many elements are added to a set. Each implementation comes with helper
+functions, like `sizeForFPR` and `sizeForBits`, that the user can leverage to
+configure filters.
 
-# Get involved!
+Both immutable (`Bloom`) and mutable (`MBloom`) bloom filters, including
+functions to convert between the two, are provided for each implementation. Note
+however that a (mutable) bloom filter can not be resized once created, and that
+elements can not be deleted once inserted.
 
-Please report bugs via the
-[github issue tracker](https://github.com/haskell-pkg-janitors/bloomfilter).
+For more information about the library and examples of how to use it, see the
+Haddock documentation of the different modules.
 
-Master [git repository](https://github.com/haskell-pkg-janitors/bloomfilter):
+# Usage notes
 
-* `git clone git://github.com/haskell-pkg-janitors/bloomfilter.git`
+User should take into account the following:
 
+* This package is not supported on 32bit systems.
 
-# Authors
+# Differences from the `bloomfilter` package
 
-This library is written by Bryan O'Sullivan, <[email protected]>.
+The library is a full rewrite of the [`bloomfilter`][bloomfilter:hackage]
+package, originally authored by Bryan O'Sullivan <[email protected]>. The main
+differences are:
+
+* `bloomfilter-blocked` supports both classic and blocked bloom filters, whereas
+  `bloomfilter` only supports the former.
+* `bloomfilter-blocked` supports bloom filters of arbitrary sizes, whereas
+  `bloomfilter` limits the sizes to powers of two. Moreover, we support sizes up
+  to `2^48` for classic bloom filters and up to `2^41` for blocked bloom
+  filters, instead of `2^32`.
+* In `bloomfilter`, the `Bloom` and `MBloom` types are parametrised over a
+  `Hashable` type class. In `bloomfilter-blocked` the hashing scheme is static.
+  This allows clean (de-)serialisation of bloom filters.
+* For hashing `bloomfilter-blocked` uses `XXH3` instead of the Jenkins'
+  `lookup3` that `bloomfilter` uses.
+
+
+<!-- Sources -->
+
+[bloom-filter:wiki]: https://en.wikipedia.org/wiki/Bloom_filter
+[bloomfilter:hackage]: https://hackage.haskell.org/package/bloomfilter

bloomfilter-blocked/src/Data/BloomFilter.hs

@@ -1,3 +1,6 @@
+-- | By default, this module re-exports the classic bloom filter implementation
+-- from "Data.BloomFilter.Classic". If you want to use the blocked bloom filter
+-- implementation, import "Data.BloomFilter.Blocked".
 module Data.BloomFilter (
     module Data.BloomFilter.Classic
   ) where