1.2.1
- Fix support for ghc-9.2 #99
- Fix performance regression for ghc-9.0 #101
- Add
uniformEnumM
and uniformEnumRM
- Add
initStdGen
#103
- Add
globalStdGen
#117
- Add
runStateGenST_
- Ensure that default implementation of
ShortByteString
generation uses
unpinned memory. #116
- Fix #54 with
#68 - if exactly one value in the
range of floating point is infinite, then
uniformRM
/randomR
returns that
value.
- Add default implementation of
uniformM
that uses Generic
#70
Random
instance for CBool
#77
- Addition of
TGen
and TGenM
#95
- Addition of tuple instances for
Random
up to 7-tuple
#72
1.2.0
- Breaking change which mostly maintains backwards compatibility, see
“Breaking Changes” below.
- Support for monadic generators e.g. mwc-random.
- Monadic adapters for pure generators (providing a uniform monadic
interface to pure and monadic generators).
- Faster in all cases except one by more than x18 (N.B. x18 not 18%) and
some cases (depending on the type) faster by more than x1000 - see
below for benchmarks.
- Passes a large number of random number test suites:
- Better quality split as judged by these
tests. Again
see random-quality for
details on how to do this yourself.
- Unbiased generation of ranges.
- Updated tests and benchmarks.
- Continuous integration.
Breaking Changes
Version 1.2.0 introduces these breaking changes:
- requires
base >= 4.8
(GHC-7.10)
StdGen
is no longer an instance of Read
randomIO
and randomRIO
were extracted from the Random
class into
separate functions
In addition, there may be import clashes with new functions, e.g. uniform
and
uniformR
.
Deprecations
Version 1.2.0 introduces genWord64
, genWord32
and similar methods to the
RandomGen
class. The significantly slower method next
and its companion
genRange
are now deprecated.
Issues Addressed
Issue Number |
Description |
Comment |
25 |
The seeds generated by split are not independent |
Fixed: changed algorithm to SplitMix, which provides a robust split operation |
26 |
Add Random instances for tuples |
Addressed: added Uniform instances for up to 6-tuples |
44 |
Add Random instance for Natural |
Addressed: added UniformRange instance for Natural |
51 |
Very low throughput |
Fixed: see benchmarks below |
53 |
incorrect distribution of randomR for floating-point numbers |
(*) |
55 |
System/Random.hs:43:1: warning: [-Wtabs] |
Fixed: No more tabs |
58 |
Why does random for Float and Double produce exactly 24 or 53 bits? |
(*) |
59 |
read :: StdGen fails for strings longer than 6 |
Addressed: StdGen is no longer an instance of Read |
Comments
(*) 1.2 samples more bits but does not sample every Float
or
Double
. There are methods to do this but they have some downsides;
see here for a
fuller discussion.
Benchmarks
Here are some benchmarks run on a 3.1 GHz Intel Core i7. The full
benchmarks can be run using e.g. stack bench
. The benchmarks are
measured in milliseconds per 100,000 generations. In some cases, the
performance is over x1000 times better; the minimum performance
increase for the types listed below is more than x36.
Name |
1.1 Mean |
1.2 Mean |
Float |
27.819 |
0.305 |
Double |
50.644 |
0.328 |
Integer |
42.332 |
0.332 |
Word |
40.739 |
0.027 |
Int |
43.847 |
0.028 |
Char |
17.009 |
0.462 |
Bool |
17.542 |
0.027 |
1.1
1.0.1.1
bump for overflow bug fixes
1.0.1.2
bump for ticket 8704, build fusion
1.0.1.0
bump for bug fixes,
1.0.0.4
bumped version for float/double range bugfix