unisonweb/unison
1
0
Fork 0
mirror of https://github.com/unisonweb/unison.git synced 2024-09-21 23:37:23 +03:00
Code Issues Projects Releases Wiki Activity

polish, lots of helper functions

Browse Source
This commit is contained in:
Paul Chiusano 2016-12-29 01:06:53 -05:00
parent 379d4d5188
commit e292987fc7
2 changed files with 121 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
yaks/easytest/README.markdown
View File

@ -5,37 +5,53 @@ import EasyTest
import Control.Monad
import Control.Applicative
main = runOnly "addition" $ do
expect (1 + 1 == 2)
fork $ do
ns <- [0..10] `forM` \n -> replicateM n (randomBetween (0 :: Int, 10))
ns `forM_` \ns -> expect (reverse (reverse ns) == ns)
scope "addition" $ expect (3 + 3 == 6)
scope "always passes" $ do
note "I'm running this test, even though it always passes!"
ok -- like `pure ()`, but records a success result
scope "failing test" $ crash "oh noes!!"
main = runOnly "addition" $ tests
[ expect (1 + 1 == 2)
, fork $ do
-- generate lists from size 0 to 10, of Ints in (0,43)
-- shorthand: listsOf [0..10] (int' 0 43)
ns <- [0..10] `forM` \n -> replicateM n (int' 0 43)
ns `forM_` \ns -> expect (reverse (reverse ns) == ns)
, scope "addition" $ expect (3 + 3 == 6)
, scope "always passes" $ do
note "I'm running this test, even though it always passes!"
ok -- like `pure ()`, but records a success result
, scope "failing test" $ crash "oh noes!!" ]
```
The library is simple: you just write ordinary Haskell code in the `Test` monad, which has access to:
The idea here is to write tests with ordinary Haskell code, with control flow explicit and under programmer control. Tests are values of type `Test a`, and `Test` forms a monad with access to:
* random numbers (the `random` and `randomBetween` functions)
* repeatable randomness (the `random` and `random'` functions for random and bounded random values, or handy specialized `int`, `int'`, `double`, `double'`, etc)
* I/O (via `liftIO`)
* failure (via `crash`)
* logging (via `note` or `noteScoped`)
* hierarchically-named subcomputations which can be switched on and off (in the above code, only the `"addition"`-scoped test would be run, and we could do `run` instead if we wanted to run the whole suite)
* failure (via `crash`, which yields a stack trace!)
* logging (via `note`, `noteScoped`, or `note'`)
* hierarchically-named subcomputations which can be switched on and off (in the above code, `"always passes" and "failing test"`-scoped tests would not be run, and we could do `run` instead of `runOnly` if we wanted to run the whole suite)
* parallelism (note the `fork` which runs that subtree of the test suite in a parallel thread).
* conjunction of tests via `MonadPlus` (the `<|>` operation runs both tests, even if the first test fails, and the `tests` function used above is just `msum`).
`Test` is an instance of everything through `MonadPlus` (the `<|>` operation runs both tests, even if the first test fails). You assemble `Test` values into a test suite using ordinary Haskell code, not framework magic. Notice that to generate a list of random values, we just `replicateM` and `forM` as usual. If this gets tedious... we can factor this logic out into helper functions!
Using any or all of these capabilities, you assemble `Test` values into a "test suite" (just another `Test` value) using ordinary Haskell code, not framework magic. Notice that to generate a list of random values, we just `replicateM` and `forM` as usual. If this gets tedious... we can factor this logic out into helper functions! For instance:
```Haskell
listOf :: Int -> Test a -> Test [a]
listOf = replicateM
listsOf :: [Int] -> Test a -> Test [[a]]
listsOf sizes gen = sizes `forM` \n -> listOf n gen
ex :: Test ()
ex = do
ns <- listsOf [0..100] int
ns `forM_` \ns -> expect (reverse (reverse ns) == ns)
```
This library is opinionated and might not be for everyone. But here's some of my thinking in writing it:
* Testing should uncomplicated, minimal friction, and ideally: FUN.
* A lot of testing frameworks are weirdly optimized for adding lots of diagnostic information up front, as if just whatever diagnostic information you happen to think to capture will magically allow you to fix whatever bugs your tests reveal. EastTest takes the opposite approach: be lazy about adding diagnostics and labeling subexpressions, but make it trivial to reproduce failing tests without running your entire suite. If a test fails, you can easily rerun just that test, with the same random seed, and add whatever diagnostics or print statements you need to track down what's wrong.
* A lot of testing frameworks are weirdly optimized for adding lots of diagnostic information up front, as if whatever diagnostic information you happen to think to capture will be exactly what is needed to fix whatever bugs your tests reveal. EasyTest takes the opposite approach: be lazy about adding diagnostics and labeling subexpressions, but make it trivial to reproduce failing tests without running your entire suite. If a test fails, you can easily rerun just that test, with the same random seed, and add whatever diagnostics or print statements you need to track down what's wrong.
* Another reason not to add diagnostics up front: you avoid needing to remember two different versions of every function or operator (the one you use in your regular code, and the one you use with your testing "framework" to supply diagnostics). HUnit has operators named `(@=?)`, `(~?=)`, and a bunch of others for asserting equality with diagnostics on failure. QuickCheck has `(.&&.)` and `(.||.)`. Just... no.
* HUnit, QuickCheck, SmallCheck, Tasty, and whatever else are frameworks that hide control flow from the programmer and make some forms of control flow difficult or impossible to specify (for instance, you can't do I/O in your QuickCheck tests!). In contrast, EasyTest is just a single data type with a monadic API and a few helper functions. You assemble your tests using ordinary monadic code, and there is never any magic. Want to abstract over something? _Write a regular function._ Need to generate some testing data? Write regular functions.
* "How do I modify the number of generated test cases for QuickCheck for just one of my properties?" Or control the maximum size for these `Gen` and `Arbitrary` types? Some arbitrary "configuration setting" that you have to look up every time.
* "How do I modify the number of generated test cases for QuickCheck for just one of my properties?" Or control the maximum size for these `Gen` and `Arbitrary` types? Some arbitrary "configuration setting" that you have to look up every time. No thanks.
* Global configuration settings are evil. I want fine-grained control over the amount of parallelism, test case sizes, and so on.
* Most of the functionality of QuickCheck is overkill anyway! There's no need for `Arbitrary` instances (explicit generation is totally fine, and even preferred in most cases), `Coarbitrary` (cute, but not useful when the HOF you are testing is parametric), or shrinking (just generate your test cases in increasing sizes, and your first failure will be the smallest).
* Most of the functionality of QuickCheck is overkill anyway! There's no need for `Arbitrary` instances (explicit generation is totally fine, and even preferred in most cases), `Coarbitrary` (cute, but not useful when the HOF you are testing is parametric), or shrinking (just generate your test cases in increasing sizes, and your first failure will be the smallest!).
I hope that you enjoy the library and that it proves useful.

99
yaks/easytest/src/EasyTest.hs
View File

@ -2,7 +2,7 @@
{-# Language FunctionalDependencies #-}
{-# Language GeneralizedNewtypeDeriving #-}
module EasyTest (Test, crash, currentScope, noteScoped, skip, ok, fork, fork', scope, note, expect, tests, random, randomBetween, run', runOnly, run, rerun, rerunOnly, parseMessages, module Control.Monad.IO.Class) where
module EasyTest where
import Control.Applicative
import Control.Concurrent
@ -12,12 +12,14 @@ import Control.Monad
import Control.Monad.IO.Class
import Control.Monad.Reader
import Data.List
import Data.Map (Map)
import Data.Word
import GHC.Stack
import qualified System.Random as Random
import System.Random (Random)
import System.Exit
import System.Random (Random)
import qualified Control.Concurrent.Async as A
import qualified Data.Map as Map
import qualified System.Random as Random
data Status = Failed | Passed | Skipped
@ -91,23 +93,28 @@ run' seed note allow (Test t) = do
Right () -> note $ "Waiting for any asynchronously spawned tests to complete ..."
atomically $ writeTQueue resultsQ Nothing
_ <- A.waitCatch rs
note line
note "\n"
resultsMap <- readTVarIO results
let
resultsList = Map.toList resultsMap
succeeded = length [ a | a@(_, Passed) <- resultsList ]
failures = [ a | (a, Failed) <- resultsList ]
failed = length failures
note $ " " ++ show succeeded ++ (if failed == 0 then " PASSED" else " passed")
note $ " " ++ show (length failures) ++ (if failed == 0 then " failed" else " FAILED (failed scopes below)")
case failures of
[] -> do
note "\n"
note line
note "✅ all tests passed! 👍 🎉"
case succeeded of
0 -> do
note "😶 hmm ... no test results recorded"
note "Tip: use `ok`, `expect`, or `crash` to record results"
note "Tip: if running via `runOnly` or `rerunOnly`, check for typos"
1 -> note $ "✅ 1 test passed, no failures! 👍 🎉"
_ -> note $ "✅ " ++ show succeeded ++ " tests passed, no failures! 👍 🎉"
(hd:_) -> do
note $ " " ++ intercalate "\n " (map showMessages failures)
note line
note "\n"
note $ " " ++ show succeeded ++ (if failed == 0 then " PASSED" else " passed")
note $ " " ++ show (length failures) ++ (if failed == 0 then " failed" else " FAILED (failed scopes below)")
note $ " " ++ intercalate "\n " (map (show . showMessages) failures)
note ""
note $ " To rerun with same random seed:\n"
note $ " EasyTest.rerun " ++ show seed
@ -142,6 +149,9 @@ note msg = do
liftIO $ note_ msg
pure ()
note' :: Show s => s -> Test ()
note' = note . show
random :: Random a => Test a
random = do
rng <- asks rng
@ -151,15 +161,78 @@ random = do
writeTVar rng rng1
pure a
randomBetween :: Random a => (a,a) -> Test a
randomBetween bounds = do
random' :: Random a => a -> a -> Test a
random' lower upper = do
rng <- asks rng
liftIO . atomically $ do
rng0 <- readTVar rng
let (a, rng1) = Random.randomR bounds rng0
let (a, rng1) = Random.randomR (lower,upper) rng0
writeTVar rng rng1
pure a
int :: Test Int
int = random
char :: Test Char
char = random
double :: Test Double
double = random
word :: Test Word
word = random
word8 :: Test Word8
word8 = random
int' :: Int -> Int -> Test Int
int' = random'
char' :: Char -> Char -> Test Char
char' = random'
double' :: Double -> Double -> Test Double
double' = random'
word' :: Word -> Word -> Test Word
word' = random'
word8' :: Word8 -> Word8 -> Test Word8
word8' = random'
-- | Sample uniformly from the given list of possibilities
pick :: [a] -> Test a
pick as = let n = length as; ind = picker n as in do
i <- int' 0 (n - 1)
Just a <- pure (ind i)
pure a
picker :: Int -> [a] -> (Int -> Maybe a)
picker _ [] = const Nothing
picker _ [a] = \i -> if i == 0 then Just a else Nothing
picker size as = go where
lsize = size `div` 2
rsize = size - lsize
(l,r) = splitAt lsize as
lpicker = picker lsize l
rpicker = picker rsize r
go i = if i < lsize then lpicker i else rpicker (i - lsize)
listOf :: Int -> Test a -> Test [a]
listOf = replicateM
listsOf :: [Int] -> Test a -> Test [[a]]
listsOf sizes gen = sizes `forM` \n -> listOf n gen
pair :: Test a -> Test b -> Test (a,b)
pair = liftA2 (,)
mapOf :: Ord k => Int -> Test k -> Test v -> Test (Map k v)
mapOf n k v = Map.fromList <$> listOf n (pair k v)
mapsOf :: Ord k => [Int] -> Test k -> Test v -> Test [Map k v]
mapsOf sizes k v = sizes `forM` \n -> mapOf n k v
wrap :: Test a -> Test a
wrap (Test t) = Test $ do
env <- ask