2 % (c) The GRASP/AQUA Project, Glasgow University, 1992-1998
4 \section[Util]{Highly random utility functions}
7 -- IF_NOT_GHC is meant to make this module useful outside the context of GHC
13 Eager, thenEager, returnEager, mapEager, appEager, runEager,
16 -- general list processing
17 zipEqual, zipWithEqual, zipWith3Equal, zipWith4Equal,
18 zipLazy, stretchZipWith,
19 mapAndUnzip, mapAndUnzip3,
20 nOfThem, lengthExceeds, isSingleton, only,
28 IF_NOT_GHC(quicksort COMMA stableSortLt COMMA mergesort COMMA)
30 IF_NOT_GHC(mergeSort COMMA) naturalMergeSortLe, -- from Carsten
31 IF_NOT_GHC(naturalMergeSort COMMA mergeSortLe COMMA)
33 -- transitive closures
37 mapAccumL, mapAccumR, mapAccumB, foldl2, count,
40 thenCmp, cmpList, prefixMatch, postfixMatch,
46 IF_NOT_GHC(cfst COMMA applyToPair COMMA applyToFst COMMA)
47 IF_NOT_GHC(applyToSnd COMMA foldPair COMMA)
51 #if __GLASGOW_HASKELL__ < 402
58 #if __GLASGOW_HASKELL__ <= 408
66 #include "HsVersions.h"
68 import IO ( hPutStrLn, stderr )
69 import List ( zipWith4 )
70 import Panic ( panic )
71 import IOExts ( IORef, newIORef, unsafePerformIO )
73 #if __GLASGOW_HASKELL__ <= 408
74 import Exception ( catchIO, justIoErrors, raiseInThread )
76 #ifndef mingw32_TARGET_OS
82 %************************************************************************
84 \subsection{The Eager monad}
86 %************************************************************************
88 The @Eager@ monad is just an encoding of continuation-passing style,
89 used to allow you to express "do this and then that", mainly to avoid
90 space leaks. It's done with a type synonym to save bureaucracy.
95 type Eager ans a = (a -> ans) -> ans
97 runEager :: Eager a a -> a
98 runEager m = m (\x -> x)
100 appEager :: Eager ans a -> (a -> ans) -> ans
101 appEager m cont = m cont
103 thenEager :: Eager ans a -> (a -> Eager ans b) -> Eager ans b
104 thenEager m k cont = m (\r -> k r cont)
106 returnEager :: a -> Eager ans a
107 returnEager v cont = cont v
109 mapEager :: (a -> Eager ans b) -> [a] -> Eager ans [b]
110 mapEager f [] = returnEager []
111 mapEager f (x:xs) = f x `thenEager` \ y ->
112 mapEager f xs `thenEager` \ ys ->
117 %************************************************************************
119 \subsection{A for loop}
121 %************************************************************************
124 -- Compose a function with itself n times. (nth rather than twice)
125 nTimes :: Int -> (a -> a) -> (a -> a)
128 nTimes n f = f . nTimes (n-1) f
132 %************************************************************************
134 \subsection[Utils-lists]{General list processing}
136 %************************************************************************
138 A paranoid @zip@ (and some @zipWith@ friends) that checks the lists
139 are of equal length. Alastair Reid thinks this should only happen if
140 DEBUGging on; hey, why not?
143 zipEqual :: String -> [a] -> [b] -> [(a,b)]
144 zipWithEqual :: String -> (a->b->c) -> [a]->[b]->[c]
145 zipWith3Equal :: String -> (a->b->c->d) -> [a]->[b]->[c]->[d]
146 zipWith4Equal :: String -> (a->b->c->d->e) -> [a]->[b]->[c]->[d]->[e]
150 zipWithEqual _ = zipWith
151 zipWith3Equal _ = zipWith3
152 zipWith4Equal _ = zipWith4
154 zipEqual msg [] [] = []
155 zipEqual msg (a:as) (b:bs) = (a,b) : zipEqual msg as bs
156 zipEqual msg as bs = panic ("zipEqual: unequal lists:"++msg)
158 zipWithEqual msg z (a:as) (b:bs)= z a b : zipWithEqual msg z as bs
159 zipWithEqual msg _ [] [] = []
160 zipWithEqual msg _ _ _ = panic ("zipWithEqual: unequal lists:"++msg)
162 zipWith3Equal msg z (a:as) (b:bs) (c:cs)
163 = z a b c : zipWith3Equal msg z as bs cs
164 zipWith3Equal msg _ [] [] [] = []
165 zipWith3Equal msg _ _ _ _ = panic ("zipWith3Equal: unequal lists:"++msg)
167 zipWith4Equal msg z (a:as) (b:bs) (c:cs) (d:ds)
168 = z a b c d : zipWith4Equal msg z as bs cs ds
169 zipWith4Equal msg _ [] [] [] [] = []
170 zipWith4Equal msg _ _ _ _ _ = panic ("zipWith4Equal: unequal lists:"++msg)
175 -- zipLazy is lazy in the second list (observe the ~)
177 zipLazy :: [a] -> [b] -> [(a,b)]
179 zipLazy (x:xs) ~(y:ys) = (x,y) : zipLazy xs ys
184 stretchZipWith :: (a -> Bool) -> b -> (a->b->c) -> [a] -> [b] -> [c]
185 -- (stretchZipWith p z f xs ys) stretches ys by inserting z in
186 -- the places where p returns *True*
188 stretchZipWith p z f [] ys = []
189 stretchZipWith p z f (x:xs) ys
190 | p x = f x z : stretchZipWith p z f xs ys
191 | otherwise = case ys of
193 (y:ys) -> f x y : stretchZipWith p z f xs ys
198 mapAndUnzip :: (a -> (b, c)) -> [a] -> ([b], [c])
200 mapAndUnzip f [] = ([],[])
204 (rs1, rs2) = mapAndUnzip f xs
208 mapAndUnzip3 :: (a -> (b, c, d)) -> [a] -> ([b], [c], [d])
210 mapAndUnzip3 f [] = ([],[],[])
211 mapAndUnzip3 f (x:xs)
214 (rs1, rs2, rs3) = mapAndUnzip3 f xs
216 (r1:rs1, r2:rs2, r3:rs3)
220 nOfThem :: Int -> a -> [a]
221 nOfThem n thing = replicate n thing
223 lengthExceeds :: [a] -> Int -> Bool
224 -- (lengthExceeds xs n) is True if length xs > n
225 (x:xs) `lengthExceeds` n = n < 1 || xs `lengthExceeds` (n - 1)
226 [] `lengthExceeds` n = n < 0
228 isSingleton :: [a] -> Bool
229 isSingleton [x] = True
230 isSingleton _ = False
241 snocView :: [a] -> ([a], a) -- Split off the last element
242 snocView xs = go xs []
244 go [x] acc = (reverse acc, x)
245 go (x:xs) acc = go xs (x:acc)
248 Debugging/specialising versions of \tr{elem} and \tr{notElem}
251 isIn, isn'tIn :: (Eq a) => String -> a -> [a] -> Bool
254 isIn msg x ys = elem__ x ys
255 isn'tIn msg x ys = notElem__ x ys
257 --these are here to be SPECIALIZEd (automagically)
259 elem__ x (y:ys) = x==y || elem__ x ys
261 notElem__ x [] = True
262 notElem__ x (y:ys) = x /= y && notElem__ x ys
266 = elem (_ILIT 0) x ys
270 | i ># _ILIT 100 = panic ("Over-long elem in: " ++ msg)
271 | otherwise = x == y || elem (i +# _ILIT(1)) x ys
274 = notElem (_ILIT 0) x ys
276 notElem i x [] = True
278 | i ># _ILIT 100 = panic ("Over-long notElem in: " ++ msg)
279 | otherwise = x /= y && notElem (i +# _ILIT(1)) x ys
285 %************************************************************************
287 \subsection[Utils-sorting]{Sorting}
289 %************************************************************************
291 %************************************************************************
293 \subsubsection[Utils-quicksorting]{Quicksorts}
295 %************************************************************************
300 -- tail-recursive, etc., "quicker sort" [as per Meira thesis]
301 quicksort :: (a -> a -> Bool) -- Less-than predicate
303 -> [a] -- Result list in increasing order
306 quicksort lt [x] = [x]
307 quicksort lt (x:xs) = split x [] [] xs
309 split x lo hi [] = quicksort lt lo ++ (x : quicksort lt hi)
310 split x lo hi (y:ys) | y `lt` x = split x (y:lo) hi ys
311 | True = split x lo (y:hi) ys
315 Quicksort variant from Lennart's Haskell-library contribution. This
316 is a {\em stable} sort.
319 stableSortLt = sortLt -- synonym; when we want to highlight stable-ness
321 sortLt :: (a -> a -> Bool) -- Less-than predicate
323 -> [a] -- Result list
325 sortLt lt l = qsort lt l []
327 -- qsort is stable and does not concatenate.
328 qsort :: (a -> a -> Bool) -- Less-than predicate
329 -> [a] -- xs, Input list
330 -> [a] -- r, Concatenate this list to the sorted input list
331 -> [a] -- Result = sort xs ++ r
335 qsort lt (x:xs) r = qpart lt x xs [] [] r
337 -- qpart partitions and sorts the sublists
338 -- rlt contains things less than x,
339 -- rge contains the ones greater than or equal to x.
340 -- Both have equal elements reversed with respect to the original list.
342 qpart lt x [] rlt rge r =
343 -- rlt and rge are in reverse order and must be sorted with an
344 -- anti-stable sorting
345 rqsort lt rlt (x : rqsort lt rge r)
347 qpart lt x (y:ys) rlt rge r =
350 qpart lt x ys (y:rlt) rge r
353 qpart lt x ys rlt (y:rge) r
355 -- rqsort is as qsort but anti-stable, i.e. reverses equal elements
357 rqsort lt [x] r = x:r
358 rqsort lt (x:xs) r = rqpart lt x xs [] [] r
360 rqpart lt x [] rle rgt r =
361 qsort lt rle (x : qsort lt rgt r)
363 rqpart lt x (y:ys) rle rgt r =
366 rqpart lt x ys rle (y:rgt) r
369 rqpart lt x ys (y:rle) rgt r
372 %************************************************************************
374 \subsubsection[Utils-dull-mergesort]{A rather dull mergesort}
376 %************************************************************************
380 mergesort :: (a -> a -> Ordering) -> [a] -> [a]
382 mergesort cmp xs = merge_lists (split_into_runs [] xs)
384 a `le` b = case cmp a b of { LT -> True; EQ -> True; GT -> False }
385 a `ge` b = case cmp a b of { LT -> False; EQ -> True; GT -> True }
387 split_into_runs [] [] = []
388 split_into_runs run [] = [run]
389 split_into_runs [] (x:xs) = split_into_runs [x] xs
390 split_into_runs [r] (x:xs) | x `ge` r = split_into_runs [r,x] xs
391 split_into_runs rl@(r:rs) (x:xs) | x `le` r = split_into_runs (x:rl) xs
392 | True = rl : (split_into_runs [x] xs)
395 merge_lists (x:xs) = merge x (merge_lists xs)
399 merge xl@(x:xs) yl@(y:ys)
401 EQ -> x : y : (merge xs ys)
402 LT -> x : (merge xs yl)
403 GT -> y : (merge xl ys)
407 %************************************************************************
409 \subsubsection[Utils-Carsten-mergesort]{A mergesort from Carsten}
411 %************************************************************************
414 Date: Mon, 3 May 93 20:45:23 +0200
415 From: Carsten Kehler Holst <kehler@cs.chalmers.se>
416 To: partain@dcs.gla.ac.uk
417 Subject: natural merge sort beats quick sort [ and it is prettier ]
419 Here is a piece of Haskell code that I'm rather fond of. See it as an
420 attempt to get rid of the ridiculous quick-sort routine. group is
421 quite useful by itself I think it was John's idea originally though I
422 believe the lazy version is due to me [surprisingly complicated].
423 gamma [used to be called] is called gamma because I got inspired by
424 the Gamma calculus. It is not very close to the calculus but does
425 behave less sequentially than both foldr and foldl. One could imagine
426 a version of gamma that took a unit element as well thereby avoiding
427 the problem with empty lists.
429 I've tried this code against
431 1) insertion sort - as provided by haskell
432 2) the normal implementation of quick sort
433 3) a deforested version of quick sort due to Jan Sparud
434 4) a super-optimized-quick-sort of Lennart's
436 If the list is partially sorted both merge sort and in particular
437 natural merge sort wins. If the list is random [ average length of
438 rising subsequences = approx 2 ] mergesort still wins and natural
439 merge sort is marginally beaten by Lennart's soqs. The space
440 consumption of merge sort is a bit worse than Lennart's quick sort
441 approx a factor of 2. And a lot worse if Sparud's bug-fix [see his
442 fpca article ] isn't used because of group.
449 group :: (a -> a -> Bool) -> [a] -> [[a]]
452 Date: Mon, 12 Feb 1996 15:09:41 +0000
453 From: Andy Gill <andy@dcs.gla.ac.uk>
455 Here is a `better' definition of group.
458 group p (x:xs) = group' xs x x (x :)
460 group' [] _ _ s = [s []]
461 group' (x:xs) x_min x_max s
462 | not (x `p` x_max) = group' xs x_min x (s . (x :))
463 | x `p` x_min = group' xs x x_max ((x :) . s)
464 | otherwise = s [] : group' xs x x (x :)
466 -- This one works forwards *and* backwards, as well as also being
467 -- faster that the one in Util.lhs.
472 let ((h1:t1):tt1) = group p xs
473 (t,tt) = if null xs then ([],[]) else
474 if x `p` h1 then (h1:t1,tt1) else
479 generalMerge :: (a -> a -> Bool) -> [a] -> [a] -> [a]
480 generalMerge p xs [] = xs
481 generalMerge p [] ys = ys
482 generalMerge p (x:xs) (y:ys) | x `p` y = x : generalMerge p xs (y:ys)
483 | otherwise = y : generalMerge p (x:xs) ys
485 -- gamma is now called balancedFold
487 balancedFold :: (a -> a -> a) -> [a] -> a
488 balancedFold f [] = error "can't reduce an empty list using balancedFold"
489 balancedFold f [x] = x
490 balancedFold f l = balancedFold f (balancedFold' f l)
492 balancedFold' :: (a -> a -> a) -> [a] -> [a]
493 balancedFold' f (x:y:xs) = f x y : balancedFold' f xs
494 balancedFold' f xs = xs
496 generalMergeSort p [] = []
497 generalMergeSort p xs = (balancedFold (generalMerge p) . map (: [])) xs
499 generalNaturalMergeSort p [] = []
500 generalNaturalMergeSort p xs = (balancedFold (generalMerge p) . group p) xs
502 mergeSort, naturalMergeSort :: Ord a => [a] -> [a]
504 mergeSort = generalMergeSort (<=)
505 naturalMergeSort = generalNaturalMergeSort (<=)
507 mergeSortLe le = generalMergeSort le
508 naturalMergeSortLe le = generalNaturalMergeSort le
511 %************************************************************************
513 \subsection[Utils-transitive-closure]{Transitive closure}
515 %************************************************************************
517 This algorithm for transitive closure is straightforward, albeit quadratic.
520 transitiveClosure :: (a -> [a]) -- Successor function
521 -> (a -> a -> Bool) -- Equality predicate
523 -> [a] -- The transitive closure
525 transitiveClosure succ eq xs
529 go done (x:xs) | x `is_in` done = go done xs
530 | otherwise = go (x:done) (succ x ++ xs)
533 x `is_in` (y:ys) | eq x y = True
534 | otherwise = x `is_in` ys
537 %************************************************************************
539 \subsection[Utils-accum]{Accumulating}
541 %************************************************************************
543 @mapAccumL@ behaves like a combination
544 of @map@ and @foldl@;
545 it applies a function to each element of a list, passing an accumulating
546 parameter from left to right, and returning a final value of this
547 accumulator together with the new list.
550 mapAccumL :: (acc -> x -> (acc, y)) -- Function of elt of input list
551 -- and accumulator, returning new
552 -- accumulator and elt of result list
553 -> acc -- Initial accumulator
555 -> (acc, [y]) -- Final accumulator and result list
557 mapAccumL f b [] = (b, [])
558 mapAccumL f b (x:xs) = (b'', x':xs') where
560 (b'', xs') = mapAccumL f b' xs
563 @mapAccumR@ does the same, but working from right to left instead. Its type is
564 the same as @mapAccumL@, though.
567 mapAccumR :: (acc -> x -> (acc, y)) -- Function of elt of input list
568 -- and accumulator, returning new
569 -- accumulator and elt of result list
570 -> acc -- Initial accumulator
572 -> (acc, [y]) -- Final accumulator and result list
574 mapAccumR f b [] = (b, [])
575 mapAccumR f b (x:xs) = (b'', x':xs') where
577 (b', xs') = mapAccumR f b xs
580 Here is the bi-directional version, that works from both left and right.
583 mapAccumB :: (accl -> accr -> x -> (accl, accr,y))
584 -- Function of elt of input list
585 -- and accumulator, returning new
586 -- accumulator and elt of result list
587 -> accl -- Initial accumulator from left
588 -> accr -- Initial accumulator from right
590 -> (accl, accr, [y]) -- Final accumulators and result list
592 mapAccumB f a b [] = (a,b,[])
593 mapAccumB f a b (x:xs) = (a'',b'',y:ys)
595 (a',b'',y) = f a b' x
596 (a'',b',ys) = mapAccumB f a' b xs
599 A combination of foldl with zip. It works with equal length lists.
602 foldl2 :: (acc -> a -> b -> acc) -> acc -> [a] -> [b] -> acc
604 foldl2 k z (a:as) (b:bs) = foldl2 k (k z a b) as bs
607 Count the number of times a predicate is true
610 count :: (a -> Bool) -> [a] -> Int
612 count p (x:xs) | p x = 1 + count p xs
613 | otherwise = count p xs
617 %************************************************************************
619 \subsection[Utils-comparison]{Comparisons}
621 %************************************************************************
624 thenCmp :: Ordering -> Ordering -> Ordering
625 {-# INLINE thenCmp #-}
627 thenCmp other any = other
629 cmpList :: (a -> a -> Ordering) -> [a] -> [a] -> Ordering
630 -- `cmpList' uses a user-specified comparer
632 cmpList cmp [] [] = EQ
633 cmpList cmp [] _ = LT
634 cmpList cmp _ [] = GT
635 cmpList cmp (a:as) (b:bs)
636 = case cmp a b of { EQ -> cmpList cmp as bs; xxx -> xxx }
640 prefixMatch :: Eq a => [a] -> [a] -> Bool
641 prefixMatch [] _str = True
642 prefixMatch _pat [] = False
643 prefixMatch (p:ps) (s:ss) | p == s = prefixMatch ps ss
646 postfixMatch :: Eq a => [a] -> [a] -> Bool
647 postfixMatch pat str = prefixMatch (reverse pat) (reverse str)
650 %************************************************************************
652 \subsection[Utils-pairs]{Pairs}
654 %************************************************************************
656 The following are curried versions of @fst@ and @snd@.
659 cfst :: a -> b -> a -- stranal-sem only (Note)
663 The following provide us higher order functions that, when applied
664 to a function, operate on pairs.
667 applyToPair :: ((a -> c),(b -> d)) -> (a,b) -> (c,d)
668 applyToPair (f,g) (x,y) = (f x, g y)
670 applyToFst :: (a -> c) -> (a,b)-> (c,b)
671 applyToFst f (x,y) = (f x,y)
673 applyToSnd :: (b -> d) -> (a,b) -> (a,d)
674 applyToSnd f (x,y) = (x,f y)
676 foldPair :: (a->a->a,b->b->b) -> (a,b) -> [(a,b)] -> (a,b)
677 foldPair fg ab [] = ab
678 foldPair fg@(f,g) ab ((a,b):abs) = (f a u,g b v)
679 where (u,v) = foldPair fg ab abs
683 unzipWith :: (a -> b -> c) -> [(a, b)] -> [c]
684 unzipWith f pairs = map ( \ (a, b) -> f a b ) pairs
689 seqList :: [a] -> b -> b
691 seqList :: (Eval a) => [a] -> b -> b
694 seqList (x:xs) b = x `seq` seqList xs b
696 #if __HASKELL1__ <= 4
697 ($!) :: (Eval a) => (a -> b) -> a -> b
703 #if __GLASGOW_HASKELL__ < 402
704 bracket :: IO a -> (a -> IO b) -> (a -> IO c) -> IO c
705 bracket before after thing = do
707 r <- (thing a) `catch` (\err -> after a >> fail err)
716 global :: a -> IORef a
717 global a = unsafePerformIO (newIORef a)
723 #if __GLASGOW_HASKELL__ <= 408
725 ioErrors = justIoErrors
726 throwTo = raiseInThread
729 #ifdef mingw32_TARGET_OS
730 foreign import "_getpid" myGetProcessID :: IO Int
732 myGetProcessID :: IO Int
733 myGetProcessID = Posix.getProcessID