hmt-0.16: Haskell Music Theory

Safe HaskellSafe
LanguageHaskell98

Music.Theory.List

Contents

Description

List functions.

Synopsis

Documentation

slice :: Int -> Int -> [a] -> [a] Source #

Data.Vector.slice, ie. starting index (zero-indexed) and number of elements.

slice 4 5 [1..] == [5,6,7,8,9]

section :: Int -> Int -> [a] -> [a] Source #

Variant of slice with start and end indices (zero-indexed).

section 4 8 [1..] == [5,6,7,8,9]

bracket :: (a, a) -> [a] -> [a] Source #

Bracket sequence with left and right values.

bracket ('<','>') "1,2,3" == "<1,2,3>"

unbracket' :: [a] -> (Maybe a, [a], Maybe a) Source #

unbracket :: [t] -> Maybe (t, [t], t) Source #

The first & middle & last elements of a list.

unbracket "[12]" == Just ('[',"12",']')

unbracket_err :: [t] -> (t, [t], t) Source #

bracket_l :: ([a], [a]) -> [a] -> [a] Source #

Variant where brackets are sequences.

bracket_l ("<:",":>") "1,2,3" == "<:1,2,3:>"

Split

separate_at :: Eq a => [a] -> [a] -> Maybe ([a], [a]) Source #

Relative of splitOn, but only makes first separation.

splitOn "//" "lhs//rhs//rem" == ["lhs","rhs","rem"]
separate_at "//" "lhs//rhs//rem" == Just ("lhs","rhs//rem")

on_elem :: Eq a => a -> Splitter a Source #

Splitter comparing single element.

split_before :: Eq a => a -> [a] -> [[a]] Source #

Split before the indicated element.

split_before 'x' "axbcxdefx" == ["a","xbc","xdef","x"]
split_before 'x' "xa" == ["","xa"]
map (flip split_before "abcde") "ae_" == [["","abcde"],["abcd","e"],["abcde"]]
map (flip break "abcde" . (==)) "ae_" == [("","abcde"),("abcd","e"),("abcde","")]

Rotate

genericRotate_left :: Integral i => i -> [a] -> [a] Source #

Generic form of rotate_left.

rotate_left :: Int -> [a] -> [a] Source #

Left rotation.

rotate_left 1 [1..3] == [2,3,1]
rotate_left 3 [1..5] == [4,5,1,2,3]

genericRotate_right :: Integral n => n -> [a] -> [a] Source #

Generic form of rotate_right.

rotate_right :: Int -> [a] -> [a] Source #

Right rotation.

rotate_right 1 [1..3] == [3,1,2]

rotate :: Integral n => n -> [a] -> [a] Source #

Rotate left by n mod #p places.

rotate 1 [1..3] == [2,3,1]
rotate 8 [1..5] == [4,5,1,2,3]

rotate_r :: Integral n => n -> [a] -> [a] Source #

Rotate right by n places.

rotate_r 8 [1..5] == [3,4,5,1,2]

rotations :: [a] -> [[a]] Source #

All rotations.

rotations [0,1,3] == [[0,1,3],[1,3,0],[3,0,1]]

rotate_starting_from :: Eq a => a -> [a] -> Maybe [a] Source #

Rotate list so that is starts at indicated element.

rotate_starting_from 'c' "abcde" == Just "cdeab"
rotate_starting_from '_' "abc" == Nothing

rotate_starting_from_err :: Eq a => a -> [a] -> [a] Source #

Erroring variant.

adj :: Int -> Int -> [a] -> [[a]] Source #

Sequence of n adjacent elements, moving forward by k places. The last element may have fewer than n places, but will reach the end of the input sequence.

adj 3 2 "adjacent" == ["adj","jac","cen","nt"]

adj' :: Int -> Int -> [a] -> [[a]] Source #

Variant of adj where the last element has n places but may not reach the end of the input sequence.

adj' 3 2 "adjacent" == ["adj","jac","cen"]

genericAdj2 :: Integral n => n -> [t] -> [(t, t)] Source #

Generic form of adj2.

adj2 :: Int -> [t] -> [(t, t)] Source #

Adjacent elements of list, at indicated distance, as pairs.

adj2 1 [1..5] == [(1,2),(2,3),(3,4),(4,5)]
let l = [1..5] in zip l (tail l) == adj2 1 l
adj2 2 [1..4] == [(1,2),(3,4)]
adj2 3 [1..5] == [(1,2),(4,5)]

close :: [a] -> [a] Source #

Append first element to end of list.

close [1..3] == [1,2,3,1]

adj2_cyclic :: Int -> [t] -> [(t, t)] Source #

adj2 . close.

adj2_cyclic 1 [1..3] == [(1,2),(2,3),(3,1)]

interleave :: [a] -> [a] -> [a] Source #

Interleave elements of p and q.

interleave [1..3] [4..6] == [1,4,2,5,3,6]
interleave ".+-" "abc" == ".a+b-c"
interleave [1..3] [] == []

interleave_set :: [[a]] -> [a] Source #

Interleave list of lists. Allows lists to be of non-equal lenghts.

interleave_set ["abcd","efgh","ijkl"] == "aeibfjcgkdhl"
interleave_set ["abc","defg","hijkl"] == "adhbeicfjgkl"

deinterleave :: Int -> [a] -> [[a]] Source #

De-interleave n lists.

deinterleave 2 ".a+b-c" == [".+-","abc"]
deinterleave 3 "aeibfjcgkdhl" == ["abcd","efgh","ijkl"]

deinterleave2 :: [t] -> ([t], [t]) Source #

Special case for two-part deinterleaving.

deinterleave2 ".a+b-c" == (".+-","abc")

interleave_continue :: [a] -> [a] -> [a] Source #

Variant that continues with the longer input.

interleave_continue ".+-" "abc" == ".a+b-c"
interleave_continue [1..3] [] == [1..3]
interleave_continue [] [1..3] == [1..3]

interleave_rotations :: Int -> Int -> [b] -> [b] Source #

interleave of rotate_left by i and j.

interleave_rotations 9 3 [1..13] == [10,4,11,5,12,6,13,7,1,8,2,9,3,10,4,11,5,12,6,13,7,1,8,2,9,3]

generic_histogram :: (Ord a, Integral i) => [a] -> [(a, i)] Source #

histogram_by :: Ord a => (a -> a -> Bool) -> [a] -> [(a, Int)] Source #

histogram :: Ord a => [a] -> [(a, Int)] Source #

Count occurences of elements in list.

map histogram ["","hohoh"] == [[],[('h',3),('o',2)]]

duplicates_by :: Ord a => (a -> a -> Bool) -> [a] -> [a] Source #

duplicates :: Ord a => [a] -> [a] Source #

Elements that appear more than once in the input.

map duplicates ["duplicates","redundant"] == ["","dn"]

segments :: Int -> Int -> [a] -> [[a]] Source #

List segments of length i at distance j.

segments 2 1 [1..5] == [[1,2],[2,3],[3,4],[4,5]]
segments 2 2 [1..5] == [[1,2],[3,4]]

intersect_l :: Eq a => [[a]] -> [a] Source #

foldl1 intersect.

intersect_l [[1,2],[1,2,3],[1,2,3,4]] == [1,2]

union_l :: Eq a => [[a]] -> [a] Source #

foldl1 union.

sort (union_l [[1,3],[2,3],[3]]) == [1,2,3]

adj_intersect :: Eq a => Int -> [[a]] -> [[a]] Source #

Intersection of adjacent elements of list at distance n.

adj_intersect 1 [[1,2],[1,2,3],[1,2,3,4]] == [[1,2],[1,2,3]]

cycles :: Int -> [a] -> [[a]] Source #

List of cycles at distance n.

cycles 2 [1..6] == [[1,3,5],[2,4,6]]
cycles 3 [1..9] == [[1,4,7],[2,5,8],[3,6,9]]
cycles 4 [1..8] == [[1,5],[2,6],[3,7],[4,8]]

filter_halt :: (a -> Bool) -> (a -> Bool) -> [a] -> [a] Source #

Variant of filter that has a predicate to halt processing, ie. filter of takeWhile.

filter_halt (even . fst) ((< 5) . snd) (zip [1..] [0..])

replace :: Eq a => [a] -> [a] -> [a] -> [a] Source #

Replace all p with q in s.

replace "_x_" "-X-" "an _x_ string" == "an -X- string"
replace "ab" "cd" "ab ab cd ab" == "cd cd cd cd"

replace_at :: Integral i => [a] -> i -> a -> [a] Source #

Replace the ith value at ns with x.

replace_at "test" 2 'n' == "tent"

Association lists

group_on :: Eq x => (a -> x) -> [a] -> [[a]] Source #

Equivalent to groupBy == on f.

let r = [[(1,'a'),(1,'b')],[(2,'c')],[(3,'d'),(3,'e')],[(4,'f')]]
in group_on fst (zip [1,1,2,3,3,4] "abcdef") == r

collate_on_adjacent :: (Eq k, Ord k) => (a -> k) -> (a -> v) -> [a] -> [(k, [v])] Source #

Given accesors for key and value collate adjacent values.

collate_adjacent :: Ord a => [(a, b)] -> [(a, [b])] Source #

collate_on_adjacent of fst and snd.

collate_adjacent (zip "TDD" "xyz") == [('T',"x"),('D',"yz")]

collate_on :: Ord k => (a -> k) -> (a -> v) -> [a] -> [(k, [v])] Source #

sortOn prior to collate_on_adjacent.

let r = [('A',"a"),('B',"bd"),('C',"ce"),('D',"f")]
in collate_on fst snd (zip "ABCBCD" "abcdef") == r

collate :: Ord a => [(a, b)] -> [(a, [b])] Source #

collate_on of fst and snd.

collate (zip "TDD" "xyz") == [('D',"yz"),('T',"x")]
collate (zip [1,2,1] "abc") == [(1,"ac"),(2,"b")]

uncollate :: [(k, [v])] -> [(k, v)] Source #

Reverse of collate, inverse if order is not considered.

uncollate [(1,"ac"),(2,"b")] == zip [1,1,2] "acb"

with_key :: k -> [v] -> [(k, v)] Source #

Make assoc list with given key.

with_key 'a' [1..3] == [('a',1),('a',2),('a',3)]

dx_d :: Num a => a -> [a] -> [a] Source #

Intervals to values, zero is n.

dx_d 5 [1,2,3] == [5,6,8,11]

dx_d' :: Num t => t -> [t] -> (t, [t]) Source #

Variant that takes initial value and separates final value. This is an appropriate function for mapAccumL.

dx_d' 5 [1,2,3] == (11,[5,6,8])
dx_d' 0 [1,1,1] == (3,[0,1,2])

d_dx_by :: (t -> t -> u) -> [t] -> [u] Source #

Apply flip of f between elements of l.

d_dx_by (,) "abcd" == [('b','a'),('c','b'),('d','c')]

d_dx :: Num a => [a] -> [a] Source #

Integrate, d_dx_by -, ie. pitch class segment to interval sequence.

d_dx [5,6,8,11] == [1,2,3]
d_dx [] == []

difference :: Eq a => [a] -> [a] -> [a] Source #

Elements of p not in q.

[1,2,3] `difference` [1,2] == [3]

is_subset :: Eq a => [a] -> [a] -> Bool Source #

Is p a subset of q, ie. is intersect of p and q == p.

is_subset [1,2] [1,2,3] == True

is_superset :: Eq a => [a] -> [a] -> Bool Source #

Is p a superset of q, ie. flip is_subset.

is_superset [1,2,3] [1,2] == True

subsequence :: Eq a => [a] -> [a] -> Bool Source #

Is p a subsequence of q, ie. synonym for isInfixOf.

subsequence [1,2] [1,2,3] == True

elem_index_unique :: Eq a => a -> [a] -> Int Source #

Variant of elemIndices that requires e to be unique in p.

elem_index_unique 'a' "abcda" == undefined

lookup_err_msg :: (Eq k, Show k) => String -> k -> [(k, v)] -> v Source #

Lookup that errors and prints message.

lookup_err :: Eq k => k -> [(k, v)] -> v Source #

Error variant.

lookup_def :: Eq k => k -> v -> [(k, v)] -> v Source #

lookup variant with default value.

reverse_lookup :: Eq b => b -> [(a, b)] -> Maybe a Source #

Reverse lookup.

reverse_lookup 'c' [] == Nothing
reverse_lookup 'c' (zip [0..4] ['a'..]) == Just 2

find_bounds' :: (t -> s -> Ordering) -> [(t, t)] -> s -> Either ((t, t), Ordering) (t, t) Source #

Basis of find_bounds_scl, indicates if x is to the left or right of the list, and it to the right whether equal or not. Right values will be correct if the list is not ascending, however Left values only make sense for ascending ranges.

map (find_bounds' compare [(0,1),(1,2)]) [-1,0,1,2,3]

decide_nearest' :: Ord o => (p -> o) -> (p, p) -> p Source #

decide_nearest :: (Num o, Ord o) => o -> (o, o) -> o Source #

Decide if value is nearer the left or right value of a range.

find_nearest_err :: (Num n, Ord n) => [n] -> n -> n Source #

Find the number that is nearest the requested value in an ascending list of numbers.

map (find_nearest_err [0,3.5,4,7]) [-1,1,3,5,7,9] == [0,0,3.5,4,7,7]

find_nearest :: (Num n, Ord n) => [n] -> n -> Maybe n Source #

find_bounds_scl :: Bool -> (t -> s -> Ordering) -> [(t, t)] -> s -> Maybe (t, t) Source #

Basis of find_bounds. There is an option to consider the last element specially, and if equal to the last span is given.

find_bounds :: Bool -> (t -> s -> Ordering) -> [t] -> s -> Maybe (t, t) Source #

Find adjacent elements of list that bound element under given comparator.

let {f = find_bounds True compare [1..5]
    ;r = [Nothing,Just (1,2),Just (3,4),Just (4,5)]}
in map f [0,1,3.5,5] == r

drop_last :: [t] -> [t] Source #

Special case of dropRight.

map drop_last ["","?","remove"] == ["","","remov"]

dropRight :: Int -> [a] -> [a] Source #

Variant of drop from right of list.

dropRight 1 [1..9] == [1..8]

dropWhileRight :: (a -> Bool) -> [a] -> [a] Source #

Variant of dropWhile from right of list.

dropWhileRight Data.Char.isDigit "A440" == "A"

take_right :: Int -> [a] -> [a] Source #

take from right.

take_right 3 "taking" == "ing"

take_while_right :: (a -> Bool) -> [a] -> [a] Source #

takeWhile from right.

take_while_right Data.Char.isDigit "A440" == "440"

at_head :: (a -> b) -> (a -> b) -> [a] -> [b] Source #

Apply f at first element, and g at all other elements.

at_head negate id [1..5] == [-1,2,3,4,5]

at_last :: (a -> b) -> (a -> b) -> [a] -> [b] Source #

Apply f at all but last element, and g at last element.

at_last (* 2) negate [1..4] == [2,4,6,-4]

separate_last' :: [a] -> ([a], Maybe a) Source #

Separate list into an initial list and perhaps the last element tuple.

separate_last' [] == ([],Nothing)

separate_last :: [a] -> ([a], a) Source #

Error on null input.

separate_last [1..5] == ([1..4],5)

indicate_repetitions :: Eq a => [a] -> [Maybe a] Source #

Replace directly repeated elements with Nothing.

indicate_repetitions "abba" == [Just 'a',Just 'b',Nothing,Just 'a']

zip_with_adj :: (a -> a -> b) -> [a] -> [b] Source #

zipWith of list and it's own tail.

zip_with_adj (,) "abcde" == [('a','b'),('b','c'),('c','d'),('d','e')]

compare_adjacent_by :: (a -> a -> Ordering) -> [a] -> [Ordering] Source #

Type-specialised zip_with_adj.

compare_adjacent :: Ord a => [a] -> [Ordering] Source #

compare_adjacent_by of compare.

compare_adjacent [0,1,3,2] == [LT,LT,GT]

adjacent_groupBy :: (a -> a -> Bool) -> [a] -> [[a]] Source #

groupBy does not make adjacent comparisons, it compares each new element to the start of the group. This function is the adjacent variant.

groupBy (<) [1,2,3,2,4,1,5,9] == [[1,2,3,2,4],[1,5,9]]
adjacent_groupBy (<) [1,2,3,2,4,1,5,9] == [[1,2,3],[2,4],[1,5,9]]

group_ranges :: (Num t, Eq t) => [t] -> [(t, t)] Source #

Reduce sequences of consecutive values to ranges.

group_ranges [-1,0,3,4,5,8,9,12] == [(-1,0),(3,5),(8,9),(12,12)]
group_ranges [3,2,3,4,3] == [(3,3),(2,4),(3,3)]

group_just :: [Maybe a] -> [[Maybe a]] Source #

groupBy on structure of Maybe, ie. all Just compare equal.

let r = [[Just 1],[Nothing,Nothing],[Just 4,Just 5]]
in group_just [Just 1,Nothing,Nothing,Just 4,Just 5] == r

all_equal :: Eq a => [a] -> Bool Source #

Predicate to determine if all elements of the list are ==.

all_equal "aaa" == True

all_eq :: Eq n => [n] -> Bool Source #

Variant using nub.

sort_group_on :: Ord b => (a -> b) -> [a] -> [[a]] Source #

group_on of sortOn.

let r = [[('1','a'),('1','c')],[('2','d')],[('3','b'),('3','e')]]
in sort_group_on fst (zip "13123" "abcde") == r

mcons :: Maybe a -> [a] -> [a] Source #

Maybe cons element onto list.

Nothing `mcons` "something" == "something"
Just 's' `mcons` "omething" == "something"

Ordering

type Compare_F a = a -> a -> Ordering Source #

Comparison function type.

two_stage_compare :: Compare_F a -> Compare_F a -> Compare_F a Source #

If f compares EQ, defer to g.

n_stage_compare :: [Compare_F a] -> Compare_F a Source #

Sequence of comparison functions, continue comparing until not EQ.

compare (1,0) (0,1) == GT
n_stage_compare [compare `on` snd,compare `on` fst] (1,0) (0,1) == LT

sort_to :: Ord i => [e] -> [i] -> [e] Source #

Sort sequence a based on ordering of sequence b.

sort_to "abc" [1,3,2] == "acb"
sort_to "adbce" [1,4,2,3,5] == "abcde"

sort_on :: Ord i => [i] -> [e] -> [e] Source #

flip of sort_to.

sort_on [1,4,2,3,5] "adbce" == "abcde"

sort_by_two_stage :: (Ord b, Ord c) => (a -> b) -> (a -> c) -> [a] -> [a] Source #

sort_by_n_stage :: Ord b => [a -> b] -> [a] -> [a] Source #

merge_by :: Compare_F a -> [a] -> [a] -> [a] Source #

Given a comparison function, merge two ascending lists.

mergeBy compare [1,3,5] [2,4] == [1..5]

merge_on :: Ord x => (a -> x) -> [a] -> [a] -> [a] Source #

merge_by_two_stage :: Ord b => (a -> b) -> Compare_F c -> (a -> c) -> [a] -> [a] -> [a] Source #

merge :: Ord a => [a] -> [a] -> [a] Source #

mergeBy compare.

merge_set_by :: (a -> a -> Ordering) -> [[a]] -> [a] Source #

Merge list of sorted lists given comparison function. Note that this is not equal to mergeAll.

merge_set :: Ord a => [[a]] -> [a] Source #

merge_set_by of compare.

merge_set [[1,3,5,7,9],[2,4,6,8],[10]] == [1..10]

merge_by_resolve :: (a -> a -> a) -> Compare_F a -> [a] -> [a] -> [a] Source #

merge_by variant that joins (resolves) equal elements.

let {left p _ = p
    ;right _ q = q
    ;cmp = compare `on` fst
    ;p = zip [1,3,5] "abc"
    ;q = zip [1,2,3] "ABC"
    ;left_r = [(1,'a'),(2,'B'),(3,'b'),(5,'c')]
    ;right_r = [(1,'A'),(2,'B'),(3,'C'),(5,'c')]}
in merge_by_resolve left cmp p q == left_r &&
   merge_by_resolve right cmp p q == right_r

find_non_ascending :: (a -> a -> Ordering) -> [a] -> Maybe (a, a) Source #

First non-ascending pair of elements.

elem_ordered :: Ord t => t -> [t] -> Bool Source #

Variant of elem that operates on a sorted list, halting. This is member.

16 `elem_ordered` [1,3 ..] == False
16 `elem` [1,3 ..] == undefined

elemIndex_ordered :: Ord t => t -> [t] -> Maybe Int Source #

Variant of elemIndex that operates on a sorted list, halting.

16 `elemIndex_ordered` [1,3 ..] == Nothing
16 `elemIndex_ordered` [0,1,4,9,16,25,36,49,64,81,100] == Just 4

zip_with_kr :: (a -> b -> c) -> [a] -> [b] -> ([c], [b]) Source #

Keep right variant of zipWith, where unused rhs values are returned.

zip_with_kr (,) [1..3] ['a'..'e'] == ([(1,'a'),(2,'b'),(3,'c')],"de")

zip_with_perhaps_rhs :: (a -> b -> Either c c) -> (a -> c) -> [a] -> [b] -> [c] Source #

A zipWith variant that always consumes an element from the left hand side (lhs), but only consumes an element from the right hand side (rhs) if the zip function is Right and not if Left. There's also a secondary function to continue if the rhs ends before the lhs.

fill_gaps_ascending :: (Enum n, Ord n) => t -> (n, n) -> [(n, t)] -> [(n, t)] Source #

Fill gaps in a sorted association list, range is inclusive at both ends.

let r = [(1,'a'),(2,'x'),(3,'x'),(4,'x'),(5,'b'),(6,'x'),(7,'c'),(8,'x'),(9,'x')]
in fill_gaps_ascending' 'x' (1,9) (zip [1,5,7] "abc") == r

fill_gaps_ascending' :: (Num n, Enum n, Ord n) => t -> (n, n) -> [(n, t)] -> [(n, t)] Source #

Direct definition.

minmax :: Ord t => [t] -> (t, t) Source #

minimum and maximum in one pass.

minmax "minimumandmaximum" == ('a','x')

Bimap

bimap1 :: (t -> u) -> (t, t) -> (u, u) Source #

Apply f to both elements of a two-tuple, ie. bimap f f.

pad_right :: a -> Int -> [a] -> [a] Source #

Append k to the right of l until result has n places.

map (pad_right '0' 2 . return) ['0' .. '9']
pad_right '0' 12 "1101" == "110100000000"
map (pad_right ' '3) ["S","E-L"] == ["S  ","E-L"]

pad_left :: a -> Int -> [a] -> [a] Source #

Append k to the left of l until result has n places.

map (pad_left '0' 2 . return) ['0' .. '9']

Embedding

embedding :: Eq t => ([t], [t]) -> Either [Int] [Int] Source #

Locate first (leftmost) embedding of q in p. Return partial indices for failure at Left.

embedding ("embedding","ming") == Right [1,6,7,8]
embedding ("embedding","mind") == Left [1,6,7]

embedding_err :: Eq t => ([t], [t]) -> [Int] Source #

is_embedding :: Eq t => [t] -> [t] -> Bool Source #

Does q occur in sequence, though not necessarily adjacently, in p.

is_embedding [1 .. 9] [1,3,7] == True
is_embedding "embedding" "ming" == True
is_embedding "embedding" "mind" == False

all_embeddings_m :: (Eq t, MonadLogic m) => [t] -> [t] -> m [Int] Source #

all_embeddings :: Eq t => [t] -> [t] -> [[Int]] Source #

Enumerate indices for all embeddings of q in p.

all_embeddings "all_embeddings" "leg" == [[1,4,12],[1,7,12],[2,4,12],[2,7,12]]

Un-list

unlist1 :: [t] -> Maybe t Source #

Unpack one element list.

unlist1_err :: [t] -> t Source #

Erroring variant.

Traversable

adopt_shape :: Traversable t => (a -> b -> c) -> [b] -> t a -> t c Source #

Replace elements at Traversable with result of joining with elements from list.

let t = Node 0 [Node 1 [Node 2 [],Node 3 []],Node 4 []]
putStrLn $ drawTree (fmap show t)
let u = (adopt_shape (\_ x -> x) "abcde" t)
putStrLn $ drawTree (fmap return u)

adopt_shape_m :: Traversable t => (a -> b -> c) -> [b] -> t (Maybe a) -> t (Maybe c) Source #

Variant of adopt_shape that considers only Just elements at Traversable.

let {s = "a(b(cd)ef)ghi"
    ;t = group_tree (begin_end_cmp_eq '(' ')') s}
in adopt_shape_m (,) [1..13] t

Tree

group_tree :: (a -> Bool, a -> Bool) -> [a] -> Tree (Maybe a) Source #

Given an Ordering predicate where LT opens a group, GT closes a group, and EQ continues current group, construct tree from list.

let {l = "a {b {c d} e f} g h i"
    ;t = group_tree ((==) '{',(==) '}') l}
in catMaybes (flatten t) == l
let {d = putStrLn . drawTree . fmap show}
in d (group_tree ((==) '(',(==) ')') "a(b(cd)ef)ghi")

Indexing

remove_ix :: Int -> [a] -> [a] Source #

Remove element at index.

remove_ix 5 "remove" == "remov"
remove_ix 5 "short" == undefined

operate_ixs :: Bool -> [Int] -> [a] -> [a] Source #

select_ixs :: [Int] -> [a] -> [a] Source #

remove_ixs :: [Int] -> [a] -> [a] Source #

replace_ix :: (a -> a) -> Int -> [a] -> [a] Source #

Replace element at i in p by application of f.

replace_ix negate 1 [1..3] == [1,-2,3]

at_cyclic :: [a] -> Int -> a Source #

Cyclic indexing function.

map (at_cyclic "cycle") [0..9] == "cyclecycle"