shellcheck/src/ShellCheck/Fixer.hs

411 lines
13 KiB
Haskell

{-
Copyright 2018-2019 Vidar Holen, Ng Zhi An
This file is part of ShellCheck.
https://www.shellcheck.net
ShellCheck is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
ShellCheck is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
-}
{-# LANGUAGE TemplateHaskell #-}
module ShellCheck.Fixer (applyFix, removeTabStops, mapPositions, Ranged(..), runTests) where
import ShellCheck.Interface
import ShellCheck.Prelude
import Control.Monad.State
import Data.Array
import Data.List
import Data.Semigroup
import GHC.Exts (sortWith)
import Test.QuickCheck
-- The Ranged class is used for types that has a start and end position.
class Ranged a where
start :: a -> Position
end :: a -> Position
overlap :: a -> a -> Bool
overlap x y =
(yStart >= xStart && yStart < xEnd) || (yStart < xStart && yEnd > xStart)
where
yStart = start y
yEnd = end y
xStart = start x
xEnd = end x
-- Set a new start and end position on a Ranged
setRange :: (Position, Position) -> a -> a
-- Tests auto-verify that overlap commutes
assertOverlap x y = overlap x y && overlap y x
assertNoOverlap x y = not (overlap x y) && not (overlap y x)
prop_overlap_contiguous = assertNoOverlap
(tFromStart 10 12 "foo" 1)
(tFromStart 12 14 "bar" 2)
prop_overlap_adjacent_zerowidth = assertNoOverlap
(tFromStart 3 3 "foo" 1)
(tFromStart 3 3 "bar" 2)
prop_overlap_enclosed = assertOverlap
(tFromStart 3 5 "foo" 1)
(tFromStart 1 10 "bar" 2)
prop_overlap_partial = assertOverlap
(tFromStart 1 5 "foo" 1)
(tFromStart 3 7 "bar" 2)
instance Ranged PositionedComment where
start = pcStartPos
end = pcEndPos
setRange (s, e) pc = pc {
pcStartPos = s,
pcEndPos = e
}
instance Ranged Replacement where
start = repStartPos
end = repEndPos
setRange (s, e) r = r {
repStartPos = s,
repEndPos = e
}
-- The Monoid instance for Fix merges fixes that do not conflict.
-- TODO: Make an efficient 'mconcat'
instance Monoid Fix where
mempty = newFix
mappend = (<>)
instance Semigroup Fix where
f1 <> f2 =
-- FIXME: This might need to also discard adjacent zero-width ranges for
-- when two fixes change the same AST node, e.g. `foo` -> "$(foo)"
if or [ r2 `overlap` r1 | r1 <- fixReplacements f1, r2 <- fixReplacements f2 ]
then f1
else newFix {
fixReplacements = fixReplacements f1 ++ fixReplacements f2
}
-- Conveniently apply a transformation to positions in a Fix
mapPositions :: (Position -> Position) -> Fix -> Fix
mapPositions f = adjustFix
where
adjustReplacement rep =
rep {
repStartPos = f $ repStartPos rep,
repEndPos = f $ repEndPos rep
}
adjustFix fix =
fix {
fixReplacements = map adjustReplacement $ fixReplacements fix
}
-- Rewrite a Ranged from a tabstop of 8 to 1
removeTabStops :: Ranged a => a -> Array Int String -> a
removeTabStops range ls =
let startColumn = realignColumn lineNo colNo range
endColumn = realignColumn endLineNo endColNo range
startPosition = (start range) { posColumn = startColumn }
endPosition = (end range) { posColumn = endColumn } in
setRange (startPosition, endPosition) range
where
realignColumn lineNo colNo c =
if lineNo c > 0 && lineNo c <= fromIntegral (length ls)
then real (ls ! fromIntegral (lineNo c)) 0 0 (colNo c)
else colNo c
real _ r v target | target <= v = r
-- hit this case at the end of line, and if we don't hit the target
-- return real + (target - v)
real [] r v target = r + (target - v)
real ('\t':rest) r v target = real rest (r+1) (v + 8 - (v `mod` 8)) target
real (_:rest) r v target = real rest (r+1) (v+1) target
lineNo = posLine . start
endLineNo = posLine . end
colNo = posColumn . start
endColNo = posColumn . end
-- A replacement that spans multiple line is applied by:
-- 1. merging the affected lines into a single string using `unlines`
-- 2. apply the replacement as if it only spanned a single line
-- The tricky part is adjusting the end column of the replacement
-- (the end line doesn't matter because there is only one line)
--
-- aaS <--- start of replacement (row 1 column 3)
-- bbbb
-- cEc
-- \------- end of replacement (row 3 column 2)
--
-- a flattened string will look like:
--
-- "aaS\nbbbb\ncEc\n"
--
-- The column of E has to be adjusted by:
-- 1. lengths of lines to be replaced, except the end row itself
-- 2. end column of the replacement
-- 3. number of '\n' by `unlines`
multiToSingleLine :: [Fix] -> Array Int String -> ([Fix], String)
multiToSingleLine fixes lines =
(map (mapPositions adjust) fixes, unlines $ elems lines)
where
-- A prefix sum tree from line number to column shift.
-- FIXME: The tree will be totally unbalanced.
shiftTree :: PSTree Int
shiftTree =
foldl (\t (n,s) -> addPSValue (n+1) (length s + 1) t) newPSTree $
assocs lines
singleString = unlines $ elems lines
adjust pos =
pos {
posLine = 1,
posColumn = (posColumn pos) +
(fromIntegral $ getPrefixSum (fromIntegral $ posLine pos) shiftTree)
}
-- Apply a fix and return resulting lines.
-- The number of lines can increase or decrease with no obvious mapping back, so
-- the function does not return an array.
applyFix :: Fix -> Array Int String -> [String]
applyFix fix fileLines =
let
untabbed = fix {
fixReplacements =
map (\c -> removeTabStops c fileLines) $
fixReplacements fix
}
(adjustedFixes, singleLine) = multiToSingleLine [untabbed] fileLines
in
lines . runFixer $ applyFixes2 adjustedFixes singleLine
-- start and end comes from pos, which is 1 based
prop_doReplace1 = doReplace 0 0 "1234" "A" == "A1234" -- technically not valid
prop_doReplace2 = doReplace 1 1 "1234" "A" == "A1234"
prop_doReplace3 = doReplace 1 2 "1234" "A" == "A234"
prop_doReplace4 = doReplace 3 3 "1234" "A" == "12A34"
prop_doReplace5 = doReplace 4 4 "1234" "A" == "123A4"
prop_doReplace6 = doReplace 5 5 "1234" "A" == "1234A"
doReplace start end o r =
let si = fromIntegral (start-1)
ei = fromIntegral (end-1)
(x, xs) = splitAt si o
z = drop (ei - si) xs
in
x ++ r ++ z
-- Fail if the 'expected' string is not result when applying 'fixes' to 'original'.
testFixes :: String -> String -> [Fix] -> Bool
testFixes expected original fixes =
actual == expected
where
actual = runFixer (applyFixes2 fixes original)
-- A Fixer allows doing repeated modifications of a string where each
-- replacement automatically accounts for shifts from previous ones.
type Fixer a = State (PSTree Int) a
-- Apply a single replacement using its indices into the original string.
-- It does not handle multiple lines, all line indices must be 1.
applyReplacement2 :: Replacement -> String -> Fixer String
applyReplacement2 rep string = do
tree <- get
let transform pos = pos + getPrefixSum pos tree
let originalPos = (repStartPos rep, repEndPos rep)
(oldStart, oldEnd) = tmap (fromInteger . posColumn) originalPos
(newStart, newEnd) = tmap transform (oldStart, oldEnd)
let (l1, l2) = tmap posLine originalPos in
when (l1 /= 1 || l2 /= 1) $
error $ pleaseReport "bad cross-line fix"
let replacer = repString rep
let shift = (length replacer) - (oldEnd - oldStart)
let insertionPoint =
case repInsertionPoint rep of
InsertBefore -> oldStart
InsertAfter -> oldEnd+1
put $ addPSValue insertionPoint shift tree
return $ doReplace newStart newEnd string replacer
where
tmap f (a,b) = (f a, f b)
-- Apply a list of Replacements in the correct order
applyReplacements2 :: [Replacement] -> String -> Fixer String
applyReplacements2 reps str =
foldM (flip applyReplacement2) str $
reverse $ sortWith repPrecedence reps
-- Apply all fixes with replacements in the correct order
applyFixes2 :: [Fix] -> String -> Fixer String
applyFixes2 fixes = applyReplacements2 (concatMap fixReplacements fixes)
-- Get the final value of a Fixer.
runFixer :: Fixer a -> a
runFixer f = evalState f newPSTree
-- A Prefix Sum Tree that lets you look up the sum of values at and below an index.
-- It's implemented essentially as a Fenwick tree without the bit-based balancing.
-- The last Num is the sum of the left branch plus current element.
data PSTree n = PSBranch n (PSTree n) (PSTree n) n | PSLeaf
deriving (Show)
newPSTree :: Num n => PSTree n
newPSTree = PSLeaf
-- Get the sum of values whose keys are <= 'target'
getPrefixSum :: (Ord n, Num n) => n -> PSTree n -> n
getPrefixSum = f 0
where
f sum _ PSLeaf = sum
f sum target (PSBranch pivot left right cumulative) =
case target `compare` pivot of
LT -> f sum target left
GT -> f (sum+cumulative) target right
EQ -> sum+cumulative
-- Add a value to the Prefix Sum tree at the given index.
-- Values accumulate: addPSValue 42 2 . addPSValue 42 3 == addPSValue 42 5
addPSValue :: (Ord n, Num n) => n -> n -> PSTree n -> PSTree n
addPSValue key value tree = if value == 0 then tree else f tree
where
f PSLeaf = PSBranch key PSLeaf PSLeaf value
f (PSBranch pivot left right sum) =
case key `compare` pivot of
LT -> PSBranch pivot (f left) right (sum + value)
GT -> PSBranch pivot left (f right) sum
EQ -> PSBranch pivot left right (sum + value)
prop_pstreeSumsCorrectly kvs targets =
let
-- Trivial O(n * m) implementation
dumbPrefixSums :: [(Int, Int)] -> [Int] -> [Int]
dumbPrefixSums kvs targets =
let prefixSum target = sum [v | (k,v) <- kvs, k <= target]
in map prefixSum targets
-- PSTree O(n * log m) implementation
smartPrefixSums :: [(Int, Int)] -> [Int] -> [Int]
smartPrefixSums kvs targets =
let tree = foldl (\tree (pos, shift) -> addPSValue pos shift tree) PSLeaf kvs
in map (\x -> getPrefixSum x tree) targets
in smartPrefixSums kvs targets == dumbPrefixSums kvs targets
-- Semi-convenient functions for constructing tests.
testFix :: [Replacement] -> Fix
testFix list = newFix {
fixReplacements = list
}
tFromStart :: Int -> Int -> String -> Int -> Replacement
tFromStart start end repl order =
newReplacement {
repStartPos = newPosition {
posLine = 1,
posColumn = fromIntegral start
},
repEndPos = newPosition {
posLine = 1,
posColumn = fromIntegral end
},
repString = repl,
repPrecedence = order,
repInsertionPoint = InsertAfter
}
tFromEnd start end repl order =
(tFromStart start end repl order) {
repInsertionPoint = InsertBefore
}
prop_simpleFix1 = testFixes "hello world" "hell world" [
testFix [
tFromEnd 5 5 "o" 1
]]
prop_anchorsLeft = testFixes "-->foobar<--" "--><--" [
testFix [
tFromStart 4 4 "foo" 1,
tFromStart 4 4 "bar" 2
]]
prop_anchorsRight = testFixes "-->foobar<--" "--><--" [
testFix [
tFromEnd 4 4 "bar" 1,
tFromEnd 4 4 "foo" 2
]]
prop_anchorsBoth1 = testFixes "-->foobar<--" "--><--" [
testFix [
tFromStart 4 4 "bar" 2,
tFromEnd 4 4 "foo" 1
]]
prop_anchorsBoth2 = testFixes "-->foobar<--" "--><--" [
testFix [
tFromEnd 4 4 "foo" 2,
tFromStart 4 4 "bar" 1
]]
prop_composeFixes1 = testFixes "cd \"$1\" || exit" "cd $1" [
testFix [
tFromStart 4 4 "\"" 10,
tFromEnd 6 6 "\"" 10
],
testFix [
tFromEnd 6 6 " || exit" 5
]]
prop_composeFixes2 = testFixes "$(\"$1\")" "`$1`" [
testFix [
tFromStart 1 2 "$(" 5,
tFromEnd 4 5 ")" 5
],
testFix [
tFromStart 2 2 "\"" 10,
tFromEnd 4 4 "\"" 10
]]
prop_composeFixes3 = testFixes "(x)[x]" "xx" [
testFix [
tFromStart 1 1 "(" 4,
tFromEnd 2 2 ")" 3,
tFromStart 2 2 "[" 2,
tFromEnd 3 3 "]" 1
]]
prop_composeFixes4 = testFixes "(x)[x]" "xx" [
testFix [
tFromStart 1 1 "(" 4,
tFromStart 2 2 "[" 3,
tFromEnd 2 2 ")" 2,
tFromEnd 3 3 "]" 1
]]
prop_composeFixes5 = testFixes "\"$(x)\"" "`x`" [
testFix [
tFromStart 1 2 "$(" 2,
tFromEnd 3 4 ")" 2,
tFromStart 1 1 "\"" 1,
tFromEnd 4 4 "\"" 1
]]
return []
runTests = $quickCheckAll