module Data.Text.ParagraphLayout.Internal.Plain (layoutPlain)
where
import Control.Applicative (ZipList (ZipList), getZipList)
import Data.Foldable (toList)
import Data.Int (Int32)
import Data.List (mapAccumL)
import Data.List.NonEmpty (NonEmpty ((:|)), nonEmpty, (<|))
import qualified Data.List.NonEmpty as NonEmpty
import Data.Maybe (catMaybes)
import Data.Text.Foreign (lengthWord8)
import Data.Text.Glyphize
( Buffer (..)
, ContentType (ContentTypeUnicode)
, FontExtents (..)
, GlyphInfo
, GlyphPos
, defaultBuffer
, fontExtentsForDir
, shape
)
import Data.Text.ICU (Breaker, LocaleName, breakCharacter, breakLine)
import qualified Data.Text.ICU as BreakStatus (Line (Hard))
import Data.Text.Internal (Text (Text))
import qualified Data.Text.Lazy as Lazy
import Data.Text.ParagraphLayout.Internal.BiDiReorder
import Data.Text.ParagraphLayout.Internal.Break
import Data.Text.ParagraphLayout.Internal.Fragment
import Data.Text.ParagraphLayout.Internal.LineHeight
import Data.Text.ParagraphLayout.Internal.Paragraph
import Data.Text.ParagraphLayout.Internal.ParagraphLayout
import qualified Data.Text.ParagraphLayout.Internal.ProtoFragment as PF
import Data.Text.ParagraphLayout.Internal.Rect
import Data.Text.ParagraphLayout.Internal.ResolvedSpan (WithSpan (WithSpan))
import qualified Data.Text.ParagraphLayout.Internal.ResolvedSpan as RS
import Data.Text.ParagraphLayout.Internal.Run
import Data.Text.ParagraphLayout.Internal.Span
import Data.Text.ParagraphLayout.Internal.TextContainer
-- | Lay out a paragraph of plain, unidirectional text using a single font.
layoutPlain :: Paragraph -> ParagraphLayout
layoutPlain p@(Paragraph _ _ _ opts) = paragraphLayout sls
where
sls = map SpanLayout fragsBySpan
fragsBySpan = take (length spans) $ RS.splitBySpanIndex frags
frags = case nonEmpty wrappedRuns of
Just xs -> layoutAndAlignLines maxWidth xs
Nothing -> []
wrappedRuns = spansToRunsWrapped spans
maxWidth = paragraphMaxWidth opts
spans = resolveSpans p
-- | Split a number of spans into a flat array of runs and add a wrapper
-- so that each run can be traced back to its originating span.
spansToRunsWrapped :: [RS.ResolvedSpan] -> [WithSpan Run]
spansToRunsWrapped ss = concat $ map spanToRunsWrapped ss
-- | Split a span into runs and add a wrapper
-- so that each run can be traced back to its originating span.
spanToRunsWrapped :: RS.ResolvedSpan -> [WithSpan Run]
spanToRunsWrapped s = map (WithSpan s) (spanToRuns s)
-- | Create a multi-line layout from the given runs, splitting them as
-- necessary to fit within the requested line width.
--
-- The output is a flat list of fragments positioned in both dimensions.
layoutAndAlignLines :: Int32 -> NonEmpty (WithSpan Run) -> [WithSpan Fragment]
layoutAndAlignLines maxWidth runs = frags
where
frags = concatMap NonEmpty.toList fragsInLines
(_, fragsInLines) = mapAccumL positionLineH originY canonicalLines
canonicalLines = fmap reorder logicalLines
logicalLines = nonEmptyItems $ layoutLines maxWidth runs
originY = paragraphOriginY
nonEmptyItems :: Foldable t => t [a] -> [NonEmpty a]
nonEmptyItems = catMaybes . map nonEmpty . toList
-- | Create a multi-line layout from the given runs, splitting them as
-- necessary to fit within the requested line width.
--
-- The output is a two-dimensional list of fragments positioned along the
-- horizontal axis.
layoutLines ::
Int32 -> NonEmpty (WithSpan Run) -> NonEmpty [WithSpan PF.ProtoFragment]
layoutLines maxWidth runs = case nonEmpty rest of
-- Everything fits. We are done.
Nothing -> NonEmpty.singleton fitting
-- Something fits, the rest goes on the next line.
Just rest' -> fitting <| layoutLines maxWidth rest'
where
(fitting, rest) = layoutAndWrapRunsH maxWidth runs
-- TODO: Allow a run across multiple spans (e.g. if they only differ by colour).
-- | Position all the given horizontal fragments on the same line,
-- using `originY` as its top edge, and return the bottom edge for continuation.
--
-- Glyphs will be aligned by their ascent line, similar to the behaviour of
-- @vertical-align: top@ in CSS.
--
-- TODO: For rich text, allow other types of vertical alignment.
positionLineH :: Int32 -> NonEmpty (WithSpan PF.ProtoFragment) ->
(Int32, NonEmpty (WithSpan Fragment))
positionLineH originY pfs = (nextY, frags)
where
nextY = maximum $ fmap y_min rects
rects = fmap (\ (WithSpan _ r) -> fragmentRect r) frags
(_, frags) = mapAccumL (positionFragmentH originY) originX pfs
originX = paragraphOriginX
-- | Position the given horizontal fragment on a line,
-- using `originY` as its top edge and `originX` as its left edge,
-- returning the X coordinate of its right edge for continuation.
positionFragmentH ::
Int32 -> Int32 -> WithSpan PF.ProtoFragment -> (Int32, WithSpan Fragment)
positionFragmentH originY originX (WithSpan rs pf) = (nextX, WithSpan rs frag)
where
nextX = originX + PF.advance pf
frag = Fragment rect (penX, penY) (PF.glyphs pf)
rect = Rect originX originY (PF.advance pf) (-lineHeight)
penX = 0
penY = descent + leading `div` 2 - lineHeight
lineHeight = case RS.spanLineHeight rs of
Normal -> normalLineHeight
Absolute h -> h
leading = lineHeight - normalLineHeight
normalLineHeight = ascent + descent
ascent = ascender extents
descent = - descender extents
extents = fontExtentsForDir font (PF.direction pf)
font = RS.spanFont rs
-- | Calculate layout for multiple horizontal runs, breaking them as necessary
-- to fit as much content as possible without exceeding the maximum line width,
-- and return the remaining runs to be placed on other lines.
layoutAndWrapRunsH :: Int32 -> NonEmpty (WithSpan Run) ->
([WithSpan PF.ProtoFragment], [WithSpan Run])
layoutAndWrapRunsH maxWidth runs = NonEmpty.head $ validLayouts
where
validLayouts = dropWhile1 tooLong layouts
tooLong (pfs, _) = totalAdvances pfs > maxWidth
layouts = fmap layoutFst splits
layoutFst (runs1, runs2) = (layoutRunsH runs1, runs2)
-- TODO: Consider optimising.
-- We do not need to look for soft breaks further than the
-- shortest hard break.
splits = hardSplit runs :| softSplits runs
-- | Treat a list of runs as a contiguous sequence, and split them into two
-- lists so that the first list contains as many non-whitespace characters as
-- possible without crossing a hard line break (typically after a newline
-- character).
--
-- If the input is non-empty and starts with a hard line break, then the first
-- output list will contain a run of zero characters. This can be used to
-- correctly size an empty line.
--
-- If there is no hard line break in the input, the first output list will
-- contain the whole input, and the second output list will be empty.
hardSplit :: NonEmpty (WithSpan Run) -> ([WithSpan Run], [WithSpan Run])
hardSplit runs = allowFstEmpty $ trimFst $ NonEmpty.last $ splits
where
trimFst (runs1, runs2) = (trim runs1, runs2)
trim
= trimTextsStartPreserve isStartSpace
. trimTextsEndPreserve isEndSpace
. trimTextsEndPreserve isNewline
-- TODO: Consider optimising.
-- We do not need to look for any line breaks further than the
-- shortest hard break.
splits = noSplit :| map allowSndEmpty hSplits
noSplit = (runs, [])
hSplits = -- from longest to shortest
splitTextsBy (map fst . filter isHard . runLineBreaks) runs
isHard (_, status) = status == BreakStatus.Hard
-- | Treat a list of runs as a contiguous sequence,
-- and find all possible ways to split them into two non-empty lists,
-- using soft line break opportunities (typically after words) and then
-- using character boundaries.
--
-- Runs of zero characters will not be created. If line breaking would result
-- in a line that consists entirely of whitespace, this whitespace will be
-- skipped, so an empty line is not created.
--
-- The results in the form (prefix, suffix) will be ordered so that items
-- closer to the start of the list are preferred for line breaking, but without
-- considering overflows.
softSplits :: NonEmpty (WithSpan Run) -> [([WithSpan Run], [WithSpan Run])]
softSplits runs = map (allowSndEmpty . trimFst) splits
where
trimFst (runs1, runs2) = (trim runs1, runs2)
trim = trimTextsStart isStartSpace . trimTextsEnd isEndSpace
splits = lSplits ++ cSplits
lSplits = splitTextsBy (map fst . runLineBreaks) runs
-- TODO: Consider optimising.
-- We do not need to look for character breaks further than the
-- shortest line break.
cSplits = splitTextsBy (map fst . runCharacterBreaks) runs
allowFstEmpty :: (NonEmpty a, b) -> ([a], b)
allowFstEmpty (a, b) = (NonEmpty.toList a, b)
allowSndEmpty :: (a, NonEmpty b) -> (a, [b])
allowSndEmpty (a, b) = (a, NonEmpty.toList b)
-- | The suffix remaining after removing the longest prefix of the list for
-- which the predicate holds, except always including at least the last element
-- of the original list.
dropWhile1 :: (a -> Bool) -> NonEmpty a -> NonEmpty a
dropWhile1 p list = case NonEmpty.uncons list of
(_, Nothing) -> list
(x, Just xs) -> if p x
then dropWhile1 p xs
else list
-- | Calculate layout for multiple horizontal runs on the same line, without
-- any breaking.
layoutRunsH :: [WithSpan Run] -> [WithSpan PF.ProtoFragment]
layoutRunsH runs = map layoutRunH runs
-- | Sum of all advances within the given fragments.
totalAdvances :: [WithSpan PF.ProtoFragment] -> Int32
totalAdvances pfs = sum $ map (\ (WithSpan _ pf) -> PF.advance pf) pfs
-- | Calculate layout for the given horizontal run and attach extra information.
layoutRunH :: WithSpan Run -> WithSpan PF.ProtoFragment
layoutRunH (WithSpan rs run) = WithSpan rs pf
where
pf = PF.protoFragmentH dir glyphs
glyphs = shapeRun (WithSpan rs run)
dir = runDirection run
-- | Calculate layout for the given run independently of its position.
shapeRun :: WithSpan Run -> [(GlyphInfo, GlyphPos)]
shapeRun (WithSpan rs run) = shape font buffer features
where
font = RS.spanFont rs
buffer = defaultBuffer
{ text = Lazy.fromStrict $ runText run
, contentType = Just ContentTypeUnicode
, direction = runDirection run
, script = runScript run
, language = Just $ RS.spanLanguage rs
-- Perhaps counter-intuitively, the `beginsText` and `endsText`
-- flags refer to everything that "Data.Text.Glyphize" can see,
-- not just the current run.
--
-- Since all runs are cut from a single continuous `Text` that
-- represents the entire paragraph, and "Data.Text.Glyphize" peeks
-- at the whole underlying byte array, HarfBuzz will be able to see
-- both the beginning and the end of the paragraph at all times,
-- so these flags can always be set.
, beginsText = True
, endsText = True
}
features = []
resolveSpans :: Paragraph -> [RS.ResolvedSpan]
resolveSpans p@(Paragraph arr pStart spans pOpts) = do
let sBounds = paragraphSpanBounds p
let pEnd = NonEmpty.last sBounds
let sStarts = NonEmpty.init sBounds
let sLengths = map spanLength spans
(i, s, sStart, sLen) <- getZipList $ (,,,)
<$> ZipList [0 ..]
<*> ZipList spans
<*> ZipList sStarts
<*> ZipList sLengths
let lang = spanLanguage $ spanOptions s
let lBreaks = paragraphBreaks breakLine p pEnd lang
let cBreaks = paragraphBreaks breakCharacter p pEnd lang
return RS.ResolvedSpan
{ RS.spanIndex = i
, RS.spanOffsetInParagraph = sStart - pStart
-- TODO: Consider adding checks for array bounds.
, RS.spanText = Text arr sStart sLen
, RS.spanFont = paragraphFont pOpts
, RS.spanLineHeight = paragraphLineHeight pOpts
, RS.spanLanguage = lang
, RS.spanLineBreaks = subOffsetsDesc (sStart - pStart) lBreaks
, RS.spanCharacterBreaks = subOffsetsDesc (sStart - pStart) cBreaks
}
paragraphBreaks :: (LocaleName -> Breaker a) -> Paragraph -> Int -> String ->
[(Int, a)]
paragraphBreaks breakFunc (Paragraph arr off _ _) end lang =
breaksDesc (breakFunc (locale lang LBAuto)) paragraphText
where
paragraphText = Text arr off (end - off)
runLineBreaks :: WithSpan Run -> [(Int, BreakStatus.Line)]
runLineBreaks (WithSpan rs run) =
runBreaksFromSpan run $ RS.spanLineBreaks rs
runCharacterBreaks :: WithSpan Run -> [(Int, ())]
runCharacterBreaks (WithSpan rs run) =
runBreaksFromSpan run $ RS.spanCharacterBreaks rs
-- | Constrain span breaks to a selected run and adjust offsets.
runBreaksFromSpan :: Run -> [(Int, a)] -> [(Int, a)]
runBreaksFromSpan run spanBreaks =
dropWhile (not . valid) $ subOffsetsDesc (runOffsetInSpan run) spanBreaks
where
valid (off, _) = off < runLength
runLength = lengthWord8 $ getText run
-- | Predicate for characters that can be potentially removed from the
-- beginning of a line according to the CSS Text Module.
isStartSpace :: Char -> Bool
isStartSpace c = c `elem` [' ', '\t']
-- | Predicate for characters that can be potentially removed from the end of
-- a line according to the CSS Text Module.
isEndSpace :: Char -> Bool
isEndSpace c = c `elem` [' ', '\t', '\x1680']
-- | Predicate for characters that should be removed from the end of a line in
-- the case of a hard line break.
isNewline :: Char -> Bool
isNewline c = c == '\n'