aboutsummaryrefslogtreecommitdiff
path: root/haddock-library/test
diff options
context:
space:
mode:
Diffstat (limited to 'haddock-library/test')
-rw-r--r--haddock-library/test/Documentation/Haddock/Parser/UtilSpec.hs22
-rw-r--r--haddock-library/test/Documentation/Haddock/ParserSpec.hs816
-rw-r--r--haddock-library/test/Documentation/Haddock/Utf8Spec.hs14
-rw-r--r--haddock-library/test/Spec.hs1
4 files changed, 853 insertions, 0 deletions
diff --git a/haddock-library/test/Documentation/Haddock/Parser/UtilSpec.hs b/haddock-library/test/Documentation/Haddock/Parser/UtilSpec.hs
new file mode 100644
index 00000000..265a4d49
--- /dev/null
+++ b/haddock-library/test/Documentation/Haddock/Parser/UtilSpec.hs
@@ -0,0 +1,22 @@
+{-# LANGUAGE OverloadedStrings #-}
+module Documentation.Haddock.Parser.UtilSpec (main, spec) where
+
+import Data.Attoparsec.ByteString.Char8
+import Data.Either
+import Documentation.Haddock.Parser.Util
+import Test.Hspec
+
+main :: IO ()
+main = hspec spec
+
+spec :: Spec
+spec = do
+ describe "takeUntil" $ do
+ it "takes everything until a specified byte sequence" $ do
+ parseOnly (takeUntil "end") "someend" `shouldBe` Right "some"
+
+ it "requires the end sequence" $ do
+ parseOnly (takeUntil "end") "someen" `shouldSatisfy` isLeft
+
+ it "takes escaped bytes unconditionally" $ do
+ parseOnly (takeUntil "end") "some\\endend" `shouldBe` Right "some\\end"
diff --git a/haddock-library/test/Documentation/Haddock/ParserSpec.hs b/haddock-library/test/Documentation/Haddock/ParserSpec.hs
new file mode 100644
index 00000000..3889d555
--- /dev/null
+++ b/haddock-library/test/Documentation/Haddock/ParserSpec.hs
@@ -0,0 +1,816 @@
+{-# LANGUAGE OverloadedStrings, StandaloneDeriving
+ , FlexibleInstances, UndecidableInstances
+ , IncoherentInstances #-}
+{-# OPTIONS_GHC -fno-warn-orphans #-}
+
+module Documentation.Haddock.ParserSpec (main, spec) where
+
+import Data.Monoid
+import Data.String
+import qualified Documentation.Haddock.Parser as Parse
+import Documentation.Haddock.Types
+import Test.Hspec
+import Test.QuickCheck
+
+type Doc id = DocH () id
+
+deriving instance Show a => Show (Header a)
+deriving instance Show a => Show (Doc a)
+deriving instance Eq a => Eq (Header a)
+deriving instance Eq a => Eq (Doc a)
+
+instance IsString (Doc String) where
+ fromString = DocString
+
+instance IsString a => IsString (Maybe a) where
+ fromString = Just . fromString
+
+parseParas :: String -> Doc String
+parseParas = Parse.toRegular . Parse.parseParas
+
+parseString :: String -> Doc String
+parseString = Parse.toRegular . Parse.parseString
+
+main :: IO ()
+main = hspec spec
+
+spec :: Spec
+spec = do
+ describe "parseString" $ do
+ let infix 1 `shouldParseTo`
+ shouldParseTo :: String -> Doc String -> Expectation
+ shouldParseTo input ast = parseString input `shouldBe` ast
+
+ it "is total" $ do
+ property $ \xs ->
+ (length . show . parseString) xs `shouldSatisfy` (> 0)
+
+ context "when parsing text" $ do
+ it "can handle unicode" $ do
+ "灼眼のシャナ" `shouldParseTo` "灼眼のシャナ"
+
+ it "accepts numeric character references" $ do
+ "foo bar baz λ" `shouldParseTo` "foo bar baz λ"
+
+ it "accepts hexadecimal character references" $ do
+ "e" `shouldParseTo` "e"
+
+ it "allows to backslash-escape characters" $ do
+ property $ \x -> ['\\', x] `shouldParseTo` DocString [x]
+
+ context "when parsing strings contaning numeric character references" $ do
+ it "will implicitly convert digits to characters" $ do
+ "AAAA" `shouldParseTo` "AAAA"
+
+ "灼眼のシャナ"
+ `shouldParseTo` "灼眼のシャナ"
+
+ it "will implicitly convert hex encoded characters" $ do
+ "eeee" `shouldParseTo` "eeee"
+
+ context "when parsing identifiers" $ do
+ it "parses identifiers enclosed within single ticks" $ do
+ "'foo'" `shouldParseTo` DocIdentifier "foo"
+
+ it "parses identifiers enclosed within backticks" $ do
+ "`foo`" `shouldParseTo` DocIdentifier "foo"
+
+ it "parses a word with an one of the delimiters in it as DocString" $ do
+ "don't" `shouldParseTo` "don't"
+
+ it "doesn't pass pairs of delimiters with spaces between them" $ do
+ "hel'lo w'orld" `shouldParseTo` "hel'lo w'orld"
+
+ it "don't use apostrophe's in the wrong place's" $ do
+ " don't use apostrophe's in the wrong place's" `shouldParseTo`
+ "don't use apostrophe's in the wrong place's"
+
+ context "when parsing URLs" $ do
+ let hyperlink :: String -> Maybe String -> Doc String
+ hyperlink url = DocHyperlink . Hyperlink url
+
+ it "parses a URL" $ do
+ "<http://example.com/>" `shouldParseTo` hyperlink "http://example.com/" Nothing
+
+ it "accepts an optional label" $ do
+ "<http://example.com/ some link>" `shouldParseTo` hyperlink "http://example.com/" "some link"
+
+ it "does not accept newlines in label" $ do
+ "<foo bar\nbaz>" `shouldParseTo` "<foo bar\nbaz>"
+
+ -- new behaviour test, this will be now consistent with other markup
+ it "allows us to escape > inside the URL" $ do
+ "<http://examp\\>le.com>" `shouldParseTo`
+ hyperlink "http://examp>le.com" Nothing
+
+ "<http://exa\\>mp\\>le.com>" `shouldParseTo`
+ hyperlink "http://exa>mp>le.com" Nothing
+
+ -- Likewise in label
+ "<http://example.com f\\>oo>" `shouldParseTo`
+ hyperlink "http://example.com" "f>oo"
+
+ it "parses inline URLs" $ do
+ "foo <http://example.com/> bar" `shouldParseTo`
+ "foo " <> hyperlink "http://example.com/" Nothing <> " bar"
+
+ it "doesn't allow for multi-line link tags" $ do
+ "<ba\nz aar>" `shouldParseTo` "<ba\nz aar>"
+
+ context "when autolinking URLs" $ do
+ it "autolinks HTTP URLs" $ do
+ "http://example.com/" `shouldParseTo` hyperlink "http://example.com/" Nothing
+
+ it "autolinks HTTPS URLs" $ do
+ "https://www.example.com/" `shouldParseTo` hyperlink "https://www.example.com/" Nothing
+
+ it "autolinks FTP URLs" $ do
+ "ftp://example.com/" `shouldParseTo` hyperlink "ftp://example.com/" Nothing
+
+ it "does not include a trailing comma" $ do
+ "http://example.com/, Some other sentence." `shouldParseTo`
+ hyperlink "http://example.com/" Nothing <> ", Some other sentence."
+
+ it "does not include a trailing dot" $ do
+ "http://example.com/. Some other sentence." `shouldParseTo`
+ hyperlink "http://example.com/" Nothing <> ". Some other sentence."
+
+ it "does not include a trailing exclamation mark" $ do
+ "http://example.com/! Some other sentence." `shouldParseTo`
+ hyperlink "http://example.com/" Nothing <> "! Some other sentence."
+
+ it "does not include a trailing question mark" $ do
+ "http://example.com/? Some other sentence." `shouldParseTo`
+ hyperlink "http://example.com/" Nothing <> "? Some other sentence."
+
+ context "when parsing pictures" $ do
+ let picture :: String -> Maybe String -> Doc String
+ picture uri = DocPic . Picture uri
+
+ it "parses a simple picture" $ do
+ "<<baz>>" `shouldParseTo` picture "baz" Nothing
+
+ it "parses a picture with a title" $ do
+ "<<b a z>>" `shouldParseTo` picture "b" (Just "a z")
+
+ it "parses a picture with unicode" $ do
+ "<<灼眼のシャナ>>" `shouldParseTo` picture "灼眼のシャナ" Nothing
+
+ it "allows for escaping of the closing tags" $ do
+ "<<ba\\>>z>>" `shouldParseTo` picture "ba>>z" Nothing
+
+ it "doesn't allow for multi-line picture tags" $ do
+ "<<ba\nz aar>>" `shouldParseTo` "<<ba\nz aar>>"
+
+ context "when parsing anchors" $ do
+ it "parses a single word anchor" $ do
+ "#foo#" `shouldParseTo` DocAName "foo"
+
+ it "parses a multi word anchor" $ do
+ "#foo bar#" `shouldParseTo` DocAName "foo bar"
+
+ it "parses a unicode anchor" $ do
+ "#灼眼のシャナ#" `shouldParseTo` DocAName "灼眼のシャナ"
+
+ it "does not accept newlines in anchors" $ do
+ "#foo\nbar#" `shouldParseTo` "#foo\nbar#"
+
+ context "when parsing emphasised text" $ do
+ it "emphasises a word on its own" $ do
+ "/foo/" `shouldParseTo` DocEmphasis "foo"
+
+ it "emphasises inline correctly" $ do
+ "foo /bar/ baz" `shouldParseTo` "foo " <> DocEmphasis "bar" <> " baz"
+
+ it "emphasises unicode" $ do
+ "/灼眼のシャナ/" `shouldParseTo` DocEmphasis "灼眼のシャナ"
+
+ it "does not emphasise multi-line strings" $ do
+ " /foo\nbar/" `shouldParseTo` "/foo\nbar/"
+
+ it "does not emphasise the empty string" $ do
+ "//" `shouldParseTo` "//"
+
+ it "parses escaped slashes literally" $ do
+ "/foo\\/bar/" `shouldParseTo` DocEmphasis "foo/bar"
+
+ it "recognizes other markup constructs within emphasised text" $ do
+ "/foo @bar@ baz/" `shouldParseTo`
+ DocEmphasis ("foo " <> DocMonospaced "bar" <> " baz")
+
+ it "allows other markup inside of emphasis" $ do
+ "/__inner bold__/" `shouldParseTo` DocEmphasis (DocBold "inner bold")
+
+ it "doesn't mangle inner markup unicode" $ do
+ "/__灼眼のシャナ &#65;__/" `shouldParseTo` DocEmphasis (DocBold "灼眼のシャナ A")
+
+ it "properly converts HTML escape sequences" $ do
+ "/&#65;&#65;&#65;&#65;/" `shouldParseTo` DocEmphasis "AAAA"
+
+ it "allows to escape the emphasis delimiter inside of emphasis" $ do
+ "/empha\\/sis/" `shouldParseTo` DocEmphasis "empha/sis"
+
+ context "when parsing monospaced text" $ do
+ it "parses simple monospaced text" $ do
+ "@foo@" `shouldParseTo` DocMonospaced "foo"
+
+ it "parses inline monospaced text" $ do
+ "foo @bar@ baz" `shouldParseTo` "foo " <> DocMonospaced "bar" <> " baz"
+
+ it "allows to escape @" $ do
+ "@foo \\@ bar@" `shouldParseTo` DocMonospaced "foo @ bar"
+
+ it "accepts unicode" $ do
+ "@foo 灼眼のシャナ bar@" `shouldParseTo` DocMonospaced "foo 灼眼のシャナ bar"
+
+ it "accepts other markup in monospaced text" $ do
+ "@/foo/@" `shouldParseTo` DocMonospaced (DocEmphasis "foo")
+
+ it "requires the closing @" $ do
+ "@foo /bar/ baz" `shouldParseTo` "@foo " <> DocEmphasis "bar" <> " baz"
+
+ context "when parsing bold strings" $ do
+ it "allows for a bold string on its own" $ do
+ "__bold string__" `shouldParseTo`
+ DocBold "bold string"
+
+ it "bolds inline correctly" $ do
+ "hello __everyone__ there" `shouldParseTo`
+ "hello "
+ <> DocBold "everyone" <> " there"
+
+ it "bolds unicode" $ do
+ "__灼眼のシャナ__" `shouldParseTo`
+ DocBold "灼眼のシャナ"
+
+ it "does not do __multi-line\\n bold__" $ do
+ " __multi-line\n bold__" `shouldParseTo` "__multi-line\n bold__"
+
+ it "allows other markup inside of bold" $ do
+ "__/inner emphasis/__" `shouldParseTo`
+ (DocBold $ DocEmphasis "inner emphasis")
+
+ it "doesn't mangle inner markup unicode" $ do
+ "__/灼眼のシャナ &#65;/__" `shouldParseTo`
+ (DocBold $ DocEmphasis "灼眼のシャナ A")
+
+ it "properly converts HTML escape sequences" $ do
+ "__&#65;&#65;&#65;&#65;__" `shouldParseTo`
+ DocBold "AAAA"
+
+ it "allows to escape the bold delimiter inside of bold" $ do
+ "__bo\\__ld__" `shouldParseTo`
+ DocBold "bo__ld"
+
+ it "doesn't allow for empty bold" $ do
+ "____" `shouldParseTo` "____"
+
+ context "when parsing module strings" $ do
+ it "should parse a module on its own" $ do
+ "\"Module\"" `shouldParseTo`
+ DocModule "Module"
+
+ it "should parse a module inline" $ do
+ "This is a \"Module\"." `shouldParseTo`
+ "This is a " <> DocModule "Module" <> "."
+
+ it "can accept a simple module name" $ do
+ "\"Hello\"" `shouldParseTo` DocModule "Hello"
+
+ it "can accept a module name with dots" $ do
+ "\"Hello.World\"" `shouldParseTo` DocModule "Hello.World"
+
+ it "can accept a module name with unicode" $ do
+ "\"Hello.Worldλ\"" `shouldParseTo` DocModule "Hello.Worldλ"
+
+ it "parses a module name with a trailing dot as regular quoted string" $ do
+ "\"Hello.\"" `shouldParseTo` "\"Hello.\""
+
+ it "parses a module name with a space as regular quoted string" $ do
+ "\"Hello World\"" `shouldParseTo` "\"Hello World\""
+
+ it "parses a module name with invalid characters as regular quoted string" $ do
+ "\"Hello&[{}(=*)+]!\"" `shouldParseTo` "\"Hello&[{}(=*)+]!\""
+
+ it "accepts a module name with unicode" $ do
+ "\"Foo.Barλ\"" `shouldParseTo` DocModule "Foo.Barλ"
+
+ it "treats empty module name as regular double quotes" $ do
+ "\"\"" `shouldParseTo` "\"\""
+
+ describe "parseParas" $ do
+ let infix 1 `shouldParseTo`
+ shouldParseTo :: String -> Doc String -> Expectation
+ shouldParseTo input ast = parseParas input `shouldBe` ast
+
+ it "is total" $ do
+ property $ \xs ->
+ (length . show . parseParas) xs `shouldSatisfy` (> 0)
+
+ context "when parsing text paragraphs" $ do
+ let filterSpecial = filter (`notElem` (".(=#-[*`\v\f\n\t\r\\\"'_/@<> " :: String))
+
+ it "parses an empty paragraph" $ do
+ "" `shouldParseTo` DocEmpty
+
+ it "parses a simple text paragraph" $ do
+ "foo bar baz" `shouldParseTo` DocParagraph "foo bar baz"
+
+ it "accepts markup in text paragraphs" $ do
+ "foo /bar/ baz" `shouldParseTo` DocParagraph ("foo " <> DocEmphasis "bar" <> " baz")
+
+ it "preserve all regular characters" $ do
+ property $ \xs -> let input = filterSpecial xs in (not . null) input ==>
+ input `shouldParseTo` DocParagraph (DocString input)
+
+ it "separates paragraphs by empty lines" $ do
+ unlines [
+ "foo"
+ , " \t "
+ , "bar"
+ ] `shouldParseTo` DocParagraph "foo" <> DocParagraph "bar"
+
+ context "when a pragraph only contains monospaced text" $ do
+ it "turns it into a code block" $ do
+ "@foo@" `shouldParseTo` DocCodeBlock "foo"
+
+ context "when parsing birdtracks" $ do
+ it "parses them as a code block" $ do
+ unlines [
+ ">foo"
+ , ">bar"
+ , ">baz"
+ ] `shouldParseTo` DocCodeBlock "foo\nbar\nbaz"
+
+ it "ignores leading whitespace" $ do
+ unlines [
+ " >foo"
+ , " \t >bar"
+ , " >baz"
+ ]
+ `shouldParseTo` DocCodeBlock "foo\nbar\nbaz"
+
+ it "strips one leading space from each line of the block" $ do
+ unlines [
+ "> foo"
+ , "> bar"
+ , "> baz"
+ ] `shouldParseTo` DocCodeBlock "foo\n bar\nbaz"
+
+ it "ignores empty lines when stripping spaces" $ do
+ unlines [
+ "> foo"
+ , ">"
+ , "> bar"
+ ] `shouldParseTo` DocCodeBlock "foo\n\nbar"
+
+ context "when any non-empty line does not start with a space" $ do
+ it "does not strip any spaces" $ do
+ unlines [
+ ">foo"
+ , "> bar"
+ ] `shouldParseTo` DocCodeBlock "foo\n bar"
+
+ it "ignores nested markup" $ do
+ unlines [
+ ">/foo/"
+ ] `shouldParseTo` DocCodeBlock "/foo/"
+
+ it "treats them as regular text inside text paragraphs" $ do
+ unlines [
+ "foo"
+ , ">bar"
+ ] `shouldParseTo` DocParagraph "foo\n>bar"
+
+ context "when parsing code blocks" $ do
+ it "accepts a simple code block" $ do
+ unlines [
+ "@"
+ , "foo"
+ , "bar"
+ , "baz"
+ , "@"
+ ] `shouldParseTo` DocCodeBlock "foo\nbar\nbaz\n"
+
+ it "ignores trailing whitespace after the opening @" $ do
+ unlines [
+ "@ "
+ , "foo"
+ , "@"
+ ] `shouldParseTo` DocCodeBlock "foo\n"
+
+ it "rejects code blocks that are not closed" $ do
+ unlines [
+ "@"
+ , "foo"
+ ] `shouldParseTo` DocParagraph "@\nfoo"
+
+ it "accepts nested markup" $ do
+ unlines [
+ "@"
+ , "/foo/"
+ , "@"
+ ] `shouldParseTo` DocCodeBlock (DocEmphasis "foo" <> "\n")
+
+ it "allows to escape the @" $ do
+ unlines [
+ "@"
+ , "foo"
+ , "\\@"
+ , "bar"
+ , "@"
+ ] `shouldParseTo` DocCodeBlock "foo\n@\nbar\n"
+
+ it "accepts horizontal space before the @" $ do
+ unlines [ " @"
+ , "foo"
+ , ""
+ , "bar"
+ , "@"
+ ] `shouldParseTo` DocCodeBlock "foo\n\nbar\n"
+
+ it "strips a leading space from a @ block if present" $ do
+ unlines [ " @"
+ , " hello"
+ , " world"
+ , " @"
+ ] `shouldParseTo` DocCodeBlock "hello\nworld\n"
+
+ unlines [ " @"
+ , " hello"
+ , ""
+ , " world"
+ , " @"
+ ] `shouldParseTo` DocCodeBlock "hello\n\nworld\n"
+
+ it "only drops whitespace if there's some before closing @" $ do
+ unlines [ "@"
+ , " Formatting"
+ , " matters."
+ , "@"
+ ]
+ `shouldParseTo` DocCodeBlock " Formatting\n matters.\n"
+
+ it "accepts unicode" $ do
+ "@foo 灼眼のシャナ bar@" `shouldParseTo` DocCodeBlock "foo 灼眼のシャナ bar"
+
+ it "requires the closing @" $ do
+ "@foo /bar/ baz"
+ `shouldParseTo` DocParagraph ("@foo " <> DocEmphasis "bar" <> " baz")
+
+
+ context "when parsing examples" $ do
+ it "parses a simple example" $ do
+ ">>> foo" `shouldParseTo` DocExamples [Example "foo" []]
+
+ it "parses an example with result" $ do
+ unlines [
+ ">>> foo"
+ , "bar"
+ , "baz"
+ ] `shouldParseTo` DocExamples [Example "foo" ["bar", "baz"]]
+
+ it "parses consecutive examples" $ do
+ unlines [
+ ">>> fib 5"
+ , "5"
+ , ">>> fib 10"
+ , "55"
+ ] `shouldParseTo` DocExamples [
+ Example "fib 5" ["5"]
+ , Example "fib 10" ["55"]
+ ]
+
+ it ("requires an example to be separated"
+ ++ " from a previous paragraph by an empty line") $ do
+ "foobar\n\n>>> fib 10\n55" `shouldParseTo`
+ DocParagraph "foobar"
+ <> DocExamples [Example "fib 10" ["55"]]
+
+ it "parses bird-tracks inside of paragraphs as plain strings" $ do
+ let xs = "foo\n>>> bar"
+ xs `shouldParseTo` DocParagraph (DocString xs)
+
+ it "skips empty lines in front of an example" $ do
+ "\n \n\n>>> foo" `shouldParseTo` DocExamples [Example "foo" []]
+
+ it "terminates example on empty line" $ do
+ unlines [
+ ">>> foo"
+ , "bar"
+ , " "
+ , "baz"
+ ]
+ `shouldParseTo`
+ DocExamples [Example "foo" ["bar"]] <> DocParagraph "baz"
+
+ it "parses a <BLANKLINE> result as an empty result" $ do
+ unlines [
+ ">>> foo"
+ , "bar"
+ , "<BLANKLINE>"
+ , "baz"
+ ]
+ `shouldParseTo` DocExamples [Example "foo" ["bar", "", "baz"]]
+
+ it "accepts unicode in examples" $ do
+ ">>> 灼眼\nシャナ" `shouldParseTo` DocExamples [Example "灼眼" ["シャナ"]]
+
+ context "when prompt is prefixed by whitespace" $ do
+ it "strips the exact same amount of whitespace from result lines" $ do
+ unlines [
+ " >>> foo"
+ , " bar"
+ , " baz"
+ ] `shouldParseTo` DocExamples [Example "foo" ["bar", "baz"]]
+
+ it "preserves additional whitespace" $ do
+ unlines [
+ " >>> foo"
+ , " bar"
+ ] `shouldParseTo` DocExamples [Example "foo" [" bar"]]
+
+ it "keeps original if stripping is not possible" $ do
+ unlines [
+ " >>> foo"
+ , " bar"
+ ] `shouldParseTo` DocExamples [Example "foo" [" bar"]]
+
+
+ context "when parsing paragraphs nested in lists" $ do
+ it "can nest the same type of list" $ do
+ "* foo\n\n * bar" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocUnorderedList [DocParagraph "bar"]]
+
+ it "can nest another type of list inside" $ do
+ "* foo\n\n 1. bar" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocOrderedList [DocParagraph "bar"]]
+
+ it "can nest a code block inside" $ do
+ "* foo\n\n @foo bar baz@" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocCodeBlock "foo bar baz"]
+
+ "* foo\n\n @\n foo bar baz\n @" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocCodeBlock "foo bar baz\n"]
+
+ it "can nest more than one level" $ do
+ "* foo\n\n * bar\n\n * baz\n qux" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocUnorderedList [ DocParagraph $ "bar"
+ <> DocUnorderedList [DocParagraph "baz\nqux"]
+ ]
+ ]
+
+ it "won't fail on not fully indented paragraph" $ do
+ "* foo\n\n * bar\n\n * qux\nquux" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocUnorderedList [ DocParagraph "bar" ]
+ , DocParagraph "qux\nquux"]
+
+
+ it "can nest definition lists" $ do
+ "[a] foo\n\n [b] bar\n\n [c] baz\n qux" `shouldParseTo`
+ DocDefList [ ("a", "foo"
+ <> DocDefList [ ("b", "bar"
+ <> DocDefList [("c", "baz\nqux")])
+ ])
+ ]
+
+ it "can come back to top level with a different list" $ do
+ "* foo\n\n * bar\n\n1. baz" `shouldParseTo`
+ DocUnorderedList [ DocParagraph $ "foo"
+ <> DocUnorderedList [ DocParagraph "bar" ]
+ ]
+ <> DocOrderedList [ DocParagraph "baz" ]
+
+ it "definition lists can come back to top level with a different list" $ do
+ "[foo] foov\n\n [bar] barv\n\n1. baz" `shouldParseTo`
+ DocDefList [ ("foo", "foov"
+ <> DocDefList [ ("bar", "barv") ])
+ ]
+ <> DocOrderedList [ DocParagraph "baz" ]
+
+ context "when parsing properties" $ do
+ it "can parse a single property" $ do
+ "prop> 23 == 23" `shouldParseTo` DocProperty "23 == 23"
+
+ it "can parse multiple subsequent properties" $ do
+ unlines [
+ "prop> 23 == 23"
+ , "prop> 42 == 42"
+ ]
+ `shouldParseTo`
+ DocProperty "23 == 23" <> DocProperty "42 == 42"
+
+ it "accepts unicode in properties" $ do
+ "prop> 灼眼のシャナ ≡ 愛" `shouldParseTo`
+ DocProperty "灼眼のシャナ ≡ 愛"
+
+ it "can deal with whitespace before and after the prop> prompt" $ do
+ " prop> xs == (reverse $ reverse xs) " `shouldParseTo`
+ DocProperty "xs == (reverse $ reverse xs)"
+
+ context "when parsing unordered lists" $ do
+ it "parses a simple list" $ do
+ unlines [
+ " * one"
+ , " * two"
+ , " * three"
+ ]
+ `shouldParseTo` DocUnorderedList [
+ DocParagraph "one"
+ , DocParagraph "two"
+ , DocParagraph "three"
+ ]
+
+ it "ignores empty lines between list items" $ do
+ unlines [
+ "* one"
+ , ""
+ , "* two"
+ ]
+ `shouldParseTo` DocUnorderedList [
+ DocParagraph "one"
+ , DocParagraph "two"
+ ]
+
+ it "accepts an empty list item" $ do
+ "*" `shouldParseTo` DocUnorderedList [DocParagraph DocEmpty]
+
+ it "accepts multi-line list items" $ do
+ unlines [
+ "* point one"
+ , " more one"
+ , "* point two"
+ , "more two"
+ ]
+ `shouldParseTo` DocUnorderedList [
+ DocParagraph "point one\n more one"
+ , DocParagraph "point two\nmore two"
+ ]
+
+ it "accepts markup in list items" $ do
+ "* /foo/" `shouldParseTo` DocUnorderedList [DocParagraph (DocEmphasis "foo")]
+
+ it "requires empty lines between list and other paragraphs" $ do
+ unlines [
+ "foo"
+ , ""
+ , "* bar"
+ , ""
+ , "baz"
+ ]
+ `shouldParseTo` DocParagraph "foo" <> DocUnorderedList [DocParagraph "bar"] <> DocParagraph "baz"
+
+ context "when parsing ordered lists" $ do
+ it "parses a simple list" $ do
+ unlines [
+ " 1. one"
+ , " (1) two"
+ , " 3. three"
+ ]
+ `shouldParseTo` DocOrderedList [
+ DocParagraph "one"
+ , DocParagraph "two"
+ , DocParagraph "three"
+ ]
+
+ it "ignores empty lines between list items" $ do
+ unlines [
+ "1. one"
+ , ""
+ , "2. two"
+ ]
+ `shouldParseTo` DocOrderedList [
+ DocParagraph "one"
+ , DocParagraph "two"
+ ]
+
+ it "accepts an empty list item" $ do
+ "1." `shouldParseTo` DocOrderedList [DocParagraph DocEmpty]
+
+ it "accepts multi-line list items" $ do
+ unlines [
+ "1. point one"
+ , " more one"
+ , "1. point two"
+ , "more two"
+ ]
+ `shouldParseTo` DocOrderedList [
+ DocParagraph "point one\n more one"
+ , DocParagraph "point two\nmore two"
+ ]
+
+ it "accepts markup in list items" $ do
+ "1. /foo/" `shouldParseTo` DocOrderedList [DocParagraph (DocEmphasis "foo")]
+
+ it "requires empty lines between list and other paragraphs" $ do
+ unlines [
+ "foo"
+ , ""
+ , "1. bar"
+ , ""
+ , "baz"
+ ]
+ `shouldParseTo` DocParagraph "foo" <> DocOrderedList [DocParagraph "bar"] <> DocParagraph "baz"
+
+ context "when parsing definition lists" $ do
+ it "parses a simple list" $ do
+ unlines [
+ " [foo] one"
+ , " [bar] two"
+ , " [baz] three"
+ ]
+ `shouldParseTo` DocDefList [
+ ("foo", "one")
+ , ("bar", "two")
+ , ("baz", "three")
+ ]
+
+ it "ignores empty lines between list items" $ do
+ unlines [
+ "[foo] one"
+ , ""
+ , "[bar] two"
+ ]
+ `shouldParseTo` DocDefList [
+ ("foo", "one")
+ , ("bar", "two")
+ ]
+
+ it "accepts an empty list item" $ do
+ "[foo]" `shouldParseTo` DocDefList [("foo", DocEmpty)]
+
+ it "accepts multi-line list items" $ do
+ unlines [
+ "[foo] point one"
+ , " more one"
+ , "[bar] point two"
+ , "more two"
+ ]
+ `shouldParseTo` DocDefList [
+ ("foo", "point one\n more one")
+ , ("bar", "point two\nmore two")
+ ]
+
+ it "accepts markup in list items" $ do
+ "[foo] /foo/" `shouldParseTo` DocDefList [("foo", DocEmphasis "foo")]
+
+ it "accepts markup for the label" $ do
+ "[/foo/] bar" `shouldParseTo` DocDefList [(DocEmphasis "foo", "bar")]
+
+ it "requires empty lines between list and other paragraphs" $ do
+ unlines [
+ "foo"
+ , ""
+ , "[foo] bar"
+ , ""
+ , "baz"
+ ]
+ `shouldParseTo` DocParagraph "foo" <> DocDefList [("foo", "bar")] <> DocParagraph "baz"
+
+ context "when parsing consecutive paragraphs" $ do
+ it "will not capture irrelevant consecutive lists" $ do
+ unlines [ " * bullet"
+ , ""
+ , ""
+ , " - different bullet"
+ , ""
+ , ""
+ , " (1) ordered"
+ , " "
+ , " 2. different bullet"
+ , " "
+ , " [cat] kitten"
+ , " "
+ , " [pineapple] fruit"
+ ] `shouldParseTo`
+ DocUnorderedList [ DocParagraph "bullet"
+ , DocParagraph "different bullet"]
+ <> DocOrderedList [ DocParagraph "ordered"
+ , DocParagraph "different bullet"
+ ]
+ <> DocDefList [ ("cat", "kitten")
+ , ("pineapple", "fruit")
+ ]
+
+ context "when parsing function documentation headers" $ do
+ it "can parse a simple header" $ do
+ "= Header 1\nHello." `shouldParseTo`
+ DocParagraph (DocHeader (Header 1 "Header 1"))
+ <> DocParagraph "Hello."
+
+ it "allow consecutive headers" $ do
+ "= Header 1\n== Header 2" `shouldParseTo`
+ DocParagraph (DocHeader (Header 1 "Header 1"))
+ <> DocParagraph (DocHeader (Header 2 "Header 2"))
+
+ it "accepts markup in the header" $ do
+ "= /Header/ __1__\nFoo" `shouldParseTo`
+ DocParagraph (DocHeader
+ (Header 1 (DocEmphasis "Header" <> " " <> DocBold "1")))
+ <> DocParagraph "Foo"
diff --git a/haddock-library/test/Documentation/Haddock/Utf8Spec.hs b/haddock-library/test/Documentation/Haddock/Utf8Spec.hs
new file mode 100644
index 00000000..47e12704
--- /dev/null
+++ b/haddock-library/test/Documentation/Haddock/Utf8Spec.hs
@@ -0,0 +1,14 @@
+module Documentation.Haddock.Utf8Spec (main, spec) where
+
+import Test.Hspec
+import Test.QuickCheck
+import Documentation.Haddock.Utf8
+
+main :: IO ()
+main = hspec spec
+
+spec :: Spec
+spec = do
+ describe "decodeUtf8" $ do
+ it "is inverse to encodeUtf8" $ do
+ property $ \xs -> (decodeUtf8 . encodeUtf8) xs `shouldBe` xs
diff --git a/haddock-library/test/Spec.hs b/haddock-library/test/Spec.hs
new file mode 100644
index 00000000..a824f8c3
--- /dev/null
+++ b/haddock-library/test/Spec.hs
@@ -0,0 +1 @@
+{-# OPTIONS_GHC -F -pgmF hspec-discover #-}