1119 lines
36 KiB
Plaintext
1119 lines
36 KiB
Plaintext
|
|
> {-# LANGUAGE TupleSections #-}
|
|
> -- | This is the module with the parser functions.
|
|
> module Language.SQL.SimpleSQL.Parser
|
|
> (parseQueryExpr
|
|
> ,parseValueExpr
|
|
> ,parseQueryExprs
|
|
> ,ParseError(..)) where
|
|
|
|
> import Control.Monad.Identity (Identity)
|
|
> import Control.Monad (guard, void)
|
|
> import Control.Applicative ((<$), (<$>), (<*>) ,(<*), (*>))
|
|
> import Data.Maybe (fromMaybe,catMaybes)
|
|
> import Data.Char (toLower)
|
|
> import Text.Parsec (errorPos,sourceLine,sourceColumn,sourceName
|
|
> ,setPosition,setSourceColumn,setSourceLine,getPosition
|
|
> ,option,between,sepBy,sepBy1,string,manyTill,anyChar
|
|
> ,try,string,many1,oneOf,digit,(<|>),choice,char,eof
|
|
> ,optionMaybe,optional,many,letter,parse
|
|
> ,chainl1, (<?>),notFollowedBy,alphaNum)
|
|
> import Text.Parsec.String (Parser)
|
|
> import qualified Text.Parsec as P (ParseError)
|
|
> import Text.Parsec.Perm (permute,(<$?>), (<|?>))
|
|
> import qualified Text.Parsec.Expr as E
|
|
|
|
> import Language.SQL.SimpleSQL.Syntax
|
|
|
|
The public API functions.
|
|
|
|
> -- | Parses a query expr, trailing semicolon optional.
|
|
> parseQueryExpr :: FilePath
|
|
> -- ^ filename to use in errors
|
|
> -> Maybe (Int,Int)
|
|
> -- ^ line number and column number of the first character
|
|
> -- in the source (to use in errors)
|
|
> -> String
|
|
> -- ^ the SQL source to parse
|
|
> -> Either ParseError QueryExpr
|
|
> parseQueryExpr = wrapParse topLevelQueryExpr
|
|
|
|
> -- | Parses a list of query expressions, with semi colons between
|
|
> -- them. The final semicolon is optional.
|
|
> parseQueryExprs :: FilePath
|
|
> -- ^ filename to use in errors
|
|
> -> Maybe (Int,Int)
|
|
> -- ^ line number and column number of the first character
|
|
> -- in the source (to use in errors)
|
|
> -> String
|
|
> -- ^ the SQL source to parse
|
|
> -> Either ParseError [QueryExpr]
|
|
> parseQueryExprs = wrapParse queryExprs
|
|
|
|
> -- | Parses a value expression.
|
|
> parseValueExpr :: FilePath
|
|
> -- ^ filename to use in errors
|
|
> -> Maybe (Int,Int)
|
|
> -- ^ line number and column number of the first character
|
|
> -- in the source (to use in errors)
|
|
> -> String
|
|
> -- ^ the SQL source to parse
|
|
> -> Either ParseError ValueExpr
|
|
> parseValueExpr = wrapParse valueExpr
|
|
|
|
This helper function takes the parser given and:
|
|
|
|
sets the position when parsing
|
|
automatically skips leading whitespace
|
|
checks the parser parses all the input using eof
|
|
converts the error return to the nice wrapper
|
|
|
|
> wrapParse :: Parser a
|
|
> -> FilePath
|
|
> -> Maybe (Int,Int)
|
|
> -> String
|
|
> -> Either ParseError a
|
|
> wrapParse parser f p src =
|
|
> either (Left . convParseError src) Right
|
|
> $ parse (setPos p *> whitespace *> parser <* eof) f src
|
|
|
|
> -- | Type to represent parse errors.
|
|
> data ParseError = ParseError
|
|
> {peErrorString :: String
|
|
> -- ^ contains the error message
|
|
> ,peFilename :: FilePath
|
|
> -- ^ filename location for the error
|
|
> ,pePosition :: (Int,Int)
|
|
> -- ^ line number and column number location for the error
|
|
> ,peFormattedError :: String
|
|
> -- ^ formatted error with the position, error
|
|
> -- message and source context
|
|
> } deriving (Eq,Show)
|
|
|
|
------------------------------------------------
|
|
|
|
= value expressions
|
|
|
|
== literals
|
|
|
|
See the stringToken lexer below for notes on string literal syntax.
|
|
|
|
> stringValue :: Parser ValueExpr
|
|
> stringValue = StringLit <$> stringToken
|
|
|
|
> number :: Parser ValueExpr
|
|
> number = NumLit <$> numberLiteral
|
|
|
|
parse SQL interval literals, something like
|
|
interval '5' day (3)
|
|
or
|
|
interval '5' month
|
|
|
|
wrap the whole lot in try, in case we get something like this:
|
|
interval '3 days'
|
|
which parses as a typed literal
|
|
|
|
> interval :: Parser ValueExpr
|
|
> interval = keyword_ "interval" >>
|
|
> mkIt <$> stringToken
|
|
> <*> optionMaybe
|
|
> ((,) <$> identifierBlacklist blacklist
|
|
> <*> optionMaybe (parens integer))
|
|
> where
|
|
> mkIt val Nothing = TypedLit (TypeName "interval") val
|
|
> mkIt val (Just (a,b)) = IntervalLit val a b
|
|
|
|
> literal :: Parser ValueExpr
|
|
> literal = number <|> stringValue <|> interval
|
|
|
|
== Names
|
|
|
|
> name :: Parser Name
|
|
> name = choice [QName <$> quotedIdentifier
|
|
> ,Name <$> identifierBlacklist blacklist]
|
|
|
|
== star
|
|
|
|
used in select *, select x.*, and agg(*) variations, and some other
|
|
places as well. Because it is quite general, the parser doesn't
|
|
attempt to check that the star is in a valid context, it parses it OK
|
|
in any value expression context.
|
|
|
|
> star :: Parser ValueExpr
|
|
> star = Star <$ symbol "*"
|
|
|
|
== parameter
|
|
|
|
use in e.g. select * from t where a = ?
|
|
|
|
> parameter :: Parser ValueExpr
|
|
> parameter = Parameter <$ questionMark
|
|
|
|
named parameter:
|
|
|
|
select x from t where x > :param
|
|
|
|
> hostParameter :: Parser ValueExpr
|
|
> hostParameter =
|
|
> HostParameter
|
|
> <$> hostParameterToken
|
|
> <*> optionMaybe (keyword "indicator" *> hostParameterToken)
|
|
|
|
== function application, aggregates and windows
|
|
|
|
this represents anything which syntactically looks like regular C
|
|
function application: an identifier, parens with comma sep value
|
|
expression arguments.
|
|
|
|
The parsing for the aggregate extensions is here as well:
|
|
|
|
aggregate([all|distinct] args [order by orderitems])
|
|
|
|
> aggOrApp :: Name -> Parser ValueExpr
|
|
> aggOrApp n =
|
|
> makeApp n
|
|
> <$> parens ((,,) <$> duplicates
|
|
> <*> choice [commaSep valueExpr]
|
|
> <*> (optionMaybe orderBy))
|
|
> where
|
|
> makeApp i (Nothing,es,Nothing) = App i es
|
|
> makeApp i (d,es,od) = AggregateApp i d es (fromMaybe [] od)
|
|
|
|
> duplicates :: Parser (Maybe SetQuantifier)
|
|
> duplicates = optionMaybe $
|
|
> choice [All <$ keyword_ "all"
|
|
> ,Distinct <$ keyword "distinct"]
|
|
|
|
parse a window call as a suffix of a regular function call
|
|
this looks like this:
|
|
functionname(args) over ([partition by ids] [order by orderitems])
|
|
|
|
No support for explicit frames yet.
|
|
|
|
The convention in this file is that the 'Suffix', erm, suffix on
|
|
parser names means that they have been left factored. These are almost
|
|
always used with the optionSuffix combinator.
|
|
|
|
> windowSuffix :: ValueExpr -> Parser ValueExpr
|
|
> windowSuffix (App f es) =
|
|
> keyword_ "over"
|
|
> *> parens (WindowApp f es
|
|
> <$> option [] partitionBy
|
|
> <*> option [] orderBy
|
|
> <*> optionMaybe frameClause)
|
|
> where
|
|
> partitionBy = keyword_ "partition" >>
|
|
> keyword_ "by" >> commaSep1 valueExpr
|
|
> frameClause =
|
|
> mkFrame <$> choice [FrameRows <$ keyword_ "rows"
|
|
> ,FrameRange <$ keyword_ "range"]
|
|
> <*> frameStartEnd
|
|
> frameStartEnd =
|
|
> choice
|
|
> [keyword_ "between" >>
|
|
> mkFrameBetween <$> frameLimit True
|
|
> <*> (keyword_ "and" *> frameLimit True)
|
|
> ,mkFrameFrom <$> frameLimit False]
|
|
> -- use the bexpression style from the between parsing for frame between
|
|
> frameLimit useB =
|
|
> choice
|
|
> [Current <$ keyword_ "current" <* keyword_ "row"
|
|
> ,keyword_ "unbounded" >>
|
|
> choice [UnboundedPreceding <$ keyword_ "preceding"
|
|
> ,UnboundedFollowing <$ keyword_ "following"]
|
|
> ,do
|
|
> e <- if useB then valueExprB else valueExpr
|
|
> choice [Preceding e <$ keyword_ "preceding"
|
|
> ,Following e <$ keyword_ "following"]
|
|
> ]
|
|
> mkFrameBetween s e rs = FrameBetween rs s e
|
|
> mkFrameFrom s rs = FrameFrom rs s
|
|
> mkFrame rs c = c rs
|
|
> windowSuffix _ = fail ""
|
|
|
|
> app :: Name -> Parser ValueExpr
|
|
> app n = aggOrApp n >>= optionSuffix windowSuffix
|
|
|
|
== iden prefix term
|
|
|
|
all the value expressions which start with an identifier
|
|
|
|
(todo: really put all of them here instead of just some of them)
|
|
|
|
> idenPrefixTerm :: Parser ValueExpr
|
|
> idenPrefixTerm =
|
|
> -- todo: work out how to left factor this
|
|
> try (TypedLit <$> typeName <*> stringToken)
|
|
> <|> (name >>= iden)
|
|
> where
|
|
> iden n = app n <|> return (Iden n)
|
|
|
|
== case expression
|
|
|
|
> caseValue :: Parser ValueExpr
|
|
> caseValue =
|
|
> Case <$> (keyword_ "case" *> optionMaybe valueExpr)
|
|
> <*> many1 whenClause
|
|
> <*> optionMaybe elseClause
|
|
> <* keyword_ "end"
|
|
> where
|
|
> whenClause = (,) <$> (keyword_ "when" *> commaSep1 valueExpr)
|
|
> <*> (keyword_ "then" *> valueExpr)
|
|
> elseClause = keyword_ "else" *> valueExpr
|
|
|
|
== miscellaneous keyword operators
|
|
|
|
These are keyword operators which don't look like normal prefix,
|
|
postfix or infix binary operators. They mostly look like function
|
|
application but with keywords in the argument list instead of commas
|
|
to separate the arguments.
|
|
|
|
cast: cast(expr as type)
|
|
|
|
> cast :: Parser ValueExpr
|
|
> cast = keyword_ "cast" >>
|
|
> parens (Cast <$> valueExpr
|
|
> <*> (keyword_ "as" *> typeName))
|
|
|
|
the special op keywords
|
|
parse an operator which is
|
|
operatorname(firstArg keyword0 arg0 keyword1 arg1 etc.)
|
|
|
|
> data SpecialOpKFirstArg = SOKNone
|
|
> | SOKOptional
|
|
> | SOKMandatory
|
|
|
|
> specialOpK :: String -- name of the operator
|
|
> -> SpecialOpKFirstArg -- has a first arg without a keyword
|
|
> -> [(String,Bool)] -- the other args with their keywords
|
|
> -- and whether they are optional
|
|
> -> Parser ValueExpr
|
|
> specialOpK opName firstArg kws =
|
|
> keyword_ opName >> do
|
|
> void openParen
|
|
> let pfa = do
|
|
> e <- valueExpr
|
|
> -- check we haven't parsed the first
|
|
> -- keyword as an identifier
|
|
> guard (case (e,kws) of
|
|
> (Iden (Name i), (k,_):_) | map toLower i == k -> False
|
|
> _ -> True)
|
|
> return e
|
|
> fa <- case firstArg of
|
|
> SOKNone -> return Nothing
|
|
> SOKOptional -> optionMaybe (try pfa)
|
|
> SOKMandatory -> Just <$> pfa
|
|
> as <- mapM parseArg kws
|
|
> void closeParen
|
|
> return $ SpecialOpK (Name opName) fa $ catMaybes as
|
|
> where
|
|
> parseArg (nm,mand) =
|
|
> let p = keyword_ nm >> valueExpr
|
|
> in fmap (nm,) <$> if mand
|
|
> then Just <$> p
|
|
> else optionMaybe (try p)
|
|
|
|
The actual operators:
|
|
|
|
EXTRACT( date_part FROM expression )
|
|
|
|
POSITION( string1 IN string2 )
|
|
|
|
SUBSTRING(extraction_string FROM starting_position [FOR length]
|
|
[COLLATE collation_name])
|
|
|
|
CONVERT(char_value USING conversion_char_name)
|
|
|
|
TRANSLATE(char_value USING translation_name)
|
|
|
|
OVERLAY(string PLACING embedded_string FROM start
|
|
[FOR length])
|
|
|
|
TRIM( [ [{LEADING | TRAILING | BOTH}] [removal_char] FROM ]
|
|
target_string
|
|
[COLLATE collation_name] )
|
|
|
|
> specialOpKs :: Parser ValueExpr
|
|
> specialOpKs = choice $ map try
|
|
> [extract, position, substring, convert, translate, overlay, trim]
|
|
|
|
> extract :: Parser ValueExpr
|
|
> extract = specialOpK "extract" SOKMandatory [("from", True)]
|
|
|
|
> position :: Parser ValueExpr
|
|
> position = specialOpK "position" SOKMandatory [("in", True)]
|
|
|
|
strictly speaking, the substring must have at least one of from and
|
|
for, but the parser doens't enforce this
|
|
|
|
> substring :: Parser ValueExpr
|
|
> substring = specialOpK "substring" SOKMandatory
|
|
> [("from", False),("for", False),("collate", False)]
|
|
|
|
> convert :: Parser ValueExpr
|
|
> convert = specialOpK "convert" SOKMandatory [("using", True)]
|
|
|
|
|
|
> translate :: Parser ValueExpr
|
|
> translate = specialOpK "translate" SOKMandatory [("using", True)]
|
|
|
|
> overlay :: Parser ValueExpr
|
|
> overlay = specialOpK "overlay" SOKMandatory
|
|
> [("placing", True),("from", True),("for", False)]
|
|
|
|
trim is too different because of the optional char, so a custom parser
|
|
the both ' ' is filled in as the default if either parts are missing
|
|
in the source
|
|
|
|
> trim :: Parser ValueExpr
|
|
> trim =
|
|
> keyword "trim" >>
|
|
> parens (mkTrim
|
|
> <$> option "both" sides
|
|
> <*> option " " stringToken
|
|
> <*> (keyword_ "from" *> valueExpr)
|
|
> <*> optionMaybe (keyword_ "collate" *> stringToken))
|
|
> where
|
|
> sides = choice ["leading" <$ keyword_ "leading"
|
|
> ,"trailing" <$ keyword_ "trailing"
|
|
> ,"both" <$ keyword_ "both"]
|
|
> mkTrim fa ch fr cl =
|
|
> SpecialOpK (Name "trim") Nothing
|
|
> $ catMaybes [Just (fa,StringLit ch)
|
|
> ,Just ("from", fr)
|
|
> ,fmap (("collate",) . StringLit) cl]
|
|
|
|
in: two variations:
|
|
a in (expr0, expr1, ...)
|
|
a in (queryexpr)
|
|
|
|
this is parsed as a postfix operator which is why it is in this form
|
|
|
|
> inSuffix :: Parser (ValueExpr -> ValueExpr)
|
|
> inSuffix =
|
|
> mkIn <$> inty
|
|
> <*> parens (choice
|
|
> [InQueryExpr <$> queryExpr
|
|
> ,InList <$> commaSep1 valueExpr])
|
|
> where
|
|
> inty = choice [True <$ keyword_ "in"
|
|
> ,False <$ keyword_ "not" <* keyword_ "in"]
|
|
> mkIn i v = \e -> In i e v
|
|
|
|
|
|
between:
|
|
expr between expr and expr
|
|
|
|
There is a complication when parsing between - when parsing the second
|
|
expression it is ambiguous when you hit an 'and' whether it is a
|
|
binary operator or part of the between. This code follows what
|
|
postgres does, which might be standard across SQL implementations,
|
|
which is that you can't have a binary and operator in the middle
|
|
expression in a between unless it is wrapped in parens. The 'bExpr
|
|
parsing' is used to create alternative value expression parser which
|
|
is identical to the normal one expect it doesn't recognise the binary
|
|
and operator. This is the call to valueExprB.
|
|
|
|
> betweenSuffix :: Parser (ValueExpr -> ValueExpr)
|
|
> betweenSuffix =
|
|
> makeOp <$> (Name <$> opName)
|
|
> <*> valueExprB
|
|
> <*> (keyword_ "and" *> valueExprB)
|
|
> where
|
|
> opName = choice
|
|
> ["between" <$ keyword_ "between"
|
|
> ,"not between" <$ try (keyword_ "not" <* keyword_ "between")]
|
|
> makeOp n b c = \a -> SpecialOp n [a,b,c]
|
|
|
|
subquery expression:
|
|
[exists|unique] (queryexpr)
|
|
|
|
> subquery :: Parser ValueExpr
|
|
> subquery = SubQueryExpr <$> sqkw <*> parens queryExpr
|
|
> where
|
|
> sqkw = choice
|
|
> [SqExists <$ keyword_ "exists"
|
|
> ,SqUnique <$ keyword_ "unique"]
|
|
|
|
|
|
a = any (select * from t)
|
|
|
|
> quantifiedComparison :: Parser (ValueExpr -> ValueExpr)
|
|
> quantifiedComparison = do
|
|
> c <- comp
|
|
> cq <- compQuan
|
|
> q <- parens queryExpr
|
|
> return $ \v -> QuantifiedComparison v c cq q
|
|
> where
|
|
> comp = Name <$> choice (map symbol
|
|
> ["=", "<>", "<=", "<", ">", ">="])
|
|
> compQuan = choice
|
|
> [CPAny <$ keyword_ "any"
|
|
> ,CPSome <$ keyword_ "some"
|
|
> ,CPAll <$ keyword_ "all"]
|
|
|
|
a match (select a from t)
|
|
|
|
> matchPredicate :: Parser (ValueExpr -> ValueExpr)
|
|
> matchPredicate = do
|
|
> keyword_ "match"
|
|
> u <- option False (True <$ keyword_ "unique")
|
|
> q <- parens queryExpr
|
|
> return $ \v -> Match v u q
|
|
|
|
|
|
> arrayPostfix :: Parser (ValueExpr -> ValueExpr)
|
|
> arrayPostfix = do
|
|
> es <- brackets (commaSep valueExpr)
|
|
> return $ \v -> Array v es
|
|
|
|
> arrayCtor :: Parser ValueExpr
|
|
> arrayCtor = keyword_ "array" >>
|
|
> choice
|
|
> [ArrayCtor <$> parens queryExpr
|
|
> ,Array (Iden (Name "array")) <$> brackets (commaSep valueExpr)]
|
|
|
|
typename: used in casts. Special cases for the multi keyword typenames
|
|
that SQL supports.
|
|
|
|
> typeName :: Parser TypeName
|
|
> typeName = choice (multiWordParsers
|
|
> ++ [TypeName <$> identifierBlacklist blacklist])
|
|
> >>= optionSuffix precision
|
|
> where
|
|
> multiWordParsers =
|
|
> flip map multiWordTypeNames
|
|
> $ \ks -> (TypeName . unwords) <$> try (mapM keyword ks)
|
|
> multiWordTypeNames = map words
|
|
> ["double precision"
|
|
> ,"character varying"
|
|
> ,"char varying"
|
|
> ,"character large object"
|
|
> ,"char large object"
|
|
> ,"national character"
|
|
> ,"national char"
|
|
> ,"national character varying"
|
|
> ,"national char varying"
|
|
> ,"national character large object"
|
|
> ,"nchar large object"
|
|
> ,"nchar varying"
|
|
> ,"bit varying"
|
|
> ]
|
|
|
|
todo: timestamp types:
|
|
|
|
| TIME [ <left paren> <time precision> <right paren> ] [ WITH TIME ZONE ]
|
|
| TIMESTAMParser [ <left paren> <timestamp precision> <right paren> ] [ WITH TIME ZONE ]
|
|
|
|
|
|
> precision t = parens (commaSep integer) >>= makeWrap t
|
|
> makeWrap (TypeName t) [a] = return $ PrecTypeName t a
|
|
> makeWrap (TypeName t) [a,b] = return $ PrecScaleTypeName t a b
|
|
> makeWrap _ _ = fail "there must be one or two precision components"
|
|
|
|
|
|
== value expression parens, row ctor and scalar subquery
|
|
|
|
> parensTerm :: Parser ValueExpr
|
|
> parensTerm = parens $ choice
|
|
> [SubQueryExpr SqSq <$> queryExpr
|
|
> ,ctor <$> commaSep1 valueExpr]
|
|
> where
|
|
> ctor [a] = Parens a
|
|
> ctor as = SpecialOp (Name "rowctor") as
|
|
|
|
|
|
== operator parsing
|
|
|
|
The 'regular' operators in this parsing and in the abstract syntax are
|
|
unary prefix, unary postfix and binary infix operators. The operators
|
|
can be symbols (a + b), single keywords (a and b) or multiple keywords
|
|
(a is similar to b).
|
|
|
|
TODO: carefully review the precedences and associativities.
|
|
|
|
> opTable :: Bool -> [[E.Operator String () Identity ValueExpr]]
|
|
> opTable bExpr =
|
|
> [-- parse match and quantified comparisons as postfix ops
|
|
> -- todo: left factor the quantified comparison with regular
|
|
> -- binary comparison, somehow
|
|
> [E.Postfix $ try quantifiedComparison
|
|
> ,E.Postfix matchPredicate]
|
|
> ,[binarySym "." E.AssocLeft]
|
|
> ,[postfix' arrayPostfix]
|
|
> ,[prefixSym "+", prefixSym "-"]
|
|
> ,[binarySym "^" E.AssocLeft]
|
|
> ,[binarySym "*" E.AssocLeft
|
|
> ,binarySym "/" E.AssocLeft
|
|
> ,binarySym "%" E.AssocLeft]
|
|
> ,[binarySym "+" E.AssocLeft
|
|
> ,binarySym "-" E.AssocLeft]
|
|
> ,[binarySym ">=" E.AssocNone
|
|
> ,binarySym "<=" E.AssocNone
|
|
> ,binarySym "!=" E.AssocRight
|
|
> ,binarySym "<>" E.AssocRight
|
|
> ,binarySym "||" E.AssocRight
|
|
> ,prefixSym "~"
|
|
> ,binarySym "&" E.AssocRight
|
|
> ,binarySym "|" E.AssocRight
|
|
> ,binaryKeyword "like" E.AssocNone
|
|
> ,binaryKeyword "overlaps" E.AssocNone]
|
|
> ++ map (`binaryKeywords` E.AssocNone)
|
|
> ["not like"
|
|
> ,"is similar to"
|
|
> ,"is not similar to"
|
|
> ,"is distinct from"
|
|
> ,"is not distinct from"]
|
|
> ++ map postfixKeywords
|
|
> ["is null"
|
|
> ,"is not null"
|
|
> ,"is true"
|
|
> ,"is not true"
|
|
> ,"is false"
|
|
> ,"is not false"
|
|
> ,"is unknown"
|
|
> ,"is not unknown"]
|
|
> -- have to use try with inSuffix because of a conflict
|
|
> -- with 'in' in position function
|
|
> -- between also has a try in it to deal with 'not'
|
|
> -- ambiguity
|
|
> ++ [E.Postfix $ try inSuffix,E.Postfix betweenSuffix]
|
|
> ]
|
|
> ++
|
|
> [[binarySym "<" E.AssocNone
|
|
> ,binarySym ">" E.AssocNone]
|
|
> ,[binarySym "=" E.AssocRight]
|
|
> ,[prefixKeyword "not"]]
|
|
> ++
|
|
> if bExpr then [] else [[binaryKeyword "and" E.AssocLeft]]
|
|
> ++
|
|
> [[binaryKeyword "or" E.AssocLeft]]
|
|
> where
|
|
> binarySym nm assoc = binary (symbol_ nm) nm assoc
|
|
> binaryKeyword nm assoc = binary (keyword_ nm) nm assoc
|
|
> -- use try with the multi keywords because of shared
|
|
> -- prefixes to the sets of keywords. Should left factor
|
|
> -- somehow
|
|
> binaryKeywords nm assoc = binary (try $ mapM_ keyword_ (words nm)) nm assoc
|
|
> binary p nm assoc =
|
|
> E.Infix (p >> return (\a b -> BinOp a (Name nm) b)) assoc
|
|
> prefixKeyword nm = prefix (keyword_ nm) nm
|
|
> prefixSym nm = prefix (symbol_ nm) nm
|
|
> prefix p nm = prefix' (p >> return (PrefixOp (Name nm)))
|
|
> postfixKeywords nm = postfix (try $ mapM_ keyword_ (words nm)) nm
|
|
> postfix p nm = postfix' (p >> return (PostfixOp (Name nm)))
|
|
> -- hack from here
|
|
> -- http://stackoverflow.com/questions/10475337/parsec-expr-repeated-prefix-postfix-operator-not-supported
|
|
> -- not implemented properly yet
|
|
> -- I don't think this will be enough for all cases
|
|
> -- at least it works for 'not not a'
|
|
> -- ok: "x is not true is not true"
|
|
> -- no work: "x is not true is not null"
|
|
> prefix' p = E.Prefix . chainl1 p $ return (.)
|
|
> postfix' p = E.Postfix . chainl1 p $ return (flip (.))
|
|
|
|
== value expressions
|
|
|
|
TODO:
|
|
left factor stuff which starts with identifier
|
|
|
|
This parses most of the value exprs.The order of the parsers and use
|
|
of try is carefully done to make everything work. It is a little
|
|
fragile and could at least do with some heavy explanation.
|
|
|
|
> valueExpr :: Parser ValueExpr
|
|
> valueExpr = E.buildExpressionParser (opTable False) term
|
|
|
|
> term :: Parser ValueExpr
|
|
> term = choice [literal
|
|
> ,parameter
|
|
> ,hostParameter
|
|
> ,caseValue
|
|
> ,cast
|
|
> ,arrayCtor
|
|
> ,specialOpKs
|
|
> ,parensTerm
|
|
> ,subquery
|
|
> ,star
|
|
> ,idenPrefixTerm]
|
|
> <?> "value expression"
|
|
|
|
expose the b expression for window frame clause range between
|
|
|
|
> valueExprB :: Parser ValueExpr
|
|
> valueExprB = E.buildExpressionParser (opTable True) term
|
|
|
|
|
|
-------------------------------------------------
|
|
|
|
= query expressions
|
|
|
|
== select lists
|
|
|
|
> selectItem :: Parser (ValueExpr,Maybe Name)
|
|
> selectItem = (,) <$> valueExpr <*> optionMaybe als
|
|
> where als = optional (keyword_ "as") *> name
|
|
|
|
> selectList :: Parser [(ValueExpr,Maybe Name)]
|
|
> selectList = commaSep1 selectItem
|
|
|
|
== from
|
|
|
|
Here is the rough grammar for joins
|
|
|
|
tref
|
|
(cross | [natural] ([inner] | (left | right | full) [outer])) join
|
|
tref
|
|
[on expr | using (...)]
|
|
|
|
> from :: Parser [TableRef]
|
|
> from = keyword_ "from" *> commaSep1 tref
|
|
> where
|
|
> tref = nonJoinTref >>= optionSuffix joinTrefSuffix
|
|
> nonJoinTref = choice
|
|
> [parens $ choice
|
|
> [TRQueryExpr <$> queryExpr
|
|
> ,TRParens <$> tref]
|
|
> ,TRLateral <$> (keyword_ "lateral"
|
|
> *> nonJoinTref)
|
|
> ,do
|
|
> n <- name
|
|
> choice [TRFunction n
|
|
> <$> parens (commaSep valueExpr)
|
|
> ,do
|
|
> choice [TRQualified n <$> (symbol "." >> name)
|
|
> ,return $ TRSimple n]]]
|
|
> >>= optionSuffix aliasSuffix
|
|
> aliasSuffix j = option j (TRAlias j <$> alias)
|
|
> joinTrefSuffix t = (do
|
|
> nat <- option False (True <$ keyword_ "natural")
|
|
> TRJoin t <$> joinType
|
|
> <*> nonJoinTref
|
|
> <*> optionMaybe (joinCondition nat))
|
|
> >>= optionSuffix joinTrefSuffix
|
|
|
|
> joinType :: Parser JoinType
|
|
> joinType = choice
|
|
> [JCross <$ keyword_ "cross" <* keyword_ "join"
|
|
> ,JInner <$ keyword_ "inner" <* keyword_ "join"
|
|
> ,JLeft <$ keyword_ "left"
|
|
> <* optional (keyword_ "outer")
|
|
> <* keyword_ "join"
|
|
> ,JRight <$ keyword_ "right"
|
|
> <* optional (keyword_ "outer")
|
|
> <* keyword_ "join"
|
|
> ,JFull <$ keyword_ "full"
|
|
> <* optional (keyword_ "outer")
|
|
> <* keyword_ "join"
|
|
> ,JInner <$ keyword_ "join"]
|
|
|
|
> joinCondition :: Bool -> Parser JoinCondition
|
|
> joinCondition nat =
|
|
> choice [guard nat >> return JoinNatural
|
|
> ,keyword_ "on" >> JoinOn <$> valueExpr
|
|
> ,keyword_ "using" >> JoinUsing <$> parens (commaSep1 name)
|
|
> ]
|
|
|
|
> alias :: Parser Alias
|
|
> alias = Alias <$> tableAlias <*> columnAliases
|
|
> where
|
|
> tableAlias = optional (keyword_ "as") *> name
|
|
> columnAliases = optionMaybe $ parens $ commaSep1 name
|
|
|
|
== simple other parts
|
|
|
|
Parsers for where, group by, having, order by and limit, which are
|
|
pretty trivial.
|
|
|
|
> whereClause :: Parser ValueExpr
|
|
> whereClause = keyword_ "where" *> valueExpr
|
|
|
|
> groupByClause :: Parser [GroupingExpr]
|
|
> groupByClause = keyword_ "group" *> keyword_ "by"
|
|
> *> commaSep1 groupingExpression
|
|
> where
|
|
> groupingExpression =
|
|
> choice
|
|
> [keyword_ "cube" >>
|
|
> Cube <$> parens (commaSep groupingExpression)
|
|
> ,keyword_ "rollup" >>
|
|
> Rollup <$> parens (commaSep groupingExpression)
|
|
> ,GroupingParens <$> parens (commaSep groupingExpression)
|
|
> ,keyword_ "grouping" >> keyword_ "sets" >>
|
|
> GroupingSets <$> parens (commaSep groupingExpression)
|
|
> ,SimpleGroup <$> valueExpr
|
|
> ]
|
|
|
|
> having :: Parser ValueExpr
|
|
> having = keyword_ "having" *> valueExpr
|
|
|
|
> orderBy :: Parser [SortSpec]
|
|
> orderBy = keyword_ "order" *> keyword_ "by" *> commaSep1 ob
|
|
> where
|
|
> ob = SortSpec
|
|
> <$> valueExpr
|
|
> <*> option Asc (choice [Asc <$ keyword_ "asc"
|
|
> ,Desc <$ keyword_ "desc"])
|
|
> <*> option NullsOrderDefault
|
|
> (keyword_ "nulls" >>
|
|
> choice [NullsFirst <$ keyword "first"
|
|
> ,NullsLast <$ keyword "last"])
|
|
|
|
allows offset and fetch in either order
|
|
+ postgresql offset without row(s) and limit instead of fetch also
|
|
|
|
> offsetFetch :: Parser (Maybe ValueExpr, Maybe ValueExpr)
|
|
> offsetFetch = permute ((,) <$?> (Nothing, Just <$> offset)
|
|
> <|?> (Nothing, Just <$> fetch))
|
|
|
|
> offset :: Parser ValueExpr
|
|
> offset = keyword_ "offset" *> valueExpr
|
|
> <* option () (choice [keyword_ "rows"
|
|
> ,keyword_ "row"])
|
|
|
|
> fetch :: Parser ValueExpr
|
|
> fetch = choice [ansiFetch, limit]
|
|
> where
|
|
> ansiFetch = keyword_ "fetch" >>
|
|
> choice [keyword_ "first",keyword_ "next"]
|
|
> *> valueExpr
|
|
> <* choice [keyword_ "rows",keyword_ "row"]
|
|
> <* keyword_ "only"
|
|
> limit = keyword_ "limit" *> valueExpr
|
|
|
|
== common table expressions
|
|
|
|
> with :: Parser QueryExpr
|
|
> with = keyword_ "with" >>
|
|
> With <$> option False (True <$ keyword_ "recursive")
|
|
> <*> commaSep1 withQuery <*> queryExpr
|
|
> where
|
|
> withQuery =
|
|
> (,) <$> (alias <* keyword_ "as")
|
|
> <*> parens queryExpr
|
|
|
|
== query expression
|
|
|
|
This parser parses any query expression variant: normal select, cte,
|
|
and union, etc..
|
|
|
|
> queryExpr :: Parser QueryExpr
|
|
> queryExpr =
|
|
> choice [with
|
|
> ,choice [values,table, select]
|
|
> >>= optionSuffix queryExprSuffix]
|
|
> where
|
|
> select = keyword_ "select" >>
|
|
> mkSelect
|
|
> <$> (fromMaybe All <$> duplicates)
|
|
> <*> selectList
|
|
> <*> optionMaybe tableExpression
|
|
> mkSelect d sl Nothing =
|
|
> makeSelect{qeSetQuantifier = d, qeSelectList = sl}
|
|
> mkSelect d sl (Just (TableExpression f w g h od ofs fe)) =
|
|
> Select d sl f w g h od ofs fe
|
|
> values = keyword_ "values"
|
|
> >> Values <$> commaSep (parens (commaSep valueExpr))
|
|
> table = keyword_ "table" >> Table <$> name
|
|
|
|
local data type to help with parsing the bit after the select list,
|
|
called 'table expression' in the ansi sql grammar. Maybe this should
|
|
be in the public syntax?
|
|
|
|
> data TableExpression
|
|
> = TableExpression
|
|
> {_teFrom :: [TableRef]
|
|
> ,_teWhere :: Maybe ValueExpr
|
|
> ,_teGroupBy :: [GroupingExpr]
|
|
> ,_teHaving :: Maybe ValueExpr
|
|
> ,_teOrderBy :: [SortSpec]
|
|
> ,_teOffset :: Maybe ValueExpr
|
|
> ,_teFetchFirst :: Maybe ValueExpr}
|
|
|
|
> tableExpression :: Parser TableExpression
|
|
> tableExpression =
|
|
> mkTe <$> from
|
|
> <*> optionMaybe whereClause
|
|
> <*> option [] groupByClause
|
|
> <*> optionMaybe having
|
|
> <*> option [] orderBy
|
|
> <*> offsetFetch
|
|
> where
|
|
> mkTe f w g h od (ofs,fe) =
|
|
> TableExpression f w g h od ofs fe
|
|
|
|
> queryExprSuffix :: QueryExpr -> Parser QueryExpr
|
|
> queryExprSuffix qe =
|
|
> (CombineQueryExpr qe
|
|
> <$> (choice
|
|
> [Union <$ keyword_ "union"
|
|
> ,Intersect <$ keyword_ "intersect"
|
|
> ,Except <$ keyword_ "except"] <?> "set operator")
|
|
> <*> (fromMaybe Distinct <$> duplicates)
|
|
> <*> option Respectively
|
|
> (Corresponding <$ keyword_ "corresponding")
|
|
> <*> queryExpr)
|
|
> >>= optionSuffix queryExprSuffix
|
|
|
|
wrapper for query expr which ignores optional trailing semicolon.
|
|
|
|
> topLevelQueryExpr :: Parser QueryExpr
|
|
> topLevelQueryExpr =
|
|
> queryExpr >>= optionSuffix ((semi *>) . return)
|
|
|
|
wrapper to parse a series of query exprs from a single source. They
|
|
must be separated by semicolon, but for the last expression, the
|
|
trailing semicolon is optional.
|
|
|
|
> queryExprs :: Parser [QueryExpr]
|
|
> queryExprs =
|
|
> (:[]) <$> queryExpr
|
|
> >>= optionSuffix ((semi *>) . return)
|
|
> >>= optionSuffix (\p -> (p++) <$> queryExprs)
|
|
|
|
------------------------------------------------
|
|
|
|
= lexing parsers
|
|
|
|
whitespace parser which skips comments also
|
|
|
|
> whitespace :: Parser ()
|
|
> whitespace =
|
|
> choice [simpleWhitespace *> whitespace
|
|
> ,lineComment *> whitespace
|
|
> ,blockComment *> whitespace
|
|
> ,return ()] <?> "whitespace"
|
|
> where
|
|
> lineComment = try (string "--")
|
|
> *> manyTill anyChar (void (char '\n') <|> eof)
|
|
> blockComment = -- no nesting of block comments in SQL
|
|
> try (string "/*")
|
|
> -- try used here so it doesn't fail when we see a
|
|
> -- '*' which isn't followed by a '/'
|
|
> *> manyTill anyChar (try $ string "*/")
|
|
> -- use many1 so we can more easily avoid non terminating loops
|
|
> simpleWhitespace = void $ many1 (oneOf " \t\n")
|
|
|
|
> lexeme :: Parser a -> Parser a
|
|
> lexeme p = p <* whitespace
|
|
|
|
> integer :: Parser Integer
|
|
> integer = read <$> lexeme (many1 digit) <?> "integer"
|
|
|
|
|
|
number literals
|
|
|
|
here is the rough grammar target:
|
|
|
|
digits
|
|
digits.[digits][e[+-]digits]
|
|
[digits].digits[e[+-]digits]
|
|
digitse[+-]digits
|
|
|
|
numbers are parsed to strings, not to a numeric type. This is to avoid
|
|
making a decision on how to represent numbers, the client code can
|
|
make this choice.
|
|
|
|
> numberLiteral :: Parser String
|
|
> numberLiteral = lexeme (
|
|
> (choice [int
|
|
> >>= optionSuffix dot
|
|
> >>= optionSuffix fracts
|
|
> >>= optionSuffix expon
|
|
> ,fract "" >>= optionSuffix expon])
|
|
> <* notFollowedBy (alphaNum <|> char '.'))
|
|
> <?> "number literal"
|
|
> where
|
|
> int = many1 digit
|
|
> fract p = dot p >>= fracts
|
|
> dot p = (p++) <$> string "."
|
|
> fracts p = (p++) <$> int
|
|
> expon p = concat <$> sequence
|
|
> [return p
|
|
> ,(:[]) <$> oneOf "eE"
|
|
> ,option "" (string "+" <|> string "-")
|
|
> ,int]
|
|
|
|
|
|
> identifier :: Parser String
|
|
> identifier = lexeme ((:) <$> firstChar <*> many nonFirstChar)
|
|
> <?> "identifier"
|
|
> where
|
|
> firstChar = letter <|> char '_'
|
|
> nonFirstChar = digit <|> firstChar
|
|
|
|
> quotedIdentifier :: Parser String
|
|
> quotedIdentifier = char '"' *> manyTill anyChar doubleQuote
|
|
> <?> "identifier"
|
|
|
|
TODO: add "" inside quoted identifiers
|
|
|
|
parses an identifier with a : prefix. The : isn't included in the
|
|
return value
|
|
|
|
> hostParameterToken :: Parser String
|
|
> hostParameterToken = lexeme $ char ':' *> identifier
|
|
|
|
todo: work out the symbol parsing better
|
|
|
|
> symbol :: String -> Parser String
|
|
> symbol s = try (lexeme $ do
|
|
> u <- choice (many1 (char '.') :
|
|
> map (try . string) [">=","<=","!=","<>","||"]
|
|
> ++ map (string . (:[])) "+-^*/%~&|<>=")
|
|
> guard (s == u)
|
|
> return s)
|
|
> <?> s
|
|
|
|
> questionMark :: Parser Char
|
|
> questionMark = lexeme $ char '?'
|
|
|
|
> openParen :: Parser Char
|
|
> openParen = lexeme $ char '('
|
|
|
|
> closeParen :: Parser Char
|
|
> closeParen = lexeme $ char ')'
|
|
|
|
> openBracket :: Parser Char
|
|
> openBracket = lexeme $ char '['
|
|
|
|
> closeBracket :: Parser Char
|
|
> closeBracket = lexeme $ char ']'
|
|
|
|
|
|
> comma :: Parser Char
|
|
> comma = lexeme $ char ','
|
|
|
|
> semi :: Parser Char
|
|
> semi = lexeme $ char ';'
|
|
|
|
> doubleQuote :: Parser Char
|
|
> doubleQuote = lexeme $ char '"'
|
|
|
|
> --stringToken :: Parser String
|
|
> --stringToken = lexeme (char '\'' *> manyTill anyChar (char '\''))
|
|
> -- todo: tidy this up, add the prefixes stuff, and add the multiple
|
|
> -- string stuff
|
|
> stringToken :: Parser String
|
|
> stringToken =
|
|
> lexeme (char '\'' *> manyTill anyChar (char '\'')
|
|
> >>= optionSuffix moreString)
|
|
> <?> "string"
|
|
> where
|
|
> moreString s0 = choice
|
|
> [-- handle two adjacent quotes
|
|
> do
|
|
> void $ char '\''
|
|
> s <- manyTill anyChar (char '\'')
|
|
> optionSuffix moreString (s0 ++ "'" ++ s)
|
|
> ,-- handle string in separate parts
|
|
> -- e.g. 'part 1' 'part 2'
|
|
> do
|
|
> try (whitespace <* char '\'')
|
|
> s <- manyTill anyChar (char '\'')
|
|
> optionSuffix moreString (s0 ++ s)
|
|
> ]
|
|
|
|
= helper functions
|
|
|
|
> keyword :: String -> Parser String
|
|
> keyword k = try (do
|
|
> i <- identifier
|
|
> guard (map toLower i == k)
|
|
> return k) <?> k
|
|
|
|
> parens :: Parser a -> Parser a
|
|
> parens = between openParen closeParen
|
|
|
|
> brackets :: Parser a -> Parser a
|
|
> brackets = between openBracket closeBracket
|
|
|
|
> commaSep :: Parser a -> Parser [a]
|
|
> commaSep = (`sepBy` comma)
|
|
|
|
> keyword_ :: String -> Parser ()
|
|
> keyword_ = void . keyword
|
|
|
|
> symbol_ :: String -> Parser ()
|
|
> symbol_ = void . symbol
|
|
|
|
> commaSep1 :: Parser a -> Parser [a]
|
|
> commaSep1 = (`sepBy1` comma)
|
|
|
|
a possible issue with the option suffix is that it enforces left
|
|
associativity when chaining it recursively. Have to review
|
|
all these uses and figure out if any should be right associative
|
|
instead, and create an alternative suffix parser
|
|
|
|
> optionSuffix :: (a -> Parser a) -> a -> Parser a
|
|
> optionSuffix p a = option a (p a)
|
|
|
|
> identifierBlacklist :: [String] -> Parser String
|
|
> identifierBlacklist bl = try (do
|
|
> i <- identifier
|
|
> guard (map toLower i `notElem` bl)
|
|
> return i)
|
|
> <?> "identifier"
|
|
|
|
> blacklist :: [String]
|
|
> blacklist =
|
|
> [-- case
|
|
> "case", "when", "then", "else", "end"
|
|
> ,--join
|
|
> "natural","inner","outer","cross","left","right","full","join"
|
|
> ,"on","using","lateral"
|
|
> ,"from","where","group","having","order","limit", "offset", "fetch"
|
|
> ,"as","in"
|
|
> ,"except", "intersect", "union"
|
|
> ]
|
|
|
|
These blacklisted names are mostly needed when we parse something with
|
|
an optional alias, e.g. select a a from t. If we write select a from
|
|
t, we have to make sure the from isn't parsed as an alias. I'm not
|
|
sure what other places strictly need the blacklist, and in theory it
|
|
could be tuned differently for each place the identifierString/
|
|
identifier parsers are used to only blacklist the bare minimum.
|
|
|
|
The standard has a weird mix of reserved keywords and unreserved
|
|
keywords (I'm not sure what exactly being an unreserved keyword
|
|
means).
|
|
|
|
--------------------------------------------
|
|
|
|
= helper functions
|
|
|
|
> setPos :: Maybe (Int,Int) -> Parser ()
|
|
> setPos Nothing = return ()
|
|
> setPos (Just (l,c)) = fmap f getPosition >>= setPosition
|
|
> where f = flip setSourceColumn c
|
|
> . flip setSourceLine l
|
|
|
|
> convParseError :: String -> P.ParseError -> ParseError
|
|
> convParseError src e =
|
|
> ParseError
|
|
> {peErrorString = show e
|
|
> ,peFilename = sourceName p
|
|
> ,pePosition = (sourceLine p, sourceColumn p)
|
|
> ,peFormattedError = formatError src e
|
|
> }
|
|
> where
|
|
> p = errorPos e
|
|
|
|
format the error more nicely: emacs format for positioning, plus
|
|
context
|
|
|
|
> formatError :: String -> P.ParseError -> String
|
|
> formatError src e =
|
|
> sourceName p ++ ":" ++ show (sourceLine p)
|
|
> ++ ":" ++ show (sourceColumn p) ++ ":"
|
|
> ++ context
|
|
> ++ show e
|
|
> where
|
|
> context =
|
|
> let lns = take 1 $ drop (sourceLine p - 1) $ lines src
|
|
> in case lns of
|
|
> [x] -> "\n" ++ x ++ "\n"
|
|
> ++ replicate (sourceColumn p - 1) ' ' ++ "^\n"
|
|
> _ -> ""
|
|
> p = errorPos e
|