dash in range not correctly parsed

[neongreen]

ChrisKuklewicz/regex-tdfa#24, originally reported by @pjljvandelaar

---

As specified in
https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap09.html

the expression "[--@]" matches any of the characters between '-' and '@' inclusive

However, ("@" =~ "[--@]") results in

Explict error in module Text.Regex.TDFA.String : 
Text.Regex.TDFA.String died: parseRegex for Text.Regex.TDFA.String failed:"[--@]" (line 1, column 4):
       unexpected A dash is in the wrong place in a bracket
       expecting "]"
       CallStack (from HasCallStack):
         error, called at .\Text\Regex\TDFA\Common.hs:29:3 in regex-tdfa-1.2.3.1-DVMXTrvIFHgDCky8s203W0:Text.Regex.TDFA.Common)

[andreasabel]

The error is also triggered for an empty range, e.g. [1-0]:

ghci> ("1" =~ "[1-0]") :: String
"*** Exception: Explict error in module Text.Regex.TDFA.String : Text.Regex.TDFA.String died: parseRegex for Text.Regex.TDFA.String failed:"[1-0]" (line 1, column 4):
unexpected A dash is in the wrong place in a bracket
expecting "]"
CallStack (from HasCallStack):
  error, called at lib/Text/Regex/TDFA/Common.hs:31:3 in regex-tdfa-1.3.1.3-inplace:Text.Regex.TDFA.Common

I think the intention was to give another error in this case:

regex-tdfa/lib/Text/Regex/TDFA/ReadRegex.hs

Lines 130 to 137 in 85c89c9

	p_set_elem_range = try $ do
	start <- noneOf "]-"
	_ <- char '-'
	end <- noneOf "]"
	-- bug fix: check start <= end before "return (BEChars [start..end])"
	if start <= end
	then return (BEChars [start..end])
	else unexpected "End point of dashed character range is less than starting point"

However, this error is caught by <|>:

regex-tdfa/lib/Text/Regex/TDFA/ReadRegex.hs

Lines 118 to 119 in 85c89c9

	p_set_elem = p_set_elem_class <|> p_set_elem_equiv <|> p_set_elem_coll
	<|> p_set_elem_range <|> p_set_elem_char <?> "Failed to parse bracketed string"

So we end up with the error triggered by the last alternative:

regex-tdfa/lib/Text/Regex/TDFA/ReadRegex.hs

Lines 139 to 144 in 85c89c9

	p_set_elem_char = do
	c <- noneOf "]"
	when (c == '-') $ do
	atEnd <- (lookAhead (char ']') >> return True) <|> (return False)
	when (not atEnd) (unexpected "A dash is in the wrong place in a bracket")
	return (BEChar c)

[andreasabel]

The solution to the OP is to simply not further restrict the appearance of - in a bracketed expression. E.g. [--] would be valid, meaning either - or -. So, we delete this check:

regex-tdfa/lib/Text/Regex/TDFA/ReadRegex.hs

Lines 141 to 143 in 85c89c9

	when (c == '-') $ do
	atEnd <- (lookAhead (char ']') >> return True) <|> (return False)
	when (not atEnd) (unexpected "A dash is in the wrong place in a bracket")