Let\'s have following parser:
parser :: GenParser Char st String
parser = choice (fmap (try . string) [\"head\", \"tail\", \"tales\"]
<
Here's what I've got with Parsec:
λ> let parser = choice $ fmap (try . string) ["head", "tail", "tales"]
λ> parseTest parser "ta"
parse error at (line 1, column 1):
unexpected "t"
expecting "head", "tail" or "tales"
If you care to try modern version of Parsec — Megaparsec, you will end up with:
λ> let parser = choice $ fmap (try . string) ["head", "tail", "tales"]
λ> parseTest parser "ta"
1:1:
unexpected "ta" or 't'
expecting "head", "tail", or "tales"
What's going on here? First, when we parse ordered collection of characters,
like with string, we display incorrect input completely. This is much
better in our opinion because:
λ> parseTest (string "when" <* eof) "well"
1:1:
unexpected "we"
expecting "when"
We're pointing to beginning of the word and we show the whole thing that is
not correct (up to first mismatching character) and the whole thing we
expect. This is more readable in my opinion. Only parsers built on tokens
work this way (that is, when we're trying to match fixed string,
case-insensitive variant is available).
Then, what about unexpected "ta" or 't', why do we get 't' part? This is
also absolutely correct, because with your collection of alternatives, the
first letter 't' can be also unexpected by itself because you have
an alternative that doesn't start with 't'. Let's see another example:
λ> let parser = choice $ fmap (try . string) ["tall", "tail", "tales"]
λ> parseTest parser "ta"
1:1:
unexpected "ta"
expecting "tail", "tales", or "tall"
Or how about:
λ> parseTest (try (string "lexer") <|> string "lexical") "lex"
1:1:
unexpected "lex"
expecting "lexer" or "lexical"
Parsec:
λ> parseTest (try (string "lexer") <|> string "lexical") "lex"
parse error at (line 1, column 1):
unexpected end of input
expecting "lexical"
Why take pains to make it work when it can “just work”?
There are many other great things about Megaparsec, if you're interested, you can learn more about it here. It's hard to compete with Parsec, but we have written our own tutorials and our docs are very good.