takeP fails with "unexpected end of input" for negative number of tokens

[olafklinke]

parseTest (takeP Nothing (-1) :: Parsec Void String String) "foo"
expected behavior: return the empty string, similar to Prelude.take
actual behavior: fails with a message unexpected end of input.

While a negative number of tokens is clearly a non-sensical argument to that function, I'd expect either a more informative parse error along the lines of "can not take negative number of tokens" or a behavior analogous to Prelude.take where take (-1) = take 0.

The implementation of takeP for ParsecT explicitly uses EndOfInput, so this seems to be intended. Why? The Haddocks say:

It's guaranteed that if the parser succeeds, the requested number of tokens will be returned.

This is logically equivalent to:

If the requested number of tokens can not be returned, then the parser fails.

So yes, the parser honors the spec, but in a way inconsistent with its name.