Improve performance about rdf4h

[wismill]

@robstewart57 what do you think about this proposal?

I would like to keep only Megaparsec for the parsing, as it is now very fast and robust.

[robstewart57]

@wismill the idea of #36 was to generalise rdf4h across numerous parsers, specificlaly attoparsec and parsec. E.g.

-- |'NTriplesParser' is an instance of 'RdfParser' using parsec based parsers.
instance RdfParser NTriplesParser where
  parseString _  = parseStringParsec
  parseFile   _  = parseFileParsec
  parseURL    _  = parseURLParsec

-- |'NTriplesParser' is an instance of 'RdfParser'.
instance RdfParser NTriplesParserCustom where
  parseString (NTriplesParserCustom Parsec)     = parseStringParsec
  parseString (NTriplesParserCustom Attoparsec) = parseStringAttoparsec
  parseFile   (NTriplesParserCustom Parsec)     = parseFileParsec
  parseFile   (NTriplesParserCustom Attoparsec) = parseFileAttoparsec
  parseURL    (NTriplesParserCustom Parsec)     = parseURLParsec
  parseURL    (NTriplesParserCustom Attoparsec) = parseURLAttoparsec

rdf4h/src/Text/RDF/RDF4H/NTriplesParser.hs

Line 44 in 896e21e

-- |'NTriplesParser' is an instance of 'RdfParser' using parsec based parsers.

.

This functionality comes from this PR: #36

The motivation of this flexibility between parsers is that each of they have trade offs. E.g. "Megaparsec vs Attoparsec" : https://github.com/mrkkrp/megaparsec#megaparsec-vs-attoparsec

Attoparsec is sometimes faster but not that feature-rich. It should be used when you want to process large amounts of data where performance matters more than quality of error messages.

This is a realistic assumption when working with RDF data, which might be 10s/100s MegaBytes or millions of triples.

There's megaparsec instances in parsers-megaparsec that we could use: https://hackage.haskell.org/package/parsers-megaparsec

The issue arises when attoparsec, parsec and megaparsec instances of the typeclasses in the parsers have different parsing semantics (which presumably shouldn't happen), meaning the rdf4h tests against w2c-tests might pass for one instance of the parsers typeclasses, e.g. parsec, but fail for others, megaparsec.

[wismill]

Ok, but at least we could drop parsec and keep only attoparsec and megaparsec. Then we could use parser-combinators.

I think we should also provide a way to stream parsing results. As there are several framework for this, I wonder if we should provide new packages such as: rdf4h-pipes, rdf4h-conduit, rdf4h-streamly, etc.

[robstewart57]

@wismill

I think we should also provide a way to stream parsing results. As there are several framework for this, I wonder if we should provide new packages such as: rdf4h-pipes, rdf4h-conduit, rdf4h-streamly, etc.

I would also really like to see this!