Per-line parser reporting per-line errors

[MaxGabriel]

Right now, the functions in Data.Csv generally return Either String (Vector a) or a similar variant, which is great in most cases.

If you want to get errors on a per-line basis, one needs to use Data.Csv.Incremental*. This is unfortunate because the complexity of it is significantly higher (because it also is supporting interleaving IO and incrementally feeding data to the parser).

I think adding convenience functions for taking a ByteString and returning Vector (Either String a), without going through the Incremental functions, could be a good addition. The main use case I have in mind for those is providing better error messages for user-provided CSVs.

• If my read of the Cassava docs is right

[andreasabel]

I think adding convenience functions for taking a ByteString and returning Vector (Either String a),

Yeah, that is thinkable. The current parser (e.g. for without header), https://github.com/haskell-hvr/cassava/blob/c821c8366ac4ce4ee3929e315ba37e694ad56f04/src/Data/Csv/Parser.hs#L69-L75 uses the parser modifier sepByEndOfLine': https://github.com/haskell-hvr/cassava/blob/c821c8366ac4ce4ee3929e315ba37e694ad56f04/src/Data/Csv/Parser.hs#L92-L105 So, it is just a single parse with a single error returned. If you want per-line parsing, you will first have to split the input using a variant of sepByEndOfLine' and then map a line parser over it.

PR welcome, if it comes with benchmarks comparing the performance of single parse versus per line parse.

[MaxGabriel]

Ok, I'm not interested in the performance implications of this, so just going to close

[andreasabel]

Let's leave it open, maybe others are interested.