Isn't csv a text-based rather than binary(bytestring)-based format?

[tysonzero]

The spec appears to only mention text, and the specific binary encoding / charset of that text seems out of scope.

Accordingly it seems to me as though cassava should generally be dealing with Text instead of ByteString, perhaps with a Data.Csv.Utf8 module for just directly treating ByteString values as encoded utf-8 text.

[jchia]

An ASCII delimiter in the undecoded ByteString corresponds a delimiter in the corresponding UTF-8-decoded Text, so under UTF-8 encoding there is no problem with making a mistake with delimiters.

However, the user is forced to use UTF-8 if there are Text/ShortText/Char fields (cassava assumes UTF-8). If he wants to use another text encoding, he needs to use ByteString fields and do the ByteString-Text conversion separately. Alternatively, he can perform transcoding between UTF-8 and the other text encoding, using UTF-8-encoded ByteStrings when interfacing with cassava.

I have no idea about the performance characteristics of each alternative, though, including the proposed Data.Csv.Utf8.

[tysonzero]

To be clear Data.Csv.Utf8 would just be the current implementation. The module name should make it clear that using utf8 for the ByteString arguments is safe, and that non-utf8 arguments should expect edge cases and require additional care.

I am also unsure of how the Data.Csv or Data.Csv.Text or whatever Text-based alternative would change performance.