I really have to second this. It's fast, it's smart, it handles almost anything, and it reliably gives good error messages. It's an important part of my data science toolbelt.
The problem is that CSV is basically non discoverable, as it has no metadata eg about encoding or delimiters or locale... Thats why Excel gives you a sample and lets you change delimiters etc. You can guess, or you can write something that seems to work for a particular input source. But it is best avoided.
I'm mostly using Python and trying to learn Go, both have great csv parsers and I used the one in PHP. Given the size and features present in .Net I was really suprised to not find a csv parser.
