some issues with migration from Protolude.Conv to Protolude.ConvertText

[robx]

some issues with migration from Protolude.Conv to Protolude.ConvertText

[sdiehl]

So this is a known breaking change in 0.3.0. since the old interface exposed some partial functions which would fail on encoding errors.

The principled fix would be to go through your codebase and replace all type errors with the function that explicitly handles the UnicodeException with Either.

decodeUtf8' :: ByteString -> Either UnicodeException Text

The less principled fix is to use old interface Protolude.Conv which is left as a compatability layer but it is silently partial. So your business logic may blow up if you pass it malformed strings. Thus the fix in this release.

[robx]

I understand (I think), but that only really addresses the third point in the list, right? It seems unfortunate that easy ways to encode to UTF8 or to convert between lazy and strict bytestrings were lost.

I'm wondering now whether instances ConvertText _ ByteString would be an option. Though I realize that UTF8-encoded text is not what's understood to be text in the current sense of ConvertText.

[sdiehl]

I suppose the lazy to strict bytestring conversion exposes potential partiality, but the strict to lazy bytestring should always be safe. I'll look to restoring those to the TextConvert class.

[domenkozar]

The loss of ConvertText Text ByteString is quite sad, what was partial about that conversion?

[adamwespiser]

ConvertText Text ByteString would use encodeUtf8 which is non-partial.

Similarly, the following instances could be added as well:

ConvertText Text ByteString
ConvertText String ByteString
ConvertText Lazy.Text ByteString
ConvertText Text Lazy.ByteString
ConvertText String Lazy.ByteString
ConvertText Lazy.Text Lazy.ByteString

As a convenience, there's definitely merit, but I'm not sure if it would be confusing that you can convert to a ByteString, but not from one using ConvertText. In that way, ConvertText is not sufficient for working with ByteString to Text / String conversions.

I'm not sure what the original intention was here (beyond removing the partial function decodeUtf8 which can cause nasty bugs).
I'm inclined to add the above instances back in. In the meantime, there is toUtf8 and toUtf8Lazy.