Major version changes (v3)

[Porges]

Ideas for v3:

• Input type should be Text, not ByteString (see Use Text instead of ByteString, since emails can be unicode #9)
• Test suite will be inherited from Dominic Sayer's isemail, and won't be inlined in the code.
• The basic EmailAddress should not provide individual access to localPart/domainPart, but instead be a newtype around Text. This will facilitate loading/storing from unvalidated stores in a simpler fashion than is currently possible (e.g. at the moment you can use unsafeEmailAddress but it takes the two parts separately, not a single email address string).
• Lift ParseOptions to type level, per this comment.
• The default parsing mode will change to something saner that excludes "obsolete" syntaxes (these can be accessed using the "Detailed" module).
• Support internationalized emails.
  • In local parts
    • Add tests for each bit (atext, ctext, etc – ignore dtext since we'll do proper handling)
  • In domains (this probably requires an IDNA2008 implementation)
• For those who want full analysis, there will be a separate "Detailed" module which can break apart an email address. (Or, just provide functions to break down the EmailAddress type...)
  • This module will provide DomainPart = HostName | IP.
    TODO: what modules to depend on for IP address types? There are currently ip, network-ip, others? Do any of them expose parsers from Text?
    Resolution: I chose ip for now, since it provides Attoparsec Text parsers.
    To pass tests I need to close these issues:
    • Support 'mixed' IPv6 representations
      • PR submitted – released in 1.3.0
    • Support for converting IPv6 to Text
      • PR submitted
• Error messages should be improved, and this checked by tests.
• Consider exposing an FFI interface for other languages to use.
• Consider improved equality (see Use case-insensitive ByteString in EmailAddress #36)
  this should fall out of better domain parsing
• Consider switching to Megaparsec, for better error messages?
• Consider which instances to upstream from https://github.com/cdepillabout/emailaddress (as pointed out by @bitemyapp)
  • NFData (per Add NFData and Binary instances to EmailAddress #25)
  • Binary (per Add NFData and Binary instances to EmailAddress #25)
  • PathPiece (per Consider adding PathPiece instance? #20)
  • Aeson instances (from emailaddress)

[Porges]

8c6e0c7 adds the isemail test-suite.

Another thing to think about: I should be testing the canonicalized versions of email addresses to ensure the normal form is as expected. (The isemail suite does not do this.)

[Porges]

Idea: use DataKinds for the options type that controls what to allow in the email address.

This way we have type EmailAddress = EmailAddress' SaneParseOptions and you can't accidentally mix email addresses parsed with different levels of flexibility.

Also: add a RestrictedEmailAddress for "Emails as they supported by most major providers".

Edit: didn't use DataKinds because that forces it onto the consumer. Instead we have a typeclass ParseOptions which supplies the configuration. This provides the extra safety without any extra burden on consumers.

[MichelBoucey]

@Porges Don't you know my IPv6Addr package? It should give all you need about IPv6 addresses. If not, feel free to contact me.

[Porges]

@MichelBoucey I didn't come across it when I was looking. It looks like it could be useful. I'll stick with ip for now until I get everything working and then have a better look.

One thing that might be required is: in order to pass the test-suite, in some situations I have to be able to parse IPv4-only addresses and not accept IPv6 ones. I'm not sure if this is possible with your current API (except if I parse and then check the result to see if it is IPv4-compatible).