net/smtp: SendMail AUTH error with PlainAuth and successful STARTTLS

[imirkin]

See issue #22145 for background. This got merged in go1.11.

It appears that google-apps-for-your-domain SMTP, when connecting on port 25, does not list AUTH even after STARTTLS. The regular gmail smtp does:

$ openssl s_client -starttls smtp -crlf -connect smtp.gmail.com:25
...
EHLO foo.com
250-smtp.gmail.com at your service, [xxx]
250-SIZE 35882577
250-8BITMIME
250-AUTH LOGIN PLAIN XOAUTH2 PLAIN-CLIENTTOKEN OAUTHBEARER XOAUTH
...

but:

$ openssl s_client -starttls smtp -crlf -connect aspmx.l.google.com:25
...
EHLO foo.com
250-mx.google.com at your service, [xxx]
250-SIZE 157286400
250-8BITMIME
250-ENHANCEDSTATUSCODES
250-PIPELINING
250-CHUNKING
250 SMTPUTF8
QUIT
DONE

The latter had been working previous to the change in question, but now, predictably, I'm getting errors about AUTH not being supported. I believe the change originally went in under the "I can't believe that'll happen" banner, but ... believe it :)

[imirkin]

Hmmmm! I did some more debugging after discussing with a colleague (which invariably happens after bug-filing).

Here's what looks like is happening -- AUTH PLAIN is legitimately not supported on aspmx.l.google.com. It was all previously working fine because I was sending emails to a domain handled by that mx (to myself, really, for debugging), which naturally got accepted without any authentication. The AUTH PLAIN command would previously just get ignored by the MX.

So ... I think this can be closed. This change did result in our email-sending not working though. In addition to my scenario, our "real" code runs in an environment with a local relay (and no AUTH), but we had been constructing an empty-user/pass PlainAuth object, due to dealing with the option of those being supplied. Changing that to nil when user/pass are not supplied was good enough to resolve there.

Up to you whether to consider this a bug/regression, or a buggy-code situation on our end.

[ianlancetaylor]

Thanks for the follow-up. Although the behavior did change, it doesn't seem to have changed unreasonably. I'm going to close this.