A few months ago I went to investigate why some automated emails from our platform eventually ended into the spam folder for a specific user.

While investigating we've finally found the peculiar case of reproduction. Were marked as spam emails meeting all these criterias:

  • They are sent to recipients hosted on Yahoo-owned services (e.g. Yahoo Mail)
  • They contain both a @ and a non-ASCII character in the display name of the sender

Using Foo @® Bar <something@example.com> as the sender, our library sent an email with the following From header:

From: =?utf-8?Q?Foo=20@=C2=AE=20Bar?= <something@example.com>

Wait, what's this format? We'll see in a moment because right now doing more tests makes me notice that the issue can also be triggered by using this display name as the recipient. Let's see how some well-known providers handle this.

The next test is to send an email from Gmail and FastMail with a crafted display name and see if it lands in the inbox of our Yahoo Mail user. For that, I'll use the following text: Kévin @ Blah — (last char being U+2014 EM DASH).

In both cases the email was correctly delivered to the inbox. Let's check what was put in the To header.

First, what did FastMail send?

To: =?UTF-8?Q?K=C3=A9vin_=40_Blah_=E2=80=94?= <kdecherf@ymail.com>

And what did Gmail send?

To: =?UTF-8?B?S8OpdmluIEAgQmxhaCDigJQ=?= <something@ymail.com>

Wait, what's this? Have a seat and let me introduce you to RFC 13421 aka "Representation of Non-ASCII Text in Internet Message Headers".

Here is a short summary of this standard:

RFC 1341 describes a mechanism for denoting textual body parts which are coded in various character sets, as well as methods for encoding such body parts as sequences of printable ASCII characters.

This memo describes similar techniques to allow the encoding of non-ASCII text in various portions of a RFC 8222 message header, in a manner which is unlikely to confuse existing message handling software

It defines two different encodings: "Q" for 'Quoted-printable' encoding and "B" for Base 64 encoding.

Now, we know that Gmail is using "B" encoded-words, thus we'll ignore it for the rest of our investigation.

Let's focus on the difference between what FastMail actually sent and what we sent: the @. FastMail encoded it in the display name as part of the whole "Q" encoded-phrase using =40 whereas we've let the @ untouched.

Interesting.

What happens if we remove the non-ASCII character from the display name? Sending an email through our library results in the following encoding, without deliverability issue:

From: "Foo @ Bar" <address...>

If we summarize, our library handles From/To header depending on the presence of non-ASCII characters:

  • If present, "Q" encode the phrase
  • Else, quote the phrase

Can a plain @ in a q encoded-word cause issue? What says the RFC?

Section "Use of encoded-words in message headers"1 states:

["Q" encoded-words can be used] As a replacement for a "word" entity within a "phrase", for example, one that precedes an address in a From, To, or Cc header. The EBNF definition for phrase from RFC 822 thus becomes:

phrase = 1*(encoded-word / word)

In this case the set of characters that may be used in a "Q"-encoded encoded-word is restricted to: <upper and lower case ASCII letters, decimal digits, "!", "*", "+", "-", "/", "=", and "_" (underscore, ASCII 95.)>.

That being said, @ is not considered as an accepted character in our case and must be encoded as part of the q encoded-word.

So it looks like there was a bug.

And here we are, it is the end of yet another day reading a RFC; and I must admit that I still enjoy learning weird things from thirty years old standards3.


  1. "Representation of Non-ASCII Text in Internet Message Headers" https://datatracker.ietf.org/doc/html/rfc1342 ↩︎ ↩︎

  2. "Standard for the format of ARPA Internet Text Message" https://datatracker.ietf.org/doc/html/rfc822 ↩︎

  3. Shit, I'm not even younger than this one ↩︎