unnecessary base64 encoding in trac (ticket) emails

[xris <xris*siliconmechanics*com>]

unnecessary base64 encoding is a sign of many spam/virus messages, and many people's mail servers block these messages outright (including mine). Is there any specific reason why trac base64-encodes its outbound emails? If not, please don't do so — I don't like having to choose between getting no trac emails, and getting more spam.

[Emmanuel Blot]

As Trac supports the UTF8 'charset', it should be able to send UTF8 encoded messages.

Base64 is the encoding choice for UTF8 (quoted printable would not be effective, btw)

[xris <xris*siliconmechanics*com>]

but it should only send them IF there is utf-8 content, no?

However, I think the main issue is for emails where the sole content of the message is base64-encoded, with no raw-text format given. But honestly, I'm not entirely sure what it takes to trigger my mailserver's "no unnecessary base64" filter.

[Emmanuel Blot]

That would be an optimisation, may be.

Everything in Trac is stored as UTF-8, I believe. If the whole message is made of pure ASCII characters, Base64 encoding might be disabled, but that means the message would have to be parsed, checked that UTF-8 stream only contains characters < 127 , and to disable Base64. Much work for a few gains: Base64 works gracefully in all cases.

Encoding pure ASCII in Base64 is valid, even not optimal.

I'm not sure to understand what you meant by "no raw-text format" ?

[xris <xris*siliconmechanics*com>]

many anti-spam measures flag messages with no plaintext portion — particularly base64-only messages because that technique is too often used by spammers to obscure their messages from things that would parse message content.

Anyway, what's wrong with quoted-printable and 'Content-Type: text/plain; charset="utf-8"' ?

[Emmanuel Blot]

Quoted Printable is a poor encoding system: 3 characters for any non ASCII byte: for any char code ≥ 128, QP generates a triplet "=XX", where XX is the hexadecimal value of the character code. This means up to a 200% increase of the message size.

Conversely, Base64 maps every character (ASCII or not) from a 8bit range to a 6bit range. This means a 25% increase of the message size.

QP is an acceptable encoding for (mostly western) european languages, where most characters are ASCII, with some percentage of special characters in the range [127..256].

With other languages (not based on the roman/latin alphabet), every character code is > 256 in Unicode, which means that every UTF8 character would be translated into many ASCII character (from 6 to 9 ?).

There's nothing wrong with UTF8/Base64. This is not an exotic combination, it is quite regular, I believe.

[xris <xris*siliconmechanics*com>]

You're worried about size on a 2k email? What about for the rest of us who would have a 0% increase with QP because there's no UTF-8 content? Nonetheless, it's your prerogative to choose, but I know that there are more people than myself out there who have had to turn off spam filtering (resulting in a 20-30% increase in spam in my case) in order to receive trac ticket messages, and it would be nice if there was at least an option to use QP instead of base64 (I mean, it's ONE option to pass into the python mail function). Trac is the only legit sender of pure-base64 content that I know of, and QP *is* the standard way (at least in my experience) to handle utf-8 content, even if it may not be the most efficient.

[Emmanuel Blot]

I never said I did not want an option for QP: I really don't mind, I only tried to explain the choice of Base64, from my understanding. It was about technical details, not personal wills.

I get a lot of spam too, hopefully destroyed by Thunderbird or Mail, but from my perspective, I would prefer to fix the issue (the SPAM filter that takes some assumptions) rather than to change a specific product to match the SPAM filter criteria.

I'm not worried about a 2k email. I'm worried about trading Base64 for QP, as from my perspective, it is a non sense for non 'latin' languages. As long as the QP encoding is is an option, I have no problem with it.

[xris <xris*siliconmechanics*com>]

Ok, just sounded from the conversation like you were saying "it's base64, and going to stay that way".. If the resolution of this ticket is "added an option for QP encoding instead of base64" I'm totally cool with that (not a python coder, or I'd have already submitted a patch), and can be patient, too.

(as for handling it in tbird, I run my own mail server, and would rather not be paying for the bandwidth to receive spam if I can block 20% or so of it before it gets to me, and tarpit the sender in the process)

[Emmanuel Blot]

No problem. BTW, I did not change the bug status, it is still reported as 'new'

[Axel.Thimm@…]

utf-8/base64 seems to be harmful for sourceforge.net mail services, too.

While it is sourceforge's mail processing (adding a signature) that is broken, it would be nice to have a way to work around it.

I just activated ticket notification onto such a list and very wierd trailing garbage poped up at the end of the notifying mails.

​http://sourceforge.net/mailarchive/forum.php?thread_id=8115721&forum_id=27233

[Emmanuel Blot]

A new notification implementation, is available here [2357]:

• ​http://projects.edgewall.com/trac/browser/sandbox/notification or
• ​http://svn.edgewall.com/repos/trac/sandbox/notification/

To setup encoding scheme, add a new line in the [notification] section of your trac.ini and select one of the available options:

[notification]
mime_encoding = base64 | qp | none

• none: for plain english only (notifications won't be sent if tickets contain non-ASCII charaters)
• qp: quoted-printable, best for western-european languages
• base64: works with any language (default value)

Feedback welcomed.

[Emmanuel Blot]

[2799] provides an option to select the encoding type

[dharana@…]

Thank you for fixing it, useful for me too.

[petr.hroudny@…]

Just found this - and I believe an important option is missing.

It's perfectly valid to generate 8-bit UTF-8 emails today, when the following MIME headers are used:

Content-Type: text/plain; charset=utf-8 
Content-Transfer-Encoding: 8bit

An example MUAs doing that are Thunderbird, Mutt, …

Thus another option for 8bit notifications is needed and I'd even say it should be the default.

[Remy Blank]

Would you mind opening a new ticket for this enhancement request? We usually close tickets against a given milestone to keep track of the version where a feature has been added.