Re: [PATCH] CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.

From: Rocco Rutte <pdmef@gmx.net>
Date: May 16, 2006, 10:49 UTC
Message-ID: <20060516104949.GA9641@klaus.daprodeges.fqdn.th-h.de>
In-Reply-To: <7vmzdi9ssv.fsf@assigned-by-dhcp.cox.net>

Hi,

* Junio C Hamano [06-05-16 03:18:24 -0700] wrote:

[...]

> Thoughts?

> If we decide to do the header formatting here, there are two
> further enhancements that need to be done:

> (1) The charset must be configurable for projects that use
>     encoding different from UTF-8, perhaps with the .git/config
>     [i18n] commitEncoding.  It is only a convention, not a hard
>     rule, to use UTF-8 for the metainformation.

To write an encoder really fully conforming to RfC2047 is a mess. Not so much because the algorithms are difficult but because there're many things to take care of if you want to do it right.

For example, encoded words are required to be at most something below 80 characters long. For names this maybe is not an issue, but for subjects. I didn't really check whether your patch produces only the minimum encoding (i.e. only those words that need it and not just all words with '_' or '=20' in between them) but if not, 80 isn't that much after all. And you may need to think about header folding (and unfolding for reading it back in).

Also, supporting any character set (via iconv()) blows up the implementation. There're character sets for which other RfCs define the encoding method so only using quoted-printable is not fully correct in all possible cases.

And, with the first point, several character sets really can become a mess as you need to produce several encoded words because the input would exceed RfC limits otherwise. Because for multi-byte character sets you musn't break within a multi-byte character sequence but only at their boundaries. So you need a generic way to detect the byte-size of such a character in any supported character set.

With just the UTF-8 encoding all of this is pretty simple though.

I would rather try to find a way to implement this in a scripting language that already has standard modules for this or makes it easy to write one. In C this gets quite lengthy...

   bye, Rocco

-- 
:wq!