[Bug: 21.5-b29] gnus can't send in 21.5.29 but in 21.4.22

[Katsumi Yamaoka]

Katsumi Yamaoka wrote:
> Lars Magne Ingebrigtsen wrote:
>> Reiner Steib <reinersteib+gmane at imap.cc> writes:

>>> There you may find a possible solution[1] or a workaround[2] there.  I
>>> don't know why[1] hasn't been implemented.
>>>
>>> ,----[ http://thread.gmane.org/gmane.emacs.gnus.general/68154/focus=68185 ]
>>>| Gnus should bind `coding-system-for-read' then to the value of
>>>| `mm-binary-coding-system' that defaults to `binary'.  However,
>>>| it seems to have been set to nil or `undecided':
>>> `----

>> I don't know why, either.  Katsumi?

> I realized what we have to fix is `mml-compute-boundary-1'.  It
> may take a couple of days (or shorter?).  I'll work on it.

> The function looks for ones that are the same as a MIME boundary
> (e.g., =-=-=) in contents of MIME parts, and updates the boundary
> pattern so as to be unique, if any.  Contents to be checked should
> be encoded ones, however it doesn't so.  As for a file to attach,
> it reads the file without binding `coding-system-for-read', doesn't
> encode it, and looks for things like the boundary in it.  That is
> the root cause of an XEmacs 21.5 error.

Done.  Not yet tested with XEmacs 21.5 though.
The new function definition encodes every part in the same way
practicing when actually sending them in advance and then looks
for things like a MIME boundary.  So it may slow message sending.
In addition, it assumes that signing and encrypting will never
generate a boundary pattern in data.