Re: stored mail differs from the sent

Re: stored mail differs from the sent

[Thomas Persson]

Jesper Harder <harder@myrealbox.com> writes:

> Thomas Persson <thomas@spacecentre.se> writes:
> 
> > I routinely sign my emails with `mc-sign'. If the email is written
> > in the English alphabet then no problem but if it contains Swedish
> > signs like å, ä and ö then the message that gets sent verifies
> > correctly but the archived message is corrupt.
> 
> It sounds like a bug in Mailcrypt.  I would suggest that you upgrade
> to Gnus 5.10 and use the built-in crypto support.
> 
> -- 
> Jesper Harder                                <http://purl.org/harder/>

I don't think the problem is with mailcrypt. As I sad, the text in the
archived mail is not the same as the text in the sent mail when it
contains Swedish characters. The signature is produced with the text
in the sent mail and therefore does not verify correctly for the
archived mail because the text in the archived mail is in a different
character set or something. The mailcrypt thing is just a symptom.

Re: stored mail differs from the sent

[Thomas Persson]

Jesper Harder <harder@myrealbox.com> writes:

> Thomas Persson <thomas@spacecentre.se> writes:
> 
> > The archived message is corrupt and the mail which I actually send
> > verifies correctly. That tells me the archiving procedure is
> > screwing up the messages.
> 
> The archived message is not corrupt.  It verifies correctly for me in
> Gnus 5.10.
> 
> -- 
> Jesper Harder                                <http://purl.org/harder/>

I tried the same procedure of signing and sending a mail to myself but
this time with an email account where the MTA didn't convert the
content-transfer-encoding from quoted-printable to 8bit and this time
neither the archived nor the received mail verified correctly.

I have done some research and my working theory is that digitally
signed messages cannot be quoted-printable. Is this correct?

In that case the problem is that when I include Swedish signs in a
message, emacs sets the content-transfer-encoding to quoted-printable
which screws up the signature. When I send it through an MTA that
converts the content-transfer-encoding from quoted-printable to 8bit
everything becomes ok again.

What I have to do then is to make sure that emacs sets the
content-transfer-encoding to 8bit whenever I send a digitally signed
message. There is a variable called
`mm-content-transfer-encoding-defaults' but I feel that I do not
enough about this yet.

Re: stored mail differs from the sent

[Jesper Harder]

Thomas Persson <thomas@spacecentre.se> writes:

> Jesper Harder <harder@myrealbox.com> writes:
>
>> The archived message is not corrupt.  It verifies correctly for me in
>> Gnus 5.10.
>
> I tried the same procedure of signing and sending a mail to myself but
> this time with an email account where the MTA didn't convert the
> content-transfer-encoding from quoted-printable to 8bit and this time
> neither the archived nor the received mail verified correctly.

This is probably a bug in Mailcrypt as your previous QP example
verified for me.

> I have done some research and my working theory is that digitally
> signed messages cannot be quoted-printable. Is this correct?

Yes and no.  If your message includes 8bit characters you shouldn't
really be using old-style inline PGP in the first place.  However, if
you do, the message should be QP encoded.

The right thing for messages with non-ASCII characters is to use
PGP/MIME which works reliably in this case.

> What I have to do then is to make sure that emacs sets the
> content-transfer-encoding to 8bit whenever I send a digitally signed
> message. There is a variable called
> `mm-content-transfer-encoding-defaults' but I feel that I do not
> enough about this yet.

That is the wrong approach.  As I suggested previously you should
rather upgrade to Gnus 5.10 (and use PGP/MIME).  If you really *must*
use inline PGP in combination with 8bit, then this works in 5.10 as
well.

-- 
Jesper Harder                                <http://purl.org/harder/>

Re: stored mail differs from the sent

[Thomas Persson]

Jesper Harder <harder@myrealbox.com> writes:

> Thomas Persson <thomas@spacecentre.se> writes:
>
>> Jesper Harder <harder@myrealbox.com> writes:
>>
>>> The archived message is not corrupt.  It verifies correctly for me in
>>> Gnus 5.10.

I installed Gnus v5.10.6 but my old archived messages still don't
verify correctly. Not with mailcrypt and not with
pgg-verify-region. In fact none of my old messages, not even the
really old ones I wrote in pine works anymore after the upgrade. They
(the old pine messages with 8bit encoding) if I export them to a file
and verify them with gpg -d. 

>> I tried the same procedure of signing and sending a mail to myself but
>> this time with an email account where the MTA didn't convert the
>> content-transfer-encoding from quoted-printable to 8bit and this time
>> neither the archived nor the received mail verified correctly.
>
> This is probably a bug in Mailcrypt as your previous QP example
> verified for me.

Yes it does seem to be a bug in mailcrypt. Mailcrypt probably creates
the signature assuming that the non ASCII characters will be sent as
8bit. I see now that the fact that the messages I sent verified
correctly was just a fluke. The MTA translated them into 8bit.

>> I have done some research and my working theory is that digitally
>> signed messages cannot be quoted-printable. Is this correct?
>
> Yes and no.  If your message includes 8bit characters you shouldn't
> really be using old-style inline PGP in the first place.  However, if
> you do, the message should be QP encoded.
>
> The right thing for messages with non-ASCII characters is to use
> PGP/MIME which works reliably in this case.
>
>> What I have to do then is to make sure that emacs sets the
>> content-transfer-encoding to 8bit whenever I send a digitally signed
>> message. There is a variable called
>> `mm-content-transfer-encoding-defaults' but I feel that I do not
>> enough about this yet.
>
> That is the wrong approach.  As I suggested previously you should
> rather upgrade to Gnus 5.10 (and use PGP/MIME).  If you really *must*
> use inline PGP in combination with 8bit, then this works in 5.10 as
> well.

I'm convinced. I'll start using PGP/MIME instead. I was just annoyed
that using inline PGP with non ASCII characters shouldn't work with
Gnus when I did it with pine for over a year.