Re: [PATCH 0/4] emacs: avoid type errors due to nil as content-type

[Tomi Ollila <tomi.ollila@iki.fi>]

On Tue, Jan 12 2021, Jonas Bernoulli wrote:

> Tomi Ollila <tomi.ollila@iki.fi> writes:
>> Note that messages 
>>
>> 20210110140112.25930-11-jonas@bernoul.li and
>> 20210110140112.25930-12-jonas@bernoul.li
>>
>> have base64 -encoded content, with CRLF line endings
>> so those don't apply as is, but CR's from the encoded
>> content must be deleted.
>
> Do you know why git would decide to send these commits that way?

I do know:

In my case:

- git format-patch ...

- in one of the patch files there are utf-8 characters

- git send-email (to mailing list and to tomi.ollila@iki.fi)

- first stop in email is in smtp.iki.fi (which I use to send email)

- the email that returns directly from iki.fi does not have base64
  encoded content *nor* CR's in email body

- the mail that is sent to notmuchmail.org, will contain CRLF line
  endings; either smtp.iki.fi adds those or notmuchmail.org smtpd
  added those

- the mailman in notmuchmail.org notices that there is 8bit content
  in email body, and decides to base64-encode the whole content
  with CRLF line endings.

- git am, the --no-keep-cr option (default) if not effective in 
  "embedded" base64-encoded content. That particular feature is
  just for that purpose that smtp content should have CRLF line
  endings and the files those patches refer to often don't have
  (--keep-cr can be used if CRLF file endings are expected). It
  is a "feature" that in base64-encoded content is handled as
  verbatim by git-am, git-apply and so on.

For reference: the email 

https://www.mail-archive.com/notmuch@notmuchmail.org/msg50113.html

got directly (via-iki-smtpd) and via mailing list (via-notmuch-mailman)
has the following headers (among other headers) in my mail storage
(and as said above, the one got directly did not have CRLF line endings)

  $ grep Content via-iki-smtpd via-notmuch-mailman

  via-iki-smtpd:Content-Type: text/plain; charset=UTF-8
  via-iki-smtpd:Content-Transfer-Encoding: 8bit

  via-notmuch-mailman:Content-Type: text/plain; charset="utf-8"
  via-notmuch-mailman:Content-Transfer-Encoding: base64


> Maybe we should try to keep it from doing that?  I didn't write
> down the prompt but git actually asked something along the line
> of "for these two messages you have to select the base64 encoding,
> does utf-8 work for you?".

Right, the base64 encoding can be used when using git send-email, but how
many knows to do that :) (Tried to look what option is it, but 
man git-send-email gave me manpage of version 1.7.1 >;/).


Other options:

- modify mailman (David Bremner contacted mailman mailing list, they don't
  think that is a bug (well IMO is it not a bug, but...) anyway one could
  hack locally ;/ (or the smtpd frontend if there is any)

- modify git-am (based on Julio Hamano's comment on some mailing list
  something like if there is clear and sound implementation -- was it
  some filter option then something could get in... (or something))

- based on irclogs (2020-08) David has tool to be offered to 
  (debian)-mailscripts to filter CR's out from base64-encoded content.
  -- which I think is better place as this problem isn't notmuch specific
  (although having such a tool in devel/ would be more convenient for me)

>
>      Jonas

Tomi