detect K-9 signatures

[sblondon]

This PR detects K-9 signatures to separate them from the main message.

The regular expression is generic in order to fit several translations of the signature.

The original signature text is available at https://github.com/k9mail/k-9/blob/581937a7e33f8c1442d0a6225485fb28f8dab1db/app/ui/src/main/res/values/strings.xml#L64

[mailgun-ci]

Can one of the admins verify this patch?

[obukhov-sergey]

Thanks for this — default K-9 footers are indeed missed on current master. I ran extract_signature on the English fixture and got the full body back with signature=None.

The -- delimiter does not save us here: TOO_LONG_SIGNATURE_LINE is 60, and

Sent from my Android device with K-9 Mail. Please excuse my brevity.

is 68 characters, so the candidate is marked cl and dropped. The existing ^sent from my… phone pattern also misses it because K-9 has a hyphen and the line has periods, which are outside [\s,!\w]. A dedicated phone-signature rule is reasonable. The one in this PR is not safe to merge.

Blocker: the new pattern is too loose

^[\S ]*?Android[\S ]*?K-9[\S ]*?\.[\S ]*?\.

RE_PHONE_SIGNATURE is search() on the entire body, then .* with re.S consumes everything after the first hit. The other branches at least start with Sent from / Enviado desde. This one matches any line that mentions Android, then K-9, then two periods.

These all match as a “signature” (remainder of the message included):

• I use Android and K-9 Mail. Please try it. followed by meeting notes
• The Android version of K-9 is 5.603.
• On Android, I prefer K-9. It is great.

That will strip real reply text in production. Please also add $ like the other alternatives in this group.

Same pattern is also too strict

It requires two periods. One-sentence footers such as Sent from my Android device with K-9 Mail. (and many translations) will miss. Builds that say “Thunderbird” instead of “K-9” will miss too.

Tests

EN + FR fixtures are a good start, but they only cover the happy path. Please add:

1. A false-positive case: Android + K-9 in the body, then real content — body must be kept.
2. A K-9 footer without -- . That is the case that actually needs this regex; with -- the current tests also rely on RE_SIGNATURE.
3. Direct expected-signature strings (same style as test_blackberry_signature) instead of msg_body[len('Heeyyoooo.\n'):].

Suggested fix

Prefer a narrow pattern in the existing style, e.g.:

^sent[ ]from[ ]my[ ]Android[\S ]*K-9[\S ]*$
|
^Envoyé[ ]de[ ]mon[ ]appareil[ ]Android[\S ]*K-9[\S ]*$

Anchor with $. Add locales explicitly rather than “Android + K-9 + two dots”.

A more general fix, if you want to go that way: treat a line that is exactly -- / -- as an RFC 3676 delimiter and take everything after it even when the next line is longer than 60. That would catch default K-9 and any other client that uses a standard delimiter plus a long disclaimer.

Please tighten the regex and add the negative test before we merge.