I get emails from one particular gmail user where the body of the email
is base64 encoded (I've only seen base64 encoded emails from this one
person, I'm 99% sure she doesn't know anything about it). Often a few
lines or a whole paragraph will be mis-encoded, coming through as
gibberish, and then returned to normal. I also use gmail and the same
email viewed via the gmail web interface is correctly displayed. In the
raw article (both in gnus and via gmail) it looks like this:

--8<---------------cut here---------------start------------->8---
Content-Type: text/plain; charset=GB2312
Content-Transfer-Encoding: base64

SGV5LAoKx+vOysTjysfU2rnK0uK2us7Sv6rQxMLwo7/Kx7vysrvKx6OssrvSqr30o6y3tNX9ztK/
--8<---------------cut here---------------end--------------->8---

And then a huge block of text thus encoded, actually two blocks. These
emails are in Chinese, hence the GB2312 encoding. They don't always come
through garbled.

The last time I noticed this, it happened to the middle section of a
paragraph:

Cannon$A8x$B2fOB(B?$B;L$(DVy$A"A$Bg[9D$AWJ+#,N$B1Ugk$AZI$B%#$A,R$BIH$A*UbJWJ+6TN$B0M:!Di$A+D$(DZQUK$A,6xN$B0l8H4<$AxK{;XPE8$(D:'(B
$A_K{N$B1U$(DW@$Bi9RZ$A,N$BLi9S$A\P$BO:$AMUbJWJ+#,R$BIH$A*K{L$(D02$Ap<$B$6$A/#,>$B!W$(DWC$A{I$B6"=_B`(B?``$BB>$ABh$BE*!$IT@'2f<L(B
$BE*(B''$BP;A3M-(B``$AN*$BN;$AK5$B<)$AK5(B''$BE*7y5?!#(B

This should look like (copied/pasted from gmail):

Cannon$A8x$B2fOB::IR$A7"$BN;0l<s$AJ+$B!$2fWL]|$AI%$B!$0x$AN*Ub$B<s$AJ+6T$B2fMh$AK5$BB@$ADQ$BN;!$<)2f$A;9$BKW(B
$A8x$BB>2s?.9p$AK_$BB>2fWL$ARE$B48!$2fITG=6U$AIMUb$B<s$AJ+$B!$0x$AN*$BB>FC$A1p$B7c$A6/$B!$?T4IB>>eMh="$AK5(B
$B!HB>$ABh$BE*!$IT@'2f<LE*!IP;A3M-!H$AN*$BN;$AK5$B<)$AK5$B!IE*7y5?!#(B

That will mean something to Chinese speakers. The encoding may or may
not be going wonky several Chinese characters after the appearance of
ascii (Cannon), but it seems pretty obvious that it's the double quotes
(appearing as two single backtics in gnus, a proper unicode left double
quotation mark in gmail) that gets the encoding back on track.

Does this ring any bells for the encoding gurus out there?

Thanks,
Eric