Hallo!

I noticed this one on a system with my work emails (which is what I
recently reported on the IRC channel), but can reproduce it on another
system, too.  :-/

The first system is using Ubuntu's emacs23 23.1+1-4ubuntu7.1+maverick1,
the other Debian's emacs23 23.2+1-7.

In the Emacs UI, do a ``M-x notmuch-seach RET tag:notmuch RET'', wait for
it to finish and save the buffer's text to a file, called emacs.

(In my DB, the notmuch mailing list is tagged as `notmuch' -- but it
doesn't matter, just use a search term that matches some hundreds of
messages; ``*'' if you want.)

    $ notmuch search tag:notmuch > stdout
    $ diff -w -U1 \
        <(sed < stdout -e 's%^thread:[0-9a-f]*\ %%' -e 's%|%,%' -e 's%;%%' ) \
        emacs \
        | cut -c -30
    --- /dev/fd/63	2011-01-30 21:4
    +++ emacs	2011-01-30 21:30:39.
    @@ -76,3 +76,2 @@
      November 21 [1/1] Dmitry Kur
    - November 21 [9/9] Tassilo Ho
      November 20 [5/5] Matthieu L
    @@ -100,3 +99,2 @@
      November 12 [9/9] Jameson Ro
    - November 12 [3/3] Jameson Ro
      November 12 [7/7] David Edmo
    @@ -125,3 +123,2 @@
      November 03 [13/13] david@te
    - November 02 [2/2] Gregor Kap
      November 01 [10/10] Felipe C
    @@ -151,3 +148,2 @@
       October 13 [11/11] Michal S
    -  October 13 [14/14] Felipe C
       October 11 [1/1] Kristoffer
    @@ -177,3 +173,2 @@
       2010-07-06 [2/2] dbp@riseup
    -  2010-07-05 [9/9] Dmitry Kur
       2010-06-27 [1/1] servilio@g
    @@ -203,3 +198,2 @@
       2010-06-04 [2/2] Tomas Carn
    -  2010-06-04 [2/2] dme@dme.or
       2010-06-04 [4/4] Sebastian 
    [...]

That is, roughly every 25th line is dropped in the Emacs notmuch-search
buffer!  (Via setting notmuch-command to a shell script, I intercepted
the notmuch / emacs pipeline with tee, and found that what comes out of
``notmuch | tee'' is still sane, so I'm fairly sure it's an Emacs /
notmuch elisp code issue.)

And, 25 times the medium length of a ``notmuch search [...]'' output line
is... 4 KiB, the standard page size.  Is this ``just'' a problem in the
notmuch elisp code, or is Emacs doing something awful with (wild
speculation...) short reads on buffer page boundaries (or whatever else)?


Grüße,
 Thomas