Re: Behaviour of flow-fill before signature

[Simon Josefsson <jas@extundo.com>]

Reiner Steib <reinersteib+gmane@imap.cc> writes:

> On Wed, Oct 26 2005, Simon Josefsson wrote:
>
>> However, this should be fixed in CVS, both for Gnus 5.10 and No Gnus, so an
>> upgrade should solve the problem.
>
> Beside the point-at-[be]ol stuff (required for compatibility with old
> Emacsen), there is this diff:
>
> --- v5-10/lisp/flow-fill.el     2005-08-26 15:41:39.000000000 +0200
> +++ No/lisp/flow-fill.el        2005-10-26 11:27:06.000000000 +0200
> @@ -113,6 +102,11 @@
>    (save-excursion
>      (set-buffer (or (current-buffer) buffer))
>      (goto-char (point-min))
> +    ;; Remove space stuffing.
> +    (while (re-search-forward "^ " nil t)
> +      (delete-char -1)
> +      (forward-line 1))
> +    (goto-char (point-min))
>      (while (re-search-forward " $" nil t)
>        (when (save-excursion
>               (beginning-of-line)
>
> Should it be added to Gnus 5.10 too?

I think that patch is incorrect -- according to RFC 3767, space
stuffing should be done after checking for quoting.  I don't see a
simple way to fix this though, right now the 5.10 logic is a bit
backwards, in that it searches for flowed lines first and then for
quoting on those lines.  So it may miss space stuffing on fixed lines.

I don't have time to work out the code properly right now, but if
someone would like to read RFC 3767 and construct examples for all
corner cases, that would really help.  It is difficult to modify the
algorithm without accidentally cause regressions.  Making a test
harness from the examples is the next step (there is one example and
very simple test harness in flow-fill.el now).  After that, tinkering
with the algorithm is much simpler.

So I don't care strongly whether 5.10 has this patch or not, as the
code appear slightly broken regardless.