Re: scroll-conservatively overflow

[Romain Francoise <romain@orebokech.com>]

no-spam@cua.dk (Kim F. Storm) writes:

> I changed some things related to scroll a few days ago which may have
> made this more visible -- I don't know.

About scrolling: I've been bitten by a bug for some weeks now, I
originally thought it was specific to the multi-tty code but I
discovered this morning that it's not.  (A fresh checkout from CVS HEAD
this morning exhibited the same issues.)

I see it in two different situations:

 1. In Gnus, when I scroll a long article buffer (several pages) by
    repeatedly hitting RET to scroll one line at a time, scrolling stops
    on truncated lines.  For example, when I read an article that
    contains one very long line (e.g. a URL), it scrolls okay until the
    long line is the first line in the Article buffer.  I hit RET again
    and the first part of the line scrolls away.  I still see in the
    buffer the continuation part of the long (truncated) line, and if I
    hit RET again, nothing happens, it doesn't scroll more.  I have
    traced the call back in the Lisp code and it shows that the
    `scroll-up' builtin function returns nil as usual but doesn't have
    any effect on the buffer.  I can keep hitting RET to no avail.

    Scrolling works fine until the "End of buffer" message as usual when
    the article does not contain long lines; it also works fine if I
    resize the window so that lines don't get truncated.  It also works
    fine if I scroll past long lines in the article with SPC, then use
    RET to scroll one line at a time.

    This bug is reproducible in all instances of Emacs, at once.

 2. In Dired, I get "End of buffer" errors when moving over truncated
    lines, from the top to the bottom of a buffer (with `n').  For
    example, if my Dired buffer contains, in the middle of a long
    listing:

    -rw-r--r--    1 romain   romain       2047 Feb 26  2003 bpf_dump.c
    lrwxr-xr-x    1 romain   romain         22 Nov 11 12:26 bpf_filter.c -> ./bpf/net/bpf_filter.c
    -rw-r--r--    1 romain   romain       4966 Feb 26  2003 bpf_image.c

    (The middle line is truncated after "-> ./bp", I see:

     bpf_filter.c -> ./bp\
f/net/bpf_filter.c

    which might not be the case for you depending on your window width.)

    I move the cursor down to the `b' in "bpf_dump.c".  I then use `n'
    which moves the point to the `b' in "bpf_filter.c".  If I then use
    `n' to move down to the next line over the symlink line, the bell is
    rung, and I get a "End of buffer" message in the minibuffer, which
    is obviously wrong since the buffer continues..  After the error,
    the point is on the last `c' in "bpf_image.c", instead of being on
    the `b', but I think this is because of some Dired magic which gets
    confused by the error.

    The fun part is that it does not only happen in Dired buffers: in
    this very Message buffer where I'm composing this post, I can move up
    to the top with C-p, but when I move down to the bottom with C-n, I
    get an "End of buffer" error over the pasted Dired line.

    This bug does not appear in a fresh Emacs session just after I
    launch it, it appears only after I've used Emacs a bit.

This only happens in console frames, in X frames I never managed to
reproduce these bugs.  They are _very_ annoying since it makes the
display flash the bell every once in a while even though no error
occurred...

I can provide more information if needed.

-- 
Romain Francoise <romain@orebokech.com> | You know that old saying,
it's a miracle -- http://orebokech.com/ | that you always hurt the ones
                                        | you love? Well it works both
                                        | ways.