bug#12600: 24.2.50; linum-mode: line numbers in fringe do not refresh when resizing frame

[martin rudalics <rudalics@gmx.at>]

 >> (1) Change the window's buffer via `set-window-buffer'.
 >
 > This is taken care of by setting windows_or_buffers_changed.

windows_or_buffers_changed is a global flag.  How does redisplay find
out _which_ window got a new buffer?  Or does redisplay not care?

 >> (2) Change the size of the window (including toggling of scrollbars and
 >>      fringes).
 >
 > Redisplay figures this out already, I think.  Which commands/functions
 > make these changes?

They all end up calling window_resize_apply.

 >> (3) Change the buffer's position in the window (usually via scrolling,
 >>      `set-window-point' and `set-window-start').
 >
 > These likewise set windows_or_buffers_changed, so they shouldn't be a
 > problem.

But usually they affect only one window.  Again, redisplay might not
care.  But I would be surprised that it doesn't handle such a simple
case of optimization.

 >> Now in all of these cases, the respective routines in window.c would set
 >> the window's last_modified_flag to t, marking the window as dirty.
 >
 > I don't think this is necessary.  Can you try without setting this
 > flag, and see if anything breaks?

I said "would".  last_modified_flag doesn't exist.

 >> So why do we currently reset last_modified and last_overlay_modified in
 >> window.c?
 >
 > See above.  Maybe I'm missing something, but
 > windows_or_buffers_changed should take care of all of this.

OK.  I'll comment them out after the feature freeze and we'll see ;-)

 > We are going in circles, so there must be some misunderstanding here.
 > Can you describe your complete change suggestion, wrt these flags?  In
 > particular, what about buffer_modified_flag, and how does it enter
 > this picture?

Replace the three struct members last_modified, last_overlay_modified
and window_end_valid by one struct member called last_modified_flag -
actually calling it window_modified would be better.  Set
window_modified to t wherever we currently reset one of the three
members.  Redisplay, when fully done, would reset window_modified to nil
for every window it completes instead of setting the other members.

I care about BUF_MODIFF/BUF_OVERLAY_MODIFF only in one place: When
calculating `window-end' after a buffer change but before the next
redisplay, `window-end' with UPDATE non-nil has to consult
BUF_MODIFF/BUF_OVERLAY_MODIFF in order to be sure whether it can reuse
window_end_pos.  window_end_pos would be nil (invalid) initially, set to
a valid value when Fwindow_end updates it, and be reset to nil wherever
we change the value of window_modified.

martin