Re: Emacs allocates unnecessary memory during scrolling

[Chong Yidong]

address@hidden (Kim F. Storm) writes:

> MIYOSHI Masanori <address@hidden> writes:
>
>> Emacs 22 on Windows (and probably on other platforms) allocates
>> unnecessary memory during scrolling.
>
> Thank you for the bug report.
>
> Chong, can you look at it, as I don't have time to dig into it right now?
>
> Maybe a fix would be to operate with it->glyph_row = NULL as we do
> elsewhere to just do "display measurements", but that's not a trivial
> change to display_string and friends.

I'm working on several other bugs at the moment, but I'll get round to
this (unless you fix it first).

It's hard for me to debug a "allocates unnecessary memory" complaint;
the memory difference (as reported by top) doesn't seem obviously
larger to me, nor does there appear to be a performance impact.

>> The following opearaion reproduces the problem.
>>
>> o emacs -q
>> o Evaluate (setq scroll-conservatively 10) on the scratch buffer.
>> o Type `C-h' to view the HELLO file.
>> o Type `C-n' on to scroll the window.
>> o Emacs allocates unnecessary memory during scrolling.
>>
>> Large files which contains multibyte characters should be better
>> for reproduction.
>>
>>
>> Findings are:
>>
>> This changes seems to have revealed the problem.
>>> 2006-09-06  Kim F. Storm  <address@hidden>
>>> 
>>>     * xdisp.c (pos_visible_p): Remove exact_mode_line_heights_p arg;
>>>     so calculate heights even when pos-visible-in-window-p is called
>>>     with partially = t.  Don't overshoot last_visible_y in move_it_to.
>>>     Return row height and row number in new rowh and vpos args.
>>
>> In the following call sequences, it->w->ncols_scale_factor is
>> incremented a number of times.
>>
>> In xdisp.c:
>> pos_visible_p()
>> -> display_mode_line()
>> -> display_mode_element() ;; case MODE_LINE_DISPLAY:
>> -> display_string()
>> -> PRODUCE_GLYPHS()
>> -> x_produce_glyphs() ;; RIF
>> -> append_glyph() ;; it->glyph_row->used[area]++
>> -> IT_EXPAND_MATRIX_WIDTH() ;; it->w->ncols_scale_factor++
>>
>> In consequence, required_matrix_width() in dispnew.c returns a
>> large number.  And adjust_glyph_matrix() in dispnew.c allocates
>> unnecessary large memory for glyphs.
>>
>>
>> YAMAMOTO Mitsuharu san says:
>>
>> In xdisp.c:
>>> static int
>>> display_mode_line (w, face_id, format)
>> snip
>>> {
>> snip
>>>   init_iterator (&it, w, -1, -1, NULL, face_id);
>>>   prepare_desired_row (it.glyph_row);
>>
>> Before prepare_desired_row(), it.glyph_row->enabled_p happens not
>> to be 0.  This might be caused by the interruption of redisplay
>> during scrolling.
>>
>> Appending the next code before prepare_desired_row() may fix the
>> problem.
>>> it.glyph_row->enabled_p = 0;
>> But he doesn't know the place is appropriate or not.
>>
>> --
>> MIYOSHI Masanori
>
> -- 
> Kim F. Storm <address@hidden> http://www.cua.dk
>
>
>
> _______________________________________________
> Emacs-devel mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/emacs-devel