GNU bug report logs - #77961
31.0.50; Rendering HTML email is very slow since commit #eab14d68b2e72b9a6b8b0cc67c9667c2bfbed4f5

Previous Next

Full log

Message #85 received at 77961 <at> debbugs.gnu.org (full text, mbox):

From: Gerd Möllmann <gerd.moellmann <at> gmail.com>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: yantar92 <at> posteo.net, inigoserna <at> gmx.com, 77961 <at> debbugs.gnu.org
Subject: Re: bug#77961: 31.0.50; Rendering HTML email is very slow since
 commit #eab14d68b2e72b9a6b8b0cc67c9667c2bfbed4f5
Date: Tue, 22 Apr 2025 15:28:04 +0200

Eli Zaretskii <eliz <at> gnu.org> writes:

>> From: Gerd Möllmann <gerd.moellmann <at> gmail.com>
>> Cc: yantar92 <at> posteo.net,  inigoserna <at> gmx.com,  77961 <at> debbugs.gnu.org
>> Date: Tue, 22 Apr 2025 15:04:59 +0200
>> 
>> Eli Zaretskii <eliz <at> gnu.org> writes:
>> 
>> >> +	      if (dim.width > matrix->matrix_w || new_rows)
>> >> +		{
>> >> +		  row->glyphs[LEFT_MARGIN_AREA]
>> >> +		    = xnrealloc (row->glyphs[LEFT_MARGIN_AREA],
>> >> +				 dim.width, sizeof (struct glyph));
>> >> +		  /* We actually need to clear only the 'frame' member, but
>> >> +		     it's easier to clear everything.  */
>> >> +		  memset (row->glyphs[LEFT_MARGIN_AREA], 0,
>> >> +			  dim.width * sizeof (struct glyph));
>> >> +		}
>> >
>> > This means a very large matrix will not be downsized when possible.
>> > Is that a good idea?  Maybe try replacing '>' with '!=' and see if it
>> > also solves the problem?
>> 
>> It's at least what I originally intended when I wrote that function. On
>> the grounds that if it grew to some size once, it woiuld likely again to
>> that size again. Compare the allocation of the row vector where I did
>> the same, for the same reason.
>> 
>> dispnew.c:
>>   416   /* Enlarge MATRIX->rows if necessary.  New rows are cleared.  */
>>   417   if (matrix->rows_allocated < dim.height)
>>   418     {
>>   419       int old_alloc = matrix->rows_allocated;
>>   420       new_rows = dim.height - matrix->rows_allocated;
>>   421       matrix->rows = xpalloc (matrix->rows, &matrix->rows_allocated,
>>   422                               new_rows, INT_MAX, sizeof *matrix->rows);
>>   423       memset (matrix->rows + old_alloc, 0,
>>   424               (matrix->rows_allocated - old_alloc)
>>   425                 * sizeof *matrix->rows);
>>   426     }
>>   427   else
>>   428     new_rows = 0;
>
> So if the user drags the right side of the frame to make it very wide,
> or just toggles it full-screen, we will grow the glyph matrices and
> never free that space, ever, for as long as the Emacs session lives?
> With today's large screens that could be a significant waste of memory
> that is never returned to the system, I think?
>
> If the problem we want to avoid is reallocating when the size doesn't
> change (which was what happened in this case), then just '!=' should
> prevent it, while simultaneously letting us free some memory when the
> window is down-sized.  Whether such a realloc actually frees memory is
> up to the implementation, but why not let it do what it supposedly
> does best?

It has always been in done in something like the if below. Note the
first two conditions. There have never been re-allocations to shrink a
matrix. 

dispnew.c:
  493       /* If MATRIX->pool is null, MATRIX is responsible for managing
  494          its own memory.  It is a window matrix for window-based redisplay.
  495          Allocate glyph memory from the heap.  */
  496       if (dim.width > matrix->matrix_w
  497           || new_rows
  498           || tab_line_changed_p
  499           || header_line_changed_p
  500           || marginal_areas_changed_p)
  501         {

So shrinking is a new idea. One could try that out, but it's not
something for me ATM, sorry.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.