bug#54646: 29.0.50; set-fontset-font and font clipping issues

[Visuwesh <visuweshm@gmail.com>]

[ஞாயிறு ஏப்ரல் 03, 2022] Visuwesh wrote:
> [Sunday April 03, 2022] Eli Zaretskii wrote:
>
>>> From: Visuwesh <visuweshm@gmail.com>
>>> Cc: rpluim@gmail.com,  54646@debbugs.gnu.org
>>> Date: Sun, 03 Apr 2022 15:56:35 +0530
>>> 
>>> > The %c format is only good for single-byte characters, which these
>>> > ones aren't.  Please use %x to print them (in hex).
>>> >
>>> 
>>> Will do, thanks.
>>> 
>>> > Also, I think printing everything is too much, and doesn't allow to
>>> > focus.  Please print only when the character's code is one of those
>>> > involved in the problematic display.
>>> 
>>> Unfortunately, the characters that are problematic tend to differ from
>>> each run.  Nevertheless, I will hand-pick the problematic characters and
>>> send it.
>>
>> That'd be good.  We need a reproducible case to work with.
>>
>>> >   if (chars[char_idx] == CHAR1
>>> >       || chars[char_idx] == CHAR2
>>> >       || chars[char_idx] == CHAR3)
>>> >     printf (...
>>> >
>>> > where CHAR1, CHAR2, and CHAR3 are the characters involved in the
>>> > problematic display, according to "C-u C-x =".
>>> >
>>> > May I suggest that you show me the code you add to hbfont.c before you
>>> > run it?  This would avoid unnecessary iterations for you.
>>> >
>>> 
>>> Sure,
>>
>> This is okay, but please don't forget to add that 'if' condition.  I
>> think the characters involved in the composition, and in addition one
>> character on each side of those, should be enough.
>
> It seems like I did not get my point across: the characters that tend
> to be rendered problematic differ from each run so I will hand-pick the
> data for the problematic characters in _that_ run and send it.
>
>>
>> Please show the final code, so that I could then interpret the
>> print-outs correctly.
>>
>> Thanks.
>
> Considering the above, it would be
>
> [....]

It took me eons to do this again, I apologise for that.  There's one
thing that I noticed about this issue: when I use this webpage
https://www.dinamalar.com/news_detail.asp?id=3012739 as a test page and
I let _all_ the scaled characters in that page be displayed, I cannot
reproduce the issue but if I let only some of the scaled characters in
that page be displayed and go to a part that was never displayed before,
the characters there have the "bad" shaping.  I'm writing this here
in the hopes that it might help in debugging.

I was not successful in getting the data for all offending sequences.
The offending sequences were,

bb9 bbf -- ஹி
ba9 bc1 -- னு
bb5 bbf	-- வி
b86 -- ஆ

and the GOOD case for bb9 bbf is

    lbearing 1 rbearing 28 width 28 ascent 9 descent 5
    1 bb9 bbf xadvance 1809 yadvance 0 xoffset 0 yoffset 0
    xpos 0 yoff 0 wadjust 28

and the BAD case is

    lbearing 1 rbearing 28 width 28 ascent 9 descent 5
    1 bb9 bbf xadvance 3193 yadvance 0 xoffset 0 yoffset 0
    xpos 0 yoff 0 wadjust 50

If you want more data, then I can try repeating this (I did not retry
since it is really tedious).

The data is for -Goss-Kurinto Seri-regular-normal-normal-*-17-*-*-*-*-0-iso10646-1
and HarfBuzz version is still at 4.2.0.

The patch I used is

diff --git a/src/hbfont.c b/src/hbfont.c
index 2721a66120..9432f75bbf 100644
--- a/src/hbfont.c
+++ b/src/hbfont.c
@@ -490,6 +490,7 @@ hbfont_shape (Lisp_Object lgstring, Lisp_Object direction)
     : NULL;
   if (!hb_font)
     return make_fixnum (0);
+  printf("position_unit begin_hb_font: %f\n", position_unit);
 
   hb_bool_t success = hb_shape_full (hb_font, hb_buffer, NULL, 0, NULL);
   if (font->driver->end_hb_font)
@@ -589,10 +590,17 @@ hbfont_shape (Lisp_Object lgstring, Lisp_Object direction)
       LGLYPH_SET_RBEARING (lglyph, metrics.rbearing);
       LGLYPH_SET_ASCENT (lglyph, metrics.ascent);
       LGLYPH_SET_DESCENT (lglyph, metrics.descent);
+      printf("lbearing %d rbearing %d width %d ascent %d descent %d\n",
+	     metrics.lbearing, metrics.rbearing, metrics.width, metrics.ascent, metrics.descent);
 
       xoff = lround (pos[i].x_offset * position_unit);
       yoff = - lround (pos[i].y_offset * position_unit);
       wadjust = lround (pos[i].x_advance * position_unit);
+      printf("%x %x %x xadvance %d yadvance %d xoffset %d yoffset %d\n",
+	     (char_idx == 0 ? 1 : chars[char_idx-1]), chars[char_idx],
+	     (char_idx == glyph_len-1 ? 1 : chars[char_idx+1]),
+	     pos[i].x_advance, pos[i].y_advance, pos[i].x_offset, pos[i].y_offset);
+      printf("xpos %d yoff %d wadjust %d\n", xoff, yoff, wadjust);
       if (xoff || yoff || wadjust != metrics.width)
 	LGLYPH_SET_ADJUSTMENT (lglyph, CALLN (Fvector,
 					      make_fixnum (xoff),