On 12/14/22 1:39 PM, Corentin wrote:
Tom,

That analysis has been performed, many times over. Literally tens if not hundreds of hours at this point.
I believe that. But it isn't reflected well in the paper.
I do not know how to convey that in other ways than what I already tried.
Replace the screenshots with an analysis like I provided. Not for all the terminals of course, but for a select subset of them.

All the terminal tested match the behavior of what is proposed in the paper modulo tofu and other rendering bugs

I looked at code of all these terminal too, it's also in the paper. The codepoints in the screenshot are the codepoints that sg-16 specifically asked about at the last meeting.

It is virtually impossible to correlate the characters displayed in the screenshots with anything:

Tom.


I am glad to see that your observations concur.


On Wed, Dec 14, 2022, 19:30 Tom Honermann <tom@honermann.net> wrote:
On 11/30/22 5:26 PM, Corentin via SG16 wrote:

Just a note: the gist has 8570 characters and that count matches the ranges specified in the D2675R1 annex.


Simply cat that file in the terminal.
The screenshot below is a render on ITerm2
You will notice the tofu for reserved codepoints is considered narrow
but doesn't quite fit so it overlaps with the next cell, same for the number in square.


Screenshot 2022-11-30 at 23.19.47.png

As discussed previously, a single screen shot that only shows a small subset of the relevant characters is not sufficient to demonstrate that the conclusions of the paper are consistent with existing behavior. I continue to have reservations about the screen shots in the paper for this reason; I don't see how they provide useful information at all. I think they are actively misleading since they do not appear to show behavior that is consistent with the intent of the paper.

I spent some time analyzing the behavior of all 8570 characters in the terminal I use (Konsole 12.12.3 with the Hack 10pt font). Here is what I found:

  • For the characters that the paper changes from width 1 to width 2 (based on the listings in the annex), the following are displayed with a width other than 2:
    • Width 0:
      • U+016FE4 (KHITAN SMALL SCRIPT FILLER)
    • Width 1: (These were all displayed as tofu; some are probably unassigned characters, others are probably unknown by the font)
      • U+01AFF0 .. U+01AFFE
      • U+01B11F ..  
      • U+01F6DC .. U+01F6DF
      • U+01F7F0
      • U+01FA75 .. U+01FA77
      • U+01FA7B .. U+01FA7C
      • U+01FA87 .. U+01FA88
      • U+01FAA9 .. U+01FAAF
      • U+01FAB7 .. U+01FABF
      • U+01FAC3 .. U+01FACF
      • U+01FAD7 .. U+01FAF8
  • For the characters that the paper changes from width 1 to width 2 (based on the listings in the annex), the following are displayed with a width other than 1:
    • Width 2:
      • U+003248 .. U+00324F (CIRCLED NUMBER TEN ON BLACK SQUARE .. CIRCLED NUMBER EIGHTY ON BLACK SQUARE)

These results strongly match the intent of the paper and that the open question regarding the last group of characters should be answered such that they do not change width.

This is the kind of analysis I would like to see performed for other terminals so that we can qualitatively compare behavior between them. I attached C++ source code I used to display the characters.

Tom.