Wrong letter space after patching Korean font

[dhnam]

🎯 Subject of the issue

Patched D2Coding font with patcher in Windows 10.
When patched, letter space of Korean charactors are getting weird.
Screenshot is attached in 'optional'.

🔧 Your Setup

• Using D2Coding with patcher.
• Now using window CMD, but it is not about patched letters.
• Now using Windows 10, and Nerd Fonts ver. 2.1.0.

★ Optional

This is what originally looks like;

This is how it looks like after patch.
And if I patch it in single-width, it looks like:

Not sure it is same with other fullwidth characters(like Chinese or Japanese).

[dhnam]

Looks like it is own problem of FontForge.
I opened ttf file of D2Coding and directly made font out of it, and it gave me same problem.

[Finii]

Maybe this can be solved by patching with --variable-width-glyphs.
But the patched font then does probably not work in 'all' terminals anymore.

Closing this now. If you still have problems, reopen. I would need a link to the problematic font for testing in that case.

[kimwonj77]

I have same issue. With same font, both with/without --variable-width-glyphs, both mono and non-mono.
Original font is on https://github.com/naver/d2codingfont/releases/tag/VER1.3.2
Patched font is without --variable-width-glyphs and with --variable-width-glyphs

Build script is here: https://github.com/kimwonj77/d2coding-nerdfonts/blob/master/build.sh

Also, I'm not a original author of this issue. So I ping you, @Finii. Sorry about this.

[Finii]

@kimwonj77 Can you show an image were I can see the problem?
If it contains Korean text it would be good to have the text also as text in the comment so that I can try it out.

[dhnam]

Original author here.

As far as I remember (It has been three years since I found this issue) it was problem of D2Coding font.

As said I exported D2Coding directly from FontForge, but not only that, but I also tried doing same thing with FontLab and python library (as far as remember, as said), and same thing happened.

(I also uploaded same issue at naver/d2codingfont#84 too, which shows attempt to open and re-export the font cause total breakdown of what is written in Korean.)

Never tested with other Korean-based font, but guess it won't break with other fonts. (Guess it's worth try.)

Guess it's better ping @kimwonj77 . Sorry if this bothered you.

[kimwonj77]

First, It's same as original author's image. 다람쥐 헌 쳇바퀴에 타고파.

mono on windows 11 settings.

double width on windows 11 settings. but...

it's still broken on notepad.

[Finii]

Thank you. I believe I see the problem now. Trying to fix...

[Finii]

다람쥐 헌 쳇바퀴에 타고파 = b2e4 b78c c950 20 d5cc 20 ccc7 bc14 d034 c5d0 20 d0c0 ace0 d30c

Choose for research

• D2Coding/D2Coding-Ver1.3.2-20180524.ttf from https://github.com/naver/d2codingfont/releases/tag/VER1.3.2
• https://github.com/kimwonj77/d2coding-nerdfonts/blob/173307ac3ddbbd20dfb6dcca28466d88bd660167/D2Coding/D2Coding%20Nerd%20Font%20Complete.ttf

Looks ok, width is maintained.

Hmm, testing in tilix

Looks ok. Maybe it is a Windows problem. (Comment will be continued on Windows box)

[Finii]

Looks ok in windows (so far)

Windows Terminal

Looks good.

Ah, you use notepad, lets see

Ah!

Looks good in writer...

So this is a wordpad notepad problem?

Edit: Mixed note- and wordpad 🙄

[Finii]

Original font looks good in notepad

[kimwonj77]

It's bit difficult to explain, But seem like the rendering method of notepad, and some other programs are using old-school character render system. and it causes the issue. (And also, it only happens on korean character. not english.)

It's also same on fontview.exe with korean locale (fontview.exe' is locked by windows locale.). Like og author's image.

Also Eclipse

[Finii]

I found the problem.

The issue is with the OS/2 xAvgCharWidth data in the font.
(OS/2 is the name of one data table within a open type font (for example .ttf))

• For OS/2 version 0 to 2 the number is the average width of the lowercase latin letters (weighted) [1][2]
• For OS/2 version 3+ it is the average width of all glyphs

Unfortunately there are multiple bugs here in parallel so I do not really know how to proceed here.

• notepad determines the advance width from xAvgCharWidth and uses double that value for Korean glyphs (2 cell width) regardless of OS/2 version 😞
• The D2Coding font says it has an OS/2 version 3 (!) table but fills the number as version 2 would (probably to circumvent the notepad bug 🙄
• fontforge always fills the version 3+ number, even if instructed to create the font with a version 2 OS/2 table 😠

Possible actions

• Fix fontforge's OS/2 table version 0-2 (will not help because font says it uses version 3)
• Add option to font-patcher to create version 2 xAvgCharWidth table entry
• Add special rule in font-patcher to specific font name to create for this font only a version 2 xAvgCharWidth

I do not really like to add options to font-patcher if it can be avoided.

Are there other - possibly Korean - fonts affected or is that only D2Coding?

[1] https://learn.microsoft.com/en-us/typography/opentype/spec/os2#xavgcharwidth
[2] https://developer.apple.com/fonts/TrueType-Reference-Manual/RM06/Chap6OS2.html

---

Fix snippet

diff --git a/font-patcher b/font-patcher
index 01e144e4c..a82e6f8c5 100755
--- a/font-patcher
+++ b/font-patcher
@@ -135,6 +135,7 @@ class TableHEADWriter:
             positions = {'checksumAdjustment': 2+2+4,
                          'flags': 2+2+4+4+4,
                          'lowestRecPPEM': 2+2+4+4+4+2+2+8+8+2+2+2+2+2,
+                         'avgWidth': 2,
                 }
             where = self.tab_offset + positions[where]
         self.f.seek(where)
@@ -421,6 +422,10 @@ class font_patcher:
                 for idx in range(source_font.num_fonts):
                     if not self.args.quiet:
                         print("{}: Tweaking {}/{}".format(projectName, idx + 1, source_font.num_fonts))
+                    source_font.find_table([b'OS/2'], idx)
+                    dest_font.find_table([b'OS/2'], idx)
+                    dest_font.putshort(source_font.getshort('avgWidth'), 'avgWidth')
+                    dest_font.reset_table_checksum()
                     source_font.find_head_table(idx)
                     dest_font.find_head_table(idx)
                     if source_font.flags & 0x08 == 0 and dest_font.flags & 0x08 != 0:

This does not calculate the version 2 number but just copies over the value from the source font. Which would be ok if we just want to fix D2Coding I guess. The code is minimal, just what I used to prove that this is the issue's cause.

Edit: Typo fix