The text layout code is expecting the quantity of glyphs is equal to the cells allocated for each line

[reli-msft]

The issue is found in CustomTextLayout:

Instead of carefully process the clusters input, CustomTextLayout drops nearly all the information from clusters, and only keeping how many columns a cluster occupy, and later it is expecting that the glyph quantity equals to the quantity of clusters. This may lead to incorrect rendering result for many cases (complex script, ligatures, etc.)

Bug #610, #696 may be caused by this.

[reli-msft]

Running the following script in Node.JS shows a lot of problems in current text layout code:

console.log(`
Combining Diacritics    | [a\u0300\u0304]
Polytonic Greek         | [ᾊ] [Α\u0313\u0300\u0345]
Ideographic Variation   | [東京都葛\u{E0101}飾区] [奈良県葛\u{E0100}城市]
Composite Hangul        | [\u1100\u116A\u11B3]
Combining Dakuten       | [か\u309A]
Emoji vs Text           | [\u2602] [\u2602\uFE0E] [\u2602\uFE0F]
Emoji                   | [\u{1F469}]
Emoji with Skin Tone    | [\u{1F469}\u{1F3FB}]
Emoji Variation and ZWJ | [\u{1F469}\u{1F3FB}\u200D\u{1F4BB}]
`)

Terminal results:

Chrome Dev Tools results:

Edge F12 results:

Word results:

[skyline75489]

Possibly also #2191

[skyline75489]

Let me add Firefox result here, so that people don't forget that there's a browser out there called "Firefox":

[DHowett-MSFT]

@skyline75489 my hero

[skyline75489]

@DHowett-MSFT my man. I'm seeing regression on current master 306e751

[skyline75489]

I've used git bisect and found that ddcc06e is the first bad commit. I've got no idea about what #3474 even is..

[skyline75489]

I've tried this on my laptop and sadly the regression is the same:

Emoji is noticeably broken. Japanese characters are different, too.

[zadjii-msft]

@skyline75489 Thanks for the report! I'm going to move that regression to it's own thread (#3553), considering you've narrowed the regression down to a particular commit. Broadly our text rendering needs help with M chars->N glyphs scenarios like this issue describes, but we probably shouldn't have things getting worse on us.

[reli-msft]

This is heavily related to this issue: #780
There's WriteCharsLegacy and Terminal::_WriteBuffer, and they should be unified into one function that writes char stream to the back buffer with proper itemization.

[skyline75489]

Without #780 we can still improve the current implementation like #3578 . I'm not sure whether that's worth the effort, though. It's more like scratching the surface.

[reli-msft]

@skyline75489
We need to change the way we write buffers to perform correct line wrap and cursor movement.
For WT there are two buffers being used: one for CONHOST, which is the "Ground Truth", and one for WT, and the write subroutine is written differently.

So what's wrong with WriteCharsLegacy? Well...

• It does not process surrogate pairs properly.
• Combining characters are not written into the "previous" cell. They are treated separately and the cursor is moved.

And to provide a proper text layout, cells should...

• Have some metadata to guide the client that whether they should be shaped together with adjacent cells.
• If the real width of glyphs is different from the desired width, whar should we do.

[skyline75489]

@reli-msft Gotta admit, the whole Unicode experience is harder to fix than I initially anticipated. I'm not sure the exact plan Console team have. But I'm positive that we are heading to the right direction.

[reli-msft]

@skyline75489
Everybody are underestimating how complex text could be.

Let me count all the features that a proper rich text box should support:

• Input methods, including...
  • Composing IME
  • Speech
  • Handwriting
  • Emoji picker
• Complex text rendering, including...
  • Bidi
  • Shaping
  • Complex-script-aware editing
  • Complex-script-aware find, replace, etc.
• Accessibility

And what they can choose to support:

• Advanced typography, including...
  • About line breaking...
    • Hyphenation
    • Optimized line break (Knuth-plass, etc.)
  • About microtypography
    • OpenType features (ligatures, etc.)
    • OpenType variations
    • Multiple master fonts
    • Proper font fallback ← It is not a simple lookup-at-each-character process
  • Advanced Middle East features, including...
    • Kashida
  • Advanced Far East features, including...
    • Kinsoku Shori
    • Auto space insertion
    • Kumimoji
    • Warichu
    • Ruby
  • Proper vertical layout, including...
    • Yoko-in-Tate
• Inline objects and paragraph-like objects, including...
  • Images
  • Hyperlinks
  • Math equations ← This is really hard.

[DHowett]

You know, with the completion of #16916 I believe we've finally cracked this nut.