Low-level tokenization mode for Glow

[fabiospampinato]

Low-level tokenization mode for Glow

[tipiirai]

Sounds like a legit idea. I was planning to implement clearer parsing/tokenization and rendering phases because there is a need for more customized highlighting per language.

I'm sorry this answer took so long. My mind has been occupied with the upcoming design system, but I'm planning to make a round of updates to Glow and Nuekit internals before launching it.

Thanks

[nobkd]

Related to "more customized highlighting per language": #197 (comment)

[fabiospampinato]

I just released the "convenient" highlighter/tokenizer on top of Glow that I had in mind: https://twitter.com/fabiospampinato/status/1762965155841773879

Generally, FWIW, I really like this approach, and if more effort could be put into refining the syntax highlighter I think it could actually be pretty decent for a lot of use cases.

Some areas that IMO would be nice if they could be improved:

1. Producing complete tokens, that cover every input character.
2. Not producing unnecessary tokens, like the ones mentioned in the message above.
3. Improving support for some languages nested inside other languages, like JS inside a <script> tag.
4. Maybe special-casing more things, like rendering things that look like unary/binary/ternary operators with the accent color too.
5. Refining keyword-detection to not consider a word to be a keyword if it comes right after a ..
6. Detecting backtick-delimited strings as strings too.
7. Possibly refining syntax highlighting for lots of other little edge cases.

IMO with relatively few tweaks it would be closer to the quality that TextMate can achieve in a lot more cases.

---

Example comparison I got, with Glow on the left and TextMate on the right:

Code I used for the example:

import shiki from 'shiki';

// Some example code

shiki
  .getHighlighter({
    theme: 'nord',
    langs: ['js'],
  })
  .then(highlighter => {
    const code = highlighter.codeToHtml(`console.log('shiki');`, { lang: 'js' })
    document.getElementById('output').innerHTML = code
  });

[tipiirai]

@fabiospampinato there is a public parseRow method that now understands inline comments with the most recent commit. It will return an array of tokens in following format:

[
  { start: 0, end: 1, tag: "i", re: /[^\w \u2022]/g, },
  { start: 11, end: 18, tag: "em", re: /'[^']*'|"[^"]*"/g, is_string: true, }
  ...
]

Where start is the start index and end is the end index in the inputted string.

Hope this helps. This method only understands individual rows so it has no clue about multi-line comments.

[fabiospampinato]

Nice thanks 👍 Are the tokens covering the entire input string? Like what should happen in that example between indexes 1 and 11?

[fabiospampinato]

@tipiirai the new function is not exported from the entrypoint, could you fix this?

[fabiospampinato]

@tipiirai the tokenization seems a bit wrong. With this code:

import {parseRow} from 'nue-glow/src/glow.js';

const code = "import shiki from 'shiki';";
const lang = "js";

const tokens = parseRow ( code, lang );

I get the following tokens:

{start: 0, end: 6, tag: 'strong', re: /\b(null|true|false|undefined|import|from|async|aw…l|until|next|bool|ns|defn|puts|require|each)\b/gi}
{start: 13, end: 17, tag: 'strong', re: /\b(null|true|false|undefined|import|from|async|aw…l|until|next|bool|ns|defn|puts|require|each)\b/gi}
{start: 18, end: 25, tag: 'em', re: /'[^']*'|"[^"]*"/g, is_string: true}
{start: 18, end: 19, tag: 'i', re: /[^\w •]/g}
{start: 24, end: 25, tag: 'i', re: /[^\w •]/g}
{start: 25, end: 26, tag: 'i', re: /[^\w •]/g}

Which are problematic because you can spot right away that there are 3 tokens wrapping around a single character, but our input string ends with shiki';, so there's no reasonable scenario where there would be the 3 length-1 tokens there at the end.

If I explicitly slice those ranges off from the input string I get this array:

['import', 'from', "'shiki'", "'", "'", ';']

So basically there are two tokens about apostrophes for the string that shouldn't exist 😢