Hello! Your DM suggested I share some thoughts on the library, but since it would be lengthy to fit into a single Twitter meessage, I put them into this issue. I think it might be too many to open one issue per item, so I've put them all here for you:

• It would be great if subroutines could be defined separately from the regex body.
  I could have defined them with variables outside and directly interpolated them without the use of subroutines, but consider:

  const dateRegex = regex`\d{4}-\d{2}-\d{2}`; // I want to but can't name the variable `date`
  const { groups: { date, time } } = regex`(?<date>${dateRegex})[T ](?<time>${timeRegexFromElsewhere})`.exec(str);

  This causes pollution to the current scope, while

  const { groups: { date, time } } = regex({
      subroutines: {
          date: regex`\d{4}-\d{2}-\d{2}`,
          time: timeRegexFromElsewhere,
      },
  })`(?<date>\g<date>)[T ](?<time>\g<time>)`.exec(str);

  doesn't.

• It might be out of scope, but you want to think about transforming Unicode properties that aren't yet supported in ECMAScript, i.e. those other than General_Category, Script and Script_Extensions. The particularly useful ones are Block (blk), East_Asian_Width (ea), Numeric_Type (nt) and Vertical_Orientation (vo). (I once needed East_Asian_Width in my project, and ended up another dependency specific to it.)

• Consider allowing direct interpolation after \u, \p, \P and \q to get rid of the tricky {${…}} syntax.

• A subtle one, but let's wrap regex.js line 109 in a try block such that the transformed RegExp is visible on error.

Some questions:

• What is the point to use patterns instead of directly interpolating regexs inside a regex, given that flags can be altered and RegExps can be nested with regex but not with pattern?

• Is there any particular reason not to use [Symbol.match] and [Symbol.replace] but a(n optional) wrapper class? Is it because using them would be insufficient to cover all cases?