Supported Modifier Flags

[rbuckton]

In the Oct, 2021 plenary, @michaelficarra asked that we outline and provide motivating examples for each flag we are considering as a supported modifier.

The flags currently under consideration are:

• i — ignore-case
  • Rationale — Toggling ignore-case is especially useful when matching patterns with varying case sensitivity, or when parsing patterns provided via JSON configuration. Especially useful when working with complex Unicode character ranges.
  • Example — Match upper case ascii letter followed by upper or lower case ascii letter or '

    const re = /^[A-Z](?i)[a-z']+$/;
    re.test("O'Neill"); // true
    re.test("o'neill"); // false
    
    // alternatively (defaulting to ignore-case):
    const re2 = /^(?-i:[A-Z])[a-z']+$/i;

  • Example — Match word starting with D followed by word starting with D or d (from .NET documentation, see ¹)

    const re = /\b(D\w+)(?ix)\s(d\w+)\b/g;
    const input = "double dare double Double a Drooling dog The Dreaded Deep";
    re.exec(input); // ["Drooling dog", "Drooling", "dog"]
    re.exec(input); // ["Dreaded Deep", "Dreaded", "Deep"]

• m — multiline
  • Rationale — Flexibility in matching beginning-of-buffer vs. beginning-of-line or end-of-buffer vs. end-of-line in a complex pattern.
  • Example — Match a frontmatter block at the start of a file

    const re = /^---(?m)$((?:^(?!---$).*$)*)^---$/;
    re.test("---a"); // false
    re.test("---\n---"); // true
    re.test("---\na: b\n---"); // true

• s — dot-all (i.e., "single line")
  • Rationale — Control over . matching semantics within a pattern.
  • Example

    const re = /a.c(?s:.)*x.z/;
    re.test("a\ncx\nz"); // flse
    re.test("abcdxyz"); // true
    re.test("aBc\nxYz"); // true

• x — Extended Mode. This flag is proposed by https://github.com/tc39/proposal-regexp-x-mode
  • Rationale — Would allow control over significant whitespace handling in a pattern.
  • Example — Disabling x mode when composing a complex pattern:

    const idPattern = `[a-z]{2} \d{4}`; // space required
    const re = new RegExp(String.raw`
      # match the id
      (?<id>(?-x:${idPattern}))
      
      # match a separator
      :\s
      
      # match the value
      (?<value>\w+)
    `, "x");
    
    re.exec("aa0123: foo")?.groups; // undefined
    re.exec("aa 0123: foo")?.groups; // { id: "aa 0123", value: "foo" }

Flags likely too complex to support:

• u — Unicode. This flag affects how a pattern is parsed, not how it is matched. Supporting it would likely require a cover grammar and additional static semantics.
• v — Extended Unicode. This flag is proposed by https://github.com/tc39/proposal-regexp-set-notation as an extension of the u flag and would have the same difficulties.

Flags that will never be supported:

• g — Global. This flag affects the index at which matching starts and not the matching behavior itself. Changing it mid pattern would have no effect.
• y — Sticky. This flag affects the index at which matching starts and not the matching behavior itself. Changing it mid pattern would have no effect.
• d — Indices. This flag affects the match result. Changing it mid pattern would have no effect.

Footnotes

1. https://docs.microsoft.com/en-us/dotnet/standard/base-types/miscellaneous-constructs-in-regular-expressions#inline-options ↩

[ljharb]

For the examples, can you share how you'd do it without the relevant proposal?

[rbuckton]

i

Simple cases like /[A-Z][A-Za-z]/ are trivial:

// match an uppercase ASCII letter followed by a mixed-case ASCII letter

// with 'i' modifier:
/[A-Z](?i)[A-Z]/

// without 'i' modifier:
/[A-Z][A-Za-z]/

However, more complex cases are far from trivial:

// match a mixed case "hello" followed by the exact characters "World"

// with 'i' modifier:
/(?i:hello) World/

// without 'i' modifier:
/[Hh][Ee][Ll][Ll][Oo] World/

m

If you are in u mode, you could emulate non-m mode when in m mode using the proposed \A and \z buffer boundaries. However, if you are not in u mode, there's no way to match the buffer boundaries when in m mode:

// with 'm' modifier:
/^---(?m)$((?:^(?!---$).*$)*)^---$/

// without the 'm' modifier, in 'u' mode:
/\A---$((?:^(?!---$).*$)*)^---$/mu

// without the 'm' modifier, not in 'u' mode: not possible to invert when in 'm' mode

s

Its fairly complicated to invert the s flag in a RegExp without modifiers, and easy to get wrong:

// match /a.b/ outside of 's' mode, then /.+/ in 's' mode, then /c.d/ outside of 's' mode
// with 's' modifier
/a.b(?s:.)+c.d/

// without 's' modifier
/a.b(?:.|[\r\n\u2028\u2029])+c.d/

// match /a.b/ inside of 's' mode, then /.+/ outside of 's' mode, then /c.d/ inside of 's' mode
// with 's' modifier
/a.b(?-s:.+)c.d/s

// without 's' modifier
/a.b(?:(?![\r\n\u2028\u2029]).)+c.d/s

[ljharb]

There's nothing with [^\s\S] for the dotAll case?

[rbuckton]

I'm not sure I understand what you mean. Can you clarify?

[rbuckton]

If you mean using [\s\S] to match everything, that's feasible for the first s example, sure. I don't see how it helps with the second example though.

[RunDevelopment]

I just want to share a little trick to emulate m and non-m mode without using ^ and $. This might be relevant for transpilers.

- /^ $/ == /(?<![\s\S]) (?![\s\S])/
- /^ $/m == /(?<!.) (?!.)/ // no `s` flag!

This works for both u and non-u mode.

[rbuckton]

The modifiers supported by this proposal will be limited to i, m, and s. These may be potentially changed by future proposals (such as the x-mode proposal), but doing so is out of scope.

[slevithan]

@rbuckton, I know this is already closed (and implemented in V8, yay!), but for interest's sake, note that it's very possible to emulate presence or lack of s and m. I do it in regex-make to locally apply the presence or absence of local flags for RegExp instances interpolated into a template.

m

If you are in u mode, you could emulate non-m mode when in m mode using the proposed \A and \z buffer boundaries. However, if you are not in u mode, there's no way to match the buffer boundaries when in m mode

Emulating is possible without u mode or buffer boundaries.

• Emulate an m mode ^: (?<=^|[\n\r\u2028\u2029])
• Emulate a non-m mode ^: (?<![^])
• Emulate an m mode $: (?=$|[\n\r\u2028\u2029])
• Emulate a non-m mode $: (?![^])

s

Its fairly complicated to invert the s flag in a RegExp without modifiers, and easy to get wrong:

[...]
// without 's' modifier
/a.b(?:.|[\r\n\u2028\u2029])+c.d/

[...]
// without 's' modifier
/a.b(?:(?![\r\n\u2028\u2029]).)+c.d/s

It's easier than that:

• Emulate an s mode .: [^]
• Emulate a non-s mode .: [[^]--[\n\r\u2028\u2029]] (with v) or (?:(?![\n\r\u2028\u2029]).) for a less efficient version without v (same as you showed in the quote).

Note that, like your (?:.|[\r\n\u2028\u2029]) example, [^] either matches full code points or doesn't based on the presence of flag u/v.

[RunDevelopment]

• Emulate a non-s mode .: [[^]--[\n\r\u2028\u2029]] (with v) or (?:(?![\n\r\u2028\u2029]).) for a less efficient version without v (same as you showed in the quote).

Or [^\n\r\u2028\u2029].