Streaming regexp matching

Isiah Meadows isiahmeadows at
Mon Jul 30 18:39:58 UTC 2018

There's two things I've found that need suspendable matching:

1. Partially matching against a string, which is useful with
interactive form validation and such.
2. Pattern matching and replacement over a character stream, which is
useful for things like matching against files without loading the
entire thing into memory or easier filtering of requests.

Also, it'd be nice if there was a facility to get *all* matches,
including duplicate group matches. This is often useful for simple
parsing, where if such support existed, you could just use a Kleene
star instead of the standard `exec` loops (which admittedly get old).

And finally, we could avoid setting regexp globals here. That would
speed up the matcher quite a bit.

So, here's my proposal:

- `regexp.matcher() -> matcher` - Create a streaming regexp matcher.
- `matcher.consume(codePoint, charSize?) -> result | undefined` -
Consume a Unicode code point or `-1` if no more characters exist, and
return a match result, `undefined` if no match occurred. `charSize` is
the number of bytes represented by `codePoint` (default: 1-2 if `/u`
is set, 1 otherwise), so it can work with other encodings flexibly.
- `matcher.nextPossibleStart -> number` - The next possible start the
matcher could have, for more effective buffering and stream
management. This is implementation-defined, but it *must* be be `-1`
after the matcher completes, and it *must* be within [0, N) otherwise,
where N is the next returned match.
- ` -> string | number | undefined` - Return the group
index/name of the current match, or `undefined` if it's just issuing a
match of the global regexp.
- `result.start -> number` - Return the matched value's start index.
- `result.end -> number` - Return the matched value's end index.
- This does *not* modify any globals or regexp instance members. It
only reads `regexp.lastIndex` on creation. (It doesn't operate on
strings, so it shouldn't return any it doesn't already have.)

Most RegExp methods could similarly be built using this as a base: if
they work on strings, they can iterate their code points.

As for the various concerns:

- Partial matching is just iterating a string's character codes and
seeing if the matcher ever returned non-`undefined`.
- Streaming pattern matching is pretty obvious from just reading the API.
- Getting all matches is just iterating the string and returning an
object with all the groups + strings it matched.


/cc Mathias Bynens, since I know you're involved in this kind of
text-heavy stuff.


Isiah Meadows
contact at

More information about the es-discuss mailing list