The second half of the fragment match, for JSON Pointer, won't match the empty pointer ($ref: "#", $ref: "foo#"). The suggested change follows the JSON Pointer ABNF.

This is still incomplete though. Fragments may contain %xx encoded characters. In a pointer their use may be necessary, but in either a pointer or anchor they are expected to be treated like the characters they represent, even when those characters don't actually need to be %-encoded.

Not too pretty, but I think it allows anything that's valid. It won't reject some invalid anchors or pointers if they're %-encoded - it would be possible to express the same character ranges as allowed combinations of %-encoded digits, but I'm not masochistic enough to do it.

I'm a little uncertain describing a uri fragment with a regex isn't more trouble than it's worth.

The comment says must be non-empty, but this does allow an empty fragment.

The change to + requires a non-empty pointer; disallowing the / from the reference-token character range is better consistent with its spec's ABNF (and iriReferenceToSchemaString above), though it doesn't ultimately change what the regex matches.

And of course %-encoding needs fixing here too.

actually I think we need to allow empty fragments, so it's the comment that is wrong here.

@karenetheridge Agreed, you need to be able to represent the root with a JSON Pointer fragment, which would be an empty fragment.