I had discovered an issue where TOML table header parsing had undesired behaviour where table declarations like [table] and [[array]] were being split across multiple tokens, creating split spans when parsed. The bug was actually, for me, in refractor but refractor uses the same logic to parsing as prism so prism also should get a fix.

I have implemented a change in behaviour to toml header parsing in this PR which keeps the entire table header as a single token, you can see the change in the diff on the test output but basically [header] becomes [header] rather than "[", "header" ,"]" (and also made a simplification by removing insertkey in place of dottedkey

However, having thought about it a bit longer, I do recognise this is perhaps not an upgrade since no longer does this allow separate styling of the punctuation in the header compared to the key.

So what I would propose is to copy the approach used in parsing ini files, where the whole thing becomes a nested structure

using this:

inside: {
	table-name': {
			pattern: RegExp('(^\\[\\[?\\s*)' + dottedKey),
			lookbehind: true,
			alias: 'class-name',
		},
		'punctuation': /\[|\]/,
	},

to produce this output:

	["table", [
		["punctuation", "["],
		["table-name", "table"],
		["punctuation", "]"]
	]],
	["table", [
		["punctuation", "["],
		["punctuation", "["],
		["table-name", "array"],
		["punctuation", "]"],
		["punctuation", "]"]
	]],

I swapped to this approach that mirrors ini as, on reflection, it is clearly the best choice.