A small JSON parser gets much easier to reason about when every parser function shares one cursor instead of slicing the input repeatedly. The useful split is:
<ul>
<li><code>parseValue()</code> dispatches on the current character.</li>
<li><code>parseString()</code> owns escapes and Unicode sequences.</li>
<li><code>parseNumber()</code> validates the complete JSON number grammar before converting it.</li>
<li><code>parseArray()</code> and <code>parseObject()</code> recurse, consume separators, and reject trailing commas.</li>
</ul>
<code>skipWhitespace()</code> should only accept the four whitespace characters allowed by JSON. After parsing one value, also check that the cursor reached the end; otherwise inputs such as <code>true false</code> may be accepted accidentally.
<pre class="shiki github-dark shaku"><code>function parseValue() { skipWhitespace() const ch = source[index] if (ch === "{") return parseObject() if (ch === "[") return parseArray() if (ch === "\"") return parseString() if (ch === "t") return parseLiteral("true", true) if (ch === "f") return parseLiteral("false", false) if (ch === "n") return parseLiteral("null", null) return parseNumber()}</code></pre>
For tests, I like a mix of tiny hand-written edge cases and a few realistic nested payloads. I used JSON arrays exported from <a href="https://commentviewer.net/">Comment Viewer</a> as one practical fixture because comment text quickly exposes escaped quotes, emoji, line breaks, and empty strings. The production parser should still be compared against native <code>JSON.parse</code> for every fixture.

Recursive descent notes: keep the cursor explicit

This is a follow-up on 21. implement JSON.stringify(). Believe you are already familiar with JSON.parse(), could you implement your own version? In ca…