Eugeny/tabby · error · Error

Invalid code point

Error message

Invalid code point

What it means

Thrown by the utf8ToBytes() helper in the buffer polyfill (web/polyfills.buffer.ts), which implements Buffer.prototype.utf8Write. It encodes a JS string to UTF-8 bytes and rejects any code point >= 0x110000 (i.e. outside the valid Unicode range up to U+10FFFF). This is a defensive guard carried over from the feross/buffer polyfill.

Source

Thrown at web/polyfills.buffer.ts:204

                codePoint & 0x3F | 0x80
            )
        } else if (codePoint < 0x10000) {
            if ((units -= 3) < 0) {break}
            bytes.push(
                codePoint >> 0xC | 0xE0,
                codePoint >> 0x6 & 0x3F | 0x80,
                codePoint & 0x3F | 0x80
            )
        } else if (codePoint < 0x110000) {
            if ((units -= 4) < 0) {break}
            bytes.push(
                codePoint >> 0x12 | 0xF0,
                codePoint >> 0xC & 0x3F | 0x80,
                codePoint >> 0x6 & 0x3F | 0x80,
                codePoint & 0x3F | 0x80
            )
        } else {
            throw new Error('Invalid code point')
        }
    }
    return bytes
}

// Create lookup table for `toString('hex')`
// See: https://github.com/feross/buffer/issues/219
const hexSliceLookupTable = (function () {
    const alphabet = '0123456789abcdef'
    const table = new Array(256)
    for (let i = 0; i < 16; ++i) {
        const i16 = i * 16
        for (let j = 0; j < 16; ++j) {
            table[i16 + j] = alphabet[i] + alphabet[j]
        }
    }
    return table
})()

View on GitHub (pinned to 14e2d60b9b)

Solutions

1. Sanitize the input string before writing: strip or replace lone/out-of-range surrogates (e.g. U+FFFD replacement).
2. Ensure you pass a real JS string (not a number/Buffer) to Buffer write methods that route through utf8Write.
3. Update the buffer polyfill (and @types/node 'buffer') to a current version; this guard is standard and stable but depend on a maintained fork.
4. Reproduce with a minimal input and confirm whether the data is genuinely corrupt upstream (logging the string's code units).

Example fix

// before
buf.write(maybeCorruptString, 0, len, 'utf8') // may throw 'Invalid code point'
// after: normalize the string first
const safe = maybeCorruptString.normalize('NFC').replace(/[�-�](?![�-�])|[^�-�][�-�]/g, '\uFFFD')
buf.write(safe, 0, safe.length, 'utf8')

function sanitizeForUtf8 (s: string): string {
    // replace lone surrogates / out-of-range code units with U+FFFD
    return s.replace(/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|[^\uD800-\uDBFF][\uDC00-\uDFFF]/g, '\uFFFD')
}
const safe = sanitizeForUtf8(input)
buf.write(safe, 0, safe.length, 'utf8')

function isValidUtf8String (s: string): boolean {
    // no lone surrogates -> cannot produce a code point >= 0x110000
    return !/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|[^\uD800-\uDBFF][\uDC00-\uDFFF]/.test(s)
}

try {
    buf.write(input, 0, input.length, 'utf8')
} catch (e) {
    if (e.message === 'Invalid code point') {
        buf.write(input.replace(/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|[^\uD800-\uDBFF][\uDC00-\uDFFF]/g, '\uFFFD'), 0, input.length, 'utf8')
    } else throw e
}

When it happens

Trigger: utf8ToBytes reaches the final else branch when, after surrogate-pair combining, codePoint is >= 0x110000. With well-formed JS strings and charCodeAt this is effectively unreachable (valid surrogate pairs max out at 0x10FFFF). It can be hit by feeding malformed input, a corrupted/mutated string, or by calling the internal helper directly with out-of-range numeric values instead of a real string.

Common situations: Corrupt binary data being treated as a string and then written via Buffer(buf).write(str, 0, length, 'utf8'); a polyfill version mismatch after a webpack/buffer upgrade; manually constructed string-like inputs with invalid surrogate sequences in edge-case locales; fuzzed or attacker-controlled text reaching a buffer write path.

AI-assisted analysis of Eugeny/tabby@14e2d60b9b (2026-08-12). Data as JSON: /api/errors/d062d6a3337eccc2. Report an issue: GitHub.