Consider not truncating lengths and instead [EnforceRange]?

[ChALkeR]

E.g. in deriveBits (but other places use blind conversion too) :

Promise<ArrayBuffer> deriveBits(
AlgorithmIdentifier algorithm,
CryptoKey baseKey,
optional unsigned long? length = null
);

webcrypto/spec/Overview.html

Lines 1289 to 1293 in 9d98db7

	Promise<ArrayBuffer> deriveBits(
	AlgorithmIdentifier algorithm,
	CryptoKey baseKey,
	optional unsigned long? length = null
	);

Here, length is unsigned long on the API level, and thus passing any non-nullish arg to it converts that arg to unsigned long.

More specifically, currently:

length = 2**32 + 8 is allowed and returns 1 byte

> await crypto.subtle.deriveBits({ name: 'PBKDF2', hash: 'SHA-256', iterations: 1, salt: Uint8Array.of(1) }, await crypto.subtle.importKey('raw', Uint8Array.of(1), 'PBKDF2', false, ['deriveBits']), 2**32 + 8)
ArrayBuffer(1)

length = -8 is allowed and returns 536870911 bytes

> await crypto.subtle.deriveBits({ name: 'PBKDF2', hash: 'SHA-256', iterations: 1, salt: Uint8Array.of(1) }, await crypto.subtle.importKey('raw', Uint8Array.of(1), 'PBKDF2', false, ['deriveBits']), -8)
ArrayBuffer(536870911)

Also length = NaN is allowed and returns an empty ArrayBuffer.

I would really have just preferred an error in both cases

Validation exists, but after conversion:

• 0 returns 0 bytes.
• 1 throws
• -9 throws
• -8 returns 536870911 bytes
• undefined throws
• null throws
• NaN reuturns 0 bytes
• Infinity returns 0 bytes
• 25769803800 returns 3 bytes

[twiss]

I agree that this would be better. The main question I'd have is whether we can change this without breaking anything. It seems unlikely that anything relies on this but it might be worth investigating.

@annevk and @davidben, have any opinions?

[annevk]

It seems worth trying.

[davidben]

...that's how unsigned long works? Lovely. I wonder how many cases in the web platform do we actually want non-EnforceRange behavior? Perhaps we should flip that default around.

Anyway, agreed this is worth trying.

[twiss]

@ChALkeR Do you want to make a PR, and write some WPT tests for this?

[ChALkeR]

@twiss Not at the moment.

[panva]

• add [EnforceRange] instead of truncating lengths #555
• add [EnforceRange] instead of truncating lengths WICG/webcrypto-modern-algos#65