Minimum `DataSize` value when validating CARv2 header

[masih]

Question: What should the minimum accepted DataSize value be in a CARv2 header?

• Could it be zero?
• or should it be the minimum valid size for a CARv1 which is header plus at least one root CID that comes to 26 bytes.

See discussion:

• Add API to regenerate index from CARv1 or CARv2 #309 (comment)_

[rvagg]

or should it be the minimum valid size for a CARv1 which is header plus at least one root CID that comes to 26 bytes.

Oh yeah, it needs an inner header..

Doesn't need a CID though. I think 0xa265726f6f7473806776657273696f6e01 0x11a265726f6f7473806776657273696f6e01 might be the shortest possible CARv1, it's just {version:1,roots:[]} with a varint length prefix. So that would be 17 18 bytes.

But we need to test what even works with both v0 and v2 CARv1 readers, maybe they don't even like that.

[masih]

If no roots are needed do you recall the rationale for this by any chance?

[rvagg]

Why yes I do! (btw I thought we'd removed that for the v2 reader, but sadly not

go-car/v2/internal/carv1/car.go

Lines 132 to 134 in 7ba9372

	if len(ch.Roots) == 0 {
	return nil, fmt.Errorf("empty car, no roots")
	}

, so I guess we need more than 18 bytes, but just enough for a valid CID). See "Unresolved issues" in the spec: https://ipld.io/specs/transport/car/carv1/#number-of-roots.

go-car was originally designed for Filecoin deals, a single root, which we widened a little for general use but never removed the >0 restriction because we didn't want to upset the Lotus apple-cart (or is that a Lotus cart?). js-car allows for zero roots, which can be useful for "this is just a bundle of blocks" and IIRC web3.storage and some of their backend stuff are using CARs in this way.

[masih]

Interesting thank you