remove redundant length fields

[marten-seemann]

Once we require hex-encoding for byte-fields (which, as far as I can see from other peoples' qlogs, is already the de-facto standard), we don't need length fields any more. This applies to the scil and dcil in the Header (as well as the proposed token length in #94), the NewTokenFrame, NewConnectionIDFrame (and probably also the QPACK events, but I'm not an expert on those).

[rmarx]

I kind of disagree, as I think an important use case is where you'd not want to log the raw values but only the lengths.

We can argue whether that's the case for the CIDs of course, but for tokens and qpack data (and, obviously, payloads), I think people should have the option to not log the raw values.

This flows from privacy/security concerns, as well as storage efficiency for larger deployments.

I do think we could add text stating something like "if you log the raw values in hex, the length field is redundant, and tools SHOULD be able to derive length from hex"?

[marten-seemann]

Good point. I can see why you want to log tokens (since they create a correlation spanning two separate connections).

I do think we could add text stating something like "if you log the raw values in hex, the length field is redundant, and tools SHOULD be able to derive length from hex"?

I'd make this an even stronger statement: ""if you log the raw values in hex, the length field is redundant and SHOULD (or MUST?) be omitted. Tools MUST be able to derive length from hex".

[rmarx]

Fixed by above commits.

Main draft now explains how truncated raw byte values should be accompanied by a length field and that tools should be able to derive the length field from the raw byte values.

Event definitions draft references this explicitly when discussing the RawInfo approach.