Whitespace is not removed from field values in HTTP/2 or HTTP/3

[martinthomson]

This claim in Section 5.5 is likely incorrect:

Leading and trailing whitespace in raw field values is removed upon field parsing (e.g., Section 5.1 of [Messaging]).

Does it need qualification (in some versions) or permission (could be removed)?

[reschke]

"could be removed" or "can be removed" seems accurate.

That said, isn't this a potential future interop problem? Shouldn't HTTP/2 and HTTP/3 mandate removal as well?

[mnot]

I think the question here is when it's removed. My understanding is that it is removed for the purposes of matching ABNF, and it's removed when the value is used, but it's not necessarily removed automatically when an implementation hands a value to an application.

[reschke]

But if the WS is not removed before the value is passed to the application, it might not be removed at all.

What do HTTP/2 implementations do here?

[mnot]

Discussed in Feb 21 interim; need to focus on interpretation of field, not e.g., forwarding. Should be a requirement.

[annevk]

It would have nice if it was an error to have leading/trailing field value whitespace in newer HTTP versions. Probably too late now, though perhaps not for H/3?