Re: [PATCH 3/7] pack-protocol.txt: Mark all LFs in push-cert as required

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 6, 2015, 18:23 UTC
Message-ID: <xmqq381116xp.fsf@gitster.dls.corp.google.com>
In-Reply-To: <xmqq7fqd17qn.fsf@gitster.dls.corp.google.com>

Junio C Hamano <gitster@pobox.com> writes:

Show 10 quoted lines

> By clarifying that "sender SHOULD terminate with LF, receiver MUST
> NOT require it" is the rule (and fixing the existing implementations
> at places where they violate the "MUST NOT" part, which I think are
> very small number of places), I think we can drop these LF (or LF?
> for that matter) from all of the PKT-LINE() in the construction in
> the pack-protocol.txt, which would be a very good thing to do.
>
> The example in your sentence will become PKT-LINE(foo SP bar) and
> the "there may be an LF at the end" would only be at one place, as a
> part of the definition of PKT-LINE().

I quickly scanned both the sources where we use packet_write() in the code and say PKT-LINE in the doc; aside from the actual packfile transfer that happens on the sideband, which technically _is_ a user of PKT-LINE, we do not send anything that does not end with a text in PKT-LINE. I just wanted to make sure that "there may or may not be an LF at the end; if there is, it is not part of the payload but is part of the framing" does not invite new implementors to break their binary transfer by reading the definition of PKT-LINE too literally to mean "ok, so I stuffed this 998 byte binary gunk to the packet and insert an optional LF before sending the remainder in separate packets".

Thanks.