- From: Julian Reschke <julian.reschke@gmx.de>
- Date: Sat, 08 Jan 2011 12:17:53 +0100
- To: Mykyta Yevstifeyev <evnikita2@gmail.com>
- CC: IETF Discussion <ietf@ietf.org>, httpbis Group <ietf-http-wg@w3.org>, ietf-message-headers@ietf.org, iesg@ietf.org
On 08.01.2011 11:19, Mykyta Yevstifeyev wrote: >> If a draft changes three times during LC, there may be a problem. I >> encourage you to go back to the drawing board, and think hard(er about >> the feedback you got, in particular >> <http://lists.w3.org/Archives/Public/ietf-http-wg/2010OctDec/0645.html>. > Why do you think that the draft shouldn't be changed during LC? Because people assume stability during LC, and are unlikely to review the spec multiple times during the same LC. Of course that does not mean that you can't start addressing issues during LC. Just do so in a way that it doesn't change what's being last-called. For instance, by publishing the current edits on a private web page. If you have non-editorial changes, the AD will have to decide whether they require a new LC. > Moreover, I had some reasons for doing that. firstly, version -09 added > the section that was not present in -08 and if I added it now, there > would not be a possibility to discuss it. Version -10 presented the most There was no stability *because* of these updates. > stable one to reflect all the comments I received during the LC. And -11 > was prepared as final for IESG review. My recommendation to the IESG is to wait for a stable spec, and then restart the LC. > I have fully answered all the question form the message you refer to. >> >> I also mentioned at least once that in many frameworks this is >> essentially un-implementable, as different types of header fields are >> processed by different, independent layers in the code, and thus >> there's no way some part of the code will ever *know* which headers >> have been "processed". Do you think that this is not a problem? > That is the issues that would be decided by the implementators. > Moreover, one code layer may support this technology while others may not. How so? If a servlet does support it, but the servlet container does not, there's no way to generate a correct header field value. >>> ... >>> * Syntax: Changed. Now is not /token**/but /1*VCAHR /for definition >>> of the name of the header. >>> ... >> >> Why? > RFC2616 gives the following definition of token: > > token = 1*<any CHAR except CTLs or separators> > separators = "(" | ")" | "<" |">" | "@" > | "," | ";" | ":" | "\" |<"> > | "/" | "[" | "]" | "?" | "=" > | "{" | "}" | SP | HT > > > > And non-visible chars are not allowed in headers. So only VCHARs. That didn't answer my question. Why did you change something that was ok? It also shows why you'll need a new LC. In this particular case the change wasn't only not necessary, but also changes the syntax significantly. VCHAR allows significantly more characters than HTTP's token, namely the comma that you already need as delimiter. Well, assuming you mean the VCHAR definition from RFC 5234, because your spec doesn't clearly say where it comes from (you could import specific productions or all of them from 5234, but you should say that). Speaking of which: you can't have a "_" character in a rule name. Best regards, Julian
Received on Saturday, 8 January 2011 11:18:44 UTC