From: John Stanley (stanley@peak.org)
Date: Mon Apr 16 2001 - 01:14:09 CDT
Henry Spencer (henry@spsystems.net):
>It's quite conceivable to have software that is
>optimized for the typical case, but can handle the more general case with
>some loss in performance. Case-insensitivity, in general, has some cost.
Yes, case insensitivity has costs. But it costs even more to test a header
for the "preferred" case convention and then test it again using the RFC
conformant case insensitivity.
>My concern is not so much that there's an obvious big win there, but that
>it costs us very little to leave the option open,
Coding multiple tests for the same headers has no costs? Two tests is
cheaper than one? Who says that you don't have the option to test for
whatever capitalization mode you want if the standard says that there is
no case sensitivity?
>Pointing to a preferred form would also discourage pointless
>variation in capitalization.
All capitalization of header names is pointless. In fact, extra characters
in header names only serve to slow things down even more. My god, do you
realize how much time could be saved if "Path: " was just "p: "? If speed
of strcmp is the issue, let's make all headers just one character. Lower
case or upper, you pick. Imagine how much space that would save on
servers. You'd have to pick different letters if you have current headers
that start with the same one. I suggest "z" for "NNTP-Posting-Host". A
16-fold increas in processing that header name.
Russ Allbery (rra@stanford.edu):
> Is there any widely deployed news software that does this?
The version of trn I use does case-sensitive testing, and rejects
NEWSGROUPS: as a header. This is particularly annoying, since it generates
X-ORIGINAL-NEWSGROUPS: when it deals with an article with a followup-to.
Erland Sommarskog (sommar@algonet.se):
>Now, that is a good argument for the text inherited from Son-of-1036.
>Headers are case-insensitive, and any agent that does not understand
>that is broken. Still, headers are not only looked at by machines. and
>only confusion if a newsreader starts to spit out headers like
>"neWsgrOUps: " just to be coOl.
Confusion where? In the servers? Hardly. In the users? What is there to be
confused about? We've already been barraged by odd capitalization, such as
NeXT, NeWS, eTrade, etc. I think that those who are 'confused' by seeing a
newsgroups header would be confused anyway.
>There is nothing of the definition of Local Headers, that says that
>this is a local header. Just because it is inserted at a certain site
>does not make it local.
The site has defined it as local.
>However, the draft does not claim to include a list of local headers at
>this point,
Wait a minute. If you argue that the header I used as an example cannot be
local because this draft did not list it as local, how can you then say
that this draft does not claim to list the local headers? And if the only
local header is xref, why bother having a definition of local header at
all?
>So as a relaying agent, site C must pass
>this scoring header on. As a serving agent it may remove them, though;
I wasn't talking about Site C as a relay. I was talking about the part
where this draft says that local headers MUST be removed. Not "MAY" be
removed.
>Charles's argument is apparently that this phone number etc at the end is
>not a signature but vanilla body text.
No, I asked if typing my name was a signature as defined by this draft and
he said it was. Apparently his argument is that there is some
interoperability issue raised by my typing my name, since I MUST have a
dash dash space CRLF in front of it or I am violating this draft.
> The draft only says MAY ...
Which is a violation of the concept that what you don't understand you
should not touch. If it isn't a 'back-reference', then it isn't a back
reference, and according to the syntax, it can be anything at all. Of
course, your agent MAY do whatever you want with your subject headers, but
making a special statement about "non-standard back-references", which
means "any text at all", is silly.
>Maybe it may all seem theoretical to you, but learn some Swedish
>and hang around in the Swedish newsgroups for a taste of the real world.
I'm sure glad to find out that Sweden thinks it has a monopoly on "the
real world". If you have a problem with "Sv: " not being an official
back-reference, take it up with someone who cares. I'm commention only on
the stripping out of unknown text at whim.