Minutes from HTTP/1.1 Implementor's dinner

I'm posting these for Rohit.

- Jim

HTTP 1.1 Implementers' Dinner
Rohit Khare, Dec 9, 1997

[The general idea was to walk through the document looking for capitalized 
MUST
and MUST NOT requirements. This discussion helped clarify that we defintely 
need
automated tools to track implementations' support for Draft Standard 
documentation
-- we didn't even touch the MAYs and SHOULDs.

What follows is a semi-transcript... Rohit]

Prescription of a new issue in 4.4 Message Length:
notification to user agents (JG:take to mlist and deal with there)
Something about warning to users seems egregious

MUST notify the user is too strong -- is it required for robustness
in the prot?

LM rules the meta-discussion out of order : are there two interoperable 
implementations?
or not? Doesn't seem to be.

Client guys: Henrik Frystyk Nielsen, Yaron Goland
Server: Rich (IBM), Scott (Agranat),
Proxy: henrik, daniel, yaron, Rich
Client: Sami Sun (URN resolver)
Server: Dave, Henry, Jigsaw, Scott
Dave Kristol, Daniel Veillard are also present

YG: there are currently 265 requirments
YG: will be releasing a spreadsheet of all 265 with what IE4 and IE5 do, by
january.

Section 3.1
could be prefaced by leading zeros. MS doesn't know, but will fix if found.
Rich: do handle it. Henry: handles it

HFN: would be more interesting to test if "1.2" works as a test case

YG: these answers I'm giving tonght are for IE4 as I know it. BUT,
we have libInet, which is eqivalent to libwww
(which means we have common support). See,
IE4 complies with "all IANA charsets", but other apps ontop of INET may
ignore it.

(ie. their 3rd party developers may be able to pass illegit stuff down 
through
the api. For example, it doesn't check that mimetypes passed in have no 
white
space.)

----- Major and minor must be treated as separate integers. (3.1 , 3 
paragraphs in )

[Rich has a server Don't list ]
[Henry has don'ts for must and should, but not mays]

JG: WHAT we need is two separate implementations that have been tested to
interoperate -- NOT necc shipped, much less commercial and supported. JUST
had to be implemented once.

Just a client and server? JG: I'm more comfortable with two each (c,s,p)
That's my personal belief

YG: given that we don't need shipment, we may try testing against IE5.

SL: the most important result of today's dinner is to list out what must be
tested ASAP

YG: now that our testers are up to speed, we're ready to hit anything that
people put up for testing.

We need a standardized template for testing ***

Need a suite we can automate fromthe client; send us a list of
URLs we should hit, so we can automate the testing process (YG)

LM: it would be great to do some multivendor proxy chains.

RK: even if we set up a chain from the same vendor...

JG: most features in here can be implemented in 1.0 and that's just fine
(so let's all put our 1.0 proxies in the test?)

HFN: put needs to be tested (with chunked, byterange, through proxys,
put things should be tested as rigorously as GET.

....

Do you [ms] have any international test sites?

Henry: we have our own satellite networks, but we need to have plug-nd-play
boxes shipped in. We have a network simulator (sits between two lans and it
sets error rate, delay, etc). If we ever do a face-to-face, we'll bring it.

LM: did anyone get back to the connectathon folks.  (Quentin Clark, 
cthon@sun.com)
Scott has been in touch, but no firm plan.

Scott asked for a test profile.

<time to order dinner>

Henry: speaking of connectathon, a few years, they did a TCPIP. ALl they
provided was powerand space, not too useful. We'd be willing to do that.

YG offers space, power, catering AND a trip to the company store...

Section 3.1 continued, lots of straightfoward ones.

JG suggests reading it, rather than reading aloud.

Rich: I'm worried about proxies up and downgrading to versions.
Henry: I know it does that. Rich thinks so, but doesn't know for sure.
(RE-VERSION)

Reference to the leach / mogul versioning draft. Final text is in the
spec since munich (JG). Jigsaw does -- I sent a 1.0 req and it upgrades
(Scott reports).

3.2.2 has a SHOULD which is too srong (numeric IP addresses). IP Addresses
are FQDNs and fully legal. SHOULD become should -- lowercase.

3.2.3
HFN: the main kludge there is spaces.
YG: we're not compliant and *can't be* . Our servers are not case-sensitive
and we're not going to change it. We had a big long meeting about it...

LM: of COURSE you do this. YG: no, we don't

JG: we have atleast two W3C environments which do it, though.

JG: we've mad w3.org case-INsensitive... /HyperText/MarkUp was confusing 
everyone...

YG: in a perfect world, the server should be case-insensitive and does the 
mapping and sends Location: . But, we work offline on DOS, so we strip case 
on the
client.

JG: my rule is protocols should be case-preserving but insensitive.

Summary: do we have a third implementation which is case-preserving?

Henry: if you have a client acting as a client-side cache, is the lookup
beyond the host-name case sensitive. IE is insensitive. IBM had to 
implement
an escaping system to be preserving -- so their proxy does.

Scott commits to adding a test case for this.

3.3 -----

must accept all three date formats (henry, scott, rich, and yaron say yes)

3.3.1 ----

Yes, they're silly, but do they work?

PASSED

3.3.2 ---
SL: WHY is it here? who uses delta-seconds? Editorial issue to JIM

3.4 ---

YG: we are compliant
Rich: we are compliant

Henry: I think the MUST is not necessary, editorial issue -- should it be
normative . NEVER MIND -- it's a quote from the MIME source text.

3.5 ---

IE supports deflate and gzip. NOT compress. NOT support x-gzip. Yaron to 
harass.

JG: issue around "identity"

RIch: we don't send idenity (though could be configured)

YG: we NEVER send qvalues.

HFN: w3c handles Identity - -HFN to check this.

Henry: we don't know about identity, since it's post-2068.

Scott to check as well.

HFN: C-E identity should never happen. ONLY in A-E.

-----

Chunked transfer encoding: sent by many.

YG: our proxy is 1.0, hence doesn't do chunked onward (same for Rich)

Henry: IMHO, pipelined PUT and POST is looking for trouble...

Henrik says W3C code is OK.

IBM: has not impl trailers. Scott: no trailers

THERE ARE NOT TWO IMPLS OF TRAILERS AT THIS TABLE -- no one is generating 
them!
THERE ARE INTEROPERABLE CHUNKS
THERE ARE INTEROPERABLE TRAILERS
BUT NOT TOGETHER>

<Dinner>


3.6.1 ----

DONE

3.7 ----

LWS in mime type.
(client-side only?)

YG: compliant
W3C: compliant

PROCESS QUESTION: When documenting for DRAFT standard, do you have to 
document
WHICH two, or that there ARE two.

in 3.7.0 -- parametrization of mimetypes, forking of viewer. Case 
insensitive.

IE ignores charset on mime-type.
editorial [sic] - YARON Has more  -- "to the and inform"

YG: we sniff charset from the stream. We don't do charset. Uses statistical
algorithms to guess charset.

LM decides IE4 is NOT compliant because it ignores charset parameters.
Charset is the RIGHT way.

YG: I spent 2 hours in a life-and-death battle, and we decided life sucks.
I had i18n experts begging me to do some fixes. We lost. URLs are OPAQUE.

JG: the answer from Jon Postel is that we should have documented who has 
done which
things. We should go by sections and note exceptions.

RESOLVED: there ARE two W3C user-agents that passes parameters correctly
(including lynx and netscape)

LM: Are there any clients that send charset on put or post.
(Henrik does put/post with chunked. IE5 will, too)

------

Multipart types

IBM decodes some multiparts.
MHTML defines a use for multipart.
LM: do we have implementatons of multipart at all?
YES: file upload is multipart, and thus:
Microsoft admits, yes on client, yes on server.

IBMs proxy may not check that the epologue isn't empty.

----
3.8
syntax of UA -- all compliant
(Kristol isn't yet)
----
3.9
qvalues
rathole
End of dinner.

Received on Wednesday, 10 December 1997 10:23:01 UTC