462243 - Create an HTML document that links or intersperses commentary with mnot's cache tests

Status: RESOLVED INCOMPLETE

(Core :: Networking, defect)

Description

[Robert Sayre]

We have a good analysis of mnot's tests, but it's hard to relate each point to the specific tests. We should make it easier before we present the document to the HTTP working group.

Comment 1

[Bjarne (:bjarne)]

Attachment: PDF document

I decided to export it as PDF in order to control formatting.

Comment 2

[u7736]

Very cool. I'll take a look and correct tests or give feedback here as appropriate; however, it may take a litle while (things are busy). Please ping me if I go quiet too long; guilt is a fantastic motivator.

Comment 3

[Bjarne (:bjarne)]

Attachment: xpcshell test for request cache-control

Implements the (claimed) proper expected behaviour for no-cache and no-store cache-control directives. Comments and/or corrections are welcome. Note that this test fails with current code from mozilla-central.

Comment 4

[Bjarne (:bjarne)]

Attachment: Test for query-freshness

Detailed testing for URIs with query-string (GET requests with '?' in the request-string)

Comment 5

[Bjarne (:bjarne)]

(In reply to comment #2)
> Very cool. I'll take a look and correct tests or give feedback here as
> appropriate; however, it may take a litle while (things are busy). Please ping
> me if I go quiet too long; guilt is a fantastic motivator.

Ping... :)

Comments on the expected results in the attached tests would be sufficient, I think. Once we agree that the tests are expecting the correct results I'll provide patches to necko.

Comment 6

[Bjarne (:bjarne)]

Testcase and patch addressing the directive-precedence has been added to bug #203271

Comment 7

[Bjarne (:bjarne)]

I'm not quite sure why this patch only fixes first half of the test on mnot's page...  there is no reason why the order of headers in the response should matter. I'll look into the raw data from the network later.

Comment 8

[Bjarne (:bjarne)]

(In reply to comment #2)
> Please ping me if I go quiet too long; guilt is a fantastic motivator.

*ping*

Comment 9

[Bjarne (:bjarne)]

(In reply to comment #8)
> (In reply to comment #2)
> > Please ping me if I go quiet too long; guilt is a fantastic motivator.
> 
> *ping*

*beep*...

Comment 10

[u7736]

OK, going through the issues you raise;

* Cache-Control: no-store request header honoured -- I agree with your analysis. The second response should not be cached. Will fix.

Regarding reload upon no-store - that's a good question, might be worth an issue in httpbis. Intuitively, no-store implies no-cache, but this isn't said explicitly.

* Cache-Control: no-cache request header honoured / Pragma: no-cache request header honoured -- these tests actually need to be re-structured. no-cache as a request directive means that the corresponding response not be taken from cache, not that the response shouldn't be cached for future use.

* Heuristic freshness w/ "?" -- was a server configuration problem (mod_expires crept in after deployment); should be fixed now. FF still seems to be failing. 

* Redirects w/ and w/o freshness information -- fixed server scripts to generate redirects correctly. FF still seems to be failing.

* Cache invalidation -- agreed, will look into fixing.

Comment 11

[Bjarne (:bjarne)]

First, thanks for making time to respond to this. Some comments...

(In reply to comment #10)
>  Intuitively, no-store implies no-cache, but this isn't said
> explicitly.

Yes, this is the problem. Looking at it from the other angle : If no-store does *not* imply no-cache, it would have to return any cached response and *then* remove all traces from the cache. Thus it would have the side-effect of deleting entries from cache and in practice mean something like : "Get the resource either from cache or from server, but make sure there are no traces in my cache afterwards". It could be spec'ed like this, I guess, but it's a rather complex semantics to attach to a request-header directive.

> * Cache-Control: no-cache request header honoured / Pragma: no-cache request
> header honoured -- these tests actually need to be re-structured. no-cache as a
> request directive means that the corresponding response not be taken from
> cache, not that the response shouldn't be cached for future use.

Do you agree with the expected pattern at page 3 in the document (assuming the intuitive behaviour of no-store above)? Attachment #346423 [details] is supposed to implement this pattern and I use it as a testcase.

> * Heuristic freshness w/ "?" -- was a server configuration problem (mod_expires
> crept in after deployment); should be fixed now. FF still seems to be failing. 

Yes...  I've added bug #468594 to address this separately.

> * Redirects w/ and w/o freshness information -- fixed server scripts to
> generate redirects correctly. FF still seems to be failing.

This is a real problem, yes. Bug #415086 and bug #341189 are probably variations of this - do you agree?

Comment 12

[Bjarne (:bjarne)]

(In reply to comment #10)
> * Redirects w/ and w/o freshness information -- fixed server scripts to
> generate redirects correctly. FF still seems to be failing.

Yes, my installed FF 3.0.3 fails on these, but they seem to be ok in my compiled version from trunk. I'm not aware of any specific patch that that may have fixed the problem, though. Maybe biesi knows?

Comment 13

[Bjarne (:bjarne)]

*ping*

Comment 14

[Bjarne (:bjarne)]

*beep*  (worked nicely last time...  :) )

Comment 15

[Bjarne (:bjarne)]

*ping*

Comment 16

[Bjarne (:bjarne)]

*ping again*

Comment 17

[Bjarne (:bjarne)]

ping....

Comment 18

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Bjarne, please stop pinging and better request review on your patches from the correct person.

Comment 19

[Bjarne (:bjarne)]

-> default owner

Comment 20

[Mark Nottingham]

FYI, I've ported the cache-related tests (finally) to WPT - see:
  https://github.com/w3c/web-platform-tests/pull/5137

Overall, looks good, with one or two minor exceptions; will create a couple of bugs.