90004 - Proxy reload doesn't "load"

Status: RESOLVED WORKSFORME

(Core :: Networking: HTTP, defect)

Description

[Ben Bucksch (:BenB)]

Reproduction:
1. Load a webpage you own with an image
2. Change the webpage on the server, both the text and the image.
3. Press the Reload button on the Navigator toolbar

Actual result:
- The page is re-rendered, but still shows the old content.

Expected result:
- The new, altered page (both text and image) gets displayed.

Additional Comments:
- I use a Squid Proxy with the following config:

refresh_pattern         (gif|jpg|jpeg|png)$             0       10%     2880
#                       images                                          2 days
refresh_pattern         .                               0       5%      1440
#                       rest (html, cgi etc.)                           1 day

I assume that there are no proxy or server bugs involved.

- I don't know, if Navigator UI uses the wrong flag or if the network layer
behaves badly. Trying with the latter first.

- In this bug, I don't ask for a full reload of the page in all cases. But
Mozilla must at least check, if the html or embedded object changed, by
comparing "last changed" dates. And it must work with standard software.

Comment 1

[tenthumbs]

I'm seeing the expected behavior with an 0706 nightly. Mozilla is
talking to my own squid. I created a page with an image, pointed mozilla
at it, changed the page and the image, and did a reload (not shift
reload). Mozilla issued If-Modified-Since and If-None-Match headers for
the page and the image. Squid did the right thing and fetched the
updated data.

Either your squid is strangely configured or there's been a regression
since 0706.

Comment 2

[Ben Bucksch (:BenB)]

tenthumbs, my squid is pretty much standard (apart from the cited configs).

Do you have an idea what could be causing this? I don't know :-(.

Do you remember our discussion about what should be the default for the reload
button, months ago?This bug is exactly what I was scared about. SHIFT-Reload did
what I expected.

Comment 3

[Ben Bucksch (:BenB)]

Oh, I used 0.9.2.

Comment 4

[tenthumbs]

Yes, I remember the discussion. I think mozilla's doing the right thing
now. I'll attach an extract from squid's log. It appears quite normal.

Maybe there's been a code change since 0.9.2.

Comment 5

[tenthumbs]

Attachment: squid log

Comment 6

[tenthumbs]

That's two successive reloads in the attachment, changing the content before
each reload. Maybe your logs will show something different.

Comment 7

[Ben Bucksch (:BenB)]

All: I sent tenthumbs my config, and he said that it is almost the same as his one.
If anyone knows, if and what code changed between 0.9.2 and 20010706, please
tell so.

Comment 8

[Doug Turner (:dougt)]

gordon, is this a dup of 79983?

Comment 9

[gordon]

Doug, no, I don't think so.  Sounds like it's fixed anyway.

Comment 10

[gordon]

Ben, can you try with a more recent build?  If it works, let's close this.

Thanks.

Comment 11

[benc]

ben: Is you cache set to "never"? Also, what platforms do you see this on?

Using MacOS 2001-07-20-03-092 (fasted build I could find)...

I found that reload sent a new request (with an if-modified-since) and loaded
fine. Communicator seemed to work in a similar way. I did a quick-and-dirty
trace, too nasty to clean up and post here at this time.

My cache is set to "always".            

Comment 12

[Ben Bucksch (:BenB)]

benc,

> Is you cache set to "never"?

It must not matter. I said reload.

> Also, what platforms do you see this on?

Linux

> Communicator seemed to work in a similar way

4.x was broken, too.

Comment 13

[Ben Bucksch (:BenB)]

I tried the Mozilla nightly trunk 2001-07-25 egcs build, and I couldn't
reproduce the bug. But I couldn't reproduce it even on the build where I
originally saw th bug. It might be something odd like bug 79983 (I did try
deleting stuff, but it didn't help with reproduction).

Comment 14

[gordon]

Bug 79983 is effectively fixed for Mac, Windows, and Linux (and OS/2 I believe).
 We're working on extending the fix to other platforms, but that may require
waiting until we land a new version of NSPR.

Ben Bucksch, since we can't reproduce this anymore I'm going to close this bug
as FIXED, but if you can reproduce the problem again please reopen it or submit
a new bug.  Thanks for the report.

Comment 15

[Ben Bucksch (:BenB)]

Nope, not fixed (REOPENing), and I have a reproducable, easy reproduction:
1. Go to <http://www2.berlin-consortium.org/wiki/html/Berlin/TestMe.htm>
2. "Edit this page"
3. Type some text
4. Submit
-> You will still see the old page. - Expected: You see your new text.
5. Normal Reload
-> You will see your new text.
6. "Edit this page" again
-> You will see the old content. - Expected: You see your new text.
7. Reload
-> No change - Expected: You see your new text.
8. Shift-Reload
-> You see your new content.

This bug is about: At least step 7 should already give the result that step 8 gives.

Comment 16

[Ben Bucksch (:BenB)]

Attachment: prefs.js used for reproduction

Comment 17

[Ben Bucksch (:BenB)]

Attachment: Corresponding squid log

Comment 18

[michael]

same thing here. i use winroute as proxy and the reload does not work 
(shift-reload works).
it works fine with netscape 4.78, but not in mozilla 092 or 093.
but mozilla 092 works on a squid proxy i tested (could'n test 093, yet)

Comment 19

[tenthumbs]

Just wondering which squid version you're using? I'm running 
squid2.4STABLE1 plus all the approved patches.

Since you have access to the server, you might try turning on
log_mime_hdrs in squid's configuration. It will bloat the log so turn it
on only for testing.

Comment 20

[michael]

tenthumbs:
if this question came flying in my direction, i'm sorry, no access to the squid 
proxy. can't even check the version right now, 'cause it's the companies proxy 
and i'm on vacation.
i have to play around with the winroute logs a bit.

Comment 21

[Ben Bucksch (:BenB)]

> which squid version you're using?

2.3.STABLE4

Comment 22

[kyles]

I'm seeing this on build 2001082203, Windows 2000.  I had a page showing old
data, reload did not show any new content.  Shift-reload did.

    - Kyle

Comment 23

[kyles]

Is this a cache problem?   After adding the previous comment, I clicked on the
"Go Back to Bug..." button.  It loaded this bug page, with the old data, not
including my comment.  I got to see my comment after a shift-reload.

    - Kyle

Comment 24

[gordon]

Darin, this looks like a HTTP proxy issue.

Comment 25

[gordon]

Assigning to Darin rather than Neeti...

Comment 26

[Darin Fisher]

kyle: i think you were seeing bug 96480, which was a blocker today and has been
fixed.

Comment 27

[Darin Fisher]

reporter: can you please confirm this bug against a recent nightly bug?  thx!

Comment 28

[Darin Fisher]

i believe this is fixed.  reporter: please verify.

-> punting from 0.9.4

Comment 29

[Ben Bucksch (:BenB)]

nope, not fixed. same result as 2001-07-30 19:22. Tested with Mozilla Linux
gcc295 2001-08-27-22.

Comment 30

[Kunihiko Aizawa]

I have same problem on 0.9.3 (Win32 and OS/2) and SmartCache http proxy.
I think a key is no-cache directive (to http proxy) in a request header.

NC4.78 and IE5.01 add "Pragma: no-cache" to request header not only when
shift-reload but also reload. For these browsers, the difference between
reload and shift-reload is only whether add "If-Modified-Since:" to a header.
But Mozilla add no-cache directive only when shift-reload.
(Reference: Bug 81125)

Comment 31

[Darin Fisher]

any pointers to where can i find more info on the SmartCache http proxy?

Comment 32

[Kunihiko Aizawa]

Smart Cache is written in Java and is distributed at
http://home.worldonline.cz/~cz210552/

Comment 33

[Ben Bucksch (:BenB)]

I just saw the bug again, with an ordinary static webpage (no query/CGI or
anything). Server: www.beonex.com. Page updated via scp. Mozilla 0.9.2.1.

Comment 34

[Darin Fisher]

ben: any chance you could repeat your testcase using a recent nightly build?

Comment 35

[Ben Bucksch (:BenB)]

Yes, if you fix it some time in thenear future :).

Still exactly same result (testcase in comment from 2001-07-30 19:22).

Comment 36

[Darin Fisher]

-> 0.9.8

Comment 37

[Michael D. Crawford]

I see the bug appearing when using WinRoute Pro 4.1.27 as my proxy on NT4
service pack 6, using Mozilla 0.9.5 on a Windows 98 machine.

I wouldn't be too suprised if this is in large part caused by WinRoute, as I've
found it to have lots of problems (corrupting FTP data, freezing ssh sessions)
but I do find the expected behaviour when I used Netscape 4.76 on Windows.  I
can use Netscape to force the proxy to update its cache and then view the
correct page with Mozilla, but no amount of reloading with just Mozilla will work.

You can get WinRoute from http://www.winroute.com/  They have a 30 day free trial.

Comment 38

[Darin Fisher]

try configuring mozilla to use HTTP/1.0 and/or disable keep-alive connections.

you can find settings for these in the debug->network preferences panel found
with nightly builds.

Comment 39

[Michael D. Crawford]

Reload through WinRoute Pro still doesn't work in 0.9.6.  As Darin Fisher
suggested, I tried using HTTP 1.0 and disabling keepalive, in various combinations:

1.1, keepalive enabled: doesn't work
1.1, keepalive disabled: doesn't work
1.0, keepalive enabled: doesn't work
1.0, keepalive disabled: doesn't work

Just to make sure the preference was being applied, I quit and relaunched
Mozilla after changing each preference.

This is with WinRoute Pro 4.1.27.  WinRoute is running on Windows NT4 Service
Pack 6, Mozilla running on a Windows 98 machine.  The NT4 box is connected to
the internet via PPP dialup, the windows 98 machine connected to the NT4 box via
10baseT ethernet.

You can get WinRoute pro from http://www.tinysoftware.com/ - it has a 30 day
free evaluation.

I see that there is a 4.1.30 available, I will download and install it and
report the results.

Comment 40

[Kunihiko Aizawa]

Same here, with Smart Cache 0.59.
Changing HTTP version/keep-alive of mozilla doesn't let Smart Cache
refresh its cached object.

Why mozilla doesn't have "Pragma: no-cache" header in request when reload?
I think mozilla should have the header at least when HTTP version is set as 1.0.
I see following note at section 14.9 "Cache-Control" of RFC2626:
> Note that HTTP/1.0 caches might not implement Cache-Control and
> might only implement Pragma: no-cache (see section 14.32).

Comment 41

[Darin Fisher]

agreed... it would be really nice if we had some code in place to detect the
proxy version, and base the reload request headers on that.

-> 0.9.9 unfortunately

Comment 42

[Darin Fisher]

Attachment: patch - partial solution

this patch partially solves the problem by forcing mozilla to send 'Pragma:
no-cache' on reload if configured to speak HTTP/1.0 in the preferences.

ultimately, mozilla should instead record information about the proxy server
and use that to determine what headers to send on reload.

Comment 43

[Darin Fisher]

cc'ing gagan for input... gagan can you review the patch?

Comment 44

[Gagan]

Comment on attachment 66255 [details] [diff] [review]
patch - partial solution

r=gagan

Comment 45

[rpotts (gone)]

Comment on attachment 66255 [details] [diff] [review]
patch - partial solution

sr=[e-mail]

Comment 46

[Darin Fisher]

checked in patch...

probably won't make any further progress on this bug in the 1.0 timeframe. 
reducing priority.

-> future

Comment 47

[Kunihiko Aizawa]

I have confirmed that Win32-2002012608 can reload pages fine via Smart Cache
if Preferences/Debug/Networking/HTTP Version is set "1.0".

Comment 48

[Asa Dotzler [:asa]]

is this bug 79990?

Comment 49

[Ben Bucksch (:BenB)]

No.

Comment 50

[basic]

remove self

Comment 51

[Darin Fisher]

-> default owner