73490 - HTTP URLs with '?' characters should not be cached unless an expiration time is explicitly provided by the server

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[Darin Fisher]

HTTP URLs with '?' characters should not be cached unless an expiration time is
explicitly provided by the server.  See section 13.9 of RFC2616 for details.

Comment 1

[Darin Fisher]

Attachment: fixes the problem

Comment 2

[Gagan]

sans the printfs, r=gagan

Comment 3

[Gagan]

you may want to add a comment to logs instead of the printfs

Comment 4

[Darin Fisher]

Attachment: cleaned up patch

Comment 5

[Darin Fisher]

dougt says "sr=dougt"

Comment 6

[Alec Flett]

the fact that we're searching the whole spec for a '?' worries me - could a
multibyte-char hostname (now that they are allowed) contain an 8 bit character
which happens to match '?'? it would suck for that entire host or domain to have
their pages uncached.

Comment 7

[Darin Fisher]

good point.  i'll have to be more careful.  perhaps it is enough to check
whether or not the URL has a non-empty "query" portion (via nsIURI::query).

would this be sufficient?

Comment 8

[Darin Fisher]

Attachment: new patch (check for non-empty query string)

Comment 9

[Judson Valeski]

that seems sufficient to me. if it's not, that would imply url parsing problems.
r=valeski on the 3/26/01 21:28 patch.

Comment 10

[Judson Valeski]

*** Bug 73600 has been marked as a duplicate of this bug. ***

Comment 11

[Bradley Baetz]

*** Bug 73572 has been marked as a duplicate of this bug. ***

Comment 12

[Alec Flett]

sr=alecf 
(though you might run into platform-specific bustage with that "if (str...)"
because I'm not sure all platforms will know to cast that nsXPIDLCString to a
const char *)

Comment 13

[Darin Fisher]

alecf: not a problem... that convention is used all over the place in HTTP.

Comment 14

[Judson Valeski]

yea, nsXPIDLCString does the right thing xp now.

Comment 15

[Peter Lairo]

Spam? (if yes, sorry): The cached url's in recent builds (2001-03-27) have
become a major annoyance when reporting bugs. Going to a bug page previously
visited and recently changed (bug reported) *shows the old state*. Pressing
SHIFT-Reload is a pain and most users will not know how to or allways remember
to do this.

suggest keyword: *nscatfood*

Comment 16

[Darin Fisher]

fix checked in.

Comment 17

[Boris Zbarsky [:bzbarsky]]

This seems to have broken going back to forms.  If I fill out a form at say
http://bugzilla.mozilla.org/show_bug.cgi?id=73490, then click on "Vote for this
bug" and then hit "back", the form data should not be cleared....  Currently it
is (linux build 2001-04-02-05).

Should I open a new bug on that?

Comment 18

[Darin Fisher]

yes please do.. that is probably a side effect of the data not being cached.

Comment 19

[Boris Zbarsky [:bzbarsky]]

bug 74639 filed.

Comment 20

[Dennis Andersen]

I think it's a bad idea not to cache all URLs with '?'. I think that all pages 
with '?' should be cached except the pages wich sends the "no-cache-header".

Comment 21

[Darin Fisher]

deniande: please read section 13.9 of RFC1616 (http://www.w3.org/Protocols) and
let me know how you interpret it... it seems very clear about this point.

Comment 22

[Alec Flett]

I think we need to cache them through session history, but not in the case of it
being typed in a url bar, or clicked on through a link..

Comment 23

[Darin Fisher]

sounds right to me too.  i've talked to radha and it looks like we can use
the same solution for POST urls to accomplish this.  namely, HTTP will add
a unique id in the cache key.  only history will know this cache key, and 
so only history will be able to retrieve data for these urls from the cache.

see bug 56346 for details.

Comment 24

[Dennis Andersen]

Darin: I know that RFC1616 says that URLs with '?' characters should not be 
cached, but let me describe what I want.

I would like that prevously visited pages should be availibe when I am browsing 
offline. For example these "bug-pages". Before this bug was fixed, I could 
browse these pages offline, but now I can't because there is a '?' character in 
the URL.

Internet Explorer does this, so I'm currently using IE, because I use this 
future a lot. Before this bug was fixed I used Mozilla, and I want to use it 
again, but not before I can browse visited pages offline.

Comment 25

[Dennis Andersen]

We could do so that the page isn't loaded from cache when you are working 
online, but when you are working offline (File => Work Offline), then the page 
is loaded from the cache...

What do you think about that?

Comment 26

[Darin Fisher]

Dennis: see bug 56346... i think you'll agree that the solution described there
will meet your concerns as well.

Comment 27

[Dennis Andersen]

Darin:

From bug 56346:

"ok, i've discussed this with gagan and gordon, and we've come to the conclusion
that HTTP should just put everything in the cache."

Does that mean that I would be able to browse all visited webpages offline
(inclusive URLs with '?')?

Comment 28

[Peter Lairo]

If everything is in the cache, how do we deal with the original problem: Going
back to a bug after making a change doesn't show the changes?

Comment 29

[Boris Zbarsky [:bzbarsky]]

Peter: "put everything in the cache" does not mean "load everything from the cache".

Verified fixed in the 2001-04-03-11 build, by the way.

Comment 30

[Boris Zbarsky [:bzbarsky]]

er, that was 2001-04-03-11 Linux build.  :)

Comment 31

[benc]

Can anyone easily provide a URL that has and does not have the expiration time?

Comment 32

[Darin Fisher]

perhaps a bugzilla query?

Comment 33

[Darin Fisher]

or rather, perhaps a bugzilla bug link such as:

http://bugzilla.mozilla.org/show_bug.cgi?id=73490

Comment 34

[Gagan]

benc: you can use the testserver to set something up. seek help from bbaetz.

Comment 35

[Tracy Walker [:tracy]]

mass remove verifyme requests greater than 4 months old