Bug 20806 – Referer is not sent when going back to an uncached page (blank page at kin.naver.com)

• Component: History
• Status: RESOLVED FIXED

[Jungshik Shin]

How to reproduce:

1. Go to http://kin.naver.com/list/list_kinup.php?d1id=6&dir_id=601 
2. Around the bottom of the page, there's a list of pages (1 2 3 4 5 6 
...). Click on one of them (other than '1') to go to the corresponding 
page.
3. If you click on '3', the url will be http://kin.naver.com/list/list_kinup.phpd1id=6&dir_id=601&sort=kinup_point&
page=3

4. Click on the 2nd column from the left in the list of Q&A items with 
'thumb-up' icons in the rightmost column.

5. The url will be something like this:
http://kin.naver.com/detail/detail.php?
d1id=6&dir_id=60101&eid=jbyUO8ElqrWTXbxk7ObSqBSon7Nxrg8Z&l_url=L2xpc3QvbGlz
dF9raW51cC5waHA/ZDFpZD02JmRpcl9pZD02MDEmc29ydD1raW51cF9wb2ludCZwYWdlPTM=

6. Press 'the back button'.

7. The url in the address bar is 
http://kin.naver.com/list/list_kinup.phpd1id=6&dir_id=601&sort=kinup_point&
page=3

Exactly the same as what you had in step 3

What's expected:

In step 7, get the same page as shown in step 3

Actual result:

In step 7, a blank screen is shown.   

Additional info : 
IE 7 and Firefox do not have this issue. Chrome has the same issue ( http://code.google.com/p/chromium/issues/detail?id=2189 : Chrome shows HTTP 404 error page. ). So does Safari with webkit trunk build on Mac OS 10.4.11

[Jungshik Shin]

FYI : Naver is #1 portal in Korea with ~70% market share. 

[Alexey Proskuryakov]

A blank page is what this server gives for the page 3 URL when loading it manually, too. Compare output of

curl -i "http://kin.naver.com/list/list_kinup.php?d1id=6&dir_id=601&sort=kinup_point&page=3" --header "Referer: http://kin.naver.com/list/list_kinup.php?d1id=6&dir_id=601"

and 

curl -i "http://kin.naver.com/list/list_kinup.php?d1id=6&dir_id=601&sort=kinup_point&page=3"

So, the issue is two-fold. First, the page doesn't go to back/forward cache, that's bug 13631. Second, we don't re-send Referer header when going back involves a reload. I think that we should fix both.

[Jungshik Shin]

Thanks for the diagnosis. 

So, if we go directly to 'http://kin.naver.com/list/list_kinup.php?d1id=6&dir_id=601&sort=kinup_point&page=3', we'll get 404 error. 

Ick ...   I guess I know why Naver does that ....

[Eric Roman]

Created attachment 24935 [details]
Preserve the request referrer across history navigations.

This patch replaces the two referrer fields in HistoryItem (m_rssFeedReferrer, m_formReferrer) with a single generic one (m_referrer). It doesn't look like the distinction was necessary.

I am not sure how to write a test for this, ideas welcome (in particular how to avoid page cache in layout test?)

[Alexey Proskuryakov]

Page cache is currently always disabled in DumpRenderTree. There are also many examples of reasons of documents not to use it in bug 4123.

But these are all bugs that need to be fixed some day.

[Eric Roman]

Created attachment 25032 [details]
Send referrer on history navigations.

Added layout test.

[Eric Seidel (no email)]

Comment on attachment 25032 [details]
Send referrer on history navigations.

This looks good from my perspective, but I think Bradee or Anders will need to chime in.  Apple will need to do an internal code search before removing the RSSReferrer methods (which could be changed to just grab referrer() (possibly conditionally on if it's an RSS url) if necessary as a crutch.

[Eric Seidel (no email)]

Comment on attachment 25032 [details]
Send referrer on history navigations.

On second thought, I would suggest *not* removing the RSSFeedReferrer methods, since that will make this harder to review.  The Apple folks who maintain those APIs should take care of that.  You should just fix those methods to do the right thing, and then the Apple folks can come back and remove them when they're not needed.

[Eric Roman]

Created attachment 26534 [details]
Send referrer on history navigations.

> I would suggest *not* removing the RSSFeedReferrer methods,
> since that will make this harder to review.  The Apple folks who maintain those
> APIs should take care of that.

Done.
I removed the deletion of the "rss" and "form" referrers, and marked them as deprecated instead.

[Eric Roman]

Created attachment 26543 [details]
Send referrer on history navigations.

I had misunderstood Eric Seidel's previous comment.
This time it should be what he had suggested.

Do I also need to add the new symbol to WebCore.order ?

Thanks!

[Eric Seidel (no email)]

Comment on attachment 26543 [details]
Send referrer on history navigations.

Apple maintains the order files.  They'll get regenerated when necessary.  Missing symbols from an order file just won't get sorted to the front of the linked binary.

This patch looks good to me, but ideally someone from Apple would comment about if this could have unintended consequences for RSS stuff.

[Eric Roman]

ping. Any feedback on this?

[Eric Seidel (no email)]

Comment on attachment 26543 [details]
Send referrer on history navigations.

This looks fine to me.  r=me.

[Eric Roman]

http://trac.webkit.org/changeset/40132

[Alexey Proskuryakov]

So, this was landed a while ago, marking as FIXED.