Bug 41797

Summary: REGRESSION (HTML5 parser?): Impossible to get past the CAPTCHA on postcode.royalmail.com
Product: WebKit Reporter: David Evans <dave.evans55>
Component: DOMAssignee: Adam Barth <abarth>
Status: RESOLVED FIXED    
Severity: Normal CC: abarth, ap, beidson, commit-queue, eric, koivisto, webkit.review.bot
Priority: P1 Keywords: InRadar, Regression
Version: 528+ (Nightly build)   
Hardware: All   
OS: All   
URL: http://postcode.royalmail.com/
Bug Depends on:    
Bug Blocks: 41115    
Attachments:
Description Flags
Patch
none
Patch none

David Evans
Reported 2010-07-07 14:03:34 PDT
At the Royal Mail postcode finder site it is impossible to get past the CAPTCHA no matter how many times you try. However, trying it with Safari without a nightly build webkit, or Firefox 3.6.4, and you will get past it in one or two goes. Safari Version 5.0 (6533.16, r62632) Try finding the postcode for: Building number: 5 Street: Rochford Avenue Town: Shenfield or any other reasonable UK address.
Attachments
Patch (7.78 KB, patch)
2010-07-12 16:19 PDT, Adam Barth 		no flags
DetailsFormatted DiffDiff
Patch (3.88 KB, patch)
2010-07-12 16:29 PDT, Adam Barth 		no flags
DetailsFormatted DiffDiff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.
Alexey Proskuryakov
Comment 1 2010-07-08 12:00:43 PDT
How does one get to see a CAPTCHA on this site? I don't see it anywhere, all I could find mentioned was a limit of 15 searches per day.
David Evans
Comment 2 2010-07-09 03:03:40 PDT
I think you have to do a few searches before you get to see the Captcha.
Alexey Proskuryakov
Comment 3 2010-07-09 11:38:14 PDT
> I think you have to do a few searches before you get to see the Captcha. I'm seeing the CAPTCHA today. The behavior is a little unstable, so it's hard to bisect to find out when this started, but I suspect HTLM5 parser. The problem here is that we request the captcha image twice, so server side state gets out of sync with what is actually displayed. The first request is made by normal parser, and the second one is made by preload scanner. This may or may not be a bug in loader, but preload scanner getting behind normal parsing makes no sense at all. This probably affects more than just one site.
Alexey Proskuryakov
Comment 4 2010-07-09 11:38:34 PDT
<rdar://problem/8175896>
Adam Barth
Comment 5 2010-07-09 11:42:47 PDT
The second request comes from the preload scanner... Interesting. That could be because the preload scanner request has lower priority than the regular request. If they get kicked off close to each other, the regular one might win. We might need to be more agressive about canceling preload scanning.
Alexey Proskuryakov
Comment 6 2010-07-09 13:04:58 PDT
> That could be because the preload scanner request has lower priority than the regular request. I was talking about Loader::Host::addRequest(), sorry for making it too terse. Preload scanner actually sees the tag after real parser does.
Adam Barth
Comment 7 2010-07-11 13:47:12 PDT
I'll look at this soon.
Adam Barth
Comment 8 2010-07-12 16:19:57 PDT
Created attachment 61282 [details] Patch
Eric Seidel (no email)
Comment 9 2010-07-12 16:22:13 PDT
Comment on attachment 61282 [details] Patch It would be cleaner if you could get rid of the AAAAs by using an expression in php and stick them in a <div> or something which you can display none.
Adam Barth
Comment 10 2010-07-12 16:29:38 PDT
Created attachment 61284 [details] Patch
Eric Seidel (no email)
Comment 11 2010-07-12 16:47:03 PDT
Comment on attachment 61284 [details] Patch Thanks!
WebKit Commit Bot
Comment 12 2010-07-12 18:28:25 PDT
Comment on attachment 61284 [details] Patch Clearing flags on attachment: 61284 Committed r63154: <http://trac.webkit.org/changeset/63154>
WebKit Commit Bot
Comment 13 2010-07-12 18:28:30 PDT
All reviewed patches have been landed. Closing bug.
WebKit Review Bot
Comment 14 2010-07-12 18:52:46 PDT
http://trac.webkit.org/changeset/63154 might have broken GTK Linux 32-bit Release
Eric Seidel (no email)
Comment 15 2010-07-12 19:05:58 PDT
Committed r63158: <http://trac.webkit.org/changeset/63158>
Note You need to log in before you can comment on or make changes to this bug.