Re: evil webkit bug, loads of junk DNS requests
On 2011/02/14 09:37, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... This is still a problem with the new version of webkit; updated diff below. Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.42 diff -u -p -r1.42 Makefile --- Makefile3 Mar 2011 13:44:10 - 1.42 +++ Makefile6 Mar 2011 10:51:55 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.7 EPOCH =0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 6 Mar 2011 10:51:55 - @@ -0,0 +1,12 @@ +$OpenBSD$ +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sat Mar 5 20:50:25 2011 WebCore/platform/network/soup/DNSSoup.cpp Sat Mar 5 20:51:40 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
yes please On Sun, Mar 06, 2011 at 10:52:16AM +, Stuart Henderson wrote: On 2011/02/14 09:37, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... This is still a problem with the new version of webkit; updated diff below. Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.42 diff -u -p -r1.42 Makefile --- Makefile 3 Mar 2011 13:44:10 - 1.42 +++ Makefile 6 Mar 2011 10:51:55 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.7 EPOCH = 0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 6 Mar 2011 10:51:55 - @@ -0,0 +1,12 @@ +$OpenBSD$ +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sat Mar 5 20:50:25 2011 WebCore/platform/network/soup/DNSSoup.cppSat Mar 5 20:51:40 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
On Mon, Feb 14, 2011 at 10:37 AM, Stuart Henderson s...@spacehopper.org wrote: ... With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... This DNS prefetch stuff is really a crap :( ok for me. David
Re: evil webkit bug, loads of junk DNS requests
I am all for this. In fact I think we should go further and remove the guess url that user typed because he/she can't type code as well and while at it the dns prefetcher. On Mon, Feb 14, 2011 at 09:37:19AM +, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.41 diff -u -p -r1.41 Makefile --- Makefile 30 Dec 2010 17:55:32 - 1.41 +++ Makefile 14 Feb 2011 09:31:47 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.6 EPOCH = 0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 14 Feb 2011 09:31:47 - @@ -0,0 +1,16 @@ +$OpenBSD$ + +don't try and resolve DNS for . for every single mouse movement +in a browser window. from https://bugs.webkit.org/show_bug.cgi?id=38263 + +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sun Feb 13 17:45:39 2011 WebCore/platform/network/soup/DNSSoup.cppSun Feb 13 17:46:05 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
why? disable it in xxterm if you don't like it, some people actually do like it. On (2011-02-14 07:27), Marco Peereboom wrote: I am all for this. In fact I think we should go further and remove the guess url that user typed because he/she can't type code as well and while at it the dns prefetcher. On Mon, Feb 14, 2011 at 09:37:19AM +, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.41 diff -u -p -r1.41 Makefile --- Makefile30 Dec 2010 17:55:32 - 1.41 +++ Makefile14 Feb 2011 09:31:47 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.6 EPOCH =0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 14 Feb 2011 09:31:47 - @@ -0,0 +1,16 @@ +$OpenBSD$ + +don't try and resolve DNS for . for every single mouse movement +in a browser window. from https://bugs.webkit.org/show_bug.cgi?id=38263 + +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sun Feb 13 17:45:39 2011 WebCore/platform/network/soup/DNSSoup.cpp Sun Feb 13 17:46:05 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
I think you don't understand what happens behind the scenes. Every domain you enter gets prefixed and tested against the domain the browser thinks you are in. you type moo on you a.b.com domain and you'll see fantastically smart dns lookups such as: moo.com.a.b.com, moo.net.a.b.com etc. Easily 10 network names that will be wrong. The prefetcher does the opposite of its promise, it slows down the entire browser. By looking up all kinds of crap a user won't click. A good example is news.google.com which does about 350 DNS lookups. All that to save 1 that the user actually clicks. They traded one perceived performance issue for a real one. There is more insanity but these are just examples. On Mon, Feb 14, 2011 at 02:54:32PM +0100, Robert Nagy wrote: why? disable it in xxterm if you don't like it, some people actually do like it. On (2011-02-14 07:27), Marco Peereboom wrote: I am all for this. In fact I think we should go further and remove the guess url that user typed because he/she can't type code as well and while at it the dns prefetcher. On Mon, Feb 14, 2011 at 09:37:19AM +, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.41 diff -u -p -r1.41 Makefile --- Makefile 30 Dec 2010 17:55:32 - 1.41 +++ Makefile 14 Feb 2011 09:31:47 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.6 EPOCH = 0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 14 Feb 2011 09:31:47 - @@ -0,0 +1,16 @@ +$OpenBSD$ + +don't try and resolve DNS for . for every single mouse movement +in a browser window. from https://bugs.webkit.org/show_bug.cgi?id=38263 + +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sun Feb 13 17:45:39 2011 WebCore/platform/network/soup/DNSSoup.cppSun Feb 13 17:46:05 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
I know, I've read the discussion and I am still saying: do not disable it by default. On (2011-02-14 08:07), Marco Peereboom wrote: I think you don't understand what happens behind the scenes. Every domain you enter gets prefixed and tested against the domain the browser thinks you are in. you type moo on you a.b.com domain and you'll see fantastically smart dns lookups such as: moo.com.a.b.com, moo.net.a.b.com etc. Easily 10 network names that will be wrong. The prefetcher does the opposite of its promise, it slows down the entire browser. By looking up all kinds of crap a user won't click. A good example is news.google.com which does about 350 DNS lookups. All that to save 1 that the user actually clicks. They traded one perceived performance issue for a real one. There is more insanity but these are just examples. On Mon, Feb 14, 2011 at 02:54:32PM +0100, Robert Nagy wrote: why? disable it in xxterm if you don't like it, some people actually do like it. On (2011-02-14 07:27), Marco Peereboom wrote: I am all for this. In fact I think we should go further and remove the guess url that user typed because he/she can't type code as well and while at it the dns prefetcher. On Mon, Feb 14, 2011 at 09:37:19AM +, Stuart Henderson wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. Chrome doesn't have this problem (it doesn't seem to have this code in its internal copy of webkit). I tracked down a patch in https://bugs.webkit.org/show_bug.cgi?id=38263 which fixes this. It seems a bit of a hack (why is it bothering to look these up internally anyway?) but it's way better than what we have now. With the port diff below, I can now start xxxterm or midori, open a page, and move the mouse over the window, without sending hundreds or thousands of junk requests to the nameserver. If you're using this on a metered or slow internet connection it's quite an improvement... Index: Makefile === RCS file: /cvs/ports/www/webkit/Makefile,v retrieving revision 1.41 diff -u -p -r1.41 Makefile --- Makefile30 Dec 2010 17:55:32 - 1.41 +++ Makefile14 Feb 2011 09:31:47 - @@ -4,6 +4,7 @@ COMMENT = open source web browser engine DISTNAME = webkit-1.2.6 EPOCH =0 +REVISION = 0 CATEGORIES = www HOMEPAGE = http://webkitgtk.org/ Index: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp === RCS file: patches/patch-WebCore_platform_network_soup_DNSSoup_cpp diff -N patches/patch-WebCore_platform_network_soup_DNSSoup_cpp --- /dev/null 1 Jan 1970 00:00:00 - +++ patches/patch-WebCore_platform_network_soup_DNSSoup_cpp 14 Feb 2011 09:31:47 - @@ -0,0 +1,16 @@ +$OpenBSD$ + +don't try and resolve DNS for . for every single mouse movement +in a browser window. from https://bugs.webkit.org/show_bug.cgi?id=38263 + +--- WebCore/platform/network/soup/DNSSoup.cpp.orig Sun Feb 13 17:45:39 2011 WebCore/platform/network/soup/DNSSoup.cpp Sun Feb 13 17:46:05 2011 +@@ -36,6 +36,8 @@ namespace WebCore { + void prefetchDNS(const String hostname) + { + #ifdef HAVE_LIBSOUP_2_29_90 ++if (hostname.isEmpty()) ++return; + String uri = http://+hostname; + GOwnPtrSoupURI soupURI(soup_uri_new(uri.utf8().data())); + // We may get invalid hostnames, so NULL-check here.
Re: evil webkit bug, loads of junk DNS requests
Kill it! I am often stuck behind slow internet connections Chris Bennett
Re: evil webkit bug, loads of junk DNS requests
On Mon, 14 Feb 2011 15:10:52 +0100 Robert Nagy wrote: I know, I've read the discussion and I am still saying: do not disable it by default. I always disable prefetching where possible in all browsers. I only want to connect to and download stuff that I click on. Is there a decent example where it's beneficial or might it break something?
Re: evil webkit bug, loads of junk DNS requests
On Mon, Feb 14, 2011 at 1:37 AM, Stuart Henderson s...@spacehopper.org wrote: As mentioned in http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=578019 browsers using ports/www/webkit make a dns request for . each time the mouse position changes. brilliant software is brilliant.
Re: evil webkit bug, loads of junk DNS requests
On Mon, Feb 14, 2011 at 08:07:57AM -0600, Marco Peereboom wrote: I think you don't understand what happens behind the scenes. Every domain you enter gets prefixed and tested against the domain the browser thinks you are in. you type moo on you a.b.com domain and you'll see fantastically smart dns lookups such as: moo.com.a.b.com, moo.net.a.b.com etc. Easily 10 network names that will be wrong. The prefetcher does the opposite of its promise, it slows down the entire browser. By looking up all kinds of crap a user won't click. A good example is news.google.com which does about 350 DNS lookups. All that to save 1 that the user actually clicks. They traded one perceived performance issue for a real one. There is more insanity but these are just examples. That discussion doesn't belong here, bring it to webkit bugzilla/lists if you want that fixed in a sane way. Landry