March 2013 - webkit-efl - lists.webkit.org

minimal build as a guard of non-default build options
by Ed Bartosh 30 Mar '13

30 Mar '13

Hi, Would it make sense to build webkit-efl with as many as possible build options turned off as a method of guarding of non-default build options? Of course it's not an absolute guarantee that some combination of options fail a build, but I believe it's much better than buildiing with default options only. So far I come up with this method: # grep option Tools/Scripts/webkitperl/FeatureList.pm | sed -n 's/.*option => "\([^"]\+\)".*/--no-\1/p' |tr '\n' ' ' --no-3d-rendering --no-accelerated-2d-canvas --no-battery-status --no-blob --no-canvas-path --no-canvas-proxy --no-channel-messaging --no-csp-next --no-css-device-adaptation --no-css-exclusions --no-css-filters --no-css3-conditional-rules --no-css3-text --no-css3-text-line-break --no-css-box-decoration-break --no-css-image-orientation --no-css-image-resolution --no-css-image-set --no-css-regions --no-css-shaders --no-css-sticky-position --no-css-compositing --no-css-transforms-animations-unprefixed --no-css-variables --no-custom-scheme-handler --no-datalist --no-data-transfer-items --no-details --no-device-orientation --no-dialog --no-directory-upload --no-dom4-events-constructor --no-download-attribute --no-file-system --no-filters --no-font-load-events --no-ftpdir --no-fullscreen-api --no-gamepad --no-geolocation --no-high-dpi-canvas --no-icon-database --no-iframe-seamless --no-indexed-database --no-input-speech --no-input-type-color --no-input-type-date --no-input-type-datetime --no-input-type-datetimelocal --no-input-type-month --no-input-type-time --no-input-type-week --no-inspector --no-javascript-debugger --no-legacy-notifications --no-legacy-vendor-prefixes --no-legacy-web-audio --no-link-prefetch --no-link-prerender --no-mathml --no-media-capture --no-media-source --no-media-statistics --no-media-stream --no-meter-tag --no-mhtml --no-microdata --no-mouse-cursor-scale --no-navigator-content-utils --no-netscape-plugin-api --no-network-info --no-nosniff --no-notifications --no-orientation-events --no-page-visibility-api --no-performance-timeline --no-progress-tag --no-proximity-events --no-quota --no-resolution-media-query --no-resource-timing --no-request-animation-frame --no-scripted-speech --no-shadow-dom --no-shared-workers --no-sql-database --no-style-scoped --no-suid-linux-sandbox --no-svg --no-svg-dom-objc-bindings --no-svg-fonts --no-system-malloc --no-template-element --no-text-autosizing --no-tiled-backing-store --no-threaded-html-parser --no-touch-events --no-touch-slider --no-touch-icon-loading --no-user-timing --no-vibration --no-video --no-video-track --no-webgl --no-web-audio --no-web-sockets --no-web-timing --no-workers --no-xhr-timeout --no-xslt Do you think this build would make sense? Would it improve the current situation? Any other ideas for the list of options that would make more sense than above list? -- BR, Ed

2 2

Fixing --no-tiled-backing-store build
by Vyacheslav Ostapenko 29 Mar '13

29 Mar '13

Hi Ed, Great work! The only thing, that "tiled-backing-store" is a misused flag in this case and, IMHO, "accelerated compositing" switch should be used in this case. This flag was used because of "historical" reasons and it might be a good time to fix it also . Also, --no-tiled-backing-store works as a workaround because it is switches off texture mapper based accelerated compositing drawing which uses opengl by default and because of this needs GLX . But texture mapper has fallback mode when it uses ImageBuffer as a "texture" . To enable this creating of EvasGLContext should be disabled. I remember Yael was fixing it. IMHO, it would be better workaround for the Xvfb GLX regression. > Another point is how to avoid this kind of breakages in the future. Have "minimal" build bot like Qt has? Slava On Fri, Mar 29, 2013 at 11:20 AM, Ed Bartosh <bartosh(a)gmail.com> wrote: > Hi *Kalyan,* > * > * > *Thank you for your suggestions. I'll split the bug. It looks like "one > patch for one bug" rule is assumed in the current workflow as tools like > webkit-patch automatically obsolete previous patches in the bug when new > one is uploaded.* > * > * > *BTW, another newbie question: How can I assign myself to the bug? Or > should I ask someone else with sufficient rights to do that? > * > > Regarding support --no-tiled-backing-store I also thought that one of the > possible ways to go is to drop support of it completely. However, in > current situation it means that developers using latest versions of major > Linux distros are blocked by this issue. > > Another point is how to avoid this kind of breakages in the future. As > Laszlo mentioned in the bug without being continuously tested builds with > non-default options will tend to regress to build failures. From my point > of view it boils down to the amount of build resources we have for testing. > It's not a big deal to run build with --no-tiled-backing-store or other > option for every commit, but this is very resource-consuming. Alternatively > it might make sense to get rid of as many options as we can if we can't > ensure that non-default builds don't fail. This will at least make code > much more readable as a lot of #ifs will be cleaned up. > > Regards, > Ed > > > 2013/3/29 kalyan kondapally <kondapallykalyan(a)gmail.com> > >> HI Ed, >> >> Thanks for the effort. >> >> Reg Adding checks: >> >> With X fix in, it boils down to whether we want to support >> --no-tiled-backing-store or not. >> AC checks make sense to me (unless we decide not to support non Ac >> composition mode on UI process side.). Coordinated graphics checks seem >> un-necessary as most of the classes are WebKit2 related. Someone more >> familiar with the code might think otherwise :) >> >> Reg Reviewers: >> You already have Laszlo,* Gyuyoung Kim participating in the discussion. >> Kenneth, MIkhail are already in cc. >> >> * >> *As you have already split the patches to smaller ones, you might want >> to create new bugs to each one and make the current one as blocked by >> others. This would at least help to focus the discussion for that >> particular check. (i.e coordinated graphics, tiledbackinstore and AC ) >> >> * >> *br, >> Kalyan >> * >> >> On Thu, Mar 28, 2013 at 10:13 PM, Ed Bartosh <bartosh(a)gmail.com> wrote: >> >>> Hi, >>> >>> Thanks for you comments guys. I really appreciate that. >>> >>> I'd like to point out that even if xserver doesn't have regression my >>> fixes still make sense as they wrap code in #ifs depending on the features >>> like 3D_GRAPHICS, ACCELERATED_COMPOSITING, etc. So, they have some value by >>> themselves, not just because they fix --no-tiled-backing-store build. >>> >>> Considering Gyuyoung's comment in the bug about removing coorditaned >>> graphics I'm going to split my patch to several patches just to make >>> them accepted easier. >>> >>> Sorry for the newby question. Any suggestions whom to ask for review? >>> I'm new to this project and it's not clear for me what to do to make my >>> changes landed in the tree faster. >>> >>> Thank you, >>> Ed >>> >>> 2013/3/28 Tiago Vignatti <tiago.vignatti(a)linux.intel.com> >>> >>>> Xorg 1.13.2 has the proper fix in place. In the ideal world, distros >>>> would update for the latest minor version of Xorg and then you would just >>>> fetch it in your machine... >>>> >>>> Tiago >>>> >>>> >>>> On 03/28/2013 03:36 AM, Gyuyoung Kim wrote: >>>> >>>>> Hi, >>>>> >>>>> As I commented on the bug, I think it would be best X server 1.13.0 >>>>> supports WebKit EFL layout test. However, if it will not support EFL >>>>> layout test in the near future, it looks we may need to support >>>>> --no-tiled-backing-store option again. Because I also heard that some >>>>> WebKit EFL engineers suffered from this issue. But, I still think best >>>>> is to support X server 1.13.0 on EFL. >>>>> >>>>> Any other opinions? >>>>> >>>>> Gyuyoung. >>>>> >>>>> On Thu, Mar 28, 2013 at 9:06 AM, Ed Bartosh <bartosh(a)gmail.com >>>>> <mailto:bartosh@gmail.com>> wrote: >>>>> >>>>> Hi, >>>>> >>>>> I need your suggestions. >>>>> >>>>> I'm trying to fix EFL build with disabled accelerated composition >>>>> (--no-tiled-backing-store build option). >>>>> I used this option because of X server 1.13.0 >>>>> <http://lists.webkit.org/**pipermail/webkit-efl/2012-** >>>>> November/000434.html<http://lists.webkit.org/pipermail/webkit-efl/2012-November/000434.html>> >>>>> issue, >>>>> >>>>> mentioned here: >>>>> http://trac.webkit.org/wiki/**WebKitEFLLayoutTest#**Knownissues<http://trac.webkit.org/wiki/WebKitEFLLayoutTest#Knownissues> >>>>> I'm trying to build on Ubuntu 12.10, which has X server 1.13.0 with >>>>> this regression. >>>>> >>>>> It was not that easy to fix this as I suspected as it's required a >>>>> lot of changes, but it was fun and I learned a bit of WebKit code. >>>>> You can see some details in this bug: >>>>> https://bugs.webkit.org/show_**bug.cgi?id=11315<https://bugs.webkit.org/show_bug.cgi?id=11315> >>>>> <https://bugs.webkit.org/show_**bug.cgi?id=113153<https://bugs.webkit.org/show_bug.cgi?id=113153> >>>>> >3 >>>>> >>>>> >>>>> Can you please suggest what should I do next? Should I split this >>>>> patch to several patches or leave it as it is? Whom to ask for >>>>> review? >>>>> >>>>> Considering the amount of changes was it at all feasible to fix >>>>> this >>>>> build or it would be better to simply get rid of >>>>> --no-tiled-backing-store option? >>>>> >>>>> Thank you, >>>>> -- >>>>> BR, >>>>> Ed >>>>> >>>>> ______________________________**_________________ >>>>> webkit-efl mailing list >>>>> webkit-efl(a)lists.webkit.org <mailto:webkit-efl@lists.**webkit.org<webkit-efl(a)lists.webkit.org> >>>>> > >>>>> https://lists.webkit.org/**mailman/listinfo/webkit-efl<https://lists.webkit.org/mailman/listinfo/webkit-efl> >>>>> >>>>> >>>>> -- >>> BR, >>> Ed >>> _______________________________________________ >>> webkit-efl mailing list >>> webkit-efl(a)lists.webkit.org >>> https://lists.webkit.org/mailman/listinfo/webkit-efl >>> >>> >> > > > -- > BR, > Ed > > _______________________________________________ > webkit-efl mailing list > webkit-efl(a)lists.webkit.org > https://lists.webkit.org/mailman/listinfo/webkit-efl > >

2 1

Fixing --no-tiled-backing-store build
by Ed Bartosh 29 Mar '13

29 Mar '13

Hi, Thanks for you comments guys. I really appreciate that. I'd like to point out that even if xserver doesn't have regression my fixes still make sense as they wrap code in #ifs depending on the features like 3D_GRAPHICS, ACCELERATED_COMPOSITING, etc. So, they have some value by themselves, not just because they fix --no-tiled-backing-store build. Considering Gyuyoung's comment in the bug about removing coorditaned graphics I'm going to split my patch to several patches just to make them accepted easier. Sorry for the newby question. Any suggestions whom to ask for review? I'm new to this project and it's not clear for me what to do to make my changes landed in the tree faster. Thank you, Ed 2013/3/28 Tiago Vignatti <tiago.vignatti(a)linux.intel.com> > Xorg 1.13.2 has the proper fix in place. In the ideal world, distros would > update for the latest minor version of Xorg and then you would just fetch > it in your machine... > > Tiago > > > On 03/28/2013 03:36 AM, Gyuyoung Kim wrote: > >> Hi, >> >> As I commented on the bug, I think it would be best X server 1.13.0 >> supports WebKit EFL layout test. However, if it will not support EFL >> layout test in the near future, it looks we may need to support >> --no-tiled-backing-store option again. Because I also heard that some >> WebKit EFL engineers suffered from this issue. But, I still think best >> is to support X server 1.13.0 on EFL. >> >> Any other opinions? >> >> Gyuyoung. >> >> On Thu, Mar 28, 2013 at 9:06 AM, Ed Bartosh <bartosh(a)gmail.com >> <mailto:bartosh@gmail.com>> wrote: >> >> Hi, >> >> I need your suggestions. >> >> I'm trying to fix EFL build with disabled accelerated composition >> (--no-tiled-backing-store build option). >> I used this option because of X server 1.13.0 >> <http://lists.webkit.org/**pipermail/webkit-efl/2012-** >> November/000434.html<http://lists.webkit.org/pipermail/webkit-efl/2012-November/000434.html>> >> issue, >> >> mentioned here: >> http://trac.webkit.org/wiki/**WebKitEFLLayoutTest#**Knownissues<http://trac.webkit.org/wiki/WebKitEFLLayoutTest#Knownissues> >> I'm trying to build on Ubuntu 12.10, which has X server 1.13.0 with >> this regression. >> >> It was not that easy to fix this as I suspected as it's required a >> lot of changes, but it was fun and I learned a bit of WebKit code. >> You can see some details in this bug: >> https://bugs.webkit.org/show_**bug.cgi?id=11315<https://bugs.webkit.org/show_bug.cgi?id=11315> >> <https://bugs.webkit.org/show_**bug.cgi?id=113153<https://bugs.webkit.org/show_bug.cgi?id=113153> >> >3 >> >> >> Can you please suggest what should I do next? Should I split this >> patch to several patches or leave it as it is? Whom to ask for review? >> >> Considering the amount of changes was it at all feasible to fix this >> build or it would be better to simply get rid of >> --no-tiled-backing-store option? >> >> Thank you, >> -- >> BR, >> Ed >> >> ______________________________**_________________ >> webkit-efl mailing list >> webkit-efl(a)lists.webkit.org <mailto:webkit-efl@lists.**webkit.org<webkit-efl(a)lists.webkit.org> >> > >> https://lists.webkit.org/**mailman/listinfo/webkit-efl<https://lists.webkit.org/mailman/listinfo/webkit-efl> >> >> >> -- BR, Ed

2 4

Fixing --no-tiled-backing-store build
by Ed Bartosh 28 Mar '13

28 Mar '13

Hi, I need your suggestions. I'm trying to fix EFL build with disabled accelerated composition (--no-tiled-backing-store build option). I used this option because of X server 1.13.0<http://lists.webkit.org/pipermail/webkit-efl/2012-November/000434.html> issue, mentioned here: http://trac.webkit.org/wiki/WebKitEFLLayoutTest#Knownissues I'm trying to build on Ubuntu 12.10, which has X server 1.13.0 with this regression. It was not that easy to fix this as I suspected as it's required a lot of changes, but it was fun and I learned a bit of WebKit code. You can see some details in this bug: https://bugs.webkit.org/show_bug.cgi?id=11315<https://bugs.webkit.org/show_bug.cgi?id=113153> 3 Can you please suggest what should I do next? Should I split this patch to several patches or leave it as it is? Whom to ask for review? Considering the amount of changes was it at all feasible to fix this build or it would be better to simply get rid of --no-tiled-backing-store option? Thank you, -- BR, Ed

3 2

About usage of state caching variables to reduce redundant OpenGL state changes
by Bruno Abinader 21 Mar '13

21 Mar '13

Hi all, I've been profiling OpenGL calls from the UIProcess and found out we are doing quite some redundant state changes and getter/setter calls. This might affect performance on platforms where the OpenGL client communicates directly with server, without using a GPU proxy to detect such redundancies (like Chromium does). Based on this, I've implemented a GraphicsContext3DState class, which would cache and thus, reduce the amount of OpenGL calls during runtime. Details are covered in bug 110883 [1] and the state cache is disabled by default, but it can be enabled on a per-platform basis as necessary. Please let me know your opinion on this matter, and I could also clarify some missing information if necessary. Best regards, Links: [1] https://bugs.webkit.org/show_bug.cgi?id=110883 -- Bruno de Oliveira Abinader Software Engineer @ basysKom GmbH WebKit committer / Nokia Certified Qt Specialist

2 1

replacing libsoup
by Michael Blumenkrantz 15 Mar '13

15 Mar '13

Hi, I am the author of Azy, a networking toolkit, and I am currently working on expanding it in a number of different areas. One of my targets is to make it work as a replacement for libsoup, as discussed on the e-devel mailing list some months ago. I investigated the webkit source tree a bit, but it's a huge mess to someone with no background in webkit, and my time away from maintaining E18 is limited; I am hoping that someone can outline the exact features necessary for replacing libsoup so that I can ensure they are implemented. I think it's most of the way there, but probably missing a few things like advanced authentication functionality. Thanks

1 0

Changing the test expectations lookup sequence
by Thiago Marcos P. Santos 07 Mar '13

07 Mar '13

Hi EWebkitters, Currently on the EFL WK2 port, the lookup sequence for test expectations is "efl-wk2 -> wk2 -> efl -> generic". The "efl" part of this sequence is shared between EFL WebKit1 and EFL WebKit2. I'm currently facing some challenges when gardening because of: 1. Currently we have some tests skipped (~90) on the "wk2" expectations that have different expectations on "efl". It gives you the impression that the test is running (because is marked as [ Failure ] when in fact is being skipped because "wk2" has precedence. The flakiness dashboard also gets confused about it and gives a false positive. 2. Removing or adding a file to the shared expectation requires testing on both ports, which IMO is a heavy burden. 3. I'm trying to remove from the flaky list some tests that are now considered stable (no failures on the past 100 runs). Because of the shared expectations, I will have to analyse the results of one extra bot, even though I have no intentions of doing gardening for WebKit1. I would like to suggest changing the lookup sequence to "efl-wk2 -> wk2 -> generic", so we can make EFL WK2 port completely independent. Cheers,

6 5

Render issue
by Daniel Robinson 04 Mar '13

04 Mar '13

I'm running WebKit EFL on a CentOS 6.3 machine. When I run the MiniBrowser, I get an all white page. I'm running on a virtual machine and using Xvfb. I am trying to load a well known site, google.com. When I do this with Firefox, I can use xwd to grab the root plane of the display and Firefox is rendering the Google page. When I run the MiniBrowser, I know that it is accessing www.google.com because I see the network traffic with tshark. I can paint an arbitrary box in the display with Xlib calls. Using xwd, I can retrieve that. Two thoughts: 1) Webkit is waiting for input of some kind. Perhaps, but when I set the DISPLAY variable to "localhost:14.0" or whatever, I see the window on my workstation and by trying a number of things happen. If I press the mouse button at various points in the window, the URL in the URL window changes to reflect new pages being loaded. 2) Some permission requirement is not being met. I have built the MiniBrowser in debug mode and when it quits, it lists the objects it has rendered. Any help would be appreciated. --dan robinson -- ----------------------- daniel g. robinson www.bornlivelovedie.com bornlivelovedie.tumblr.com @daguro

1 0