Re: [webkit-dev] Committing EFL baselines
On 10/17/2011 04:13 PM, Raphael Kubo da Costa wrote: Anyway, are there any options besides simply svn commit'ting all these files? I'm thinking in uploading the text baselines (which are small) first. When these are in place, we could work together to diminish the amount of pixel baselines by adopting things like ScrollbarMock and such. Sounds like a plan? Cheers, Leandro ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
Leandro Pereira lean...@profusion.mobi writes: What's the preferred way to commit a ton of baselines for a port that currently has none? On our internal EFL tree, there is about 125MB of baselines for both pixel and text tests. Unfortunately we were unable to share our baselines with similar ports, due to slight differences in results. This is pretty much unreviewable, so I pretend to commit this directly, in batches (one commit per toplevel directory in LayoutTests/platform/efl) in the next weeks. Any objections or suggestions? Bringing the thread back from the dead. We're almost there with upstreaming our DumpRenderTree code, so the time to upload the baselines is coming. Right now, our LayoutTests/platform/efl is about 126M in size. One nice thing we've managed to do since the last time we talked about this is that we've started using smfr's mock scrollbar theme commits to reduce the differences in scrollbar size and appearance, so at least this should be common with other ports. Anyway, are there any options besides simply svn commit'ting all these files? -- Raphael Kubo da Costa ProFUSION embedded systems http://profusion.mobi ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
[webkit-dev] Committing EFL baselines
Hi. What's the preferred way to commit a ton of baselines for a port that currently has none? On our internal EFL tree, there is about 125MB of baselines for both pixel and text tests. Unfortunately we were unable to share our baselines with similar ports, due to slight differences in results. This is pretty much unreviewable, so I pretend to commit this directly, in batches (one commit per toplevel directory in LayoutTests/platform/efl) in the next weeks. Any objections or suggestions? Cheers, Leandro ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On Mon, Sep 12, 2011 at 1:32 PM, Leandro Pereira lean...@profusion.mobiwrote: Hi. What's the preferred way to commit a ton of baselines for a port that currently has none? On our internal EFL tree, there is about 125MB of baselines for both pixel and text tests. Unfortunately we were unable to share our baselines with similar ports, due to slight differences in results. What are differences you're seeing? This is pretty much unreviewable, so I pretend to commit this directly, in batches (one commit per toplevel directory in LayoutTests/platform/efl) in the next weeks. Any objections or suggestions? It'll be nice if we could spend some time analyzing the differences between EFL and other ports to minimize the size of patch first. - Ryosuke ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On Mon, Sep 12, 2011 at 1:48 PM, Peter Kasting pkast...@chromium.orgwrote: On Mon, Sep 12, 2011 at 1:36 PM, Ryosuke Niwa rn...@webkit.org wrote: This is pretty much unreviewable, so I pretend to commit this directly, in batches (one commit per toplevel directory in LayoutTests/platform/efl) in the next weeks. Any objections or suggestions? It'll be nice if we could spend some time analyzing the differences between EFL and other ports to minimize the size of patch first. In particular, if we have pixel tests that don't need to be pixel tests at all, or font rendering differences due to explanatory text that could be moved to an HTML comment inside the test itself, we can obviate the need for port-specific baselines in many of those cases (and eliminate more baselines from ports already in the tree, and reduce the burden on other ports that want to run pixel tests, and reduce the maintenance cost of changing the tests). Are y'all suggesting that efl port should do these items (converting pixel tests to non-pixel tests) before committing their baselines? dave PS fwiw, at one point, I did some work to figure out which tests were changing most often to see where people would get the most return on their investment and to my memory many tests with a high churn rate were svg (which seem like they would be harder to convert to a non-pixel test format). PK ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On Mon, Sep 12, 2011 at 1:56 PM, David Levin le...@chromium.org wrote: On Mon, Sep 12, 2011 at 1:48 PM, Peter Kasting pkast...@chromium.orgwrote: On Mon, Sep 12, 2011 at 1:36 PM, Ryosuke Niwa rn...@webkit.org wrote: This is pretty much unreviewable, so I pretend to commit this directly, in batches (one commit per toplevel directory in LayoutTests/platform/efl) in the next weeks. Any objections or suggestions? It'll be nice if we could spend some time analyzing the differences between EFL and other ports to minimize the size of patch first. In particular, if we have pixel tests that don't need to be pixel tests at all, or font rendering differences due to explanatory text that could be moved to an HTML comment inside the test itself, we can obviate the need for port-specific baselines in many of those cases (and eliminate more baselines from ports already in the tree, and reduce the burden on other ports that want to run pixel tests, and reduce the maintenance cost of changing the tests). Are y'all suggesting that efl port should do these items (converting pixel tests to non-pixel tests) before committing their baselines? No. But in general, we try to match DRT's behavior to that of Mac port to avoid adding redundant baselines so EFL port should do the same. - Ryosuke ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On Mon, Sep 12, 2011 at 1:56 PM, David Levin le...@chromium.org wrote: On Mon, Sep 12, 2011 at 1:48 PM, Peter Kasting pkast...@chromium.orgwrote: In particular, if we have pixel tests that don't need to be pixel tests at all, or font rendering differences due to explanatory text that could be moved to an HTML comment inside the test itself, we can obviate the need for port-specific baselines in many of those cases (and eliminate more baselines from ports already in the tree, and reduce the burden on other ports that want to run pixel tests, and reduce the maintenance cost of changing the tests). Are y'all suggesting that efl port should do these items (converting pixel tests to non-pixel tests) before committing their baselines? I don't think it's fair to force the EFL folks to do all this. I do think all ports would benefit from it and all ports should be spending some effort on this kind of thing. Inasmuch as there is an opportunity here for doing this work to minimize the landing cost for the EFL baselines, I'm raising this so that those folks can keep the option in mind and make use of it opportunistically. PK ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On 09/12/2011 05:36 PM, Ryosuke Niwa wrote: On our internal EFL tree, there is about 125MB of baselines for both pixel and text tests. Unfortunately we were unable to share our baselines with similar ports, due to slight differences in results. What are differences you're seeing? Most differences are because the EFL port doesn't change the viewport size because of the scrollbar: it is drawn on top of the content on the EFL port by default. So there is a little more room horizontally, and then text results are slightly different from GTK+'s (which uses the same font backends, for instance). In those tests that only draws colored rectangles, the render tree is pretty much the same as any other port's. However, in these tests, the pixel results are rendered with colors slightly different from the expected. Almost no difference to the naked eye. I suspect either some rounding problems inside Cairo, or wrong color profile. Leandro ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
Re: [webkit-dev] Committing EFL baselines
On Mon, Sep 12, 2011 at 2:50 PM, Leandro Pereira lean...@profusion.mobiwrote: Most differences are because the EFL port doesn't change the viewport size because of the scrollbar: it is drawn on top of the content on the EFL port by default. So there is a little more room horizontally, and then text results are slightly different from GTK+'s (which uses the same font backends, for instance). In those tests that only draws colored rectangles, the render tree is pretty much the same as any other port's. However, in these tests, the pixel results are rendered with colors slightly different from the expected. Almost no difference to the naked eye. I suspect either some rounding problems inside Cairo, or wrong color profile. Chromium port has made a significant effort into matching scrollbar/color with Mac port to mitigate this issue. Maybe you can talk to one of our friendly contributors and get some help in reducing the number of baselines you need to check in. I don't intend to claim that you're obligated to do this but if you check in a lot of port-specific baselines, then you're very likely end up having to rebaseline hundreds of tests each day to catch up with new changes landed in trunk. This is a very tedious job and people from other ports spend a significant engineering time into this. - Ryosuke ___ webkit-dev mailing list webkit-dev@lists.webkit.org http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev