User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9b4pre) Gecko/2008021204 Minefield/3.0b4pre Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9b4pre) Gecko/2008021204 Minefield/3.0b4pre ID:2008021204 In FF3 the top border of an element is rendered with a smaller height if: * the DIV is empty and * border-top-style is not (dotted or dashed) and border-bottom-style is (dotted or dashed) and * border-top-width is greater than border-bottom-width. See attachment. Reproducible: Always

See attachment 303034 [details] for bug 379303

This does not reproduce on FF2.

I see this on linux, too.

How about steps to reproduce?

See attachment 303034 [details] of the bug 379303 You can change the top and bottom border styles and widths to see the bug in action. You can trigger the bug with any combination of the border styles marked with a red background. In FF3: * create an empty DIV and * set border-top-style other than (dotted or dashed) and border-bottom-style to (dotted or dashed) and * set border-top-width to a value s greater than the border-bottom-width Actual results: the top border is rendered with a smaller width than specified. Use the following css for any empty DIV (or other tag): { background-color: red; /*so we can see the background*/ border-top: 30px solid lime; border-bottom: 10px dashed blue; } Results: - the top border should be 30px all but instead it appears to be 20px - the red background can be seen through what should have been the lower 10px of the top border - the bottom border appears solid instead of dashed.

This may be same as bug 417543

This is almost certainly a regression from Vlad's border work. Shouldn't be hard to fix and I think we should fix it.

(In reply to comment #7) > Created an attachment (id=306423) [details] > an actual testcase > Could use attachment 303034 [details] of the bug 379303

It's much better to have a very simple testcase than a rather complex dynamic testcase.

That's true only if the testcase covers all the possible combinations, as described in the bug description. Your testcase covers only one. It's true, the developer may need to verify his code against only one case, but I really tried to make sure he get it right in all cases.

If you want to test a lot of cases, make a set of testcases, or one testcase that contains a lot of elements that test the different cases. Asking people to click on a lot of options to test everything is not helpful.

So the problem is this code here: http://mxr.mozilla.org/seamonkey/source/layout/base/nsCSSRendering.cpp#1121 The way rendering with dotted bits work is we first render the corners, and then we go back and fill in the sides. This is done to avoid weird discontinuities at the corners when using dotted/dashed borders, especially with rounded borders. The problem in this case is that there really aren't any corners. But this halfWidth/halfHeight stuff, I can't remember what that was protecting against. Removing the halfWidth/halfHeight clamping fixes the problem, and I don't see anything wrong with any of my testcases. Running reftests now.

Reftests say everything's ok, so let's get rid of that code.

So what happens now if you have a very small DIV with very large border-radius?

Nothing different; the radii are already clamped at http://mxr.mozilla.org/seamonkey/source/layout/base/nsCSSRendering.cpp#2259 .

checked in, along with the reftests.