construction.
The iteration was incorrect here. Although it accidentally worked
before, it will cause problems when we have incremental style
recalculation.
The `after_block_iteration` reftest will become interesting once we have
incremental style recalc.
Also fixes calculation of background-position percentages:
Rather than multiplying the container size by a percent and aligning the top
left of the image at the resulting width, we also need to subtract a
corresponding percent of the image size, per
http://dev.w3.org/csswg/css2/colors.html#propdef-background-position
"A value pair of '100% 100%' places the lower right corner of the image in the
lower right corner of the padding box. With a value pair of '14% 84%', the
point 14% across and 84% down the image is to be placed at the point 14%
across and 84% down the padding box."
… when parsing as an absolute URL (without a base) fails.
Previously, arguments were parsed as URLs with the current working directory
converted to an URL and used as the base URL. This mostly worked for
relative filenames, except that `?` and `#` had a special meaning
and needed to be percent-encoded.
Fix#3340.
This still needs a lot of work, but it covers the basic
cases and improves wikipedia while passing all existing tests.
Tweak reftest to deal with linux/travis black background.
Add iframe tests to their own subdirectory and add another test case
that used to trigger a fatal error. The new test case uses the
"allow-scripts" sandbox attribute to work around a script task failure
caused by the child frame sharing the same script task as the parent.
This fixes the visual artifacts seen at the top of wikipedia pages by:
- Setting clipping rect to avoid images going outside their bounds.
- Handling case of background-position being >= 100% such that wrapping is required.
However, the gradient is not currently visible on wikipedia. This relies on bug #1997 being fixed.
When calculating the preferred width for a block, accumulate
the left and right float widths of children separately, which
is then max'ed with the normal flow widths later on.
Ref bug #2554 - improves the layout of the top bar.
