The history is now recorded per frame, but needs to be exposed in a followup PR.
Also fixes a race condition that occurs loading iframes under heavy CPU load.
This ensures that iframes never do a reflow / layout until they have a valid
window size set from their parent frame.
If the root layer has a zero size rectangle, it will never be
painted, so assume the pipeline is idle in that case.
Re-enable two reftests that were disabled due to this issue.
@jdm This initial version has a few outstanding issues that I wanted to invite input on. Specifically:
1. I had some difficulty finding a home for the `StorageType` enum. Structs defined outside of the `script` module don't seem to be able to use the `#[jstraceable]` annotation and the `net` module (where `StorageTask` lives) doesn't have access to `script`. Per Simon Sapin's suggestion, I worked around this temporarily by creating a `TraceableStorageType` stand-in struct that was traceable and which could be translated into a regular `StorageType` when being sent to the `StorageTask`. Unsure of the best way to resolve this hack. Thoughts?
2. Apart from the `Storage` constructor used in `Window::SessionStorage` and the new `Window::LocalStorage`, there's also a method called `Storage::Constructor`. I'm unclear on what (if anything) will actually invoke this, so I'm not sure which variant of `StorageType` to use here. I've temporarily created an `Unknown` variant of `StorageType` as a placeholder.
3. I discovered that the web platform tests directory's localStorage tests. Many of them now pass despite the configured expectation that they fail. However, several do not pass. Is there a good way for me to add debug logging or otherwise get a sense of which assertion failed / what went wrong?
Thanks for your continued help!
Cycles were being created in the flow tree since absolutely positioned
descendants had pointers to their containing blocks. This adds
WeakFlowRef (based on Weak<T>) and makes these backpointers weak
references. This also harmonizes our custom Arc<T>, FlowRef, to be
consistent with the upstream implementation.
Fixes#4915.
Cycles were being created in the flow tree since absolutely positioned
descendants had pointers to their containing blocks. This adds
WeakFlowRef (based on Weak<T>) and makes these backpointers weak
references. This also harmonizes our custom Arc<T>, FlowRef, to be
consistent with the upstream implementation.
Fixes#4915.
`./mach test tests/wpt/web-platform-tests/html/browsers/history/the-location-interface/security_location_0.sub.htm` is still failing with the same message as reported in #3219.
CGDataProviderCreateWithData just wraps the underlying buffer. The
underlying buffer needs to be kept around until the data provider is
freed. Adding the buffer to the FontTemplateData struct ensures it
sticks around.
`./mach test-wpt` will fail in non-obvious ways unless all wpt submodules have recursively been checked out first. This ensure they have been when running the command in a checkout of Servo that hasn't been bootstrapped yet.
CGDataProviderCreateWithData just wraps the underlying buffer. The
underlying buffer needs to be kept around until the data provider is
freed. Adding the buffer to the FontTemplateData struct ensures it
sticks around.
SSL is broken-ish (eg tw.yahoo.com, html.spec.whatwg.org don't work since we don't verify SAN properly), this flag can let devs bypass the protection for testing purposes.
Looking for some review, due to a couple of issues:
* The vmin/vmax tests fail due to exit status being 1 instead of 0. However, when I manually check them via `./mach run tests/ref/viewport_percentage_vmax_vmin{,a,b,c}.html`, the pages render correctly
* Resizing is incredibly flaky. I've checked (via logging), and the actual calculation of the computed length is correct for a given viewport size. However, the box dimensions end up all over the place. I've attached a screenshot to demonstrate (the box is supposed to cover the top-left quarter of the window).

Fixes#5165.
When a viewport is resized, the computed values for a style containing viewport percentage length units become stale. However, there's no way for those styles to be invalidated after a resize. As a solution, this commit invalidates the computed values cache after a resize has occurred, which is probably over-kill.
A better solution would probably be to track under what conditions computed values remain valid, and invalidate them as indicated.
Fixes#5158
Part of this involved switching to OsStrings for process arguments. Those now interface awkwardly with some remaining use old_io, but that will only be needed until rustc is newer.
Table layout code has been refactored to push the spacing down to
rowgroups and rows; this will aid the implementation of
`border-collapse` as well.
r? @SimonSapin
`cellspacing` attribute per HTML5 § 14.3.9.
Table layout code has been refactored to push the spacing down to
rowgroups and rows; this will aid the implementation of
`border-collapse` as well.
This commit also fixes two nasty issues in table layout:
* In fixed layout, extra space would not be divided among columns that
had auto width but had nonzero minimum width.
* In automatic layout, extra space would be distributed to constrained
columns as well even if unconstrained columns with percentage equal to
zero were present.
This is GIF specific. It's also done when the image is PNG but PNG is
handled separately with the PNG crate, whereas GIFs are handled by the
stb-image crate and the distinction between alpha and non-alpha-supporting
images was missing.
Fix for issue #2108. That issue mentions an `unwrap_object` which doesn't seem to exist so I renamed `unwrap` to `native_from_reflector` and `unwrap_jsmanaged` to `native_from_reflector_jsmanaged`. The latter is a bit unweildy - maybe a shorter name might be better?