Added panic message to failures.
Added the panic message to failures. This is a step towards #10334, since it gives us access to the panic error message when we fire a `mozbrowsererror` event. The remaining steps are also to record the backtrace, and to report the failure in the event.
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10587)
<!-- Reviewable:end -->
In rust-url 1.0 the `Url` struct is going to have private fields, and there
is no way to to create an aribitrary one without going through the parser.
The plugin never had a clear demonstrated performance benefit,
it was made mostly because it was possible and relatively easy at the time.
Without this `./mach run -b -- -M` would start fine but navigating to a site would fail with:
`thread 'ScriptThread PipelineId { namespace_id: PipelineNamespaceId(0), index: PipelineIndex(0) }' panicked at 'assertion failed: mozbrowser_enabled()', /Users/ulf/Documents/Code/servo/components/script/dom/htmliframeelement.rs:163`
The main issue was resources_dir_path. Every time it was called it would start from the executable's path and walk up the hierarchy to find a directory named "resources". The sandbox was granted permission to read from the found resources dir, but after the sandbox had been activated resources_dir_path would again start from the executable's path and try to find the resources dir. It would then fail with "Operation not permitted" when trying to canonicalize the path because it didn't have permissions to read metadata under ./target.
To fix this the resources dir path is now cached between resources_dir_path calls.
This allows keeping the VertexAttrib* calls asynchronous.
Another option would be to do the validation in the apply() function,
but that'd require us passing an unnecessary channel around and add
extra synchronization.
The counterpart of this is that it has to be updated when the context
changes, but that's less problem.
compositing: Send only one mouse move event if WebRender is in use.
In WebRender mode, we were sending two mouse move events: one with the
proper coordinates and one with the wrong coordinates, because of
incorrect fall-through. The script task would usually (but not always,
depending on timing) ignore the first event in favor of the second
event, resulting in incorrect mouse move event coordinates in most
cases.
Closes servo/webrender#238.
Closes#10298.
r? @metajack
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10321)
<!-- Reviewable:end -->
Removed panicking when frame or pipeline lookup fails.
Removed the methods `pipeline(id)`, `pipeline_mut(id)`, `frame(id)` and `frame_mut(id)` from constellation, which panicked when the table lookup failed.
The panics were causing race conditions, e.g. visiting google.com and resizing the page would cause a panic, most likely due to an iframe being added and removed, with the `DOMLoad` event arriving after the iframe had been removed, causing a panic.
This patch fixes#10017 and #8769 (although in non-webrender builds there's now a different panic, see https://github.com/servo/servo/issues/10017#issuecomment-198160200).
There are a few `TODO` items in the initial commit, for cases where it's not completely obvious what to do in the case of failure.
<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="35" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/servo/10082)
<!-- Reviewable:end -->
In WebRender mode, we were sending two mouse move events: one with the
proper coordinates and one with the wrong coordinates, because of
incorrect fall-through. The script task would usually (but not always,
depending on timing) ignore the first event in favor of the second
event, resulting in incorrect mouse move event coordinates in most
cases.
Closes servo/webrender#238.
Closes#10298.
This changes headless operation to strictly be a runtime option, rather
than a compile-time one. Note that the old headless version still relied
on a display server to support WebGL, while it now requires one all the
time.
Fixes#8573
has loaded.
This avoids a flash of unstyled content, which looks especially bad in
browser.html since unstyled content is white and browser.html has a
transparent background.
Closes#9996.
