Shared JS is confirmed working fine in Windows after a week of testing.
Non-Windows will have to wait unfortunately. The JS_SHARED_LIBRARY option is
kept for easy testing on those unsupported platforms.
The trace in Spidermonkey's old-configure is causing the MOZ_CONFIG_ICU function
to be called twice, which is not desirable as it can cause the MOZ_SHARED_ICU
check in build/autoconf/icu.m4 to fail and build an ICU data file even if we're
building ICU as a shared library.
It seems that if we're going to enable JS_SHARED_LIBRARY, we need ICU to be a
shared library too in order to prevent code duplication in mozjs.dll bloating
it to a whopping 21 MB.
This unfortunately means we have to fold icudata back to icu itself. But since
icudata is still separated from libxul (which is what really matters for #1451),
I think this is a fair compromise.
Building ICU as shared is offered as an option via --enable-shared-icu. But
technically you really need --enable-shared-icu if you're going to use
--enable-shared-js. Unfortunately I can't make configure exit with an error if
MOZ_SHARED_ICU is not defined when JS_SHARED_LIBRARY is enabled, because options
are controlled by different configure scripts (icu is based on bash, while js is
based on python), and I have no idea how to pass MOZ_SHARED_ICU to js's
moz.configure.
Tag #62
This is the one I don't understand how it fixed the bustage. Why is it needed
to remove JS_PUBLIC_API..? It would seem counter-intuitive, but it wasn't...
The unsafeSet method is based on Jon Coppeard's part 2 patch for Mozilla bug
1306008. This is needed because of the use of unsafeSet in rekey() of
MovableCellHasher<JS::Heap<T>>
Skia does some nasty things that break with DLL calls like having virtual dtors
without a body definition (and it's honestly surprising that it even builds to
begin with the way it's done, but apparently they use some linking symbol voodoo
to work around that in dependent classes).
Not wanting to have to pull Google code apart to try and fix that for something
we're only using for canvas anyway, it's better to just let Skia reside in
libxul instead.
We need to use StyleChildrenIterator which iterates over them, instead.
This is a bit of a hack but we can't always have our cake.
We need to be a bit careful though, since ::before and ::after are owned
by their own frame, and thus could be unbound from the tree or even dead
after removing the frame.
This basically ports bugs 1435566 and 1438467
Currently, we dispatch keypress event when Enter is pressed without
modifiers or only with the Shift key (line break).
However, other browsers dispatch keypress events for Ctrl + Enter also
even if it doesn't cause any text input.
So, we should fire keypress events for Ctrl + Enter, even in strict
keypress dispatching mode. Note that with other modifiers, it depends on
the browser and/or platform and we can't dispatch the event for
consistent behavior.
This means web developers shouldn't rely one keypress events to catch
Alt + Enter, Meta + Enter and two or more modifiers + Enter.
Based on BZ 1438133
Resolves#2161
Pseudo-classes with a forgiving selector list argument are allowed to follow a pseudo-element, but must treat any selector that is not of the same type as invalid. It doesn't make any sense, but that's the behavior of other tainted browsers.
- Check against all selector parts and not the leftmost selector only for ::slotted()
- Walk rules for ::slotted() regardless if the shadow root is opened/closed
- Ensure that ::slotted() rules are walked in the right order
- Fix ::slotted inheritance from topmost shadow root
- Block slot elements from being matched by ::slotted
- Ensure ::slotted() is serialized as a pseudo-element
- Add pref to control whether the pseudo-class is enabled
Previously, we'd match :host despite the element having a different shadow root from the one that we currently have. Also, there's a test where :host should be blocked from matching if it is a descendant of an explicit universal selector.
(x) = function() {..} is an extremely sparsely documented way to prevent function name inference.
Modeled after how BytecodeEmitter::emitInitializer does it.
Attn: this plain patch port still contains issues caused by differences in the
ParseNode tree that will be fixed in the following commits.
Based-on: m-c 1466000/{7-10}
This removed a bit of reduncancy and makes stack building easier to follow.
- Move call context handling from BytecodeEmitter::emitGetNameAtLocation to BytecodeEmitter::emitCalleeAndThis
- Add stack comments for prop and elem related methods in BytecodeEmitter
- Remove emitAtopOp with NameNode*
- Pass JSOp to BytecodeEmitter::emitAssignment (modified)
- Add BytecodeEmitter::emitGetFunctionThis variant with offset
Based-on: m-c 1466000/{2-6}
