Some initialization handlings of ClientLayerManager exists in nsBaseWidget::CreateCompositor(). It is not good. Move them to ClientLayerManager::Initialize().
Differential Revision: https://phabricator.services.mozilla.com/D39476
--HG--
extra : moz-landing-system : lando
This requires replacing inclusions of it with inclusions of more specific prefs
files.
The exception is that StaticPrefsAll.h, which is equivalent to StaticPrefs.h,
and is used in `Codegen.py` because doing something smarter is tricky and
suitable for a follow-up. As a result, any change to StaticPrefList.yaml will
still trigger recompilation of all the generated DOM bindings files, but that's
still a big improvement over trigger recompilation of every file that uses
static prefs.
Most of the changes in this commit are very boring. The only changes that are
not boring are modules/libpref/*, Codegen.py, and ServoBindings.toml.
Differential Revision: https://phabricator.services.mozilla.com/D39138
--HG--
extra : moz-landing-system : lando
This is to ensure that multiple completions cannot be attempted on the same `GeckoResult`, resulting in crashes.
Differential Revision: https://phabricator.services.mozilla.com/D36929
--HG--
extra : moz-landing-system : lando
This is caused by a race condition when the compositor is detached. Because the actual detachment happens in a new thread, `detach` can complete and release the lock on `mLayerViewSupport`, and `RecvScreenPixels` can obtain the lock, before `mLayerViewSupport` is properly cleaned up. We therefore check to ensure that `lvs` is not null before calling a method on it.
Differential Revision: https://phabricator.services.mozilla.com/D36928
--HG--
extra : moz-landing-system : lando
For the URI handling case, we still want to parse the URI to look for any
malformation. OTOH, IShellDispatch2::ShellExecute does not accept PIDLs as
arguments, we we need an overload that converts the absolute PIDL back to a
path for the purposes of passing on to that interface.
Differential Revision: https://phabricator.services.mozilla.com/D38944
--HG--
extra : moz-landing-system : lando
This is just so that both the launcher process and other Gecko code can share
this method.
Differential Revision: https://phabricator.services.mozilla.com/D38943
--HG--
extra : moz-landing-system : lando
While there, transform the MOZ_ASSERTs into static_asserts, so that
discrepancies are caught at build time rather than runtime.
Differential Revision: https://phabricator.services.mozilla.com/D38165
--HG--
extra : moz-landing-system : lando
We're seeing crashes in bug 1545381 in appshell shutdown, which appear
to point to appshell initialization somehow falling over non-obviously.
Appshell initialization should basically never fail, so let's complain
loudly if that ever happens.
Differential Revision: https://phabricator.services.mozilla.com/D38915
--HG--
extra : moz-landing-system : lando
In some cases, other tools may show selected content in their UI. E.g.,
character palette of macOS. Therefore, we shouldn't allow them to show
masked password.
Differential Revision: https://phabricator.services.mozilla.com/D38430
--HG--
extra : moz-landing-system : lando
This patch creates editor API to get/set unmask-range of password field.
Its design is similar to `setSelectionRange()` and `selectionStart`/
`selectionEnd` attributes. The unmasked range is automatically
masked if `aTimeout` of `unmask()` is set to non-zero.
Otherwise, unmasked range won't be masked automatically even after user
or web apps modifies the editor, and inserting new character expands
unmasking range.
The following patch makes `TextEditRules` use these API to implement
delayed masking of password.
Note that editor has never supported dynamic `eEditorPasswordMask` change.
E.g., if you have typed some characters into an editor and toggle the flag,
the characters are not unmasked nor masked. Then, if you type new characters,
only they are correctly masked or unmasked. This patch puts `MOZ_ASSERT()`
to reject this feature completely on debug build for making the unmasking
implementation simpler.
Differential Revision: https://phabricator.services.mozilla.com/D38004
--HG--
extra : moz-landing-system : lando
Currently it's completely unclear at use sites that the getters for `once`
static prefs return the pref value from startup, rather than the current pref
value. (Bugs have been caused by this.) This commit improves things by changing
the getter name to make it clear that the pref value obtained is from startup.
This required changing things within libpref so it distinguishes between the
"base id" (`foo_bar`) and the "full id" (`foo_bar` or
`foo_bar_DoNotUseDirectly` or `foo_bar_AtStartup` or
`foo_bar_AtStartup_DoNotUseDirectly`; the name used depends on the `mirror` and
`do_not_use_directly` values in the YAML definition.) The "full id" is used in
most places, while the "base id" is used for the `GetPrefName_*` and
`GetPrefDefault_*` functions.
(This is a nice demonstration of the benefits of the YAML file, BTW. Making
this change with the old code would have involved adding an entry to every
single pref in StaticPrefList.h.)
The patch also rejigs the comment at the top of StaticPrefList.yaml, to clarify
some things.
Differential Revision: https://phabricator.services.mozilla.com/D38604
--HG--
extra : moz-landing-system : lando
This is latent bug in the code. The layers id used in the parent process'
call to SetTargetAPZC was always the one that the APZ hit-test produced.
But in the case where the parent process had a chrome event listener that
does a preventDefault on the event, that is the wrong layers id to use,
because we want to use the parent process' layers id instead of the content
process' layers id.
The reason the test in this bug hits this is because with WebRender enabled
the code in APZCTreeManagerParent that receives the SetTargetAPZC message
checks the layers id to see if it matches expectations (if it doesn't, it
assumes a malicious content process). In this scenario the layers id doesn't
match and causes an assertion failure. With this fix the layers id matches
expectations.
I don't believe this has any functional effect beyond the malicious content
process check.
Differential Revision: https://phabricator.services.mozilla.com/D38238
--HG--
extra : moz-landing-system : lando
CompositorInitiallyPaused() uses mNeedsUpdatingEGLSurface. But it is not good. mNeedsUpdatingEGLSurface is set to true in moz_container_set_initial_draw_callback(). If compositor is created before moz_container_set_initial_draw_callback(), compositor is not initially paused. It happens sometimes with popup window.
Differential Revision: https://phabricator.services.mozilla.com/D38200
--HG--
extra : moz-landing-system : lando
This was the order of calls for CompositorOGL+WebRender before this patch:
- PreRender, calls [mView preRender:]
- [compositing happens]
- PostRender, calls [mView postRender:]
And this was the order of calls for BasicCompositor before this patch:
- PreRender, ignored
- StartRemoteDrawing(InRegion)
- [software compositing happens]
- EndRemoteDrawing, calls [mView preRender:], does a GL composition, then calls [mView postRender:]
- PostRender, ignored
After this patch, all paths call [mView preRender:] and [mView postRender:] from
PreRender and PostRender.
This changeset also makes it so that we can't tear down the ChildView while
BasicCompositor compositing is happening.
Differential Revision: https://phabricator.services.mozilla.com/D37915
--HG--
extra : moz-landing-system : lando
This is an existing problem: Whenever you enter DOM fullscreen mode on a
window that has drawsContentsIntoWindowFrame == YES, we drop the information
about whether the title should be shown in the titlebar on top of Gecko's
drawing. Then, when you leave DOM fullscreen again, the freshly-created
ToolbarWindow will have mDrawTitle and titleVisibility set to their default
values: mDrawTitle defaults to NO and titleVisibility defaults to
NSWindowTitleVisible.
The title can be drawn in two different modes:
- If the ChildView is covering the titlebar and drawing it in its OpenGL
context, the ChildView handles the drawing of the title text. That drawing
code respects the window's mDrawTitle field, and ignores titleVisibility.
- If Cocoa is drawing the titlebar, it respects the titleVisibility property.
At the moment, Cocoa's drawing is never visible, because it is covered up by
the ChildView's OpenGL context. As a consequence, the extraneous title is never
actually visible on the screen and the bug doesn't actually cause a visible
glitch.
Once we use CoreAnimation, Cocoa's drawing will become visible, and the wrong
value of the titleVisibility property would become apparent.
Differential Revision: https://phabricator.services.mozilla.com/D26403
--HG--
extra : moz-landing-system : lando
386947-1.xul, now has one assertion since we take a different code
path with chrome URL's and XBL files. The assertion is triggered since the
binding is invalid.
Differential Revision: https://phabricator.services.mozilla.com/D34542
--HG--
extra : moz-landing-system : lando
In bug 1551961, a WM_NCMOUSELEAVE message handler was added to the Windows widget
backend so that we can detect when the mouse leaves a draggable region (since
draggable regions are hittest'ed as HTCAPTION, and are therefore "non-client").
Inside that message handler, we used the WindowAtMouse handler to determine whether
or not the WM_NCMOUSELEAVE was firing because the mouse was moving off of the
non-client area to client area, or moving out of the window entirely.
For reasons that are not particularly clear, on Windows Aero Glass, when clicking on
one of the min/max/close caption buttons, a WM_NCMOUSELEAVE message fires, and the
::WindowFromPoint call that WindowAtMouse uses causes another WM_NCMOUSELEAVE message
to be queued immediately after, so we end up in this situation where the message
loop is getting spammed by WM_NCMOUSELEAVE messages. Moving the mouse away from
the caption buttons seems to stop the spamming.
We have a function similar to WindowAtMouse called EventIsInsideWindow which
does not use ::WindowFromPoint, and this seems to sidestep the issue, while
being functionally equivalent.
Depends on D37421
Differential Revision: https://phabricator.services.mozilla.com/D37422
--HG--
extra : moz-landing-system : lando
We need to do this in order to not override the mouse cursor that we set in the
widget layer when hit-testing in the non-client region.
We were accidentally overriding before because the PresShell normally queues up a
synthetic mousemove event when the mouse is moving within the client region. That
mousemove cause the EventStateManager to update the cursor to match Gecko's
reckoning of the cursor CSS style of the underlying frame (which overrides the
cursor we may have set in the Windows non-client region hittest - for example,
one of the window resize cursors).
By clearing the mMouseInDraggableArea boolean when transitioning from a draggable
to non-draggable region in the non-client region, we make sure that we process the
WM_MOUSELEAVE message, which sends the eMouseExitFromWidget event into Gecko, which
effectively cancels handling of the synthetic mousemove.
Differential Revision: https://phabricator.services.mozilla.com/D37421
--HG--
extra : moz-landing-system : lando
Previously, `runRemoveContentTest()` was the last test which synthesizes
composition in the `<textarea>`. However, new test order move it to middle
of the other tests. Then, it hits an `MOZ_ASSERT()` in
`TextComposition::EditorWillHandleCompositionChangeEvent()` that detects a
bug. When editable element is removed from the DOM tree during a composition,
editor cannot listen to `eCompositionEnd` event. Therefore, when the editor
gets back, it still has destroyed `TextComposition` and keeps handling new
composition with the old one. Therefore this patch makes
`EditorBase::InstallEventListeners()` forget the destroyed composition.
Differential Revision: https://phabricator.services.mozilla.com/D37061
--HG--
extra : moz-landing-system : lando
