In order to prevent the deadlock, we need to release sAppShellLock when
we start waiting in SyncRunEvent. However, we cannot simply unlock it
before the wait because that introduces an out-of-order unlocking wrt
mSyncRunMonitor, which can cause further deadlocks. So this patch
converts mSyncRunMoitor to a condvar and make it use sAppShellLock. That
then involves making aAppShellLock a Mutex instead of a StaticMutex. The
final result is having one lock (sAppShellLock), which supports any
other condvars that we have like mSyncRunFinished.
This patch moves the SyncRunEvent logic from GLControllerSupport to
nsAppShell, as it could be useful elsewhere. This patch fixes a race
condition related to shutdown, where a deadlock could occur if Gecko
shuts down when another thread is waiting for a synchronous event to
finish running. This patch also fixes a crash on shutdown when we tried
to create a mutex after the deadlock detector has shut down.
When getting nsAppShell from another thread, there could be a race with
nsAppShell being destroyed on the main thread. This patch makes the raw
nsAppShell pointer only accessible from the main thread, and use a
static mutex to coordinate accessing nsAppShell from other threads.
Right now LayerView depends on geckoConnected being called by GeckoApp
or GeckoView during its initialization. However, we can get rid of it
and let LayerView handle the task itself. As part of this change, screen
depth overriding is moved to native code in nsAppShell.
GeckoEvent.KEY_EVENT and GeckoEvent.IME_EVENT used to count as UI events
for the hang monitor. We should count the new native calls as native
events too through this patch.
The bulk of this commit was generated with a script, executed at the top
level of a typical source code checkout. The only non-machine-generated
part was modifying MFBT's moz.build to reflect the new naming.
CLOSED TREE makes big refactorings like this a piece of cake.
# The main substitution.
find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \
xargs perl -p -i -e '
s/nsRefPtr\.h/RefPtr\.h/g; # handle includes
s/nsRefPtr ?</RefPtr</g; # handle declarations and variables
'
# Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h.
perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h
# Handle nsRefPtr.h itself, a couple places that define constructors
# from nsRefPtr, and code generators specially. We do this here, rather
# than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename
# things like nsRefPtrHashtable.
perl -p -i -e 's/nsRefPtr/RefPtr/g' \
mfbt/nsRefPtr.h \
xpcom/glue/nsCOMPtr.h \
xpcom/base/OwningNonNull.h \
ipc/ipdl/ipdl/lower.py \
ipc/ipdl/ipdl/builtin.py \
dom/bindings/Codegen.py \
python/lldbutils/lldbutils/utils.py
# In our indiscriminate substitution above, we renamed
# nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up.
find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \
xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g'
if [ -d .git ]; then
git mv mfbt/nsRefPtr.h mfbt/RefPtr.h
else
hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h
fi
Right now we don't allow quitting Fennec when the last nsWindow closes
(e.g. when the last GeckoView is destroyed) because we want to keep the
Gecko thread running throughout the app process lifetime. However, when
we are asked to quit explicitly through nsIAppStartup::Quit, we should
release the hold on nsAppStartup and allow quitting to continue.
The PROCESS_OBJECT GeckoEvent is used to set the layer client object in
Gecko once Gecko is done loading. This patch converts it to a native
call in GeckoView.Window.
The PROCESS_OBJECT GeckoEvent is used to set the layer client object in
Gecko once Gecko is done loading. This patch converts it to a native
call in GeckoView.Window.
The PROCESS_OBJECT GeckoEvent is used to set the layer client object in
Gecko once Gecko is done loading. This patch converts it to a native
call in GeckoView.Window.
We have a pretty messy system of coalescing viewport events that
introduced a race condition during the recent JNI refactoring. This
patch makes that code simpler and fixes the race condition. Instead of
keeping track of a previous viewport event, we now scan the event queue
for previous viewport events. This shouldn't be a perf concern because
we only scan the queue for viewport and native callback events, and stop
scanning as soon as we find another kind of event.
Originally, the GeckoThread PROFILE_READY state was chosen to correspond
to the profile-do-change event, to give priority to JNI code (e.g.
window creation) over other events that may be registered under
profile-after-change event. However, this leads to broken tests because
our testing infra expects things like window creation to happen during
profile-after-change at the earliest. This is because we have to wait
for addons like SpecialPowers to be loaded between profile-do-change and
profile-after-change. This patch changes the PROFILE_READY state to
correspond to the profile-after-change event, so things are consistent
again.
nsWindow will implement native methods of GeckoView.Window. This patch
implements the open method, which opens a new window in the same manner
as the CLH, and associates the new nsWindow with the GeckoView.Window
instance.
With the new nsAppShell event loop based on runnable events, we need to
implement AndroidGeckoEvent handling as a runnable event. This patch adds
nsAppShell::LegacyGeckoEvent and adopts its implementation from existing
code that handle AndroidGeckoEvent.
nsAppShell is currently based on AndroidGeckoEvent objects, which mirror
GeckoEvent on the Java side. With GeckoEvent going away, we will be
gradually removing AndroidGeckoEvent as well. This patch makes the
nsAppShell event loop based on runnable objects, which derive from
nsAppShell::Event. Using runnable objects is much more flexible and allows
us, for example, to post a lambda to the event loop to be run later.
Gecko on Android follows the Android app model where it never stops
until it is killed by the system or told explicitly to quit. Therefore,
we should exit Gecko when there is no window or the last window is closed.
One thing we do in the Fennec CLH is to make a speculative connection
based on the URI that's passed in. However, by the time the CLH runs,
we're far along into startup, and the advantage of a speculative
connection is reduced. This patch implements making speculative
connection as a method in GeckoThread, so that Fennec can make a
speculative connection without relying on the Fennec CLH.
Many Gecko operations depend on the profile being available. This
patch adds a PROFILE_READY Gecko state so that we can queue calls
until profile is loaded.
First we need to set the Gecko thread JNIEnv* in nsAndroidStartup, but
after that we can initialize and deinitialize the rest of JNI, including
AndroidBridge, in GeckoAppShell. This makes nsAppShell control the
AndroidBridge lifetime. Over time, parts of the AndroidBridge
functionality will be migrated to nsAppShell.
