Right now we report Java exception stacks through two code paths:
- GeckoAppShell.reportJavaCrash for exceptions caught by Java code
- jni::HandleUncaughtException for exceptions caught by native code
This patch combines the code for reporting the Java stack trace into
ReportException, to make it easier to add new features such as reporting
OOM exceptions.
We used to only handle certain exceptions in native code, and otherwise
forward the exception to Java handling code. Handling exceptions in
native code gives us better native stack traces in crash reports, so
this patch makes us always handle uncaught Java exceptions in native
code if possible, and only forward to Java as a fallback.
The patch is generated from following command:
rgrep -l unused.h|xargs sed -i -e s,mozilla/unused.h,mozilla/Unused.h,
MozReview-Commit-ID: AtLcWApZfES
--HG--
rename : mfbt/unused.h => mfbt/Unused.h
Refactor the dispatching code to be more streamlined, and add two
optimizations to the way we dispatch a call:
* avoid a pair of unnecessary calls to add/delete the global class ref,
when dispatching a static call to the Gecko thread without a class ref
parameter.
* avoid an extra allocation when dispatching to a proxy function.
Implement checking the calling thread of a JNI call based on the
calledFrom attribute set in WrapForJNI. Also implement automatic call
dispatching based on the dispatchTo attribute set in WrapForJNI. This
eliminates the use of UsesNativeCallProxy and UsesGeckoThreadProxy.
Update the code generator and related classes in annotation processor to
use the new WrapForJNI flags. Also add some more sanity checking to make
sure the flags are used correctly.
Add native method calls to AndroidGamepadManager to replace the gamepad
events in GeckoEvent. Implement those calls in AndroidGamepad.cpp. The
jni/Refs.h change is necessary to fix a compile error when using
jni::BooleanArray.
PrefsHelper uses Maybe<> as a replacement for a ternary expression, in
order to work around jni::StringParam's lack of copy constructor.
However, we can add a move constructor to StringParam, which lets us use
it in a ternary expression, and avoid the awkwardness of Maybe<>.
This removes the unnecessary setting of c-basic-offset from all
python-mode files.
This was automatically generated using
perl -pi -e 's/; *c-basic-offset: *[0-9]+//'
... on the affected files.
The bulk of these files are moz.build files but there a few others as
well.
MozReview-Commit-ID: 2pPf3DEiZqx
--HG--
extra : rebase_source : 0a7dcac80b924174a2c429b093791148ea6ac204
PrefsHelper uses Maybe<> as a replacement for a ternary expression, in
order to work around jni::StringParam's lack of copy constructor.
However, we can add a move constructor to StringParam, which lets us use
it in a ternary expression, and avoid the awkwardness of Maybe<>.
clang complains that it's unable to instantiate this template because
the functions being passed as arguments are MOZ_JNICALL, while the
actual parameter to the function has no such attribute. Adding the
attribute makes everything line up.
clang complains about specializations of NativeStubImpl not being able
to see the private constructor of ProxyNativeCall. While
ProxyNativeCall includes a friend declaration for NativeStubImpl, it's
not obvious *which* NativeStubImpl is being friended, as NativeStubImpl
hasn't been forward declared and exists in a completely separate
namespace. Forward declaring NativeStubImpl and adjusting the friend
declaration makes everything more correct.
Even though a class may not be multi-threaded, its object references can
still be passed to different threads. As long as the references are not
accessed, we should allow it, and therefore GlobalRef should use
GetEnvForThread for ref/unref, instead of using FindEnv, which can
trigger assertions on non-main threads.
WindowEvent is not specific to GeckoViewSupport, and classes other than
GeckoViewSupport use it, so this patch moves it to outside of
GeckoViewSupport.
Improve the performance of JNI calls by making JNI calls require a
Context object. LocalRef inherits from Context and can make calls
directly. Non-local Ref classes will generate a Context object when
making a call. The patch also makes the template design cleaner in
several cases.
Implement the PrefsHelper native methods. The previous browser.js
implementation supported "pseudo-prefs" that did not exist as actual
prefs, but was accessible through PrefsHelper. In order to accommodate
these pseudo-prefs, we send observer notifications in order to
communicate with browser.js about prefs that we don't support.
This patch adds specializations of jni::Ref for primitive arrays like
IntArray and FloatArray, so that the arrays can be accessed through the
GetElement, GetElements, and Length members.
When we have a Java exception in native code, the Java stack in the
exception will not be very useful because the top frame is the native
entry point. In this case, the native stack is more useful. However,
currently we don't get a good native stack in this situation because we
go through Java when handling the exception, and the native stack we get
will have a lot of unknown frames inside libdvm or libart. This patch
makes us stay in native code when handling an uncaught exception from
native code, so that we get a good native stack.
GLController instances are associated with a particular nsWindow, rather
than a particular View. Therefore, we need to let GeckoView manage
GLController instances, as part of GeckoView's handling of saving and
restoring states.
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.
It's helpful to have a static DisposeNative when disposing of instances
that are still associated with a native object through an null weak
pointer. In that case, we can't call instance methods because the weak
pointer is null, but we can still call static methods.
This patch adds a specialization for jni::Ref<jni::ObjectArray>, which
includes members for getting the length of the array and accessing
array elements.
This patch adds a specialization for jni::Ref<jni::ObjectArray>, which
includes members for getting the length of the array and accessing
array elements.
The patch removes 455 occurrences of FAIL_ON_WARNINGS from moz.build files, and
adds 78 instances of ALLOW_COMPILER_WARNINGS. About half of those 78 are in
code we control and which should be removable with a little effort.
--HG--
extra : rebase_source : 82e3387abfbd5f1471e953961d301d3d97ed2973
If a C++ class implements native calls that all return void, it can
choose to have those calls go through a custom proxy function by
inheriting from mozilla::jni::UsesNativeCallProxy and override the
ProxyNativeCall member.
ProxyNativeCall accepts a rvalue reference to a
functor object specific to each call, and can alter the calling
environment (e.g. dispatch the call to another thread) before issuing
the actual native call through the functor's operator().
Any JNI refs contained in the call are automatically turned into global
refs so that the call can continue to work outside of the original JNI
call.
Native call proxy will be used to implement automatic dispatching of
native calls from the main thread to run on the Gecko thread.
This patch fixes a compile error when using WeakPtr, where Impl* was
expected but SupportsWeakPtr<Impl>* was given.
This patch also fixes a compile error when using the DisposeNative
implementation provided by the autogenerated Natives class. e.g.,
> struct Foo : Bar::Natives<Foo> {
> using Bar::Natives<Foo>::DisposeNative;
> };
This uses a default implementation of DisposeNative instead of a custom
implementation, and resulted in a compile error that this patch fixes.
The mangled name of a NativeStubImpl instantiation is longer than
necessary because of the ReturnType arg. This patch turns it into a bool
parameter. It also reorders the parameters for cleanness.
Add more constructors in LocalRef and GlobalRef to accommodate use cases
where a JNIEnv is already available for performance reasons. Also add
more null-checks when creating references for performance reasons.
clang doesn't like it when we declare things as a struct in one place
and a class in another. This change makes all the current usages
consistent and clang happy.