MozReview-Commit-ID: LVML7EZaSYD
In non-e10s AccessibleWrap::HandleAccEvent, we special case our handling of
CARET_MOVED and FOCUS events with a call to UpdateSystemCaretFor. In e10s mode
we were not doing the same thing for proxied events sent from content. This
threw JAWS for a loop and presumably messes up other ATs as well.
This patch modifies the IPDL messages for these two events so that we may
send the caret rect along with the event, thus allowing us to update the
system caret for proxied events as well.
--HG--
extra : rebase_source : e1502c12b038739520afd5c7078d011e25ea669e
Here is a scenario under which the assert is invalid:
RecvHideEvent is received on a DocAccessibleParent DAP1 that has a child document DAP2... that in turn has a child document DAP3. DAP1::RecvHideEvent calls Shutdown on its root ProxyAccessible. This recursively calls Shutdown on child ProxyAccessibles until it reaches the one that is the parent of the child document DAP2. ProxyAccessibleBase::Shutdown then attempts to Unbind DAP2, which calls DAP1::RemoveChildDoc, which trips the assert because DAP2 has child document DAP3.
In content processes, we want to avoid a situation where a DocAccessible has lost its IPC actor (DocAccessibleChild) but finds itself processing a delayed request to take on child documents. In this case, we shut down the child documents as they are no longer valid.
Now that we inherit the recommended configurations, many definitions are handled semi-automatically.
MozReview-Commit-ID: 1H2MXzk6VdQ
--HG--
extra : rebase_source : 25e9e8b9a9f62d4e22873998ff4532fbcbe2e367
It is awkward to need to have a DocAccessibleChild to send events with before
DocAccessible::DoInitialUpdate() is called because we might not have a
TabParent for the document until then. However we could try to fire reload
events on documents before DoInitialUpdate() was called which would require
having a DocAccessibleChild there to send the load event to the parent process.
However before DocAccessible::DoInitialUpdate() is called the document already
has a pending load event. If a document hasn't yet been the subject of a load
event it doesn't make a lot of sense to fire a reload event for that document,
and then the initial load event. So it should be safe to skip firing
reload events for documents where mLoadEvent specifies a load event to
fire in the future.
We sometimes briefly have more than one root DocAccessible associated with a TabChild, for example, while navigating links in a page. This patch makes sure that we use the correct accessible when delaying messages that we forward to the parent process.
This avoids conflicts with mozilla::dom::FrameType.
MozReview-Commit-ID: 7aEMbHRaTFk
--HG--
extra : rebase_source : 2d01321f5ce0ec8c0e3f70984674f82678034b3c
We can lower the eslint cyclomatic complexity threshold in some directories without adding eslint suppression comments in any .js source files. We need to specify the complexity rule in accessible/.eslintrc because it doesn't inherit the mozilla/recommended rules. eslint's default complexity threshold is 20.
Also bump the eslint-plugin-mozilla version because we modified the mozilla/recommended rules.
MozReview-Commit-ID: 57T4gAjPH7z
--HG--
extra : rebase_source : 4565abfa722b9459cfb4e006e843da13ed7cffd4
extra : intermediate-source : 658588564c08c9fd5e60633d1457f24087de8570
extra : source : 7e0526e3b943419a80c0cd2fa462cabbf8925eb1
eslint's default max-nested-callbacks threshold is 10, but now we make it an error. We could further lower the max-nested-callbacks threshold globally to 8, like browser/.eslintrc.js, but that would require adding suppression comments in (two) more .js test files. 10 seems good enough for now since it's the eslint default.
We need to specify max-nested-callbacks in accessible/.eslintrc because it doesn't inherit the mozilla/recommended rules.
Also bump the eslint-plugin-mozilla version because we modified the mozilla/recommended rules.
MozReview-Commit-ID: JA41vsi4U7j
--HG--
extra : rebase_source : 2dd211ebd3b8cf83f67f26cac5244bec8978f0e2
extra : intermediate-source : 6f5e75502be394038543029e3cfd474c5b1c2e98
extra : source : a13437d73c97fabd073ab8a6f93e85a5084b7405
The "no-spaced-func" name was deprecated in ESLint v3.3.0 and replaced by "func-call-spacing", which is already specified in the mozilla/recommended rules and some other .eslintrc.js files. We need to specify func-call-spacing in accessible/tests/browser/.eslintrc.js because it doesn't inherit the mozilla/recommended rules.
MozReview-Commit-ID: 7L8fuVtTu0X
--HG--
extra : rebase_source : 9cbd3717e6360d47b1a4589e8d5658ccf4bcba59
extra : intermediate-source : 61d723ca5f9b4dd9a22f2956c2f49d998e9db6c9
extra : source : e589244b9db5a744166ed151ff3e2e77432fdb04
MozReview-Commit-ID: Kp8x5o66nrY
I want AccessibleHandler.dll to use different UUIDs based on release channel.
The way I was doing it before wasn't working correctly because I also wanted
local builds to have their own set of UUIDs vs our regular Nightly/Beta/Release
builds.
I also want the beta channel to have its own set of UUIDs that are distinct
from release.
I'm using MOZ_UPDATE_CHANNEL to distinguish between the channels when
NIGHTLY_BUILD and BETA_OR_RELEASE are insufficient.
--HG--
extra : rebase_source : 8cb28a22a3cac16fb743a8fe81db5e120c1fdf6d
We need to update mChildDocs on child document removal. This also cleans up some code related to setting a new child doc when one is already set -- we now assert that this cannot happen.
In the next patch we want to add a method called
GetUnanimatedStyleContextForElementNoFlush but that's much too long. Instead it
seems better to just drop 'ForElement' from all these methods since it should be
fairly obvious we are getting the style context for an element given that the
first argument is an element.
MozReview-Commit-ID: JQKaEuCKV2F
--HG--
extra : rebase_source : 3ba51f3b00d1ec7bc91102629d9c0abb88992fef
MozReview-Commit-ID: HkiIgDQURtK
The a11y COM handler consists of two classes. AccessibleHandler is the one that
is actually interposed by COM between the client and the proxy. It implements
IAccessible2_3, IServiceProvider, and IProvideClassInfo. It is derived from
mscom::Handler which takes care of most of the COM boilerplate for handler
instantiation. AccessibleHandler must override a few virtual functions from
mscom::Handler:
* QueryHandlerInterface - mscom::Handler must implement the "real" QI. This is
the method that mscom::Handler uses to delegate to the implementer.
* ReadHandlerPayload - deserializing the payload. This uses
mscom::StructFromStream and the IA2Data_Decode function that was generated by
midl. It should essentially be the inverse of the serialization code in the
IHandlerProvider implementation inside Gecko.
* MarshalAs - identical to IHandlerProvider::MarshalAs
* GetHandlerPayloadSize and WriteHandlerPayload - it sucks that we have to
include these, but they are neccessary for the scenario when COM wants to
make a proxy of a proxy. When that happens, we need to get COM to serialize
the payload again; it isn't smart enough to recognize payloads in existing
proxies and copy them itself.
In order to reduce round trips, AccessibleHandler implements the most
sophisticated IA2 interface. This way the client has access to any of those
interfaces from IDispatch all the way down to IAccessible2_3 without needing to
incur additional round-trips to obtain them.
We also implement IServiceProvider so that a client may QueryService for an IA2
interface without needing to incur more round trips. In addition, we maintain
a list of service IDs that we definitely do not support, and handle those
locally instead of incurring an IPC round-trip.
We also implement IProvideClassInfo because NVDA's python code (and presumably
other ATs that use interpreted languages) supports using that interface to
assist its FFI library with resolution of our COM objects.
COM wraps each instance of an IAccessible* proxy with an AccessibleHandler
object. Proxies whose interfaces are not part of the IDispatch ... IAccessible2_3
hierarchy are not wrapped with handlers (though this might change in the future
if we need to expand coverage). AccessibleHandler implements IAccessible2_3 such
that, if we have cached data available for a particular IA2 method, we provide
that cached data to the client. Otherwise we pass on the request to the COM
proxy for retrieval via IPC.
The other class in the handler dll is AccessibleHandlerControl. This class
implements the IHandlerControl interface that is provided to IGeckoBackChannel.
Gecko dispatches cache invalidation events using this interface. This class
is a singleton within the handler dll and manages shared, handler-wide resources.
--HG--
extra : rebase_source : 4fc3910f789b0ed92c90655344ffb89f576ca5bd
MozReview-Commit-ID: GTQF3x1pBtX
A general outline of the COM handler (a.k.a. the "smart proxy"):
COM handlers are pieces of code that are loaded by the COM runtime along with
a proxy and are layered above that proxy. This enables the COM handler to
interpose itself between the caller and the proxy, thus providing the
opportunity for the handler to manipulate an interface's method calls before
those calls reach the proxy.
Handlers are regular COM components that live in DLLs and are declared in the
Windows registry. In order to allow for the specifying of a handler (and an
optional payload to be sent with the proxy), the mscom library allows its
clients to specify an implementation of the IHandlerProvider interface.
IHandlerProvider consists of 5 functions:
* GetHandler returns the CLSID of the component that should be loaded into
the COM client's process. If GetHandler returns a failure code, then no
handler is loaded.
* GetHandlerPayloadSize and WriteHandlerPayload are for obtaining the payload
data. These calls are made on a background thread but need to do their work
on the main thread. We declare the payload struct in IDL. MIDL generates two
functions, IA2Payload_Encode and IA2Payload_Decode, which are used by
mscom::StructToStream to read and write that struct to and from buffers.
* The a11y payload struct also includes an interface, IGeckoBackChannel, that
allows the handler to communicate directly with Gecko. IGeckoBackChannel
currently provides two methods: one to allow the handler to request fresh
cache information, and the other to provide Gecko with its IHandlerControl
interface.
* MarshalAs accepts an IID that specifies the interface that is about to be
proxied. We may want to send a more sophisticated proxy than the one that
is requested. The desired IID is returned by this function. In the case of
a11y interfaces, we should always return IAccessible2_3 if we are asked for
one of its parent interfaces. This allows us to eliminate round trips to
resolve more sophisticated interfaces later on.
* NewInstance, which is needed to ensure that all descendent proxies are also
imbued with the same handler code.
The main focus of this patch is as follows:
1. Provide an implementation of the IHandlerProvider interface;
2. Populate the handler payload (ie, the cache) with data;
3. Modify CreateHolderFromAccessible to specify the HandlerPayload object;
4. Receive the IHandlerControl interface from the handler DLL and move it
into the chrome process.
Some more information about IHandlerControl:
There is one IHandlerControl per handler DLL instance. It is the interface that
we call in Gecko when we need to dispatch an event to the handler. In order to
ensure that events are dispatched in the correct order, we need to dispatch
those events from the chrome main thread so that they occur in sequential order
with calls to NotifyWinEvent.
--HG--
extra : rebase_source : acb44dead7cc5488424720e1bf58862b7b30374f
unfortunately because creating a DocAccessibleParent is not atomic one can be
created by AllocPDocAccessible(), but then RecvPDocAccessibleConstructor() can
fail. If that happens we will call actor destroyed on the new
DocAccessibleParent, however because the constructor messaged failed it will
not have a wrapper. That means this assert does not necessarily hold.
The Manager is set by IPDL for remotely constructed objects but our DocAccessibleChilds are created on the child process side, so we need to assign a manager in the constructor so that we can find it when needed.
--HG--
extra : rebase_source : 8bf76534860ed73fbdc71df494130f6028400fa3
Its possible to coalesce away events such that the first two events in the
queue are reorder events where the second reorder can be coalesced with the
first. In that case there's no point in shuffling the list before removing
the second reorder event.
Really, there are only two cases we need to worry about. Either
IsSelectionCached(), and then our SelectionProperties has the data we want, or
not and then we have a non-null mSelCon which has the data we want.
Since we are now using cached selection state a lot more (instead of
initializing the editor whenever someone asks for selection state), we need to
actually update it more correctly when .value is set.
And since we now update the cached selection state for the case when .value has
been set (to point to the end of the text), we need to change
HTMLInputElement::HasCachedSelection to return false for that case. Otherwise
we will always do eager editor init on value set. We handle that by not doing
eager init if the cached selection is collapsed.
The web platform test changes test the "update on .value set" behavior. They
fail without this patch, pass with it.
MozReview-Commit-ID: DDU8U4MGb23
Really, there are only two cases we need to worry about. Either
IsSelectionCached(), and then our SelectionProperties has the data we want, or
not and then we have a non-null mSelCon which has the data we want.
Since we are now using cached selection state a lot more (instead of
initializing the editor whenever someone asks for selection state), we need to
actually update it more correctly when .value is set.
And since we now update the cached selection state for the case when .value has
been set (to point to the end of the text), we need to change
HTMLInputElement::HasCachedSelection to return false for that case. Otherwise
we will always do eager editor init on value set. We handle that by not doing
eager init if the cached selection is collapsed.
The web platform test changes test the "update on .value set" behavior. They
fail without this patch, pass with it.
MozReview-Commit-ID: DDU8U4MGb23
a BindChildDoc message can race with the parent process shutting down a tab.
The result of that is that RecvBindChildDoc() can be called on a
DocAccessibleParent that has already been shut down by the destruction of the
owning TabParent.
sending the emulated window information is basically unrelated to sending the
parent COM proxy to the child process, and in the future it will be useful to
send these at different times.
Use strcmp() instead of strncmp() when comparing the AtkObject name
before changing it. 9133117f410c06ad7fa25323a168bc97200043b9 changed
from Equals() to strncmp() to prevent issues related to embedded NUL
bytes, but the change lead to only checking if the AtkObject name has
the new value as prefix. Instead, use strcmp() to check the whole
C strings, but still stopping at the first NUL byte in the new name.
strcmp() has to be safe here, as NS_ConvertUTF16toUTF8::get() has to
return a NUL-terminated string or the code would crash a few lines
below when calling strdup() on it.
the outer doc of a top level content document does not necessarily exist when
we recieve the PDocAccessibleConstructor message for it. So we shouldn't asser
there is an outer doc here.
Follow-up to fix breakage in accessibility caused by the bundle
conversion. In particular, optString(foo) should have been converted to
getString(foo, "") because optString returns "" by default.
Also fix a small bug in Presentation.jsm where an array or null should
be used instead of a string.
There are cases where there is no good reason to shutdown the child document
actors before the parent. For example when TabParent::DestroyInternal() is
about to send the Destroy message we might as well just shut down the top level
document and let that tear down the rest.
Since it can be called in strange orderings wrt SetChildDoc() it seems
reasonable to believe we can get an ordering where the proxy has a document as
a child but is not marked as having document children.
It seems likely that some documents are created in content processes without
a DocAccessibleChild actor because there is no docshell or tabchild
associated with the document. However DocAccessible::DoInitialUpdate()
already calls functions that assume the document is associated with a
docshell. So hopefully trying to create the child actor there will mean it
is more successful.
After bug 1329814 a DocAccessibleParent can be shutdown by the parent process
before a content process has started shutting down the document that
DocAccessibleParent represents. That means the content processcan send an
event from a live document to a DocAccessibleParent that has already been
destroyed.
When TabChild recieves the Destroy message from its parent the doc accessibles
for that tab are shut down. However if the Shutdown message sent by
DocAccessible::Shutdown() isn't recieved before the child process handles the
runnable that sends __delete__ to the tab parent actor the parent can send a
message to the already shut down doc accessible child actor.
It causes warnings if the logging define is undefined as it is on aurora,
because the field is then unused. Since the only uses can easily be
replaced with the mQueueEvents field we can just do that and remove the
useless field.
Bug 1321418 - 1. Use GekcoBundle events in GeckoApp; r=snorp r=sebastian
Switch GeckoApp to using GeckoBundle events everywhere. UI or Gecko
events are used if the event requires the UI or Gecko thread,
respectively, and background events are used for all other events.
There are changes to some other Java classes, such as SnackbarBuilder
and GeckoAccessibility, due to the switch to GeckoBundle.
For "Snackbar:Show", we need the global EventDispatcher because the
event can be sent to both GeckoApp and GeckoPreferences. Howveer, we
only want one listener registered at the same time, so we register and
unregister in GeckoApp's and GeckoPreferences's onPause and onResume
methods.
Bug 1321418 - 2. Use appropriate JS EventDispatcher to send GeckoApp events; r=snorp r=sebastian
Change JS code that sends events to GeckoApp to use either the global
EventDispatcher or the per-window EventDispatcher.
"Session:StatePurged" is not used so it's removed. "Gecko:Ready" in
geckoview.js is not necessary because it's only used for GeckoApp, so
it's removed from geckoview.js.
Bug 1321418 - 3. Use GeckoBundle events in BrowserApp; r=snorp r=sebastian
Switch BrowserApp to using GeckoBundle events, in a similar vein as
GeckoApp. UI or Gecko events are used if the event handlers required UI
or Gecko thread, respectively, and background events are used for all
other events.
Some other Java classes also have to be modified as a result of
switching to GeckoBundle.
Bug 1321418 - 4. Use global EventDispatcher to send BrowserApp events; r=snorp r=sebastian
Change JS code that sends events to BrowserApp to use the global
EventDispatcher instead of "Messaging".
Bug 1321418 - 5. Update usages of events in tests; r=gbrown
Update cases where we use or wait for events in tests.
Dan Banner
tiago
Eitan Isaacson
Aaron Klotz
David Parks
Mark Banner
Alexander Surkov
Wes Kocher
Carsten "Tomcat" Book
Tom Tromey
Trevor Saunders
Emilio Cobos Álvarez
Masatoshi Kimura
Chris Peterson
Florian Queze
Sebastian Hengst
Eugen Sawin
Ryan VanderMeulen
Brian Birtles
Paul Bignier
Jeff Muizelaar
Boris Zbarsky
Andreas Farre
Michael Layzell
Jim Chen
David Major
Mats Palmgren
Yura Zenevich
Phil Ringnalda
Xidorn Quan
Yura Zenevich
Sylvestre Ledru
Colomban Wendling
Marco Zehe
Geoff Brown
Nathan Froyd
Michelangelo De Simone
Jean-Luc Bonnafoux
Joel Maher
Cameron McCormack
Tim Nguyen
Ting-Yu Chou
Jessica Jong
Robert Longson
Kan-Ru Chen
Dão Gottwald