Change only right arrows to left arrows when direction is RTL
MozReview-Commit-ID: 3mMCAGCIUYP
--HG--
extra : rebase_source : 54fb3bc6455f2117cb0ddae06c4f895f58204d0f
extra : amend_source : 1398b66e4184c71755801c94ecdc2c22ab961ec0
extra : histedit_source : c17fae6ae45191a685af4274d5a6789fb040d691
The cause of bug 1309515 is, HandleKeyDownEvent() dispatches eKeyPress events even after InsertText() dispatches composition events via InsertTextAsCommittingComposition().
Therefore, this patch consumes the current key event after dispatching composition events.
Note that for consistency with Windows, InsertText() should use eKeyPress events rather than composition events at least in this case. However, changing the behavior has some risk. So, we should fix this bug with the safest hack for uplift.
MozReview-Commit-ID: 7FYR5N2lATe
--HG--
extra : rebase_source : 4485bd76a68567e8c20a84c2fbca78c626f592c5
When ContentCacheInParent receives eCompositionStart, it temporarily sets mCompositionStart to selection start offset. However, if there is a composition in the remote process, the selection start is caret position in the composition string. Therefore, it's not useful information. Instead, the composition start offset should be used because around there are a lot of information.
For that, ContentCacheInParent should always store compostion start offset in the remote process with mCompositionStartInChild even if mWidgetHasComposition is false. And when it receives eCompositionStart, mCompositionStart should be set to mCompositionStartInChild.
MozReview-Commit-ID: DksPNEsi6Ec
--HG--
extra : rebase_source : bcf2946273d24a4c37c33fa18a321660115e3fb6
ContentCacheInParent::mCompositionStart was set to ContentCacheInChild::mCompositionStart without any check. However, that's clearly wrong approach. For example, when the remote process handles some composition events after eCompositionCommit(AsIs) in the parent process, mCompositionStart is valid offset even after mWidgetHasComposition is set to false. Similarly, even after parent process sends eCompositionStart, the remote process may send invalid offset for mCompositionStart due to no composition in it.
For solving this issue, ContentCacheInParent should check mWidgetHasComposition.
If it's true and coming offset is valid, let's use it (even if mPendingCompositionCount is 2 or bigger since widget shouldn't use WidgetQueryContentEvent when there are some pending events).
If the coming offset is invalid but mWidgetHasComposition is false, let's use selection start instead because HandleQueryContentEvent() can work around selection start offset only.
Otherwise, i.e., mWidgetHasComposition is false, we should always set mCompositionStart to invalid offset.
MozReview-Commit-ID: IONU0Cbhpil
--HG--
extra : rebase_source : 05ae9eb0e7a0bb63b65be7a54103eb798274c8ff
When ContentCacheInParent receives content information from the remote process, it notifies TextComposition of the latest composition start offset in the remote process. However, the information may be older composition's, i.e., the composition was already committed in the process but is still being handled by the remote process. TextComposition shouldn't work with such obsolete information.
Note that TextComposition instance is created and destroyed when WidgetCompostionEvent is handled by IMEStateManager. Then, TextComposition instance guarantees that all following composition events for a composition are sent to same EventTarget (including TabParent). So, TextComposition is always synced with a composition in widget.
MozReview-Commit-ID: 78NuvpE2rPx
--HG--
extra : rebase_source : 3ffbf2513750f4b940bfedd19a8b191f28611115
If the remote process is busy or user restarts composition too quickly, there could be 2 or more compositions in ContentCache. For managing such case, ContentCacheInParent should manage the pending composition count which is increased at dispatching eCompositionStart event to the remote process and decreased at receiving eCompositionCommit(AsIs) event from the remote process.
MozReview-Commit-ID: KbTsK20NEZD
--HG--
extra : rebase_source : 428b14646ccde190b446b5c820e0e84866a855f2
For making the meaning of ContentCacheInParent::mIsComposing clearer, let's rename it to mWidgetHasComposition. It becomes true when the parent process sends eCompositionStart to the remote process and false when the parent process sends eCompositionCommit(AsIs). So, it represents if the widget (i.e., the native IME handler in the chrome process) has composition.
MozReview-Commit-ID: 5k05IXMgJxw
--HG--
extra : rebase_source : 69be70f22fb9b9dd6125ce390026b0740bd6de8f
Using direction-changed signal, we detect keyboard change for bidi.
When system uses fcitx's IM and ibus's arabic keyboard layout, this signal might fire often when switching layout and gdk_keymap_get_direction might return invalid bidi information. But I think that this is rare issue. Most users don't use Firefox Arabic version (it means that bidi.browser.ui = true) with ibus Arabic layout and fcitx CJK IM. Since there is no GTK3 API to get current IM module, I cannot find workaround for this.
MozReview-Commit-ID: DL8uUXJFWYz
--HG--
extra : rebase_source : a93e8969b878d01048726e03b24f362ead7e2059
Similarly, NativeKey is a stack class and mFollowingCharMsgs which stores following char messages of WM_(SYS)KEYDOWN should be AutoTArray for preventing to use heap at handling inputs with usual keyboard layout.
5 is enough size for handling usual keyboard layout because we support only 5 or less characters per dead key sequence.
MozReview-Commit-ID: IphcIOmPW0C
--HG--
extra : rebase_source : 74ce6f9fef73df4b9bbccbbdd713470ff7357889
Fortunately, UniCharsAndModifiers instances are only in stack. Therefore, we can make it a stack class and use nsAutoString and AutoTArray for not using heap at handling inputs from usual keyboard layouts.
MozReview-Commit-ID: 9ZPbdjGst64
--HG--
extra : rebase_source : 06a21c20575747591c6bf1a55dff7e6e38884333
Now, we have an security issue. mCommittedCharsAndModifiers may be initialized with multiple WM_(SYS)CHAR messages. However, if it's generated by odd (or malicious) middleware, mCommittedCharsAndModifiers may be overflown because it has only fixed array. For fixing this issue, first, we should hide the members for making the users not depend on the design of UniCharsAndModifiers.
This patch changes UniCharsAndModifiers to a class and hiding mChars and adding |CharAt() const|.
MozReview-Commit-ID: 5EjrIhmCdE4
--HG--
extra : rebase_source : a3dc68998f5f177582addd8587c4da86b2d17b3b
Now, NativeKey respects following WM_CHAR message. Therefore, we can create a test for bug 1293505 which a function key causes a printable character.
Additionally, bug 1307703 is now fixed by the previous patch. So, let's add automated test for it too.
Finally, now, I found a way to test with some keyboard layouts which are not available on old Windows. Therefore, we should add automated tests for bug 1297985 too.
MozReview-Commit-ID: IqCEPbPYrcQ
--HG--
extra : rebase_source : 451d0264f1180cae7d7035a498f1c13416d53246
This patch creates NativeKey::DispatchKeyPressEventsWithRetrievedCharMessages() for dispatching eKeyPress event with mCommittedCharsAndModifiers when it stores following printable WM_(SYS)CHAR messages.
Using loop for dispatching eKeyPress event for every WM_(SYS)CHAR message is wrong because WidgetKeyboardEvent::mKeyValue is initialized with mCommittedCharsAndModifiers and it causes TextEventDispatcher dispatching multiple eKeyPress events at every call of MaybeDispatchKeypressEvents(). Therefore, if mKeyValue is "^^", eKeyPress event is dispatched 4 times --for the first message, eKeyPress events are fired for each "^" and for the second message, eKeyPress events are fired again for each "^"--. Therefore, when it handles WM_(SYS)KEYDOWN and it causes inputting one or more printable characters, it's the easiest way not to use HandleCharMessage().
The new method calls TextEventDispatcher::MaybeDispatchKeypressEvents() only once and it requests to call the callback method with new argument of MaybeDispatchKeypressEvents() when it needs to dispatch 2 or more eKeyPress events. Then, NativeKey::WillDispatchKeyboardEvent() can set each eKeyPress event to raw information of the message and proper modifier state.
With this change, we can dispatch multiple eKeyPress events with retrieved WM_(SYS)CHAR message information rather than retrieved information from active keyboard layout. Therefore, NeedsToHandleWithoutFollowingCharMessages() doesn't return true even when mCommittedCharsAndModifiers stores two or more characters.
FYI: there is a bug in test_keycodes.xul. That is, Alt+'A' of Greek keyboard layout should cause WM_SYSCHAR with a corresponding Greek character but ASCII characters are specified. Therefore, this patch includes the fix of these bugs
MozReview-Commit-ID: JVm7ZJVug0O
--HG--
extra : rebase_source : 414ecbe2c01c53f294d1346414b1a289aa0abfe8
First, mCommittedCharsAndModifiers should be initialized with following char messages because the messages could be different from the information of current keyboard layout.
So, this patch guarantees that mCommittedCharsAndModifiers are same as the user expected when there is one or more WM_CHAR or WM_SYSCHAR messages.
MozReview-Commit-ID: I5Ack0xccoL
--HG--
extra : rebase_source : e4c7af75fd200f30ec52b53dc05f49ae8887c6f0
This is a simple mistake and blocks following patchs' automated tests.
For example, when Alt+Shift+foo doesn't cause text, this returns false even though it should return true.
MozReview-Commit-ID: 91L33vZhouT
--HG--
extra : rebase_source : be7d4e7ad670c1b03e6497c1c313c187ece80c1c
Now, NativeKey::HandleCharMessage() has almost same code, one is for dispatching eKeyPress event for non-printable keys or printable keys when one of Alt or Ctrl key is pressed, the other is for printable keys when Alt or Ctrl key is pressed.
The difference of them is, the former block removes Alt state and Ctrl state for handling AltGr key. When AltGr key is pressed, both Alt and Ctrl state are true. However, EditorBase treas keypress events whose altKey or ctrlKey is true as non-printable key event. Therefore, we need to set these modifier state to false when AltGr key is pressed and the key causes some text.
Note that as far as we know, when a key press with AltGr doesn't cause any characters, WM_CHAR isn't generated. Therefore, we don't need to check with complicated logic if the key event is actually inputting a character.
MozReview-Commit-ID: BRNWfICvkSm
--HG--
extra : rebase_source : 3b8a2aaeb4c53be1567f7a8d17dfef1af90406dd
Current shortcut key handling is really difference from what we did before struggling with "key hell". Therefore, remaining hacks for charCode in NativeKey::HandleCharMessage() are not necessary because they are for old code.
MozReview-Commit-ID: 3hvsBOiJ6VV
--HG--
extra : rebase_source : 7387fc89930f9b15d5d01612eb4a7a5726924bcd
This patch makes NativeKey::HandleCharMessage() stop dispatching eKeyPress event when the message is inputting a control character. NativeKey::HandleCharMessage() runs following cases:
1. WM_KEYDOWN followed by a WM_CHAR message whose wParam is not a control character causes inputting printable characters.
2. WM_CHAR message is received without WM_KEYDOWN message.
So, when HandleCharMessage() is called for a control character, it's unusual case. E.g., another application sends only WM_CHAR message with a control character.
On the other hand, Gecko is the only browser which dispatches "keypress" event even if pressed printable key doesn't cause inputting any characters. And such "keypress" event is used for shortcut key handling and some default action handling. So, even if we stop dispatching eKeyPress event from HandleCharMessage(), it shouldn't affect most users.
Note that this patch does NOT change the behavior when the user inputs a control character with usual keyboard layout such as Ctrl+A of en-US keyboard layout because DispatchKeyPressEventsWithoutCharMessage() dispatches eKeyPress event in such cases.
This patch also adds a lot of tests with Ctrl or Alt state for detecting regressions.
MozReview-Commit-ID: KUNqTp7RDSc
--HG--
extra : rebase_source : f91e37b0b82a8e8a95444a08f6d3101f14fadc50
Currently, Enter and Backspace keys are handled without following WM_(SYS)CHAR message. However, older code needs hack for them for avoiding conflict with Ctrl+J vs. Ctrl+Enter, etc.
So, we can get rid of them now. And when a keydown causes inputting a control character, NativeKey should handle it with keyboard layout (i.e., without following char message(s)).
MozReview-Commit-ID: 6bVuK0BzFbx
--HG--
extra : rebase_source : 08ff45c1990ef8f3ed7703c69b2dc22e5f1dd6f5
Currently, NativeKey::HandleCharMessage() is only a public method and it takes any char message. However, when it's called outside of NativeKey, it should work only with NativeKey::mMsg.
Therefore, we should make current HandleCharMessage() a private method and create new overload method which doesn't take MSG as an argument.
MozReview-Commit-ID: LowV2FUmR3U
--HG--
extra : rebase_source : 271c61515752cf1ea56cf12c4f7ef6f1a90c9cac
NativeKey shouldn't include characters which are provided by WM_SYSCHAR message or WM_DEADCHAR message into mCommittedCharsAndModifiers.
MozReview-Commit-ID: Ax1BmO5wTy0
--HG--
extra : rebase_source : 5d7d1735a783213f7c6a47cca0cadb1d4c5104c0
For consistency with IsPrintableCharMessage(), IsFollowedByNonControlCharMessage() should be renamed to IsFollowedByPrintableCharMessage().
MozReview-Commit-ID: CBJFPO4FZej
--HG--
extra : rebase_source : f5d677743be35a3bb5fca35e7658e907aa46dbcb
Currently, NativeKey::IsFollowedByNonControlCharMessage() returns true only when the first char message is a printable char message. Although, I don't know actual cases coming printable WM_CHAR message after non-printable char message, i.e.,
1. WM_SYSCHAR or WM_DEADCHAR
2. WM_CHAR for a printable character
or
1. WM_CHAR with a non-printable character (a control character)
2. WM_CHAR with a printable character
, we should make it return true because when one or more characters are being inputted, we should ignore non-printable char messages and handle printable char messages in the path handling text input.
MozReview-Commit-ID: 1v7v5mCRFCP
--HG--
extra : rebase_source : f82f7b77376450168bad34be031164ecf7338621
Currently, NativeKey::IsPrintableCharMessage() returns true if given message is WM_CHAR or WM_SYSCHAR. However, WM_SYSCHAR never causes inputting any character and even if WM_CHAR has a control character (i.e., non-printable character), it returns true.
MozReview-Commit-ID: IoU5F06WImz
--HG--
extra : rebase_source : c48f424f3d2960e6e9ceaa6d2fb09952ca634c3f
This guarantees that the runnable will be processed before shutdown leak checking.
MozReview-Commit-ID: B89AcDUWb3y
--HG--
extra : rebase_source : bfbd311fba367c1856cc5d972e6e999ccd4de1ac
Bug 1305498 - 1. Remove NotificationClient task queue; r=sebastian
Not sure why we needed a task queue for NotificationClient actions. The
actions all go through IPC and are non-blocking, so it's perfectly fine
to perform them off of whatever thread we're on.
Bug 1305498 - 2. Integrate NotificationHandler et al into NotificationCllient; r=sebastian
There's no reason to have NotificationHandler, AppNotificationClient,
and ServiceNotificationClient all separate from the base
NotificationClient class. This patch adds the functionality of
those three classes to NotificationClient.
The notifications hash map is changed from a ConcurrentHashMap to a
regular HashMap with synchronization because I think the use case here
doesn't warrant the added performance and overhead of ConcurrentHashMap.
NotificationService is changed to match the new NotificationClient. Now
the only job for NotificationService is to set a notification as
foreground, rather than to manage all notifications like before.
NotificationHandler, AppNotificationClient, and
ServiceNotificationClient will be removed in a later patch.
Bug 1305498 - 3. Set NotificationListener in GeckoApplication; r=sebastian
Set NotificationListener once in GeckoApplication.onCreate, instead of
spreading it out in GeckoApp, BrowserApp, and GeckoService. This is
possible because there's no longer a distinction between
AppNotificationClient and ServiceNotificationClient in the new,
consolidated NotificationClient.
Bug 1305498 - 4. Remove obsolete notification classes; r=sebastian
Remove AppNotificationClient, ServiceNotificationClient, and
NotificationHandler, now that they've all been replaced by the new,
consolidated NotificationClient.
Bug 1305498 - 5. Use NotificationReceiver for web notification callbacks; r=sebastian
Previously, web notification callbacks went to GeckoApp directly, but
that presented some problems such as not being able to implement the
on-close callback, because we don't want to launch GeckoApp when the
notification is closed by swiping. This patch makes us use
NotificationReceiver for callbacks, and let NotificationReceiver launch
GeckoApp if necessary.
Bug 1305498 - 6. Don't keep notification cookie in native code; r=sebastian
Keep the notification cookie a single location (in the notification
intent itself), and simplify the native notification handling code.
Bug 1305498 - 7. Use NotificationReceiver for persistent notifications; r=sebastian
Currently, persistent notification callbacks go through a different code
path, but it'd be more consistent and correct to let persistent
notification callbacks go through NotificationReceiver as well.
This takes care of some housekeeping work that was missing for
persistent notifications, such as deleting the mNotifications entry when
the notification is closed.
Sebastian Hengst
Carsten "Tomcat" Book
Makoto Kato
Masayuki Nakano
Marcos Caceres
Kartikaya Gupta
Neil Deakin
Stone Shih
Edwin Flores
Ryan VanderMeulen
Nicolas Silva
Markus Stange
Jim Mathies
Benjamin Smedberg
David Major
Jim Chen
Wes Kocher