Different from Windows and macOS, we cannot check if active keyboard layout
works as "IME" actually. Therefore, this patch add the telemetry probe
to the dispatcher of eCompositionStart. However, composition string is also
used by some Wester keyboard layouts which have dead keys. So, the meaning
of the result is deferent from the other platforms, but it must be useful
information which IM (e.g., fcitx, ibus) is used by most users.
MozReview-Commit-ID: A7vYuGtcrRw
On some Linux environment, GTK_IM_MODULE env may be "xim". Then, actual
IM is specified with XMODIFIERS env with "@im=". Therefore, if active IM
context ID is xim, IMContextWrapper::Init() needs to look for actual IM name
in XMODIFIERS.
MozReview-Commit-ID: 1aGjBkF4AQn
IIIMF is really old IME framework. In these days, it's not used as default IM
module of any major distributions. However, ATOK X which is old proprietary
IME requires IIIMF and it's still used by some Japanese IME users. Therefore,
we need to take back the hack to prevent crash caused by IIIMF.
We did increment refcount of GtkIMContextIIIM class to prevent IIIM module
from being unloaded. However, it was not ported when we changed default
toolkit from GTK2 to GTK3. So, this is doing the porting.
Unfortunately, the instance of GtkIMContextIIIM is wrapped by opacity struct.
So, it's not safe to access the pointer with declaring a mimic struct.
Therefore, we should directly get GType from the name with calling
g_type_from_name("GtkIMContextIIIM") instead of using G_TYPE_FROM_INSTANCE()
and g_type_name().
MozReview-Commit-ID: GCZaSUtPiS9
uim is an old IM which uses key snooper to listen to key events rather than
via filter key event API which should be called by applications. It's still
used by Debian 9.x, so, we still need to support this.
Unfortunately, we cannot detect if uim actually uses key snooper because it's
switch by build option of uim. Currently, Debian builds uim as using key
snooper. So, we should assume uim uses key snooper always. On the other
hand, somebody *might* use uim built as not using key snooper, so, let's
decide if uim uses key snooper with new pref,
"intl.ime.hack.uim.using_key_snooper", but its default should be true.
Note that ibus and Fcitx also have the mode to use key snooper (perhaps for
backward compatibility with uim). However, it's not enabled in default
settings and even if it's enabled, Firefox is in whitelist in the default
settings of them for stop using key snooper. Therefore, we don't need to
support key snooper mode for them unless we'll get some requests to
support their key snooping mode.
MozReview-Commit-ID: 6fTsfKrHzvo
ibus and fcitx usually post key event to other IME process, then, if it causes
some signals to updating composition string, they may not send the posted
key event to us again. Then, IMContextWrapper dispatches neither eKeyDown nor
eKeyUp event since mProcessingKeyEvent becomes non-nullptr only while
OnKeyEvent() is called. So, IMContextWrapper need to store key event if
OnKeyEvent() assumes that given key event is posted to different process.
Then, if IMContextWrapper receives some signals, it should dispatch eKeyDown
and eKeyUp event with stored key event.
Note that we cannot compare the pointer of first event and following event
directly even though usually both events are same address as far as I checked
because according to the source code of ibus, fcitx and GDK, they use
gdk_event_copy() to keep storing original event. According to the document of
the API, it might just increment refcount. However, the actual implementation
of the API always creates another instance and return it. So, it might be
used same address by arena allocation or something accidentally. Anyway,
we shouldn't compare them. Instead, we need to compare each information of
two key events. Unfortunately, we also cannot compare them simply. Both
ibus and fcitx set unused bits of GdkEventKey::state to true when they send
back the event to us. Therefore, we should compare some of or all of the
members by ourselves. I think that matching time must be enough in most
cases since its value of native key events are properly set. However, for
safer code, this patch also checks type, keyval and part of state.
MozReview-Commit-ID: FZSwN61v0Sd
ibus and fcitx have asynchronous key event handling mode and it's enabled in
default settings. That is, when they receive a key event from application via
a call of gtk_im_context_filter_keypress(), they may post the key event
information to other IME process, then does nothing but store the copy of the
event with gdk_event_copy() and returns true for the result of
gtk_im_context_filter_keypress(). When the other IME process handles the
event, returns the result to them in our process. Then, they send the stored
key event to us again. Finally, they actually handles the event in our process
actually.
Therefore, we may receive every key event twice. So, this causes dispatching
eKeyDown event and eKeyUp event twice. Preceding key event is always marked
as "processed by IME" since gtk_im_context_filter_keypress() returns true
temporarily and following key event is dispatched as expected. So, we need
to ignore the first event only when gtk_im_context_filter_keypress() returns
true but the event is posted to different process.
Unfortunately, we cannot know if the key event is actually posted to different
process directly. However, we can know if active IM is ibus, fcitx or another
one and if ibus or fcitx is in asynchronous key handling mode.
The former information is provided by gtk_im_multicontext_get_context_id().
It returns a string which is set to the IM multicontext instance by creator.
We'll get "ibus" if IM is ibus, get "fcitx" if IM is fcitx.
The latter information is not provided. However, they consider the mode from
env value. ibus checks IBUS_ENABLE_SYNC_MODE. fcitx checks both
IBUS_ENABLE_SYNC_MODE and FCITX_ENABLE_SYNC_MODE.
Additionally, we can know if received key event has already been posted to
other IME process. They use undefined bit of GdkEventKey::state to store
if the key event has already been posted (1 << 25, they called "ignored" flag).
Although their approach is really hacky but we can refer the information at
least for now.
Finally, when we guess a key event is posted to other IME process, let's
IMContextWrapper::OnKeyEvent() not dispatch eKeyDown nor eKeyUp event.
Note that if it's handled synchronously as unexpected, it may causes
dispatching one or more composition events and/or delete content event.
So, in such case, we dispatch a keyboard event for processing key event
anyway. There is only once case we'll fail to dispatch keyboard event.
If we receive signals to dispatch composition events or delete content
command event when IM receives the result from other IME process but
it doesn't send the key event to us. This will be fixed by the following
patch.
MozReview-Commit-ID: 94PrlnmQ3uJ
uim is an old IM which uses key snooper to listen to key events rather than
via filter key event API which should be called by applications. It's still
used by Debian 9.x, so, we still need to support this.
Unfortunately, we cannot detect if uim actually uses key snooper because it's
switch by build option of uim. Currently, Debian builds uim as using key
snooper. So, we should assume uim uses key snooper always. On the other
hand, somebody *might* use uim built as not using key snooper, so, let's
decide if uim uses key snooper with new pref,
"intl.ime.hack.uim.using_key_snooper", but its default should be true.
Note that ibus and Fcitx also have the mode to use key snooper (perhaps for
backward compatibility with uim). However, it's not enabled in default
settings and even if it's enabled, Firefox is in whitelist in the default
settings of them for stop using key snooper. Therefore, we don't need to
support key snooper mode for them unless we'll get some requests to
support their key snooping mode.
MozReview-Commit-ID: 6fTsfKrHzvo
ibus and fcitx usually post key event to other IME process, then, if it causes
some signals to updating composition string, they may not send the posted
key event to us again. Then, IMContextWrapper dispatches neither eKeyDown nor
eKeyUp event since mProcessingKeyEvent becomes non-nullptr only while
OnKeyEvent() is called. So, IMContextWrapper need to store key event if
OnKeyEvent() assumes that given key event is posted to different process.
Then, if IMContextWrapper receives some signals, it should dispatch eKeyDown
and eKeyUp event with stored key event.
Note that we cannot compare the pointer of first event and following event
directly even though usually both events are same address as far as I checked
because according to the source code of ibus, fcitx and GDK, they use
gdk_event_copy() to keep storing original event. According to the document of
the API, it might just increment refcount. However, the actual implementation
of the API always creates another instance and return it. So, it might be
used same address by arena allocation or something accidentally. Anyway,
we shouldn't compare them. Instead, we need to compare each information of
two key events. Unfortunately, we also cannot compare them simply. Both
ibus and fcitx set unused bits of GdkEventKey::state to true when they send
back the event to us. Therefore, we should compare some of or all of the
members by ourselves. I think that matching time must be enough in most
cases since its value of native key events are properly set. However, for
safer code, this patch also checks type, keyval and part of state.
MozReview-Commit-ID: FZSwN61v0Sd
ibus and fcitx have asynchronous key event handling mode and it's enabled in
default settings. That is, when they receive a key event from application via
a call of gtk_im_context_filter_keypress(), they may post the key event
information to other IME process, then does nothing but store the copy of the
event with gdk_event_copy() and returns true for the result of
gtk_im_context_filter_keypress(). When the other IME process handles the
event, returns the result to them in our process. Then, they send the stored
key event to us again. Finally, they actually handles the event in our process
actually.
Therefore, we may receive every key event twice. So, this causes dispatching
eKeyDown event and eKeyUp event twice. Preceding key event is always marked
as "processed by IME" since gtk_im_context_filter_keypress() returns true
temporarily and following key event is dispatched as expected. So, we need
to ignore the first event only when gtk_im_context_filter_keypress() returns
true but the event is posted to different process.
Unfortunately, we cannot know if the key event is actually posted to different
process directly. However, we can know if active IM is ibus, fcitx or another
one and if ibus or fcitx is in asynchronous key handling mode.
The former information is provided by gtk_im_multicontext_get_context_id().
It returns a string which is set to the IM multicontext instance by creator.
We'll get "ibus" if IM is ibus, get "fcitx" if IM is fcitx.
The latter information is not provided. However, they consider the mode from
env value. ibus checks IBUS_ENABLE_SYNC_MODE. fcitx checks both
IBUS_ENABLE_SYNC_MODE and FCITX_ENABLE_SYNC_MODE.
Additionally, we can know if received key event has already been posted to
other IME process. They use undefined bit of GdkEventKey::state to store
if the key event has already been posted (1 << 25, they called "ignored" flag).
Although their approach is really hacky but we can refer the information at
least for now.
Finally, when we guess a key event is posted to other IME process, let's
IMContextWrapper::OnKeyEvent() not dispatch eKeyDown nor eKeyUp event.
Note that if it's handled synchronously as unexpected, it may causes
dispatching one or more composition events and/or delete content event.
So, in such case, we dispatch a keyboard event for processing key event
anyway. There is only once case we'll fail to dispatch keyboard event.
If we receive signals to dispatch composition events or delete content
command event when IM receives the result from other IME process but
it doesn't send the key event to us. This will be fixed by the following
patch.
MozReview-Commit-ID: 94PrlnmQ3uJ
On Linux, dead key is implemented with "table-based input methods" which are
available even on GtkIMContextSimple (i.e., available even in password fields).
Therefore, IMContextWrapper handles dead key sequence as usual composition of
IME. However, on the other platforms, we dispatch "Dead" eKeyDown and eKeyUp
events for dead key.
We started to mark keyboard events which are handled by IME as "processed by
IME" since bug 1343451, i.e., we started to set mKeyNameIndex to
KEY_NAME_INDEX_Process. However, we should keep previous behavior, i.e., keep
setting it to KEY_NAME_INDEX_Dead. Fortunately, GDK's key event tells us
whether the keyboard event is a dead key event with keysym. So, we can detect
if we're in a dead key sequence simply.
MozReview-Commit-ID: Dv336WptfXN
Currently, IMContextWrapper doesn't dispatch eKeyDown event and eKeyUp event
if it's handled by IME. However, for conforming to UI Events, it should
not eat given keyboard events completely.
This patch makes IMContextWrapper dispatches eKeyDown event or eKeyUp event
before dispatching first event of composition events or content command
event.
MozReview-Commit-ID: H2jHpViTH5Q
For conforming UI Events spec, KeymapWrapper::InitKeyEvent() should initialize
mKeyCode and mKeyNameIndex with NS_VK_PROCESSKEY and KEY_NAME_INDEX_Process if
given keyboard event has already been handled by IME.
For making it know if given keyboard event has been handled by IME, this patch
adds additional bool argument to it and its callers.
Note that this patch changes keyCode value and key value of "keydown" event if
it's fired before "compositionstart" since Chromium does so on Linux.
MozReview-Commit-ID: FC3tfyeeopU
Currently, IMContextWrapper doesn't dispatch eKeyDown event and eKeyUp event
if it's handled by IME. However, for conforming to UI Events, it should
not eat given keyboard events completely.
This patch makes IMContextWrapper dispatches eKeyDown event or eKeyUp event
before dispatching first event of composition events or content command
event.
MozReview-Commit-ID: H2jHpViTH5Q
For conforming UI Events spec, KeymapWrapper::InitKeyEvent() should initialize
mKeyCode and mKeyNameIndex with NS_VK_PROCESSKEY and KEY_NAME_INDEX_Process if
given keyboard event has already been handled by IME.
For making it know if given keyboard event has been handled by IME, this patch
adds additional bool argument to it and its callers.
Note that this patch changes keyCode value and key value of "keydown" event if
it's fired before "compositionstart" since Chromium does so on Linux.
MozReview-Commit-ID: FC3tfyeeopU
If a keydown event handler moves focus like Ctrl+PageDown handler, IM context
may be changed to DISABLED or something. In such case, native IME would stop
current composition because focus moving in Gecko causes making IME blurred.
However, IMContextWrapper::DispatchCompositionStart() always dispatches
eCompositionStart even in such case.
So, it should stop dispatching eCompositionStart if IME enabled state is
changed during dispatching the preceding keydown event.
Note that this patch moves the setter of mComposingContext from
OnStartCompositionNative() which is a signal listener of "preedit_start" to
DispatchCompositionStart() because if IME starts composition without
"preedit_start" signal, DispatchCompositionStart() will be called but
OnStartCompositionNative() isn't called. However, this fix needs
mComposingContext.
MozReview-Commit-ID: F3F6NuCOrkJ
When you start new composition during converting with Mozc in e10s mode, the following things occur:
1. Mozc commits previous composition.
2. Gecko dispatches eCompositionCommit event.
3. Mozc sets new composition string (skipping composition start signal).
4. Gecko dispatches eCompositionStart and eCompositionChange event.
5. Selection is changed asynchronously.
6. Gecko sets position of IME windows.
At #4, Gecko stores start of composition as selection start, then, trying to adjust it at #5. However, new selection is caret position in new composition string. Therefore, it's not used for the adjustment. This causes that stored composition start offset is always the start of the previous composition (if the previous patch didn't change EnsureToCacheSelection() behavior). So, IMContextWrapper needs to compute proper composition start offset in this case.
The simplest fix is, modifying selection at #2 as which will be occurred in focused editor. So, this patch makes the selection cache collapsed to the end of committing string.
Note that actual selection may be different if JS changes selection and/or the text in the focused editor. However, it doesn't matter. IMContextWrapper should behave as expected while current composition is active.
MozReview-Commit-ID: 221mDUd8yRP
IMContextWrapper::EnsureToCacheSelection() always queries actual selection when the caller needs selected string. However, this may be expensive and this is bad behavior for the following patch because it wants to emulate selection range until receiving next selection change notification.
Therefore, this patch makes IMContextWrapper::Selection store selected string instead of just its length like other native IME handlers
Additionally, this patch renames IMContextWrapper::mSelectedString to IMContextWrapper::mSelectedStringRemovedByComposition for making the difference between it and the new string in Selection clearer.
MozReview-Commit-ID: 3bygvW7sKf4
Bug 1343075 - 1a. Add TextEventDispatcherListener::GetIMEUpdatePreference; r=masayuki
Add a GetIMEUpdatePreference method to TextEventDispatcherListener to
optionally control which IME notifications are received by NotifyIME.
This patch also makes nsBaseWidget forward its GetIMEUpdatePreference
call to the widget's native TextEventDispatcherListener.
Bug 1343075 - 1b. Implement GetIMEUpdatePreference for all TextEventDispatcherListener; r=masayuki
This patch implements GetIMEUpdatePreference for all
TextEventDispatcherListener implementations, by moving previous
implementations of nsIWidget::GetIMEUpdatePreference.
Bug 1343075 - 2. Allow setting a PuppetWidget's native TextEventDispatcherListener; r=masayuki
In PuppetWidget, add getter and setter for the widget's native
TextEventDispatcherListener. This allows overriding of PuppetWidget's
default IME handling. For example, on Android, the PuppetWidget's native
TextEventDispatcherListener will communicate directly with Java IME code
in the main process.
Bug 1343075 - 3. Add AIDL interface for main process; r=rbarker
Add AIDL definition and implementation for an interface for the main
process that child processes can access.
Bug 1343075 - 4. Set Gecko thread JNIEnv for child process; r=snorp
Add a JNIEnv* parameter to XRE_SetAndroidChildFds, which is used to set
the Gecko thread JNIEnv for child processes. XRE_SetAndroidChildFds is
the only Android-specific entry point for child processes, so I think
it's the most logical place to initialize JNI.
Bug 1343075 - 5. Support multiple remote GeckoEditableChild; r=esawin
Support remote GeckoEditableChild instances that are created in the
content processes and connect to the parent process GeckoEditableParent
through binders.
Support having multiple GeckoEditableChild instances in GeckoEditable by
keeping track of which child is currently focused, and only allow
calls to/from the focused child by using access tokens.
Bug 1343075 - 6. Add method to get GeckoEditableParent instance; r=esawin
Add IProcessManager.getEditableParent, which a content process can call
to get the GeckoEditableParent instance that corresponds to a given
content process tab, from the main process.
Bug 1343075 - 7. Support GeckoEditableSupport in content processes; r=esawin
Support creating and running GeckoEditableSupport attached to a
PuppetWidget in content processes.
Because we don't know PuppetWidget's lifetime as well as nsWindow's,
when attached to PuppetWidget, we need to attach/detach our native
object on focus/blur, respectively.
Bug 1343075 - 8. Connect GeckoEditableSupport on PuppetWidget creation; r=esawin
Listen to the "tab-child-created" notification and attach our content
process GeckoEditableSupport to the new PuppetWidget.
Bug 1343075 - 9. Update auto-generated bindings; r=me
You know, when Korean IME commits string, then it sometimes set next preedit string. So reseting context causes that next preedit string is committed.
So we shouldn't reset IME context with preedit_end.
MozReview-Commit-ID: CZJJvYjcrKY
ibus-pinyin has a bug. When application calls gtk_im_context_reset(), which means selection is changed in application, ibus-pinyin sents a set of composition signals with empty commit string. Therefore, selecting text causes removing it.
For preventing it but not breaking the other IMEs which use surrounding text, we should give up to call gtk_im_context_reset() if IME hasn't retrieved surrounding text after the last selection change. Not having retrieved surrounding text means that the IME doesn't have any cache of contents. Therefore, not calling gtk_im_context_reset() at selection change must be safe for such IMEs.
MozReview-Commit-ID: 5cbIZjpd7zN
TargetClauseOffset is the offset from start composition offset. So we should add start composition to calaculate correct rects.
MozReview-Commit-ID: 4qCpKvw2Eb4
Here is the patched build's log:
[Main Thread]: I/nsGtkIMModuleWidgets GTKIM: 7fab5a60a2c0 CreateTextRangeArray(aContext=7fab5a7bbbf0, aCompositionString="÷" (Length()=1))
[Main Thread]: W/nsGtkIMModuleWidgets GTKIM: 7fab5a60a2c0 SetTextRange(), FAILED, due to no attr, aTextRange= { mStartOffset=0, mEndOffset=1 }
[Main Thread]: W/nsGtkIMModuleWidgets GTKIM: 7fab5a60a2c0 SetTextRange(), FAILED, due to current clause length is 0
[Main Thread]: E/nsGtkIMModuleWidgets GTKIM: 7fab5a60a2c0 SetTextRange(), FAILED, due to g_utf8_to_utf16() failure (retrieving current clause)
[Main Thread]: W/nsGtkIMModuleWidgets GTKIM: 7fab5a60a2c0 CreateTextRangeArray(), inserting a dummy clause at the beginning of the composition string mStartOffset=0, mEndOffset=1, mRangeType=TextRangeType::eRawClause
iBus Chewing IME has two clauses when user presses Shift+p, one doesn't have pango_attr, the other is empty. These clauses are not useful in Gecko. Additionally, TextRangeArray assumes that there is a clause at beginning of the composition when there is one or more clauses. Therefore, this patch tries to insert dummy clause at the beggining of composition in such case.
MozReview-Commit-ID: 3hVGVmvFrhA
I think that we can drop nsIMEUpdatePreference::DontNotifyChangesCausedByComposition(), i.e., nsIMEUpdatePreference::NOTIFY_CHANGES_CAUSED_BY_COMPOSITION because it's now used only by TSFTextStore but TSFTextStore ignores if SelectionChangeDataBase::mCausedByComposition or TextChangeDataBase::mCausedOnlyByComposition is true (for supporting async changes in e10s mode). So, only issue is, dropping the flag might cause increasing computing TextChangeData cost during composition in TSF mode. However, now, it's already enough fast and even if it'd cause performance regression, we could add a hack with TextComposition's offset information. Therefore, we don't need to worry about the performance regression so seriously.
MozReview-Commit-ID: HNT3G4isONj
Currently, all widgets request selection change notifications to IMEContentObserver. Additionally, IMEContentObserver needs to listen selection changes for caching latest selection for eQuerySelectedText. Therefore, it doesn't make sense to keep defining nsIMEUpdatePreference::NOTIFY_SELECTION_CHANGE.
If widgets didn't need selection change notifications, they could just ignore the unnecessary notifications.
Note that all widgets don't need selection change notifications if a plugin has focus and IMEContentObserver cannot observe selection changes in the plugin. Therefore, if IMEContentObserver is initialized with a plugin, it shouldn't listen selection changes (and doesn't need to notify widgets of selection changes).
MozReview-Commit-ID: FOVFFgA2nOz
Selection cache might be updated on commit string, so it should be updated correctly. Also, IMContextWrapper::OnSelectionChange() already has the check for composing using IsComposing().
MozReview-Commit-ID: 2n3f3I3aAjg
For making our code clearer by the stronger type check, we should change the anonymous enum for NS_TEXTRANGE_* to enum class whose name is "TextRangeType" and whose type is "RawTextRangeType" which is an alias of uint8_t.
Additionally, this also adds some utility methods for them.
Note that some lines which are changed by this patch become over 80 characters but it will be fixed by the following patches.
MozReview-Commit-ID: 76izA1WqTkp
It's not clear to me what NOTIFY_IME_OF_COMPOSITION_UPDATE means only from the name. For making the name clearer, this patch renames it to NOTIFY_IME_OF_COMPOSITION_EVENT_HANDLED and add some explanation to the definition.
MozReview-Commit-ID: 8ySYCNJ1Ytz
