EditorBase stores a text node, offset in it and length in it of composition
string directly. However, this wastes memory space if user never uses IME
or user only sometimes uses IME. Additionally, storing all data in
TextComposition is better than current design when other classes like
CompositionTransaction wants some information of both EditorBase and
TextComposition.
This patch moves those 3 members from EditorBase to TextComposition.
MozReview-Commit-ID: 4N7wmGGfxmt
According to the log in crash reports, eCompositionCommitRequestHandled is
sent to ContentCacheInParent twice or more for a composition. This causes
breaking mPendingCompositionCount and mPendingEventsNeedingAck management.
Currently, nsIWidget::NotifyIME() should be called only by
TextComposition::RequestToCommit(). Therefore, the method should manage if
it should request it actually. If the composition has already received
eCompositionCommit(AsIs) event, it shouldn't request it because parent process
may have already stated new composition and it shouldn't be broken by request
for old composition.
MozReview-Commit-ID: 2ekSa6EIeRP
When TextComposition::RequestIMEToCommitComposition() is called but the request is handled by native IME asynchronously, TextComposition synthesizes eCompositionCommit(AsIs) event to commit composition synchronously. (This behavior helps web apps which tries to commit composition from JS.) Then, TextComposition::DispatchCompositionEvent() should ignore following composition events which are actually sent from native IME.
However, if composition is in a remote process, TextComposition::DispatchCompositionEvent() in the main process doesn't discard the following events coming from native IME. Therefore, ContentCacheInParent is confused by this unexpected events. So, TextComposition::DispatchCompositionEvent() shouldn't send the following unnecessary composition events to TabParent.
Note that in remote process, composition is committed synchronously with the hack mentioned above and synthesized eCompositionCommit(AsIs) event of TextComposition::RequestIMEToCommitComposition() in the parent process looks like coming from native IME. Therefore, following composition events coming from native IME are not necessary for the remote process.
MozReview-Commit-ID: 2uKFxcAq88J
IME should receive notifications and requests only from proper process. E.g., IME shouldn't commit composition by a request which came from previous focused process.
This patch makes that IMEStateManager::NotifyIME() takes pointer to TabParent optionally. If the request or notification came from remote process, it should be non-nullptr. Then, this makes it ignore notifications and requests from unexpected process.
Note that this patch also touches some gfx headers because they use |ipc::| but compiler is confused at the ambiguousness between |mozilla::ipc::| and |mozilla::dom::ipc::|.
Finally, this patch changes the NS_ASSERTION in IMEHandler::OnDestroyWindow() to MOZ_ASSERT because the orange caused by the NS_ASSERTION was not realized since there was already an intermittent orange bug caused by different NS_ASSERTION.
MozReview-Commit-ID: 9CgKXQRJWmN
TextComposition in the main process is destroyed when the main process sends eCompositionCommit(AsIs) to focused remote process. Therefore, ContentCacheInParent::mCompositionPendingCount is never 2 or more now.
It may cause ContentCacheInParent::Assign() setting older composition's start offset to current composition's start offset in the main process.
For making uplift the following patch easier, the wrong patch should be backed out first.
MozReview-Commit-ID: IHWc7qZBQtc
ContentCacheInParent::mPendingCompositionCount is now managed with composition events which TabParent received. However, TextComposition doesn't dispatch composition events after coming request to commit active composition. Therefore, composition is committed forcibly in a remote process over 255 times, the main process crashes.
It's the safest way to use TextComposition to manage ContentCacheInParent::mPendingCompositionCount.
MozReview-Commit-ID: DEhzYcK1zcW
According to ATOK's behavior, IME may send different line breaker from its platform's standard. Therefore, we should treat \r as \n too.
Additionally, currently, TextComposition doesn't allow to input \n with composition. However, this was added for preventing to see odd control characters as boxes with code point. Therefore, we should allow \n for IMEs. (It was allowed, this limitation is unexpected when I reviewed the patch to reject control characters in TextComposition.)
MozReview-Commit-ID: DzGSMgp89Av
TextComposition::GetSelectionStartOffset() tries to query the start offset of the composition if there is a composition. Therefore, it tries to query first clause when it has some clause information of the composition string. However, even if the composition string is empty, TextComposition::mRanges which refers CompositionEvent::mRanges of the last eCompositionChange event may have empty clause information. In such case, this method tries to query not existing IME selection range. Then, this returns UINT32_MAX and ContentCacheInChild will query character rects around the offset and ContentEventHandler::SetRangeFromFlatTextOffset() will meet odd case.
For avoiding this issue, TextComposition::GetSelectionStartOffset() should query normal selection when composition string in the editor is empty. Note that TextComposition::mString will be updated after the editor completes updating its DOM contents. Therefore, TextComposition should refer mLastData which is set before dispatching eCompositionChange in the tree.
MozReview-Commit-ID: F2UGCQLXLSz
TextComposition should update its composition start offset after every DOM event dispatch with first clause's selection type if there is composition string.
MozReview-Commit-ID: HFkePci1PhU
TextComposition queries selection start offset a lot of times. Therefore, for reducing the runtime cost, it should use IMEContentObserver if it's available or ContentEventHandler, otherwise.
MozReview-Commit-ID: 61GgQZDX2HP
When composition string hasn't been non-empty, insertion point of the composition string can be changed by a DOM event handler. E.g., compositionstart, first compositionupdate and first text. Therefore, TextComposition should update the composition start offset cache after every event dispatch.
MozReview-Commit-ID: FOPewPTRuCn
It's enough to store target clause offset from start of the composition and better to modify mCompositionStartOffset because when even if mCompositionStartOffset is changed, we don't need to modify the target clause offset.
This patch renames mCompositionTargetOffset to mTargetClauseOffsetInComposition.
MozReview-Commit-ID: 1wt2OTUUjkY
Currently, when TextComposition tries to forcibly commit composition synchronously, it cancels the composition if there is only an ideographic space since legacy Chinese IMEs for Windows were used an ideographic space as a placeholder and shows actual composition string in its owning window (called reading window).
However, Japanese TIPs basically use composition to input an ideographic space. Unfortunately, this intentional input of an ideographic space is always canceled if an editor commits to composition at every input event in TSF mode because TSF cannot commit during a call of ITextStore::RequestLock(). Additionally, we will enable e10s mode, then, on all platforms, requesting commit composition is handled asynchronously.
Therefore, we should make the hack disabled in default settings now. If we'll find a way to distinguish if an ideographic space is a placeholder, we should recover this hack. Note that such input fields cannot handle such legacy IMEs, so, disabling the hack in default settings must be fine.
MozReview-Commit-ID: IdBcfBxeJum
For sending NOTIFY_IME_OF_COMPOSITION_EVENT_HANDLED after the other change notifications which was caused by the user input, we need to use IMEContentObserver::IMENotificationSender because it sends the notifications when it's safe to do it.
This patch makes TextComposition use IMEContentObserver to send the notification. However, if there is no active IMEContentObserver, e.g., composition events are fired on unfocused window, TextComposition sends it by itself (same as current implementation).
If IMEContentObserver stops observing when it has pending NOTIFY_IME_OF_COMPOSITION_EVENT_HANDLED, it cannot send the notification (i.e., it is discarded completely in such case). However, in such case, IMEContentObserver sends NOTIFY_IME_OF_BLUR. So, anyway, native IME handler should treat the blur notification as it including NOTIFY_IME_OF_COMPOSITION_EVENT_HANDLED.
On the other hand, we're buggy if composition events are fired in non-active window. Even in such case, IMEContentObserver should be created for active editor in each document and it notifies IME of the changes. But this is out of the scope of this bug.
MozReview-Commit-ID: 7Q0ZsJTh4hX
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
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
