* Rename KeyboardActionListener.onPress to onPressKey
* Rename KeyboardActionListener.onRelease to onReleaseKey
* Merge KeyboardSwicther.onPressShift, onPressSymbol, and onPressOtherKey to onPressKey.
* Merge KeyboardSwitcher.onReleaseShift and onReleaseSymbol to onReleaseKey.
* Merge KeyboardState.onPressShift, onPressSymbol, and onPressOtherKey to onPressKey.
* Merge KeyboardState.onReleaseShift and onReleaseSymbol to onReleaseKey.
Change-Id: Icf28fd18e238c5e534c292893e4ab5b6b98e72f8
The basic idea is that in some situations, we would restart
suggestions, but before evaluating them we would bail out
because suggestions were not requested. This would lead to
the "correction" set to null, so we would commit a null word.
This fix does two things:
- Do not restart suggestions when not requested.
- If we still end up with a composing word when suggestions
are not requested, we select the typed word as the
correct version.
Bug: 5846646
Change-Id: Ic35351841d0cb20afa99092ef681ecb7bd68bec6
The method deleteWordAtCursor would call finishComposingText,
which is a very unexpected state change and would probably cause
surprising bugs to any parent that would call it.
As it happens, it's not used anywhere any more so let's just
remove it.
Change-Id: Iaa9200866e40b4e914f76baa987279948300679c
We need revised scoring test because dictionary look up algorithm has
been changed significantly since this SuggestTests was made.
Change-Id: I69d527a6bfa24ed0dc189fddad1afab0566c6eb8
Check if any character is both a magic space swapper and
a magic space stripper for the current language, and throw
an exception if found. Since this is expensive, it's done
only in debug mode.
Change-Id: Ibd166db87c91495b76878ea0e4f420a62c7bb276
This would crash the program when inputting a punctuation
symbol that should not swap with a weak space through the
suggestion strip, then press backspace.
Those include all two-part punctuation symbols in French for
example, as well as magic space strippers like the dash in
English. For English all suggestion strips symbols happen to
be magic space swappers so this would not happen in English
by default.
Also fix a typo.
Change-Id: Ia40678c264e978509b5786290a75a3c57181648e
This was introduced by a badly-rebased commit. The culprit was
I103d6851. There was also another bug lurking there introduced
by a previous commit that this fixes.
Bug: 5775347
Change-Id: I133b54a5159e19714f4bcae8ffbb4824cff9f466
Also this change moves LatinKeyboard attributes,
autoCorrectionSpacebarLedEnabled, autoCorrectionSpacebarLedIcon,
spacebarTextRatio, spacebarTextColor, and spacebarTextShadowColor to
LatinKeyboardView.
Change-Id: I7cc27ce1fc550e9f620a9ed0fbe4b8172902d5a1
This is the last one (whew)
Also move a somewhat unrelated initialization out of the
input attribute initialization function.
Change-Id: Icc77b32fd16b112ab43786715061b1c9dca31c04
The ultimate intent is to make these variables final,
which will allow to move the computing function to the
InputAttributes class.
Change-Id: Ifb1fc7321a19b10d60aa63fa0daaa2708deda2cf
...instead of the hard-to-understand mHasUncommittedTypedChars.
This is possible because now the word composer is actually aware
of commits.
Change-Id: I36b664ce8402a280f801e87b9ebe161f416b0853
Currently, these variables hold the info about the composing
word, or maybe some outdated info, and it's not very clear
which it is. LatinIME is maintaining the freshness info in
a separate boolean, and uses it throughout the code for many,
many things, leading to much confusion.
The idea in grouping this info is, it can be saved in another
instance and restored later. It can be tested against to know
whether there is actually outdated but kept info or not, and
it should allow to straighten out what is actually currently
being typed. Ultimately, it will eliminate the need for
LatinIME to keep track of the status of the info in the
word composer.
Change-Id: I00e2c690f303f8320c9be35590a6df4583e9e456
Also rename the function to reflect better the behavior.
An upcoming change will clean up the remnants
Change-Id: I19bdf0d9e95f62ee749e1f64fe17132c4a5ef0bc
This change introduces KeyboardSet and SubKeyboard XML definitions to
represent a set of keyboard layouts.
Bug: 5002108
Bug: 5679585
Change-Id: Ib6c8d5936187381bb6725c9fe574e93871c01a86
Upon autocorrection, there is visual feedback in the text
view that quickly flashes the background of the text that
just changed. This fixes a race condition that happens
upon autocorrection when typing fast, and that results in
flashing one character too far left (typically, the flashing
area includes the whitespace before the corrected word and
not the last character of the corrected word).
This happens because the call to commitCorrection may
happen before or after the IPC sent by sendKeyChar, but the
arguments are fit only for the case where it arrives first.
Change-Id: I9b5442a665aad5a9bc66cd49228075b9056b37fa
This place is very confusing because a member variable with
a consistent meaning is hijacked for the duration of a function
to mean something else. This is in the way of easy-to-understand
refactoring.
Change-Id: If79bc771950d6bfc0ad5f0e9c51c7ef1dbb45b66
- Stop the word composer from escaping - take a page from the law
of Demeter and only report what is actually needed.
- Fix typos in comments.
- Add a comment for a fishy processing.
- Remove a useless local variable.
Change-Id: I5fa78901cbb5483fc9683bfb7094f47244b85df6
mBestWord has a confusing name - it's actually an auto-correction.
It's cleaner if it lives in the word composer because an
auto-correction should be tied to a specific user input, and
should be reset each time the user input changes to avoid
race conditions.
Change-Id: I718d29395bc747372067e6440e090c6a181994ae
Both tests have the same meaning. They should be merged.
Also, if "deleteLast()" deletes more than one char (which
never happens in the current implementation, but it's not
guaranteed), the new code is more correct.
Change-Id: I216df7cd45f4a7d76d1d5d6aa7f8d7f9a6e35ea3
If mHasUncommittedTypedChars is true, then
mWordSavedForAutoCorrectCancellation must always be null, and
spaceState must always be SPACE_STATE_NONE.
Hence, this change is supposed to be very no-op.
Change-Id: I2716f7372d996cc4dec40a4b7ac1ee7f5f4afe91
postUpdateShiftKeyState used to be called also when
mEnteredText is not null => this is a bugfix.
The rest does not change the logic, as posting a message can
be done anywhere within the function with no impact.
Change-Id: I7888797c0778702d64f96701e35b611a55a6a259
If there are no uncommitted chars, we shouldn't enter any of
the tests that follow. We didn't use to, but a change today
made it possible - it should not happen.
There is no point in doing the rest of the tests, they are
sure to fail.
Change-Id: I580dd104aff3585de72a93b38989bfd9713f615b
This also includes a new memorized string with simpler
logic that will allow to remove other members
Change-Id: I0b97243084902e68fcb835b6163c86fef8190ebc
Reorder so that they match the order in the preference xml
file. Also add TODOs for missing raw variables.
Change-Id: I06332c131a624d6c359512d75c808f3f7313fa91
The ultimate intent is to close it up in the settings, as it does not
change while typing, only when the keyboard is open again.
Change-Id: I12ff56e7482c1584877fe5531473bf03e95a60f6
These calls, setAlphabetKeyboard, setSymbolsKeyboard, and
setSymbolsShiftedKeyboard have to be call backed from KeyboardState.
Bug: 5708602
Change-Id: Ibbe1a21bd10bf942e17886869c0ab0fa1735b87e
This also tries to make the code as easy to extend as possible
for future developments.
Bug: 5701241
Change-Id: I1ed48e6a5cc7aab94c5d6e309930cc004247d7e7
Note that this is not enough: we still need to create a
reasonable proximity table for Cyrillic characters, or we
won't be able to show up suggestions.
Bug: 5701241
Change-Id: Idb141f7a230a6e1a46094308c26f43c01ab3b97a
I14dd3815 special-cased single quotes at the start of a word, but
did not do so correctly - it would kill a composing word if entered
just after some punctuations.
The right test was not isCursorTouchingWord() but
mHasUncommittedTypedChars, which actually makes for a simpler
implementation.
Bug: 5648032
Change-Id: Icd21d213e0cad6da68cacfeb921502cd8dcfac95
Single quote at start of word is not considered a part of a
word any more.
Single quote at the end of a word now behave like capitalization:
lookup in the dictionary is done *disregarding* a final quote,
and it is forcefully added back into the suggestions afterwards.
Bug: 5566368
Change-Id: I14dd3815f4b743edba56d64a3abdf4b73d863a6a
This introduces an elementary undo facility that accounts
for magic space, strong space, double space and its cancelling,
and swapped punctuation and its cancelling.
The former three were existing behavior ; this change adds
the swapped punctuation cancelling behavior.
Bug: 5454442
Change-Id: I87af633f30caa2788e9af725e556a7f0746d9a14
The spell checker cannot afford to return static objects,
seeing as the framework will then use the same objects to
pair the cookie and sequence ids to the request.
This is a backport of Ia9c3a933
Bug: 5503243
Change-Id: Ia9c3a933bfb30cf5525418b240ef60632d72c9d0
Remove a state that can never be reached, and ajust all the
callers that would always have received false.
Change-Id: Iac025568be11743428419e0772da306a4f0a0bf1
On space swap or space removal in some occasions there would be
a small blink on the screen. This change fixes that.
Change-Id: I486cbcc5dfcafd531b3fa92c4f9a3255832f22cc
The spell checker cannot afford to return static objects,
seeing as the framework will then use the same objects to
pair the cookie and sequence ids to the request.
Bug: 5503243
Change-Id: Ia9c3a933bfb30cf5525418b240ef60632d72c9d0
This bug would kill the case where the whitelist contains
a word to be autocorrected to an uppercased version of
itself, and the user would enter the uppercase version.
In this case, this bug would cause the typed word to be
killed off the list of candidates, and possibly autocorrected
to the *next* candidate.
When the whitelist checks whether this the typed word is
a candidate for whitelisting, this change has it check whether
the whitelisting results in the typed word before returning.
Hence, it can keep the case-insensitive behavior of the
whitelist.
Coincidentally, this change renames the method used to do
this, because it does not comply with the general contract
of Dictionary. This happens to be in the way of another
upcoming change.
Bug: 5420371
Change-Id: Ifb305271acc5f171adf9b18c762ae7975b14be0a
When long press triggers caps lock, the keyboard also triggers haptic
feedback.
This change also fixes trivial harmless bug in KeyboardSwitcher.
Bug: 5424681
Change-Id: I62706b49abd7be1dcebc3c5166ea03f426fc8c86
This goes together with I6b8628b9acc32449e4147a2a754b222fbb76c754
or it will break the build
Bug: 5402436
Change-Id: I07c6266b713773a8de80bb22afdd4c566261f78a
This removes the calls, but another change will be needed to remove
the messages LatinIME used to send itself to update the suggestion
strip.
Bug: 5402537
Change-Id: I5d1aa63a892516f339f3ceac21f43771b5ffda34
This change updates suggestions when the cursor is moved.
It is now reasonable to remove the explicit test for
TextEntryState.isAcceptedDefault because it is now shielded
by mExpectingUpdateSelection : actually, this probably fixes
a long-standing bug.
Bug: 5337309
Change-Id: Iee4046420c6a88d1a07d428230f93c3ebef25c39
In effect, this stops the spell checker from suggesting overly
long words.
More precisely, it takes advantage of the new facility that
takes into account the whole length of the dictionary word when
computing scores, so words much longer than the input word will
see their score demoted accordingly.
Bug: 5384578
Change-Id: I326cd7c87c3080e7fa8729f78517f8ba13672a9b
IME is called back four methods for each input field as a IME life
cycle. The four methods are onStartInput, onStartInputView,
onFinishInputView and onFinishInput.
After orientation changed, Those quartet methods will be called back
twice. This behavior of the framework might be a bug.
In order to restore the previous keyboard layout, we should skip
onFinishInputView and onFinishInput of the first quartet and
onStartInput and onStartInputView of the second quartet.
Bug: 4311428
Change-Id: I450ddc0cce5d00abc971ffd42a507a8a86682548
It now follows the following logic:
- If the word should be filtered out => false
- Else => !IN_THE_DICTIONARY
This defines the behavior for ICS MR0, and prepares for addition
of a new HAS_LIKELY_SUGGESTIONS flag in MR1.
Bug: 5383800
Change-Id: I530b1404ae8cf3337ff68ef5ab0f4d95f2dad0e8
The gutter area between the suggestions strip and the top-row keys
looks like a part of the suggestions strip, and the touch events
landing on the area should be forwarded to the suggestions strip.
Bug: 5246673
Change-Id: I92af763be0feed21aa36ceffb5d575abe554f19e
CL https://android-git.corp.google.com/g/#/c/136474 refactored
the fullscreen test, but ithe IME_FLAG_NO_EXTRACT_UI test was
lost in the process.
Note that there is still a problem (orthogonal to that change
with key_preview_backing, which appears opaque and black sometimes.
I'll re-open 5315001.
Change-Id: If3a73179d21eaca10bdc948db7bac4b4f7a88d34
Because "TRHEE DOT LEADER" glyph of Roboto font is broken, we decide
to use "..." instead.
Bug: 5344295
Change-Id: I0fa5eefc00baf80747ff5215c018619a7e15a58e
This change also separates controlling visibility of "key preview
backing view" from suggestions strip visibility.
Bug: 5315001
Change-Id: I190a71f7956c804e5d89d2d5bacecc62d565ac2c
This is a follow up change of If6c0edef.
This is a cherry-pick of Idb415f53 from Master.
Bug: 5328922
Change-Id: I36d8bda9fb95e4809598296226c598a9f08bd8bb
This should not be used lightly, as it violates the general
contract of locale, and does kill some legitimate (albeit
alledgedly rare) use patterns.
Currently, the spell checker uses this because it uses a
negative logic: it should match broadly, and when in doubt,
match even more broadly, which is almost never the case of
something that uses the locale.
In other words: don't use this option unless you are
very, VERY sure that's what you want. Hint: it isn't
Bug: 5280929
Change-Id: Ib3cae319c692161d653630038c5bcde1f4340c05
There is no definite path known for this to end up being
touched by other classes, but we could imagine through
some way or some other it ends up shoved in the stringbuilder
pool, leading to catastrophic results.
Hopefully related to
Bug: 5248688
Change-Id: Ib8abfc31263cbf31d515ed607ced5d8253971938
If the spellchecker encounters a bug and happens to crash,
it may be sensible to avoid killing the keyboard in response.
This is a possible way to do it, which comes with the big
drawback of making bugs in the spell checker harder to find.
Change-Id: Idb26fb592b9718e1dbdadeda8fbd1a8a1d805c28
Note that this affects only the results of the spell checker if
actually passed such a word. For example, the spell checker will
not flag "http://oju" as a typo, because it looks like a URL.
But in the current implementation, TextView passes "http" and
"oju" separately, so "oju" is still flagged as a typo.
Bug: 5281875
Change-Id: I9d721fd3af34edc51b11908cf2e8fe994b164242
The locale is specified by KeyboardLocale extra key in method.xml,
LatinIME will use the specified locale for keyboard layout.
Bug: 5238658
Change-Id: I8e6cb66c73a7ac1bf611d9910b42fa9cff38eba0
Autotext correction would check whether the first suggestion
so far was the same as what Autotext would return, and if it
was indeed the same, would not send its result as
autocorrect. However, the first suggestion is not guaranteed
to have a high enough score to trigger autocorrection, and
there would be cases where a word in autotext would not get
autocorrected because the word came out of bigram
suggestions. These occurrences would be extremely rare, as
they would require concomitant insert between autotext for
one char and bigram suggestion. It is, in fact, probably
limited to the capitalization of "I".
This did not happen in gingerbread because gingerbread would
not register 1-letter words as valid bigrams.
This fix works by just always sending the result of autotext
regardless of whether it is already the first suggestion or
not. This is okay because duplicates are removed afterwards
anyway - and this processing is absolutely necessary because
the autotext'd word may actually be somewhere else in the
suggestion, so it made really no sense checking for only the
first one.
Please note that there is also a race condition that can
result in "i" not being converted to "I": at the moment,
Latin IME relies on having the suggestions evaluated
at the time autocorrection is performed, but when typing
very, very fast, those messages may have been canceled.
This is not limited to the autocorrection of "i", but
affects all autocorrections. It requires a nearly
inhumane typing speed to trigger, but hitting "i" and
space in turn as fast as one can it's possible to
reproduce occasionally.
Bug: 5135113
Change-Id: I530ea6212487300001a2c0fc5b25a5c7716bdf63
Checking the magic number of a file upon decoding is necessary,
because if the file is corrupt and we don't check it, we will
fall back to a simple copy of the corrupted file. Latin IME
would realize this and would not crash, but would not use the
corrupted dictionary. If this happened to be a main dictionary,
then the user would lose the ability to use the correct
built-in dictionary.
Not the same, but kinda similar to
Bug: 5223031
Change-Id: Ic2783dc9dd5f3dcf2865623d9452765fe3778db7
This fixes a race condition that would end up with the spell
checker not finding some words in the user dictionary when it
just booted.
Bug: 5194627
Change-Id: I1ba911cc53e6ae3b111d54a6f91d1d5feef3f5de
The message this removes gets printed under normal conditions.
Normally dictionary files are compressed then crypted, but not
compressed a second time; however LatinIME tries to open a
compressed-crypted-compressed file first, because it could not
do it afterwards and we want to support this case. So under
normal operations, the first method LatinIME tries is actually
expected to fail.
Also, if we decide to stop compressing or/and encrypting dicts
LatinIME supports it as a valid use case. It should not print
errors to the log.
If the file cannot be open at all, then it is an unexpected
case, and Latin IME still reports to the log.
Change-Id: Ic5228c51365a101af1d03e2c893484d3050b5a1c
This change also tweaks more_suggestions_hint drawables to eliminate
unnecessarily padding at the bottom.
Bug: 5204563
Change-Id: If5ec731c9a490da0116a020e8fd405c7d025d707
This change also introduces "AsciiCapable" extra value for subtype
that indicates the subtype can input any ASCII code point from its
keyboard layout.
Only if the input field has "forceAscii" private IME option and the
current subtype doesn't have "AsciiCapable" extra value, the en_US
keyboard layout will be used for the input field.
Bug: 3384942
Change-Id: I25e2553e37ecb5002df1164e45f6273845fe463b
LatinIME has several fallbacks to ensure a dictionary is used.
However, it should now be possible to deactivate the main dictionary
through the dictionary pack settings. In this case, Latin IME should
not fallback to the built-in dictionary.
Change-Id: Ibd992ad89793169f04a968b40781fce819b87b6f
This also adds some optional debug code to detect more easily possible
future occurrences of the same problem.
Bug: 5195017
Change-Id: I2558b468e46f7090de868f1ec2dc9e24895d670f
The new behavior is as follows:
- If the word in the dictionary is not fully lower case, then the
exact case is required to match.
- If the word in the dictionary is fully lower case, then any of
the following patterns match:
- fully lower case
- only the first char capitalized
- all caps
Any other capitalization is rejected.
This is probably what people want. If you type a name in all lower
case, it should be marked as a typo, but if you type a word with a
capital for emphasis or just because it's the start of the sentence,
it should match a lower case word in the dictionary. If you have
a spurious capital letter in the middle of a word because of a typo,
it should be marked as such.
Accents are not affected, and should not be. An accented letter
is a different letter and a missing accent should be reported.
We should maybe consider again for some common transpositions
like the "ue" digraph for German, which is now considered a typo,
but will suggest the correct diacritics as the first suggestion.
Bug: 5145751
Change-Id: I651e24f13c90fb94700a1674ad380e95336e7dca
by ignoring its preference value, when the "show settings key option" in
the settings is disabled.
Change-Id: I934be2d9eb0bf6bf731e03f5d9b752de4a7a1c7c
This change refactors the dictionary selection code so that the
cached dictionary files list and the settings tests are more
cleanly separated.
This will also help with future refactorings that will test for
the presence of the main dictionary and insert the fall back if
it's not supplied by the dictionary pack.
Bug: 5095140
Change-Id: I8d7caad7c054031df71fe78b043801a774d50f65
The next step is to move this over to the Utils class.
The ultimate goal is to make use of the duplicate removal code
also in the spell checker as per
Bug: 5175740
Change-Id: Ica36691b843b0713b832c56ffc65e5b2ec427c4a
Now that the dictionary pack can return several files, it's better
to handle IO exceptions for each file rather than globally. This
also will help with next implementation steps.
Bug: 5095140
Change-Id: I5ed135ad2ad4f55f61f9b3f92c48a35d5c24bdb2
The dictionaries and proximities are not thread-safe. In order to
be able to check spelling in parallel, make a dictionary pool to
call upon when a spelling check is necessary.
Bug: 5156851
Change-Id: Ie3796164187dd7b7abf5ccd5d014073d43d74408
Until the dictionary itself is re-entrant, unfortunately we have
to synchronize accesses to it.
Bug: 5156851
Change-Id: I1629a160d707e6fa1ad80cba828798abb24b68b1
Dictionaries are now copied over from the dictionary pack to Latin IME.
This change enables Latin IME to use all dictionaries that have been
cached until now.
Bug: 5095140
Change-Id: Id9a2bacf9dc1c693189b0ac8aa3f75756dc1e3e6
Using only a single thread may slow down the application, if there is already another thread running.
Bug: 5142142
Change-Id: I314480a58af20ea09458ff82e586304fc3c7c070
A recent change had the contacts reloaded every time a new field
is touched. This change not only fixes the problem, but also removes
reloading contacts when changing language, which should make language
switch within LatinIME lighter.
Bug: 5125034
Change-Id: Ia61c4f75a8617113cdce88a2e2c6fdf073146a2d
The dictionary filename used to be passed directly to Latin IME.
This change implements, on the part of Latin IME, the passing of
them as an id that should then be passed through
openAssetFileDescriptor.
Bug: 5095140
Change-Id: I7d1e9d57c19f0645045368f68681680f238189fc
This does not really fix the underlying bug, but it does fix
the apparent symptoms.
When the user presses space and a letter quick, the
onUpdateSelection handler may be called after the letter has been
actually committed. The keyboard then happily proceeds to clearing
the composition because it thinks space was pressed (or the user
moved the cursor, since it can't guess which happened).
This change removes this behavior when we are expecting an update
event from a keypress. This means the bug still exists if the user
presses space twice and a letter, and all events come after the
letter, but it is very very hard to reproduce this. There may be
other collateral damage when the user moves the cursor in the form
of race conditions, but likewise, they should be really hard to
reproduce.
Bug: 5100521
Change-Id: Ib05328c9b451bf6fe288ae00296fd283a9a4e863
The "deleteChar" argument of this method is acutally always true in the current code path.
Also, in this method, textToTheLeft virtually never has a punctuation letter as its initial
character. This change is one of preparatory changes for bug:4983945.
Change-Id: I531a32570a35634c21c1d74b2b461e40a1b7f660
This should help with LatinIME opening performance.
It should not be merged before Ia9e4741f (if it is, there is a
race condition where the wrong dictionary is opened).
Bug: 5023141
Change-Id: I39c52de70a449929a622dddfdd92b38425ad9416
This change removes everything related to the quick fixes
Latin IME option item.
Note that Autotext is still used if the current subtype language
is the same as the system language.
Bug: 4985058
Change-Id: I51728a2190543ffb9e004470674ef6257b14f6c2
