You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
koreader/plugins/japanese.koplugin/main.lua

248 lines
11 KiB
Lua

--- Japanese language support for KOReader, modelled after Yomichan.
-- This plugin extends KOReader's built-in dictionary and selection system to
-- support Yomichan-style deinflection and text scanning, allowing for one-tap
-- searches of inflected verbs and multi-character words and phrases. As such,
-- this plugin removes the need for synonym-based deinflection rules for
-- StarDict-converted Japanese dictionaries.
--
-- @module koplugin.japanese
-- @alias Japanese
-- Copyright (C) 2021 Aleksa Sarai <cyphar@cyphar.com>
-- Licensed under the GPLv3 or later.
--
-- The deinflection logic is heavily modelled after Yomichan
-- <https://github.com/FooSoft/yomichan>, up to and including the deinflection
-- table. The way we try to find candidate words is also fairly similar (the
-- naive approach), though because dictionary lookups are quite expensive (we
-- have to call sdcv each time) we batch as many candidates as possible
-- together in order to reduce the impact we have on text selection.
local Deinflector = require("deinflector")
local LanguageSupport = require("languagesupport")
local ReaderDictionary = require("apps/reader/modules/readerdictionary")
local UIManager = require("ui/uimanager")
local WidgetContainer = require("ui/widget/container/widgetcontainer")
local logger = require("logger")
local util = require("util")
local _ = require("gettext")
local N_ = _.ngettext
local T = require("ffi/util").template
Clarify our OOP semantics across the codebase (#9586) Basically: * Use `extend` for class definitions * Use `new` for object instantiations That includes some minor code cleanups along the way: * Updated `Widget`'s docs to make the semantics clearer. * Removed `should_restrict_JIT` (it's been dead code since https://github.com/koreader/android-luajit-launcher/pull/283) * Minor refactoring of LuaSettings/LuaData/LuaDefaults/DocSettings to behave (mostly, they are instantiated via `open` instead of `new`) like everything else and handle inheritance properly (i.e., DocSettings is now a proper LuaSettings subclass). * Default to `WidgetContainer` instead of `InputContainer` for stuff that doesn't actually setup key/gesture events. * Ditto for explicit `*Listener` only classes, make sure they're based on `EventListener` instead of something uselessly fancier. * Unless absolutely necessary, do not store references in class objects, ever; only values. Instead, always store references in instances, to avoid both sneaky inheritance issues, and sneaky GC pinning of stale references. * ReaderUI: Fix one such issue with its `active_widgets` array, with critical implications, as it essentially pinned *all* of ReaderUI's modules, including their reference to the `Document` instance (i.e., that was a big-ass leak). * Terminal: Make sure the shell is killed on plugin teardown. * InputText: Fix Home/End/Del physical keys to behave sensibly. * InputContainer/WidgetContainer: If necessary, compute self.dimen at paintTo time (previously, only InputContainers did, which might have had something to do with random widgets unconcerned about input using it as a baseclass instead of WidgetContainer...). * OverlapGroup: Compute self.dimen at *init* time, because for some reason it needs to do that, but do it directly in OverlapGroup instead of going through a weird WidgetContainer method that it was the sole user of. * ReaderCropping: Under no circumstances should a Document instance member (here, self.bbox) risk being `nil`ed! * Kobo: Minor code cleanups.
2 years ago
local SingleInstanceDeinflector = Deinflector:new{}
Clarify our OOP semantics across the codebase (#9586) Basically: * Use `extend` for class definitions * Use `new` for object instantiations That includes some minor code cleanups along the way: * Updated `Widget`'s docs to make the semantics clearer. * Removed `should_restrict_JIT` (it's been dead code since https://github.com/koreader/android-luajit-launcher/pull/283) * Minor refactoring of LuaSettings/LuaData/LuaDefaults/DocSettings to behave (mostly, they are instantiated via `open` instead of `new`) like everything else and handle inheritance properly (i.e., DocSettings is now a proper LuaSettings subclass). * Default to `WidgetContainer` instead of `InputContainer` for stuff that doesn't actually setup key/gesture events. * Ditto for explicit `*Listener` only classes, make sure they're based on `EventListener` instead of something uselessly fancier. * Unless absolutely necessary, do not store references in class objects, ever; only values. Instead, always store references in instances, to avoid both sneaky inheritance issues, and sneaky GC pinning of stale references. * ReaderUI: Fix one such issue with its `active_widgets` array, with critical implications, as it essentially pinned *all* of ReaderUI's modules, including their reference to the `Document` instance (i.e., that was a big-ass leak). * Terminal: Make sure the shell is killed on plugin teardown. * InputText: Fix Home/End/Del physical keys to behave sensibly. * InputContainer/WidgetContainer: If necessary, compute self.dimen at paintTo time (previously, only InputContainers did, which might have had something to do with random widgets unconcerned about input using it as a baseclass instead of WidgetContainer...). * OverlapGroup: Compute self.dimen at *init* time, because for some reason it needs to do that, but do it directly in OverlapGroup instead of going through a weird WidgetContainer method that it was the sole user of. * ReaderCropping: Under no circumstances should a Document instance member (here, self.bbox) risk being `nil`ed! * Kobo: Minor code cleanups.
2 years ago
local Japanese = WidgetContainer:extend{
name = "japanese",
pretty_name = "Japanese",
Clarify our OOP semantics across the codebase (#9586) Basically: * Use `extend` for class definitions * Use `new` for object instantiations That includes some minor code cleanups along the way: * Updated `Widget`'s docs to make the semantics clearer. * Removed `should_restrict_JIT` (it's been dead code since https://github.com/koreader/android-luajit-launcher/pull/283) * Minor refactoring of LuaSettings/LuaData/LuaDefaults/DocSettings to behave (mostly, they are instantiated via `open` instead of `new`) like everything else and handle inheritance properly (i.e., DocSettings is now a proper LuaSettings subclass). * Default to `WidgetContainer` instead of `InputContainer` for stuff that doesn't actually setup key/gesture events. * Ditto for explicit `*Listener` only classes, make sure they're based on `EventListener` instead of something uselessly fancier. * Unless absolutely necessary, do not store references in class objects, ever; only values. Instead, always store references in instances, to avoid both sneaky inheritance issues, and sneaky GC pinning of stale references. * ReaderUI: Fix one such issue with its `active_widgets` array, with critical implications, as it essentially pinned *all* of ReaderUI's modules, including their reference to the `Document` instance (i.e., that was a big-ass leak). * Terminal: Make sure the shell is killed on plugin teardown. * InputText: Fix Home/End/Del physical keys to behave sensibly. * InputContainer/WidgetContainer: If necessary, compute self.dimen at paintTo time (previously, only InputContainers did, which might have had something to do with random widgets unconcerned about input using it as a baseclass instead of WidgetContainer...). * OverlapGroup: Compute self.dimen at *init* time, because for some reason it needs to do that, but do it directly in OverlapGroup instead of going through a weird WidgetContainer method that it was the sole user of. * ReaderCropping: Under no circumstances should a Document instance member (here, self.bbox) risk being `nil`ed! * Kobo: Minor code cleanups.
2 years ago
}
-- Yomichan uses 10 characters as the default look-ahead, but crengine's
-- getNextVisibleChar counts furigana if any are present, so use a higher
-- threshold to be able to look-ahead an equivalent number of characters.
local DEFAULT_TEXT_SCAN_LENGTH = 20
function Japanese:init()
self.deinflector = SingleInstanceDeinflector
self.dictionary = (self.ui and self.ui.dictionary) or ReaderDictionary:new()
self.max_scan_length = G_reader_settings:readSetting("language_japanese_text_scan_length") or DEFAULT_TEXT_SCAN_LENGTH
LanguageSupport:registerPlugin(self)
end
function Japanese:supportsLanguage(language_code)
return language_code == "ja" or language_code == "jpn"
end
--- Called from @{languagesupport.extraDictionaryFormCandidates} for Japanese
-- text. Tries to find and return any possible deinflections for the given text.
-- @param args arguments from language support
-- @treturn {string,...} extra dictionary form candiadates found (or nil)
-- @see languagesupport.extraDictionaryFormCandidates
-- @see languagesupport.registerPlugin
function Japanese:onWordLookup(args)
local text = args.text
-- If there are no CJK characters in the text, there's nothing to do.
if not util.hasCJKChar(text) then
return
end
--- @todo Try to repeatedly reduce the text and deinflect the shortened text
-- to provide more candidates. This is particularly needed because
-- JMDict has a habit of creating entries for compounds or phrases
-- that do not exist in monolingual dictionaries (even in 大辞林 or
-- 広辞苑) and our onWordSelection expansion accepts any dictionary's
-- largest entry. Unfortunately doing this nicely requires a bit of
-- extra work to be efficient (since we need to remove the last
-- character in the string).
local results = self.deinflector:deinflect(text)
logger.dbg("japanese.koplugin: deinflection of", text, "results:", results)
--- @todo Pass up the reasons list (formatted Yomichan style) to the
-- dictionary pop-up so you can get some more information about the
-- inflection. But this would require adding some kind of tag
-- metadata that we have to pass through from the lookup to the
-- dictionary pop-up.
local candidates = {}
for i, result in ipairs(results) do
candidates[i] = result.term
end
return candidates
end
-- @todo Switch this to utf8proc_category or something similar.
local JAPANESE_PUNCTUATION = "「」『』【】〘〙〖〗・・、、,。。.!?  \n"
local function isPossibleJapaneseWord(str)
for c in str:gmatch(util.UTF8_CHAR_PATTERN) do
if not util.isCJKChar(c) or JAPANESE_PUNCTUATION:find(c) ~= nil then
return false
end
end
return true
end
--- Called from @{languagesupport.improveWordSelection} for Japanese text.
-- Tries to expand the word selection defined by args.
-- @param args arguments from language support
-- @treturn {pos0,pos1} the new selection range (or nil)
-- @see languagesupport.improveWordSelection
-- @see languagesupport.registerPlugin
function Japanese:onWordSelection(args)
local callbacks = args.callbacks
local current_text = args.text
-- If the initial selection contains only non-CJK characters, then there's
-- no point trying to expand it because no Japanese words mix CJK and
-- non-CJK characters (there are non-CJK words in Japanese -- CM, NG, TKG
-- and their full-width equivalents for instance -- but they are selected
-- by crengine correctly already and are full words by themselves).
if current_text ~= "" and not util.hasCJKChar(current_text) then
return
end
-- We reset the end of the range to pos0+1 because crengine will select
-- half-width katakana (カタカナ) in strange ways that often overshoots the
-- end of words.
local pos0, pos1 = args.pos0, callbacks.get_next_char_pos(args.pos0)
-- We try to advance the end position until we hit a word.
--
-- Unfortunately it's possible for the complete word to be longer than the
-- first match (obvious examples include 読み込む or similar compound verbs
-- where it would be less than ideal to match 読み as the full word, but
-- there are more subtle kana-only cases as well) so we need to keep
-- looking forward, but unfortunately there isn't a great endpoint defined
-- either (aside from punctuation). So we just copy Yomichan and set a hard
-- limit (20 characters) and stop early if we ever hit punctuation. We then
-- select the longest word present in one of the user's installed
-- dictionaries (after deinflection).
local all_candidates = {}
local all_words = {}
local current_end = pos1
local num_expansions = 0
repeat
-- Move to the next character.
current_end = callbacks.get_next_char_pos(current_end)
current_text = callbacks.get_text_in_range(pos0, current_end)
num_expansions = num_expansions + 1
-- If the text could not be a complete Japanese word (i.e. it contains
-- a punctuation or some other special character), quit early. We test
-- the whole string rather than the last character because finding the
-- last character requires a linear walk through the string anyway, and
-- get_next_char_pos() skips over newlines.
if not isPossibleJapaneseWord(current_text) then
logger.dbg("japanese.koplugin: stopping expansion at", current_text, "because in contains non-word characters")
break
end
-- Get the selection and try to deinflect it.
local candidates = self.deinflector:deinflect(current_text)
local terms = {}
for _, candidate in ipairs(candidates) do
table.insert(terms, candidate.term)
end
-- Add the candidates to the set of words to attempt.
for _, term in ipairs(terms) do
table.insert(all_candidates, {
pos0 = pos0,
pos1 = current_end,
text = term,
})
table.insert(all_words, term)
end
until current_end == nil or num_expansions >= self.max_scan_length
logger.dbg("japanese.koplugin: attempted", num_expansions, "expansions up to", current_text)
-- Calling sdcv is fairly expensive, so reduce the cost by trying every
-- candidate in one shot and then picking the longest one which gave us a
-- result.
--- @todo Given there is a limit to how many command-line arguments you can
-- pass, we should split up the candidate list if it's too long.
local best_word
local cancelled, all_results = self.dictionary:rawSdcv(all_words)
if not cancelled and all_results ~= nil then
for i, term_results in ipairs(all_results) do
if #term_results ~= 0 then
best_word = all_candidates[i]
end
end
end
if best_word ~= nil then
return {best_word.pos0, best_word.pos1}
end
end
function Japanese:genMenuItem()
local sub_item_table = {
-- self.max_scan_length configuration
{
text_func = function()
return T(N_("Text scan length: %1 character", "Text scan length: %1 characters", self.max_scan_length), self.max_scan_length)
end,
help_text = _("Number of characters to look ahead when trying to expand tap-and-hold word selection in documents."),
keep_menu_open = true,
callback = function(touchmenu_instance)
local SpinWidget = require("ui/widget/spinwidget")
local Screen = require("device").screen
local items = SpinWidget:new{
title_text = _("Text scan length"),
info_text = T(_([[
The maximum number of characters to look ahead when trying to expand tap-and-hold word selection in documents.
Larger values allow longer phrases to be selected automatically, but with the trade-off that selections may become slower.
Default value: %1]]), DEFAULT_TEXT_SCAN_LENGTH),
width = math.floor(Screen:getWidth() * 0.75),
value = self.max_scan_length,
value_min = 0,
value_max = 1000,
value_step = 1,
value_hold_step = 10,
ok_text = _("Set scan length"),
default_value = DEFAULT_TEXT_SCAN_LENGTH,
callback = function(spin)
self.max_scan_length = spin.value
G_reader_settings:saveSetting("language_japanese_text_scan_length", self.max_scan_length)
if touchmenu_instance then touchmenu_instance:updateItems() end
end,
}
UIManager:show(items)
end,
},
}
-- self.deinflector configuration
util.arrayAppend(sub_item_table, self.deinflector:genMenuItems())
return {
text = _("Japanese"),
sub_item_table = sub_item_table,
}
end
return Japanese