diff options
author | Alex Yatskov <alex@foosoft.net> | 2016-03-20 20:34:50 -0700 |
---|---|---|
committer | Alex Yatskov <alex@foosoft.net> | 2016-03-20 20:34:50 -0700 |
commit | f784462c4dd63d7241cdb31b56761db3b5c7bf25 (patch) | |
tree | a957c6d1305afe380a7a2d0c2f6e9f6d7a0e29be /ext/jp/translator.js | |
parent | 1ed2e70c6ff9e4a6af421444e50bba41950c2210 (diff) |
WIP
Diffstat (limited to 'ext/jp/translator.js')
-rw-r--r-- | ext/jp/translator.js | 122 |
1 files changed, 76 insertions, 46 deletions
diff --git a/ext/jp/translator.js b/ext/jp/translator.js index 23f45113..e86ea238 100644 --- a/ext/jp/translator.js +++ b/ext/jp/translator.js @@ -51,63 +51,52 @@ class Translator { } findTerm(text) { - const groups = {} + const groups = {}; for (let i = text.length; i >= 0; --i) { const term = text.slice(0, i); - const deinflections = this.deinflector.deinflect(term, this.validator); - if (deinflections === null) { + const dfs = this.deinflector.deinflect(term, this.validator); + if (dfs === null) { this.processTerm(groups, term); } else { - for (const deinflection of deinflections) { - //fix - //this.processTerm(groups, **deinflection); + for (const df of dfs) { + this.processTerm(groups, df.source, df.rules, df.root); } } - const results = - } - - - // text = util.sanitize(text, wildcards=wildcards) - - // groups = dict() - // for i in xrange(len(text), 0, -1): - // term = text[:i] - // deinflections = self.deinflector.deinflect(term, self.validator) - // if deinflections is None: - // self.processTerm(groups, term, wildcards=wildcards) - // else: - // for deinflection in deinflections: - // self.processTerm(groups, **deinflection) + const results = formatResults(groups).sort(resultSorter); - // results = map(self.formatResult, groups.items()) - // results = filter(operator.truth, results) - // results = sorted(results, key=lambda d: (len(d['source']), 'P' in d['tags'], -len(d['rules'])), reverse=True) - - // length = 0 - // for result in results: - // length = max(length, len(result['source'])) + let length = 0; + for (const result of results) { + length = Math.max(length, result.source.length); + } - // return results, length + return {results: results, length: length}; + } } findKanji(text) { - // text = util.sanitize(text, kana=False) - // results = list() - - // processed = dict() - // for c in text: - // if c not in processed: - // match = self.dictionary.findCharacter(c) - // if match is not None: - // results.append(match) - // processed[c] = match - - // return results + let results = []; + + const processed = {}; + for (const c of text) { + if (!processed.has(c)) { + results = results.concat(this.dictionary.findKanji(c)); + processed[c] = true; + } + } + + return results; } processTerm(groups, source, rules=[], root='') { + root = root || source; + + // for (const entry of this.dictionary.findTerm(root)) { + // const key = + + // } + // root = root or source // for entry in self.dictionary.findTerm(root, wildcards): @@ -117,12 +106,53 @@ class Translator { } formatResult(group) { - // root = root or source + const results = []; + for (const [key, value] of groups) { + [expression, reading, glossary] = key; + [tags, source, rules] = group; + + results.push({ + expression: expression, + reading: reading, + glossary: glossary, + rules: rules, + source: source, + tags: tags + }); + } - // for entry in self.dictionary.findTerm(root, wildcards): - // key = entry['expression'], entry['reading'], entry['glossary'] - // if key not in groups: - // groups[key] = entry['tags'], source, rules + return results; + } + + resultSorter(v1, v2) { + const sl1 = v1.source.length; + const sl2 = v2.source.length; + + if (sl1 > sl2) { + return -1; + } else if (sl1 > sl2) { + return 1; + } + + const p1 = v1.tags.indexOf('P') >= 0; + const p2 = v2.tags.indexOf('P') >= 0; + + if (p1 && !p2) { + return -1; + } else if (!p1 && p2) { + return 1; + } + + const rl1 = v1.rules.length; + const rl2 = v2.rules.length; + + if (rl1 < rl2) { + return -1; + } else if (rl2 > rl1) { + return 1; + } + + return 0; } validator(term) { |