MainLogBuffer.java revision 0aafbcf879a31afc8361078bd9574915d95694c0
1/* 2 * Copyright (C) 2012 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); you may not 5 * use this file except in compliance with the License. You may obtain a copy of 6 * the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT 12 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the 13 * License for the specific language governing permissions and limitations under 14 * the License. 15 */ 16 17package com.android.inputmethod.research; 18 19import android.util.Log; 20 21import com.android.inputmethod.latin.Dictionary; 22import com.android.inputmethod.latin.Suggest; 23import com.android.inputmethod.latin.define.ProductionFlag; 24 25import java.util.Random; 26 27public class MainLogBuffer extends LogBuffer { 28 private static final String TAG = MainLogBuffer.class.getSimpleName(); 29 private static final boolean DEBUG = false && ProductionFlag.IS_EXPERIMENTAL_DEBUG; 30 31 // The size of the n-grams logged. E.g. N_GRAM_SIZE = 2 means to sample bigrams. 32 private static final int N_GRAM_SIZE = 2; 33 // The number of words between n-grams to omit from the log. 34 private static final int DEFAULT_NUMBER_OF_WORDS_BETWEEN_SAMPLES = 35 ProductionFlag.IS_EXPERIMENTAL_DEBUG ? 2 : 18; 36 37 private final ResearchLog mResearchLog; 38 private Suggest mSuggest; 39 40 // The minimum periodicity with which n-grams can be sampled. E.g. mWinWordPeriod is 10 if 41 // every 10th bigram is sampled, i.e., words 1-8 are not, but the bigram at words 9 and 10, etc. 42 // for 11-18, and the bigram at words 19 and 20. If an n-gram is not safe (e.g. it contains a 43 // number in the middle or an out-of-vocabulary word), then sampling is delayed until a safe 44 // n-gram does appear. 45 /* package for test */ int mMinWordPeriod; 46 47 // Counter for words left to suppress before an n-gram can be sampled. Reset to mMinWordPeriod 48 // after a sample is taken. 49 /* package for test */ int mWordsUntilSafeToSample; 50 51 public MainLogBuffer(final ResearchLog researchLog) { 52 super(N_GRAM_SIZE); 53 mResearchLog = researchLog; 54 mMinWordPeriod = DEFAULT_NUMBER_OF_WORDS_BETWEEN_SAMPLES + N_GRAM_SIZE; 55 final Random random = new Random(); 56 mWordsUntilSafeToSample = random.nextInt(mMinWordPeriod); 57 } 58 59 public void setSuggest(Suggest suggest) { 60 mSuggest = suggest; 61 } 62 63 @Override 64 public void shiftIn(final LogUnit newLogUnit) { 65 super.shiftIn(newLogUnit); 66 if (newLogUnit.hasWord()) { 67 if (mWordsUntilSafeToSample > 0) { 68 mWordsUntilSafeToSample--; 69 } 70 } 71 if (DEBUG) { 72 Log.d(TAG, "shiftedIn " + (newLogUnit.hasWord() ? newLogUnit.getWord() : "")); 73 } 74 } 75 76 public void resetWordCounter() { 77 mWordsUntilSafeToSample = mMinWordPeriod; 78 } 79 80 /** 81 * Determines whether the content of the MainLogBuffer can be safely uploaded in its complete 82 * form and still protect the user's privacy. 83 * 84 * The size of the MainLogBuffer is just enough to hold one n-gram, its corrections, and any 85 * non-character data that is typed between words. The decision about privacy is made based on 86 * the buffer's entire content. If it is decided that the privacy risks are too great to upload 87 * the contents of this buffer, a censored version of the LogItems may still be uploaded. E.g., 88 * the screen orientation and other characteristics about the device can be uploaded without 89 * revealing much about the user. 90 */ 91 public boolean isSafeToLog() { 92 // Check that we are not sampling too frequently. Having sampled recently might disclose 93 // too much of the user's intended meaning. 94 if (mWordsUntilSafeToSample > 0) { 95 return false; 96 } 97 if (mSuggest == null || !mSuggest.hasMainDictionary()) { 98 // Main dictionary is unavailable. Since we cannot check it, we cannot tell if a word 99 // is out-of-vocabulary or not. Therefore, we must judge the entire buffer contents to 100 // potentially pose a privacy risk. 101 return false; 102 } 103 // Reload the dictionary in case it has changed (e.g., because the user has changed 104 // languages). 105 final Dictionary dictionary = mSuggest.getMainDictionary(); 106 if (dictionary == null) { 107 return false; 108 } 109 // Check each word in the buffer. If any word poses a privacy threat, we cannot upload the 110 // complete buffer contents in detail. 111 final int length = mLogUnits.size(); 112 for (int i = 0; i < length; i++) { 113 final LogUnit logUnit = mLogUnits.get(i); 114 final String word = logUnit.getWord(); 115 if (word == null) { 116 // Digits outside words are a privacy threat. 117 if (logUnit.mayContainDigit()) { 118 return false; 119 } 120 } else { 121 // Words not in the dictionary are a privacy threat. 122 if (ResearchLogger.hasLetters(word) && !(dictionary.isValidWord(word))) { 123 if (DEBUG) { 124 Log.d(TAG, "NOT SAFE!: hasLetters: " + ResearchLogger.hasLetters(word) 125 + ", isValid: " + (dictionary.isValidWord(word))); 126 } 127 return false; 128 } 129 } 130 } 131 // All checks have passed; this buffer's content can be safely uploaded. 132 return true; 133 } 134 135 @Override 136 protected void onShiftOut(LogUnit logUnit) { 137 if (mResearchLog != null) { 138 mResearchLog.publish(logUnit, false /* isIncludingPrivateData */); 139 } 140 } 141} 142