Add claifying comment
This is a follow-up to I7c01c3dd3ac33d7e96c00836256bae9c14b124ed
Change-Id: I9e3f8968c4bbf9525d5dfe101f71373c42f88361
diff --git a/java/src/com/android/inputmethod/research/MainLogBuffer.java b/java/src/com/android/inputmethod/research/MainLogBuffer.java
index 9aa3499..7e8f166 100644
--- a/java/src/com/android/inputmethod/research/MainLogBuffer.java
+++ b/java/src/com/android/inputmethod/research/MainLogBuffer.java
@@ -196,6 +196,22 @@
}
}
+ /**
+ * If there is a safe n-gram at the front of this log buffer, publish it with all details, and
+ * remove the LogUnits that constitute it.
+ *
+ * An n-gram might not be "safe" if it violates privacy controls. E.g., it might contain
+ * numbers, an out-of-vocabulary word, or another n-gram may have been published recently. If
+ * there is no safe n-gram, then the LogUnits up through the first word-containing LogUnit are
+ * published, but without disclosing any privacy-related details, such as the word the LogUnit
+ * generated, motion data, etc.
+ *
+ * Note that a LogUnit can hold more than one word if the user types without explicit spaces.
+ * In this case, the words may be grouped together in such a way that pulling an n-gram off the
+ * front would require splitting a LogUnit. Splitting a LogUnit is not possible, so this case
+ * is treated just as the unsafe n-gram case. This may cause n-grams to be sampled at slightly
+ * less than the target frequency.
+ */
protected final void publishLogUnitsAtFrontOfBuffer() throws IOException {
// TODO: Refactor this method to require fewer passes through the LogUnits. Should really
// require only one pass.