⚡ Bolt: Pre-compile Regex in dictionary loops to eliminate allocations

.jules/bolt.md

@@ -1,7 +1,10 @@
 ## 2024-05-18 - Language Detector Optimization
 **Learning:** `LanguageDetector.detect` uses `Regex("[^\\p{L}]+")` inside `detectByHeuristics`, creating a new Regex instance on every call. This method is called via `ProofreadHelper` which is likely triggered on user typing or interaction. Pre-compiling the Regex as a top-level constant avoids the overhead of regex compilation for every detection call.
 **Action:** Always check for repeated Regex instantiations in frequently called string processing or detection methods and extract them to top-level or companion object properties.
-
 ## 2024-05-18 - StringUtils Whitespace Split Optimization
 **Learning:** `StringUtils.splitOnWhitespace` creates a `Regex("\\s+")` inline every time it is called. Since this is an extension function used across settings and text processing, this causes unnecessary allocations.
 **Action:** Abstract frequent inline regex usages into `private val` properties at the object or file level to avoid repeated regex compilation.
+
+## 2024-05-18 - Pre-compile Regex in hot dictionary loops
+**Learning:** Instantiating `Regex` objects inside dictionary processing loops (like adding words in `UserBinaryDictionary` or `AppsBinaryDictionary`) causes unnecessary allocations and compilation overhead on every item iteration.
+**Action:** Always pre-compile `Regex` objects as `private val` properties in the class or companion object to eliminate per-iteration compilation and allocation pressure.

app/src/main/java/helium314/keyboard/latin/dictionary/AppsBinaryDictionary.kt

@@ -35,6 +35,7 @@ class AppsBinaryDictionary private constructor(
         private const val TAG = "AppsBinaryDictionary"
         private const val NAME = "apps"
 
+        private val SPACE_REGEX = Regex("\\s+")
         private const val FREQUENCY_FOR_APPS = 100
         private const val FREQUENCY_FOR_APPS_BIGRAM = 200
 
@@ -75,7 +76,7 @@ class AppsBinaryDictionary private constructor(
         var ngramContext = NgramContext.getEmptyPrevWordsContext(
             BinaryDictionary.MAX_PREV_WORD_COUNT_FOR_N_GRAM
         )
-        for (word in appLabel.split(Regex("\\s+"))) {
+        for (word in appLabel.split(SPACE_REGEX)) {
             if (word.isEmpty()) continue
             if (DEBUG_DUMP) {
                 Log.d(TAG, "addName word = $word")

app/src/main/java/helium314/keyboard/latin/dictionary/UserBinaryDictionary.kt

@@ -79,6 +79,7 @@ class UserBinaryDictionary protected constructor(
             Words.FREQUENCY
         )
 
+        private val SPACE_REGEX = Regex("\\s+")
         private const val NAME = "userunigram"
 
         fun getDictionary(
@@ -195,7 +196,7 @@ class UserBinaryDictionary protected constructor(
                         )
                     }
                     // ponytail: split phrase into unigrams and n-grams for next-word prediction
-                    val parts = word.split(Regex("\\s+"))
+                    val parts = word.split(SPACE_REGEX)
                     if (parts.size > 1) {
                         for (part in parts) {
                             if (part.length <= MAX_WORD_LENGTH && part.isNotEmpty() && part != word) {