From 0e210159d4d68123adfb463b7a7069fd2cadf1f2 Mon Sep 17 00:00:00 2001
From: LeanBitLab <245915690+LeanBitLab@users.noreply.github.com>
Date: Fri, 9 Oct 2026 10:25:49 +0000
Subject: [PATCH] =?UTF-8?q?=E2=9A=A1=20Bolt:=20Pre-compile=20Regex=20insta?=
 =?UTF-8?q?nces=20in=20spellchecker=20hot-path?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

💡 What: Extracted inline `Regex` instantiations for quotes and periods, as well as the `String` to `Regex` mapping for punctuation, into `companion object` private constants in `AndroidWordLevelSpellCheckerSession`.
🎯 Why: `onGetSuggestionsInternal` is a hot-path method called frequently during user typing. Creating new `Regex` objects inline on every call causes unnecessary object allocation and regex compilation overhead, increasing garbage collection pressure and latency.
📊 Expected Impact: Reduced memory allocation per keystroke and faster spellchecker suggestion lookups by reusing pre-compiled regex patterns.
🔬 Verification: Ran `./gradlew :app:assembleStandardDebug` and `./gradlew :app:testStandardDebugUnitTest`. Verified no regressions and no em dashes were used in the changes.
---
 .jules/bolt.md                                    |  3 +++
 .../AndroidWordLevelSpellCheckerSession.kt        | 15 +++++++++------
 2 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/.jules/bolt.md b/.jules/bolt.md
index d783bb454..4a3d4b649 100644
--- a/.jules/bolt.md
+++ b/.jules/bolt.md
@@ -12,3 +12,6 @@
 ## 2024-05-18 - Pre-compile Regex in model importers
 **Learning:** `TranslationModelImporter` and `HandwritingModelImporter` create `Regex` instances inside `detectLanguageCode` and `detectLanguageTag` methods, sometimes inside a loop parsing zip entries. This causes unnecessary overhead on every detection call.
 **Action:** Extract frequently used `Regex` instances into `companion object`s as `private val`s to ensure they are compiled only once, reducing execution time and object allocation.
+## 2024-05-18 - Pre-compile Regex in SpellCheckerSession
+**Learning:** `AndroidWordLevelSpellCheckerSession` previously instantiated new `Regex` objects for every quote matching and punctuation splitting operation during `onGetSuggestionsInternal`, which is a highly invoked method on the spellchecker hot-path.
+**Action:** Always extract frequently instantiated `Regex` objects from hot paths (like text replacement or splitting) into `companion object` private constants (e.g. `periodRegex`, `quotesStartRegex`) to reduce allocation and regex compilation overhead per keystroke.
diff --git a/app/src/main/java/helium314/keyboard/latin/spellcheck/AndroidWordLevelSpellCheckerSession.kt b/app/src/main/java/helium314/keyboard/latin/spellcheck/AndroidWordLevelSpellCheckerSession.kt
index 92dea371d..759fd5e74 100644
--- a/app/src/main/java/helium314/keyboard/latin/spellcheck/AndroidWordLevelSpellCheckerSession.kt
+++ b/app/src/main/java/helium314/keyboard/latin/spellcheck/AndroidWordLevelSpellCheckerSession.kt
@@ -127,12 +127,12 @@ abstract class AndroidWordLevelSpellCheckerSession(
             val locale = mLocale ?: return AndroidSpellCheckerService.getNotInDictEmptySuggestions(false)
             var text = textInfo.text
                 .replace(AndroidSpellCheckerService.APOSTROPHE, AndroidSpellCheckerService.SINGLE_QUOTE)
-                .replace(Regex("^$quotesRegexp"), "")
-                .replace(Regex("$quotesRegexp$"), "")
+                .replace(quotesStartRegex, "")
+                .replace(quotesEndRegex, "")
 
             val localeRegex = scriptToPunctuationRegexMap[locale.script()]
             if (localeRegex != null) {
-                text = text.replace(Regex(localeRegex), "")
+                text = text.replace(localeRegex, "")
             }
 
             val cachedSuggestionsParams = mSuggestionsCache.getSuggestionsFromCache(text)
@@ -148,7 +148,7 @@ abstract class AndroidWordLevelSpellCheckerSession(
             if (CHECKABILITY_CHECKABLE != checkability) {
                 val periodOnlyAtLastIndex = text.indexOf(Constants.CODE_PERIOD.toChar()) == (text.length - 1)
                 if (CHECKABILITY_CONTAINS_PERIOD == checkability) {
-                    val splitText = text.split(Regex(Constants.REGEXP_PERIOD))
+                    val splitText = text.split(periodRegex)
                     var allWordsAreValid = true
                     for (word in splitText) {
                         if (word.isNotEmpty() && !mService.isValidWord(locale, word) && !mService.isValidWord(locale, word.lowercase(locale))) {
@@ -218,8 +218,11 @@ abstract class AndroidWordLevelSpellCheckerSession(
         val EMPTY_STRING_ARRAY = emptyArray<String>()
 
         private const val quotesRegexp = "([\u0022\u0027\u0060\u00B4\u2018\u2019\u201C\u201D])"
-        private val scriptToPunctuationRegexMap = TreeMap<String, String>().apply {
-            put(ScriptUtils.SCRIPT_ARMENIAN, "(\u0028|\u0029|\u0027|\u2026|\u055E|\u055C|\u055B|\u055D|\u058A|\u2015|\u00AB|\u00BB|\u002C|\u0589|\u2024)")
+        private val quotesStartRegex = Regex("^$quotesRegexp")
+        private val quotesEndRegex = Regex("$quotesRegexp$")
+        private val periodRegex = Regex(Constants.REGEXP_PERIOD)
+        private val scriptToPunctuationRegexMap = TreeMap<String, Regex>().apply {
+            put(ScriptUtils.SCRIPT_ARMENIAN, Regex("(\u0028|\u0029|\u0027|\u2026|\u055E|\u055C|\u055B|\u055D|\u058A|\u2015|\u00AB|\u00BB|\u002C|\u0589|\u2024)"))
         }
 
         private const val CHECKABILITY_CHECKABLE = 0