fix: guard opencv/phonemizer code paths when features are disabled

Fixes build failures when RNE_ENABLE_OPENCV=OFF or RNE_ENABLE_PHONEMIZER=OFF:

- RnExecutorchInstaller.cpp: guard CV model includes and JSI registrations
  with #ifdef RNE_ENABLE_OPENCV, phonemizer registration with
  #ifdef RNE_ENABLE_PHONEMIZER
- LLM.cpp: guard VisionEncoder and MultimodalRunner (VLM support) with
  #ifdef RNE_ENABLE_OPENCV — text-only LLM still works without OpenCV
- CMakeLists.txt: add runner/encoders/vision_encoder.cpp,
  runner/multimodal_prefiller.cpp and runner/multimodal_runner.cpp to
  OPENCV_CPP_SOURCES so they are excluded when OpenCV is disabled

Verified: Android build succeeds with enableOpencv=false, enablePhonemizer=false.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: msluszniak

packages/react-native-executorch/android/src/main/cpp/CMakeLists.txt

@@ -30,6 +30,9 @@ file(GLOB_RECURSE OPENCV_CPP_SOURCES CONFIGURE_DEPENDS
   "${COMMON_CPP_DIR}/rnexecutorch/utils/FrameProcessor.cpp"
   "${COMMON_CPP_DIR}/rnexecutorch/utils/FrameTransform.cpp"
   "${COMMON_CPP_DIR}/rnexecutorch/utils/computer_vision/*.cpp"
+  "${COMMON_CPP_DIR}/runner/encoders/vision_encoder.cpp"
+  "${COMMON_CPP_DIR}/runner/multimodal_prefiller.cpp"
+  "${COMMON_CPP_DIR}/runner/multimodal_runner.cpp"
 )
 
 # Phonemizer-dependent sources: Kokoro TTS (only user of phonemis)

packages/react-native-executorch/common/rnexecutorch/RnExecutorchInstaller.cpp

@@ -2,22 +2,28 @@
 
 #include <rnexecutorch/TokenizerModule.h>
 #include <rnexecutorch/host_objects/JsiConversions.h>
+#include <rnexecutorch/models/embeddings/text/TextEmbeddings.h>
+#include <rnexecutorch/models/llm/LLM.h>
+#include <rnexecutorch/models/speech_to_text/SpeechToText.h>
+#include <rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.h>
+#include <rnexecutorch/threads/GlobalThreadPool.h>
+#include <rnexecutorch/threads/utils/ThreadUtils.h>
+
+#ifdef RNE_ENABLE_OPENCV
 #include <rnexecutorch/models/classification/Classification.h>
 #include <rnexecutorch/models/embeddings/image/ImageEmbeddings.h>
-#include <rnexecutorch/models/embeddings/text/TextEmbeddings.h>
 #include <rnexecutorch/models/instance_segmentation/BaseInstanceSegmentation.h>
-#include <rnexecutorch/models/llm/LLM.h>
 #include <rnexecutorch/models/object_detection/ObjectDetection.h>
 #include <rnexecutorch/models/ocr/OCR.h>
 #include <rnexecutorch/models/semantic_segmentation/BaseSemanticSegmentation.h>
-#include <rnexecutorch/models/speech_to_text/SpeechToText.h>
 #include <rnexecutorch/models/style_transfer/StyleTransfer.h>
 #include <rnexecutorch/models/text_to_image/TextToImage.h>
-#include <rnexecutorch/models/text_to_speech/TextToSpeech.h>
 #include <rnexecutorch/models/vertical_ocr/VerticalOCR.h>
-#include <rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.h>
-#include <rnexecutorch/threads/GlobalThreadPool.h>
-#include <rnexecutorch/threads/utils/ThreadUtils.h>
+#endif
+
+#ifdef RNE_ENABLE_PHONEMIZER
+#include <rnexecutorch/models/text_to_speech/TextToSpeech.h>
+#endif
 
 #if defined(__ANDROID__) && defined(__aarch64__)
 #include <executorch/extension/threadpool/cpuinfo_utils.h>
@@ -40,6 +46,7 @@ void RnExecutorchInstaller::injectJSIBindings(
   jsiRuntime->global().setProperty(*jsiRuntime, "__rne_isEmulator",
                                    jsi::Value(isEmulator));
 
+#ifdef RNE_ENABLE_OPENCV
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadStyleTransfer",
       RnExecutorchInstaller::loadModel<models::style_transfer::StyleTransfer>(
@@ -72,6 +79,7 @@ void RnExecutorchInstaller::injectJSIBindings(
       RnExecutorchInstaller::loadModel<
           models::object_detection::ObjectDetection>(jsiRuntime, jsCallInvoker,
                                                      "loadObjectDetection"));
+#endif // RNE_ENABLE_OPENCV
 
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadExecutorchModule",
@@ -83,10 +91,12 @@ void RnExecutorchInstaller::injectJSIBindings(
       RnExecutorchInstaller::loadModel<TokenizerModule>(
           jsiRuntime, jsCallInvoker, "loadTokenizerModule"));
 
+#ifdef RNE_ENABLE_OPENCV
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadImageEmbeddings",
       RnExecutorchInstaller::loadModel<models::embeddings::ImageEmbeddings>(
           jsiRuntime, jsCallInvoker, "loadImageEmbeddings"));
+#endif // RNE_ENABLE_OPENCV
 
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadTextEmbeddings",
@@ -98,6 +108,7 @@ void RnExecutorchInstaller::injectJSIBindings(
       RnExecutorchInstaller::loadModel<models::llm::LLM>(
           jsiRuntime, jsCallInvoker, "loadLLM"));
 
+#ifdef RNE_ENABLE_OPENCV
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadOCR",
       RnExecutorchInstaller::loadModel<models::ocr::OCR>(
@@ -107,16 +118,19 @@ void RnExecutorchInstaller::injectJSIBindings(
       *jsiRuntime, "loadVerticalOCR",
       RnExecutorchInstaller::loadModel<models::ocr::VerticalOCR>(
           jsiRuntime, jsCallInvoker, "loadVerticalOCR"));
+#endif // RNE_ENABLE_OPENCV
 
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadSpeechToText",
       RnExecutorchInstaller::loadModel<models::speech_to_text::SpeechToText>(
           jsiRuntime, jsCallInvoker, "loadSpeechToText"));
 
+#ifdef RNE_ENABLE_PHONEMIZER
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadTextToSpeechKokoro",
       RnExecutorchInstaller::loadModel<models::text_to_speech::kokoro::Kokoro>(
           jsiRuntime, jsCallInvoker, "loadTextToSpeechKokoro"));
+#endif // RNE_ENABLE_PHONEMIZER
 
   jsiRuntime->global().setProperty(
       *jsiRuntime, "loadVAD",

packages/react-native-executorch/common/rnexecutorch/models/llm/LLM.cpp

@@ -6,9 +6,11 @@
 #include <rnexecutorch/Error.h>
 #include <rnexecutorch/Log.h>
 #include <rnexecutorch/threads/GlobalThreadPool.h>
+#include <runner/text_runner.h>
+#ifdef RNE_ENABLE_OPENCV
 #include <runner/encoders/vision_encoder.h>
 #include <runner/multimodal_runner.h>
-#include <runner/text_runner.h>
+#endif
 
 namespace rnexecutorch::models::llm {
 namespace llm = ::executorch::extension::llm;
@@ -22,10 +24,8 @@ LLM::LLM(const std::string &modelSource, const std::string &tokenizerSource,
          std::shared_ptr<react::CallInvoker> callInvoker)
     : BaseModel(modelSource, callInvoker, Module::LoadMode::File) {
 
-  if (capabilities.empty()) {
-    runner_ =
-        std::make_unique<llm::TextRunner>(std::move(module_), tokenizerSource);
-  } else {
+#ifdef RNE_ENABLE_OPENCV
+  if (!capabilities.empty()) {
     std::map<llm::MultimodalType, std::unique_ptr<llm::IEncoder>> encoders;
     for (const auto &cap : capabilities) {
       if (cap == "vision") {
@@ -35,7 +35,13 @@ LLM::LLM(const std::string &modelSource, const std::string &tokenizerSource,
     }
     runner_ = std::make_unique<llm::MultimodalRunner>(
         std::move(module_), tokenizerSource, std::move(encoders));
+  } else {
+#endif
+    runner_ =
+        std::make_unique<llm::TextRunner>(std::move(module_), tokenizerSource);
+#ifdef RNE_ENABLE_OPENCV
   }
+#endif
 
   auto loadResult = runner_->load();
   if (loadResult != Error::Ok) {