trying to add user words/patterns again:

src/ccmain/tessedit.cpp

@@ -185,8 +185,7 @@ bool Tesseract::init_tesseract_lang_data(
 #endif  // ndef DISABLED_LEGACY_ENGINE
     if (mgr->IsComponentAvailable(TESSDATA_LSTM)) {
       lstm_recognizer_ = new LSTMRecognizer;
-      ASSERT_HOST(
-          lstm_recognizer_->Load(lstm_use_matrix ? language : nullptr, mgr));
+      ASSERT_HOST(lstm_recognizer_->Load(this->params(), lstm_use_matrix ? language : nullptr, mgr));
     } else {
       tprintf("Error: LSTM requested, but not present!! Loading tesseract.\n");
       tessedit_ocr_engine_mode.set_value(OEM_TESSERACT_ONLY);

src/ccutil/params.cpp

@@ -213,4 +213,52 @@ void ParamUtils::ResetToDefaults(ParamsVectors* member_params) {
   }
 }
 
+// Resets all shared parameters by their pattern values
+void ParamUtils::ResetFromParams(ParamsVectors* member_params,
+                                 const ParamsVectors* pattern_params) {
+  if (member_params != nullptr && pattern_params != nullptr) {
+    int i;
+    for (i = 0; i < member_params->int_params.size(); ++i) {
+      IntParam& member = *member_params->int_params[i];
+      IntParam* param = ParamUtils::FindParam<IntParam>(member.name_str(),
+                                                        pattern_params->int_params,
+                                                        pattern_params->int_params);
+      if (param) {
+        //printf("overriding param %s=%d by =%d\n", member.name_str(), member, *param);
+        member.set_value(*param);
+      }
+    }
+    for (i = 0; i < member_params->bool_params.size(); ++i) {
+      BoolParam& member = *member_params->bool_params[i];
+      BoolParam* param = ParamUtils::FindParam<BoolParam>(member.name_str(),
+                                                          pattern_params->bool_params,
+                                                          pattern_params->bool_params);
+      if (param) {
+        //printf("overriding param %s=%s by =%s\n", member.name_str(), member ? "true" : "false", *param ? "true" : "false");
+        member.set_value(*param);
+      }
+    }
+    for (i = 0; i < member_params->string_params.size(); ++i) {
+      StringParam& member = *member_params->string_params[i];
+      StringParam* param = ParamUtils::FindParam<StringParam>(member.name_str(),
+                                                              pattern_params->string_params,
+                                                              pattern_params->string_params);
+      if (param) {
+        //printf("overriding param %s=%s by =%s\n", member.name_str(), member, *param);
+        member.set_value(*param);
+      }
+    }
+    for (i = 0; i < member_params->double_params.size(); ++i) {
+      DoubleParam& member = *member_params->double_params[i];
+      DoubleParam* param = ParamUtils::FindParam<DoubleParam>(member.name_str(),
+                                                              pattern_params->double_params,
+                                                              pattern_params->double_params);
+      if (param) {
+        //printf("overriding param %s=%f by =%f\n", member.name_str(), member, *param);
+        member.set_value(*param);
+      }
+    }
+  }
+}
+
 }  // namespace tesseract

src/ccutil/params.h

@@ -106,6 +106,10 @@ class ParamUtils {
 
   // Resets all parameters back to default values;
   static void ResetToDefaults(ParamsVectors* member_params);
+
+  // Resets all shared parameters by their pattern values
+  static void ResetFromParams(ParamsVectors* member_params,
+                              const ParamsVectors* pattern_params);
 };
 
 // Definition of various parameter types.

src/dict/dict.cpp

@@ -298,6 +298,9 @@ void Dict::Load(const STRING &lang, TessdataManager *data_file) {
                             getUnicharset().size(), dawg_debug_level);
 }
 
+// TODO(bertsky): unused, remove unless reasons appear to keep separate loaders
+// for Tesseract and LSTMRecognizer (same as Dict::Load without user words/patterns,
+// and document/pending words)
 // Loads the dawgs needed by the LSTM model. Call FinishLoad() after.
 void Dict::LoadLSTM(const STRING &lang, TessdataManager *data_file) {
   // Load dawgs_.

src/lstm/lstmrecognizer.cpp

@@ -66,13 +66,13 @@ LSTMRecognizer::~LSTMRecognizer() {
 }
 
 // Loads a model from mgr, including the dictionary only if lang is not null.
-bool LSTMRecognizer::Load(const char* lang, TessdataManager* mgr) {
+bool LSTMRecognizer::Load(const ParamsVectors* params, const char* lang, TessdataManager* mgr) {
   TFile fp;
   if (!mgr->GetComponent(TESSDATA_LSTM, &fp)) return false;
   if (!DeSerialize(mgr, &fp)) return false;
   if (lang == nullptr) return true;
   // Allow it to run without a dictionary.
-  LoadDictionary(lang, mgr);
+  LoadDictionary(params, lang, mgr);
   return true;
 }
 
@@ -154,11 +154,13 @@ bool LSTMRecognizer::LoadRecoder(TFile* fp) {
 // on the unicharset matching. This enables training to deserialize a model
 // from checkpoint or restore without having to go back and reload the
 // dictionary.
-bool LSTMRecognizer::LoadDictionary(const char* lang, TessdataManager* mgr) {
+bool LSTMRecognizer::LoadDictionary(const ParamsVectors* params, const char* lang, TessdataManager* mgr) {
   delete dict_;
   dict_ = new Dict(&ccutil_);
+  // get the variables configured at Tesseract instance via config/API:
+  ParamUtils::ResetFromParams(ccutil_.params(), params); 
   dict_->SetupForLoad(Dict::GlobalDawgCache());
-  dict_->LoadLSTM(lang, mgr);
+  dict_->Load(lang, mgr);
   if (dict_->FinishLoad()) return true;  // Success.
   tprintf("Failed to load any lstm-specific dictionaries for lang %s!!\n",
           lang);

src/lstm/lstmrecognizer.h

@@ -25,6 +25,7 @@
 #include "matrix.h"
 #include "network.h"
 #include "networkscratch.h"
+#include "params.h"
 #include "recodebeam.h"
 #include "series.h"
 #include "strngs.h"
@@ -154,7 +155,7 @@ class LSTMRecognizer {
   int null_char() const { return null_char_; }
 
   // Loads a model from mgr, including the dictionary only if lang is not null.
-  bool Load(const char* lang, TessdataManager* mgr);
+  bool Load(const ParamsVectors* params, const char* lang, TessdataManager* mgr);
 
   // Writes to the given file. Returns false in case of error.
   // If mgr contains a unicharset and recoder, then they are not encoded to fp.
@@ -174,7 +175,7 @@ class LSTMRecognizer {
   // on the unicharset matching. This enables training to deserialize a model
   // from checkpoint or restore without having to go back and reload the
   // dictionary.
-  bool LoadDictionary(const char* lang, TessdataManager* mgr);
+  bool LoadDictionary(const ParamsVectors* params, const char* lang, TessdataManager* mgr);
 
   // Recognizes the line image, contained within image_data, returning the
   // recognized tesseract WERD_RES for the words.