Use const char* for filename and network_spec parameters

This replaces the proprietary STRING data type
(764 instead of 838 lines remaining).

It also removes STRING from osdetect.h and serialis.h.

Signed-off-by: Stefan Weil <sw@weilnetz.de>

src/lstm/convolve.cpp

@@ -28,7 +28,7 @@
 
 namespace tesseract {
 
-Convolve::Convolve(const STRING& name, int ni, int half_x, int half_y)
+Convolve::Convolve(const std::string& name, int ni, int half_x, int half_y)
   : Network(NT_CONVOLVE, name, ni, ni * (2*half_x + 1) * (2*half_y + 1)),
     half_x_(half_x), half_y_(half_y) {
 }

src/lstm/convolve.h

@@ -33,7 +33,7 @@ class Convolve : public Network {
  public:
   // The area of convolution is 2*half_x + 1 by 2*half_y + 1, forcing it to
   // always be odd, so the center is the current pixel.
-  Convolve(const STRING& name, int ni, int half_x, int half_y);
+  Convolve(const std::string& name, int ni, int half_x, int half_y);
   ~Convolve() override = default;
 
   STRING spec() const override {

src/lstm/fullyconnected.cpp

@@ -39,7 +39,7 @@ const int kNumThreads = 1;
 
 namespace tesseract {
 
-FullyConnected::FullyConnected(const STRING& name, int ni, int no,
+FullyConnected::FullyConnected(const std::string& name, int ni, int no,
                                NetworkType type)
   : Network(type, name, ni, no), external_source_(nullptr), int_mode_(false) {
 }

src/lstm/fullyconnected.h

@@ -2,7 +2,6 @@
 // File:        fullyconnected.h
 // Description: Simple feed-forward layer with various non-linearities.
 // Author:      Ray Smith
-// Created:     Wed Feb 26 14:46:06 PST 2014
 //
 // (C) Copyright 2014, Google Inc.
 // Licensed under the Apache License, Version 2.0 (the "License");
@@ -27,7 +26,7 @@ namespace tesseract {
 // C++ Implementation of the Softmax (output) class from lstm.py.
 class FullyConnected : public Network {
  public:
-  FullyConnected(const STRING& name, int ni, int no, NetworkType type);
+  FullyConnected(const std::string& name, int ni, int no, NetworkType type);
   ~FullyConnected() override = default;
 
   // Returns the shape output from the network given an input shape (which may

src/lstm/input.cpp

@@ -27,9 +27,9 @@ namespace tesseract {
 // Max height for variable height inputs before scaling anyway.
 const int kMaxInputHeight = 48;
 
-Input::Input(const STRING& name, int ni, int no)
+Input::Input(const std::string& name, int ni, int no)
     : Network(NT_INPUT, name, ni, no), cached_x_scale_(1) {}
-Input::Input(const STRING& name, const StaticShape& shape)
+Input::Input(const std::string& name, const StaticShape& shape)
     : Network(NT_INPUT, name, shape.height(), shape.depth()),
       shape_(shape),
       cached_x_scale_(1) {

src/lstm/input.h

@@ -26,8 +26,8 @@ namespace tesseract {
 
 class Input : public Network {
  public:
-  Input(const STRING& name, int ni, int no);
-  Input(const STRING& name, const StaticShape& shape);
+  Input(const std::string& name, int ni, int no);
+  Input(const std::string& name, const StaticShape& shape);
   ~Input() override = default;
 
   STRING spec() const override {

src/lstm/lstm.cpp

@@ -26,6 +26,7 @@
 #endif
 #include <cstdio>
 #include <cstdlib>
+#include <sstream>    // for std::ostringstream
 
 #if !defined(__GNUC__) && defined(_MSC_VER)
 #include <intrin.h>     // _BitScanReverse
@@ -99,7 +100,7 @@ static inline uint32_t ceil_log2(uint32_t n)
   return (n == (1u << l2)) ? l2 : l2 + 1;
 }
 
-LSTM::LSTM(const STRING& name, int ni, int ns, int no, bool two_dimensional,
+LSTM::LSTM(const std::string& name, int ni, int ns, int no, bool two_dimensional,
            NetworkType type)
     : Network(type, name, ni, no),
       na_(ni + ns),
@@ -197,9 +198,9 @@ void LSTM::ConvertToInt() {
 void LSTM::DebugWeights() {
   for (int w = 0; w < WT_COUNT; ++w) {
     if (w == GFS && !Is2D()) continue;
-    STRING msg = name_;
-    msg.add_str_int(" Gate weights ", w);
-    gate_weights_[w].Debug2D(msg.c_str());
+    std::ostringstream msg;
+    msg << name_ << " Gate weights " << w;
+    gate_weights_[w].Debug2D(msg.str().c_str());
   }
   if (softmax_ != nullptr) {
     softmax_->DebugWeights();

src/lstm/lstm.h

@@ -2,7 +2,6 @@
 // File:        lstm.h
 // Description: Long-term-short-term-memory Recurrent neural network.
 // Author:      Ray Smith
-// Created:     Wed May 01 17:33:06 PST 2013
 //
 // (C) Copyright 2013, Google Inc.
 // Licensed under the Apache License, Version 2.0 (the "License");
@@ -47,7 +46,7 @@ class LSTM : public Network {
   // 2-d and bidi softmax LSTMs are not rejected, but are impossible to build
   // in the conventional way because the output feedback both forwards and
   // backwards in time does become impossible.
-  LSTM(const STRING& name, int num_inputs, int num_states, int num_outputs,
+  LSTM(const std::string& name, int num_inputs, int num_states, int num_outputs,
        bool two_dimensional, NetworkType type);
   ~LSTM() override;
 

src/lstm/maxpool.cpp

@@ -19,7 +19,7 @@
 
 namespace tesseract {
 
-Maxpool::Maxpool(const STRING& name, int ni, int x_scale, int y_scale)
+Maxpool::Maxpool(const char* name, int ni, int x_scale, int y_scale)
   : Reconfig(name, ni, x_scale, y_scale) {
   type_ = NT_MAXPOOL;
   no_ = ni;

src/lstm/maxpool.h

@@ -28,7 +28,7 @@ namespace tesseract {
 // Backprop propagates only to the position that was the max.
 class Maxpool : public Reconfig {
  public:
-  Maxpool(const STRING& name, int ni, int x_scale, int y_scale);
+  Maxpool(const char* name, int ni, int x_scale, int y_scale);
   ~Maxpool() override = default;
 
   // Accessors.

src/lstm/network.cpp

@@ -84,7 +84,7 @@ Network::Network()
       forward_win_(nullptr),
       backward_win_(nullptr),
       randomizer_(nullptr) {}
-Network::Network(NetworkType type, const STRING& name, int ni, int no)
+Network::Network(NetworkType type, const std::string& name, int ni, int no)
     : type_(type),
       training_(TS_ENABLED),
       needs_to_backprop_(true),
@@ -161,7 +161,7 @@ bool Network::Serialize(TFile* fp) const {
   if (!fp->Serialize(&ni_)) return false;
   if (!fp->Serialize(&no_)) return false;
   if (!fp->Serialize(&num_weights_)) return false;
-  if (!name_.Serialize(fp)) return false;
+  if (!fp->Serialize(name_.c_str(), name_.length())) return false;
   return true;
 }
 
@@ -208,40 +208,40 @@ Network* Network::CreateFromFile(TFile* fp) {
 
   switch (type) {
     case NT_CONVOLVE:
-      network = new Convolve(name, ni, 0, 0);
+      network = new Convolve(name.c_str(), ni, 0, 0);
       break;
     case NT_INPUT:
-      network = new Input(name, ni, no);
+      network = new Input(name.c_str(), ni, no);
       break;
     case NT_LSTM:
     case NT_LSTM_SOFTMAX:
     case NT_LSTM_SOFTMAX_ENCODED:
     case NT_LSTM_SUMMARY:
       network =
-          new LSTM(name, ni, no, no, false, type);
+          new LSTM(name.c_str(), ni, no, no, false, type);
       break;
     case NT_MAXPOOL:
-      network = new Maxpool(name, ni, 0, 0);
+      network = new Maxpool(name.c_str(), ni, 0, 0);
       break;
     // All variants of Parallel.
     case NT_PARALLEL:
     case NT_REPLICATED:
     case NT_PAR_RL_LSTM:
     case NT_PAR_UD_LSTM:
     case NT_PAR_2D_LSTM:
-      network = new Parallel(name, type);
+      network = new Parallel(name.c_str(), type);
       break;
     case NT_RECONFIG:
-      network = new Reconfig(name, ni, 0, 0);
+      network = new Reconfig(name.c_str(), ni, 0, 0);
       break;
     // All variants of reversed.
     case NT_XREVERSED:
     case NT_YREVERSED:
     case NT_XYTRANSPOSE:
-      network = new Reversed(name, type);
+      network = new Reversed(name.c_str(), type);
       break;
     case NT_SERIES:
-      network = new Series(name);
+      network = new Series(name.c_str());
       break;
     case NT_TENSORFLOW:
 #ifdef INCLUDE_TENSORFLOW
@@ -259,7 +259,7 @@ Network* Network::CreateFromFile(TFile* fp) {
     case NT_LOGISTIC:
     case NT_POSCLIP:
     case NT_SYMCLIP:
-      network = new FullyConnected(name, ni, no, type);
+      network = new FullyConnected(name.c_str(), ni, no, type);
       break;
     default:
       break;
@@ -298,7 +298,7 @@ void Network::DisplayForward(const NetworkIO& matrix) {
 // Displays the image of the matrix to the backward window.
 void Network::DisplayBackward(const NetworkIO& matrix) {
   Pix* image = matrix.ToPix();
-  STRING window_name = name_ + "-back";
+  std::string window_name = name_ + "-back";
   ClearWindow(false, window_name.c_str(), pixGetWidth(image),
               pixGetHeight(image), &backward_win_);
   DisplayImage(image, backward_win_);

src/lstm/network.h

@@ -105,7 +105,7 @@ enum TrainingState {
 class Network {
  public:
   Network();
-  Network(NetworkType type, const STRING& name, int ni, int no);
+  Network(NetworkType type, const std::string& name, int ni, int no);
   virtual ~Network() = default;
 
   // Accessors.
@@ -135,7 +135,7 @@ class Network {
     result.set_depth(no_);
     return result;
   }
-  const STRING& name() const {
+  const std::string& name() const {
     return name_;
   }
   virtual STRING spec() const {
@@ -297,7 +297,7 @@ class Network {
   int32_t ni_;                // Number of input values.
   int32_t no_;                // Number of output values.
   int32_t num_weights_;       // Number of weights in this and sub-network.
-  STRING name_;               // A unique name for this layer.
+  std::string name_;          // A unique name for this layer.
 
   // NOT-serialized debug data.
   ScrollView* forward_win_;   // Recognition debug display window.

src/lstm/parallel.cpp

@@ -31,7 +31,7 @@
 namespace tesseract {
 
 // ni_ and no_ will be set by AddToStack.
-Parallel::Parallel(const STRING& name, NetworkType type) : Plumbing(name) {
+Parallel::Parallel(const char* name, NetworkType type) : Plumbing(name) {
   type_ = type;
 }
 

src/lstm/parallel.h

@@ -27,7 +27,7 @@ namespace tesseract {
 class Parallel : public Plumbing {
  public:
   // ni_ and no_ will be set by AddToStack.
-  Parallel(const STRING& name, NetworkType type);
+  Parallel(const char* name, NetworkType type);
   ~Parallel() override = default;
 
   // Returns the shape output from the network given an input shape (which may

src/lstm/plumbing.cpp

@@ -22,7 +22,7 @@
 namespace tesseract {
 
 // ni_ and no_ will be set by AddToStack.
-Plumbing::Plumbing(const STRING& name)
+Plumbing::Plumbing(const std::string& name)
   : Network(NT_PARALLEL, name, 0, 0) {
 }
 

src/lstm/plumbing.h

@@ -3,7 +3,6 @@
 // Description: Base class for networks that organize other networks
 //              eg series or parallel.
 // Author:      Ray Smith
-// Created:     Mon May 12 08:11:36 PST 2014
 //
 // (C) Copyright 2014, Google Inc.
 // Licensed under the Apache License, Version 2.0 (the "License");
@@ -30,7 +29,7 @@ namespace tesseract {
 class Plumbing : public Network {
  public:
   // ni_ and no_ will be set by AddToStack.
-  explicit Plumbing(const STRING& name);
+  explicit Plumbing(const std::string& name);
   ~Plumbing() override = default;
 
   // Returns the required shape input to the network.

src/lstm/reconfig.cpp

@@ -20,7 +20,7 @@
 
 namespace tesseract {
 
-Reconfig::Reconfig(const STRING& name, int ni, int x_scale, int y_scale)
+Reconfig::Reconfig(const char* name, int ni, int x_scale, int y_scale)
   : Network(NT_RECONFIG, name, ni, ni * x_scale * y_scale),
     x_scale_(x_scale), y_scale_(y_scale) {
 }

src/lstm/reconfig.h

@@ -31,7 +31,7 @@ namespace tesseract {
 // input stride is a multiple of the y_scale factor!
 class Reconfig : public Network {
  public:
-  Reconfig(const STRING& name, int ni, int x_scale, int y_scale);
+  Reconfig(const char* name, int ni, int x_scale, int y_scale);
   ~Reconfig() override = default;
 
   // Returns the shape output from the network given an input shape (which may

src/lstm/reversed.cpp

@@ -2,7 +2,6 @@
 // File:        reversed.cpp
 // Description: Runs a single network on time-reversed input, reversing output.
 // Author:      Ray Smith
-// Created:     Thu May 02 08:42:06 PST 2013
 //
 // (C) Copyright 2013, Google Inc.
 // Licensed under the Apache License, Version 2.0 (the "License");
@@ -24,7 +23,7 @@
 
 namespace tesseract {
 
-Reversed::Reversed(const STRING& name, NetworkType type) : Plumbing(name) {
+Reversed::Reversed(const std::string& name, NetworkType type) : Plumbing(name) {
   type_ = type;
 }
 

src/lstm/reversed.h

@@ -27,7 +27,7 @@ namespace tesseract {
 // C++ Implementation of the Reversed class from lstm.py.
 class Reversed : public Plumbing {
  public:
-  explicit Reversed(const STRING& name, NetworkType type);
+  explicit Reversed(const std::string& name, NetworkType type);
   ~Reversed() override = default;
 
   // Returns the shape output from the network given an input shape (which may

src/lstm/series.cpp

@@ -25,7 +25,7 @@
 namespace tesseract {
 
 // ni_ and no_ will be set by AddToStack.
-Series::Series(const STRING& name) : Plumbing(name) {
+Series::Series(const char* name) : Plumbing(name) {
   type_ = NT_SERIES;
 }
 

src/lstm/series.h

@@ -27,7 +27,7 @@ namespace tesseract {
 class Series : public Plumbing {
  public:
   // ni_ and no_ will be set by AddToStack.
-  explicit Series(const STRING& name);
+  explicit Series(const char* name);
   ~Series() override = default;
 
   // Returns the shape output from the network given an input shape (which may

src/training/lstmtrainer.cpp

@@ -141,11 +141,11 @@ bool LSTMTrainer::TryLoadingCheckpoint(const char* filename,
 // are implemented.
 // For other args see NetworkBuilder::InitNetwork.
 // Note: Be sure to call InitCharSet before InitNetwork!
-bool LSTMTrainer::InitNetwork(const STRING& network_spec, int append_index,
+bool LSTMTrainer::InitNetwork(const char* network_spec, int append_index,
                               int net_flags, float weight_range,
                               float learning_rate, float momentum,
                               float adam_beta) {
-  mgr_.SetVersionString(mgr_.VersionString() + ":" + network_spec.c_str());
+  mgr_.SetVersionString(mgr_.VersionString() + ":" + network_spec);
   adam_beta_ = adam_beta;
   learning_rate_ = learning_rate;
   momentum_ = momentum;
@@ -157,7 +157,7 @@ bool LSTMTrainer::InitNetwork(const STRING& network_spec, int append_index,
   }
   network_str_ += network_spec;
   tprintf("Built network:%s from request %s\n",
-          network_->spec().c_str(), network_spec.c_str());
+          network_->spec().c_str(), network_spec);
   tprintf(
       "Training parameters:\n  Debug interval = %d,"
       " weights = %g, learning rate = %g, momentum=%g\n",

src/training/lstmtrainer.h

@@ -107,7 +107,7 @@ class LSTMTrainer : public LSTMRecognizer {
   // are implemented.
   // For other args see NetworkBuilder::InitNetwork.
   // Note: Be sure to call InitCharSet before InitNetwork!
-  bool InitNetwork(const STRING& network_spec, int append_index, int net_flags,
+  bool InitNetwork(const char* network_spec, int append_index, int net_flags,
                    float weight_range, float learning_rate, float momentum,
                    float adam_beta);
   // Initializes a trainer from a serialized TFNetworkModel proto.