Ooops, forgot to svn add the new code gen files

Summary: Python generator


git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@664780 13f79535-47bb-0310-9956-ffa450edef68
diff --git a/compiler/cpp/src/generate/t_py_generator.cc b/compiler/cpp/src/generate/t_py_generator.cc
new file mode 100644
index 0000000..1707d03
--- /dev/null
+++ b/compiler/cpp/src/generate/t_py_generator.cc
@@ -0,0 +1,1116 @@
+#include <stdlib.h>
+#include <sys/stat.h>
+#include <sstream>
+#include "t_py_generator.h"
+using namespace std;
+
+/**
+ * Prepares for file generation by opening up the necessary file output
+ * streams.
+ *
+ * @param tprogram The program to generate
+ */
+void t_py_generator::init_generator(t_program* tprogram) {
+  // Make output directory
+  mkdir(T_PY_DIR, S_IREAD | S_IWRITE | S_IEXEC);
+
+  // Make output file
+  string f_types_name = string(T_PY_DIR)+"/"+program_name_+"_types.py";
+  f_types_.open(f_types_name.c_str());
+
+  // Print header
+  f_types_ <<
+    py_autogen_comment() << endl <<
+    py_imports() << endl <<
+    endl;
+}
+
+/**
+ * Autogen'd comment
+ */
+string t_py_generator::py_autogen_comment() {
+  return
+    std::string("#\n") +
+    "# Autogenerated by Thrift\n" +
+    "# " + g_time_str +
+    "#\n" +
+    "# DO NOT EDIT UNLESS YOU ARE SURE THAT YOU KNOW WHAT YOU ARE DOING\n" +
+    "#\n";
+}
+
+/**
+ * Prints standard java imports
+ */
+string t_py_generator::py_imports() {
+  return
+    string("from thrift.protocol.TProtocol import *");
+}
+
+/**
+ * Does nothing in Python
+ */
+void t_py_generator::close_generator(t_program *tprogram) {
+  // Close types file
+  f_types_.close();
+}
+
+/**
+ * Generates a typedef. This is not done in Python, types are all implicit.
+ *
+ * @param ttypedef The type definition
+ */
+void t_py_generator::generate_typedef(t_typedef* ttypedef) {}
+
+/**
+ * Generates code for an enumerated type. Since define is expensive to lookup
+ * in PHP, we use a global array for this.
+ *
+ * @param tenum The enumeration
+ */
+void t_py_generator::generate_enum(t_enum* tenum) {
+  f_types_ <<
+    "class " << tenum->get_name() << ":" << endl;
+  indent_up();
+  
+  vector<t_constant*> constants = tenum->get_constants();
+  vector<t_constant*>::iterator c_iter;
+  int value = -1;
+  for (c_iter = constants.begin(); c_iter != constants.end(); ++c_iter) {
+    if ((*c_iter)->has_value()) {
+      value = (*c_iter)->get_value();
+    } else {
+      ++value;
+    }
+
+    f_types_ <<
+      indent() << (*c_iter)->get_name() << " = " << value << endl;
+  }
+
+  indent_down();
+  f_types_ << endl;
+}
+
+void t_py_generator::generate_struct(t_struct* tstruct) {
+  generate_py_struct(tstruct, false);
+}
+
+/**
+ * Generates a struct definition for a thrift exception. Basically the same
+ * as a struct but extends the Exception class.
+ *
+ * @param txception The struct definition
+ */
+void t_py_generator::generate_xception(t_struct* txception) {
+  generate_py_struct(txception, true);  
+}
+
+void t_py_generator::generate_py_struct(t_struct* tstruct,
+                                        bool is_exception) {
+  generate_py_struct_definition(f_types_, tstruct, is_exception);
+}
+
+/**
+ * Generates a struct definition for a thrift data type. This is nothing in PHP
+ * where the objects are all just associative arrays (unless of course we
+ * decide to start using objects for them...)
+ *
+ * @param tstruct The struct definition
+ */
+void t_py_generator::generate_py_struct_definition(ofstream& out,
+                                                   t_struct* tstruct,
+                                                   bool is_exception,
+                                                   bool is_result) {
+  const vector<t_field*>& members = tstruct->get_members();
+  vector<t_field*>::const_iterator m_iter; 
+
+  out <<
+    "class " << tstruct->get_name();
+  if (is_exception) {
+    out << "(Exception)";
+  }
+  out <<
+    ":" << endl;
+  indent_up();
+
+  out << endl;
+
+  out <<
+    indent() << "def __init__(self):" << endl;
+  indent_up();
+
+  if (members.size() == 0) {
+    indent(out) <<
+      "pass" <<endl;
+  } else {
+    for (m_iter = members.begin(); m_iter != members.end(); ++m_iter) {
+      if (is_result) {
+        indent(out) <<
+          "self." << (*m_iter)->get_name() << " = None" << endl;
+      } else {
+        // This fills in default values, as opposed to nulls
+        indent(out) <<
+          declare_field(*m_iter, true) << endl;
+      }
+    }
+  }
+
+  indent_down();
+ 
+  out << endl;
+
+  generate_py_struct_reader(out, tstruct);
+  if (is_result) {
+    generate_py_struct_result_writer(out, tstruct);
+  } else {
+    generate_py_struct_writer(out, tstruct);
+  }
+
+  indent_down();
+}
+
+void t_py_generator::generate_py_struct_reader(ofstream& out,
+                                                t_struct* tstruct) {
+  const vector<t_field*>& fields = tstruct->get_members();
+  vector<t_field*>::const_iterator f_iter;
+
+  indent(out) <<
+    "def read(self, iprot, itrans):" << endl;
+  indent_up();
+    
+  indent(out) <<
+    "iprot.readStructBegin(itrans)" << endl;   
+
+  // Loop over reading in fields
+  indent(out) <<
+    "while True:" << endl;
+    indent_up();
+    
+    // Read beginning field marker
+    indent(out) <<
+      "(fname, ftype, fid) = iprot.readFieldBegin(itrans)" << endl;
+
+    // Check for field STOP marker and break
+    indent(out) <<
+      "if ftype == TType.STOP:" << endl;
+    indent_up();
+    indent(out) <<
+      "break" << endl;
+    indent_down();
+   
+    // Switch statement on the field we are reading
+    bool first = true;
+    
+    // Generate deserialization code for known cases
+    for (f_iter = fields.begin(); f_iter != fields.end(); ++f_iter) {
+      if (first) {
+        first = false;
+        out <<
+          indent() << "if ";
+      } else {
+        out <<
+          indent() << "elif ";
+      }
+      out << "fid == " << (*f_iter)->get_key() << ":" << endl;
+      indent_up();
+      generate_deserialize_field(out, *f_iter, "self.");
+      indent_down();
+    }
+    
+    // In the default case we skip the field
+    out <<
+      indent() <<  "else:" << endl <<
+      indent() <<  "  iprot.skip(itrans, ftype)" << endl;
+      
+    // Read field end marker
+    indent(out) <<
+      "iprot.readFieldEnd(itrans)" << endl;
+    
+    indent_down();
+    
+    indent(out) <<
+      "iprot.readStructEnd(itrans)" << endl;
+
+    indent_down();
+  out << endl;
+}
+
+void t_py_generator::generate_py_struct_writer(ofstream& out,
+                                               t_struct* tstruct) {
+  string name = tstruct->get_name();
+  const vector<t_field*>& fields = tstruct->get_members();
+  vector<t_field*>::const_iterator f_iter;
+
+  indent(out) <<
+    "def write(self, oprot, otrans):" << endl;
+  indent_up();
+  
+  indent(out) <<
+    "oprot.writeStructBegin(otrans, '" << name << "')" << endl;
+
+  for (f_iter = fields.begin(); f_iter != fields.end(); ++f_iter) {
+    // Write field header
+    indent(out) <<
+      "oprot.writeFieldBegin(otrans, " <<
+      "'" << (*f_iter)->get_name() << "', " <<
+      type_to_enum((*f_iter)->get_type()) << ", " <<
+      (*f_iter)->get_key() << ")" << endl;
+
+    // Write field contents
+    generate_serialize_field(out, *f_iter, "self.");
+
+    // Write field closer
+    indent(out) <<
+      "oprot.writeFieldEnd(otrans)" << endl;
+  }
+
+  // Write the struct map
+  out <<
+    indent() << "oprot.writeFieldStop(otrans)" << endl <<
+    indent() << "oprot.writeStructEnd(otrans)" << endl;
+
+  indent_down();
+  out <<
+    endl;
+}
+
+void t_py_generator::generate_py_struct_result_writer(ofstream& out,
+                                                      t_struct* tstruct) {
+  string name = tstruct->get_name();
+  const vector<t_field*>& fields = tstruct->get_members();
+  vector<t_field*>::const_iterator f_iter;
+
+  indent(out) <<
+    "def write(self, oprot, otrans):" << endl;
+  indent_up();
+  
+  indent(out) <<
+    "oprot.writeStructBegin(otrans, '" << name << "')" << endl;
+
+  for (f_iter = fields.begin(); f_iter != fields.end(); ++f_iter) {
+    // Write field header
+    indent(out) <<
+      "if self." << (*f_iter)->get_name() << " != None:" << endl;
+    indent_up();
+    indent(out) <<
+      "oprot.writeFieldBegin(otrans, " <<
+      "'" << (*f_iter)->get_name() << "', " <<
+      type_to_enum((*f_iter)->get_type()) << ", " <<
+      (*f_iter)->get_key() << ")" << endl;
+
+    // Write field contents
+    generate_serialize_field(out, *f_iter, "self.");
+
+    // Write field closer
+    indent(out) <<
+      "oprot.writeFieldEnd(otrans)" << endl;
+
+    indent_down();
+  }
+
+  // Write the struct map
+  out <<
+    indent() << "oprot.writeFieldStop(otrans)" << endl <<
+    indent() << "oprot.writeStructEnd(otrans)" << endl;
+
+  indent_down();
+  out <<
+    endl;
+}
+
+/**
+ * Generates a thrift service.
+ *
+ * @param tservice The service definition
+ */
+void t_py_generator::generate_service(t_service* tservice) {
+  string f_service_name = string(T_PY_DIR)+"/"+service_name_+".py";
+  f_service_.open(f_service_name.c_str());
+
+  f_service_ <<
+    py_autogen_comment() << endl <<
+    py_imports() << endl;
+
+  f_service_ <<
+    "from " << program_name_ << "_types import *" << endl << endl;
+
+  // Generate the three main parts of the service (well, two for now in PHP)
+  generate_service_interface(tservice);
+  generate_service_client(tservice);
+  generate_service_helpers(tservice);
+  // generate_service_server(tservice);
+  
+  // Close service file
+  f_service_ << endl;
+  f_service_.close();
+}
+
+/**
+ * Generates helper functions for a service.
+ *
+ * @param tservice The service to generate a header definition for
+ */
+void t_py_generator::generate_service_helpers(t_service* tservice) {
+  vector<t_function*> functions = tservice->get_functions();
+  vector<t_function*>::iterator f_iter;
+
+  f_service_ <<
+    "# HELPER FUNCTIONS AND STRUCTURES" << endl << endl;
+
+  for (f_iter = functions.begin(); f_iter != functions.end(); ++f_iter) {
+    t_struct* ts = (*f_iter)->get_arglist();
+    generate_py_struct_definition(f_service_, ts, false);
+    generate_py_function_helpers(*f_iter);
+  }
+}
+
+/**
+ * Generates a struct and helpers for a function.
+ *
+ * @param tfunction The function
+ */
+void t_py_generator::generate_py_function_helpers(t_function* tfunction) {
+  t_struct result(tfunction->get_name() + "_result");
+  t_field success(tfunction->get_returntype(), "success", 0);
+  if (!tfunction->get_returntype()->is_void()) {
+    result.append(&success);
+  }
+
+  t_struct* xs = tfunction->get_xceptions();
+  const vector<t_field*>& fields = xs->get_members();
+  vector<t_field*>::const_iterator f_iter;
+  for (f_iter = fields.begin(); f_iter != fields.end(); ++f_iter) {
+    result.append(*f_iter);
+  }
+
+  generate_py_struct_definition(f_service_, &result, false, true);
+}
+
+/**
+ * Generates a service interface definition.
+ *
+ * @param tservice The service to generate a header definition for
+ */
+void t_py_generator::generate_service_interface(t_service* tservice) {
+  f_service_ <<
+    "class Iface:" << endl;
+  indent_up();
+  vector<t_function*> functions = tservice->get_functions();
+  vector<t_function*>::iterator f_iter; 
+  for (f_iter = functions.begin(); f_iter != functions.end(); ++f_iter) {
+    f_service_ <<
+      indent() << "def " << function_signature(*f_iter) << ":" << endl <<
+      indent() << "  pass" << endl << endl;
+  }
+  indent_down();
+  f_service_ <<
+    endl;
+}
+
+/**
+ * Generates a service client definition.
+ *
+ * @param tservice The service to generate a server for.
+ */
+void t_py_generator::generate_service_client(t_service* tservice) {
+  f_service_ <<
+    "class Client(Iface):" << endl;
+  indent_up();
+
+  // Constructor function
+  f_service_ <<
+    indent() << "def __init__(self, one, two, three=None, four=None):" << endl <<
+    indent() << "  if three == None or four == None:" << endl <<
+    indent() << "    self.__otrans = one" << endl <<
+    indent() << "    self.__itrans = one" << endl <<
+    indent() << "    self.__iprot = two" << endl <<
+    indent() << "    self.__oprot = two" << endl <<
+    indent() << "  else:" << endl <<
+    indent() << "    self.__otrans = one" << endl <<
+    indent() << "    self.__itrans = two" << endl <<
+    indent() << "    self.__iprot = three" << endl <<
+    indent() << "    self.__oprot = four" << endl <<
+    indent() << "  self.__seqid = 0" << endl <<
+    endl;
+
+  // Generate client method implementations
+  vector<t_function*> functions = tservice->get_functions();
+  vector<t_function*>::const_iterator f_iter;    
+  for (f_iter = functions.begin(); f_iter != functions.end(); ++f_iter) {
+    t_struct* arg_struct = (*f_iter)->get_arglist();
+    const vector<t_field*>& fields = arg_struct->get_members();
+    vector<t_field*>::const_iterator fld_iter;
+    string funname = (*f_iter)->get_name();
+
+    // Open function
+    indent(f_service_) <<
+      "def " << function_signature(*f_iter) << ":" << endl;
+    indent_up();
+      indent(f_service_) <<
+        "self.send_" << funname << "(";
+
+      bool first = true;
+      for (fld_iter = fields.begin(); fld_iter != fields.end(); ++fld_iter) {
+        if (first) {
+          first = false;
+        } else {
+          f_service_ << ", ";
+        }
+        f_service_ << (*fld_iter)->get_name();
+      }
+      f_service_ << ")" << endl;
+
+      if (!(*f_iter)->is_async()) {
+        f_service_ << indent();
+        if (!(*f_iter)->get_returntype()->is_void()) {
+          f_service_ << "return ";
+        }
+        f_service_ <<
+          "self.recv_" << funname << "()" << endl;
+      }
+    indent_down();
+    f_service_ << endl;
+
+    indent(f_service_) <<
+      "def send_" << function_signature(*f_iter) << ":" << endl;
+    indent_up();
+
+      std::string argsname = (*f_iter)->get_name() + "_args";
+
+      // Serialize the request header
+      f_service_ <<
+        indent() << "self.__oprot.writeMessageBegin(self.__otrans, '" << (*f_iter)->get_name() << "', TMessageType.CALL, self.__seqid)" << endl;
+      
+      f_service_ <<
+        indent() << "__args = " << argsname << "()" << endl;
+      
+      for (fld_iter = fields.begin(); fld_iter != fields.end(); ++fld_iter) {
+        f_service_ <<
+          indent() << "__args." << (*fld_iter)->get_name() << " = " << (*fld_iter)->get_name() << endl;
+      }
+           
+      // Write to the stream
+      f_service_ <<
+        indent() << "__args.write(self.__oprot, self.__otrans)" << endl <<
+        indent() << "self.__oprot.writeMessageEnd(self.__otrans)" << endl <<
+        indent() << "self.__otrans.flush()" << endl;
+
+    indent_down();
+      
+
+    if (!(*f_iter)->is_async()) {
+      std::string resultname = (*f_iter)->get_name() + "_result";
+      t_struct noargs;
+      
+      t_function recv_function((*f_iter)->get_returntype(),
+                               string("recv_") + (*f_iter)->get_name(),
+                               &noargs);
+      // Open function
+      f_service_ <<
+        endl <<
+        indent() << "def " << function_signature(&recv_function) << ":" << endl;
+      indent_up();
+
+      f_service_ <<
+        indent() << "(fname, mtype, rseqid) = self.__iprot.readMessageBegin(self.__itrans)" << endl;
+
+      // TODO(mcslee): Validate message reply here
+
+      f_service_ <<
+        indent() << "__result = " << resultname << "()" << endl <<
+        indent() << "__result.read(self.__iprot, self.__otrans)" << endl <<
+        indent() << "self.__iprot.readMessageEnd(self.__itrans)" << endl;
+
+      // Careful, only return _result if not a void function
+      if (!(*f_iter)->get_returntype()->is_void()) {
+        f_service_ <<
+          indent() << "if __result.success != None:" << endl <<
+          indent() << "  return __result.success" << endl;
+      }
+
+      t_struct* xs = (*f_iter)->get_xceptions();
+      const std::vector<t_field*>& xceptions = xs->get_members();
+      vector<t_field*>::const_iterator x_iter;
+      for (x_iter = xceptions.begin(); x_iter != xceptions.end(); ++x_iter) {
+        f_service_ <<
+          indent() << "if __result." << (*x_iter)->get_name() << " != None:" << endl <<
+          indent() << "  raise __result." << (*x_iter)->get_name() << "" << endl;
+      }
+
+      // Careful, only return _result if not a void function
+      if ((*f_iter)->get_returntype()->is_void()) {
+        indent(f_service_) <<
+          "return" << endl;
+      } else {
+        f_service_ <<
+          indent() << "raise Exception(\"" << (*f_iter)->get_name() << " failed: unknown result\");" << endl;
+      }     
+    }      
+
+    // Close function
+    indent_down();
+    f_service_ << endl;
+    
+  }
+
+  indent_down();
+  f_service_ <<
+    endl;
+}
+
+/**
+ * Deserializes a field of any type.
+ */
+void t_py_generator::generate_deserialize_field(ofstream &out,
+                                                t_field* tfield,
+                                                string prefix,
+                                                bool inclass) {
+  t_type* type = tfield->get_type();
+  while (type->is_typedef()) {
+    type = ((t_typedef*)type)->get_type();
+  }
+
+  if (type->is_void()) {
+    throw "CANNOT GENERATE DESERIALIZE CODE FOR void TYPE: " +
+      prefix + tfield->get_name();
+  }
+
+  string name = prefix + tfield->get_name();
+
+  if (type->is_struct() || type->is_xception()) {
+    generate_deserialize_struct(out,
+                                (t_struct*)(tfield->get_type()),
+                                 name);
+  } else if (type->is_container()) {
+    generate_deserialize_container(out, tfield->get_type(), name);
+  } else if (type->is_base_type() || type->is_enum()) {
+    indent(out) <<
+      name << " = iprot.";
+    
+    if (type->is_base_type()) {
+      t_base_type::t_base tbase = ((t_base_type*)type)->get_base();
+      switch (tbase) {
+      case t_base_type::TYPE_VOID:
+        throw "compiler error: cannot serialize void field in a struct: " +
+          name;
+        break;
+      case t_base_type::TYPE_STRING:        
+        out << "readString(itrans);";
+        break;
+      case t_base_type::TYPE_BOOL:
+        out << "readBool(itrans);";
+        break;
+      case t_base_type::TYPE_BYTE:
+        out << "readByte(itrans);";
+        break;
+      case t_base_type::TYPE_I16:
+        out << "readI16(itrans);";
+        break;
+      case t_base_type::TYPE_I32:
+        out << "readI32(itrans);";
+        break;
+      case t_base_type::TYPE_I64:
+        out << "readI64(itrans);";
+        break;
+      default:
+        throw "compiler error: no PHP name for base type " + tbase;
+      }
+    } else if (type->is_enum()) {
+      out << "readI32(itrans);";
+    }
+    out << endl;
+
+  } else {
+    printf("DO NOT KNOW HOW TO DESERIALIZE FIELD '%s' TYPE '%s'\n",
+           tfield->get_name().c_str(), type_name(type).c_str());
+  }  
+}
+
+/**
+ * Generates an unserializer for a variable. This makes two key assumptions,
+ * first that there is a const char* variable named data that points to the
+ * buffer for deserialization, and that there is a variable protocol which
+ * is a reference to a TProtocol serialization object.
+ */
+void t_py_generator::generate_deserialize_struct(ofstream &out,
+                                                  t_struct* tstruct,
+                                                  string prefix) {
+  out <<
+    indent() << prefix << " = " << tstruct->get_name() << "()" << endl <<
+    indent() << prefix << ".read(iprot, itrans)" << endl;
+}
+
+void t_py_generator::generate_deserialize_container(ofstream &out,
+                                                     t_type* ttype,
+                                                     string prefix) {
+  string size = tmp("_size");
+  string ktype = tmp("_ktype");
+  string vtype = tmp("_vtype");
+  string etype = tmp("_etype");
+  
+  t_field fsize(g_program->get_i32_type(), size);
+  t_field fktype(g_program->get_byte_type(), ktype);
+  t_field fvtype(g_program->get_byte_type(), vtype);
+  t_field fetype(g_program->get_byte_type(), etype);
+
+  // Declare variables, read header
+  if (ttype->is_map()) {
+    out <<
+      indent() << "(" << ktype << ", " << vtype << ", " << size << " ) = iprot.readMapBegin(itrans) " << endl;
+  } else if (ttype->is_set()) {
+    out <<
+      indent() << "(" << etype << ", " << size << ") = iprot.readSetBegin(itrans)" << endl;
+  } else if (ttype->is_list()) {
+    out <<
+      indent() << "(" << etype << ", " << size << ") = iprot.readListBegin(itrans)" << endl;
+  }
+
+  // For loop iterates over elements
+  string i = tmp("_i");
+  indent(out) <<
+    "for " << i << " in xrange(" << size << "):" << endl;
+  
+    indent_up();
+    
+    if (ttype->is_map()) {
+      generate_deserialize_map_element(out, (t_map*)ttype, prefix);
+    } else if (ttype->is_set()) {
+      generate_deserialize_set_element(out, (t_set*)ttype, prefix);
+    } else if (ttype->is_list()) {
+      generate_deserialize_list_element(out, (t_list*)ttype, prefix);
+    }
+    
+    indent_down();
+
+  // Read container end
+  if (ttype->is_map()) {
+    indent(out) << "iprot.readMapEnd(itrans)" << endl;
+  } else if (ttype->is_set()) {
+    indent(out) << "iprot.readSetEnd(itrans)" << endl;
+  } else if (ttype->is_list()) {
+    indent(out) << "iprot.readListEnd(itrans)" << endl;
+  }
+}
+
+
+/**
+ * Generates code to deserialize a map
+ */
+void t_py_generator::generate_deserialize_map_element(ofstream &out,
+                                                       t_map* tmap,
+                                                       string prefix) {
+  string key = tmp("_key");
+  string val = tmp("_val");
+  t_field fkey(tmap->get_key_type(), key);
+  t_field fval(tmap->get_val_type(), val);
+
+  /*
+  indent(out) <<
+    declare_field(&fkey, true, true) << endl;
+  indent(out) <<
+    declare_field(&fval, true, true) << endl;
+  */
+
+  generate_deserialize_field(out, &fkey);
+  generate_deserialize_field(out, &fval);
+
+  indent(out) <<
+    prefix << "[" << key << "] = " << val << endl;
+}
+
+void t_py_generator::generate_deserialize_set_element(ofstream &out,
+                                                       t_set* tset,
+                                                       string prefix) {
+  string elem = tmp("_elem");
+  t_field felem(tset->get_elem_type(), elem);
+
+  /*
+  indent(out) <<
+    "$" << elem << " = null;" << endl;
+  */
+
+  generate_deserialize_field(out, &felem);
+
+  indent(out) <<
+    prefix << ".append(" << elem << ")" << endl;
+}
+
+void t_py_generator::generate_deserialize_list_element(ofstream &out,
+                                                        t_list* tlist,
+                                                        string prefix) {
+  string elem = tmp("_elem");
+  t_field felem(tlist->get_elem_type(), elem);
+
+  /*
+  indent(out) <<
+    "$" << elem << " = null;" << endl;
+  */
+
+  generate_deserialize_field(out, &felem);
+
+  indent(out) <<
+    prefix << ".append(" << elem << ")" << endl;
+}
+
+
+/**
+ * Serializes a field of any type.
+ *
+ * @param tfield The field to serialize
+ * @param prefix Name to prepend to field name
+ */
+void t_py_generator::generate_serialize_field(ofstream &out,
+                                               t_field* tfield,
+                                               string prefix) {
+  t_type* type = tfield->get_type();
+  while (type->is_typedef()) {
+    type = ((t_typedef*)type)->get_type();
+  }
+
+  // Do nothing for void types
+  if (type->is_void()) {
+    throw "CANNOT GENERATE SERIALIZE CODE FOR void TYPE: " +
+      prefix + tfield->get_name();
+  }
+  
+  if (type->is_struct() || type->is_xception()) {
+    generate_serialize_struct(out,
+                              (t_struct*)(tfield->get_type()),
+                              prefix + tfield->get_name());
+  } else if (type->is_container()) {
+    generate_serialize_container(out,
+                                 tfield->get_type(),
+                                 prefix + tfield->get_name());
+  } else if (type->is_base_type() || type->is_enum()) {
+
+    string name = prefix + tfield->get_name();
+
+    indent(out) <<
+      "oprot.";
+    
+    if (type->is_base_type()) {
+      t_base_type::t_base tbase = ((t_base_type*)type)->get_base();
+      switch (tbase) {
+      case t_base_type::TYPE_VOID:
+        throw
+          "compiler error: cannot serialize void field in a struct: " + name;
+        break;
+      case t_base_type::TYPE_STRING:
+        out << "writeString(otrans, " << name << ")";
+        break;
+      case t_base_type::TYPE_BOOL:
+        out << "writeBool(otrans, " << name << ")";
+        break;
+      case t_base_type::TYPE_BYTE:
+        out << "writeByte(otrans, " << name << ")";
+        break;
+      case t_base_type::TYPE_I16:
+        out << "writeI16(otrans, " << name << ")";
+        break;
+      case t_base_type::TYPE_I32:
+        out << "writeI32(otrans, " << name << ")";
+        break;
+      case t_base_type::TYPE_I64:
+        out << "writeI64(otrans, " << name << ")";
+        break;
+      default:
+        throw "compiler error: no PHP name for base type " + tbase;
+      }
+    } else if (type->is_enum()) {
+      out << "writeI32(otrans, " << name << ")";
+    }
+    out << endl;
+  } else {
+    printf("DO NOT KNOW HOW TO SERIALIZE FIELD '%s%s' TYPE '%s'\n",
+           prefix.c_str(),
+           tfield->get_name().c_str(),
+           type_name(type).c_str());
+  }
+}
+
+/**
+ * Serializes all the members of a struct.
+ *
+ * @param tstruct The struct to serialize
+ * @param prefix  String prefix to attach to all fields
+ */
+void t_py_generator::generate_serialize_struct(ofstream &out,
+                                                t_struct* tstruct,
+                                                string prefix) {
+  indent(out) <<
+    prefix << ".write(oprot, otrans)" << endl;
+}
+
+void t_py_generator::generate_serialize_container(ofstream &out,
+                                                   t_type* ttype,
+                                                   string prefix) {
+  if (ttype->is_map()) {
+    indent(out) <<
+      "oprot.writeMapBegin(otrans, " <<
+      type_to_enum(((t_map*)ttype)->get_key_type()) << ", " <<
+      type_to_enum(((t_map*)ttype)->get_val_type()) << ", " <<
+      "len(" << prefix << "))" << endl;
+  } else if (ttype->is_set()) {
+    indent(out) <<
+      "oprot.writeSetBegin(otrans, " <<
+      type_to_enum(((t_set*)ttype)->get_elem_type()) << ", " <<
+      "len(" << prefix << "))" << endl;
+  } else if (ttype->is_list()) {
+    indent(out) <<
+      "oprot.writeListBegin(otrans, " <<
+      type_to_enum(((t_list*)ttype)->get_elem_type()) << ", " <<
+      "len(" << prefix << "))" << endl;
+  }
+
+    if (ttype->is_map()) {
+      string kiter = tmp("_kiter");
+      string viter = tmp("_viter");
+      indent(out) << 
+        "for " << kiter << "," << viter << " in " << prefix << ":" << endl;
+      indent_up();
+      generate_serialize_map_element(out, (t_map*)ttype, kiter, viter);
+      indent_down();
+    } else if (ttype->is_set()) {
+      string iter = tmp("_iter");
+      indent(out) << 
+        "for " << iter << " in " << prefix << ":" << endl;
+      indent_up();
+      generate_serialize_set_element(out, (t_set*)ttype, iter);
+      indent_down();
+    } else if (ttype->is_list()) {
+      string iter = tmp("_iter");
+      indent(out) << 
+        "for " << iter << " in " << prefix << ":" << endl;
+      indent_up();
+      generate_serialize_list_element(out, (t_list*)ttype, iter);
+      indent_down();
+    }
+    
+
+  if (ttype->is_map()) {
+    indent(out) <<
+      "oprot.writeMapEnd(otrans)" << endl;
+  } else if (ttype->is_set()) {
+    indent(out) <<
+      "oprot.writeSetEnd(otrans)" << endl;
+  } else if (ttype->is_list()) {
+    indent(out) <<
+      "oprot.writeListEnd(otrans)" << endl;
+  }
+}
+
+/**
+ * Serializes the members of a map.
+ *
+ */
+void t_py_generator::generate_serialize_map_element(ofstream &out,
+                                                     t_map* tmap,
+                                                     string kiter,
+                                                     string viter) {
+  t_field kfield(tmap->get_key_type(), kiter);
+  generate_serialize_field(out, &kfield, "");
+
+  t_field vfield(tmap->get_val_type(), viter);
+  generate_serialize_field(out, &vfield, "");
+}
+
+/**
+ * Serializes the members of a set.
+ */
+void t_py_generator::generate_serialize_set_element(ofstream &out,
+                                                     t_set* tset,
+                                                     string iter) {
+  t_field efield(tset->get_elem_type(), iter);
+  generate_serialize_field(out, &efield, "");
+}
+
+/**
+ * Serializes the members of a list.
+ */
+void t_py_generator::generate_serialize_list_element(ofstream &out,
+                                                      t_list* tlist,
+                                                      string iter) {
+  t_field efield(tlist->get_elem_type(), iter);
+  generate_serialize_field(out, &efield, "");
+}
+
+/**
+ * Returns a Java type name
+ *
+ * @param ttype The type
+ */
+string t_py_generator::type_name(t_type* ttype) {
+  // In Java typedefs are just resolved to their real type
+  while (ttype->is_typedef()) {
+    ttype = ((t_typedef*)ttype)->get_type();
+  }
+
+  if (ttype->is_base_type()) {
+    return base_type_name(((t_base_type*)ttype)->get_base());
+  } else if (ttype->is_enum()) {
+    return "Int32";
+  } else if (ttype->is_map()) {
+    t_map* tmap = (t_map*) ttype;
+    return "HashMap<" +
+      type_name(tmap->get_key_type()) + "," +
+      type_name(tmap->get_val_type()) + ">";
+  } else if (ttype->is_set()) {
+    t_set* tset = (t_set*) ttype;
+    return "HashSet<" + type_name(tset->get_elem_type()) + ">";
+  } else if (ttype->is_list()) {
+    t_list* tlist = (t_list*) ttype;
+    return "ArrayList<" + type_name(tlist->get_elem_type()) + ">";
+  } else {
+    return ttype->get_name();
+  }
+}
+
+/**
+ * Returns the C++ type that corresponds to the thrift type.
+ *
+ * @param tbase The base type
+ */
+string t_py_generator::base_type_name(t_base_type::t_base tbase) {
+  switch (tbase) {
+  case t_base_type::TYPE_VOID:
+    return "void";
+  case t_base_type::TYPE_STRING:
+    return "TString";
+  case t_base_type::TYPE_BOOL:
+    return "bool";
+  case t_base_type::TYPE_BYTE:
+    return "UInt8";
+  case t_base_type::TYPE_I16:
+    return "Int16";
+  case t_base_type::TYPE_I32:
+    return "Int32";
+  case t_base_type::TYPE_I64:
+    return "Int64";
+  default:
+    throw "compiler error: no PHP name for base type " + tbase;
+  }
+}
+
+/**
+ * Declares a field, which may include initialization as necessary.
+ *
+ * @param ttype The type
+ */
+string t_py_generator::declare_field(t_field* tfield, bool init, bool obj) {
+  string result = "self." + tfield->get_name();
+  if (init) {
+    t_type* type = tfield->get_type();
+    while (type->is_typedef()) {
+      type = ((t_typedef*)type)->get_type();
+    }
+    if (type->is_base_type()) {
+      t_base_type::t_base tbase = ((t_base_type*)type)->get_base();
+      switch (tbase) {
+      case t_base_type::TYPE_VOID:
+        break;
+      case t_base_type::TYPE_STRING:
+        result += " = ''";
+        break;
+      case t_base_type::TYPE_BOOL:
+        result += " = false";
+        break;
+      case t_base_type::TYPE_BYTE:
+      case t_base_type::TYPE_I16:
+      case t_base_type::TYPE_I32:
+      case t_base_type::TYPE_I64:
+        result += " = 0";
+        break;
+      default:
+        throw "compiler error: no PHP initializer for base type " + tbase;
+      }
+    } else if (type->is_enum()) {
+      result += " = 0";
+    } else if (type->is_container()) {
+      if (type->is_map()) {
+        result += " = {}";
+      } else {
+        result += " = []";
+      }
+    } else if (type->is_struct() || type->is_xception()) {
+      if (obj) {
+        result += " = " + type->get_name() + "()";
+      } else {
+        result += " = None";
+      }
+    }
+  }
+  return result;
+}
+
+/**
+ * Renders a function signature of the form 'type name(args)'
+ *
+ * @param tfunction Function definition
+ * @return String of rendered function definition
+ */
+string t_py_generator::function_signature(t_function* tfunction,
+                                           string prefix) {
+  return
+    prefix + tfunction->get_name() +
+    "(self, " + argument_list(tfunction->get_arglist()) + ")";
+}
+
+/**
+ * Renders a field list
+ */
+string t_py_generator::argument_list(t_struct* tstruct) {
+  string result = "";
+
+  const vector<t_field*>& fields = tstruct->get_members();
+  vector<t_field*>::const_iterator f_iter;
+  bool first = true;
+  for (f_iter = fields.begin(); f_iter != fields.end(); ++f_iter) {
+    if (first) {
+      first = false;
+    } else {
+      result += ", ";
+    }
+    result += (*f_iter)->get_name();
+  }
+  return result;
+}
+
+/**
+ * Converts the parse type to a C++ enum string for the given type.
+ */
+string t_py_generator::type_to_enum(t_type* type) {
+  while (type->is_typedef()) {
+    type = ((t_typedef*)type)->get_type();
+  }
+  
+  if (type->is_base_type()) {
+    t_base_type::t_base tbase = ((t_base_type*)type)->get_base();
+    switch (tbase) {
+    case t_base_type::TYPE_VOID:
+      throw "NO T_VOID CONSTRUCT";
+    case t_base_type::TYPE_STRING:
+      return "TType.STRING";
+    case t_base_type::TYPE_BOOL:
+      return "TType.BOOL";
+    case t_base_type::TYPE_BYTE:
+      return "TType.BYTE";
+    case t_base_type::TYPE_I16:
+      return "TType.I16";
+    case t_base_type::TYPE_I32:
+      return "TType.I32";
+    case t_base_type::TYPE_I64:
+      return "TType.I64";
+    }
+  } else if (type->is_enum()) {
+    return "TType.I32";
+  } else if (type->is_struct() || type->is_xception()) {
+    return "TType.STRUCT";
+  } else if (type->is_map()) {
+    return "TType.MAP";
+  } else if (type->is_set()) {
+    return "TType.SET";
+  } else if (type->is_list()) {
+    return "TType.LIST";
+  }
+
+  throw "INVALID TYPE IN type_to_enum: " + type->get_name();
+}
diff --git a/compiler/cpp/src/generate/t_py_generator.h b/compiler/cpp/src/generate/t_py_generator.h
new file mode 100644
index 0000000..2703df3
--- /dev/null
+++ b/compiler/cpp/src/generate/t_py_generator.h
@@ -0,0 +1,122 @@
+#ifndef T_PY_GENERATOR_H
+#define T_PY_GENERATOR_H
+
+#include <string>
+#include <fstream>
+#include <iostream>
+#include <vector>
+
+#include "t_oop_generator.h"
+
+#define T_PY_DIR "gen-py"
+
+/**
+ * Python code generator.
+ *
+ * @author Mark Slee <mcslee@facebook.com>
+ */
+class t_py_generator : public t_oop_generator {
+ public:
+  t_py_generator() {}
+  ~t_py_generator() {}
+
+  /** Init and close methods */
+
+  void init_generator(t_program *tprogram);
+  void close_generator(t_program *tprogram);
+
+  /** Program-level generation functions */
+
+  void generate_typedef  (t_typedef*  ttypedef);
+  void generate_enum     (t_enum*     tenum);
+  void generate_struct   (t_struct*   tstruct);
+  void generate_xception (t_struct*   txception);
+  void generate_service  (t_service*  tservice);
+
+
+  void generate_py_struct(t_struct* tstruct, bool is_exception);
+  void generate_py_struct_definition(std::ofstream& out, t_struct* tstruct, bool is_xception=false, bool is_result=false);
+  void generate_py_struct_reader(std::ofstream& out, t_struct* tstruct);
+  void generate_py_struct_result_writer(std::ofstream& out, t_struct* tstruct);
+  void generate_py_struct_writer(std::ofstream& out, t_struct* tstruct);
+
+  void generate_py_function_helpers  (t_function* tfunction);
+
+  /** Service-level generation functions */
+
+  void generate_service_helpers(t_service*  tservice);
+  void generate_service_interface (t_service* tservice);
+  void generate_service_client    (t_service* tservice);
+
+  /** Serialization constructs */
+
+  void generate_deserialize_field        (std::ofstream &out,
+                                          t_field*    tfield, 
+                                          std::string prefix="",
+                                          bool inclass=false);
+  
+  void generate_deserialize_struct       (std::ofstream &out,
+                                          t_struct*   tstruct,
+                                          std::string prefix="");
+  
+  void generate_deserialize_container    (std::ofstream &out,
+                                          t_type*     ttype,
+                                          std::string prefix="");
+  
+  void generate_deserialize_set_element  (std::ofstream &out,
+                                          t_set*      tset,
+                                          std::string prefix="");
+
+  void generate_deserialize_map_element  (std::ofstream &out,
+                                          t_map*      tmap,
+                                          std::string prefix="");
+
+  void generate_deserialize_list_element (std::ofstream &out,
+                                          t_list*     tlist,
+                                          std::string prefix="");
+
+  void generate_serialize_field          (std::ofstream &out,
+                                          t_field*    tfield,
+                                          std::string prefix="");
+
+  void generate_serialize_struct         (std::ofstream &out,
+                                          t_struct*   tstruct,
+                                          std::string prefix="");
+
+  void generate_serialize_container      (std::ofstream &out,
+                                          t_type*     ttype,
+                                          std::string prefix="");
+
+  void generate_serialize_map_element    (std::ofstream &out,
+                                          t_map*      tmap,
+                                          std::string kiter,
+                                          std::string viter);
+
+  void generate_serialize_set_element    (std::ofstream &out,
+                                          t_set*      tmap,
+                                          std::string iter);
+
+  void generate_serialize_list_element   (std::ofstream &out,
+                                          t_list*     tlist,
+                                          std::string iter);
+
+  /** Helper rendering functions */
+
+  std::string py_autogen_comment();
+  std::string py_imports();
+  std::string type_name(t_type* ttype);
+  std::string base_type_name(t_base_type::t_base tbase);
+  std::string declare_field(t_field* tfield, bool init=false, bool obj=false);
+  std::string function_signature(t_function* tfunction, std::string prefix="");
+  std::string argument_list(t_struct* tstruct);
+  std::string type_to_enum(t_type* ttype);
+
+ private:
+
+  /** File streams */
+  std::ofstream f_types_;
+  std::ofstream f_service_;
+
+};
+
+#endif