1 // Protocol Buffers - Google's data interchange format
2 // Copyright 2008 Google Inc. All rights reserved.
3 // http://code.google.com/p/protobuf/
4 //
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are
7 // met:
8 //
9 // * Redistributions of source code must retain the above copyright
10 // notice, this list of conditions and the following disclaimer.
11 // * Redistributions in binary form must reproduce the above
12 // copyright notice, this list of conditions and the following disclaimer
13 // in the documentation and/or other materials provided with the
14 // distribution.
15 // * Neither the name of Google Inc. nor the names of its
16 // contributors may be used to endorse or promote products derived from
17 // this software without specific prior written permission.
18 //
19 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
30
31 // Author: robinson@google.com (Will Robinson)
32 //
33 // This module outputs pure-Python protocol message classes that will
34 // largely be constructed at runtime via the metaclass in reflection.py.
35 // In other words, our job is basically to output a Python equivalent
36 // of the C++ *Descriptor objects, and fix up all circular references
37 // within these objects.
38 //
39 // Note that the runtime performance of protocol message classes created in
40 // this way is expected to be lousy. The plan is to create an alternate
41 // generator that outputs a Python/C extension module that lets
42 // performance-minded Python code leverage the fast C++ implementation
43 // directly.
44
45 #include <limits>
46 #include <map>
47 #include <utility>
48 #include <string>
49 #include <vector>
50
51 #include <google/protobuf/compiler/python/python_generator.h>
52 #include <google/protobuf/descriptor.pb.h>
53
54 #include <google/protobuf/stubs/common.h>
55 #include <google/protobuf/io/printer.h>
56 #include <google/protobuf/descriptor.h>
57 #include <google/protobuf/io/zero_copy_stream.h>
58 #include <google/protobuf/stubs/strutil.h>
59 #include <google/protobuf/stubs/substitute.h>
60
61 namespace google {
62 namespace protobuf {
63 namespace compiler {
64 namespace python {
65
66 namespace {
67
68 // Returns a copy of |filename| with any trailing ".protodevel" or ".proto
69 // suffix stripped.
70 // TODO(robinson): Unify with copy in compiler/cpp/internal/helpers.cc.
StripProto(const string & filename)71 string StripProto(const string& filename) {
72 const char* suffix = HasSuffixString(filename, ".protodevel")
73 ? ".protodevel" : ".proto";
74 return StripSuffixString(filename, suffix);
75 }
76
77
78 // Returns the Python module name expected for a given .proto filename.
ModuleName(const string & filename)79 string ModuleName(const string& filename) {
80 string basename = StripProto(filename);
81 StripString(&basename, "-", '_');
82 StripString(&basename, "/", '.');
83 return basename + "_pb2";
84 }
85
86
87 // Returns the name of all containing types for descriptor,
88 // in order from outermost to innermost, followed by descriptor's
89 // own name. Each name is separated by |separator|.
90 template <typename DescriptorT>
NamePrefixedWithNestedTypes(const DescriptorT & descriptor,const string & separator)91 string NamePrefixedWithNestedTypes(const DescriptorT& descriptor,
92 const string& separator) {
93 string name = descriptor.name();
94 for (const Descriptor* current = descriptor.containing_type();
95 current != NULL; current = current->containing_type()) {
96 name = current->name() + separator + name;
97 }
98 return name;
99 }
100
101
102 // Name of the class attribute where we store the Python
103 // descriptor.Descriptor instance for the generated class.
104 // Must stay consistent with the _DESCRIPTOR_KEY constant
105 // in proto2/public/reflection.py.
106 const char kDescriptorKey[] = "DESCRIPTOR";
107
108
109 // Should we generate generic services for this file?
HasGenericServices(const FileDescriptor * file)110 inline bool HasGenericServices(const FileDescriptor *file) {
111 return file->service_count() > 0 &&
112 file->options().py_generic_services();
113 }
114
115
116 // Prints the common boilerplate needed at the top of every .py
117 // file output by this generator.
PrintTopBoilerplate(io::Printer * printer,const FileDescriptor * file,bool descriptor_proto)118 void PrintTopBoilerplate(
119 io::Printer* printer, const FileDescriptor* file, bool descriptor_proto) {
120 // TODO(robinson): Allow parameterization of Python version?
121 printer->Print(
122 "# Generated by the protocol buffer compiler. DO NOT EDIT!\n"
123 "\n"
124 "from google.protobuf import descriptor\n"
125 "from google.protobuf import message\n"
126 "from google.protobuf import reflection\n");
127 if (HasGenericServices(file)) {
128 printer->Print(
129 "from google.protobuf import service\n"
130 "from google.protobuf import service_reflection\n");
131 }
132
133 // Avoid circular imports if this module is descriptor_pb2.
134 if (!descriptor_proto) {
135 printer->Print(
136 "from google.protobuf import descriptor_pb2\n");
137 }
138 printer->Print(
139 "# @@protoc_insertion_point(imports)\n");
140 printer->Print("\n\n");
141 }
142
143
144 // Returns a Python literal giving the default value for a field.
145 // If the field specifies no explicit default value, we'll return
146 // the default default value for the field type (zero for numbers,
147 // empty string for strings, empty list for repeated fields, and
148 // None for non-repeated, composite fields).
149 //
150 // TODO(robinson): Unify with code from
151 // //compiler/cpp/internal/primitive_field.cc
152 // //compiler/cpp/internal/enum_field.cc
153 // //compiler/cpp/internal/string_field.cc
StringifyDefaultValue(const FieldDescriptor & field)154 string StringifyDefaultValue(const FieldDescriptor& field) {
155 if (field.is_repeated()) {
156 return "[]";
157 }
158
159 switch (field.cpp_type()) {
160 case FieldDescriptor::CPPTYPE_INT32:
161 return SimpleItoa(field.default_value_int32());
162 case FieldDescriptor::CPPTYPE_UINT32:
163 return SimpleItoa(field.default_value_uint32());
164 case FieldDescriptor::CPPTYPE_INT64:
165 return SimpleItoa(field.default_value_int64());
166 case FieldDescriptor::CPPTYPE_UINT64:
167 return SimpleItoa(field.default_value_uint64());
168 case FieldDescriptor::CPPTYPE_DOUBLE: {
169 double value = field.default_value_double();
170 if (value == numeric_limits<double>::infinity()) {
171 // Python pre-2.6 on Windows does not parse "inf" correctly. However,
172 // a numeric literal that is too big for a double will become infinity.
173 return "1e10000";
174 } else if (value == -numeric_limits<double>::infinity()) {
175 // See above.
176 return "-1e10000";
177 } else if (value != value) {
178 // infinity * 0 = nan
179 return "(1e10000 * 0)";
180 } else {
181 return SimpleDtoa(value);
182 }
183 }
184 case FieldDescriptor::CPPTYPE_FLOAT: {
185 float value = field.default_value_float();
186 if (value == numeric_limits<float>::infinity()) {
187 // Python pre-2.6 on Windows does not parse "inf" correctly. However,
188 // a numeric literal that is too big for a double will become infinity.
189 return "1e10000";
190 } else if (value == -numeric_limits<float>::infinity()) {
191 // See above.
192 return "-1e10000";
193 } else if (value != value) {
194 // infinity - infinity = nan
195 return "(1e10000 * 0)";
196 } else {
197 return SimpleFtoa(value);
198 }
199 }
200 case FieldDescriptor::CPPTYPE_BOOL:
201 return field.default_value_bool() ? "True" : "False";
202 case FieldDescriptor::CPPTYPE_ENUM:
203 return SimpleItoa(field.default_value_enum()->number());
204 case FieldDescriptor::CPPTYPE_STRING:
205 if (field.type() == FieldDescriptor::TYPE_STRING) {
206 return "unicode(\"" + CEscape(field.default_value_string()) +
207 "\", \"utf-8\")";
208 } else {
209 return "\"" + CEscape(field.default_value_string()) + "\"";
210 }
211 case FieldDescriptor::CPPTYPE_MESSAGE:
212 return "None";
213 }
214 // (We could add a default case above but then we wouldn't get the nice
215 // compiler warning when a new type is added.)
216 GOOGLE_LOG(FATAL) << "Not reached.";
217 return "";
218 }
219
220
221
222 } // namespace
223
224
Generator()225 Generator::Generator() : file_(NULL) {
226 }
227
~Generator()228 Generator::~Generator() {
229 }
230
Generate(const FileDescriptor * file,const string & parameter,OutputDirectory * output_directory,string * error) const231 bool Generator::Generate(const FileDescriptor* file,
232 const string& parameter,
233 OutputDirectory* output_directory,
234 string* error) const {
235
236 // Completely serialize all Generate() calls on this instance. The
237 // thread-safety constraints of the CodeGenerator interface aren't clear so
238 // just be as conservative as possible. It's easier to relax this later if
239 // we need to, but I doubt it will be an issue.
240 // TODO(kenton): The proper thing to do would be to allocate any state on
241 // the stack and use that, so that the Generator class itself does not need
242 // to have any mutable members. Then it is implicitly thread-safe.
243 MutexLock lock(&mutex_);
244 file_ = file;
245 string module_name = ModuleName(file->name());
246 string filename = module_name;
247 StripString(&filename, ".", '/');
248 filename += ".py";
249
250 FileDescriptorProto fdp;
251 file_->CopyTo(&fdp);
252 fdp.SerializeToString(&file_descriptor_serialized_);
253
254
255 scoped_ptr<io::ZeroCopyOutputStream> output(output_directory->Open(filename));
256 GOOGLE_CHECK(output.get());
257 io::Printer printer(output.get(), '$');
258 printer_ = &printer;
259
260 PrintTopBoilerplate(printer_, file_, GeneratingDescriptorProto());
261 PrintFileDescriptor();
262 PrintTopLevelEnums();
263 PrintTopLevelExtensions();
264 PrintAllNestedEnumsInFile();
265 PrintMessageDescriptors();
266 // We have to print the imports after the descriptors, so that mutually
267 // recursive protos in separate files can successfully reference each other.
268 PrintImports();
269 FixForeignFieldsInDescriptors();
270 PrintMessages();
271 // We have to fix up the extensions after the message classes themselves,
272 // since they need to call static RegisterExtension() methods on these
273 // classes.
274 FixForeignFieldsInExtensions();
275 if (HasGenericServices(file)) {
276 PrintServices();
277 }
278
279 printer.Print(
280 "# @@protoc_insertion_point(module_scope)\n");
281
282 return !printer.failed();
283 }
284
285 // Prints Python imports for all modules imported by |file|.
PrintImports() const286 void Generator::PrintImports() const {
287 for (int i = 0; i < file_->dependency_count(); ++i) {
288 string module_name = ModuleName(file_->dependency(i)->name());
289 printer_->Print("import $module$\n", "module",
290 module_name);
291 }
292 printer_->Print("\n");
293 }
294
295 // Prints the single file descriptor for this file.
PrintFileDescriptor() const296 void Generator::PrintFileDescriptor() const {
297 map<string, string> m;
298 m["descriptor_name"] = kDescriptorKey;
299 m["name"] = file_->name();
300 m["package"] = file_->package();
301 const char file_descriptor_template[] =
302 "$descriptor_name$ = descriptor.FileDescriptor(\n"
303 " name='$name$',\n"
304 " package='$package$',\n";
305 printer_->Print(m, file_descriptor_template);
306 printer_->Indent();
307 printer_->Print(
308 "serialized_pb='$value$'",
309 "value", strings::CHexEscape(file_descriptor_serialized_));
310
311 // TODO(falk): Also print options and fix the message_type, enum_type,
312 // service and extension later in the generation.
313
314 printer_->Outdent();
315 printer_->Print(")\n");
316 printer_->Print("\n");
317 }
318
319 // Prints descriptors and module-level constants for all top-level
320 // enums defined in |file|.
PrintTopLevelEnums() const321 void Generator::PrintTopLevelEnums() const {
322 vector<pair<string, int> > top_level_enum_values;
323 for (int i = 0; i < file_->enum_type_count(); ++i) {
324 const EnumDescriptor& enum_descriptor = *file_->enum_type(i);
325 PrintEnum(enum_descriptor);
326 printer_->Print("\n");
327
328 for (int j = 0; j < enum_descriptor.value_count(); ++j) {
329 const EnumValueDescriptor& value_descriptor = *enum_descriptor.value(j);
330 top_level_enum_values.push_back(
331 make_pair(value_descriptor.name(), value_descriptor.number()));
332 }
333 }
334
335 for (int i = 0; i < top_level_enum_values.size(); ++i) {
336 printer_->Print("$name$ = $value$\n",
337 "name", top_level_enum_values[i].first,
338 "value", SimpleItoa(top_level_enum_values[i].second));
339 }
340 printer_->Print("\n");
341 }
342
343 // Prints all enums contained in all message types in |file|.
PrintAllNestedEnumsInFile() const344 void Generator::PrintAllNestedEnumsInFile() const {
345 for (int i = 0; i < file_->message_type_count(); ++i) {
346 PrintNestedEnums(*file_->message_type(i));
347 }
348 }
349
350 // Prints a Python statement assigning the appropriate module-level
351 // enum name to a Python EnumDescriptor object equivalent to
352 // enum_descriptor.
PrintEnum(const EnumDescriptor & enum_descriptor) const353 void Generator::PrintEnum(const EnumDescriptor& enum_descriptor) const {
354 map<string, string> m;
355 m["descriptor_name"] = ModuleLevelDescriptorName(enum_descriptor);
356 m["name"] = enum_descriptor.name();
357 m["full_name"] = enum_descriptor.full_name();
358 m["file"] = kDescriptorKey;
359 const char enum_descriptor_template[] =
360 "$descriptor_name$ = descriptor.EnumDescriptor(\n"
361 " name='$name$',\n"
362 " full_name='$full_name$',\n"
363 " filename=None,\n"
364 " file=$file$,\n"
365 " values=[\n";
366 string options_string;
367 enum_descriptor.options().SerializeToString(&options_string);
368 printer_->Print(m, enum_descriptor_template);
369 printer_->Indent();
370 printer_->Indent();
371 for (int i = 0; i < enum_descriptor.value_count(); ++i) {
372 PrintEnumValueDescriptor(*enum_descriptor.value(i));
373 printer_->Print(",\n");
374 }
375 printer_->Outdent();
376 printer_->Print("],\n");
377 printer_->Print("containing_type=None,\n");
378 printer_->Print("options=$options_value$,\n",
379 "options_value",
380 OptionsValue("EnumOptions", CEscape(options_string)));
381 EnumDescriptorProto edp;
382 PrintSerializedPbInterval(enum_descriptor, edp);
383 printer_->Outdent();
384 printer_->Print(")\n");
385 printer_->Print("\n");
386 }
387
388 // Recursively prints enums in nested types within descriptor, then
389 // prints enums contained at the top level in descriptor.
PrintNestedEnums(const Descriptor & descriptor) const390 void Generator::PrintNestedEnums(const Descriptor& descriptor) const {
391 for (int i = 0; i < descriptor.nested_type_count(); ++i) {
392 PrintNestedEnums(*descriptor.nested_type(i));
393 }
394
395 for (int i = 0; i < descriptor.enum_type_count(); ++i) {
396 PrintEnum(*descriptor.enum_type(i));
397 }
398 }
399
PrintTopLevelExtensions() const400 void Generator::PrintTopLevelExtensions() const {
401 const bool is_extension = true;
402 for (int i = 0; i < file_->extension_count(); ++i) {
403 const FieldDescriptor& extension_field = *file_->extension(i);
404 string constant_name = extension_field.name() + "_FIELD_NUMBER";
405 UpperString(&constant_name);
406 printer_->Print("$constant_name$ = $number$\n",
407 "constant_name", constant_name,
408 "number", SimpleItoa(extension_field.number()));
409 printer_->Print("$name$ = ", "name", extension_field.name());
410 PrintFieldDescriptor(extension_field, is_extension);
411 printer_->Print("\n");
412 }
413 printer_->Print("\n");
414 }
415
416 // Prints Python equivalents of all Descriptors in |file|.
PrintMessageDescriptors() const417 void Generator::PrintMessageDescriptors() const {
418 for (int i = 0; i < file_->message_type_count(); ++i) {
419 PrintDescriptor(*file_->message_type(i));
420 printer_->Print("\n");
421 }
422 }
423
PrintServices() const424 void Generator::PrintServices() const {
425 for (int i = 0; i < file_->service_count(); ++i) {
426 PrintServiceDescriptor(*file_->service(i));
427 PrintServiceClass(*file_->service(i));
428 PrintServiceStub(*file_->service(i));
429 printer_->Print("\n");
430 }
431 }
432
PrintServiceDescriptor(const ServiceDescriptor & descriptor) const433 void Generator::PrintServiceDescriptor(
434 const ServiceDescriptor& descriptor) const {
435 printer_->Print("\n");
436 string service_name = ModuleLevelServiceDescriptorName(descriptor);
437 string options_string;
438 descriptor.options().SerializeToString(&options_string);
439
440 printer_->Print(
441 "$service_name$ = descriptor.ServiceDescriptor(\n",
442 "service_name", service_name);
443 printer_->Indent();
444 map<string, string> m;
445 m["name"] = descriptor.name();
446 m["full_name"] = descriptor.full_name();
447 m["file"] = kDescriptorKey;
448 m["index"] = SimpleItoa(descriptor.index());
449 m["options_value"] = OptionsValue("ServiceOptions", options_string);
450 const char required_function_arguments[] =
451 "name='$name$',\n"
452 "full_name='$full_name$',\n"
453 "file=$file$,\n"
454 "index=$index$,\n"
455 "options=$options_value$,\n";
456 printer_->Print(m, required_function_arguments);
457
458 ServiceDescriptorProto sdp;
459 PrintSerializedPbInterval(descriptor, sdp);
460
461 printer_->Print("methods=[\n");
462 for (int i = 0; i < descriptor.method_count(); ++i) {
463 const MethodDescriptor* method = descriptor.method(i);
464 string options_string;
465 method->options().SerializeToString(&options_string);
466
467 m.clear();
468 m["name"] = method->name();
469 m["full_name"] = method->full_name();
470 m["index"] = SimpleItoa(method->index());
471 m["serialized_options"] = CEscape(options_string);
472 m["input_type"] = ModuleLevelDescriptorName(*(method->input_type()));
473 m["output_type"] = ModuleLevelDescriptorName(*(method->output_type()));
474 m["options_value"] = OptionsValue("MethodOptions", options_string);
475 printer_->Print("descriptor.MethodDescriptor(\n");
476 printer_->Indent();
477 printer_->Print(
478 m,
479 "name='$name$',\n"
480 "full_name='$full_name$',\n"
481 "index=$index$,\n"
482 "containing_service=None,\n"
483 "input_type=$input_type$,\n"
484 "output_type=$output_type$,\n"
485 "options=$options_value$,\n");
486 printer_->Outdent();
487 printer_->Print("),\n");
488 }
489
490 printer_->Outdent();
491 printer_->Print("])\n\n");
492 }
493
PrintServiceClass(const ServiceDescriptor & descriptor) const494 void Generator::PrintServiceClass(const ServiceDescriptor& descriptor) const {
495 // Print the service.
496 printer_->Print("class $class_name$(service.Service):\n",
497 "class_name", descriptor.name());
498 printer_->Indent();
499 printer_->Print(
500 "__metaclass__ = service_reflection.GeneratedServiceType\n"
501 "$descriptor_key$ = $descriptor_name$\n",
502 "descriptor_key", kDescriptorKey,
503 "descriptor_name", ModuleLevelServiceDescriptorName(descriptor));
504 printer_->Outdent();
505 }
506
PrintServiceStub(const ServiceDescriptor & descriptor) const507 void Generator::PrintServiceStub(const ServiceDescriptor& descriptor) const {
508 // Print the service stub.
509 printer_->Print("class $class_name$_Stub($class_name$):\n",
510 "class_name", descriptor.name());
511 printer_->Indent();
512 printer_->Print(
513 "__metaclass__ = service_reflection.GeneratedServiceStubType\n"
514 "$descriptor_key$ = $descriptor_name$\n",
515 "descriptor_key", kDescriptorKey,
516 "descriptor_name", ModuleLevelServiceDescriptorName(descriptor));
517 printer_->Outdent();
518 }
519
520 // Prints statement assigning ModuleLevelDescriptorName(message_descriptor)
521 // to a Python Descriptor object for message_descriptor.
522 //
523 // Mutually recursive with PrintNestedDescriptors().
PrintDescriptor(const Descriptor & message_descriptor) const524 void Generator::PrintDescriptor(const Descriptor& message_descriptor) const {
525 PrintNestedDescriptors(message_descriptor);
526
527 printer_->Print("\n");
528 printer_->Print("$descriptor_name$ = descriptor.Descriptor(\n",
529 "descriptor_name",
530 ModuleLevelDescriptorName(message_descriptor));
531 printer_->Indent();
532 map<string, string> m;
533 m["name"] = message_descriptor.name();
534 m["full_name"] = message_descriptor.full_name();
535 m["file"] = kDescriptorKey;
536 const char required_function_arguments[] =
537 "name='$name$',\n"
538 "full_name='$full_name$',\n"
539 "filename=None,\n"
540 "file=$file$,\n"
541 "containing_type=None,\n";
542 printer_->Print(m, required_function_arguments);
543 PrintFieldsInDescriptor(message_descriptor);
544 PrintExtensionsInDescriptor(message_descriptor);
545
546 // Nested types
547 printer_->Print("nested_types=[");
548 for (int i = 0; i < message_descriptor.nested_type_count(); ++i) {
549 const string nested_name = ModuleLevelDescriptorName(
550 *message_descriptor.nested_type(i));
551 printer_->Print("$name$, ", "name", nested_name);
552 }
553 printer_->Print("],\n");
554
555 // Enum types
556 printer_->Print("enum_types=[\n");
557 printer_->Indent();
558 for (int i = 0; i < message_descriptor.enum_type_count(); ++i) {
559 const string descriptor_name = ModuleLevelDescriptorName(
560 *message_descriptor.enum_type(i));
561 printer_->Print(descriptor_name.c_str());
562 printer_->Print(",\n");
563 }
564 printer_->Outdent();
565 printer_->Print("],\n");
566 string options_string;
567 message_descriptor.options().SerializeToString(&options_string);
568 printer_->Print(
569 "options=$options_value$,\n"
570 "is_extendable=$extendable$",
571 "options_value", OptionsValue("MessageOptions", options_string),
572 "extendable", message_descriptor.extension_range_count() > 0 ?
573 "True" : "False");
574 printer_->Print(",\n");
575
576 // Extension ranges
577 printer_->Print("extension_ranges=[");
578 for (int i = 0; i < message_descriptor.extension_range_count(); ++i) {
579 const Descriptor::ExtensionRange* range =
580 message_descriptor.extension_range(i);
581 printer_->Print("($start$, $end$), ",
582 "start", SimpleItoa(range->start),
583 "end", SimpleItoa(range->end));
584 }
585 printer_->Print("],\n");
586
587 // Serialization of proto
588 DescriptorProto edp;
589 PrintSerializedPbInterval(message_descriptor, edp);
590
591 printer_->Outdent();
592 printer_->Print(")\n");
593 }
594
595 // Prints Python Descriptor objects for all nested types contained in
596 // message_descriptor.
597 //
598 // Mutually recursive with PrintDescriptor().
PrintNestedDescriptors(const Descriptor & containing_descriptor) const599 void Generator::PrintNestedDescriptors(
600 const Descriptor& containing_descriptor) const {
601 for (int i = 0; i < containing_descriptor.nested_type_count(); ++i) {
602 PrintDescriptor(*containing_descriptor.nested_type(i));
603 }
604 }
605
606 // Prints all messages in |file|.
PrintMessages() const607 void Generator::PrintMessages() const {
608 for (int i = 0; i < file_->message_type_count(); ++i) {
609 PrintMessage(*file_->message_type(i));
610 printer_->Print("\n");
611 }
612 }
613
614 // Prints a Python class for the given message descriptor. We defer to the
615 // metaclass to do almost all of the work of actually creating a useful class.
616 // The purpose of this function and its many helper functions above is merely
617 // to output a Python version of the descriptors, which the metaclass in
618 // reflection.py will use to construct the meat of the class itself.
619 //
620 // Mutually recursive with PrintNestedMessages().
PrintMessage(const Descriptor & message_descriptor) const621 void Generator::PrintMessage(
622 const Descriptor& message_descriptor) const {
623 printer_->Print("class $name$(message.Message):\n", "name",
624 message_descriptor.name());
625 printer_->Indent();
626 printer_->Print("__metaclass__ = reflection.GeneratedProtocolMessageType\n");
627 PrintNestedMessages(message_descriptor);
628 map<string, string> m;
629 m["descriptor_key"] = kDescriptorKey;
630 m["descriptor_name"] = ModuleLevelDescriptorName(message_descriptor);
631 printer_->Print(m, "$descriptor_key$ = $descriptor_name$\n");
632
633 printer_->Print(
634 "\n"
635 "# @@protoc_insertion_point(class_scope:$full_name$)\n",
636 "full_name", message_descriptor.full_name());
637
638 printer_->Outdent();
639 }
640
641 // Prints all nested messages within |containing_descriptor|.
642 // Mutually recursive with PrintMessage().
PrintNestedMessages(const Descriptor & containing_descriptor) const643 void Generator::PrintNestedMessages(
644 const Descriptor& containing_descriptor) const {
645 for (int i = 0; i < containing_descriptor.nested_type_count(); ++i) {
646 printer_->Print("\n");
647 PrintMessage(*containing_descriptor.nested_type(i));
648 }
649 }
650
651 // Recursively fixes foreign fields in all nested types in |descriptor|, then
652 // sets the message_type and enum_type of all message and enum fields to point
653 // to their respective descriptors.
654 // Args:
655 // descriptor: descriptor to print fields for.
656 // containing_descriptor: if descriptor is a nested type, this is its
657 // containing type, or NULL if this is a root/top-level type.
FixForeignFieldsInDescriptor(const Descriptor & descriptor,const Descriptor * containing_descriptor) const658 void Generator::FixForeignFieldsInDescriptor(
659 const Descriptor& descriptor,
660 const Descriptor* containing_descriptor) const {
661 for (int i = 0; i < descriptor.nested_type_count(); ++i) {
662 FixForeignFieldsInDescriptor(*descriptor.nested_type(i), &descriptor);
663 }
664
665 for (int i = 0; i < descriptor.field_count(); ++i) {
666 const FieldDescriptor& field_descriptor = *descriptor.field(i);
667 FixForeignFieldsInField(&descriptor, field_descriptor, "fields_by_name");
668 }
669
670 FixContainingTypeInDescriptor(descriptor, containing_descriptor);
671 for (int i = 0; i < descriptor.enum_type_count(); ++i) {
672 const EnumDescriptor& enum_descriptor = *descriptor.enum_type(i);
673 FixContainingTypeInDescriptor(enum_descriptor, &descriptor);
674 }
675 }
676
677 // Sets any necessary message_type and enum_type attributes
678 // for the Python version of |field|.
679 //
680 // containing_type may be NULL, in which case this is a module-level field.
681 //
682 // python_dict_name is the name of the Python dict where we should
683 // look the field up in the containing type. (e.g., fields_by_name
684 // or extensions_by_name). We ignore python_dict_name if containing_type
685 // is NULL.
FixForeignFieldsInField(const Descriptor * containing_type,const FieldDescriptor & field,const string & python_dict_name) const686 void Generator::FixForeignFieldsInField(const Descriptor* containing_type,
687 const FieldDescriptor& field,
688 const string& python_dict_name) const {
689 const string field_referencing_expression = FieldReferencingExpression(
690 containing_type, field, python_dict_name);
691 map<string, string> m;
692 m["field_ref"] = field_referencing_expression;
693 const Descriptor* foreign_message_type = field.message_type();
694 if (foreign_message_type) {
695 m["foreign_type"] = ModuleLevelDescriptorName(*foreign_message_type);
696 printer_->Print(m, "$field_ref$.message_type = $foreign_type$\n");
697 }
698 const EnumDescriptor* enum_type = field.enum_type();
699 if (enum_type) {
700 m["enum_type"] = ModuleLevelDescriptorName(*enum_type);
701 printer_->Print(m, "$field_ref$.enum_type = $enum_type$\n");
702 }
703 }
704
705 // Returns the module-level expression for the given FieldDescriptor.
706 // Only works for fields in the .proto file this Generator is generating for.
707 //
708 // containing_type may be NULL, in which case this is a module-level field.
709 //
710 // python_dict_name is the name of the Python dict where we should
711 // look the field up in the containing type. (e.g., fields_by_name
712 // or extensions_by_name). We ignore python_dict_name if containing_type
713 // is NULL.
FieldReferencingExpression(const Descriptor * containing_type,const FieldDescriptor & field,const string & python_dict_name) const714 string Generator::FieldReferencingExpression(
715 const Descriptor* containing_type,
716 const FieldDescriptor& field,
717 const string& python_dict_name) const {
718 // We should only ever be looking up fields in the current file.
719 // The only things we refer to from other files are message descriptors.
720 GOOGLE_CHECK_EQ(field.file(), file_) << field.file()->name() << " vs. "
721 << file_->name();
722 if (!containing_type) {
723 return field.name();
724 }
725 return strings::Substitute(
726 "$0.$1['$2']",
727 ModuleLevelDescriptorName(*containing_type),
728 python_dict_name, field.name());
729 }
730
731 // Prints containing_type for nested descriptors or enum descriptors.
732 template <typename DescriptorT>
FixContainingTypeInDescriptor(const DescriptorT & descriptor,const Descriptor * containing_descriptor) const733 void Generator::FixContainingTypeInDescriptor(
734 const DescriptorT& descriptor,
735 const Descriptor* containing_descriptor) const {
736 if (containing_descriptor != NULL) {
737 const string nested_name = ModuleLevelDescriptorName(descriptor);
738 const string parent_name = ModuleLevelDescriptorName(
739 *containing_descriptor);
740 printer_->Print(
741 "$nested_name$.containing_type = $parent_name$;\n",
742 "nested_name", nested_name,
743 "parent_name", parent_name);
744 }
745 }
746
747 // Prints statements setting the message_type and enum_type fields in the
748 // Python descriptor objects we've already output in ths file. We must
749 // do this in a separate step due to circular references (otherwise, we'd
750 // just set everything in the initial assignment statements).
FixForeignFieldsInDescriptors() const751 void Generator::FixForeignFieldsInDescriptors() const {
752 for (int i = 0; i < file_->message_type_count(); ++i) {
753 FixForeignFieldsInDescriptor(*file_->message_type(i), NULL);
754 }
755 printer_->Print("\n");
756 }
757
758 // We need to not only set any necessary message_type fields, but
759 // also need to call RegisterExtension() on each message we're
760 // extending.
FixForeignFieldsInExtensions() const761 void Generator::FixForeignFieldsInExtensions() const {
762 // Top-level extensions.
763 for (int i = 0; i < file_->extension_count(); ++i) {
764 FixForeignFieldsInExtension(*file_->extension(i));
765 }
766 // Nested extensions.
767 for (int i = 0; i < file_->message_type_count(); ++i) {
768 FixForeignFieldsInNestedExtensions(*file_->message_type(i));
769 }
770 }
771
FixForeignFieldsInExtension(const FieldDescriptor & extension_field) const772 void Generator::FixForeignFieldsInExtension(
773 const FieldDescriptor& extension_field) const {
774 GOOGLE_CHECK(extension_field.is_extension());
775 // extension_scope() will be NULL for top-level extensions, which is
776 // exactly what FixForeignFieldsInField() wants.
777 FixForeignFieldsInField(extension_field.extension_scope(), extension_field,
778 "extensions_by_name");
779
780 map<string, string> m;
781 // Confusingly, for FieldDescriptors that happen to be extensions,
782 // containing_type() means "extended type."
783 // On the other hand, extension_scope() will give us what we normally
784 // mean by containing_type().
785 m["extended_message_class"] = ModuleLevelMessageName(
786 *extension_field.containing_type());
787 m["field"] = FieldReferencingExpression(extension_field.extension_scope(),
788 extension_field,
789 "extensions_by_name");
790 printer_->Print(m, "$extended_message_class$.RegisterExtension($field$)\n");
791 }
792
FixForeignFieldsInNestedExtensions(const Descriptor & descriptor) const793 void Generator::FixForeignFieldsInNestedExtensions(
794 const Descriptor& descriptor) const {
795 // Recursively fix up extensions in all nested types.
796 for (int i = 0; i < descriptor.nested_type_count(); ++i) {
797 FixForeignFieldsInNestedExtensions(*descriptor.nested_type(i));
798 }
799 // Fix up extensions directly contained within this type.
800 for (int i = 0; i < descriptor.extension_count(); ++i) {
801 FixForeignFieldsInExtension(*descriptor.extension(i));
802 }
803 }
804
805 // Returns a Python expression that instantiates a Python EnumValueDescriptor
806 // object for the given C++ descriptor.
PrintEnumValueDescriptor(const EnumValueDescriptor & descriptor) const807 void Generator::PrintEnumValueDescriptor(
808 const EnumValueDescriptor& descriptor) const {
809 // TODO(robinson): Fix up EnumValueDescriptor "type" fields.
810 // More circular references. ::sigh::
811 string options_string;
812 descriptor.options().SerializeToString(&options_string);
813 map<string, string> m;
814 m["name"] = descriptor.name();
815 m["index"] = SimpleItoa(descriptor.index());
816 m["number"] = SimpleItoa(descriptor.number());
817 m["options"] = OptionsValue("EnumValueOptions", options_string);
818 printer_->Print(
819 m,
820 "descriptor.EnumValueDescriptor(\n"
821 " name='$name$', index=$index$, number=$number$,\n"
822 " options=$options$,\n"
823 " type=None)");
824 }
825
OptionsValue(const string & class_name,const string & serialized_options) const826 string Generator::OptionsValue(
827 const string& class_name, const string& serialized_options) const {
828 if (serialized_options.length() == 0 || GeneratingDescriptorProto()) {
829 return "None";
830 } else {
831 string full_class_name = "descriptor_pb2." + class_name;
832 return "descriptor._ParseOptions(" + full_class_name + "(), '"
833 + CEscape(serialized_options)+ "')";
834 }
835 }
836
837 // Prints an expression for a Python FieldDescriptor for |field|.
PrintFieldDescriptor(const FieldDescriptor & field,bool is_extension) const838 void Generator::PrintFieldDescriptor(
839 const FieldDescriptor& field, bool is_extension) const {
840 string options_string;
841 field.options().SerializeToString(&options_string);
842 map<string, string> m;
843 m["name"] = field.name();
844 m["full_name"] = field.full_name();
845 m["index"] = SimpleItoa(field.index());
846 m["number"] = SimpleItoa(field.number());
847 m["type"] = SimpleItoa(field.type());
848 m["cpp_type"] = SimpleItoa(field.cpp_type());
849 m["label"] = SimpleItoa(field.label());
850 m["has_default_value"] = field.has_default_value() ? "True" : "False";
851 m["default_value"] = StringifyDefaultValue(field);
852 m["is_extension"] = is_extension ? "True" : "False";
853 m["options"] = OptionsValue("FieldOptions", options_string);
854 // We always set message_type and enum_type to None at this point, and then
855 // these fields in correctly after all referenced descriptors have been
856 // defined and/or imported (see FixForeignFieldsInDescriptors()).
857 const char field_descriptor_decl[] =
858 "descriptor.FieldDescriptor(\n"
859 " name='$name$', full_name='$full_name$', index=$index$,\n"
860 " number=$number$, type=$type$, cpp_type=$cpp_type$, label=$label$,\n"
861 " has_default_value=$has_default_value$, default_value=$default_value$,\n"
862 " message_type=None, enum_type=None, containing_type=None,\n"
863 " is_extension=$is_extension$, extension_scope=None,\n"
864 " options=$options$)";
865 printer_->Print(m, field_descriptor_decl);
866 }
867
868 // Helper for Print{Fields,Extensions}InDescriptor().
PrintFieldDescriptorsInDescriptor(const Descriptor & message_descriptor,bool is_extension,const string & list_variable_name,int (Descriptor::* CountFn)()const,const FieldDescriptor * (Descriptor::* GetterFn)(int)const) const869 void Generator::PrintFieldDescriptorsInDescriptor(
870 const Descriptor& message_descriptor,
871 bool is_extension,
872 const string& list_variable_name,
873 int (Descriptor::*CountFn)() const,
874 const FieldDescriptor* (Descriptor::*GetterFn)(int) const) const {
875 printer_->Print("$list$=[\n", "list", list_variable_name);
876 printer_->Indent();
877 for (int i = 0; i < (message_descriptor.*CountFn)(); ++i) {
878 PrintFieldDescriptor(*(message_descriptor.*GetterFn)(i),
879 is_extension);
880 printer_->Print(",\n");
881 }
882 printer_->Outdent();
883 printer_->Print("],\n");
884 }
885
886 // Prints a statement assigning "fields" to a list of Python FieldDescriptors,
887 // one for each field present in message_descriptor.
PrintFieldsInDescriptor(const Descriptor & message_descriptor) const888 void Generator::PrintFieldsInDescriptor(
889 const Descriptor& message_descriptor) const {
890 const bool is_extension = false;
891 PrintFieldDescriptorsInDescriptor(
892 message_descriptor, is_extension, "fields",
893 &Descriptor::field_count, &Descriptor::field);
894 }
895
896 // Prints a statement assigning "extensions" to a list of Python
897 // FieldDescriptors, one for each extension present in message_descriptor.
PrintExtensionsInDescriptor(const Descriptor & message_descriptor) const898 void Generator::PrintExtensionsInDescriptor(
899 const Descriptor& message_descriptor) const {
900 const bool is_extension = true;
901 PrintFieldDescriptorsInDescriptor(
902 message_descriptor, is_extension, "extensions",
903 &Descriptor::extension_count, &Descriptor::extension);
904 }
905
GeneratingDescriptorProto() const906 bool Generator::GeneratingDescriptorProto() const {
907 return file_->name() == "google/protobuf/descriptor.proto";
908 }
909
910 // Returns the unique Python module-level identifier given to a descriptor.
911 // This name is module-qualified iff the given descriptor describes an
912 // entity that doesn't come from the current file.
913 template <typename DescriptorT>
ModuleLevelDescriptorName(const DescriptorT & descriptor) const914 string Generator::ModuleLevelDescriptorName(
915 const DescriptorT& descriptor) const {
916 // FIXME(robinson):
917 // We currently don't worry about collisions with underscores in the type
918 // names, so these would collide in nasty ways if found in the same file:
919 // OuterProto.ProtoA.ProtoB
920 // OuterProto_ProtoA.ProtoB # Underscore instead of period.
921 // As would these:
922 // OuterProto.ProtoA_.ProtoB
923 // OuterProto.ProtoA._ProtoB # Leading vs. trailing underscore.
924 // (Contrived, but certainly possible).
925 //
926 // The C++ implementation doesn't guard against this either. Leaving
927 // it for now...
928 string name = NamePrefixedWithNestedTypes(descriptor, "_");
929 UpperString(&name);
930 // Module-private for now. Easy to make public later; almost impossible
931 // to make private later.
932 name = "_" + name;
933 // We now have the name relative to its own module. Also qualify with
934 // the module name iff this descriptor is from a different .proto file.
935 if (descriptor.file() != file_) {
936 name = ModuleName(descriptor.file()->name()) + "." + name;
937 }
938 return name;
939 }
940
941 // Returns the name of the message class itself, not the descriptor.
942 // Like ModuleLevelDescriptorName(), module-qualifies the name iff
943 // the given descriptor describes an entity that doesn't come from
944 // the current file.
ModuleLevelMessageName(const Descriptor & descriptor) const945 string Generator::ModuleLevelMessageName(const Descriptor& descriptor) const {
946 string name = NamePrefixedWithNestedTypes(descriptor, ".");
947 if (descriptor.file() != file_) {
948 name = ModuleName(descriptor.file()->name()) + "." + name;
949 }
950 return name;
951 }
952
953 // Returns the unique Python module-level identifier given to a service
954 // descriptor.
ModuleLevelServiceDescriptorName(const ServiceDescriptor & descriptor) const955 string Generator::ModuleLevelServiceDescriptorName(
956 const ServiceDescriptor& descriptor) const {
957 string name = descriptor.name();
958 UpperString(&name);
959 name = "_" + name;
960 if (descriptor.file() != file_) {
961 name = ModuleName(descriptor.file()->name()) + "." + name;
962 }
963 return name;
964 }
965
966 // Prints standard constructor arguments serialized_start and serialized_end.
967 // Args:
968 // descriptor: The cpp descriptor to have a serialized reference.
969 // proto: A proto
970 // Example printer output:
971 // serialized_start=41,
972 // serialized_end=43,
973 //
974 template <typename DescriptorT, typename DescriptorProtoT>
PrintSerializedPbInterval(const DescriptorT & descriptor,DescriptorProtoT & proto) const975 void Generator::PrintSerializedPbInterval(
976 const DescriptorT& descriptor, DescriptorProtoT& proto) const {
977 descriptor.CopyTo(&proto);
978 string sp;
979 proto.SerializeToString(&sp);
980 int offset = file_descriptor_serialized_.find(sp);
981 GOOGLE_CHECK_GE(offset, 0);
982
983 printer_->Print("serialized_start=$serialized_start$,\n"
984 "serialized_end=$serialized_end$,\n",
985 "serialized_start", SimpleItoa(offset),
986 "serialized_end", SimpleItoa(offset + sp.size()));
987 }
988
989 } // namespace python
990 } // namespace compiler
991 } // namespace protobuf
992 } // namespace google
993