protobuf/tools/upbc.c

/*
 * upb - a minimalist implementation of protocol buffers.
 *
 * upbc is the upb compiler.
 *
 * Copyright (c) 2009 Joshua Haberman.  See LICENSE for details.
 */

#include <ctype.h>
#include <inttypes.h>
#include "descriptor.h"
#include "upb_context.h"
#include "upb_enum.h"

/* These are in-place string transformations that do not change the length of
 * the string (and thus never need to re-allocate). */
static void to_cident(struct upb_string str)
{
  for(uint32_t i = 0; i < str.byte_len; i++)
    if(str.ptr[i] == '.' || str.ptr[i] == '/')
      str.ptr[i] = '_';
}

static void to_preproc(struct upb_string str)
{
  to_cident(str);
  for(uint32_t i = 0; i < str.byte_len; i++)
    str.ptr[i] = toupper(str.ptr[i]);
}

static int memrchr(char *data, char c, size_t len)
{
  int off = len-1;
  while(off > 0 && data[off] != c) --off;
  return off;
}

void *strtable_to_array(struct upb_strtable *t, int *size)
{
  *size = t->t.count;
  void **array = malloc(*size * sizeof(void*));
  struct upb_symtab_entry *e;
  int i = 0;
  for(e = upb_strtable_begin(t); e && i < *size; e = upb_strtable_next(t, &e->e))
    array[i++] = e;
  assert(i == *size && e == NULL);
  return array;
}

/* The .h file defines structs for the types defined in the .proto file.  It
 * also defines constants for the enum values.
 *
 * Assumes that d has been validated. */
static void write_h(struct upb_symtab_entry *entries[], int num_entries,
                    char *outfile_name, FILE *stream)
{
  /* Header file prologue. */
  struct upb_string include_guard_name = upb_strdupc(outfile_name);
  to_preproc(include_guard_name);
  fputs("/* This file was generated by upbc (the upb compiler).  "
        "Do not edit. */\n\n", stream),
  fprintf(stream, "#ifndef " UPB_STRFMT "\n", UPB_STRARG(include_guard_name));
  fprintf(stream, "#define " UPB_STRFMT "\n\n", UPB_STRARG(include_guard_name));
  fputs("#include <upb_msg.h>\n\n", stream);
  fputs("#ifdef __cplusplus\n", stream);
  fputs("extern \"C\" {\n", stream);
  fputs("#endif\n\n", stream);

  /* Enums. */
  fprintf(stream, "/* Enums. */\n\n");
  for(int i = 0; i < num_entries; i++) {  /* Foreach enum */
    if(entries[i]->type != UPB_SYM_ENUM) continue;
    struct upb_symtab_entry *entry = entries[i];
    struct upb_enum *e = entry->ref._enum;
    google_protobuf_EnumDescriptorProto *ed = e->descriptor;
    /* We use entry->e.key (the fully qualified name) instead of ed->name. */
    struct upb_string enum_name = upb_strdup(entry->e.key);
    to_cident(enum_name);

    struct upb_string enum_val_prefix = upb_strdup(entry->e.key);
    enum_val_prefix.byte_len = memrchr(enum_val_prefix.ptr,
                                       UPB_SYMBOL_SEPARATOR,
                                       enum_val_prefix.byte_len);
    enum_val_prefix.byte_len++;
    to_preproc(enum_val_prefix);

    fprintf(stream, "typedef enum " UPB_STRFMT " {\n", UPB_STRARG(enum_name));
    if(ed->set_flags.has.value) {
      for(uint32_t j = 0; j < ed->value->len; j++) {  /* Foreach enum value. */
        google_protobuf_EnumValueDescriptorProto *v = ed->value->elements[j];
        struct upb_string value_name = upb_strdup(*v->name);
        to_preproc(value_name);
        /* "  GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_UINT32 = 13," */
        fprintf(stream, "  " UPB_STRFMT UPB_STRFMT " = %" PRIu32,
                UPB_STRARG(enum_val_prefix), UPB_STRARG(value_name), v->number);
        if(j != ed->value->len-1) fputc(',', stream);
        fputc('\n', stream);
        upb_strfree(value_name);
      }
    }
    fprintf(stream, "} " UPB_STRFMT ";\n\n", UPB_STRARG(enum_name));
    upb_strfree(enum_name);
    upb_strfree(enum_val_prefix);
  }

  /* Forward declarations. */
  fputs("/* Forward declarations of all message types.\n", stream);
  fputs(" * So they can refer to each other in ", stream);
  fputs("possibly-recursive ways. */\n\n", stream);

  for(int i = 0; i < num_entries; i++) {  /* Foreach message */
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    struct upb_symtab_entry *entry = entries[i];
    /* We use entry->e.key (the fully qualified name). */
    struct upb_string msg_name = upb_strdup(entry->e.key);
    to_cident(msg_name);
    fprintf(stream, "struct " UPB_STRFMT ";\n", UPB_STRARG(msg_name));
    fprintf(stream, "typedef struct " UPB_STRFMT "\n    " UPB_STRFMT ";\n\n",
            UPB_STRARG(msg_name), UPB_STRARG(msg_name));
    upb_strfree(msg_name);
  }

  /* Message Declarations. */
  fputs("/* The message definitions themselves. */\n\n", stream);
  for(int i = 0; i < num_entries; i++) {  /* Foreach message */
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    struct upb_symtab_entry *entry = entries[i];
    struct upb_msg *m = entry->ref.msg;
    /* We use entry->e.key (the fully qualified name). */
    struct upb_string msg_name = upb_strdup(entry->e.key);
    to_cident(msg_name);
    fprintf(stream, "struct " UPB_STRFMT " {\n", UPB_STRARG(msg_name));
    fputs("  union {\n", stream);
    fprintf(stream, "    uint8_t bytes[%" PRIu32 "];\n", m->set_flags_bytes);
    fputs("    struct {\n", stream);
    for(uint32_t j = 0; j < m->num_fields; j++) {
      static char* labels[] = {"", "optional", "required", "repeated"};
      struct google_protobuf_FieldDescriptorProto *fd = m->field_descriptors[j];
      fprintf(stream, "      bool " UPB_STRFMT ":1;  /* = %" PRIu32 ", %s. */\n",
              UPB_STRARG(*fd->name), fd->number, labels[fd->label]);
    }
    fputs("    } has;\n", stream);
    fputs("  } set_flags;\n", stream);
    for(uint32_t j = 0; j < m->num_fields; j++) {
      struct upb_msg_field *f = &m->fields[j];
      struct google_protobuf_FieldDescriptorProto *fd = m->field_descriptors[j];
      if(f->type == GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_GROUP ||
         f->type == GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_MESSAGE) {
        /* Submessages get special treatment, since we have to use the message
         * name directly. */
        struct upb_string type_name_ref = *fd->type_name;
        if(type_name_ref.ptr[0] == UPB_SYMBOL_SEPARATOR) {
          /* Omit leading '.'. */
          type_name_ref.ptr++;
          type_name_ref.byte_len--;
        }
        struct upb_string type_name = upb_strdup(type_name_ref);
        to_cident(type_name);
        if(f->label == GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REPEATED) {
          fprintf(stream, "  UPB_MSG_ARRAY(" UPB_STRFMT ")* " UPB_STRFMT ";\n",
                  UPB_STRARG(type_name), UPB_STRARG(*fd->name));
        } else {
          fprintf(stream, "  " UPB_STRFMT "* " UPB_STRFMT ";\n",
                  UPB_STRARG(type_name), UPB_STRARG(*fd->name));
        }
        upb_strfree(type_name);
      } else if(f->label == GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REPEATED) {
        static char* c_types[] = {
          "", "struct upb_double_array*", "struct upb_float_array*",
          "struct upb_int64_array*", "struct upb_uint64_array*",
          "struct upb_int32_array*", "struct upb_uint64_array*",
          "struct upb_uint32_array*", "struct upb_bool_array*",
          "struct upb_string_array*", "", "",
          "struct upb_string_array*", "struct upb_uint32_array*",
          "struct upb_uint32_array*", "struct upb_int32_array*",
          "struct upb_int64_array*", "struct upb_int32_array*",
          "struct upb_int64_array*"
        };
        fprintf(stream, "  %s " UPB_STRFMT ";\n",
                c_types[fd->type], UPB_STRARG(*fd->name));
      } else {
        static char* c_types[] = {
          "", "double", "float", "int64_t", "uint64_t", "int32_t", "uint64_t",
          "uint32_t", "bool", "struct upb_string*", "", "",
          "struct upb_string*", "uint32_t", "uint32_t", "int32_t", "int64_t",
          "int32_t", "int64_t"
        };
        fprintf(stream, "  %s " UPB_STRFMT ";\n",
                c_types[fd->type], UPB_STRARG(*fd->name));
      }
    }
    fputs("};\n", stream);
    fprintf(stream, "UPB_DEFINE_MSG_ARRAY(" UPB_STRFMT ")\n\n",
            UPB_STRARG(msg_name));
    upb_strfree(msg_name);
  }

  /* Epilogue. */
  fputs("#ifdef __cplusplus\n", stream);
  fputs("}  /* extern \"C\" */\n", stream);
  fputs("#endif\n\n", stream);
  fprintf(stream, "#endif  /* " UPB_STRFMT " */\n", UPB_STRARG(include_guard_name));
  upb_strfree(include_guard_name);
}

struct strtable_entry {
  struct upb_strtable_entry e;
  int offset;
  int num;
};

int compare_entries(const void *_e1, const void *_e2)
{
  struct strtable_entry *const*e1 = _e1, *const*e2 = _e2;
  return upb_strcmp((*e1)->e.key, (*e2)->e.key);
}

/* The .c file defines the descriptor as data (in C structs).
 *
 * Assumes that d has been validated. */
static void write_c(struct upb_symtab_entry *entries[], int num_entries,
                    char *hfile_name, char *outfile_name, FILE *stream)
{
  (void)outfile_name;
  fputs("/* This file was generated by upbc (the upb compiler).  "
        "Do not edit. */\n\n", stream),
  fprintf(stream, "#include \"%s\"\n\n", hfile_name);

  /* Gather all strings into a giant string.  Use a hash */
  struct upb_strtable t;

#define ADDSTR(msg, field) \
  if(msg->set_flags.has.field) { \
    struct strtable_entry e = {.e = {.key = *msg->field}}; \
    if(upb_strtable_lookup(&t, &e.e.key) == NULL) upb_strtable_insert(&t, &e.e); \
  }

  upb_strtable_init(&t, 16, sizeof(struct strtable_entry));
  for(int i = 0; i < num_entries; i++) {
    struct strtable_entry e = {.e = {.key = entries[i]->e.key}};
    if(upb_strtable_lookup(&t, &e.e.key) == NULL) upb_strtable_insert(&t, &e.e);
    switch(entries[i]->type) {
      case UPB_SYM_MESSAGE: {
        struct upb_msg *m = entries[i]->ref.msg;
        ADDSTR(m->descriptor, name);
        for(uint32_t i = 0; i < m->num_fields; i++) {
          google_protobuf_FieldDescriptorProto *fd = m->field_descriptors[i];
          ADDSTR(fd, name);
          ADDSTR(fd, type_name);
          //ADDSTR(fd, extendee);
          //ADDSTR(fd, default_value);
          /* Neglect fd->options, doubtful that they're needed. */
        }
      }

      case UPB_SYM_ENUM: {
        google_protobuf_EnumDescriptorProto *ed = entries[i]->ref._enum->descriptor;
        ADDSTR(ed, name);
        if(ed->set_flags.has.value) {
          for(uint32_t i = 0; i < ed->value->len; i++) {
            ADDSTR(ed->value->elements[i], name);
            /* Neglect ed->value[i]->options, doubtful that they're needed. */
          }
        }
      }

      case UPB_SYM_SERVICE:
      case UPB_SYM_EXTENSION: break;  /* TODO */
    }
  }

  int size;
  struct strtable_entry **str_entries = strtable_to_array(&t, &size);
  /* Sort for nice size and reproduceability. */
  qsort(str_entries, size, sizeof(void*), compare_entries);

  /* Emit strings. */
  fputs("static char strdata[] =\n  \"", stream);
  int col = 2;
  int offset = 0;
  for(int i = 0; i < size; i++) {
    struct upb_string *s = &str_entries[i]->e.key;
    str_entries[i]->offset = offset;
    str_entries[i]->num = i;
    for(uint32_t j = 0; j < s->byte_len; j++) {
      if(++col == 80) {
        fputs("\"\n  \"", stream);
        col = 3;
      }
      fputc(s->ptr[j], stream);
    }
    offset += s->byte_len;
  }
  fputs("\";\n\n", stream);

  fputs("static struct upb_string strings[] = {\n", stream);
  for(int i = 0; i < size; i++) {
    struct strtable_entry *e = str_entries[i];
    fprintf(stream, "  {.ptr = &strdata[%d], .byte_len=%d},\n", e->offset, e->e.key.byte_len);
  }
  fputs("};\n\n", stream);

  /* Emit fields. */
  fputs("static google_protobuf_FieldDescriptorProto fields[] = {\n", stream);
  int total = 0;
  for(int i = 0; i < num_entries; i++) {
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    struct upb_msg *m = entries[i]->ref.msg;
    for(uint32_t j = 0; j < m->num_fields; j++) {
      struct google_protobuf_FieldDescriptorProto *fd = m->field_descriptors[j];
      struct strtable_entry *e = upb_strtable_lookup(&t, fd->name);
      fprintf(stream, "  {.set_flags = {.bytes={0x%02hhx}}, .name=&strings[%3d], .number=%d, .label=%d, .type=%2d, .type_name=NULL},\n",
              fd->set_flags.bytes[0], e->num, fd->number, fd->label, fd->type);
      total ++;
    }
  }
  fputs("};\n\n", stream);

  fputs("static google_protobuf_FieldDescriptorProto *field_pointers[] = {\n", stream);
  for(int i = 0; i < total; i++) {
    fprintf(stream, "  &fields[%d],\n", i);
  }
  fputs("};\n\n", stream);

  offset = 0;
  fputs("static UPB_MSG_ARRAY(google_protobuf_FieldDescriptorProto) field_arrays[] = {\n", stream);
  for(int i = 0; i < num_entries; i++) {
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    struct upb_msg *m = entries[i]->ref.msg;
    fprintf(stream, "  {.elements=&field_pointers[%d], .len=%d},\n", offset, m->num_fields);
    offset += m->num_fields;
  }
  fputs("};\n\n", stream);

  /* Emit messages. */
  fputs("static google_protobuf_DescriptorProto messages[] = {\n", stream);
  offset = 0;
  for(int i = 0; i < num_entries; i++) {
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    struct upb_msg *m = entries[i]->ref.msg;
    struct google_protobuf_DescriptorProto *d = m->descriptor;
    struct strtable_entry *e = upb_strtable_lookup(&t, d->name);
    fprintf(stream, "  {.set_flags = {.bytes={0x%02hhx}}, .name=&strings[%3d], .field=&field_arrays[%d]},\n",
            d->set_flags.bytes[0], e->num, offset);
  }
  fputs("};\n\n", stream);

  fputs("static google_protobuf_DescriptorProto *message_pointers[] = {\n", stream);
  offset = 0;
  for(int i = 0; i < num_entries; i++) {
    if(entries[i]->type != UPB_SYM_MESSAGE) continue;
    fprintf(stream, "  &messages[%d],\n", offset);
  }
  fputs("};\n\n", stream);
}

const char usage[] =
  "upbc -- upb compiler.\n"
  "upb v0.1  http://blog.reverberate.org/upb/\n"
  "\n"
  "Usage: upbc [options] input-file\n"
  "\n"
  "  -o OUTFILE-BASE    Write to OUTFILE-BASE.h and OUTFILE-BASE.c instead\n"
  "                     of using the input file as a basename.\n"
;

void usage_err(char *err)
{
  fprintf(stderr, "upbc: %s\n\n", err);
  fputs(usage, stderr);
  exit(1);
}

void error(char *err)
{
  fprintf(stderr, "upbc: %s\n\n", err);
  exit(1);
}

int main(int argc, char *argv[])
{
  /* Parse arguments. */
  char *outfile_base = NULL, *input_file = NULL;
  for(int i = 1; i < argc; i++) {
    if(strcmp(argv[i], "-o") == 0) {
      if(++i == argc)
        usage_err("-o must be followed by a FILE-BASE.");
      else if(outfile_base)
        usage_err("-o was specified multiple times.");
      outfile_base = argv[i];
    } else {
      if(input_file)
        usage_err("You can only specify one input file.");
      input_file = argv[i];
    }
  }
  if(!input_file) usage_err("You must specify an input file.");
  if(!outfile_base) outfile_base = input_file;

  /* Read input file. */
  struct upb_string descriptor;
  if(!upb_strreadfile(input_file, &descriptor))
    error("Couldn't read input file.");

  /* Parse input file. */
  struct upb_context c;
  upb_context_init(&c);
  if(!upb_context_parsefds(&c, &descriptor))
    error("Failed to parse input file descriptor.");

  /* Emit output files. */
  const int maxsize = 256;
  char h_filename[maxsize], c_filename[maxsize];
  if(snprintf(h_filename, maxsize, "%s.h", outfile_base) >= maxsize ||
     snprintf(c_filename, maxsize, "%s.c", outfile_base) >= maxsize)
    error("File base too long.\n");

  FILE *h_file = fopen(h_filename, "w"), *c_file = fopen(c_filename, "w");
  if(!h_file || !c_file)
    error("Failed to open output file(s)");

  int symcount;
  struct upb_symtab_entry **entries = strtable_to_array(&c.symtab, &symcount);
  write_h(entries, symcount, h_filename, h_file);
  write_c(entries, symcount, h_filename, c_filename, c_file);
  upb_context_free(&c);
  upb_strfree(descriptor);
  fclose(h_file);
  fclose(c_file);

  return 0;
}