diff options
Diffstat (limited to 'src/compiler/codegen_c_json_parser.c')
| -rw-r--r-- | src/compiler/codegen_c_json_parser.c | 1850 |
1 files changed, 1850 insertions, 0 deletions
diff --git a/src/compiler/codegen_c_json_parser.c b/src/compiler/codegen_c_json_parser.c new file mode 100644 index 0000000..307ce76 --- /dev/null +++ b/src/compiler/codegen_c_json_parser.c @@ -0,0 +1,1850 @@ +#include <stdlib.h> +#include "codegen_c.h" +#include "flatcc/flatcc_types.h" +#include "catalog.h" + +/* -DFLATCC_PORTABLE may help if inttypes.h is missing. */ +#ifndef PRId64 +#include <inttypes.h> +#endif + +#define PRINTLN_SPMAX 64 +static char println_spaces[PRINTLN_SPMAX]; + +static void println(fb_output_t *out, const char * format, ...) +{ + int i = out->indent * out->opts->cgen_spacing; + va_list ap; + + if (println_spaces[0] == 0) { + memset(println_spaces, 0x20, PRINTLN_SPMAX); + } + /* Don't indent on blank lines. */ + if (*format) { + while (i > PRINTLN_SPMAX) { + fprintf(out->fp, "%.*s", (int)PRINTLN_SPMAX, println_spaces); + i -= PRINTLN_SPMAX; + } + /* Use modulo to reset margin if we go too far. */ + fprintf(out->fp, "%.*s", i, println_spaces); + va_start (ap, format); + vfprintf (out->fp, format, ap); + va_end (ap); + } + fprintf(out->fp, "\n"); +} + +/* + * Unknown fields and unknown union members can be failed + * rather than ignored with a config flag. + * + * Default values an be forced with a config flat. + * + * Forward schema isn't perfect: Unknown symbolic constants + * cannot be used with known fields but will be ignored + * in ignored fields. + */ + +static int gen_json_parser_pretext(fb_output_t *out) +{ + println(out, "#ifndef %s_JSON_PARSER_H", out->S->basenameup); + println(out, "#define %s_JSON_PARSER_H", out->S->basenameup); + println(out, ""); + println(out, "/* " FLATCC_GENERATED_BY " */"); + println(out, ""); + println(out, "#include \"flatcc/flatcc_json_parser.h\""); + fb_gen_c_includes(out, "_json_parser.h", "_JSON_PARSER_H"); + gen_prologue(out); + println(out, ""); + return 0; +} + +static int gen_json_parser_footer(fb_output_t *out) +{ + gen_epilogue(out); + println(out, "#endif /* %s_JSON_PARSER_H */", out->S->basenameup); + return 0; +} + +typedef struct dict_entry dict_entry_t; +struct dict_entry { + const char *text; + int len; + void *data; + int hint; +}; + +/* Returns length of name that reminds after tag at current position. */ +static int get_dict_suffix_len(dict_entry_t *de, int pos) +{ + int n; + + n = de->len; + if (pos + 8 > n) { + return 0; + } + return n - pos - 8; +} + +/* + * Returns the length name that reminds if it terminates at the tag + * and 0 if it has a suffix. + */ +static int get_dict_tag_len(dict_entry_t *de, int pos) +{ + int n; + + n = de->len; + if (pos + 8 >= n) { + return n - pos; + } + return 0; +} + +/* + * 8 byte word part of the name starting at characert `pos` in big + * endian encoding with first char always at msb, zero padded at lsb. + * Returns length of tag [0;8]. + */ +static int get_dict_tag(dict_entry_t *de, int pos, uint64_t *tag, uint64_t *mask, + const char **tag_name, int *tag_len) +{ + int i, n = 0; + const char *a = 0; + uint64_t w = 0; + + if (pos > de->len) { + goto done; + } + a = de->text + pos; + n = de->len - pos; + if (n > 8) { + n = 8; + } + i = n; + while (i--) { + w |= ((uint64_t)a[i]) << (56 - (i * 8)); + } + *tag = w; + *mask = ~(((uint64_t)(1) << (8 - n) * 8) - 1); +done: + if (tag_name) { + *tag_name = a; + } + if (tag_len) { + *tag_len = n; + } + return n; +} + + +/* + * Find the median, but move earlier if the previous entry + * is a strict prefix within the range. + * + * `b` is inclusive. + * + * The `pos` is a window into the key at an 8 byte multiple. + * + * Only consider the range `[pos;pos+8)` and move the median + * up if an earlier key is a prefix or match within this + * window. This is needed to handle trailing data in + * a compared external key, and also to handle sub-tree + * branching when two keys has same tag at pos. + * + * Worst case we get a linear search of length 8 if all + * keys are perfect prefixes of their successor key: + * `a, ab, abc, ..., abcdefgh` + * While the midpoint stills seeks towards 'a' for longer + * such sequences, the branch logic will pool those + * squences the share prefix groups of length 8. + */ +static int split_dict_left(dict_entry_t *dict, int a, int b, int pos) +{ + int m = a + (b - a) / 2; + uint64_t wf = 0, wg = 0, wmf = 0, wmg = 0; + + while (m > a) { + get_dict_tag(&dict[m - 1], pos, &wf, &wmf, 0, 0); + get_dict_tag(&dict[m], pos, &wg, &wmg, 0, 0); + if (((wf ^ wg) & wmf) != 0) { + return m; + } + --m; + } + return m; +} + +/* + * When multiple tags are identical after split_dict_left has moved + * intersection up so a == m, we need to split in the opposite direction + * to ensure progress untill all tags in the range are identical + * at which point the trie must descend. + * + * If all tags are the same from intersection to end, b + 1 is returned + * which is not a valid element. + */ +static int split_dict_right(dict_entry_t *dict, int a, int b, int pos) +{ + int m = a + (b - a) / 2; + uint64_t wf = 0, wg = 0, wmf = 0, wmg = 0; + + while (m < b) { + get_dict_tag(&dict[m], pos, &wf, &wmf, 0, 0); + get_dict_tag(&dict[m + 1], pos, &wg, &wmg, 0, 0); + if (((wf ^ wg) & wmf) != 0) { + return m + 1; + } + ++m; + } + return m + 1; +} + +/* + * Returns the first index where the tag does not terminate at + * [pos..pos+7], or b + 1 if none exists. + */ +static int split_dict_descend(dict_entry_t *dict, int a, int b, int pos) +{ + while (a <= b) { + if (0 < get_dict_suffix_len(&dict[a], pos)) { + break; + } + ++a; + } + return a; +} + + +static int dict_cmp(const void *x, const void *y) +{ + const dict_entry_t *a = x, *b = y; + int k, n = a->len > b->len ? b->len : a->len; + + k = memcmp(a->text, b->text, (size_t)n); + return k ? k : a->len - b->len; +} + +/* Includes union vectors. */ +static inline int is_union_member(fb_member_t *member) +{ + return (member->type.type == vt_compound_type_ref || member->type.type == vt_vector_compound_type_ref) + && member->type.ct->symbol.kind == fb_is_union; +} + +static dict_entry_t *build_compound_dict(fb_compound_type_t *ct, int *count_out) +{ + fb_symbol_t *sym; + fb_member_t *member; + size_t n; + dict_entry_t *dict, *de; + char *strbuf = 0; + size_t strbufsiz = 0; + int is_union; + size_t union_index = 0; + + n = 0; + for (sym = ct->members; sym; sym = sym->link) { + member = (fb_member_t *)sym; + if (member->metadata_flags & fb_f_deprecated) { + continue; + } + is_union = is_union_member(member); + if (is_union) { + ++n; + strbufsiz += (size_t)member->symbol.ident->len + 6; + } + ++n; + } + *count_out = (int)n; + if (n == 0) { + return 0; + } + dict = malloc(n * sizeof(dict_entry_t) + strbufsiz); + if (!dict) { + return 0; + } + strbuf = (char *)dict + n * sizeof(dict_entry_t); + de = dict; + for (sym = ct->members; sym; sym = sym->link) { + member = (fb_member_t *)sym; + if (member->metadata_flags & fb_f_deprecated) { + continue; + } + de->text = member->symbol.ident->text; + de->len = (int)member->symbol.ident->len; + de->data = member; + de->hint = 0; + ++de; + is_union = is_union_member(member); + if (is_union) { + member->export_index = union_index++; + de->len = (int)member->symbol.ident->len + 5; + de->text = strbuf; + memcpy(strbuf, member->symbol.ident->text, (size_t)member->symbol.ident->len); + strbuf += member->symbol.ident->len; + strcpy(strbuf, "_type"); + strbuf += 6; + de->data = member; + de->hint = 1; + ++de; + } + } + qsort(dict, n, sizeof(dict[0]), dict_cmp); + return dict; +} + +typedef struct { + int count; + fb_schema_t *schema; + dict_entry_t *de; +} install_enum_context_t; + +static void count_visible_enum_symbol(void *context, fb_symbol_t *sym) +{ + install_enum_context_t *p = context; + + if (get_enum_if_visible(p->schema, sym)) { + p->count++; + } +} + +static void install_visible_enum_symbol(void *context, fb_symbol_t *sym) +{ + install_enum_context_t *p = context; + + if (get_enum_if_visible(p->schema, sym)) { + p->de->text = sym->ident->text; + p->de->len = (int)sym->ident->len; + p->de->data = sym; + p->de++; + } +} + +/* + * A scope dictionary contains all the enum types defined under the given + * namespace of the scope. The actually namespace is not contained in + * the name - it is an implicit prefix. It is used when looking up a + * symbolic constant assigned to a field such that the constant is first + * searched for in the same scope (namespace) as the one that defined + * the table owning the field assigned to. If that fails, a global + * namespace prefixed lookup is needed, but this is separate from this + * dictionary. In case of conflicts the local scope takes precedence + * and must be searched first. Because each table parsed can a have a + * unique local scope, we cannot install the the unprefixed lookup in + * the same dictionary as the global lookup. + * + * NOTE: the scope may have been contanimated by being expanded by a + * parent schema so we check that each symbol is visible to the current + * schema. If we didn't do this, we would risk referring to enum parsers + * that are not included in the generated source. The default empty + * namespace (i.e. scope) is an example where this easily could happen. + */ +static dict_entry_t *build_local_scope_dict(fb_schema_t *schema, fb_scope_t *scope, int *count_out) +{ + dict_entry_t *dict; + install_enum_context_t iec; + + fb_clear(iec); + + iec.schema = schema; + + fb_symbol_table_visit(&scope->symbol_index, count_visible_enum_symbol, &iec); + *count_out = iec.count; + + if (iec.count == 0) { + return 0; + } + dict = malloc((size_t)iec.count * sizeof(dict[0])); + if (!dict) { + return 0; + } + iec.de = dict; + fb_symbol_table_visit(&scope->symbol_index, install_visible_enum_symbol, &iec); + qsort(dict, (size_t)iec.count, sizeof(dict[0]), dict_cmp); + return dict; +} + +static dict_entry_t *build_global_scope_dict(catalog_t *catalog, int *count_out) +{ + size_t i, n = (size_t)catalog->nenums; + dict_entry_t *dict; + + *count_out = (int)n; + if (n == 0) { + return 0; + } + dict = malloc(n * sizeof(dict[0])); + if (!dict) { + return 0; + } + for (i = 0; i < (size_t)catalog->nenums; ++i) { + dict[i].text = catalog->enums[i].name; + dict[i].len = (int)strlen(catalog->enums[i].name); + dict[i].data = catalog->enums[i].ct; + dict[i].hint = 0; + } + qsort(dict, (size_t)catalog->nenums, sizeof(dict[0]), dict_cmp); + *count_out = catalog->nenums; + return dict; +} + +static void clear_dict(dict_entry_t *dict) +{ + if (dict) { + free(dict); + } +} + +static int gen_field_match_handler(fb_output_t *out, fb_compound_type_t *ct, void *data, int is_union_type) +{ + fb_member_t *member = data; + fb_scoped_name_t snref; + fb_symbol_text_t scope_name; + + int is_struct_container; + int is_string = 0; + int is_enum = 0; + int is_vector = 0; + int is_offset = 0; + int is_scalar = 0; + int is_optional = 0; + int is_table = 0; + int is_struct = 0; + int is_union = 0; + int is_union_vector = 0; + int is_union_type_vector = 0; + int is_base64 = 0; + int is_base64url = 0; + int is_nested = 0; + int is_array = 0; + int is_char_array = 0; + size_t array_len = 0; + fb_scalar_type_t st = 0; + const char *tname_prefix = "n/a", *tname = "n/a"; /* suppress compiler warnigns */ + fb_literal_t literal; + + fb_clear(snref); + + fb_copy_scope(ct->scope, scope_name); + is_struct_container = ct->symbol.kind == fb_is_struct; + is_optional = !!(member->flags & fb_fm_optional); + + switch (member->type.type) { + case vt_vector_type: + case vt_vector_compound_type_ref: + case vt_vector_string_type: + is_vector = 1; + break; + } + + switch (member->type.type) { + case vt_fixed_array_compound_type_ref: + case vt_vector_compound_type_ref: + case vt_compound_type_ref: + fb_compound_name(member->type.ct, &snref); + is_enum = member->type.ct->symbol.kind == fb_is_enum; + is_struct = member->type.ct->symbol.kind == fb_is_struct; + is_table = member->type.ct->symbol.kind == fb_is_table; + is_union = member->type.ct->symbol.kind == fb_is_union && !is_union_type; + if (is_enum) { + st = member->type.ct->type.st; + is_scalar = 1; + } + break; + case vt_vector_string_type: + case vt_string_type: + is_string = 1; + break; + case vt_vector_type: + /* Nested types are processed twice, once as an array, once as an object. */ + is_nested = member->nest != 0; + is_base64 = member->metadata_flags & fb_f_base64; + is_base64url = member->metadata_flags & fb_f_base64url; + is_scalar = 1; + st = member->type.st; + break; + case vt_fixed_array_type: + is_scalar = 1; + is_array = 1; + array_len = member->type.len; + st = member->type.st; + break; + case vt_scalar_type: + is_scalar = 1; + st = member->type.st; + break; + } + if (member->type.type == vt_fixed_array_compound_type_ref) { + assert(is_struct_container); + is_array = 1; + array_len = member->type.len; + } + if (is_base64 || is_base64url) { + /* Even if it is nested, parse it as a regular base64 or base64url encoded vector. */ + if (st != fb_ubyte || !is_vector) { + gen_panic(out, "internal error: unexpected base64 or base64url field type\n"); + return -1; + } + is_nested = 0; + is_vector = 0; + is_scalar = 0; + } + if (is_union_type) { + is_scalar = 0; + } + if (is_vector && is_union_type) { + is_union_type_vector = 1; + is_vector = 0; + } + if (is_vector && is_union) { + is_union_vector = 1; + is_vector = 0; + } + if (is_array && is_scalar && st == fb_char) { + is_array = 0; + is_scalar = 0; + is_char_array = 1; + } + if (is_nested == 1) { + println(out, "if (buf != end && *buf == '[') { /* begin nested */"); indent(); + } +repeat_nested: + if (is_nested == 2) { + unindent(); println(out, "} else { /* nested */"); indent(); + fb_compound_name((fb_compound_type_t *)&member->nest->symbol, &snref); + if (member->nest->symbol.kind == fb_is_table) { + is_table = 1; + } else { + is_struct = 1; + } + is_vector = 0; + is_scalar = 0; + println(out, "if (flatcc_builder_start_buffer(ctx->ctx, 0, 0, 0)) goto failed;"); + } + is_offset = !is_scalar && !is_struct && !is_union_type; + + if (is_scalar) { + tname_prefix = scalar_type_prefix(st); + tname = st == fb_bool ? "uint8_t" : scalar_type_name(st); + } + + /* Other types can also be vector, so we wrap. */ + if (is_vector) { + if (is_offset) { + println(out, "if (flatcc_builder_start_offset_vector(ctx->ctx)) goto failed;"); + } else { + println(out, + "if (flatcc_builder_start_vector(ctx->ctx, %"PRIu64", %hu, UINT64_C(%"PRIu64"))) goto failed;", + (uint64_t)member->size, (short)member->align, + (uint64_t)FLATBUFFERS_COUNT_MAX(member->size)); + } + } + if (is_array) { + if (is_scalar) { + println(out, "size_t count = %d;", array_len); + println(out, "%s *base = (%s *)((size_t)struct_base + %"PRIu64");", + tname, tname, (uint64_t)member->offset); + } + else { + println(out, "size_t count = %d;", array_len); + println(out, "void *base = (void *)((size_t)struct_base + %"PRIu64");", + (uint64_t)member->offset); + } + } + if (is_char_array) { + println(out, "char *base = (char *)((size_t)struct_base + %"PRIu64");", + (uint64_t)member->offset); + println(out, "buf = flatcc_json_parser_char_array(ctx, buf, end, base, %d);", array_len); + } + if (is_array || is_vector) { + println(out, "buf = flatcc_json_parser_array_start(ctx, buf, end, &more);"); + /* Note that we reuse `more` which is safe because it is updated at the end of the main loop. */ + println(out, "while (more) {"); indent(); + } + if (is_scalar) { + println(out, "%s val = 0;", tname); + println(out, "static flatcc_json_parser_integral_symbol_f *symbolic_parsers[] = {"); + indent(); indent(); + /* + * The scope name may be empty when no namespace is used. In that + * case the global scope is the same, but performance the + * duplicate doesn't matter. + */ + if (is_enum) { + println(out, "%s_parse_json_enum,", snref.text); + println(out, "%s_local_%sjson_parser_enum,", out->S->basename, scope_name); + println(out, "%s_global_json_parser_enum, 0 };", out->S->basename); + } else { + println(out, "%s_local_%sjson_parser_enum,", out->S->basename, scope_name); + println(out, "%s_global_json_parser_enum, 0 };", out->S->basename); + } + unindent(); unindent(); + } + /* It is not safe to acquire the pointer before building element table or string. */ + if (is_vector && !is_offset) { + println(out, "if (!(pval = flatcc_builder_extend_vector(ctx->ctx, 1))) goto failed;"); + } + if (is_struct_container) { + if (!is_array && !is_char_array) { + /* `struct_base` is given as argument to struct parsers. */ + println(out, "pval = (void *)((size_t)struct_base + %"PRIu64");", (uint64_t)member->offset); + } + } else if (is_struct && !is_vector) { + /* Same logic as scalars in tables, but scalars must be tested for default. */ + println(out, + "if (!(pval = flatcc_builder_table_add(ctx->ctx, %"PRIu64", %"PRIu64", %"PRIu16"))) goto failed;", + (uint64_t)member->id, (uint64_t)member->size, (uint16_t)member->align); + } + if (is_scalar) { + println(out, "buf = flatcc_json_parser_%s(ctx, (mark = buf), end, &val);", tname_prefix); + println(out, "if (mark == buf) {"); indent(); + println(out, "buf = flatcc_json_parser_symbolic_%s(ctx, (mark = buf), end, symbolic_parsers, &val);", tname_prefix); + println(out, "if (buf == mark || buf == end) goto failed;"); + unindent(); println(out, "}"); + if (!is_struct_container && !is_vector && !is_base64 && !is_base64url) { +#if !FLATCC_JSON_PARSE_FORCE_DEFAULTS + /* We need to create a check for the default value and create a table field if not the default. */ + if (!is_optional) { + if (!print_literal(st, &member->value, literal)) return -1; + println(out, "if (val != %s || (ctx->flags & flatcc_json_parser_f_force_add)) {", literal); indent(); + } +#endif + println(out, "if (!(pval = flatcc_builder_table_add(ctx->ctx, %"PRIu64", %"PRIu64", %hu))) goto failed;", + (uint64_t)member->id, (uint64_t)member->size, (short)member->align); +#if !FLATCC_JSON_PARSE_FORCE_DEFAULTS +#endif + } + /* For scalars in table field, and in struct container. */ + if (is_array) { + println(out, "if (count) {"); indent(); + println(out, "%s%s_write_to_pe(base, val);", out->nsc, tname_prefix); + println(out, "--count;"); + println(out, "++base;"); + unindent(); println(out, "} else if (!(ctx->flags & flatcc_json_parser_f_skip_array_overflow)) {"); indent(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_array_overflow);"); + unindent(); println(out, "}"); + } else { + println(out, "%s%s_write_to_pe(pval, val);", out->nsc, tname_prefix); + } + if (!is_struct_container && !is_vector && !(is_scalar && is_optional)) { + unindent(); println(out, "}"); + } + } else if (is_struct) { + if (is_array) { + println(out, "if (count) {"); indent(); + println(out, "buf = %s_parse_json_struct_inline(ctx, buf, end, base);", snref.text); + println(out, "--count;"); + println(out, "base = (void *)((size_t)base + %"PRIu64");", member->type.ct->size); + unindent(); println(out, "} else if (!(ctx->flags & flatcc_json_parser_f_skip_array_overflow)) {"); indent(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_array_overflow);"); + unindent(); println(out, "}"); + } else { + println(out, "buf = %s_parse_json_struct_inline(ctx, buf, end, pval);", snref.text); + } + } else if (is_string) { + println(out, "buf = flatcc_json_parser_build_string(ctx, buf, end, &ref);"); + } else if (is_base64 || is_base64url) { + println(out, "buf = flatcc_json_parser_build_uint8_vector_base64(ctx, buf, end, &ref, %u);", + !is_base64); + } else if (is_table) { + println(out, "buf = %s_parse_json_table(ctx, buf, end, &ref);", snref.text); + } else if (is_union) { + if (is_union_vector) { + println(out, "buf = flatcc_json_parser_union_vector(ctx, buf, end, %"PRIu64", %"PRIu64", h_unions, %s_parse_json_union);", + (uint64_t)member->export_index, member->id, snref.text); + } else { + println(out, "buf = flatcc_json_parser_union(ctx, buf, end, %"PRIu64", %"PRIu64", h_unions, %s_parse_json_union);", + (uint64_t)member->export_index, member->id, snref.text); + } + } else if (is_union_type) { + println(out, "static flatcc_json_parser_integral_symbol_f *symbolic_parsers[] = {"); + indent(); indent(); + println(out, "%s_parse_json_enum,", snref.text); + println(out, "%s_local_%sjson_parser_enum,", out->S->basename, scope_name); + println(out, "%s_global_json_parser_enum, 0 };", out->S->basename); + unindent(); unindent(); + if (is_union_type_vector) { + println(out, "buf = flatcc_json_parser_union_type_vector(ctx, buf, end, %"PRIu64", %"PRIu64", h_unions, symbolic_parsers, %s_parse_json_union, %s_json_union_accept_type);", + (uint64_t)member->export_index, member->id, snref.text, snref.text); + } else { + println(out, "buf = flatcc_json_parser_union_type(ctx, buf, end, %"PRIu64", %"PRIu64", h_unions, symbolic_parsers, %s_parse_json_union);", + (uint64_t)member->export_index, member->id, snref.text); + } + } else if (!is_vector && !is_char_array) { + gen_panic(out, "internal error: unexpected type for trie member\n"); + return -1; + } + if (is_vector) { + if (is_offset) { + /* Deal with table and string vector elements - unions cannot be elements. */ + println(out, "if (!ref || !(pref = flatcc_builder_extend_offset_vector(ctx->ctx, 1))) goto failed;"); + /* We don't need to worry about endian conversion - offsets vectors fix this automatically. */ + println(out, "*pref = ref;"); + } + println(out, "buf = flatcc_json_parser_array_end(ctx, buf, end, &more);"); + unindent(); println(out, "}"); + if (is_offset) { + println(out, "ref = flatcc_builder_end_offset_vector(ctx->ctx);"); + } else { + println(out, "ref = flatcc_builder_end_vector(ctx->ctx);"); + } + } + if (is_array) { + println(out, "buf = flatcc_json_parser_array_end(ctx, buf, end, &more);"); + unindent(); println(out, "}"); + println(out, "if (count) {"); indent(); + println(out, "if (ctx->flags & flatcc_json_parser_f_reject_array_underflow) {"); indent(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_array_underflow);"); + unindent(); println(out, "}"); + if (is_scalar) { + println(out, "memset(base, 0, count * sizeof(*base));"); + } else { + println(out, "memset(base, 0, count * %"PRIu64");", (uint64_t)member->type.ct->size); + } + unindent(); println(out, "}"); + } + if (is_nested == 1) { + is_nested = 2; + goto repeat_nested; + } + if (is_nested == 2) { + println(out, "if (!ref) goto failed;"); + println(out, "ref = flatcc_builder_end_buffer(ctx->ctx, ref);"); + unindent(); println(out, "} /* end nested */"); + } + if (is_nested || is_vector || is_table || is_string || is_base64 || is_base64url) { + println(out, "if (!ref || !(pref = flatcc_builder_table_add_offset(ctx->ctx, %"PRIu64"))) goto failed;", member->id); + println(out, "*pref = ref;"); + } + return 0; +} + +static void gen_field_match(fb_output_t *out, fb_compound_type_t *ct, void *data, int hint, int n) +{ + println(out, "buf = flatcc_json_parser_match_symbol(ctx, (mark = buf), end, %d);", n); + println(out, "if (mark != buf) {"); indent(); + gen_field_match_handler(out, ct, data, hint); + unindent(); println(out, "} else {"); indent(); +} + +/* This also handles union type enumerations. */ +static void gen_enum_match_handler(fb_output_t *out, fb_compound_type_t *ct, void *data, int unused_hint) +{ + fb_member_t *member = data; + + (void)unused_hint; + + /* + * This is rather unrelated to the rest, we just use the same + * trie generation logic. Here we simply need to assign a known + * value to the enum parsers output arguments. + */ + switch (ct->type.st) { + case fb_bool: + case fb_ubyte: + case fb_ushort: + case fb_uint: + case fb_ulong: + println(out, "*value = UINT64_C(%"PRIu64"), *value_sign = 0;", + member->value.u); + break; + case fb_byte: + case fb_short: + case fb_int: + case fb_long: + if (member->value.i < 0) { + println(out, "*value = UINT64_C(%"PRIu64"), *value_sign = 1;", member->value.i); + } else { + println(out, "*value = UINT64_C(%"PRIu64"), *value_sign = 0;", member->value.i); + } + break; + default: + gen_panic(out, "internal error: invalid enum type\n"); + } +} + +static void gen_enum_match(fb_output_t *out, fb_compound_type_t *ct, void *data, int hint, int n) +{ + println(out, "buf = flatcc_json_parser_match_constant(ctx, (mark = buf), end, %d, aggregate);", n); + println(out, "if (buf != mark) {"); indent(); + gen_enum_match_handler(out, ct, data, hint); + unindent(); println(out, "} else {"); indent(); +} + +static void gen_scope_match_handler(fb_output_t *out, fb_compound_type_t *unused_ct, void *data, int unused_hint) +{ + fb_compound_type_t *ct = data; + fb_scoped_name_t snt; + + (void)unused_ct; + (void)unused_hint; + assert(ct->symbol.kind == fb_is_enum || ct->symbol.kind == fb_is_union); + + fb_clear(snt); + fb_compound_name(ct, &snt); + /* May be included from another file. Unions also have _enum parsers. */ + println(out, "buf = %s_parse_json_enum(ctx, buf, end, value_type, value, aggregate);", snt.text); +} + +static void gen_scope_match(fb_output_t *out, fb_compound_type_t *ct, void *data, int hint, int n) +{ + println(out, "buf = flatcc_json_parser_match_scope(ctx, (mark = buf), end, %d);", n); + println(out, "if (buf != mark) {"); indent(); + gen_scope_match_handler(out, ct, data, hint); + unindent(); println(out, "} else {"); indent(); +} + +static void gen_field_unmatched(fb_output_t *out) +{ + println(out, "buf = flatcc_json_parser_unmatched_symbol(ctx, buf, end);"); +} + +static void gen_enum_unmatched(fb_output_t *out) +{ + println(out, "return unmatched;"); +} + +static void gen_scope_unmatched(fb_output_t *out) +{ + println(out, "return unmatched;"); +} + +/* + * Generate a trie for all members or a compound type. + * This may be a struct or a table. + * + * We have a ternary trie where a search word w compares: + * w < wx_tag is one branch [a;x), iff a < x. + * w > wx_tag is another branch (y;b], iff b > y + * and w == wx_tag is a third branch [x;y]. + * + * The sets [a;x) and (y;b] may be empty in which case a non-match + * action is triggered. + * + * [x..y] is a set of one or more fields that share the same tag at the + * current position. The first (and only the first) field name in this + * set may terminate withint the current tag (when suffix length k == + * 0). There is therefore potentially both a direct field action and a + * sub-tree action. Once there is only one field in the set and the + * field name terminates within the current tag, the search word is + * masked and tested against the field tag and the search word is also + * tested for termination in the buffer at the first position after the + * field match. If the termination was not found a non-match action is + * triggered. + * + * A non-match action may be to silently consume the rest of the + * search identifier and then the json value, or to report and + * error. + * + * A match action triggers a json value parse of a known type + * which updates into a flatcc builder object. If the type is + * basic (string or scalar) the update simple, otherwise if + * the type is within the same schema, we push context + * and switch to parse the nested type, otherwise we call + * a parser in another schema. When a trie is done, we + * switch back context if in the same schema. The context + * lives on a stack. This avoids deep recursion because + * schema parsers are not mutually recursive. + * + * The trie is also used to parse enums and scopes (namespace prefixes) + * with a slight modification. + */ + +enum trie_type { table_trie, struct_trie, enum_trie, local_scope_trie, global_scope_trie }; +typedef struct trie trie_t; + +typedef void gen_match_f(fb_output_t *out, fb_compound_type_t *ct, void *data, int hint, int n); +typedef void gen_unmatched_f(fb_output_t *out); + +struct trie { + dict_entry_t *dict; + gen_match_f *gen_match; + gen_unmatched_f *gen_unmatched; + /* Not used with scopes. */ + fb_compound_type_t *ct; + int type; + int union_total; + int label; +}; + +/* + * This function is a final handler of the `gen_trie` function. Often + * just to handle a single match, but also to handle a prefix range + * special case like keys in `{ a, alpha, alpha2 }`. + * + * (See also special case of two non-prefix keys below). + * + * We know that all keys [a..b] have length in the range [pos..pos+8) + * and also that key x is proper prefix of key x + 1, x in [a..b). + * + * It is possible that `a == b`. + * + * We conduct a binary search by testing the middle for masked match and + * gradually refine until we do not have a match or have a single + * element match. + * + * (An alternative algorithm xors 8 byte tag with longest prefix and + * finds ceiling of log 2 using a few bit logic operations or intrinsic + * zero count and creates a jump table of at most 8 elements, but is + * hardly worthwhile vs 3 comparisons and 3 AND operations and often + * less than that.) + * + * Once we have a single element match we need to confirm the successor + * symbol is not any valid key - this differs among trie types and is + * therefore the polymorph match logic handles the final confirmed match + * or mismatch. + * + * Each trie type has special operation for implementing a matched and + * a failed match. Our job is to call these for each key in the range. + * + * While not the original intention, the `gen_prefix_trie` also handles the + * special case where the set has two keys where one is not a prefix of + * the other, but both terminate in the same tag. In this case we can + * immediately do an exact match test and skip the less than + * comparision. We need no special code for this, assuming the function + * is called correctly. This significantly reduces the branching in a + * case like "Red, Green, Blue". + * + * If `label` is positive, it is used to jump to additional match logic + * when a prefix was not matched. If 0 there is no additional logic and + * the symbol is considered unmatched immediately. + */ +static void gen_prefix_trie(fb_output_t *out, trie_t *trie, int a, int b, int pos, int label) +{ + int m, n; + uint64_t tag = 00, mask = 0; + const char *name; + int len; + + /* + * Weigh the intersection towards the longer prefix. Notably if we + * have two keys it makes no sense to check the shorter key first. + */ + m = a + (b - a + 1) / 2; + + n = get_dict_tag(&trie->dict[m], pos, &tag, &mask, &name, &len); + if (n == 8) { + println(out, "if (w == 0x%"PRIx64") { /* \"%.*s\" */", tag, len, name); indent(); + } else { + println(out, "if ((w & 0x%"PRIx64") == 0x%"PRIx64") { /* \"%.*s\" */", + mask, tag, len, name); indent(); + } + if (m == a) { + /* There can be only one. */ + trie->gen_match(out, trie->ct, trie->dict[m].data, trie->dict[m].hint, n); + if (label > 0) { + println(out, "goto pfguard%d;", label); + } else { + trie->gen_unmatched(out); + } + unindent(); println(out, "}"); + unindent(); println(out, "} else { /* \"%.*s\" */", len, name); indent(); + if (label > 0) { + println(out, "goto pfguard%d;", label); + } else { + trie->gen_unmatched(out); + } + } else { + if (m == b) { + trie->gen_match(out, trie->ct, trie->dict[m].data, trie->dict[m].hint, n); + if (label > 0) { + println(out, "goto pfguard%d;", label); + } else { + trie->gen_unmatched(out); + } + unindent(); println(out, "}"); + } else { + gen_prefix_trie(out, trie, m, b, pos, label); + } + unindent(); println(out, "} else { /* \"%.*s\" */", len, name); indent(); + gen_prefix_trie(out, trie, a, m - 1, pos, label); + } + unindent(); println(out, "} /* \"%.*s\" */", len, name); +} + +static void gen_trie(fb_output_t *out, trie_t *trie, int a, int b, int pos) +{ + int x, k; + uint64_t tag = 0, mask = 0; + const char *name = ""; + int len = 0, has_prefix_key = 0, prefix_guard = 0, has_descend; + int label = 0; + + /* + * Process a trie at the level given by pos. A single level covers + * one tag. + * + * A tag is a range of 8 characters [pos..pos+7] that is read as a + * single big endian word and tested as against a ternary trie + * generated in code. In generated code the tag is stored in "w". + * + * Normally trailing data in a tag is not a problem + * because the difference between two keys happen in the middle and + * trailing data is not valid key material. When the difference is + * at the end, we get a lot of special cases to handle. + * + * Regardless, when we believe we have a match, a final check is + * made to ensure that the next character after the match is not a + * valid key character - for quoted keys a valid termiantot is a + * quote, for unquoted keys it can be one of several characters - + * therefore quoted keys are faster to parse, even if they consume + * more space. The trie does not care about these details, the + * gen_match function handles this transparently for different + * symbol types. + */ + + + /* + * If we have one or two keys that terminate in this tag, there is no + * need to do a branch test before matching exactly. + * + * We observe that `gen_prefix_trie` actually handles this + * case well, even though it was not designed for it. + */ + if ((get_dict_suffix_len(&trie->dict[a], pos) == 0) && + (b == a || (b == a + 1 && get_dict_suffix_len(&trie->dict[b], pos) == 0))) { + gen_prefix_trie(out, trie, a, b, pos, 0); + return; + } + + /* + * Due trie nature, we have a left, middle, and right range where + * the middle range all compare the same at the current trie level + * when masked against shortest (and first) key in middle range. + */ + x = split_dict_left(trie->dict, a, b, pos); + + if (x > a) { + /* + * This is normal early branch with a key `a < x < b` such that + * any shared prefix ranges do not span x. + */ + get_dict_tag(&trie->dict[x], pos, &tag, &mask, &name, &len); + println(out, "if (w < 0x%"PRIx64") { /* branch \"%.*s\" */", tag, len, name); indent(); + gen_trie(out, trie, a, x - 1, pos); + unindent(); println(out, "} else { /* branch \"%.*s\" */", len, name); indent(); + gen_trie(out, trie, x, b, pos); + unindent(); println(out, "} /* branch \"%.*s\" */", len, name); + return; + } + x = split_dict_right(trie->dict, a, b, pos); + + /* + * [a .. x-1] is a non-empty sequence of prefixes, + * for example { a123, a1234, a12345 }. + * The keys might not terminate in the current tag. To find those + * that do, we will evaluate k such that: + * [a .. k-1] are prefixes that terminate in the current tag if any + * such exists. + * [x..b] are keys that are prefixes up to at least pos + 7 but + * do not terminate in the current tag. + * [k..x-1] are prefixes that do not termiante in the current tag. + * Note that they might not be prefixes when considering more than the + * current tag. + * The range [a .. x-1] can ge generated with `gen_prefix_trie`. + * + * We generally have the form + * + * [a..b] = + * (a)<prefixes>, (k-1)<descend-prefix>, (k)<descend>, (x)<reminder> + * + * Where <prefixes> are keys that terminate at the current tag. + * <descend> are keys that have the prefixes as prefix but do not + * terminate at the current tag. + * <descend-prerfix> is a single key that terminates exactly + * where the tag ends. If there are no descend keys it is part of + * prefixes, otherwise it is tested as a special case. + * <reminder> are any keys larger than the prefixes. + * + * The reminder keys cannot be tested before we are sure that no + * prefix is matching at least no prefixes that is not a + * descend-prefix. This is because less than comparisons are + * affected by trailing data within the tag caused by prefixes + * terminating early. Trailing data is not a problem if two keys are + * longer than the point where they differ even if they terminate + * within the current tag. + * + * Thus, if we have non-empty <descend> and non-empty <reminder>, + * the reminder must guard against any matches in prefix but not + * against any matches in <descend>. If <descend> is empty and + * <prefixes> == <descend-prefix> a guard is also not needed. + */ + + /* Find first prefix that does not terminate at the current level, or x if absent */ + k = split_dict_descend(trie->dict, a, x - 1, pos); + has_descend = k < x; + + /* If we have a descend, process that in isolation. */ + if (has_descend) { + has_prefix_key = k > a && get_dict_tag_len(&trie->dict[k - 1], pos) == 8; + get_dict_tag(&trie->dict[k], pos, &tag, &mask, &name, &len); + println(out, "if (w == 0x%"PRIx64") { /* descend \"%.*s\" */", tag, len, name); indent(); + if (has_prefix_key) { + /* We have a key that terminates at the descend prefix. */ + println(out, "/* descend prefix key \"%.*s\" */", len, name); + trie->gen_match(out, trie->ct, trie->dict[k - 1].data, trie->dict[k - 1].hint, 8); + println(out, "/* descend suffix \"%.*s\" */", len, name); + } + println(out, "buf += 8;"); + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, trie, k, x - 1, pos + 8); + if (has_prefix_key) { + unindent(); println(out, "} /* desend suffix \"%.*s\" */", len, name); + /* Here we move the <descend-prefix> key out of the <descend> range. */ + --k; + } + unindent(); println(out, "} else { /* descend \"%.*s\" */", len, name); indent(); + } + prefix_guard = a < k && x <= b; + if (prefix_guard) { + label = ++trie->label; + } + if (a < k) { + gen_prefix_trie(out, trie, a, k - 1, pos, label); + } + if (prefix_guard) { + /* All prefixes tested, but none matched. */ + println(out, "goto endpfguard%d;", label); + margin(); + println(out, "pfguard%d:", label); + unmargin(); + } + if (x <= b) { + gen_trie(out, trie, x, b, pos); + } else if (a >= k) { + trie->gen_unmatched(out); + } + if (prefix_guard) { + margin(); + println(out, "endpfguard%d:", label); + unmargin(); + println(out, "(void)0;"); + } + if (has_descend) { + unindent(); println(out, "} /* descend \"%.*s\" */", len, name); + } +} + + +/* + * Parsing symbolic constants: + * + * An enum parser parses the local symbols and translate them into + * numeric values. + * + * If a symbol wasn't matched, e.g. "Red", it might be matched with + * "Color.Red" but the enum parser does not handle this. + * + * Instead a scope parser maps each type in the scope to a call + * to an enum parser, e.g. "Color." maps to a color enum parser + * that understands "Red". If this also fails, a call is made + * to a global scope parser that maps a namespace to a local + * scope parser, for example "Graphics.Color.Red" first + * recognizes the namespace "Graphics." which may or may not + * be the same as the local scope tried earlier, then "Color." + * is matched and finally "Red". + * + * The scope and namespace parsers may cover extend namespaces from + * include files so each file calls into dependencies as necessary. + * This means the same scope can have multiple parsers and must + * therefore be name prefixed by the basename of the include file. + * + * The enums can only exist in a single file. + * + * The local scope is defined as the scope in which the consuming + * fields container is defined, so if Pen is a table in Graphics + * with a field named "ink" and the pen is parsed as + * { "ink": "Color.Red" }, then Color would be parsed in the + * Graphics scope. If ink was and enum of type Color, the enum + * parser would be tried first. If ink was, say, an integer + * type, it would not try an enum parse first but try the local + * scope, then the namespace scope. + * + * It is permitted to have multiple symbols in a string when + * the enum type has flag attribute so values can be or'ed together. + * The parser does not attempt to validate this and will simple + * 'or' together multiple values after coercing each to the + * receiving field type: "Has.ink Has.shape Has.brush". + */ + + +/* + * Used by scalar/enum/union_type table fields to look up symbolic + * constants in same scope as the table was defined, thus avoiding + * namespace prefix. + * + * Theh matched name then calls into the type specific parser which + * may be in a dependent file. + * + * Because each scope may be extended in dependent schema files + * we recreate the scope in full in each file. + */ +static void gen_local_scope_parser(void *context, fb_scope_t *scope) +{ + fb_output_t *out = context; + int n = 0; + trie_t trie; + fb_symbol_text_t scope_name; + + fb_clear(trie); + fb_copy_scope(scope, scope_name); + if (((trie.dict = build_local_scope_dict(out->S, scope, &n)) == 0) && n > 0) { + gen_panic(out, "internal error: could not build dictionary for json parser\n"); + return; + } + /* Not used for scopes. */ + trie.ct = 0; + trie.type = local_scope_trie; + trie.gen_match = gen_scope_match; + trie.gen_unmatched = gen_scope_unmatched; + println(out, "static const char *%s_local_%sjson_parser_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", + out->S->basename, scope_name); + indent(); indent(); + println(out, "int *value_type, uint64_t *value, int *aggregate)"); + unindent(); unindent(); + println(out, "{"); indent(); + if (n == 0) { + println(out, "/* Scope has no enum / union types to look up. */"); + println(out, "return buf; /* unmatched; */"); + unindent(); println(out, "}"); + } else { + println(out, "const char *unmatched = buf;"); + println(out, "const char *mark;"); + println(out, "uint64_t w;"); + println(out, ""); + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, &trie, 0, n - 1, 0); + println(out, "return buf;"); + unindent(); println(out, "}"); + } + println(out, ""); + clear_dict(trie.dict); +} + +/* + * This parses namespace prefixed types. Because scopes can be extended + * in dependent schema files, each file has its own global scope parser. + * The matched types call into type specific parsers that may be in + * a dependent file. + * + * When a local scope is also parsed, it should be tried before the + * global scope. + */ +static int gen_global_scope_parser(fb_output_t *out) +{ + int n = 0; + trie_t trie; + catalog_t catalog; + + fb_clear(trie); + if (build_catalog(&catalog, out->S, 1, &out->S->root_schema->scope_index)) { + return -1; + } + + if ((trie.dict = build_global_scope_dict(&catalog, &n)) == 0 && n > 0) { + clear_catalog(&catalog); + gen_panic(out, "internal error: could not build dictionary for json parser\n"); + return -1; + } + /* Not used for scopes. */ + trie.ct = 0; + trie.type = global_scope_trie; + trie.gen_match = gen_scope_match; + trie.gen_unmatched = gen_scope_unmatched; + println(out, "static const char *%s_global_json_parser_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", out->S->basename); + indent(); indent(); + println(out, "int *value_type, uint64_t *value, int *aggregate)"); + unindent(); unindent(); + println(out, "{"); indent(); + if (n == 0) { + println(out, "/* Global scope has no enum / union types to look up. */"); + println(out, "return buf; /* unmatched; */"); + unindent(); println(out, "}"); + } else { + println(out, "const char *unmatched = buf;"); + println(out, "const char *mark;"); + println(out, "uint64_t w;"); + println(out, ""); + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, &trie, 0, n - 1, 0); + println(out, "return buf;"); + unindent(); println(out, "}"); + } + println(out, ""); + clear_dict(trie.dict); + clear_catalog(&catalog); + return 0; +} + +/* + * Constants have the form `"Red"` or `Red` but may also be part + * of a list of flags: `"Normal High Average"` or `Normal High + * Average`. `more` indicates more symbols follow. + * + * Returns input argument if there was no valid match, + * `end` on syntax error, and `more=1` if matched and + * there are more constants to parse. + * Applies the mached and coerced constant to `pval` + * with a binary `or` operation so `pval` must be initialized + * to 0 before teh first constant in a list. + */ +static int gen_enum_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + int n = 0; + trie_t trie; + + fb_clear(trie); + assert(ct->symbol.kind == fb_is_enum || ct->symbol.kind == fb_is_union); + + if ((trie.dict = build_compound_dict(ct, &n)) == 0 && n > 0) { + gen_panic(out, "internal error: could not build dictionary for json parser\n"); + return -1; + } + trie.ct = ct; + trie.type = enum_trie; + trie.gen_match = gen_enum_match; + trie.gen_unmatched = gen_enum_unmatched; + + fb_clear(snt); + fb_compound_name(ct, &snt); + + println(out, "static const char *%s_parse_json_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", snt.text); + indent(); indent(); + println(out, "int *value_sign, uint64_t *value, int *aggregate)"); + unindent(); unindent(); + println(out, "{"); indent(); + if (n == 0) { + println(out, "/* Enum has no fields. */"); + println(out, "*aggregate = 0;"); + println(out, "return buf; /* unmatched; */"); + unindent(); println(out, "}"); + } else { + println(out, "const char *unmatched = buf;"); + println(out, "const char *mark;"); + println(out, "uint64_t w;"); + println(out, ""); + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, &trie, 0, n - 1, 0); + println(out, "return buf;"); + unindent(); println(out, "}"); + } + println(out, ""); + clear_dict(trie.dict); + return 0; +} + +/* + * We do not check for duplicate settings or missing struct fields. + * Missing fields are zeroed. + * + * TODO: we should track nesting level because nested structs do not + * interact with the builder so the builders level limit will not kick + * in. As long as we get input from our own parser we should, however, + * be reasonable safe as nesting is bounded. + */ +static int gen_struct_parser_inline(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + int n; + trie_t trie; + + fb_clear(trie); + assert(ct->symbol.kind == fb_is_struct); + if ((trie.dict = build_compound_dict(ct, &n)) == 0 && n > 0) { + gen_panic(out, "internal error: could not build dictionary for json parser\n"); + return -1; + } + trie.ct = ct; + trie.type = struct_trie; + trie.gen_match = gen_field_match; + trie.gen_unmatched = gen_field_unmatched; + + fb_clear(snt); + fb_compound_name(ct, &snt); + println(out, "static const char *%s_parse_json_struct_inline(flatcc_json_parser_t *ctx, const char *buf, const char *end, void *struct_base)", snt.text); + println(out, "{"); indent(); + println(out, "int more;"); + if (n > 0) { + println(out, "flatcc_builder_ref_t ref;"); + println(out, "void *pval;"); + println(out, "const char *mark;"); + println(out, "uint64_t w;"); + } + println(out, ""); + println(out, "buf = flatcc_json_parser_object_start(ctx, buf, end, &more);"); + println(out, "while (more) {"); indent(); + if (n == 0) { + println(out, "/* Empty struct. */"); + println(out, "buf = flatcc_json_parser_unmatched_symbol(ctx, buf, end);"); + } else { + println(out, "buf = flatcc_json_parser_symbol_start(ctx, buf, end);"); + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, &trie, 0, n - 1, 0); + } + println(out, "buf = flatcc_json_parser_object_end(ctx, buf, end , &more);"); + unindent(); println(out, "}"); + println(out, "return buf;"); + if (n > 0) { + /* Set runtime error if no other error was set already. */ + margin(); + println(out, "failed:"); + unmargin(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_runtime);"); + } + unindent(); println(out, "}"); + println(out, ""); + clear_dict(trie.dict); + return 0; +} + +static int gen_struct_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + + assert(ct->symbol.kind == fb_is_struct); + fb_clear(snt); + fb_compound_name(ct, &snt); + println(out, "static const char *%s_parse_json_struct(flatcc_json_parser_t *ctx, const char *buf, const char *end, flatcc_builder_ref_t *result)", snt.text); + println(out, "{"); indent(); + println(out, "void *pval;"); + println(out, ""); + println(out, "*result = 0;"); + println(out, "if (!(pval = flatcc_builder_start_struct(ctx->ctx, %"PRIu64", %"PRIu16"))) goto failed;", + (uint64_t)ct->size, (uint16_t)ct->align); + println(out, "buf = %s_parse_json_struct_inline(ctx, buf, end, pval);", snt.text); + println(out, "if (ctx->error || !(*result = flatcc_builder_end_struct(ctx->ctx))) goto failed;"); + println(out, "return buf;"); + margin(); + println(out, "failed:"); + unmargin(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_runtime);"); + unindent(); println(out, "}"); + println(out, ""); + println(out, "static inline int %s_parse_json_as_root(flatcc_builder_t *B, flatcc_json_parser_t *ctx, const char *buf, size_t bufsiz, flatcc_json_parser_flags_t flags, const char *fid)", snt.text); + println(out, "{"); indent(); + println(out, "return flatcc_json_parser_struct_as_root(B, ctx, buf, bufsiz, flags, fid, %s_parse_json_struct);", + snt.text); + unindent(); println(out, "}"); + println(out, ""); + return 0; +} + +static int gen_table_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + fb_member_t *member; + int first, i, n; + int is_union, is_required; + trie_t trie; + + fb_clear(trie); + assert(ct->symbol.kind == fb_is_table); + if ((trie.dict = build_compound_dict(ct, &n)) == 0 && n > 0) { + gen_panic(out, "internal error: could not build dictionary for json parser\n"); + return -1; + } + trie.ct = ct; + trie.type = table_trie; + trie.gen_match = gen_field_match; + trie.gen_unmatched = gen_field_unmatched; + + trie.union_total = 0; + for (i = 0; i < n; ++i) { + trie.union_total += !!trie.dict[i].hint; + } + + fb_clear(snt); + fb_compound_name(ct, &snt); + println(out, "static const char *%s_parse_json_table(flatcc_json_parser_t *ctx, const char *buf, const char *end, flatcc_builder_ref_t *result)", snt.text); + println(out, "{"); indent(); + println(out, "int more;"); + + if (n > 0) { + println(out, "void *pval;"); + println(out, "flatcc_builder_ref_t ref, *pref;"); + println(out, "const char *mark;"); + println(out, "uint64_t w;"); + } + if (trie.union_total) { + println(out, "size_t h_unions;"); + } + println(out, ""); + println(out, "*result = 0;"); + println(out, "if (flatcc_builder_start_table(ctx->ctx, %"PRIu64")) goto failed;", + ct->count); + if (trie.union_total) { + println(out, "if (end == flatcc_json_parser_prepare_unions(ctx, buf, end, %"PRIu64", &h_unions)) goto failed;", (uint64_t)trie.union_total); + } + println(out, "buf = flatcc_json_parser_object_start(ctx, buf, end, &more);"); + println(out, "while (more) {"); indent(); + println(out, "buf = flatcc_json_parser_symbol_start(ctx, buf, end);"); + if (n > 0) { + println(out, "w = flatcc_json_parser_symbol_part(buf, end);"); + gen_trie(out, &trie, 0, n - 1, 0); + } else { + println(out, "/* Table has no fields. */"); + println(out, "buf = flatcc_json_parser_unmatched_symbol(ctx, buf, end);"); + } + println(out, "buf = flatcc_json_parser_object_end(ctx, buf, end, &more);"); + unindent(); println(out, "}"); + println(out, "if (ctx->error) goto failed;"); + for (first = 1, i = 0; i < n; ++i) { + member = trie.dict[i].data; + if (member->metadata_flags & fb_f_deprecated) { + continue; + } + is_union = is_union_member(member); + is_required = member->metadata_flags & fb_f_required; + if (is_required) { + if (first) { + println(out, "if (!flatcc_builder_check_required_field(ctx->ctx, %"PRIu64")", member->id - !!is_union); + indent(); + } else { + println(out, "|| !flatcc_builder_check_required_field(ctx->ctx, %"PRIu64")", member->id - !!is_union); + } + first = 0; + } + } + if (!first) { + unindent(); println(out, ") {"); indent(); + println(out, "buf = flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_required);"); + println(out, "goto failed;"); + unindent(); println(out, "}"); + } + if (trie.union_total) { + println(out, "buf = flatcc_json_parser_finalize_unions(ctx, buf, end, h_unions);"); + } + println(out, "if (!(*result = flatcc_builder_end_table(ctx->ctx))) goto failed;"); + println(out, "return buf;"); + /* Set runtime error if no other error was set already. */ + margin(); + println(out, "failed:"); + unmargin(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_runtime);"); + unindent(); println(out, "}"); + println(out, ""); + println(out, "static inline int %s_parse_json_as_root(flatcc_builder_t *B, flatcc_json_parser_t *ctx, const char *buf, size_t bufsiz, flatcc_json_parser_flags_t flags, const char *fid)", snt.text); + println(out, "{"); indent(); + println(out, "return flatcc_json_parser_table_as_root(B, ctx, buf, bufsiz, flags, fid, %s_parse_json_table);", + snt.text); + unindent(); println(out, "}"); + println(out, ""); + clear_dict(trie.dict); + return 0; +} + +static int gen_union_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt, snref; + fb_symbol_t *sym; + fb_member_t *member; + int n; + const char *s; + + fb_clear(snt); + fb_clear(snref); + fb_compound_name(ct, &snt); + println(out, "static const char *%s_parse_json_union(flatcc_json_parser_t *ctx, const char *buf, const char *end, uint8_t type, flatcc_builder_ref_t *result)", snt.text); + println(out, "{"); indent(); + println(out, ""); + println(out, "*result = 0;"); + println(out, "switch (type) {"); + println(out, "case 0: /* NONE */"); indent(); + println(out, "return flatcc_json_parser_none(ctx, buf, end);"); + unindent(); + for (sym = ct->members; sym; sym = sym->link) { + member = (fb_member_t *)sym; + symbol_name(sym, &n, &s); + switch (member->type.type) { + case vt_missing: + /* NONE is of type vt_missing and already handled. */ + continue; + case vt_compound_type_ref: + fb_compound_name(member->type.ct, &snref); + println(out, "case %u: /* %.*s */", (unsigned)member->value.u, n, s); indent(); + switch (member->type.ct->symbol.kind) { + case fb_is_table: + println(out, "buf = %s_parse_json_table(ctx, buf, end, result);", snref.text); + break; + case fb_is_struct: + println(out, "buf = %s_parse_json_struct(ctx, buf, end, result);", snref.text); + break; + default: + gen_panic(out, "internal error: unexpected compound union member type\n"); + return -1; + } + println(out, "break;"); + unindent(); + continue; + case vt_string_type: + println(out, "case %u: /* %.*s */", (unsigned)member->value.u, n, s); indent(); + println(out, "buf = flatcc_json_parser_build_string(ctx, buf, end, result);"); + println(out, "break;"); + unindent(); + continue; + default: + gen_panic(out, "internal error: unexpected union member type\n"); + return -1; + } + } + /* Unknown union, but not an error if we allow schema forwarding. */ + println(out, "default:"); indent(); + println(out, "if (!(ctx->flags & flatcc_json_parser_f_skip_unknown)) {"); indent(); + println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_unknown_union);"); + unindent(); println(out, "} else {"); indent(); + println(out, "return flatcc_json_parser_generic_json(ctx, buf, end);"); + unindent(); println(out, "}"); + unindent(); println(out, "}"); + println(out, "if (ctx->error) return buf;"); + println(out, "if (!*result) {"); + indent(); println(out, "return flatcc_json_parser_set_error(ctx, buf, end, flatcc_json_parser_error_runtime);"); + unindent(); println(out, "}"); + println(out, "return buf;"); + unindent(); println(out, "}"); + println(out, ""); + return 0; +} + +static int gen_union_accept_type(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt, snref; + fb_symbol_t *sym; + fb_member_t *member; + int n; + const char *s; + + fb_clear(snt); + fb_clear(snref); + fb_compound_name(ct, &snt); + println(out, "static int %s_json_union_accept_type(uint8_t type)", snt.text); + println(out, "{"); indent(); + println(out, "switch (type) {"); + for (sym = ct->members; sym; sym = sym->link) { + member = (fb_member_t *)sym; + symbol_name(sym, &n, &s); + if (member->type.type == vt_missing) { + println(out, "case 0: return 1; /* NONE */"); + continue; + } + println(out, "case %u: return 1; /* %.*s */", (unsigned)member->value.u, n, s); + } + /* Unknown union, but not an error if we allow schema forwarding. */ + println(out, "default: return 0;"); indent(); + unindent(); println(out, "}"); + unindent(); println(out, "}"); + println(out, ""); + return 0; +} + +static void gen_local_scope_prototype(void *context, fb_scope_t *scope) +{ + fb_output_t *out = context; + fb_symbol_text_t scope_name; + + fb_copy_scope(scope, scope_name); + + println(out, "static const char *%s_local_%sjson_parser_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", + out->S->basename, scope_name); + println(out, "int *value_type, uint64_t *value, int *aggregate);"); +} + +static int gen_root_table_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + + fb_clear(snt); + fb_compound_name(ct, &snt); + + println(out, "static int %s_parse_json(flatcc_builder_t *B, flatcc_json_parser_t *ctx,", out->S->basename); + indent(); indent(); + println(out, "const char *buf, size_t bufsiz, flatcc_json_parser_flags_t flags)"); + unindent(); unindent(); + println(out, "{"); indent(); + println(out, "flatcc_json_parser_t parser;"); + println(out, "flatcc_builder_ref_t root;"); + println(out, ""); + println(out, "ctx = ctx ? ctx : &parser;"); + println(out, "flatcc_json_parser_init(ctx, B, buf, buf + bufsiz, flags);"); + if (out->S->file_identifier.type == vt_string) { + println(out, "if (flatcc_builder_start_buffer(B, \"%.*s\", 0, 0)) return -1;", + out->S->file_identifier.s.len, out->S->file_identifier.s.s); + } else { + println(out, "if (flatcc_builder_start_buffer(B, 0, 0, 0)) return -1;"); + } + println(out, "%s_parse_json_table(ctx, buf, buf + bufsiz, &root);", snt.text); + println(out, "if (ctx->error) {"); indent(); + println(out, "return ctx->error;"); + unindent(); println(out, "}"); + println(out, "if (!flatcc_builder_end_buffer(B, root)) return -1;"); + println(out, "ctx->end_loc = buf;"); + println(out, "return 0;"); + unindent(); println(out, "}"); + println(out, ""); + return 0; +} + +static int gen_root_struct_parser(fb_output_t *out, fb_compound_type_t *ct) +{ + fb_scoped_name_t snt; + + fb_clear(snt); + fb_compound_name(ct, &snt); + + println(out, "static int %s_parse_json(flatcc_builder_t *B, flatcc_json_parser_t *ctx,", out->S->basename); + indent(); indent(); + println(out, "const char *buf, size_t bufsiz, int flags)"); + unindent(); unindent(); + println(out, "{"); indent(); + println(out, "flatcc_json_parser_t ctx_;"); + println(out, "flatcc_builder_ref_t root;"); + println(out, ""); + println(out, "ctx = ctx ? ctx : &ctx_;"); + println(out, "flatcc_json_parser_init(ctx, B, buf, buf + bufsiz, flags);"); + if (out->S->file_identifier.type == vt_string) { + println(out, "if (flatcc_builder_start_buffer(B, \"%.*s\", 0, 0)) return -1;", + out->S->file_identifier.s.len, out->S->file_identifier.s.s); + } else { + println(out, "if (flatcc_builder_start_buffer(B, 0, 0, 0)) return -1;"); + } + println(out, "buf = %s_parse_json_struct(ctx, buf, buf + bufsiz, &root);", snt.text); + println(out, "if (ctx->error) {"); indent(); + println(out, "return ctx->error;"); + unindent(); println(out, "}"); + println(out, "if (!flatcc_builder_end_buffer(B, root)) return -1;"); + println(out, "ctx->end_loc = buf;"); + println(out, "return 0;"); + unindent(); println(out, "}"); + println(out, ""); + return 0; +} + + +static int gen_root_parser(fb_output_t *out) +{ + fb_symbol_t *root_type = out->S->root_type.type; + + if (!root_type) { + return 0; + } + if (root_type) { + switch (root_type->kind) { + case fb_is_table: + return gen_root_table_parser(out, (fb_compound_type_t *)root_type); + case fb_is_struct: + return gen_root_struct_parser(out, (fb_compound_type_t *)root_type); + default: + break; + } + } + return 0; +} + +static int gen_json_parser_prototypes(fb_output_t *out) +{ + fb_symbol_t *sym; + fb_scoped_name_t snt; + fb_symbol_t *root_type = out->S->root_type.type; + + fb_clear(snt); + + if (root_type) + switch (root_type->kind) { + case fb_is_table: + case fb_is_struct: + println(out, "/*"); + println(out, " * Parses the default root table or struct of the schema and constructs a FlatBuffer."); + println(out, " *"); + println(out, " * Builder `B` must be initialized. `ctx` can be null but will hold"); + println(out, " * hold detailed error info on return when available."); + println(out, " * Returns 0 on success, or error code."); + println(out, " * `flags` : 0 by default, `flatcc_json_parser_f_skip_unknown` silently"); + println(out, " * ignores unknown table and structs fields, and union types."); + println(out, " */"); + println(out, "static int %s_parse_json(flatcc_builder_t *B, flatcc_json_parser_t *ctx,", + out->S->basename); + indent(); indent(); + println(out, "const char *buf, size_t bufsiz, flatcc_json_parser_flags_t flags);"); + unindent(); unindent(); + println(out, ""); + break; + default: + break; + } + for (sym = out->S->symbols; sym; sym = sym->link) { + switch (sym->kind) { + case fb_is_union: + fb_compound_name((fb_compound_type_t *)sym, &snt); + println(out, "static const char *%s_parse_json_union(flatcc_json_parser_t *ctx, const char *buf, const char *end, uint8_t type, flatcc_builder_ref_t *pref);", snt.text); + println(out, "static int %s_json_union_accept_type(uint8_t type);", snt.text); + /* A union also has an enum parser to get the type. */ + println(out, "static const char *%s_parse_json_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", snt.text); + indent(); indent(); + println(out, "int *value_type, uint64_t *value, int *aggregate);"); + unindent(); unindent(); + break; + case fb_is_struct: + fb_compound_name((fb_compound_type_t *)sym, &snt); + println(out, "static const char *%s_parse_json_struct_inline(flatcc_json_parser_t *ctx, const char *buf, const char *end, void *struct_base);", snt.text); + println(out, "static const char *%s_parse_json_struct(flatcc_json_parser_t *ctx, const char *buf, const char *end, flatcc_builder_ref_t *result);", snt.text); + break; + case fb_is_table: + fb_compound_name((fb_compound_type_t *)sym, &snt); + println(out, "static const char *%s_parse_json_table(flatcc_json_parser_t *ctx, const char *buf, const char *end, flatcc_builder_ref_t *result);", snt.text); + break; + case fb_is_enum: + fb_compound_name((fb_compound_type_t *)sym, &snt); + println(out, "static const char *%s_parse_json_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", snt.text); + indent(); indent(); + println(out, "int *value_type, uint64_t *value, int *aggregate);", snt.text); + unindent(); unindent(); + break; + } + } + fb_scope_table_visit(&out->S->root_schema->scope_index, gen_local_scope_prototype, out); + println(out, "static const char *%s_global_json_parser_enum(flatcc_json_parser_t *ctx, const char *buf, const char *end,", out->S->basename); + indent(); indent(); + println(out, "int *value_type, uint64_t *value, int *aggregate);"); + unindent(); unindent(); + println(out, ""); + return 0; +} + +static int gen_json_parsers(fb_output_t *out) +{ + fb_symbol_t *sym; + + for (sym = out->S->symbols; sym; sym = sym->link) { + switch (sym->kind) { + case fb_is_union: + gen_union_parser(out, (fb_compound_type_t *)sym); + gen_union_accept_type(out, (fb_compound_type_t *)sym); + gen_enum_parser(out, (fb_compound_type_t *)sym); + break; + case fb_is_struct: + gen_struct_parser_inline(out, (fb_compound_type_t *)sym); + gen_struct_parser(out, (fb_compound_type_t *)sym); + break; + case fb_is_table: + gen_table_parser(out, (fb_compound_type_t *)sym); + break; + case fb_is_enum: + gen_enum_parser(out, (fb_compound_type_t *)sym); + break; + } + } + fb_scope_table_visit(&out->S->root_schema->scope_index, gen_local_scope_parser, out); + gen_global_scope_parser(out); + gen_root_parser(out); + return 0; +} + +int fb_gen_c_json_parser(fb_output_t *out) +{ + gen_json_parser_pretext(out); + gen_json_parser_prototypes(out); + gen_json_parsers(out); + gen_json_parser_footer(out); + return 0; +} |