JSON

To read, validate and write JSON data interchange material.

§1. Introduction. JSON (Douglas Crockford, c. 2000) stands for "JavaScript Object Notation", but is now a standardised data interchange format used in many contexts. It's especially suitable for passing small amounts of data over the Internet, or between programs, or for making small data files human-readable. It's not good for larger data sets: it is really designed for messages, metadata or preference files.

This section provides encoding and decoding facilities. It is intended to comply with ECMA-404, except that (i) it disallows repetition the same key in the same object, and (ii) text can only be used in the Basic Multilingual Plane of Unicode points 0x0000 to 0xffff.

There are no size maxima or limitations. Still, this code was written at typing speed, and no effort has gone into reducing memory usage or running time in the face of large (or malicious) JSON content. Error reporting is also limited in fulsomeness.

See the foundation-test test case json for many exercises of the code below; do not change this section without checking that it continues to pass.

§2. Data model. JSON has a simple data model which we need to replicate in memory. Each value will be a pointer to a (permanently held in memory) JSON_value object. This is in effect a union, in that its type is always one of the following, and then only certain elements are meaningful depending on type.

These are exactly the JSON types except that numbers are split between integer and floating-point versions (the conflation of the two is where the Javascript origins of JSON show through), and that the type ERROR_JSONTYPE represents invalid data resulting from attempting to decode erroneous JSON.

enum NUMBER_JSONTYPE from 1
enum DOUBLE_JSONTYPE
enum STRING_JSONTYPE
enum BOOLEAN_JSONTYPE
enum ARRAY_JSONTYPE
enum OBJECT_JSONTYPE
enum NULL_JSONTYPE
enum ERROR_JSONTYPE

void JSON::write_type(OUTPUT_STREAM, int t) {
    switch (t) {
        case NUMBER_JSONTYPE:  WRITE("number"); break;
        case DOUBLE_JSONTYPE:  WRITE("double"); break;
        case STRING_JSONTYPE:  WRITE("string"); break;
        case BOOLEAN_JSONTYPE: WRITE("boolean"); break;
        case ARRAY_JSONTYPE:   WRITE("array"); break;
        case OBJECT_JSONTYPE:  WRITE("object"); break;
        case NULL_JSONTYPE:    WRITE("null"); break;
        case ERROR_JSONTYPE:   WRITE("<error>"); break;
        default:               WRITE("<invalid %d>", t); break;
    }
}

§3.

typedef struct JSON_value {
    int JSON_type;
    int if_integer;
    double if_double;
    struct text_stream *if_string;
    int if_boolean;
    struct linked_list *if_list;  of JSON_value
    struct dictionary *dictionary_if_object;  to JSON_value
    struct linked_list *list_if_object;  of text_stream
    struct text_stream *if_error;
    CLASS_DEFINITION
} JSON_value;

The structure JSON_value is private to this section.

§4. Now some constructor functions to create data of each JSON type:

JSON_value *JSON::new_null(void) {
    JSON_value *value = CREATE(JSON_value);
    value->JSON_type = NULL_JSONTYPE;
    value->if_integer = 0;
    value->if_double = 0;
    value->if_string = NULL;
    value->if_boolean = NOT_APPLICABLE;
    value->if_list = NULL;
    value->dictionary_if_object = NULL;
    value->list_if_object = NULL;
    value->if_error = NULL;
    return value;
}

JSON_value *JSON::new_boolean(int b) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = BOOLEAN_JSONTYPE;
    value->if_boolean = b;
    if ((b != TRUE) && (b != FALSE)) internal_error("improper JSON boolean");
    return value;
}

JSON_value *JSON::new_number(int b) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = NUMBER_JSONTYPE;
    value->if_integer = b;
    return value;
}

JSON_value *JSON::new_double(double d) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = DOUBLE_JSONTYPE;
    value->if_double = d;
    return value;
}

JSON_value *JSON::new_string(text_stream *S) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = STRING_JSONTYPE;
    value->if_string = Str::duplicate(S);
    return value;
}

§5. JSON arrays — lists, in effect — should be created in an empty state, and then have entries added sequentially:

JSON_value *JSON::new_array(void) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = ARRAY_JSONTYPE;
    value->if_list = NEW_LINKED_LIST(JSON_value);
    return value;
}

JSON_value *JSON::add_to_array(JSON_value *array, JSON_value *new_entry) {
    if (array == NULL) internal_error("no array");
    if (array->JSON_type == ERROR_JSONTYPE) return array;
    if (array->JSON_type != ARRAY_JSONTYPE) internal_error("not an array");
    if (new_entry == NULL) internal_error("no new entry");
    if (new_entry->JSON_type == ERROR_JSONTYPE) return new_entry;
    ADD_TO_LINKED_LIST(new_entry, JSON_value, array->if_list);
    return array;
}

§6. Similarly, JSON objects — dictionaries of key-value pairs, in effect — should be created in an empty state, and then have key-value pairs added as needed:

JSON_value *JSON::new_object(void) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = OBJECT_JSONTYPE;
    value->dictionary_if_object = Dictionaries::new(16, FALSE);
    value->list_if_object = NEW_LINKED_LIST(text_stream);
    return value;
}

JSON_value *JSON::add_to_object(JSON_value *obj, text_stream *key, JSON_value *value) {
    if (obj == NULL) internal_error("no object");
    if (obj->JSON_type == ERROR_JSONTYPE) return obj;
    if (obj->JSON_type != OBJECT_JSONTYPE) internal_error("not an object");
    if (value == NULL) internal_error("no new entry");
    if (value->JSON_type == ERROR_JSONTYPE) return value;
    key = Str::duplicate(key);
    ADD_TO_LINKED_LIST(key, text_stream, obj->list_if_object);
    dict_entry *de = Dictionaries::create(obj->dictionary_if_object, key);
    if (de) de->value = value;
    return obj;
}

§7. The following looks up a key in an object, returning NULL if and only if it is not present:

JSON_value *JSON::look_up_object(JSON_value *obj, text_stream *key) {
    if (obj == NULL) internal_error("no object");
    if (obj->JSON_type == ERROR_JSONTYPE) return NULL;
    if (obj->JSON_type != OBJECT_JSONTYPE) internal_error("not an object");
    dict_entry *de = Dictionaries::find(obj->dictionary_if_object, key);
    if (de == NULL) return NULL;
    return de->value;
}

§8. The following changes the value for a key in an object, creating if necessary:

void JSON::change_object(JSON_value *obj, text_stream *key, JSON_value *val) {
    if (obj == NULL) internal_error("no object");
    if (obj->JSON_type == ERROR_JSONTYPE) internal_error("erroneous object");
    if (obj->JSON_type != OBJECT_JSONTYPE) internal_error("not an object");
    dict_entry *de = Dictionaries::find(obj->dictionary_if_object, key);
    if (de == NULL) JSON::add_to_object(obj, key, val);
    else de->value = val;
}

§9. One last constructor creates an invalid JSON value resulting from incorrect JSON input:

JSON_value *JSON::error(text_stream *msg) {
    JSON_value *value = JSON::new_null();
    value->JSON_type = ERROR_JSONTYPE;
    value->if_error = Str::duplicate(msg);
    return value;
}

§10. This is a very limited form of comparison, since it cannot test equality of arrays or objects.

int JSON::eq(JSON_value *val1, JSON_value *val2) {
    if ((val1 == NULL) && (val2)) return FALSE;
    if ((val1) && (val2 == NULL)) return FALSE;
    if (val1 == NULL) return TRUE;
    if (val1->JSON_type != val2->JSON_type) return FALSE;
    switch (val1->JSON_type) {
        case NUMBER_JSONTYPE:  if (val1->if_integer == val2->if_integer) return TRUE; break;
        case STRING_JSONTYPE:  if (Str::eq(val1->if_string, val2->if_string)) return TRUE; break;
        case BOOLEAN_JSONTYPE: if (val1->if_boolean == val2->if_boolean) return TRUE; break;
        case NULL_JSONTYPE:    return TRUE;
    }
    return FALSE;
}

§11. Decoding JSON. We do no actual file-handling in this section, but the following decoder can be pointed to the contents of UTF-8 text file as needed.

The decoder returns a non-NULL pointer in all cases. If the text contains any malformed JSON anywhere inside it, this pointer will be to a value of type ERROR_JSONTYPE. Such a value should be thrown away as soon as the error message is made use of.

JSON_value *JSON::decode(text_stream *T, text_file_position *tfp) {
    return JSON::decode_range(T, 0, Str::len(T), tfp);
}

JSON_value *JSON::decode_error(text_stream *err, text_file_position *tfp) {
    TEMPORARY_TEXT(msg)
    if (tfp) WRITE_TO(msg, "%f: ", tfp->text_file_filename);
    WRITE_TO(msg, "%S", err);
    JSON_value *value = JSON::error(msg);
    DISCARD_TEXT(msg)
    return value;
}

JSON_value *JSON::decode_error_q(text_stream *err, text_file_position *tfp,
    text_stream *T, int from, int to) {
    TEMPORARY_TEXT(msg)
    WRITE_TO(msg, "%S", err);
    if ((T) && (from < to)) {
        WRITE_TO(msg, ": '");
        for (int i=from; ((i<to) && (i-from < 40)); i++) {
            if (Str::get_at(T, i) == '\n') WRITE_TO(msg, " ");
            else WRITE_TO(msg, "%c", Str::get_at(T, i));
        }
        WRITE_TO(msg, "'");
    }
    JSON_value *value = JSON::decode_error(msg, tfp);
    DISCARD_TEXT(msg)
    return value;
}

§12. This decodes the text in the character position range [from, to) as a JSON value.

The possibilities here are [ ... ] for an array, { ... } for an object, "..." for a string, a token beginning with a digit or a minus sign for a number (note that + and . are not allowed to open a number according to the JSON standard), and the special cases true, false and null.

JSON_value *JSON::decode_range(text_stream *T, int from, int to, text_file_position *tfp) {
    int first_nws = -1, last_nws = -1;
    inchar32_t first_c = 0, last_c = 0;
    Find the first and last non-whitespace character12.1;
    switch (first_c) {
        case '[':
            if (last_c != ']') return JSON::decode_error(I"mismatched '[' ... ']'", tfp);
            JSON_value *array = JSON::new_array();
            return JSON::decode_array(array, T, first_nws+1, last_nws, tfp);
        case '{':
            if (last_c != '}') return JSON::decode_error(I"mismatched '{' ... '}'", tfp);
            JSON_value *obj = JSON::new_object();
            return JSON::decode_object(obj, T, first_nws+1, last_nws, tfp);
        case '"':
            if (last_c != '"') return JSON::decode_error(I"mismatched quotation marks", tfp);
            return JSON::decode_string(T, first_nws+1, last_nws, tfp);
    }
    if ((Characters::isdigit(first_c)) || (first_c == '-'))
        return JSON::decode_number(T, first_nws, last_nws+1, tfp);
    if ((Str::includes_at(T, first_nws, I"true")) && (last_nws - first_nws == 3))
        return JSON::new_boolean(TRUE);
    if ((Str::includes_at(T, first_nws, I"false")) && (last_nws - first_nws == 4))
        return JSON::new_boolean(FALSE);
    if ((Str::includes_at(T, first_nws, I"null")) && (last_nws - first_nws == 3))
        return JSON::new_null();
    return JSON::decode_error(I"unknown JSON value", tfp);
}

§12.1. Find the first and last non-whitespace character12.1 =

    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE) {
            first_nws = i; break;
        }
    for (int i=to-1; i>=from; i--)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE) {
            last_nws = i; break;
        }
    if (first_nws < 0) return JSON::decode_error(I"whitespace where JSON value expected", tfp);
    first_c = Str::get_at(T, first_nws);
    last_c = Str::get_at(T, last_nws);

This code is used in §12.

§13. So now we have individual decoder functions for each type. First, arrays, where now the range [from, to) represents what is inside the square brackets: this needs to be a comma-separated list. We follow ECMA strictly in disallowing a final comma before the ], unlike some JSON-like parsers.

JSON_value *JSON::decode_array(JSON_value *array, text_stream *T, int from, int to,
    text_file_position *tfp) {
    int content = FALSE;
    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE)
            content = TRUE;
    if (content == FALSE) return array;
    NextEntry: ;
    int first_comma = -1, bl = 0;
    for (int i=from, quoted = FALSE; i<to; i++) {
        inchar32_t c = Str::get_at(T, i);
        switch (c) {
            case '"': quoted = (quoted)?FALSE:TRUE; break;
            case '\\': if (quoted) i++; break;
            case ',': if ((quoted == FALSE) && ((first_comma < 0) && (bl == 0)))
                    first_comma = i;
                break;
            case '[': case '{': if (quoted == FALSE) bl++; break;
            case ']': case '}': if (quoted == FALSE) bl--; break;
        }
    }
    if (first_comma >= 0) {
        array = JSON::decode_array_entry(array, T, from, first_comma, tfp);
        from = first_comma + 1;
        goto NextEntry;
    }
    return JSON::decode_array_entry(array, T, from, to, tfp);
}

JSON_value *JSON::decode_array_entry(JSON_value *array, text_stream *T, int from, int to,
    text_file_position *tfp) {
    JSON_value *value = JSON::decode_range(T, from, to, tfp);
    return JSON::add_to_array(array, value);
}

§14. And similarly for objects.

JSON_value *JSON::decode_object(JSON_value *obj, text_stream *T, int from, int to,
    text_file_position *tfp) {
    int content = FALSE;
    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE)
            content = TRUE;
    if (content == FALSE) return obj;
    NextEntry: ;
    int first_comma = -1, bl = 0;
    for (int i=from, quoted = FALSE; i<to; i++) {
        inchar32_t c = Str::get_at(T, i);
        switch (c) {
            case '"': quoted = (quoted)?FALSE:TRUE; break;
            case '\\': if (quoted) i++; break;
            case ',': if ((quoted == FALSE) && ((first_comma < 0) && (bl == 0)))
                    first_comma = i;
                break;
            case '[': case '{': if (quoted == FALSE) bl++; break;
            case ']': case '}': if (quoted == FALSE) bl--; break;
        }
    }
    if (first_comma >= 0) {
        obj = JSON::decode_object_entry(obj, T, from, first_comma, tfp);
        from = first_comma + 1;
        goto NextEntry;
    }
    return JSON::decode_object_entry(obj, T, from, to, tfp);
}

§15. Note that we allow key names to include all kinds of unconscionable garbage, as ECMA requires. \u0003\"\t\t\t is a valid JSON key name; so is the empty string.

We are however slightly stricter than ECMA in that we disallow duplicate keys in the same object. ECMA says this is a "semantic consideration that may be defined by JSON processors". We are hereby defining it.

JSON_value *JSON::decode_object_entry(JSON_value *obj, text_stream *T, int from, int to,
    text_file_position *tfp) {
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    int saved_from = from, saved_to = to;
    if (from >= to)
        return JSON::decode_error(I"object body ends with comma", tfp);
    if (Str::get_at(T, from) != '"')
        return JSON::decode_error_q(I"key does not begin with quotation mark", tfp, T, saved_from, saved_to);
    from++;
    int ended = FALSE;
    TEMPORARY_TEXT(key)
    while (from < to) {
        inchar32_t c = Str::get_at(T, from++);
        if (c == '\"') { ended = TRUE; break; }
        PUT_TO(key, c);
        if ((c == '\\') && (from+1 < to)) {
            c = Str::get_at(T, from++);
            PUT_TO(key, c);
        }
    }
    if (ended == FALSE) return JSON::decode_error_q(I"key does not end with quotation mark", tfp, T, saved_from, saved_to);
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    if ((from >= to) || (Str::get_at(T, from) != ':'))
        return JSON::decode_error_q(I"key is not followed by ':'", tfp, T, saved_from, saved_to);
    from++;
    if (JSON::look_up_object(obj, key)) return JSON::decode_error_q(I"duplicate key", tfp, T, saved_from, saved_to);
    JSON_value *value = JSON::decode_range(T, from, to, tfp);
    obj = JSON::add_to_object(obj, key, value);
    DISCARD_TEXT(key)
    return obj;
}

§16. Numbers are annoying to decode since they can be given either in a restricted floating-point syntax, or in decimal. ECMA is slippery on the question of exactly what floating-point numbers can be represented, but it's common to consider them as being double, so we'll follow suit.

JSON_value *JSON::decode_number(text_stream *T, int from, int to, text_file_position *tfp) {
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    while ((to > from) && (Characters::is_whitespace(Str::get_at(T, to-1)))) to--;
    if (to <= from) return JSON::decode_error(I"whitespace where number expected", tfp);
    TEMPORARY_TEXT(integer)
    int at = from;
    if ((Str::get_at(T, at) == '-') && (to > at+1)) { PUT_TO(integer, '-'); at++; }
    int double_me = FALSE;
    for (int i=at; i<to; i++)
        if (Characters::isdigit(Str::get_at(T, i)))
            PUT_TO(integer, Str::get_at(T, i));
        else if ((Str::get_at(T, i) == 'E') || (Str::get_at(T, i) == 'e') ||
            (Str::get_at(T, i) == '.') || (Str::get_at(T, i) == '+'))
            double_me = TRUE;
        else
            return JSON::decode_error(I"number is not a decimal integer", tfp);
    JSON_value *value = NULL;
    if (double_me) {
        char double_buffer[32];
        for (int i=0; i<32; i++) double_buffer[i] = 0;
        for (int i=from; (i<to) && (i-from<31); i++)
            double_buffer[i-from] = (char) Str::get_at(T, i);
        double d = atof(double_buffer);
        if (isnan(d)) return JSON::decode_error(I"number is not allowed to be NaN", tfp);
        value = JSON::new_double(d);
    } else {
        int N = Str::atoi(integer, 0);
        value = JSON::new_number(N);
    }
    DISCARD_TEXT(integer)
    return value;
}

§17. Strings are easy except for escape characters. I have no idea why JSON wants to allow the escaping of forward slash, but the standard requires it.

JSON_value *JSON::decode_string(text_stream *T, int from, int to, text_file_position *tfp) {
    TEMPORARY_TEXT(string)
    for (int i=from; i<to; i++) {
        inchar32_t c = Str::get_at(T, i);
        if (c == '\\') {
            i++;
            c = Str::get_at(T, i);
            if (Characters::iscntrl(c)) return JSON::decode_error(I"unescaped control character", tfp);
            switch (c) {
                case 'b': c = 8; break;
                case 't': c = 9; break;
                case 'n': c = 10; break;
                case 'f': c = 12; break;
                case 'r': c = 13; break;
                case '\\': break;
                case '/': break;
                case '"': break;
                case 'u': Decode a hexadecimal Unicode escape17.1; break;
                default: return JSON::decode_error(I"bad '\\' escape in string", tfp);
            }
            PUT_TO(string, c);
        } else {
            PUT_TO(string, c);
        }
    }
    JSON_value *val = JSON::new_string(string);
    DISCARD_TEXT(string)
    return val;
}

§17.1. We don't quite fully implement ECMA here: the following is fine for code points in the Basic Multilingual Plane, but we don't handle the curious UTF-16 surrogate pair rule for code points between 0x10000 and 0x10fff.

Decode a hexadecimal Unicode escape17.1 =

    if (i+4 >= to) return JSON::decode_error(I"incomplete '\\u' escape", tfp);
    int hex = 0;
    for (int j=0; j<4; j++) {
        int v = 0;
        inchar32_t digit = Str::get_at(T, i+1+j);
        if ((digit >= '0') && (digit <= '9')) v = (int) (digit-'0');
        else if ((digit >= 'a') && (digit <= 'f')) v = 10 + ((int) (digit-'a'));
        else if ((digit >= 'A') && (digit <= 'F')) v = 10 + ((int) (digit-'A'));
        else return JSON::decode_error(I"garbled '\\u' escape", tfp);
        hex = hex * 16 + v;
    }
    c = (inchar32_t) hex;
    i += 4;

This code is used in §17.

§18. Encoding JSON.

void JSON::encode(OUTPUT_STREAM, JSON_value *J) {
    if (J == NULL) internal_error("no JSON value supplied");
    switch (J->JSON_type) {
        case ERROR_JSONTYPE:
            internal_error("tried to encode erroneous JSON");
        case NUMBER_JSONTYPE:
            WRITE("%d", J->if_integer);
            break;
        case DOUBLE_JSONTYPE:
            WRITE("%g", J->if_double);
            break;
        case STRING_JSONTYPE:
            WRITE("\""); JSON::encode_string(OUT, J->if_string); WRITE("\"");
            break;
        case BOOLEAN_JSONTYPE:
            if (J->if_boolean == TRUE) WRITE("true");
            else if (J->if_boolean == FALSE) WRITE("false");
            else internal_error("improper boolean JSON value");
            break;
        case ARRAY_JSONTYPE: {
            WRITE("[");
            int count = 0;
            JSON_value *E;
            LOOP_OVER_LINKED_LIST(E, JSON_value, J->if_list) {
                if (count++ > 0) WRITE(",");
                WRITE(" ");
                JSON::encode(OUT, E);
            }
            if (count > 0) WRITE(" ");
            WRITE("]");
            break;
        }
        case OBJECT_JSONTYPE: {
            WRITE("{\n"); INDENT;
            int count = 0;
            text_stream *key;
            LOOP_OVER_LINKED_LIST(key, text_stream, J->list_if_object) {
                if (count++ > 0) WRITE(",\n");
                JSON_value *E = Dictionaries::read_value(J->dictionary_if_object, key);
                if (E == NULL) internal_error("broken JSON object dictionary");
                WRITE("\"");
                JSON::encode_string(OUT, key);
                WRITE("\": ");
                JSON::encode(OUT, E);
            }
            if (count > 0) WRITE("\n");
            OUTDENT; WRITE("}");
            break;
        }
        case NULL_JSONTYPE:
            WRITE("null");
            break;
        default: internal_error("unsupported JSON value type");
    }
}

§19. Note that we elect not to escape the slash character, or any Unicode code points above 32.

void JSON::encode_string(OUTPUT_STREAM, text_stream *T) {
    LOOP_THROUGH_TEXT(pos, T) {
        inchar32_t c = Str::get(pos);
        switch (c) {
            case '\\': WRITE("\\\\"); break;
            case 8: WRITE("\\b"); break;
            case 9: WRITE("\\t"); break;
            case 10: WRITE("\\n"); break;
            case 12: WRITE("\\f"); break;
            case 13: WRITE("\\r"); break;
            default:
                if (Characters::iscntrl(c)) WRITE("\\u%04x", (int)c);
                else PUT(c);
                break;
        }
    }
}

§20. Requirements. Of course, the trouble with JSON is that it's a soup of undifferentiated data. Just because you're expecting a pair of numbers, there's no reason to suppose that's what you've been given.

A JSON_requirement is a sort of JSON schema: a specification for the structure of a JSON_value. At the top level, it's a list of one or more equally good alternative specifications. Note that the empty list is not allowed.

typedef struct JSON_requirement {
    struct linked_list *alternatives;  of JSON_single_requirement
    CLASS_DEFINITION
} JSON_requirement;

JSON_requirement *JSON::single_choice(JSON_single_requirement *sing) {
    JSON_requirement *req = CREATE(JSON_requirement);
    req->alternatives = NEW_LINKED_LIST(JSON_single_requirement);
    ADD_TO_LINKED_LIST(sing, JSON_single_requirement, req->alternatives);
    return req;
}

JSON_requirement *JSON::add_alternative(JSON_requirement *so_far,
    JSON_single_requirement *sing) {
    if (so_far == NULL) return JSON::single_choice(sing);
    ADD_TO_LINKED_LIST(sing, JSON_single_requirement, so_far->alternatives);
    return so_far;
}

The structure JSON_requirement is private to this section.

§21. A "single requirement" is a little more than what a type would be in Javascript, if Javascript actually had types. It can communicate something like "a number" or "a list of strings"; but it can also say "the value has to be exactly this".

typedef struct JSON_single_requirement {
    struct JSON_requirement *this_requirement;
    struct JSON_value *this_value;
    struct JSON_type *this_type;
    CLASS_DEFINITION
} JSON_single_requirement;

The structure JSON_single_requirement is private to this section.

§22. Exactly one of this_requirement, this_value and this_type should be non-NULL, so we have one constructor function for each case:

JSON_single_requirement *JSON::require_requirement(JSON_requirement *req) {
    JSON_single_requirement *sing = CREATE(JSON_single_requirement);
    sing->this_requirement = req;
    sing->this_value = NULL;
    sing->this_type = NULL;
    return sing;
}

JSON_single_requirement *JSON::require_value(JSON_value *value) {
    JSON_single_requirement *sing = CREATE(JSON_single_requirement);
    sing->this_requirement = NULL;
    sing->this_value = value;
    sing->this_type = NULL;
    return sing;
}

JSON_single_requirement *JSON::require_type(int t) {
    JSON_single_requirement *sing = CREATE(JSON_single_requirement);
    sing->this_requirement = NULL;
    sing->this_value = NULL;
    sing->this_type = JSON::new_type_requirement(t);
    return sing;
}

§23. JSON types, in our model, look very like //JSON_value//s.

typedef struct JSON_type {
    int JSON_type;

    struct linked_list *if_list;  of JSON_requirement
    struct JSON_requirement *all_if_list;

    struct dictionary *dictionary_if_object;  to JSON_pair_requirement
    struct linked_list *list_if_object;  of text_stream

    struct text_stream *if_error;
    CLASS_DEFINITION
} JSON_type;

typedef struct JSON_pair_requirement {
    struct JSON_requirement *req;
    int optional;
    CLASS_DEFINITION
} JSON_pair_requirement;

JSON_type *JSON::new_type_requirement(int t) {
    JSON_type *type = CREATE(JSON_type);
    type->JSON_type = t;

    if (t == ARRAY_JSONTYPE) {
        type->if_list = NEW_LINKED_LIST(JSON_requirement);
        type->all_if_list = NULL;
    } else {
        type->if_list = NULL;
        type->all_if_list = NULL;
    }

    if (t == OBJECT_JSONTYPE) {
        type->dictionary_if_object = Dictionaries::new(16, FALSE);
        type->list_if_object = NEW_LINKED_LIST(text_stream);
    } else {
        type->dictionary_if_object = NULL;
        type->list_if_object = NULL;
    }

    type->if_error = NULL;
    return type;
}

The structure JSON_type is private to this section.
The structure JSON_pair_requirement is accessed in 4/prp and here.

§24. A convenience for "the value must be an array of any number of entries, each of which meets the requirement E_req":

JSON_single_requirement *JSON::require_array_of(JSON_requirement *E_req) {
    JSON_single_requirement *req = JSON::require_type(ARRAY_JSONTYPE);
    req->this_type->all_if_list = E_req;
    return req;
}

§25. If an array wants to be a tuple with a fixed number of entries, each with its own requirement, then instead call JSON::require_type(ARRAY_JSONTYPE) and then make a number of calls to the following in sequence:

void JSON::require_entry(JSON_single_requirement *array_sr, JSON_requirement *entry_sr) {
    if (array_sr == NULL) internal_error("no array");
    if ((array_sr->this_type == NULL) ||
        (array_sr->this_type->JSON_type != ARRAY_JSONTYPE)) internal_error("not an array");
    if (entry_sr == NULL) internal_error("no new entry");
    ADD_TO_LINKED_LIST(entry_sr, JSON_requirement, array_sr->this_type->if_list);
}

§26. Similarly, create an object requirement with JSON::require_type(OBJECT_JSONTYPE) and then either require or allow key-value pairs with:

void JSON::require_pair(JSON_single_requirement *obj_sr, text_stream *key, JSON_requirement *req) {
    JSON::require_pair_inner(obj_sr, key, req, FALSE);
}

void JSON::allow_pair(JSON_single_requirement *obj_sr, text_stream *key, JSON_requirement *req) {
    JSON::require_pair_inner(obj_sr, key, req, TRUE);
}

void JSON::require_pair_inner(JSON_single_requirement *obj_sr, text_stream *key,
    JSON_requirement *req, int opt) {
    if (obj_sr == NULL) internal_error("no object");
    if ((obj_sr->this_type == NULL) ||
        (obj_sr->this_type->JSON_type != OBJECT_JSONTYPE)) internal_error("not an object");
    if (req == NULL) internal_error("no val req");
    key = Str::duplicate(key);
    ADD_TO_LINKED_LIST(key, text_stream, obj_sr->this_type->list_if_object);
    JSON_pair_requirement *pr = CREATE(JSON_pair_requirement);
    pr->req = req;
    pr->optional = opt;
    dict_entry *de = Dictionaries::create(obj_sr->this_type->dictionary_if_object, key);
    if (de) de->value = pr;
}

§27. This then extracts the requirement on a given key, or returns NULL is if is not permitted:

JSON_pair_requirement *JSON::look_up_pair(JSON_single_requirement *obj_sr, text_stream *key) {
    if (obj_sr == NULL) internal_error("no object");
    if ((obj_sr->this_type == NULL) ||
        (obj_sr->this_type->JSON_type != OBJECT_JSONTYPE)) internal_error("not an object");
    dict_entry *de = Dictionaries::find(obj_sr->this_type->dictionary_if_object, key);
    if (de == NULL) return NULL;
    return de->value;
}

§28. This is used when parsing textual requirements, to indicate a syntax error; but it is not valid as a requirement itself.

JSON_single_requirement *JSON::error_sr(text_stream *msg) {
    JSON_single_requirement *req = JSON::require_type(ERROR_JSONTYPE);
    req->this_type->if_error = Str::duplicate(msg);
    return req;
}

§29. Validation. To "validate" a JSON value is to determine that it meets some JSON_requirement.

The following returns TRUE if the value meets the requirement in full; if not, FALSE, and then if errs is not null, a list of error messages is appended to the linked list errs.

The stack here is used to give better error messages by locating where the problem was: e.g. "object.coordinates[1]" is the result of the stack holding "object" > ".cooordinates" > "[1]".

int JSON::validate(JSON_value *val, JSON_requirement *req, linked_list *errs) {
    lifo_stack *location = NEW_LIFO_STACK(text_stream);
    if ((val) && (val->JSON_type == ARRAY_JSONTYPE)) {
        PUSH_TO_LIFO_STACK(I"array", text_stream, location);
    }
    if ((val) && (val->JSON_type == OBJECT_JSONTYPE)) {
        PUSH_TO_LIFO_STACK(I"object", text_stream, location);
    }
    return JSON::validate_r(val, req, errs, location);
}

void JSON::validation_error(linked_list *errs, text_stream *err, lifo_stack *location) {
    if (errs) {
        text_stream *msg = Str::new();
        int S = LinkedLists::len(location);
        for (int i=S-1; i>=0; i--) {
            int c = 0;
            text_stream *seg;
            LOOP_OVER_LINKED_LIST(seg, text_stream, location)
                if (c++ == i)
                    WRITE_TO(msg, "%S", seg);
        }
        if (Str::len(msg) > 0) WRITE_TO(msg, ": ");
        WRITE_TO(msg, "%S", err);
        ADD_TO_LINKED_LIST(msg, text_stream, errs);
    }
}

§30. So this is the recursive verification function. At the top level, it says the value must match one of the single requirements in the list. (We can stop as soon as it has met one.) If it meets none of them, we produce error messages for the reason it fails just the first.

int JSON::validate_r(JSON_value *val, JSON_requirement *req, linked_list *errs,
    lifo_stack *location) {
    if (val == NULL) internal_error("no value");
    if (req == NULL) internal_error("no req");
    JSON_single_requirement *sing;
    LOOP_OVER_LINKED_LIST(sing, JSON_single_requirement, req->alternatives) {
        int rv = JSON::validate_single_r(val, sing, NULL, location);
        if (rv) return TRUE;
    }
    LOOP_OVER_LINKED_LIST(sing, JSON_single_requirement, req->alternatives) {
        JSON::validate_single_r(val, sing, errs, location);
        break;
    }
    return FALSE;
}

§31. Bad data always fails, and otherwise we split into the three cases.

int JSON::validate_single_r(JSON_value *val, JSON_single_requirement *req,
    linked_list *errs, lifo_stack *location) {
    if (val->JSON_type == ERROR_JSONTYPE) {
        JSON::validation_error(errs,
            I"erroneous JSON value from parsing bad text", location);
        return FALSE;
    }
    if (req->this_requirement) Validate against this requirement31.1;
    if (req->this_value) Validate against this value31.2;
    if (req->this_type) Validate against this type31.3;
    internal_error("bad single requirement");
}

§31.1. Validate against this requirement31.1 =

    return JSON::validate_r(val, req->this_requirement, errs, location);

This code is used in §31.

§31.2. Validate against this value31.2 =

    if (JSON::eq(val, req->this_value) == FALSE) {
        TEMPORARY_TEXT(msg)
        WRITE_TO(msg, "value ");
        JSON::encode(msg, val);
        WRITE_TO(msg, " not one of those allowed");
        JSON::validation_error(errs, msg, location);
        DISCARD_TEXT(msg)
        return FALSE;
    }
    return TRUE;

This code is used in §31.

§31.3. Validate against this type31.3 =

    Verify that the JSON type constructors match31.3.1;
    int outcome = TRUE;
    if (val->JSON_type == ARRAY_JSONTYPE)
        Verify that the array entries meet requirements31.3.2;
    if (val->JSON_type == OBJECT_JSONTYPE)
        Verify that the object members meet requirements31.3.3;
    return outcome;

This code is used in §31.

§31.3.1. Verify that the JSON type constructors match31.3.1 =

    if (val->JSON_type != req->this_type->JSON_type) {
        if (errs) {
            TEMPORARY_TEXT(msg)
            WRITE_TO(msg, "expected ");
            JSON::write_type(msg, req->this_type->JSON_type);
            WRITE_TO(msg, " but found ");
            JSON::write_type(msg, val->JSON_type);
            JSON::validation_error(errs, msg, location);
            DISCARD_TEXT(msg)
        }
        return FALSE;
    }

This code is used in §31.3.

§31.3.2. Verify that the array entries meet requirements31.3.2 =

    int count = 0;
    JSON_value *E;
    LOOP_OVER_LINKED_LIST(E, JSON_value, val->if_list) {
        JSON_requirement *E_req = req->this_type->all_if_list;
        if (E_req == NULL) {
            JSON_requirement *A_req;
            int rcount = 0;
            LOOP_OVER_LINKED_LIST(A_req, JSON_requirement, req->this_type->if_list)
                if (rcount++ == count)
                    E_req = A_req;
        }
        TEMPORARY_TEXT(at)
        WRITE_TO(at, "[%d]", count);
        PUSH_TO_LIFO_STACK(at, text_stream, location);
        if (E_req == NULL) {
            JSON::validation_error(errs, I"unexpected array entry", location);
            outcome = FALSE;
        } else {
            if (JSON::validate_r(E, E_req, errs, location) == FALSE) outcome = FALSE;
        }
        POP_LIFO_STACK(text_stream, location);
        DISCARD_TEXT(at)
        count++;
    }

This code is used in §31.3.

§31.3.3. Verify that the object members meet requirements31.3.3 =

    text_stream *key;
    LOOP_OVER_LINKED_LIST(key, text_stream, val->list_if_object)
        Verify that the member with this key is allowed and contains the right data31.3.3.1;
    LOOP_OVER_LINKED_LIST(key, text_stream, req->this_type->list_if_object) {
        JSON_pair_requirement *pr =
            Dictionaries::read_value(req->this_type->dictionary_if_object, key);
        if (pr == NULL) internal_error("broken JSON object requirement");
        if (pr->optional == FALSE)
            Verify that the value object does provide this mandatory member31.3.3.2;
    }

This code is used in §31.3.

§31.3.3.1. Verify that the member with this key is allowed and contains the right data31.3.3.1 =

    JSON_value *E = Dictionaries::read_value(val->dictionary_if_object, key);
    if (E == NULL) internal_error("broken JSON object dictionary");
    JSON_pair_requirement *pr = JSON::look_up_pair(req, key);
    TEMPORARY_TEXT(at)
    WRITE_TO(at, ".%S", key);
    PUSH_TO_LIFO_STACK(at, text_stream, location);
    if (pr == NULL) {
        TEMPORARY_TEXT(msg)
        WRITE_TO(msg, "unexpected member '%S'", key);
        JSON::validation_error(errs, msg, location);
        DISCARD_TEXT(msg)
        outcome = FALSE;
    } else {
        if (JSON::validate_r(E, pr->req, errs, location) == FALSE) outcome = FALSE;
    }
    POP_LIFO_STACK(text_stream, location);
    DISCARD_TEXT(at)

This code is used in §31.3.3.

§31.3.3.2. Verify that the value object does provide this mandatory member31.3.3.2 =

    JSON_value *E = JSON::look_up_object(val, key);
    if (E == NULL) {
        TEMPORARY_TEXT(msg)
        WRITE_TO(msg, "member '%S' missing", key);
        JSON::validation_error(errs, msg, location);
        DISCARD_TEXT(msg)
        outcome = FALSE;
    }

This code is used in §31.3.3.

§32. Decoding JSON requirements. It's convenient to be able to read and write these requirements to textual form, exactly as we do with JSON itself, and here goes.

This is an example of the syntax we parse. It's JSON except that

(a) the type names number, double, string, boolean and null are used in place of their respective values;
(b) a question mark ? before the name of a key means that it is optional;
(c) if an array has one entry followed by an asterisk *, it means "any number of entries, each of which must match this";
(d) <name> refers to a requirement recorded in the known_names dictionary.

For example:

{
    "coordinates": [ double, double, string ],
    ?"jurisdiction": string,
    "journal": [ {
        "date": number,
        "entry": string
    }* ]
}

This function is essentially the same as JSON::decode, but returning a requirement rather than a value.

Note that known_names can be NULL to have it not recognise any such names; there's no need to create an empty dictionary if this feature is unwanted.

JSON_requirement *JSON::decode_req(text_stream *T, dictionary *known_names) {
    return JSON::decode_req_range(T, 0, Str::len(T), known_names);
}

§33. This decodes the text in the character position range [from, to) as a JSON requirement.

JSON_requirement *JSON::decode_req_range(text_stream *T, int from, int to,
    dictionary *known_names) {
    int first_nws = -1, last_nws = -1;
    inchar32_t first_c = 0, last_c = 0;
    Find the first and last non-whitespace character in requirement33.2;
    if (first_c == '(') {
        if (last_c != ')')
            return JSON::single_choice(JSON::error_sr(I"mismatched '(' ... ')'"));
        from = first_nws + 1;
        to = last_nws;
        JSON_requirement *req = NULL;
        NextEntry: ;
        int first_pipe = -1, bl = 0;
        for (int i=from, quoted = FALSE; i<to; i++) {
            inchar32_t c = Str::get_at(T, i);
            switch (c) {
                case '"': quoted = (quoted)?FALSE:TRUE; break;
                case '\\': if (quoted) i++; break;
                case '|': if ((first_pipe < 0) && (bl == 0)) first_pipe = i; break;
                case '[': case '{': case '(': if (quoted == FALSE) bl++; break;
                case ']': case '}': case ')': if (quoted == FALSE) bl--; break;
            }
        }
        if (first_pipe >= 0) {
            req = JSON::decode_req_alternative(req, T, from, first_pipe, known_names);
            from = first_pipe + 1;
            goto NextEntry;
        }
        return JSON::decode_req_alternative(req, T, from, to, known_names);
    }
    return JSON::single_choice(JSON::decode_sreq_range(T, from, to, known_names));
}

JSON_requirement *JSON::decode_req_alternative(JSON_requirement *req, text_stream *T,
    int from, int to, dictionary *known_names) {
    JSON_single_requirement *sing = JSON::decode_sreq_range(T, from, to, known_names);
    return JSON::add_alternative(req, sing);
}

§33.1. Note that the keyword null is ambiguous in the grammar for JSON requirements: does it mean "the value null", or does it mean "any value of the type null"? This makes no difference because the type null admits only the value null, but for what it's worth, we opt for the value.

JSON_single_requirement *JSON::decode_sreq_range(text_stream *T, int from, int to,
    dictionary *known_names) {
    int first_nws = -1, last_nws = -1;
    inchar32_t first_c = 0, last_c = 0;
    Find the first and last non-whitespace character in requirement33.2;
    if (first_nws < 0) return JSON::error_sr(I"whitespace where requirement expected");
    switch (first_c) {
        case '[':
            if (last_c != ']') return JSON::error_sr(I"mismatched '[' ... ']'");
            JSON_single_requirement *array_sr = JSON::require_type(ARRAY_JSONTYPE);
            return JSON::decode_req_array(array_sr, T, first_nws+1, last_nws, known_names);
        case '{':
            if (last_c != '}') return JSON::error_sr(I"mismatched '{' ... '}'");
            JSON_single_requirement *obj_sr = JSON::require_type(OBJECT_JSONTYPE);
            return JSON::decode_req_object(obj_sr, T, first_nws+1, last_nws, known_names);
        case '<':
            if (last_c != '>') return JSON::error_sr(I"mismatched '<' ... '>'");
            JSON_requirement *known = NULL;
            TEMPORARY_TEXT(name)
            for (int i = first_nws+1; i<last_nws; i++)
                PUT_TO(name, Str::get_at(T, i));
            if (known_names) {
                dict_entry *de = Dictionaries::find(known_names, name);
                if (de == NULL) return JSON::error_sr(I"unknown '<name>'");
                known = de->value;
            } else {
                return JSON::error_sr(I"'<' ... '>' not allowed");
            }
            DISCARD_TEXT(name)
            if (known) return JSON::require_requirement(known);
            return NULL;
    }

    int require_value = FALSE;

    if ((first_c == '"') || (first_c == '-') || (Characters::isdigit(first_c)))
        require_value = TRUE;
    if ((Str::includes_at(T, first_nws, I"true")) && (last_nws - first_nws == 3))
        require_value = TRUE;
    if ((Str::includes_at(T, first_nws, I"false")) && (last_nws - first_nws == 4))
        require_value = TRUE;
    if ((Str::includes_at(T, first_nws, I"null")) && (last_nws - first_nws == 3))
        require_value = TRUE;

    if (require_value) {
        JSON_value *value = JSON::decode_range(T, from, to, NULL);
        if (value->JSON_type == ERROR_JSONTYPE) {
            TEMPORARY_TEXT(err)
            WRITE_TO(err, "JSON value error: %S", value->if_error);
            JSON_single_requirement *sing = JSON::error_sr(err);
            DISCARD_TEXT(err)
            return sing;
        }
        return JSON::require_value(value);
    }

    if ((Str::includes_at(T, first_nws, I"number")) && (last_nws - first_nws == 5))
        return JSON::require_type(NUMBER_JSONTYPE);
    if ((Str::includes_at(T, first_nws, I"double")) && (last_nws - first_nws == 5))
        return JSON::require_type(DOUBLE_JSONTYPE);
    if ((Str::includes_at(T, first_nws, I"string")) && (last_nws - first_nws == 5))
        return JSON::require_type(STRING_JSONTYPE);
    if ((Str::includes_at(T, first_nws, I"boolean")) && (last_nws - first_nws == 6))
        return JSON::require_type(BOOLEAN_JSONTYPE);

    text_stream *msg = Str::new();
    WRITE_TO(msg, "unknown JSON type '");
    for (int i=first_nws; i<last_nws; i++) PUT_TO(msg, Str::get_at(T, i));
    WRITE_TO(msg, "'");
    return JSON::error_sr(msg);
}

§33.2. Find the first and last non-whitespace character in requirement33.2 =

    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE) {
            first_nws = i; break;
        }
    for (int i=to-1; i>=from; i--)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE) {
            last_nws = i; break;
        }
    first_c = Str::get_at(T, first_nws);
    last_c = Str::get_at(T, last_nws);

This code is used in §33, §33.1.

§34. Array requirements:

JSON_single_requirement *JSON::decode_req_array(JSON_single_requirement *array_sr,
    text_stream *T, int from, int to, dictionary *known_names) {
    int content = FALSE;
    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE)
            content = TRUE;
    if (content == FALSE) return array_sr;
    while ((to > from) && (Characters::is_whitespace(Str::get_at(T, to-1)))) to--;
    if (Str::get_at(T, to-1) == '*') {
        to--;
        return JSON::require_array_of(JSON::decode_req_range(T, from, to, known_names));
    }
    NextEntry: ;
    int first_comma = -1, bl = 0;
    for (int i=from, quoted = FALSE; i<to; i++) {
        inchar32_t c = Str::get_at(T, i);
        switch (c) {
            case '"': quoted = (quoted)?FALSE:TRUE; break;
            case '\\': if (quoted) i++; break;
            case ',': if ((quoted == FALSE) && ((first_comma < 0) && (bl == 0)))
                    first_comma = i;
                break;
            case '[': case '{': case '(': if (quoted == FALSE) bl++; break;
            case ']': case '}': case ')': if (quoted == FALSE) bl--; break;
        }
    }
    if (first_comma >= 0) {
        array_sr = JSON::decode_req_array_entry(array_sr, T, from, first_comma, known_names);
        from = first_comma + 1;
        goto NextEntry;
    }
    return JSON::decode_req_array_entry(array_sr, T, from, to, known_names);
}

JSON_single_requirement *JSON::decode_req_array_entry(JSON_single_requirement *array_sr,
    text_stream *T, int from, int to, dictionary *known_names) {
    JSON_requirement *req = JSON::decode_req_range(T, from, to, known_names);
    JSON::require_entry(array_sr, req);
    return array_sr;
}

§35. And similarly for objects.

JSON_single_requirement *JSON::decode_req_object(JSON_single_requirement *obj,
    text_stream *T, int from, int to, dictionary *known_names) {
    int content = FALSE;
    for (int i=from; i<to; i++)
        if (Characters::is_whitespace(Str::get_at(T, i)) == FALSE)
            content = TRUE;
    if (content == FALSE) return obj;
    NextEntry: ;
    int first_comma = -1, bl = 0;
    for (int i=from, quoted = FALSE; i<to; i++) {
        inchar32_t c = Str::get_at(T, i);
        switch (c) {
            case '"': quoted = (quoted)?FALSE:TRUE; break;
            case '\\': if (quoted) i++; break;
            case ',': if ((quoted == FALSE) && ((first_comma < 0) && (bl == 0)))
                    first_comma = i;
                break;
            case '[': case '{': case '(': if (quoted == FALSE) bl++; break;
            case ']': case '}': case ')': if (quoted == FALSE) bl--; break;
        }
    }
    if (first_comma >= 0) {
        obj = JSON::decode_req_object_entry(obj, T, from, first_comma, known_names);
        from = first_comma + 1;
        goto NextEntry;
    }
    return JSON::decode_req_object_entry(obj, T, from, to, known_names);
}

JSON_single_requirement *JSON::decode_req_object_entry(JSON_single_requirement *obj,
    text_stream *T, int from, int to, dictionary *known_names) {
    int optional = FALSE;
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    if (Str::get_at(T, from) == '?') { optional = TRUE; from++; }
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    if (Str::get_at(T, from) != '"')
        return JSON::error_sr(I"key does not begin with quotation mark");
    from++;
    int ended = FALSE;
    TEMPORARY_TEXT(key)
    while (from < to) {
        inchar32_t c = Str::get_at(T, from++);
        if (c == '\"') { ended = TRUE; break; }
        PUT_TO(key, c);
        if ((c == '\\') && (from+1 < to)) {
            c = Str::get_at(T, from++);
            PUT_TO(key, c);
        }
    }
    if (ended == FALSE) return JSON::error_sr(I"key does not end with quotation mark");
    while (Characters::is_whitespace(Str::get_at(T, from))) from++;
    if ((from >= to) || (Str::get_at(T, from) != ':'))
        return JSON::error_sr(I"key is not followed by ':'");
    from++;
    if (JSON::look_up_pair(obj, key)) return JSON::error_sr(I"duplicate key");
    JSON_requirement *req = JSON::decode_req_range(T, from, to, known_names);
    if (optional) JSON::allow_pair(obj, key, req);
    else JSON::require_pair(obj, key, req);
    DISCARD_TEXT(key)
    return obj;
}

§36. Encoding JSON requirements. This is now simple, with one caveat. It's possible to set up requirement trees so that they are not well-founded. For example:

    JSON_single_requirement *set = JSON::require_type(ARRAY_JSONTYPE);
    set->all_if_list = JSON::single_choice(set);

This is not useless: it matches, say, [], [ [] ] and [ [], [ [] ] ] and other constructions giving amusement to set theorists. But it would cause the following to hang. Note that requirements read in from files (see below) are always well-founded, and so do not have this issue.

void JSON::encode_req(OUTPUT_STREAM, JSON_requirement *req) {
    JSON::encode_req_r(OUT, req);
}

void JSON::encode_req_r(OUTPUT_STREAM, JSON_requirement *req) {
    if (req == NULL) internal_error("no JSON value supplied");
    int L = LinkedLists::len(req->alternatives);
    if (L > 1) WRITE("( ");
    int c = 0;
    JSON_single_requirement *sing;
    LOOP_OVER_LINKED_LIST(sing, JSON_single_requirement, req->alternatives) {
        if (c++ > 0) WRITE(" | ");
        JSON::encode_sreq_r(OUT, sing);
    }
    if (L > 1) WRITE(" )");
}

void JSON::encode_sreq_r(OUTPUT_STREAM, JSON_single_requirement *sing) {
    if (sing->this_requirement) JSON::encode_req_r(OUT, sing->this_requirement);
    if (sing->this_value) JSON::encode(OUT, sing->this_value);
    if (sing->this_type) JSON::encode_type(OUT, sing->this_type);
}

void JSON::encode_type(OUTPUT_STREAM, JSON_type *type) {
    switch (type->JSON_type) {
        case ARRAY_JSONTYPE: {
            WRITE("[");
            if (type->all_if_list) {
                WRITE(" ");
                JSON::encode_req_r(OUT, type->all_if_list);
                WRITE("* ");
            } else {
                int count = 0;
                JSON_requirement *E_req;
                LOOP_OVER_LINKED_LIST(E_req, JSON_requirement, type->if_list) {
                    if (count++ > 0) WRITE(",");
                    WRITE(" ");
                    JSON::encode_req_r(OUT, E_req);
                }
                if (count > 0) WRITE(" ");
            }
            WRITE("]");
            break;
        }
        case OBJECT_JSONTYPE: {
            WRITE("{\n"); INDENT;
            int count = 0;
            text_stream *key;
            LOOP_OVER_LINKED_LIST(key, text_stream, type->list_if_object) {
                if (count++ > 0) WRITE(",\n");
                JSON_pair_requirement *pr =
                    Dictionaries::read_value(type->dictionary_if_object, key);
                if (pr == NULL) internal_error("broken JSON req dictionary");
                if (pr->optional) WRITE("?");
                WRITE("\"");
                JSON::encode_string(OUT, key);
                WRITE("\": ");
                JSON::encode_req_r(OUT, pr->req);
            }
            if (count > 0) WRITE("\n");
            OUTDENT; WRITE("}");
            break;
        }
        default: JSON::write_type(OUT, type->JSON_type);
    }
}

§37. Reading requirements files. This convenient function reads in a set of requirements from a text file. Each requirement should begin <name> ::=, and then continues until the next such header, or the end of the file. So for example:

    ! My scheme for JSON files describing geographical locations

    <optional-letter> ::= ( "alpha" | "beta" | null )

    <position> ::= {
        "category": <optional-letter>,
        "latitude": double,
        "longitude": double,
    }

is a valid file declaring two requirements. Forward references are not allowed — e.g., <position> can refer to <optional-letter> but not vice versa — and therefore the requirements read in will always be well-founded. Comments are lines beginning with !; other than comments, only white space is permitted before the first requirement begins.

Note that the function JSON::read_requirements_file returns a dictionary of the requirements it has read, by name (but without their angle-brackets): here, it would have two keys, optional-letter and position.

typedef struct JSON_rrf_state {
    struct text_stream *name;
    struct text_stream *defn;
    struct dictionary *dict;
    struct text_file_position at;
} JSON_rrf_state;

dictionary *JSON::read_requirements_file(dictionary *known, filename *F) {
    if (known == NULL) known = Dictionaries::new(32, FALSE);
    JSON_rrf_state state;
    state.name = Str::new();
    state.defn = Str::new();
    state.dict = known;
    TextFiles::read(F, FALSE, "unable to read file of JSON requirements", TRUE,
        &JSON::read_requirements_file_helper, NULL, (void *) &state);
    JSON::process_req_defn(&state);
    return known;
}

void JSON::read_requirements_file_helper(text_stream *text, text_file_position *tfp,
    void *v_state) {
    JSON_rrf_state *state = (JSON_rrf_state *) v_state;
    match_results mr = Regexp::create_mr();
    if (Regexp::match(&mr, text, U" *<(%C+)> *::= *(%c*)")) {
        JSON::process_req_defn(state);
        WRITE_TO(state->name, "%S", mr.exp[0]);
        WRITE_TO(state->defn, "%S", mr.exp[1]);
        state->at = *tfp;
    } else if (Regexp::match(&mr, text, U" *!%c*")) {
         do nothing: this line is a comment
    } else if (Regexp::match(&mr, text, U" *")) {
         do nothing: this line is blank
    } else if (Str::len(state->name) > 0) {
        WRITE_TO(state->defn, "%S\n", text);
    } else {
        Errors::in_text_file_S(I"spurious text before first requirement", tfp);
    }
}

The structure JSON_rrf_state is accessed in 2/trs, 4/prp, 5/mrk, 5/mpi, 5/mpi2, 5/mv, 5/im, 7/ld and here.

§38. This is called when the end of a definition is reached, either because another is about to start, or because the end of the file has come:

void JSON::process_req_defn(JSON_rrf_state *state) {
    if (Str::len(state->name) > 0) {
        JSON_requirement *req =
            JSON::decode_printing_errors(state->defn, state->dict, &(state->at));
        if (req) {
            dict_entry *de = Dictionaries::create(state->dict, state->name);
            if (de) de->value = req;
        }
    }
    Str::clear(state->name);
    Str::clear(state->defn);
}

JSON_requirement *JSON::decode_printing_errors(text_stream *defn, dictionary *dict,
    text_file_position *tfp) {
    JSON_requirement *req = JSON::decode_req(defn, dict);
    if (req == NULL) internal_error("decode_req returned NULL");
    int errors_found = FALSE;
    JSON_single_requirement *sing;
    LOOP_OVER_LINKED_LIST(sing, JSON_single_requirement, req->alternatives) {
        if ((sing->this_type) && (sing->this_type->JSON_type == ERROR_JSONTYPE)) {
            TEMPORARY_TEXT(err)
            WRITE_TO(err, "JSON requirement error: %S", sing->this_type->if_error);
            Errors::in_text_file_S(err, tfp);
            errors_found = TRUE;
            DISCARD_TEXT(err)
        }
    }
    if (errors_found == FALSE) return req;
    return NULL;
}

JSON_requirement *JSON::look_up_requirements(dictionary *known, text_stream *name) {
    dict_entry *de = Dictionaries::find(known, name);
    if (de == NULL) return NULL;
    return de->value;
}