1
0
mirror of https://github.com/janet-lang/janet synced 2024-11-16 05:34:48 +00:00
janet/core/parse.c

568 lines
18 KiB
C
Raw Normal View History

/*
* Copyright (c) 2017 Calvin Rose
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
#include <dst/dst.h>
2017-12-08 20:57:02 +00:00
#include "strtod.h"
/* Checks if a string slice is equal to a string constant */
static int check_str_const(const char *ref, const uint8_t *start, const uint8_t *end) {
while (*ref && start < end) {
if (*ref != *(char *)start) return 0;
++ref;
++start;
}
return !*ref && start == end;
}
/* Quote a value */
static Dst quote(Dst x) {
Dst *t = dst_tuple_begin(2);
t[0] = dst_csymbolv("quote");
t[1] = x;
return dst_wrap_tuple(dst_tuple_end(t));
}
/* Check if a character is whitespace */
static int is_whitespace(uint8_t c) {
2017-11-01 21:53:43 +00:00
return c == ' '
|| c == '\t'
|| c == '\n'
|| c == '\r'
|| c == '\0'
|| c == ';'
2017-11-01 21:53:43 +00:00
|| c == ',';
}
/* Code gen
printf("static uint32_t symchars[8] = {\n\t");
for (int i = 0; i < 256; i += 32) {
uint32_t block = 0;
for (int j = 0; j < 32; j++) {
block |= is_symbol_char_gen(i + j) << j;
}
printf("0x%08x%s", block, (i == (256 - 32)) ? "" : ", ");
}
printf("\n};\n");
static int is_symbol_char_gen(uint8_t c) {
if (c >= 'a' && c <= 'z') return 1;
if (c >= 'A' && c <= 'Z') return 1;
if (c >= '0' && c <= '9') return 1;
return (c == '!' ||
c == '$' ||
c == '%' ||
c == '&' ||
c == '*' ||
c == '+' ||
c == '-' ||
c == '.' ||
c == '/' ||
c == ':' ||
c == '<' ||
c == '=' ||
c == '>' ||
c == '@' ||
c == '\\' ||
c == '^' ||
c == '_' ||
c == '~' ||
c == '|');
}
The table contains 256 bits, where each bit is 1
if the corresponding ascci code is a symbol char, and 0
if not. The upper characters are also considered symbol
chars and are then checked for utf-8 compliance. */
2018-01-13 00:23:36 +00:00
static uint32_t symchars[8] = {
0x00000000, 0x77ffec72, 0xd7ffffff, 0x57fffffe,
0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff
};
/* Check if a character is a valid symbol character
* symbol chars are A-Z, a-z, 0-9, or one of !$&*+-./:<=>@\^_~| */
static int is_symbol_char(uint8_t c) {
return symchars[c >> 5] & (1 << (c & 0x1F));
}
/* Validate some utf8. Useful for identifiers. Only validates
* the encoding, does not check for valid codepoints (they
* are less well defined than the encoding). */
static int valid_utf8(const uint8_t *str, int32_t len) {
int32_t i = 0;
int32_t j;
while (i < len) {
int32_t nexti;
uint8_t c = str[i];
/* Check the number of bytes in code point */
if (c < 0x80) nexti = i + 1;
else if ((c >> 5) == 0x06) nexti = i + 2;
else if ((c >> 4) == 0x0E) nexti = i + 3;
else if ((c >> 3) == 0x1E) nexti = i + 4;
/* Don't allow 5 or 6 byte code points */
else return 0;
/* No overflow */
if (nexti > len)
return 0;
/* Ensure trailing bytes are well formed (10XX XXXX) */
for (j = i + 1; j < nexti; j++) {
if ((str[j] >> 6) != 2)
return 0;
}
/* Check for overlong encodings */
if ((nexti == i + 2) && str[i] < 0xC2) return 0;
if ((str[i] == 0xE0) && str[i + 1] < 0xA0) return 0;
if ((str[i] == 0xF0) && str[i + 1] < 0x90) return 0;
i = nexti;
}
return 1;
}
/* Get hex digit from a letter */
static int to_hex(uint8_t c) {
if (c >= '0' && c <= '9') {
return c - '0';
} else if (c >= 'A' && c <= 'F') {
return 10 + c - 'A';
} else if (c >= 'a' && c <= 'f') {
return 10 + c - 'a';
} else {
return -1;
}
}
/* Make source mapping for atom (non recursive structure) */
static Dst atom_map(int32_t start, int32_t end) {
Dst *t = dst_tuple_begin(2);
t[0] = dst_wrap_integer(start);
t[1] = dst_wrap_integer(end);
return dst_wrap_tuple(dst_tuple_end(t));
}
/* Create mappingd for recursive data structure */
static Dst ds_map(int32_t start, int32_t end, Dst submap) {
Dst *t = dst_tuple_begin(3);
t[0] = dst_wrap_integer(start);
t[1] = dst_wrap_integer(end);
t[2] = submap;
return dst_wrap_tuple(dst_tuple_end(t));
}
/* Create a sourcemapping for a key value pair */
static Dst kv_map(Dst k, Dst v) {
Dst *t = dst_tuple_begin(2);
t[0] = k;
t[1] = v;
return dst_wrap_tuple(dst_tuple_end(t));
}
typedef struct {
DstArray stack;
DstArray mapstack;
const uint8_t *srcstart;
const uint8_t *end;
2017-11-01 21:53:43 +00:00
const char *errmsg;
DstParseStatus status;
} ParseArgs;
2017-11-01 21:53:43 +00:00
/* Entry point of the recursive descent parser */
static const uint8_t *parse_recur(
ParseArgs *args,
const uint8_t *src,
int32_t recur) {
const uint8_t *end = args->end;
const uint8_t *mapstart;
int32_t qcount = 0;
Dst ret;
Dst submapping;
/* Prevent stack overflow */
if (recur == 0) goto too_much_recur;
/* try parsing again */
begin:
/* Trim leading whitespace and count quotes */
while (src < end && (is_whitespace(*src) || *src == '\'')) {
if (*src == '\'') {
++qcount;
}
++src;
}
/* Check for end of source */
if (src >= end) {
if (qcount || recur != DST_RECURSION_GUARD) {
goto unexpected_eos;
} else {
goto nodata;
}
}
/* Open mapping */
mapstart = src;
submapping = dst_wrap_nil();
/* Detect token type based on first character */
switch (*src) {
/* Numbers, symbols, simple literals */
default:
atom:
{
int ascii = 1;
Dst numcheck;
const uint8_t *tokenend = src;
if (!is_symbol_char(*src)) goto unexpected_character;
while (tokenend < end && is_symbol_char(*tokenend)) {
if (*tokenend++ > 127) ascii = 0;
}
2017-12-08 20:57:02 +00:00
numcheck = dst_scan_number(src, tokenend - src);
if (!dst_checktype(numcheck, DST_NIL)) {
ret = numcheck;
} else if (check_str_const("nil", src, tokenend)) {
ret = dst_wrap_nil();
} else if (check_str_const("false", src, tokenend)) {
ret = dst_wrap_boolean(0);
} else if (check_str_const("true", src, tokenend)) {
ret = dst_wrap_boolean(1);
} else {
if (*src >= '0' && *src <= '9') {
goto sym_nodigits;
} else {
/* Don't do full utf8 check unless we have seen non ascii characters. */
int valid = ascii || valid_utf8(src, tokenend - src);
if (!valid)
goto invalid_utf8;
if (*src == ':') {
ret = dst_stringv(src + 1, tokenend - src - 1);
} else {
ret = dst_symbolv(src, tokenend - src);
}
}
}
src = tokenend;
break;
}
case '#':
{
/* Jump to next newline */
while (src < end && *src != '\n')
++src;
goto begin;
}
/* String literals */
case '"':
{
const uint8_t *strend = ++src;
const uint8_t *strstart = strend;
int32_t len = 0;
int containsEscape = 0;
/* Preprocess string to check for escapes and string end */
while (strend < end && *strend != '"') {
len++;
if (*strend++ == '\\') {
2017-11-01 21:53:43 +00:00
containsEscape = 1;
if (strend >= end) goto unexpected_eos;
if (*strend == 'h') {
strend += 3;
if (strend >= end) goto unexpected_eos;
2017-11-01 21:53:43 +00:00
} else {
strend++;
if (strend >= end) goto unexpected_eos;
}
}
}
if (containsEscape) {
uint8_t *buf = dst_string_begin(len);
uint8_t *write = buf;
2017-11-01 21:53:43 +00:00
while (src < strend) {
if (*src == '\\') {
src++;
switch (*src++) {
case 'n': *write++ = '\n'; break;
case 'r': *write++ = '\r'; break;
case 't': *write++ = '\t'; break;
case 'f': *write++ = '\f'; break;
case '0': *write++ = '\0'; break;
case '"': *write++ = '"'; break;
case '\'': *write++ = '\''; break;
case 'z': *write++ = '\0'; break;
case 'e': *write++ = 27; break;
case 'h': {
2017-11-01 21:53:43 +00:00
int d1 = to_hex(*src++);
int d2 = to_hex(*src++);
if (d1 < 0 || d2 < 0) goto invalid_hex;
2017-11-01 21:53:43 +00:00
*write++ = 16 * d1 + d2;
break;
}
default:
goto unknown_strescape;
}
} else {
2017-11-01 21:53:43 +00:00
*write++ = *src++;
}
}
ret = dst_wrap_string(dst_string_end(buf));
} else {
ret = dst_wrap_string(dst_string(strstart, strend - strstart));
}
src = strend + 1;
break;
}
/* Data Structure literals */
case '@':
if (src[1] != '{')
goto atom;
2017-11-01 21:53:43 +00:00
case '(':
case '[':
case '{':
{
int32_t n = 0, i = 0;
int32_t istable = 0;
uint8_t close;
switch (*src++) {
case '[': close = ']'; break;
case '{': close = '}'; break;
case '@': close = '}'; src++; istable = 1; break;
default: close = ')'; break;
}
/* Trim trailing whitespace */
while (src < end && (is_whitespace(*src)))
++src;
/* Recursively parse inside literal */
while (*src != close) {
2017-11-01 21:53:43 +00:00
src = parse_recur(args, src, recur - 1);
if (args->errmsg || !src) return src;
n++;
/* Trim trailing whitespace */
while (src < end && (is_whitespace(*src)))
++src;
}
src++;
switch (close) {
case ')':
{
Dst *tup = dst_tuple_begin(n);
Dst *subtup = dst_tuple_begin(n);
for (i = n; i > 0; i--) {
tup[i - 1] = dst_array_pop(&args->stack);
subtup[i - 1] = dst_array_pop(&args->mapstack);
}
ret = dst_wrap_tuple(dst_tuple_end(tup));
submapping = dst_wrap_tuple(dst_tuple_end(subtup));
break;
}
case ']':
{
DstArray *arr = dst_array(n);
DstArray *subarr = dst_array(n);
for (i = n; i > 0; i--) {
arr->data[i - 1] = dst_array_pop(&args->stack);
subarr->data[i - 1] = dst_array_pop(&args->mapstack);
}
arr->count = n;
subarr->count = n;
ret = dst_wrap_array(arr);
submapping = dst_wrap_array(subarr);
break;
}
case '}':
{
if (n & 1) {
if (istable)
goto table_oddargs;
goto struct_oddargs;
}
if (istable) {
DstTable *t = dst_table(n);
DstTable *subt = dst_table(n);
for (i = n; i > 0; i -= 2) {
Dst val = dst_array_pop(&args->stack);
Dst key = dst_array_pop(&args->stack);
Dst subval = dst_array_pop(&args->mapstack);
Dst subkey = dst_array_pop(&args->mapstack);
dst_table_put(t, key, val);
dst_table_put(subt, key, kv_map(subkey, subval));
}
ret = dst_wrap_table(t);
submapping = dst_wrap_table(subt);
} else {
DstKV *st = dst_struct_begin(n >> 1);
DstKV *subst = dst_struct_begin(n >> 1);
for (i = n; i > 0; i -= 2) {
Dst val = dst_array_pop(&args->stack);
Dst key = dst_array_pop(&args->stack);
Dst subval = dst_array_pop(&args->mapstack);
Dst subkey = dst_array_pop(&args->mapstack);
dst_struct_put(st, key, val);
dst_struct_put(subst, key, kv_map(subkey, subval));
}
ret = dst_wrap_struct(dst_struct_end(st));
submapping = dst_wrap_struct(dst_struct_end(subst));
}
break;
}
}
break;
}
}
/* Push source mapping */
if (dst_checktype(submapping, DST_NIL)) {
/* We just parsed an atom */
dst_array_push(&args->mapstack, atom_map(
mapstart - args->srcstart,
src - args->srcstart));
} else {
/* We just parsed a recursive data structure */
dst_array_push(&args->mapstack, ds_map(
mapstart - args->srcstart,
src - args->srcstart,
submapping));
}
/* Quote the returned value qcount times */
while (qcount--) {
int32_t start = mapstart - args->srcstart;
int32_t end = src - args->srcstart;
Dst sourcemap = dst_array_pop(&args->mapstack);
Dst* tup = dst_tuple_begin(2);
tup[0] = atom_map(start, end);
tup[1] = sourcemap;
ret = quote(ret);
dst_array_push(&args->mapstack, ds_map(
start,
end,
dst_wrap_tuple(dst_tuple_end(tup))));
}
/* Push the result to the stack */
dst_array_push(&args->stack, ret);
/* Return the new source position for further calls */
return src;
/* Errors below */
nodata:
args->status = DST_PARSE_NODATA;
return NULL;
unexpected_eos:
2017-11-01 21:53:43 +00:00
args->errmsg = "unexpected end of source";
args->status = DST_PARSE_UNEXPECTED_EOS;
return NULL;
unexpected_character:
2017-11-01 21:53:43 +00:00
args->errmsg = "unexpected character";
args->status = DST_PARSE_ERROR;
return src;
sym_nodigits:
2017-11-01 21:53:43 +00:00
args->errmsg = "symbols cannot start with digits";
args->status = DST_PARSE_ERROR;
return src;
table_oddargs:
args->errmsg = "table literal needs an even number of arguments";
args->status = DST_PARSE_ERROR;
return src;
struct_oddargs:
2017-11-01 21:53:43 +00:00
args->errmsg = "struct literal needs an even number of arguments";
args->status = DST_PARSE_ERROR;
return src;
unknown_strescape:
2017-11-01 21:53:43 +00:00
args->errmsg = "unknown string escape sequence";
args->status = DST_PARSE_ERROR;
return src;
invalid_hex:
2017-11-01 21:53:43 +00:00
args->errmsg = "invalid hex escape in string";
args->status = DST_PARSE_ERROR;
return src;
invalid_utf8:
args->errmsg = "identifier is not valid utf-8";
args->status = DST_PARSE_ERROR;
return src;
too_much_recur:
2017-11-01 21:53:43 +00:00
args->errmsg = "recursed too deeply in parsing";
args->status = DST_PARSE_ERROR;
return src;
}
/* Parse an array of bytes. Return value in the fiber return value. */
DstParseResult dst_parse(const uint8_t *src, int32_t len) {
DstParseResult res;
ParseArgs args;
const uint8_t *newsrc;
dst_array_init(&args.stack, 10);
args.status = DST_PARSE_OK;
args.srcstart = src;
args.end = src + len;
args.errmsg = NULL;
2017-11-01 21:53:43 +00:00
dst_array_init(&args.mapstack, 10);
newsrc = parse_recur(&args, src, DST_RECURSION_GUARD);
res.status = args.status;
res.bytes_read = (int32_t) (newsrc - src);
if (args.errmsg) {
res.error = dst_cstring(args.errmsg);
res.value = dst_wrap_nil();
2017-12-08 20:57:02 +00:00
res.map = NULL;
} else {
res.value = dst_array_pop(&args.stack);
res.error = NULL;
2017-12-08 20:57:02 +00:00
res.map = dst_unwrap_tuple(dst_array_pop(&args.mapstack));
}
2017-11-01 21:53:43 +00:00
dst_array_deinit(&args.stack);
dst_array_deinit(&args.mapstack);
return res;
}
/* Parse a c string */
DstParseResult dst_parsec(const char *src) {
int32_t len = 0;
while (src[len]) ++len;
return dst_parse((const uint8_t *)src, len);
}