tokenizer.c

/* Tokenizer implementation */

#include "Python.h"
#include "pycore_call.h"          // _PyObject_CallNoArgs()

#include <assert.h>

#include "tokenizer.h"
#include "errcode.h"

/* Alternate tab spacing */
#define ALTTABSIZE 1

#define is_potential_identifier_start(c) (\
              (c >= 'a' && c <= 'z')\
               || (c >= 'A' && c <= 'Z')\
               || c == '_'\
               || (c >= 128))

#define is_potential_identifier_char(c) (\
              (c >= 'a' && c <= 'z')\
               || (c >= 'A' && c <= 'Z')\
               || (c >= '0' && c <= '9')\
               || c == '_'\
               || (c >= 128))


/* Don't ever change this -- it would break the portability of Python code */
#define TABSIZE 8

#define MAKE_TOKEN(token_type) token_setup(tok, token, token_type, p_start, p_end)
#define MAKE_TYPE_COMMENT_TOKEN(token_type, col_offset, end_col_offset) (\
                type_comment_token_setup(tok, token, token_type, col_offset, end_col_offset, p_start, p_end))
#define ADVANCE_LINENO() \
            tok->lineno++; \
            tok->col_offset = 0;

#define INSIDE_FSTRING(tok) (tok->tok_mode_stack_index > 0)
#define INSIDE_FSTRING_EXPR(tok) (tok->curly_bracket_expr_start_depth >= 0)
#ifdef Py_DEBUG
static inline tokenizer_mode* TOK_GET_MODE(struct tok_state* tok) {
    assert(tok->tok_mode_stack_index >= 0);
    assert(tok->tok_mode_stack_index < MAXFSTRINGLEVEL);
    return &(tok->tok_mode_stack[tok->tok_mode_stack_index]);
}
static inline tokenizer_mode* TOK_NEXT_MODE(struct tok_state* tok) {
    assert(tok->tok_mode_stack_index >= 0);
    assert(tok->tok_mode_stack_index + 1 < MAXFSTRINGLEVEL);
    return &(tok->tok_mode_stack[++tok->tok_mode_stack_index]);
}
#else
#define TOK_GET_MODE(tok) (&(tok->tok_mode_stack[tok->tok_mode_stack_index]))
#define TOK_NEXT_MODE(tok) (&(tok->tok_mode_stack[++tok->tok_mode_stack_index]))
#endif

/* Forward */
static struct tok_state *tok_new(void);
static int tok_nextc(struct tok_state *tok);
static void tok_backup(struct tok_state *tok, int c);
static int syntaxerror(struct tok_state *tok, const char *format, ...);

/* Spaces in this constant are treated as "zero or more spaces or tabs" when
   tokenizing. */
static const char* type_comment_prefix = "# type: ";

/* Create and initialize a new tok_state structure */

static struct tok_state *
tok_new(void)
{
    struct tok_state *tok = (struct tok_state *)PyMem_Malloc(
                                            sizeof(struct tok_state));
    if (tok == NULL)
        return NULL;
    tok->buf = tok->cur = tok->inp = NULL;
    tok->fp_interactive = 0;
    tok->interactive_src_start = NULL;
    tok->interactive_src_end = NULL;
    tok->start = NULL;
    tok->end = NULL;
    tok->done = E_OK;
    tok->fp = NULL;
    tok->input = NULL;
    tok->tabsize = TABSIZE;
    tok->indent = 0;
    tok->indstack[0] = 0;
    tok->atbol = 1;
    tok->pendin = 0;
    tok->prompt = tok->nextprompt = NULL;
    tok->lineno = 0;
    tok->starting_col_offset = -1;
    tok->col_offset = -1;
    tok->level = 0;
    tok->altindstack[0] = 0;
    tok->decoding_state = STATE_INIT;
    tok->decoding_erred = 0;
    tok->enc = NULL;
    tok->encoding = NULL;
    tok->cont_line = 0;
    tok->filename = NULL;
    tok->decoding_readline = NULL;
    tok->decoding_buffer = NULL;
    tok->readline = NULL;
    tok->type_comments = 0;
    tok->interactive_underflow = IUNDERFLOW_NORMAL;
    tok->str = NULL;
    tok->report_warnings = 1;
    tok->tok_extra_tokens = 0;
    tok->comment_newline = 0;
    tok->implicit_newline = 0;
    tok->tok_mode_stack[0] = (tokenizer_mode){.kind =TOK_REGULAR_MODE, .f_string_quote='\0', .f_string_quote_size = 0, .f_string_debug=0};
    tok->tok_mode_stack_index = 0;
#ifdef Py_DEBUG
    tok->debug = _Py_GetConfig()->parser_debug;
#endif
    return tok;
}

static char *
new_string(const char *s, Py_ssize_t len, struct tok_state *tok)
{
    char* result = (char *)PyMem_Malloc(len + 1);
    if (!result) {
        tok->done = E_NOMEM;
        return NULL;
    }
    memcpy(result, s, len);
    result[len] = '\0';
    return result;
}

static char *
error_ret(struct tok_state *tok) /* XXX */
{
    tok->decoding_erred = 1;
    if ((tok->fp != NULL || tok->readline != NULL) && tok->buf != NULL) {/* see _PyTokenizer_Free */
        PyMem_Free(tok->buf);
    }
    tok->buf = tok->cur = tok->inp = NULL;
    tok->start = NULL;
    tok->end = NULL;
    tok->done = E_DECODE;
    return NULL;                /* as if it were EOF */
}


static const char *
get_normal_name(const char *s)  /* for utf-8 and latin-1 */
{
    char buf[13];
    int i;
    for (i = 0; i < 12; i++) {
        int c = s[i];
        if (c == '\0')
            break;
        else if (c == '_')
            buf[i] = '-';
        else
            buf[i] = Py_TOLOWER(c);
    }
    buf[i] = '\0';
    if (strcmp(buf, "utf-8") == 0 ||
        strncmp(buf, "utf-8-", 6) == 0)
        return "utf-8";
    else if (strcmp(buf, "latin-1") == 0 ||
             strcmp(buf, "iso-8859-1") == 0 ||
             strcmp(buf, "iso-latin-1") == 0 ||
             strncmp(buf, "latin-1-", 8) == 0 ||
             strncmp(buf, "iso-8859-1-", 11) == 0 ||
             strncmp(buf, "iso-latin-1-", 12) == 0)
        return "iso-8859-1";
    else
        return s;
}

/* Return the coding spec in S, or NULL if none is found.  */

static int
get_coding_spec(const char *s, char **spec, Py_ssize_t size, struct tok_state *tok)
{
    Py_ssize_t i;
    *spec = NULL;
    /* Coding spec must be in a comment, and that comment must be
     * the only statement on the source code line. */
    for (i = 0; i < size - 6; i++) {
        if (s[i] == '#')
            break;
        if (s[i] != ' ' && s[i] != '\t' && s[i] != '\014')
            return 1;
    }
    for (; i < size - 6; i++) { /* XXX inefficient search */
        const char* t = s + i;
        if (memcmp(t, "coding", 6) == 0) {
            const char* begin = NULL;
            t += 6;
            if (t[0] != ':' && t[0] != '=')
                continue;
            do {
                t++;
            } while (t[0] == ' ' || t[0] == '\t');

            begin = t;
            while (Py_ISALNUM(t[0]) ||
                   t[0] == '-' || t[0] == '_' || t[0] == '.')
                t++;

            if (begin < t) {
                char* r = new_string(begin, t - begin, tok);
                const char* q;
                if (!r)
                    return 0;
                q = get_normal_name(r);
                if (r != q) {
                    PyMem_Free(r);
                    r = new_string(q, strlen(q), tok);
                    if (!r)
                        return 0;
                }
                *spec = r;
                break;
            }
        }
    }
    return 1;
}

/* Check whether the line contains a coding spec. If it does,
   invoke the set_readline function for the new encoding.
   This function receives the tok_state and the new encoding.
   Return 1 on success, 0 on failure.  */

static int
check_coding_spec(const char* line, Py_ssize_t size, struct tok_state *tok,
                  int set_readline(struct tok_state *, const char *))
{
    char *cs;
    if (tok->cont_line) {
        /* It's a continuation line, so it can't be a coding spec. */
        tok->decoding_state = STATE_NORMAL;
        return 1;
    }
    if (!get_coding_spec(line, &cs, size, tok)) {
        return 0;
    }
    if (!cs) {
        Py_ssize_t i;
        for (i = 0; i < size; i++) {
            if (line[i] == '#' || line[i] == '\n' || line[i] == '\r')
                break;
            if (line[i] != ' ' && line[i] != '\t' && line[i] != '\014') {
                /* Stop checking coding spec after a line containing
                 * anything except a comment. */
                tok->decoding_state = STATE_NORMAL;
                break;
            }
        }
        return 1;
    }
    tok->decoding_state = STATE_NORMAL;
    if (tok->encoding == NULL) {
        assert(tok->decoding_readline == NULL);
        if (strcmp(cs, "utf-8") != 0 && !set_readline(tok, cs)) {
            error_ret(tok);
            PyErr_Format(PyExc_SyntaxError, "encoding problem: %s", cs);
            PyMem_Free(cs);
            return 0;
        }
        tok->encoding = cs;
    } else {                /* then, compare cs with BOM */
        if (strcmp(tok->encoding, cs) != 0) {
            error_ret(tok);
            PyErr_Format(PyExc_SyntaxError,
                         "encoding problem: %s with BOM", cs);
            PyMem_Free(cs);
            return 0;
        }
        PyMem_Free(cs);
    }
    return 1;
}

/* See whether the file starts with a BOM. If it does,
   invoke the set_readline function with the new encoding.
   Return 1 on success, 0 on failure.  */

static int
check_bom(int get_char(struct tok_state *),
          void unget_char(int, struct tok_state *),
          int set_readline(struct tok_state *, const char *),
          struct tok_state *tok)
{
    int ch1, ch2, ch3;
    ch1 = get_char(tok);
    tok->decoding_state = STATE_SEEK_CODING;
    if (ch1 == EOF) {
        return 1;
    } else if (ch1 == 0xEF) {
        ch2 = get_char(tok);
        if (ch2 != 0xBB) {
            unget_char(ch2, tok);
            unget_char(ch1, tok);
            return 1;
        }
        ch3 = get_char(tok);
        if (ch3 != 0xBF) {
            unget_char(ch3, tok);
            unget_char(ch2, tok);
            unget_char(ch1, tok);
            return 1;
        }
    } else {
        unget_char(ch1, tok);
        return 1;
    }
    if (tok->encoding != NULL)
        PyMem_Free(tok->encoding);
    tok->encoding = new_string("utf-8", 5, tok);
    if (!tok->encoding)
        return 0;
    /* No need to set_readline: input is already utf-8 */
    return 1;
}

static int
tok_concatenate_interactive_new_line(struct tok_state *tok, const char *line) {
    assert(tok->fp_interactive);

    if (!line) {
        return 0;
    }

    Py_ssize_t current_size = tok->interactive_src_end - tok->interactive_src_start;
    Py_ssize_t line_size = strlen(line);
    char last_char = line[line_size > 0 ? line_size - 1 : line_size];
    if (last_char != '\n') {
        line_size += 1;
    }
    char* new_str = tok->interactive_src_start;

    new_str = PyMem_Realloc(new_str, current_size + line_size + 1);
    if (!new_str) {
        if (tok->interactive_src_start) {
            PyMem_Free(tok->interactive_src_start);
        }
        tok->interactive_src_start = NULL;
        tok->interactive_src_end = NULL;
        tok->done = E_NOMEM;
        return -1;
    }
    strcpy(new_str + current_size, line);
    tok->implicit_newline = 0;
    if (last_char != '\n') {
        /* Last line does not end in \n, fake one */
        new_str[current_size + line_size - 1] = '\n';
        new_str[current_size + line_size] = '\0';
        tok->implicit_newline = 1;
    }
    tok->interactive_src_start = new_str;
    tok->interactive_src_end = new_str + current_size + line_size;
    return 0;
}

/* Traverse and remember all f-string buffers, in order to be able to restore
   them after reallocating tok->buf */
static void
remember_fstring_buffers(struct tok_state *tok)
{
    int index;
    tokenizer_mode *mode;

    for (index = tok->tok_mode_stack_index; index >= 0; --index) {
        mode = &(tok->tok_mode_stack[index]);
        mode->f_string_start_offset = mode->f_string_start - tok->buf;
        mode->f_string_multi_line_start_offset = mode->f_string_multi_line_start - tok->buf;
    }
}

/* Traverse and restore all f-string buffers after reallocating tok->buf */
static void
restore_fstring_buffers(struct tok_state *tok)
{
    int index;
    tokenizer_mode *mode;

    for (index = tok->tok_mode_stack_index; index >= 0; --index) {
        mode = &(tok->tok_mode_stack[index]);
        mode->f_string_start = tok->buf + mode->f_string_start_offset;
        mode->f_string_multi_line_start = tok->buf + mode->f_string_multi_line_start_offset;
    }
}

static int
set_fstring_expr(struct tok_state* tok, struct token *token, char c) {
    assert(token != NULL);
    assert(c == '}' || c == ':' || c == '!');
    tokenizer_mode *tok_mode = TOK_GET_MODE(tok);

    if (!tok_mode->f_string_debug || token->metadata) {
        return 0;
    }

    PyObject *res = PyUnicode_DecodeUTF8(
        tok_mode->last_expr_buffer,
        tok_mode->last_expr_size - tok_mode->last_expr_end,
        NULL
    );
    if (!res) {
        return -1;
    }
    token->metadata = res;
    return 0;
}

static int
update_fstring_expr(struct tok_state *tok, char cur)
{
    assert(tok->cur != NULL);

    Py_ssize_t size = strlen(tok->cur);
    tokenizer_mode *tok_mode = TOK_GET_MODE(tok);

    switch (cur) {
       case 0:
            if (!tok_mode->last_expr_buffer || tok_mode->last_expr_end >= 0) {
                return 1;
            }
            char *new_buffer = PyMem_Realloc(
                tok_mode->last_expr_buffer,
                tok_mode->last_expr_size + size
            );
            if (new_buffer == NULL) {
                PyMem_Free(tok_mode->last_expr_buffer);
                goto error;
            }
            tok_mode->last_expr_buffer = new_buffer;
            strncpy(tok_mode->last_expr_buffer + tok_mode->last_expr_size, tok->cur, size);
            tok_mode->last_expr_size += size;
            break;
        case '{':
            if (tok_mode->last_expr_buffer != NULL) {
                PyMem_Free(tok_mode->last_expr_buffer);
            }
            tok_mode->last_expr_buffer = PyMem_Malloc(size);
            if (tok_mode->last_expr_buffer == NULL) {
                goto error;
            }
            tok_mode->last_expr_size = size;
            tok_mode->last_expr_end = -1;
            strncpy(tok_mode->last_expr_buffer, tok->cur, size);
            break;
        case '}':
        case '!':
        case ':':
            if (tok_mode->last_expr_end == -1) {
                tok_mode->last_expr_end = strlen(tok->start);
            }
            break;
        default:
            Py_UNREACHABLE();
    }
    return 1;
error:
    tok->done = E_NOMEM;
    return 0;
}

static void
free_fstring_expressions(struct tok_state *tok)
{
    int index;
    tokenizer_mode *mode;

    for (index = tok->tok_mode_stack_index; index >= 0; --index) {
        mode = &(tok->tok_mode_stack[index]);
        if (mode->last_expr_buffer != NULL) {
            PyMem_Free(mode->last_expr_buffer);
            mode->last_expr_buffer = NULL;
            mode->last_expr_size = 0;
            mode->last_expr_end = -1;
        }
    }
}

/* Read a line of text from TOK into S, using the stream in TOK.
   Return NULL on failure, else S.

   On entry, tok->decoding_buffer will be one of:
     1) NULL: need to call tok->decoding_readline to get a new line
     2) PyUnicodeObject *: decoding_feof has called tok->decoding_readline and
       stored the result in tok->decoding_buffer
     3) PyByteArrayObject *: previous call to tok_readline_recode did not have enough room
       (in the s buffer) to copy entire contents of the line read
       by tok->decoding_readline.  tok->decoding_buffer has the overflow.
       In this case, tok_readline_recode is called in a loop (with an expanded buffer)
       until the buffer ends with a '\n' (or until the end of the file is
       reached): see tok_nextc and its calls to tok_reserve_buf.
*/

static int
tok_reserve_buf(struct tok_state *tok, Py_ssize_t size)
{
    Py_ssize_t cur = tok->cur - tok->buf;
    Py_ssize_t oldsize = tok->inp - tok->buf;
    Py_ssize_t newsize = oldsize + Py_MAX(size, oldsize >> 1);
    if (newsize > tok->end - tok->buf) {
        char *newbuf = tok->buf;
        Py_ssize_t start = tok->start == NULL ? -1 : tok->start - tok->buf;
        Py_ssize_t line_start = tok->start == NULL ? -1 : tok->line_start - tok->buf;
        Py_ssize_t multi_line_start = tok->multi_line_start - tok->buf;
        remember_fstring_buffers(tok);
        newbuf = (char *)PyMem_Realloc(newbuf, newsize);
        if (newbuf == NULL) {
            tok->done = E_NOMEM;
            return 0;
        }
        tok->buf = newbuf;
        tok->cur = tok->buf + cur;
        tok->inp = tok->buf + oldsize;
        tok->end = tok->buf + newsize;
        tok->start = start < 0 ? NULL : tok->buf + start;
        tok->line_start = line_start < 0 ? NULL : tok->buf + line_start;
        tok->multi_line_start = multi_line_start < 0 ? NULL : tok->buf + multi_line_start;
        restore_fstring_buffers(tok);
    }
    return 1;
}

static inline int
contains_null_bytes(const char* str, size_t size) {
    return memchr(str, 0, size) != NULL;
}

static int
tok_readline_recode(struct tok_state *tok) {
    PyObject *line;
    const  char *buf;
    Py_ssize_t buflen;
    line = tok->decoding_buffer;
    if (line == NULL) {
        line = PyObject_CallNoArgs(tok->decoding_readline);
        if (line == NULL) {
            error_ret(tok);
            goto error;
        }
    }
    else {
        tok->decoding_buffer = NULL;
    }
    buf = PyUnicode_AsUTF8AndSize(line, &buflen);
    if (buf == NULL) {
        error_ret(tok);
        goto error;
    }
    // Make room for the null terminator *and* potentially
    // an extra newline character that we may need to artificially
    // add.
    size_t buffer_size = buflen + 2;
    if (!tok_reserve_buf(tok, buffer_size)) {
        goto error;
    }
    memcpy(tok->inp, buf, buflen);
    tok->inp += buflen;
    *tok->inp = '\0';
    if (tok->fp_interactive &&
        tok_concatenate_interactive_new_line(tok, buf) == -1) {
        goto error;
    }
    Py_DECREF(line);
    return 1;
error:
    Py_XDECREF(line);
    return 0;
}

/* Set the readline function for TOK to a StreamReader's
   readline function. The StreamReader is named ENC.

   This function is called from check_bom and check_coding_spec.

   ENC is usually identical to the future value of tok->encoding,
   except for the (currently unsupported) case of UTF-16.

   Return 1 on success, 0 on failure. */

static int
fp_setreadl(struct tok_state *tok, const char* enc)
{
    PyObject *readline, *open, *stream;
    int fd;
    long pos;

    fd = fileno(tok->fp);
    /* Due to buffering the file offset for fd can be different from the file
     * position of tok->fp.  If tok->fp was opened in text mode on Windows,
     * its file position counts CRLF as one char and can't be directly mapped
     * to the file offset for fd.  Instead we step back one byte and read to
     * the end of line.*/
    pos = ftell(tok->fp);
    if (pos == -1 ||
        lseek(fd, (off_t)(pos > 0 ? pos - 1 : pos), SEEK_SET) == (off_t)-1) {
        PyErr_SetFromErrnoWithFilename(PyExc_OSError, NULL);
        return 0;
    }

    open = _PyImport_GetModuleAttrString("io", "open");
    if (open == NULL) {
        return 0;
    }
    stream = PyObject_CallFunction(open, "isisOOO",
                    fd, "r", -1, enc, Py_None, Py_None, Py_False);
    Py_DECREF(open);
    if (stream == NULL) {
        return 0;
    }

    readline = PyObject_GetAttr(stream, &_Py_ID(readline));
    Py_DECREF(stream);
    if (readline == NULL) {
        return 0;
    }
    Py_XSETREF(tok->decoding_readline, readline);

    if (pos > 0) {
        PyObject *bufobj = _PyObject_CallNoArgs(readline);
        if (bufobj == NULL) {
            return 0;
        }
        Py_DECREF(bufobj);
    }

    return 1;
}

/* Fetch the next byte from TOK. */

static int fp_getc(struct tok_state *tok) {
    return getc(tok->fp);
}

/* Unfetch the last byte back into TOK.  */

static void fp_ungetc(int c, struct tok_state *tok) {
    ungetc(c, tok->fp);
}

/* Check whether the characters at s start a valid
   UTF-8 sequence. Return the number of characters forming
   the sequence if yes, 0 if not.  The special cases match
   those in stringlib/codecs.h:utf8_decode.
*/
static int
valid_utf8(const unsigned char* s)
{
    int expected = 0;
    int length;
    if (*s < 0x80) {
        /* single-byte code */
        return 1;
    }
    else if (*s < 0xE0) {
        /* \xC2\x80-\xDF\xBF -- 0080-07FF */
        if (*s < 0xC2) {
            /* invalid sequence
               \x80-\xBF -- continuation byte
               \xC0-\xC1 -- fake 0000-007F */
            return 0;
        }
        expected = 1;
    }
    else if (*s < 0xF0) {
        /* \xE0\xA0\x80-\xEF\xBF\xBF -- 0800-FFFF */
        if (*s == 0xE0 && *(s + 1) < 0xA0) {
            /* invalid sequence
               \xE0\x80\x80-\xE0\x9F\xBF -- fake 0000-0800 */
            return 0;
        }
        else if (*s == 0xED && *(s + 1) >= 0xA0) {
            /* Decoding UTF-8 sequences in range \xED\xA0\x80-\xED\xBF\xBF
               will result in surrogates in range D800-DFFF. Surrogates are
               not valid UTF-8 so they are rejected.
               See https://www.unicode.org/versions/Unicode5.2.0/ch03.pdf
               (table 3-7) and http://www.rfc-editor.org/rfc/rfc3629.txt */
            return 0;
        }
        expected = 2;
    }
    else if (*s < 0xF5) {
        /* \xF0\x90\x80\x80-\xF4\x8F\xBF\xBF -- 10000-10FFFF */
        if (*(s + 1) < 0x90 ? *s == 0xF0 : *s == 0xF4) {
            /* invalid sequence -- one of:
               \xF0\x80\x80\x80-\xF0\x8F\xBF\xBF -- fake 0000-FFFF
               \xF4\x90\x80\x80- -- 110000- overflow */
            return 0;
        }
        expected = 3;
    }
    else {
        /* invalid start byte */
        return 0;
    }
    length = expected + 1;
    for (; expected; expected--)
        if (s[expected] < 0x80 || s[expected] >= 0xC0)
            return 0;
    return length;
}

static int
ensure_utf8(char *line, struct tok_state *tok)
{
    int badchar = 0;
    unsigned char *c;
    int length;
    for (c = (unsigned char *)line; *c; c += length) {
        if (!(length = valid_utf8(c))) {
            badchar = *c;
            break;
        }
    }
    if (badchar) {
        PyErr_Format(PyExc_SyntaxError,
                     "Non-UTF-8 code starting with '\\x%.2x' "
                     "in file %U on line %i, "
                     "but no encoding declared; "
                     "see https://peps.python.org/pep-0263/ for details",
                     badchar, tok->filename, tok->lineno);
        return 0;
    }
    return 1;
}

/* Fetch a byte from TOK, using the string buffer. */

static int
buf_getc(struct tok_state *tok) {
    return Py_CHARMASK(*tok->str++);
}

/* Unfetch a byte from TOK, using the string buffer. */

static void
buf_ungetc(int c, struct tok_state *tok) {
    tok->str--;
    assert(Py_CHARMASK(*tok->str) == c);        /* tok->cur may point to read-only segment */
}

/* Set the readline function for TOK to ENC. For the string-based
   tokenizer, this means to just record the encoding. */

static int
buf_setreadl(struct tok_state *tok, const char* enc) {
    tok->enc = enc;
    return 1;
}

/* Return a UTF-8 encoding Python string object from the
   C byte string STR, which is encoded with ENC. */

static PyObject *
translate_into_utf8(const char* str, const char* enc) {
    PyObject *utf8;
    PyObject* buf = PyUnicode_Decode(str, strlen(str), enc, NULL);
    if (buf == NULL)
        return NULL;
    utf8 = PyUnicode_AsUTF8String(buf);
    Py_DECREF(buf);
    return utf8;
}


static char *
translate_newlines(const char *s, int exec_input, int preserve_crlf,
                   struct tok_state *tok) {
    int skip_next_lf = 0;
    size_t needed_length = strlen(s) + 2, final_length;
    char *buf, *current;
    char c = '\0';
    buf = PyMem_Malloc(needed_length);
    if (buf == NULL) {
        tok->done = E_NOMEM;
        return NULL;
    }
    for (current = buf; *s; s++, current++) {
        c = *s;
        if (skip_next_lf) {
            skip_next_lf = 0;
            if (c == '\n') {
                c = *++s;
                if (!c)
                    break;
            }
        }
        if (!preserve_crlf && c == '\r') {
            skip_next_lf = 1;
            c = '\n';
        }
        *current = c;
    }
    /* If this is exec input, add a newline to the end of the string if
       there isn't one already. */
    if (exec_input && c != '\n' && c != '\0') {
        *current = '\n';
        current++;
    }
    *current = '\0';
    final_length = current - buf + 1;
    if (final_length < needed_length && final_length) {
        /* should never fail */
        char* result = PyMem_Realloc(buf, final_length);
        if (result == NULL) {
            PyMem_Free(buf);
        }
        buf = result;
    }
    return buf;
}

/* Decode a byte string STR for use as the buffer of TOK.
   Look for encoding declarations inside STR, and record them
   inside TOK.  */

static char *
decode_str(const char *input, int single, struct tok_state *tok, int preserve_crlf)
{
    PyObject* utf8 = NULL;
    char *str;
    const char *s;
    const char *newl[2] = {NULL, NULL};
    int lineno = 0;
    tok->input = str = translate_newlines(input, single, preserve_crlf, tok);
    if (str == NULL)
        return NULL;
    tok->enc = NULL;
    tok->str = str;
    if (!check_bom(buf_getc, buf_ungetc, buf_setreadl, tok))
        return error_ret(tok);
    str = tok->str;             /* string after BOM if any */
    assert(str);
    if (tok->enc != NULL) {
        utf8 = translate_into_utf8(str, tok->enc);
        if (utf8 == NULL)
            return error_ret(tok);
        str = PyBytes_AsString(utf8);
    }
    for (s = str;; s++) {
        if (*s == '\0') break;
        else if (*s == '\n') {
            assert(lineno < 2);
            newl[lineno] = s;
            lineno++;
            if (lineno == 2) break;
        }
    }
    tok->enc = NULL;
    /* need to check line 1 and 2 separately since check_coding_spec
       assumes a single line as input */
    if (newl[0]) {
        if (!check_coding_spec(str, newl[0] - str, tok, buf_setreadl)) {
            return NULL;
        }
        if (tok->enc == NULL && tok->decoding_state != STATE_NORMAL && newl[1]) {
            if (!check_coding_spec(newl[0]+1, newl[1] - newl[0],
                                   tok, buf_setreadl))
                return NULL;
        }
    }
    if (tok->enc != NULL) {
        assert(utf8 == NULL);
        utf8 = translate_into_utf8(str, tok->enc);
        if (utf8 == NULL)
            return error_ret(tok);
        str = PyBytes_AS_STRING(utf8);
    }
    assert(tok->decoding_buffer == NULL);
    tok->decoding_buffer = utf8; /* CAUTION */
    return str;
}

/* Set up tokenizer for string */

struct tok_state *
_PyTokenizer_FromString(const char *str, int exec_input, int preserve_crlf)
{
    struct tok_state *tok = tok_new();
    char *decoded;

    if (tok == NULL)
        return NULL;
    decoded = decode_str(str, exec_input, tok, preserve_crlf);
    if (decoded == NULL) {
        _PyTokenizer_Free(tok);
        return NULL;
    }

    tok->buf = tok->cur = tok->inp = decoded;
    tok->end = decoded;
    return tok;
}

struct tok_state *
_PyTokenizer_FromReadline(PyObject* readline, const char* enc,
                          int exec_input, int preserve_crlf)
{
    struct tok_state *tok = tok_new();
    if (tok == NULL)
        return NULL;
    if ((tok->buf = (char *)PyMem_Malloc(BUFSIZ)) == NULL) {
        _PyTokenizer_Free(tok);
        return NULL;
    }
    tok->cur = tok->inp = tok->buf;
    tok->end = tok->buf + BUFSIZ;
    tok->fp = NULL;
    if (enc != NULL) {
        tok->encoding = new_string(enc, strlen(enc), tok);
        if (!tok->encoding) {
            _PyTokenizer_Free(tok);
            return NULL;
        }
    }
    tok->decoding_state = STATE_NORMAL;
    Py_INCREF(readline);
    tok->readline = readline;
    return tok;
}

/* Set up tokenizer for UTF-8 string */

struct tok_state *
_PyTokenizer_FromUTF8(const char *str, int exec_input, int preserve_crlf)
{
    struct tok_state *tok = tok_new();
    char *translated;
    if (tok == NULL)
        return NULL;
    tok->input = translated = translate_newlines(str, exec_input, preserve_crlf, tok);
    if (translated == NULL) {
        _PyTokenizer_Free(tok);
        return NULL;
    }
    tok->decoding_state = STATE_NORMAL;
    tok->enc = NULL;
    tok->str = translated;
    tok->encoding = new_string("utf-8", 5, tok);
    if (!tok->encoding) {
        _PyTokenizer_Free(tok);
        return NULL;
    }

    tok->buf = tok->cur = tok->inp = translated;
    tok->end = translated;
    return tok;
}

/* Set up tokenizer for file */

struct tok_state *
_PyTokenizer_FromFile(FILE *fp, const char* enc,
                      const char *ps1, const char *ps2)
{
    struct tok_state *tok = tok_new();
    if (tok == NULL)
        return NULL;
    if ((tok->buf = (char *)PyMem_Malloc(BUFSIZ)) == NULL) {
        _PyTokenizer_Free(tok);
        return NULL;
    }
    tok->cur = tok->inp = tok->buf;
    tok->end = tok->buf + BUFSIZ;
    tok->fp = fp;
    tok->prompt = ps1;
    tok->nextprompt = ps2;
    if (enc != NULL) {
        /* Must copy encoding declaration since it
           gets copied into the parse tree. */
        tok->encoding = new_string(enc, strlen(enc), tok);
        if (!tok->encoding) {
            _PyTokenizer_Free(tok);
            return NULL;
        }
        tok->decoding_state = STATE_NORMAL;
    }
    return tok;
}

/* Free a tok_state structure */

void
_PyTokenizer_Free(struct tok_state *tok)
{
    if (tok->encoding != NULL) {
        PyMem_Free(tok->encoding);
    }
    Py_XDECREF(tok->decoding_readline);
    Py_XDECREF(tok->decoding_buffer);
    Py_XDECREF(tok->readline);
    Py_XDECREF(tok->filename);
    if ((tok->readline != NULL || tok->fp != NULL ) && tok->buf != NULL) {
        PyMem_Free(tok->buf);
    }
    if (tok->input) {
        PyMem_Free(tok->input);
    }
    if (tok->interactive_src_start != NULL) {
        PyMem_Free(tok->interactive_src_start);
    }
    free_fstring_expressions(tok);
    PyMem_Free(tok);
}

void
_PyToken_Free(struct token *token) {
    Py_XDECREF(token->metadata);
}

void
_PyToken_Init(struct token *token) {
    token->metadata = NULL;
}

static int
tok_readline_raw(struct tok_state *tok)
{
    do {
        if (!tok_reserve_buf(tok, BUFSIZ)) {
            return 0;
        }
        int n_chars = (int)(tok->end - tok->inp);
        size_t line_size = 0;
        char *line = _Py_UniversalNewlineFgetsWithSize(tok->inp, n_chars, tok->fp, NULL, &line_size);
        if (line == NULL) {
            return 1;
        }
        if (tok->fp_interactive &&
            tok_concatenate_interactive_new_line(tok, line) == -1) {
            return 0;
        }
        tok->inp += line_size;
        if (tok->inp == tok->buf) {
            return 0;
        }
    } while (tok->inp[-1] != '\n');
    return 1;
}

static int
tok_readline_string(struct tok_state* tok) {
    PyObject* line = NULL;
    PyObject* raw_line = PyObject_CallNoArgs(tok->readline);
    if (raw_line == NULL) {
        if (PyErr_ExceptionMatches(PyExc_StopIteration)) {
            PyErr_Clear();
            return 1;
        }
        error_ret(tok);
        goto error;
    }
    if(tok->encoding != NULL) {
        if (!PyBytes_Check(raw_line)) {
            PyErr_Format(PyExc_TypeError, "readline() returned a non-bytes object");
            error_ret(tok);
            goto error;
        }
        line = PyUnicode_Decode(PyBytes_AS_STRING(raw_line), PyBytes_GET_SIZE(raw_line),
                                tok->encoding, "replace");
        Py_CLEAR(raw_line);
        if (line == NULL) {
            error_ret(tok);
            goto error;
        }
    } else {
        if(!PyUnicode_Check(raw_line)) {
            PyErr_Format(PyExc_TypeError, "readline() returned a non-string object");
            error_ret(tok);
            goto error;
        }
        line = raw_line;
        raw_line = NULL;
    }
    Py_ssize_t buflen;
    const char* buf = PyUnicode_AsUTF8AndSize(line, &buflen);
    if (buf == NULL) {
        error_ret(tok);
        goto error;
    }

    // Make room for the null terminator *and* potentially
    // an extra newline character that we may need to artificially
    // add.
    size_t buffer_size = buflen + 2;
    if (!tok_reserve_buf(tok, buffer_size)) {
        goto error;
    }
    memcpy(tok->inp, buf, buflen);
    tok->inp += buflen;
    *tok->inp = '\0';

    tok->line_start = tok->cur;
    Py_DECREF(line);
    return 1;
error:
    Py_XDECREF(raw_line);
    Py_XDECREF(line);
    return 0;
}

static int
tok_underflow_string(struct tok_state *tok) {
    char *end = strchr(tok->inp, '\n');
    if (end != NULL) {
        end++;
    }
    else {
        end = strchr(tok->inp, '\0');
        if (end == tok->inp) {
            tok->done = E_EOF;
            return 0;
        }
    }
    if (tok->start == NULL) {
        tok->buf = tok->cur;
    }
    tok->line_start = tok->cur;
    ADVANCE_LINENO();
    tok->inp = end;
    return 1;
}

static int
tok_underflow_interactive(struct tok_state *tok) {
    if (tok->interactive_underflow == IUNDERFLOW_STOP) {
        tok->done = E_INTERACT_STOP;
        return 1;
    }
    char *newtok = PyOS_Readline(tok->fp ? tok->fp : stdin, stdout, tok->prompt);
    if (newtok != NULL) {
        char *translated = translate_newlines(newtok, 0, 0, tok);
        PyMem_Free(newtok);
        if (translated == NULL) {
            return 0;
        }
        newtok = translated;
    }
    if (tok->encoding && newtok && *newtok) {
        /* Recode to UTF-8 */
        Py_ssize_t buflen;
        const char* buf;
        PyObject *u = translate_into_utf8(newtok, tok->encoding);
        PyMem_Free(newtok);
        if (u == NULL) {
            tok->done = E_DECODE;
            return 0;
        }
        buflen = PyBytes_GET_SIZE(u);
        buf = PyBytes_AS_STRING(u);
        newtok = PyMem_Malloc(buflen+1);
        if (newtok == NULL) {
            Py_DECREF(u);
            tok->done = E_NOMEM;
            return 0;
        }
        strcpy(newtok, buf);
        Py_DECREF(u);
    }
    if (tok->fp_interactive &&
        tok_concatenate_interactive_new_line(tok, newtok) == -1) {
        PyMem_Free(newtok);
        return 0;
    }
    if (tok->nextprompt != NULL) {
        tok->prompt = tok->nextprompt;
    }
    if (newtok == NULL) {
        tok->done = E_INTR;
    }
    else if (*newtok == '\0') {
        PyMem_Free(newtok);
        tok->done = E_EOF;
    }
    else if (tok->start != NULL) {
        Py_ssize_t cur_multi_line_start = tok->multi_line_start - tok->buf;
        remember_fstring_buffers(tok);
        size_t size = strlen(newtok);
        ADVANCE_LINENO();
        if (!tok_reserve_buf(tok, size + 1)) {
            PyMem_Free(tok->buf);
            tok->buf = NULL;
            PyMem_Free(newtok);
            return 0;
        }
        memcpy(tok->cur, newtok, size + 1);
        PyMem_Free(newtok);
        tok->inp += size;
        tok->multi_line_start = tok->buf + cur_multi_line_start;
        restore_fstring_buffers(tok);
    }
    else {
        remember_fstring_buffers(tok);
        ADVANCE_LINENO();
        PyMem_Free(tok->buf);
        tok->buf = newtok;
        tok->cur = tok->buf;
        tok->line_start = tok->buf;
        tok->inp = strchr(tok->buf, '\0');
        tok->end = tok->inp + 1;
        restore_fstring_buffers(tok);
    }
    if (tok->done != E_OK) {
        if (tok->prompt != NULL) {
            PySys_WriteStderr("\n");
        }
        return 0;
    }

    if (tok->tok_mode_stack_index && !update_fstring_expr(tok, 0)) {
        return 0;
    }
    return 1;
}

static int
tok_underflow_file(struct tok_state *tok) {
    if (tok->start == NULL && !INSIDE_FSTRING(tok)) {
        tok->cur = tok->inp = tok->buf;
    }
    if (tok->decoding_state == STATE_INIT) {
        /* We have not yet determined the encoding.
           If an encoding is found, use the file-pointer
           reader functions from now on. */
        if (!check_bom(fp_getc, fp_ungetc, fp_setreadl, tok)) {
            error_ret(tok);
            return 0;
        }
        assert(tok->decoding_state != STATE_INIT);
    }
    /* Read until '\n' or EOF */
    if (tok->decoding_readline != NULL) {
        /* We already have a codec associated with this input. */
        if (!tok_readline_recode(tok)) {
            return 0;
        }
    }
    else {
        /* We want a 'raw' read. */
        if (!tok_readline_raw(tok)) {
            return 0;
        }
    }
    if (tok->inp == tok->cur) {
        tok->done = E_EOF;
        return 0;
    }
    tok->implicit_newline = 0;
    if (tok->inp[-1] != '\n') {
        assert(tok->inp + 1 < tok->end);
        /* Last line does not end in \n, fake one */
        *tok->inp++ = '\n';
        *tok->inp = '\0';
        tok->implicit_newline = 1;
    }

    if (tok->tok_mode_stack_index && !update_fstring_expr(tok, 0)) {
        return 0;
    }

    ADVANCE_LINENO();
    if (tok->decoding_state != STATE_NORMAL) {
        if (tok->lineno > 2) {
            tok->decoding_state = STATE_NORMAL;
        }
        else if (!check_coding_spec(tok->cur, strlen(tok->cur),
                                    tok, fp_setreadl))
        {
            return 0;
        }
    }
    /* The default encoding is UTF-8, so make sure we don't have any
       non-UTF-8 sequences in it. */
    if (!tok->encoding && !ensure_utf8(tok->cur, tok)) {
        error_ret(tok);
        return 0;
    }
    assert(tok->done == E_OK);
    return tok->done == E_OK;
}

static int
tok_underflow_readline(struct tok_state* tok) {
    assert(tok->decoding_state == STATE_NORMAL);
    assert(tok->fp == NULL && tok->input == NULL && tok->decoding_readline == NULL);
    if (tok->start == NULL && !INSIDE_FSTRING(tok)) {
        tok->cur = tok->inp = tok->buf;
    }
    if (!tok_readline_string(tok)) {
        return 0;
    }
    if (tok->inp == tok->cur) {
        tok->done = E_EOF;
        return 0;
    }
    tok->implicit_newline = 0;
    if (tok->inp[-1] != '\n') {
        assert(tok->inp + 1 < tok->end);
        /* Last line does not end in \n, fake one */
        *tok->inp++ = '\n';
        *tok->inp = '\0';
        tok->implicit_newline = 1;
    }

    if (tok->tok_mode_stack_index && !update_fstring_expr(tok, 0)) {
        return 0;
    }

    ADVANCE_LINENO();
    /* The default encoding is UTF-8, so make sure we don't have any
       non-UTF-8 sequences in it. */
    if (!tok->encoding && !ensure_utf8(tok->cur, tok)) {
        error_ret(tok);
        return 0;
    }
    assert(tok->done == E_OK);
    return tok->done == E_OK;
}

#if defined(Py_DEBUG)
static void
print_escape(FILE *f, const char *s, Py_ssize_t size)
{
    if (s == NULL) {
        fputs("NULL", f);
        return;
    }
    putc('"', f);
    while (size-- > 0) {
        unsigned char c = *s++;
        switch (c) {
            case '\n': fputs("\\n", f); break;
            case '\r': fputs("\\r", f); break;
            case '\t': fputs("\\t", f); break;
            case '\f': fputs("\\f", f); break;
            case '\'': fputs("\\'", f); break;
            case '"': fputs("\\\"", f); break;
            default:
                if (0x20 <= c && c <= 0x7f)
                    putc(c, f);
                else
                    fprintf(f, "\\x%02x", c);
        }
    }
    putc('"', f);
}
#endif

/* Get next char, updating state; error code goes into tok->done */

static int
tok_nextc(struct tok_state *tok)
{
    int rc;
    for (;;) {
        if (tok->cur != tok->inp) {
            tok->col_offset++;
            return Py_CHARMASK(*tok->cur++); /* Fast path */
        }
        if (tok->done != E_OK) {
            return EOF;
        }
        if (tok->readline) {
            rc = tok_underflow_readline(tok);
        }
        else if (tok->fp == NULL) {
            rc = tok_underflow_string(tok);
        }
        else if (tok->prompt != NULL) {
            rc = tok_underflow_interactive(tok);
        }
        else {
            rc = tok_underflow_file(tok);
        }
#if defined(Py_DEBUG)
        if (tok->debug) {
            fprintf(stderr, "line[%d] = ", tok->lineno);
            print_escape(stderr, tok->cur, tok->inp - tok->cur);
            fprintf(stderr, "  tok->done = %d\n", tok->done);
        }
#endif
        if (!rc) {
            tok->cur = tok->inp;
            return EOF;
        }
        tok->line_start = tok->cur;

        if (contains_null_bytes(tok->line_start, tok->inp - tok->line_start)) {
            syntaxerror(tok, "source code cannot contain null bytes");
            tok->cur = tok->inp;
            return EOF;
        }
    }
    Py_UNREACHABLE();
}

/* Back-up one character */

static void
tok_backup(struct tok_state *tok, int c)
{
    if (c != EOF) {
        if (--tok->cur < tok->buf) {
            Py_FatalError("tokenizer beginning of buffer");
        }
        if ((int)(unsigned char)*tok->cur != Py_CHARMASK(c)) {
            Py_FatalError("tok_backup: wrong character");
        }
        tok->col_offset--;
    }
}

static int
_syntaxerror_range(struct tok_state *tok, const char *format,
                   int col_offset, int end_col_offset,
                   va_list vargs)
{
    // In release builds, we don't want to overwrite a previous error, but in debug builds we
    // want to fail if we are not doing it so we can fix it.
    assert(tok->done != E_ERROR);
    if (tok->done == E_ERROR) {
        return ERRORTOKEN;
    }
    PyObject *errmsg, *errtext, *args;
    errmsg = PyUnicode_FromFormatV(format, vargs);
    if (!errmsg) {
        goto error;
    }

    errtext = PyUnicode_DecodeUTF8(tok->line_start, tok->cur - tok->line_start,
                                   "replace");
    if (!errtext) {
        goto error;
    }

    if (col_offset == -1) {
        col_offset = (int)PyUnicode_GET_LENGTH(errtext);
    }
    if (end_col_offset == -1) {
        end_col_offset = col_offset;
    }

    Py_ssize_t line_len = strcspn(tok->line_start, "\n");
    if (line_len != tok->cur - tok->line_start) {
        Py_DECREF(errtext);
        errtext = PyUnicode_DecodeUTF8(tok->line_start, line_len,
                                       "replace");
    }
    if (!errtext) {
        goto error;
    }

    args = Py_BuildValue("(O(OiiNii))", errmsg, tok->filename, tok->lineno,
                         col_offset, errtext, tok->lineno, end_col_offset);
    if (args) {
        PyErr_SetObject(PyExc_SyntaxError, args);
        Py_DECREF(args);
    }

error:
    Py_XDECREF(errmsg);
    tok->done = E_ERROR;
    return ERRORTOKEN;
}

static int
syntaxerror(struct tok_state *tok, const char *format, ...)
{
    // This errors are cleaned on startup. Todo: Fix it.
    va_list vargs;
    va_start(vargs, format);
    int ret = _syntaxerror_range(tok, format, -1, -1, vargs);
    va_end(vargs);
    return ret;
}

static int
syntaxerror_known_range(struct tok_state *tok,
                        int col_offset, int end_col_offset,
                        const char *format, ...)
{
    va_list vargs;
    va_start(vargs, format);
    int ret = _syntaxerror_range(tok, format, col_offset, end_col_offset, vargs);
    va_end(vargs);
    return ret;
}

static int
indenterror(struct tok_state *tok)
{
    tok->done = E_TABSPACE;
    tok->cur = tok->inp;
    return ERRORTOKEN;
}

static int
parser_warn(struct tok_state *tok, PyObject *category, const char *format, ...)
{
    if (!tok->report_warnings) {
        return 0;
    }

    PyObject *errmsg;
    va_list vargs;
    va_start(vargs, format);
    errmsg = PyUnicode_FromFormatV(format, vargs);
    va_end(vargs);
    if (!errmsg) {
        goto error;
    }

    if (PyErr_WarnExplicitObject(category, errmsg, tok->filename,
                                 tok->lineno, NULL, NULL) < 0) {
        if (PyErr_ExceptionMatches(category)) {
            /* Replace the DeprecationWarning exception with a SyntaxError
               to get a more accurate error report */
            PyErr_Clear();
            syntaxerror(tok, "%U", errmsg);
        }
        goto error;
    }
    Py_DECREF(errmsg);
    return 0;

error:
    Py_XDECREF(errmsg);
    tok->done = E_ERROR;
    return -1;
}

static int
warn_invalid_escape_sequence(struct tok_state *tok, int first_invalid_escape_char)
{
    if (!tok->report_warnings) {
        return 0;
    }

    PyObject *msg = PyUnicode_FromFormat(
        "invalid escape sequence '\\%c'",
        (char) first_invalid_escape_char
    );

    if (msg == NULL) {
        return -1;
    }

    if (PyErr_WarnExplicitObject(PyExc_SyntaxWarning, msg, tok->filename,
                                 tok->lineno, NULL, NULL) < 0) {
        Py_DECREF(msg);

        if (PyErr_ExceptionMatches(PyExc_SyntaxWarning)) {
            /* Replace the SyntaxWarning exception with a SyntaxError
               to get a more accurate error report */
            PyErr_Clear();
            return syntaxerror(tok, "invalid escape sequence '\\%c'", (char) first_invalid_escape_char);
        }

        return -1;
    }

    Py_DECREF(msg);
    return 0;
}

static int
lookahead(struct tok_state *tok, const char *test)
{
    const char *s = test;
    int res = 0;
    while (1) {
        int c = tok_nextc(tok);
        if (*s == 0) {
            res = !is_potential_identifier_char(c);
        }
        else if (c == *s) {
            s++;
            continue;
        }

        tok_backup(tok, c);
        while (s != test) {
            tok_backup(tok, *--s);
        }
        return res;
    }
}

static int
verify_end_of_number(struct tok_state *tok, int c, const char *kind) {
    if (tok->tok_extra_tokens) {
        // When we are parsing extra tokens, we don't want to emit warnings
        // about invalid literals, because we want to be a bit more liberal.
        return 1;
    }
    /* Emit a deprecation warning only if the numeric literal is immediately
     * followed by one of keywords which can occur after a numeric literal
     * in valid code: "and", "else", "for", "if", "in", "is" and "or".
     * It allows to gradually deprecate existing valid code without adding
     * warning before error in most cases of invalid numeric literal (which
     * would be confusing and break existing tests).
     * Raise a syntax error with slightly better message than plain
     * "invalid syntax" if the numeric literal is immediately followed by
     * other keyword or identifier.
     */
    int r = 0;
    if (c == 'a') {
        r = lookahead(tok, "nd");
    }
    else if (c == 'e') {
        r = lookahead(tok, "lse");
    }
    else if (c == 'f') {
        r = lookahead(tok, "or");
    }
    else if (c == 'i') {
        int c2 = tok_nextc(tok);
        if (c2 == 'f' || c2 == 'n' || c2 == 's') {
            r = 1;
        }
        tok_backup(tok, c2);
    }
    else if (c == 'o') {
        r = lookahead(tok, "r");
    }
    else if (c == 'n') {
        r = lookahead(tok, "ot");
    }
    if (r) {
        tok_backup(tok, c);
        if (parser_warn(tok, PyExc_SyntaxWarning,
                "invalid %s literal", kind))
        {
            return 0;
        }
        tok_nextc(tok);
    }
    else /* In future releases, only error will remain. */
    if (c < 128 && is_potential_identifier_char(c)) {
        tok_backup(tok, c);
        syntaxerror(tok, "invalid %s literal", kind);
        return 0;
    }
    return 1;
}

/* Verify that the identifier follows PEP 3131.
   All identifier strings are guaranteed to be "ready" unicode objects.
 */
static int
verify_identifier(struct tok_state *tok)
{
    if (tok->tok_extra_tokens) {
        return 1;
    }
    PyObject *s;
    if (tok->decoding_erred)
        return 0;
    s = PyUnicode_DecodeUTF8(tok->start, tok->cur - tok->start, NULL);
    if (s == NULL) {
        if (PyErr_ExceptionMatches(PyExc_UnicodeDecodeError)) {
            tok->done = E_DECODE;
        }
        else {
            tok->done = E_ERROR;
        }
        return 0;
    }
    Py_ssize_t invalid = _PyUnicode_ScanIdentifier(s);
    if (invalid < 0) {
        Py_DECREF(s);
        tok->done = E_ERROR;
        return 0;
    }
    assert(PyUnicode_GET_LENGTH(s) > 0);
    if (invalid < PyUnicode_GET_LENGTH(s)) {
        Py_UCS4 ch = PyUnicode_READ_CHAR(s, invalid);
        if (invalid + 1 < PyUnicode_GET_LENGTH(s)) {
            /* Determine the offset in UTF-8 encoded input */
            Py_SETREF(s, PyUnicode_Substring(s, 0, invalid + 1));
            if (s != NULL) {
                Py_SETREF(s, PyUnicode_AsUTF8String(s));
            }
            if (s == NULL) {
                tok->done = E_ERROR;
                return 0;
            }
            tok->cur = (char *)tok->start + PyBytes_GET_SIZE(s);
        }
        Py_DECREF(s);
        if (Py_UNICODE_ISPRINTABLE(ch)) {
            syntaxerror(tok, "invalid character '%c' (U+%04X)", ch, ch);
        }
        else {
            syntaxerror(tok, "invalid non-printable character U+%04X", ch);
        }
        return 0;
    }
    Py_DECREF(s);
    return 1;
}

static int
tok_decimal_tail(struct tok_state *tok)
{
    int c;

    while (1) {
        do {
            c = tok_nextc(tok);
        } while (Py_ISDIGIT(c));
        if (c != '_') {
            break;
        }
        c = tok_nextc(tok);
        if (!Py_ISDIGIT(c)) {
            tok_backup(tok, c);
            syntaxerror(tok, "invalid decimal literal");
            return 0;
        }
    }
    return c;
}


static inline int
tok_continuation_line(struct tok_state *tok) {
    int c = tok_nextc(tok);
    if (c == '\r') {
        c = tok_nextc(tok);
    }
    if (c != '\n') {
        tok->done = E_LINECONT;
        return -1;
    }
    c = tok_nextc(tok);
    if (c == EOF) {
        tok->done = E_EOF;
        tok->cur = tok->inp;
        return -1;
    } else {
        tok_backup(tok, c);
    }
    return c;
}

static int
type_comment_token_setup(struct tok_state *tok, struct token *token, int type, int col_offset,
                         int end_col_offset, const char *start, const char *end)
{
    token->level = tok->level;
    token->lineno = token->end_lineno = tok->lineno;
    token->col_offset = col_offset;
    token->end_col_offset = end_col_offset;
    token->start = start;
    token->end = end;
    return type;
}

static int
token_setup(struct tok_state *tok, struct token *token, int type, const char *start, const char *end)
{
    assert((start == NULL && end == NULL) || (start != NULL && end != NULL));
    token->level = tok->level;
    if (ISSTRINGLIT(type)) {
        token->lineno = tok->first_lineno;
    }
    else {
        token->lineno = tok->lineno;
    }
    token->end_lineno = tok->lineno;
    token->col_offset = token->end_col_offset = -1;
    token->start = start;
    token->end = end;

    if (start != NULL && end != NULL) {
        token->col_offset = tok->starting_col_offset;
        token->end_col_offset = tok->col_offset;
    }
    return type;
}


static int
tok_get_normal_mode(struct tok_state *tok, tokenizer_mode* current_tok, struct token *token)
{
    int c;
    int blankline, nonascii;

    const char *p_start = NULL;
    const char *p_end = NULL;
  nextline:
    tok->start = NULL;
    tok->starting_col_offset = -1;
    blankline = 0;


    /* Get indentation level */
    if (tok->atbol) {
        int col = 0;
        int altcol = 0;
        tok->atbol = 0;
        int cont_line_col = 0;
        for (;;) {
            c = tok_nextc(tok);
            if (c == ' ') {
                col++, altcol++;
            }
            else if (c == '\t') {
                col = (col / tok->tabsize + 1) * tok->tabsize;
                altcol = (altcol / ALTTABSIZE + 1) * ALTTABSIZE;
            }
            else if (c == '\014')  {/* Control-L (formfeed) */
                col = altcol = 0; /* For Emacs users */
            }
            else if (c == '\\') {
                // Indentation cannot be split over multiple physical lines
                // using backslashes. This means that if we found a backslash
                // preceded by whitespace, **the first one we find** determines
                // the level of indentation of whatever comes next.
                cont_line_col = cont_line_col ? cont_line_col : col;
                if ((c = tok_continuation_line(tok)) == -1) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
            else {
                break;
            }
        }
        tok_backup(tok, c);
        if (c == '#' || c == '\n' || c == '\r') {
            /* Lines with only whitespace and/or comments
               shouldn't affect the indentation and are
               not passed to the parser as NEWLINE tokens,
               except *totally* empty lines in interactive
               mode, which signal the end of a command group. */
            if (col == 0 && c == '\n' && tok->prompt != NULL) {
                blankline = 0; /* Let it through */
            }
            else if (tok->prompt != NULL && tok->lineno == 1) {
                /* In interactive mode, if the first line contains
                   only spaces and/or a comment, let it through. */
                blankline = 0;
                col = altcol = 0;
            }
            else {
                blankline = 1; /* Ignore completely */
            }
            /* We can't jump back right here since we still
               may need to skip to the end of a comment */
        }
        if (!blankline && tok->level == 0) {
            col = cont_line_col ? cont_line_col : col;
            altcol = cont_line_col ? cont_line_col : altcol;
            if (col == tok->indstack[tok->indent]) {
                /* No change */
                if (altcol != tok->altindstack[tok->indent]) {
                    return MAKE_TOKEN(indenterror(tok));
                }
            }
            else if (col > tok->indstack[tok->indent]) {
                /* Indent -- always one */
                if (tok->indent+1 >= MAXINDENT) {
                    tok->done = E_TOODEEP;
                    tok->cur = tok->inp;
                    return MAKE_TOKEN(ERRORTOKEN);
                }
                if (altcol <= tok->altindstack[tok->indent]) {
                    return MAKE_TOKEN(indenterror(tok));
                }
                tok->pendin++;
                tok->indstack[++tok->indent] = col;
                tok->altindstack[tok->indent] = altcol;
            }
            else /* col < tok->indstack[tok->indent] */ {
                /* Dedent -- any number, must be consistent */
                while (tok->indent > 0 &&
                    col < tok->indstack[tok->indent]) {
                    tok->pendin--;
                    tok->indent--;
                }
                if (col != tok->indstack[tok->indent]) {
                    tok->done = E_DEDENT;
                    tok->cur = tok->inp;
                    return MAKE_TOKEN(ERRORTOKEN);
                }
                if (altcol != tok->altindstack[tok->indent]) {
                    return MAKE_TOKEN(indenterror(tok));
                }
            }
        }
    }

    tok->start = tok->cur;
    tok->starting_col_offset = tok->col_offset;

    /* Return pending indents/dedents */
    if (tok->pendin != 0) {
        if (tok->pendin < 0) {
            if (tok->tok_extra_tokens) {
                p_start = tok->cur;
                p_end = tok->cur;
            }
            tok->pendin++;
            return MAKE_TOKEN(DEDENT);
        }
        else {
            if (tok->tok_extra_tokens) {
                p_start = tok->buf;
                p_end = tok->cur;
            }
            tok->pendin--;
            return MAKE_TOKEN(INDENT);
        }
    }

    /* Peek ahead at the next character */
    c = tok_nextc(tok);
    tok_backup(tok, c);

 again:
    tok->start = NULL;
    /* Skip spaces */
    do {
        c = tok_nextc(tok);
    } while (c == ' ' || c == '\t' || c == '\014');

    /* Set start of current token */
    tok->start = tok->cur == NULL ? NULL : tok->cur - 1;
    tok->starting_col_offset = tok->col_offset - 1;

    /* Skip comment, unless it's a type comment */
    if (c == '#') {

        const char* p = NULL;
        const char *prefix, *type_start;
        int current_starting_col_offset;

        while (c != EOF && c != '\n' && c != '\r') {
            c = tok_nextc(tok);
        }

        if (tok->tok_extra_tokens) {
            p = tok->start;
        }

        if (tok->type_comments) {
            p = tok->start;
            current_starting_col_offset = tok->starting_col_offset;
            prefix = type_comment_prefix;
            while (*prefix && p < tok->cur) {
                if (*prefix == ' ') {
                    while (*p == ' ' || *p == '\t') {
                        p++;
                        current_starting_col_offset++;
                    }
                } else if (*prefix == *p) {
                    p++;
                    current_starting_col_offset++;
                } else {
                    break;
                }

                prefix++;
            }

            /* This is a type comment if we matched all of type_comment_prefix. */
            if (!*prefix) {
                int is_type_ignore = 1;
                // +6 in order to skip the word 'ignore'
                const char *ignore_end = p + 6;
                const int ignore_end_col_offset = current_starting_col_offset + 6;
                tok_backup(tok, c);  /* don't eat the newline or EOF */

                type_start = p;

                /* A TYPE_IGNORE is "type: ignore" followed by the end of the token
                 * or anything ASCII and non-alphanumeric. */
                is_type_ignore = (
                    tok->cur >= ignore_end && memcmp(p, "ignore", 6) == 0
                    && !(tok->cur > ignore_end
                         && ((unsigned char)ignore_end[0] >= 128 || Py_ISALNUM(ignore_end[0]))));

                if (is_type_ignore) {
                    p_start = ignore_end;
                    p_end = tok->cur;

                    /* If this type ignore is the only thing on the line, consume the newline also. */
                    if (blankline) {
                        tok_nextc(tok);
                        tok->atbol = 1;
                    }
                    return MAKE_TYPE_COMMENT_TOKEN(TYPE_IGNORE, ignore_end_col_offset, tok->col_offset);
                } else {
                    p_start = type_start;
                    p_end = tok->cur;
                    return MAKE_TYPE_COMMENT_TOKEN(TYPE_COMMENT, current_starting_col_offset, tok->col_offset);
                }
            }
        }
        if (tok->tok_extra_tokens) {
            tok_backup(tok, c);  /* don't eat the newline or EOF */
            p_start = p;
            p_end = tok->cur;
            tok->comment_newline = blankline;
            return MAKE_TOKEN(COMMENT);
        }
    }

    if (tok->done == E_INTERACT_STOP) {
        return MAKE_TOKEN(ENDMARKER);
    }

    /* Check for EOF and errors now */
    if (c == EOF) {
        if (tok->level) {
            return MAKE_TOKEN(ERRORTOKEN);
        }
        return MAKE_TOKEN(tok->done == E_EOF ? ENDMARKER : ERRORTOKEN);
    }

    /* Identifier (most frequent token!) */
    nonascii = 0;
    if (is_potential_identifier_start(c)) {
        /* Process the various legal combinations of b"", r"", u"", and f"". */
        int saw_b = 0, saw_r = 0, saw_u = 0, saw_f = 0;
        while (1) {
            if (!(saw_b || saw_u || saw_f) && (c == 'b' || c == 'B'))
                saw_b = 1;
            /* Since this is a backwards compatibility support literal we don't
               want to support it in arbitrary order like byte literals. */
            else if (!(saw_b || saw_u || saw_r || saw_f)
                     && (c == 'u'|| c == 'U')) {
                saw_u = 1;
            }
            /* ur"" and ru"" are not supported */
            else if (!(saw_r || saw_u) && (c == 'r' || c == 'R')) {
                saw_r = 1;
            }
            else if (!(saw_f || saw_b || saw_u) && (c == 'f' || c == 'F')) {
                saw_f = 1;
            }
            else {
                break;
            }
            c = tok_nextc(tok);
            if (c == '"' || c == '\'') {
                if (saw_f) {
                    goto f_string_quote;
                }
                goto letter_quote;
            }
        }
        while (is_potential_identifier_char(c)) {
            if (c >= 128) {
                nonascii = 1;
            }
            c = tok_nextc(tok);
        }
        tok_backup(tok, c);
        if (nonascii && !verify_identifier(tok)) {
            return MAKE_TOKEN(ERRORTOKEN);
        }

        p_start = tok->start;
        p_end = tok->cur;

        return MAKE_TOKEN(NAME);
    }

    if (c == '\r') {
        c = tok_nextc(tok);
    }

    /* Newline */
    if (c == '\n') {
        tok->atbol = 1;
        if (blankline || tok->level > 0) {
            if (tok->tok_extra_tokens) {
                if (tok->comment_newline) {
                    tok->comment_newline = 0;
                }
                p_start = tok->start;
                p_end = tok->cur;
                return MAKE_TOKEN(NL);
            }
            goto nextline;
        }
        if (tok->comment_newline && tok->tok_extra_tokens) {
            tok->comment_newline = 0;
            p_start = tok->start;
            p_end = tok->cur;
            return MAKE_TOKEN(NL);
        }
        p_start = tok->start;
        p_end = tok->cur - 1; /* Leave '\n' out of the string */
        tok->cont_line = 0;
        return MAKE_TOKEN(NEWLINE);
    }

    /* Period or number starting with period? */
    if (c == '.') {
        c = tok_nextc(tok);
        if (Py_ISDIGIT(c)) {
            goto fraction;
        } else if (c == '.') {
            c = tok_nextc(tok);
            if (c == '.') {
                p_start = tok->start;
                p_end = tok->cur;
                return MAKE_TOKEN(ELLIPSIS);
            }
            else {
                tok_backup(tok, c);
            }
            tok_backup(tok, '.');
        }
        else {
            tok_backup(tok, c);
        }
        p_start = tok->start;
        p_end = tok->cur;
        return MAKE_TOKEN(DOT);
    }

    /* Number */
    if (Py_ISDIGIT(c)) {
        if (c == '0') {
            /* Hex, octal or binary -- maybe. */
            c = tok_nextc(tok);
            if (c == 'x' || c == 'X') {
                /* Hex */
                c = tok_nextc(tok);
                do {
                    if (c == '_') {
                        c = tok_nextc(tok);
                    }
                    if (!Py_ISXDIGIT(c)) {
                        tok_backup(tok, c);
                        return MAKE_TOKEN(syntaxerror(tok, "invalid hexadecimal literal"));
                    }
                    do {
                        c = tok_nextc(tok);
                    } while (Py_ISXDIGIT(c));
                } while (c == '_');
                if (!verify_end_of_number(tok, c, "hexadecimal")) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
            else if (c == 'o' || c == 'O') {
                /* Octal */
                c = tok_nextc(tok);
                do {
                    if (c == '_') {
                        c = tok_nextc(tok);
                    }
                    if (c < '0' || c >= '8') {
                        if (Py_ISDIGIT(c)) {
                            return MAKE_TOKEN(syntaxerror(tok,
                                    "invalid digit '%c' in octal literal", c));
                        }
                        else {
                            tok_backup(tok, c);
                            return MAKE_TOKEN(syntaxerror(tok, "invalid octal literal"));
                        }
                    }
                    do {
                        c = tok_nextc(tok);
                    } while ('0' <= c && c < '8');
                } while (c == '_');
                if (Py_ISDIGIT(c)) {
                    return MAKE_TOKEN(syntaxerror(tok,
                            "invalid digit '%c' in octal literal", c));
                }
                if (!verify_end_of_number(tok, c, "octal")) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
            else if (c == 'b' || c == 'B') {
                /* Binary */
                c = tok_nextc(tok);
                do {
                    if (c == '_') {
                        c = tok_nextc(tok);
                    }
                    if (c != '0' && c != '1') {
                        if (Py_ISDIGIT(c)) {
                            return MAKE_TOKEN(syntaxerror(tok, "invalid digit '%c' in binary literal", c));
                        }
                        else {
                            tok_backup(tok, c);
                            return MAKE_TOKEN(syntaxerror(tok, "invalid binary literal"));
                        }
                    }
                    do {
                        c = tok_nextc(tok);
                    } while (c == '0' || c == '1');
                } while (c == '_');
                if (Py_ISDIGIT(c)) {
                    return MAKE_TOKEN(syntaxerror(tok, "invalid digit '%c' in binary literal", c));
                }
                if (!verify_end_of_number(tok, c, "binary")) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
            else {
                int nonzero = 0;
                /* maybe old-style octal; c is first char of it */
                /* in any case, allow '0' as a literal */
                while (1) {
                    if (c == '_') {
                        c = tok_nextc(tok);
                        if (!Py_ISDIGIT(c)) {
                            tok_backup(tok, c);
                            return MAKE_TOKEN(syntaxerror(tok, "invalid decimal literal"));
                        }
                    }
                    if (c != '0') {
                        break;
                    }
                    c = tok_nextc(tok);
                }
                char* zeros_end = tok->cur;
                if (Py_ISDIGIT(c)) {
                    nonzero = 1;
                    c = tok_decimal_tail(tok);
                    if (c == 0) {
                        return MAKE_TOKEN(ERRORTOKEN);
                    }
                }
                if (c == '.') {
                    c = tok_nextc(tok);
                    goto fraction;
                }
                else if (c == 'e' || c == 'E') {
                    goto exponent;
                }
                else if (c == 'j' || c == 'J') {
                    goto imaginary;
                }
                else if (nonzero && !tok->tok_extra_tokens) {
                    /* Old-style octal: now disallowed. */
                    tok_backup(tok, c);
                    return MAKE_TOKEN(syntaxerror_known_range(
                            tok, (int)(tok->start + 1 - tok->line_start),
                            (int)(zeros_end - tok->line_start),
                            "leading zeros in decimal integer "
                            "literals are not permitted; "
                            "use an 0o prefix for octal integers"));
                }
                if (!verify_end_of_number(tok, c, "decimal")) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
        }
        else {
            /* Decimal */
            c = tok_decimal_tail(tok);
            if (c == 0) {
                return MAKE_TOKEN(ERRORTOKEN);
            }
            {
                /* Accept floating point numbers. */
                if (c == '.') {
                    c = tok_nextc(tok);
        fraction:
                    /* Fraction */
                    if (Py_ISDIGIT(c)) {
                        c = tok_decimal_tail(tok);
                        if (c == 0) {
                            return MAKE_TOKEN(ERRORTOKEN);
                        }
                    }
                }
                if (c == 'e' || c == 'E') {
                    int e;
                  exponent:
                    e = c;
                    /* Exponent part */
                    c = tok_nextc(tok);
                    if (c == '+' || c == '-') {
                        c = tok_nextc(tok);
                        if (!Py_ISDIGIT(c)) {
                            tok_backup(tok, c);
                            return MAKE_TOKEN(syntaxerror(tok, "invalid decimal literal"));
                        }
                    } else if (!Py_ISDIGIT(c)) {
                        tok_backup(tok, c);
                        if (!verify_end_of_number(tok, e, "decimal")) {
                            return MAKE_TOKEN(ERRORTOKEN);
                        }
                        tok_backup(tok, e);
                        p_start = tok->start;
                        p_end = tok->cur;
                        return MAKE_TOKEN(NUMBER);
                    }
                    c = tok_decimal_tail(tok);
                    if (c == 0) {
                        return MAKE_TOKEN(ERRORTOKEN);
                    }
                }
                if (c == 'j' || c == 'J') {
                    /* Imaginary part */
        imaginary:
                    c = tok_nextc(tok);
                    if (!verify_end_of_number(tok, c, "imaginary")) {
                        return MAKE_TOKEN(ERRORTOKEN);
                    }
                }
                else if (!verify_end_of_number(tok, c, "decimal")) {
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
        }
        tok_backup(tok, c);
        p_start = tok->start;
        p_end = tok->cur;
        return MAKE_TOKEN(NUMBER);
    }

  f_string_quote:
    if (((Py_TOLOWER(*tok->start) == 'f' || Py_TOLOWER(*tok->start) == 'r') && (c == '\'' || c == '"'))) {
        int quote = c;
        int quote_size = 1;             /* 1 or 3 */

        /* Nodes of type STRING, especially multi line strings
           must be handled differently in order to get both
           the starting line number and the column offset right.
           (cf. issue 16806) */
        tok->first_lineno = tok->lineno;
        tok->multi_line_start = tok->line_start;

        /* Find the quote size and start of string */
        int after_quote = tok_nextc(tok);
        if (after_quote == quote) {
            int after_after_quote = tok_nextc(tok);
            if (after_after_quote == quote) {
                quote_size = 3;
            }
            else {
                // TODO: Check this
                tok_backup(tok, after_after_quote);
                tok_backup(tok, after_quote);
            }
        }
        if (after_quote != quote) {
            tok_backup(tok, after_quote);
        }


        p_start = tok->start;
        p_end = tok->cur;
        if (tok->tok_mode_stack_index + 1 >= MAXFSTRINGLEVEL) {
            return MAKE_TOKEN(syntaxerror(tok, "too many nested f-strings"));
        }
        tokenizer_mode *the_current_tok = TOK_NEXT_MODE(tok);
        the_current_tok->kind = TOK_FSTRING_MODE;
        the_current_tok->f_string_quote = quote;
        the_current_tok->f_string_quote_size = quote_size;
        the_current_tok->f_string_start = tok->start;
        the_current_tok->f_string_multi_line_start = tok->line_start;
        the_current_tok->f_string_line_start = tok->lineno;
        the_current_tok->f_string_start_offset = -1;
        the_current_tok->f_string_multi_line_start_offset = -1;
        the_current_tok->last_expr_buffer = NULL;
        the_current_tok->last_expr_size = 0;
        the_current_tok->last_expr_end = -1;
        the_current_tok->f_string_debug = 0;

        switch (*tok->start) {
            case 'F':
            case 'f':
                the_current_tok->f_string_raw = Py_TOLOWER(*(tok->start + 1)) == 'r';
                break;
            case 'R':
            case 'r':
                the_current_tok->f_string_raw = 1;
                break;
            default:
                Py_UNREACHABLE();
        }

        the_current_tok->curly_bracket_depth = 0;
        the_current_tok->curly_bracket_expr_start_depth = -1;
        return MAKE_TOKEN(FSTRING_START);
    }

  letter_quote:
    /* String */
    if (c == '\'' || c == '"') {
        int quote = c;
        int quote_size = 1;             /* 1 or 3 */
        int end_quote_size = 0;

        /* Nodes of type STRING, especially multi line strings
           must be handled differently in order to get both
           the starting line number and the column offset right.
           (cf. issue 16806) */
        tok->first_lineno = tok->lineno;
        tok->multi_line_start = tok->line_start;

        /* Find the quote size and start of string */
        c = tok_nextc(tok);
        if (c == quote) {
            c = tok_nextc(tok);
            if (c == quote) {
                quote_size = 3;
            }
            else {
                end_quote_size = 1;     /* empty string found */
            }
        }
        if (c != quote) {
            tok_backup(tok, c);
        }

        /* Get rest of string */
        while (end_quote_size != quote_size) {
            c = tok_nextc(tok);
            if (tok->done == E_ERROR) {
                return MAKE_TOKEN(ERRORTOKEN);
            }
            if (tok->done == E_DECODE) {
                break;
            }
            if (c == EOF || (quote_size == 1 && c == '\n')) {
                assert(tok->multi_line_start != NULL);
                // shift the tok_state's location into
                // the start of string, and report the error
                // from the initial quote character
                tok->cur = (char *)tok->start;
                tok->cur++;
                tok->line_start = tok->multi_line_start;
                int start = tok->lineno;
                tok->lineno = tok->first_lineno;

                if (INSIDE_FSTRING(tok)) {
                    /* When we are in an f-string, before raising the
                     * unterminated string literal error, check whether
                     * does the initial quote matches with f-strings quotes
                     * and if it is, then this must be a missing '}' token
                     * so raise the proper error */
                    tokenizer_mode *the_current_tok = TOK_GET_MODE(tok);
                    if (the_current_tok->f_string_quote == quote &&
                        the_current_tok->f_string_quote_size == quote_size) {
                        return MAKE_TOKEN(syntaxerror(tok, "f-string: expecting '}'", start));
                    }
                }

                if (quote_size == 3) {
                    syntaxerror(tok, "unterminated triple-quoted string literal"
                                     " (detected at line %d)", start);
                    if (c != '\n') {
                        tok->done = E_EOFS;
                    }
                    return MAKE_TOKEN(ERRORTOKEN);
                }
                else {
                    syntaxerror(tok, "unterminated string literal (detected at"
                                     " line %d)", start);
                    if (c != '\n') {
                        tok->done = E_EOLS;
                    }
                    return MAKE_TOKEN(ERRORTOKEN);
                }
            }
            if (c == quote) {
                end_quote_size += 1;
            }
            else {
                end_quote_size = 0;
                if (c == '\\') {
                    c = tok_nextc(tok);  /* skip escaped char */
                    if (c == '\r') {
                        c = tok_nextc(tok);
                    }
                }
            }
        }

        p_start = tok->start;
        p_end = tok->cur;
        return MAKE_TOKEN(STRING);
    }

    /* Line continuation */
    if (c == '\\') {
        if ((c = tok_continuation_line(tok)) == -1) {
            return MAKE_TOKEN(ERRORTOKEN);
        }
        tok->cont_line = 1;
        goto again; /* Read next line */
    }

    /* Punctuation character */
    int is_punctuation = (c == ':' || c == '}' || c == '!' || c == '{');
    if (is_punctuation && INSIDE_FSTRING(tok) && INSIDE_FSTRING_EXPR(current_tok)) {
        /* This code block gets executed before the curly_bracket_depth is incremented
         * by the `{` case, so for ensuring that we are on the 0th level, we need
         * to adjust it manually */
        int cursor = current_tok->curly_bracket_depth - (c != '{');
        if (cursor == 0 && !update_fstring_expr(tok, c)) {
            return MAKE_TOKEN(ENDMARKER);
        }
        if (cursor == 0 && c != '{' && set_fstring_expr(tok, token, c)) {
            return MAKE_TOKEN(ERRORTOKEN);
        }

        if (c == ':' && cursor == current_tok->curly_bracket_expr_start_depth) {
            current_tok->kind = TOK_FSTRING_MODE;
            p_start = tok->start;
            p_end = tok->cur;
            return MAKE_TOKEN(_PyToken_OneChar(c));
        }
    }

    /* Check for two-character token */
    {
        int c2 = tok_nextc(tok);
        int current_token = _PyToken_TwoChars(c, c2);
        if (current_token != OP) {
            int c3 = tok_nextc(tok);
            int current_token3 = _PyToken_ThreeChars(c, c2, c3);
            if (current_token3 != OP) {
                current_token = current_token3;
            }
            else {
                tok_backup(tok, c3);
            }
            p_start = tok->start;
            p_end = tok->cur;
            return MAKE_TOKEN(current_token);
        }
        tok_backup(tok, c2);
    }

    /* Keep track of parentheses nesting level */
    switch (c) {
    case '(':
    case '[':
    case '{':
        if (tok->level >= MAXLEVEL) {
            return MAKE_TOKEN(syntaxerror(tok, "too many nested parentheses"));
        }
        tok->parenstack[tok->level] = c;
        tok->parenlinenostack[tok->level] = tok->lineno;
        tok->parencolstack[tok->level] = (int)(tok->start - tok->line_start);
        tok->level++;
        if (INSIDE_FSTRING(tok)) {
            current_tok->curly_bracket_depth++;
        }
        break;
    case ')':
    case ']':
    case '}':
        if (INSIDE_FSTRING(tok) && !current_tok->curly_bracket_depth && c == '}') {
            return MAKE_TOKEN(syntaxerror(tok, "f-string: single '}' is not allowed"));
        }
        if (!tok->tok_extra_tokens && !tok->level) {
            return MAKE_TOKEN(syntaxerror(tok, "unmatched '%c'", c));
        }
        if (tok->level > 0) {
            tok->level--;
            int opening = tok->parenstack[tok->level];
            if (!tok->tok_extra_tokens && !((opening == '(' && c == ')') ||
                                            (opening == '[' && c == ']') ||
                                            (opening == '{' && c == '}'))) {
                /* If the opening bracket belongs to an f-string's expression
                part (e.g. f"{)}") and the closing bracket is an arbitrary
                nested expression, then instead of matching a different
                syntactical construct with it; we'll throw an unmatched
                parentheses error. */
                if (INSIDE_FSTRING(tok) && opening == '{') {
                    assert(current_tok->curly_bracket_depth >= 0);
                    int previous_bracket = current_tok->curly_bracket_depth - 1;
                    if (previous_bracket == current_tok->curly_bracket_expr_start_depth) {
                        return MAKE_TOKEN(syntaxerror(tok, "f-string: unmatched '%c'", c));
                    }
                }
                if (tok->parenlinenostack[tok->level] != tok->lineno) {
                    return MAKE_TOKEN(syntaxerror(tok,
                            "closing parenthesis '%c' does not match "
                            "opening parenthesis '%c' on line %d",
                            c, opening, tok->parenlinenostack[tok->level]));
                }
                else {
                    return MAKE_TOKEN(syntaxerror(tok,
                            "closing parenthesis '%c' does not match "
                            "opening parenthesis '%c'",
                            c, opening));
                }
            }
        }

        if (INSIDE_FSTRING(tok)) {
            current_tok->curly_bracket_depth--;
            if (c == '}' && current_tok->curly_bracket_depth == current_tok->curly_bracket_expr_start_depth) {
                current_tok->curly_bracket_expr_start_depth--;
                current_tok->kind = TOK_FSTRING_MODE;
                current_tok->f_string_debug = 0;
            }
        }
        break;
    default:
        break;
    }

    if (!Py_UNICODE_ISPRINTABLE(c)) {
        return MAKE_TOKEN(syntaxerror(tok, "invalid non-printable character U+%04X", c));
    }

    if( c == '=' && INSIDE_FSTRING_EXPR(current_tok)) {
        current_tok->f_string_debug = 1;
    }

    /* Punctuation character */
    p_start = tok->start;
    p_end = tok->cur;
    return MAKE_TOKEN(_PyToken_OneChar(c));
}

static int
tok_get_fstring_mode(struct tok_state *tok, tokenizer_mode* current_tok, struct token *token)
{
    const char *p_start = NULL;
    const char *p_end = NULL;
    int end_quote_size = 0;
    int unicode_escape = 0;

    tok->start = tok->cur;
    tok->first_lineno = tok->lineno;
    tok->starting_col_offset = tok->col_offset;

    // If we start with a bracket, we defer to the normal mode as there is nothing for us to tokenize
    // before it.
    int start_char = tok_nextc(tok);
    if (start_char == '{') {
        int peek1 = tok_nextc(tok);
        tok_backup(tok, peek1);
        tok_backup(tok, start_char);
        if (peek1 != '{') {
            current_tok->curly_bracket_expr_start_depth++;
            if (current_tok->curly_bracket_expr_start_depth >= MAX_EXPR_NESTING) {
                return MAKE_TOKEN(syntaxerror(tok, "f-string: expressions nested too deeply"));
            }
            TOK_GET_MODE(tok)->kind = TOK_REGULAR_MODE;
            return tok_get_normal_mode(tok, current_tok, token);
        }
    }
    else {
        tok_backup(tok, start_char);
    }

    // Check if we are at the end of the string
    for (int i = 0; i < current_tok->f_string_quote_size; i++) {
        int quote = tok_nextc(tok);
        if (quote != current_tok->f_string_quote) {
            tok_backup(tok, quote);
            goto f_string_middle;
        }
    }

    if (current_tok->last_expr_buffer != NULL) {
        PyMem_Free(current_tok->last_expr_buffer);
        current_tok->last_expr_buffer = NULL;
        current_tok->last_expr_size = 0;
        current_tok->last_expr_end = -1;
    }

    p_start = tok->start;
    p_end = tok->cur;
    tok->tok_mode_stack_index--;
    return MAKE_TOKEN(FSTRING_END);

f_string_middle:

    // TODO: This is a bit of a hack, but it works for now. We need to find a better way to handle
    // this.
    tok->multi_line_start = tok->line_start;
    while (end_quote_size != current_tok->f_string_quote_size) {
        int c = tok_nextc(tok);
        if (tok->done == E_ERROR) {
            return MAKE_TOKEN(ERRORTOKEN);
        }
        if (c == EOF || (current_tok->f_string_quote_size == 1 && c == '\n')) {
            if (tok->decoding_erred) {
                return MAKE_TOKEN(ERRORTOKEN);
            }

            assert(tok->multi_line_start != NULL);
            // shift the tok_state's location into
            // the start of string, and report the error
            // from the initial quote character
            tok->cur = (char *)current_tok->f_string_start;
            tok->cur++;
            tok->line_start = current_tok->f_string_multi_line_start;
            int start = tok->lineno;

            tokenizer_mode *the_current_tok = TOK_GET_MODE(tok);
            tok->lineno = the_current_tok->f_string_line_start;

            if (current_tok->f_string_quote_size == 3) {
                return MAKE_TOKEN(syntaxerror(tok,
                                    "unterminated triple-quoted f-string literal"
                                    " (detected at line %d)", start));
            }
            else {
                return MAKE_TOKEN(syntaxerror(tok,
                                    "unterminated f-string literal (detected at"
                                    " line %d)", start));
            }
        }

        if (c == current_tok->f_string_quote) {
            end_quote_size += 1;
            continue;
        } else {
            end_quote_size = 0;
        }

        int in_format_spec = (
                current_tok->last_expr_end != -1
                &&
                INSIDE_FSTRING_EXPR(current_tok)
        );
        if (c == '{') {
            int peek = tok_nextc(tok);
            if (peek != '{' || in_format_spec) {
                tok_backup(tok, peek);
                tok_backup(tok, c);
                current_tok->curly_bracket_expr_start_depth++;
                if (current_tok->curly_bracket_expr_start_depth >= MAX_EXPR_NESTING) {
                    return MAKE_TOKEN(syntaxerror(tok, "f-string: expressions nested too deeply"));
                }
                TOK_GET_MODE(tok)->kind = TOK_REGULAR_MODE;
                p_start = tok->start;
                p_end = tok->cur;
            } else {
                p_start = tok->start;
                p_end = tok->cur - 1;
            }
            return MAKE_TOKEN(FSTRING_MIDDLE);
        } else if (c == '}') {
            if (unicode_escape) {
                p_start = tok->start;
                p_end = tok->cur;
                return MAKE_TOKEN(FSTRING_MIDDLE);
            }
            int peek = tok_nextc(tok);

            // The tokenizer can only be in the format spec if we have already completed the expression
            // scanning (indicated by the end of the expression being set) and we are not at the top level
            // of the bracket stack (-1 is the top level). Since format specifiers can't legally use double
            // brackets, we can bypass it here.
            if (peek == '}' && !in_format_spec) {
                p_start = tok->start;
                p_end = tok->cur - 1;
            } else {
                tok_backup(tok, peek);
                tok_backup(tok, c);
                TOK_GET_MODE(tok)->kind = TOK_REGULAR_MODE;
                p_start = tok->start;
                p_end = tok->cur;
            }
            return MAKE_TOKEN(FSTRING_MIDDLE);
        } else if (c == '\\') {
            int peek = tok_nextc(tok);
            if (peek == '\r') {
                peek = tok_nextc(tok);
            }
            // Special case when the backslash is right before a curly
            // brace. We have to restore and return the control back
            // to the loop for the next iteration.
            if (peek == '{' || peek == '}') {
                if (!current_tok->f_string_raw) {
                    if (warn_invalid_escape_sequence(tok, peek)) {
                        return MAKE_TOKEN(ERRORTOKEN);
                    }
                }
                tok_backup(tok, peek);
                continue;
            }

            if (!current_tok->f_string_raw) {
                if (peek == 'N') {
                    /* Handle named unicode escapes (\N{BULLET}) */
                    peek = tok_nextc(tok);
                    if (peek == '{') {
                        unicode_escape = 1;
                    } else {
                        tok_backup(tok, peek);
                    }
                }
            } /* else {
                skip the escaped character
            }*/
        }
    }

    // Backup the f-string quotes to emit a final FSTRING_MIDDLE and
    // add the quotes to the FSTRING_END in the next tokenizer iteration.
    for (int i = 0; i < current_tok->f_string_quote_size; i++) {
        tok_backup(tok, current_tok->f_string_quote);
    }
    p_start = tok->start;
    p_end = tok->cur;
    return MAKE_TOKEN(FSTRING_MIDDLE);
}


static int
tok_get(struct tok_state *tok, struct token *token)
{
    tokenizer_mode *current_tok = TOK_GET_MODE(tok);
    if (current_tok->kind == TOK_REGULAR_MODE) {
        return tok_get_normal_mode(tok, current_tok, token);
    } else {
        return tok_get_fstring_mode(tok, current_tok, token);
    }
}

int
_PyTokenizer_Get(struct tok_state *tok, struct token *token)
{
    int result = tok_get(tok, token);
    if (tok->decoding_erred) {
        result = ERRORTOKEN;
        tok->done = E_DECODE;
    }
    return result;
}

#if defined(__wasi__) || (defined(__EMSCRIPTEN__) && (__EMSCRIPTEN_major__ >= 3))
// fdopen() with borrowed fd. WASI does not provide dup() and Emscripten's
// dup() emulation with open() is slow.
typedef union {
    void *cookie;
    int fd;
} borrowed;

static ssize_t
borrow_read(void *cookie, char *buf, size_t size)
{
    borrowed b = {.cookie = cookie};
    return read(b.fd, (void *)buf, size);
}

static FILE *
fdopen_borrow(int fd) {
    // supports only reading. seek fails. close and write are no-ops.
    cookie_io_functions_t io_cb = {borrow_read, NULL, NULL, NULL};
    borrowed b = {.fd = fd};
    return fopencookie(b.cookie, "r", io_cb);
}
#else
static FILE *
fdopen_borrow(int fd) {
    fd = _Py_dup(fd);
    if (fd < 0) {
        return NULL;
    }
    return fdopen(fd, "r");
}
#endif

/* Get the encoding of a Python file. Check for the coding cookie and check if
   the file starts with a BOM.

   _PyTokenizer_FindEncodingFilename() returns NULL when it can't find the
   encoding in the first or second line of the file (in which case the encoding
   should be assumed to be UTF-8).

   The char* returned is malloc'ed via PyMem_Malloc() and thus must be freed
   by the caller. */

char *
_PyTokenizer_FindEncodingFilename(int fd, PyObject *filename)
{
    struct tok_state *tok;
    FILE *fp;
    char *encoding = NULL;

    fp = fdopen_borrow(fd);
    if (fp == NULL) {
        return NULL;
    }
    tok = _PyTokenizer_FromFile(fp, NULL, NULL, NULL);
    if (tok == NULL) {
        fclose(fp);
        return NULL;
    }
    if (filename != NULL) {
        tok->filename = Py_NewRef(filename);
    }
    else {
        tok->filename = PyUnicode_FromString("<string>");
        if (tok->filename == NULL) {
            fclose(fp);
            _PyTokenizer_Free(tok);
            return encoding;
        }
    }
    struct token token;
    // We don't want to report warnings here because it could cause infinite recursion
    // if fetching the encoding shows a warning.
    tok->report_warnings = 0;
    while (tok->lineno < 2 && tok->done == E_OK) {
        _PyToken_Init(&token);
        _PyTokenizer_Get(tok, &token);
        _PyToken_Free(&token);
    }
    fclose(fp);
    if (tok->encoding) {
        encoding = (char *)PyMem_Malloc(strlen(tok->encoding) + 1);
        if (encoding) {
            strcpy(encoding, tok->encoding);
        }
    }
    _PyTokenizer_Free(tok);
    return encoding;
}

#ifdef Py_DEBUG
void
tok_dump(int type, char *start, char *end)
{
    fprintf(stderr, "%s", _PyParser_TokenNames[type]);
    if (type == NAME || type == NUMBER || type == STRING || type == OP)
        fprintf(stderr, "(%.*s)", (int)(end - start), start);
}
#endif  // Py_DEBUG