[3.10] Ensure the str member of the tokenizer is always initialised (GH-29681). (GH-29683)

(cherry picked from commit 4f006a789a35f5d1a7ef142bd1304ce167392457)

Co-authored-by: Pablo Galindo Salgado <Pablogsal@gmail.com>

4 files changed, 428 insertions, 3 deletions

diff --git a/Parser/pegen.c b/Parser/pegen.c
index 464a902..8946aa3 100644
--- a/Parser/pegen.c
+++ b/Parser/pegen.c
@@ -432,7 +432,7 @@ get_error_line(Parser *p, Py_ssize_t lineno)
      * (multi-line) statement are stored in p->tok->interactive_src_start.
      * If not, we're parsing from a string, which means that the whole source
      * is stored in p->tok->str. */
-    assert(p->tok->fp == NULL || p->tok->fp == stdin);
+    assert((p->tok->fp == NULL && p->tok->str != NULL) || p->tok->fp == stdin);
 
     char *cur_line = p->tok->fp_interactive ? p->tok->interactive_src_start : p->tok->str;
     assert(cur_line != NULL);
diff --git a/Parser/pegen_errors.c b/Parser/pegen_errors.c
new file mode 100644
index 0000000..694184a
--- /dev/null
+++ b/Parser/pegen_errors.c
@@ -0,0 +1,425 @@
+#include <Python.h>
+#include <errcode.h>
+
+#include "tokenizer.h"
+#include "pegen.h"
+
+// TOKENIZER ERRORS
+
+void
+_PyPegen_raise_tokenizer_init_error(PyObject *filename)
+{
+    if (!(PyErr_ExceptionMatches(PyExc_LookupError)
+          || PyErr_ExceptionMatches(PyExc_SyntaxError)
+          || PyErr_ExceptionMatches(PyExc_ValueError)
+          || PyErr_ExceptionMatches(PyExc_UnicodeDecodeError))) {
+        return;
+    }
+    PyObject *errstr = NULL;
+    PyObject *tuple = NULL;
+    PyObject *type;
+    PyObject *value;
+    PyObject *tback;
+    PyErr_Fetch(&type, &value, &tback);
+    errstr = PyObject_Str(value);
+    if (!errstr) {
+        goto error;
+    }
+
+    PyObject *tmp = Py_BuildValue("(OiiO)", filename, 0, -1, Py_None);
+    if (!tmp) {
+        goto error;
+    }
+
+    tuple = PyTuple_Pack(2, errstr, tmp);
+    Py_DECREF(tmp);
+    if (!value) {
+        goto error;
+    }
+    PyErr_SetObject(PyExc_SyntaxError, tuple);
+
+error:
+    Py_XDECREF(type);
+    Py_XDECREF(value);
+    Py_XDECREF(tback);
+    Py_XDECREF(errstr);
+    Py_XDECREF(tuple);
+}
+
+static inline void
+raise_unclosed_parentheses_error(Parser *p) {
+       int error_lineno = p->tok->parenlinenostack[p->tok->level-1];
+       int error_col = p->tok->parencolstack[p->tok->level-1];
+       RAISE_ERROR_KNOWN_LOCATION(p, PyExc_SyntaxError,
+                                  error_lineno, error_col, error_lineno, -1,
+                                  "'%c' was never closed",
+                                  p->tok->parenstack[p->tok->level-1]);
+}
+
+int
+_Pypegen_tokenizer_error(Parser *p)
+{
+    if (PyErr_Occurred()) {
+        return -1;
+    }
+
+    const char *msg = NULL;
+    PyObject* errtype = PyExc_SyntaxError;
+    Py_ssize_t col_offset = -1;
+    switch (p->tok->done) {
+        case E_TOKEN:
+            msg = "invalid token";
+            break;
+        case E_EOF:
+            if (p->tok->level) {
+                raise_unclosed_parentheses_error(p);
+            } else {
+                RAISE_SYNTAX_ERROR("unexpected EOF while parsing");
+            }
+            return -1;
+        case E_DEDENT:
+            RAISE_INDENTATION_ERROR("unindent does not match any outer indentation level");
+            return -1;
+        case E_INTR:
+            if (!PyErr_Occurred()) {
+                PyErr_SetNone(PyExc_KeyboardInterrupt);
+            }
+            return -1;
+        case E_NOMEM:
+            PyErr_NoMemory();
+            return -1;
+        case E_TABSPACE:
+            errtype = PyExc_TabError;
+            msg = "inconsistent use of tabs and spaces in indentation";
+            break;
+        case E_TOODEEP:
+            errtype = PyExc_IndentationError;
+            msg = "too many levels of indentation";
+            break;
+        case E_LINECONT: {
+            col_offset = p->tok->cur - p->tok->buf - 1;
+            msg = "unexpected character after line continuation character";
+            break;
+        }
+        default:
+            msg = "unknown parsing error";
+    }
+
+    RAISE_ERROR_KNOWN_LOCATION(p, errtype, p->tok->lineno,
+                               col_offset >= 0 ? col_offset : 0,
+                               p->tok->lineno, -1, msg);
+    return -1;
+}
+
+int
+_Pypegen_raise_decode_error(Parser *p)
+{
+    assert(PyErr_Occurred());
+    const char *errtype = NULL;
+    if (PyErr_ExceptionMatches(PyExc_UnicodeError)) {
+        errtype = "unicode error";
+    }
+    else if (PyErr_ExceptionMatches(PyExc_ValueError)) {
+        errtype = "value error";
+    }
+    if (errtype) {
+        PyObject *type;
+        PyObject *value;
+        PyObject *tback;
+        PyObject *errstr;
+        PyErr_Fetch(&type, &value, &tback);
+        errstr = PyObject_Str(value);
+        if (errstr) {
+            RAISE_SYNTAX_ERROR("(%s) %U", errtype, errstr);
+            Py_DECREF(errstr);
+        }
+        else {
+            PyErr_Clear();
+            RAISE_SYNTAX_ERROR("(%s) unknown error", errtype);
+        }
+        Py_XDECREF(type);
+        Py_XDECREF(value);
+        Py_XDECREF(tback);
+    }
+
+    return -1;
+}
+
+static int
+_PyPegen_tokenize_full_source_to_check_for_errors(Parser *p) {
+    // Tokenize the whole input to see if there are any tokenization
+    // errors such as mistmatching parentheses. These will get priority
+    // over generic syntax errors only if the line number of the error is
+    // before the one that we had for the generic error.
+
+    // We don't want to tokenize to the end for interactive input
+    if (p->tok->prompt != NULL) {
+        return 0;
+    }
+
+    PyObject *type, *value, *traceback;
+    PyErr_Fetch(&type, &value, &traceback);
+
+    Token *current_token = p->known_err_token != NULL ? p->known_err_token : p->tokens[p->fill - 1];
+    Py_ssize_t current_err_line = current_token->lineno;
+
+    int ret = 0;
+
+    for (;;) {
+        const char *start;
+        const char *end;
+        switch (_PyTokenizer_Get(p->tok, &start, &end)) {
+            case ERRORTOKEN:
+                if (p->tok->level != 0) {
+                    int error_lineno = p->tok->parenlinenostack[p->tok->level-1];
+                    if (current_err_line > error_lineno) {
+                        raise_unclosed_parentheses_error(p);
+                        ret = -1;
+                        goto exit;
+                    }
+                }
+                break;
+            case ENDMARKER:
+                break;
+            default:
+                continue;
+        }
+        break;
+    }
+
+
+exit:
+    if (PyErr_Occurred()) {
+        Py_XDECREF(value);
+        Py_XDECREF(type);
+        Py_XDECREF(traceback);
+    } else {
+        PyErr_Restore(type, value, traceback);
+    }
+    return ret;
+}
+
+// PARSER ERRORS
+
+void *
+_PyPegen_raise_error(Parser *p, PyObject *errtype, const char *errmsg, ...)
+{
+    if (p->fill == 0) {
+        va_list va;
+        va_start(va, errmsg);
+        _PyPegen_raise_error_known_location(p, errtype, 0, 0, 0, -1, errmsg, va);
+        va_end(va);
+        return NULL;
+    }
+
+    Token *t = p->known_err_token != NULL ? p->known_err_token : p->tokens[p->fill - 1];
+    Py_ssize_t col_offset;
+    Py_ssize_t end_col_offset = -1;
+    if (t->col_offset == -1) {
+        if (p->tok->cur == p->tok->buf) {
+            col_offset = 0;
+        } else {
+            const char* start = p->tok->buf  ? p->tok->line_start : p->tok->buf;
+            col_offset = Py_SAFE_DOWNCAST(p->tok->cur - start, intptr_t, int);
+        }
+    } else {
+        col_offset = t->col_offset + 1;
+    }
+
+    if (t->end_col_offset != -1) {
+        end_col_offset = t->end_col_offset + 1;
+    }
+
+    va_list va;
+    va_start(va, errmsg);
+    _PyPegen_raise_error_known_location(p, errtype, t->lineno, col_offset, t->end_lineno, end_col_offset, errmsg, va);
+    va_end(va);
+
+    return NULL;
+}
+
+static PyObject *
+get_error_line_from_tokenizer_buffers(Parser *p, Py_ssize_t lineno)
+{
+    /* If the file descriptor is interactive, the source lines of the current
+     * (multi-line) statement are stored in p->tok->interactive_src_start.
+     * If not, we're parsing from a string, which means that the whole source
+     * is stored in p->tok->str. */
+    assert((p->tok->fp == NULL && p->tok->str != NULL) || p->tok->fp == stdin);
+
+    char *cur_line = p->tok->fp_interactive ? p->tok->interactive_src_start : p->tok->str;
+    assert(cur_line != NULL);
+
+    for (int i = 0; i < lineno - 1; i++) {
+        cur_line = strchr(cur_line, '\n') + 1;
+    }
+
+    char *next_newline;
+    if ((next_newline = strchr(cur_line, '\n')) == NULL) { // This is the last line
+        next_newline = cur_line + strlen(cur_line);
+    }
+    return PyUnicode_DecodeUTF8(cur_line, next_newline - cur_line, "replace");
+}
+
+void *
+_PyPegen_raise_error_known_location(Parser *p, PyObject *errtype,
+                                    Py_ssize_t lineno, Py_ssize_t col_offset,
+                                    Py_ssize_t end_lineno, Py_ssize_t end_col_offset,
+                                    const char *errmsg, va_list va)
+{
+    PyObject *value = NULL;
+    PyObject *errstr = NULL;
+    PyObject *error_line = NULL;
+    PyObject *tmp = NULL;
+    p->error_indicator = 1;
+
+    if (end_lineno == CURRENT_POS) {
+        end_lineno = p->tok->lineno;
+    }
+    if (end_col_offset == CURRENT_POS) {
+        end_col_offset = p->tok->cur - p->tok->line_start;
+    }
+
+    if (p->start_rule == Py_fstring_input) {
+        const char *fstring_msg = "f-string: ";
+        Py_ssize_t len = strlen(fstring_msg) + strlen(errmsg);
+
+        char *new_errmsg = PyMem_Malloc(len + 1); // Lengths of both strings plus NULL character
+        if (!new_errmsg) {
+            return (void *) PyErr_NoMemory();
+        }
+
+        // Copy both strings into new buffer
+        memcpy(new_errmsg, fstring_msg, strlen(fstring_msg));
+        memcpy(new_errmsg + strlen(fstring_msg), errmsg, strlen(errmsg));
+        new_errmsg[len] = 0;
+        errmsg = new_errmsg;
+    }
+    errstr = PyUnicode_FromFormatV(errmsg, va);
+    if (!errstr) {
+        goto error;
+    }
+
+    if (p->tok->fp_interactive) {
+        error_line = get_error_line_from_tokenizer_buffers(p, lineno);
+    }
+    else if (p->start_rule == Py_file_input) {
+        error_line = _PyErr_ProgramDecodedTextObject(p->tok->filename,
+                                                     (int) lineno, p->tok->encoding);
+    }
+
+    if (!error_line) {
+        /* PyErr_ProgramTextObject was not called or returned NULL. If it was not called,
+           then we need to find the error line from some other source, because
+           p->start_rule != Py_file_input. If it returned NULL, then it either unexpectedly
+           failed or we're parsing from a string or the REPL. There's a third edge case where
+           we're actually parsing from a file, which has an E_EOF SyntaxError and in that case
+           `PyErr_ProgramTextObject` fails because lineno points to last_file_line + 1, which
+           does not physically exist */
+        assert(p->tok->fp == NULL || p->tok->fp == stdin || p->tok->done == E_EOF);
+
+        if (p->tok->lineno <= lineno && p->tok->inp > p->tok->buf) {
+            Py_ssize_t size = p->tok->inp - p->tok->buf;
+            error_line = PyUnicode_DecodeUTF8(p->tok->buf, size, "replace");
+        }
+        else if (p->tok->fp == NULL || p->tok->fp == stdin) {
+            error_line = get_error_line_from_tokenizer_buffers(p, lineno);
+        }
+        else {
+            error_line = PyUnicode_FromStringAndSize("", 0);
+        }
+        if (!error_line) {
+            goto error;
+        }
+    }
+
+    if (p->start_rule == Py_fstring_input) {
+        col_offset -= p->starting_col_offset;
+        end_col_offset -= p->starting_col_offset;
+    }
+
+    Py_ssize_t col_number = col_offset;
+    Py_ssize_t end_col_number = end_col_offset;
+
+    if (p->tok->encoding != NULL) {
+        col_number = _PyPegen_byte_offset_to_character_offset(error_line, col_offset);
+        if (col_number < 0) {
+            goto error;
+        }
+        if (end_col_number > 0) {
+            Py_ssize_t end_col_offset = _PyPegen_byte_offset_to_character_offset(error_line, end_col_number);
+            if (end_col_offset < 0) {
+                goto error;
+            } else {
+                end_col_number = end_col_offset;
+            }
+        }
+    }
+    tmp = Py_BuildValue("(OiiNii)", p->tok->filename, lineno, col_number, error_line, end_lineno, end_col_number);
+    if (!tmp) {
+        goto error;
+    }
+    value = PyTuple_Pack(2, errstr, tmp);
+    Py_DECREF(tmp);
+    if (!value) {
+        goto error;
+    }
+    PyErr_SetObject(errtype, value);
+
+    Py_DECREF(errstr);
+    Py_DECREF(value);
+    if (p->start_rule == Py_fstring_input) {
+        PyMem_Free((void *)errmsg);
+    }
+    return NULL;
+
+error:
+    Py_XDECREF(errstr);
+    Py_XDECREF(error_line);
+    if (p->start_rule == Py_fstring_input) {
+        PyMem_Free((void *)errmsg);
+    }
+    return NULL;
+}
+
+void
+_Pypegen_set_syntax_error(Parser* p, Token* last_token) {
+    // Existing sintax error
+    if (PyErr_Occurred()) {
+        // Prioritize tokenizer errors to custom syntax errors raised
+        // on the second phase only if the errors come from the parser.
+        if (p->tok->done == E_DONE && PyErr_ExceptionMatches(PyExc_SyntaxError)) {
+            _PyPegen_tokenize_full_source_to_check_for_errors(p);
+        }
+        // Propagate the existing syntax error.
+        return;
+    }
+    // Initialization error
+    if (p->fill == 0) {
+        RAISE_SYNTAX_ERROR("error at start before reading any input");
+    }
+    // Parser encountered EOF (End of File) unexpectedtly
+    if (p->tok->done == E_EOF) {
+        if (p->tok->level) {
+            raise_unclosed_parentheses_error(p);
+        } else {
+            RAISE_SYNTAX_ERROR("unexpected EOF while parsing");
+        }
+        return;
+    }
+    // Indentation error in the tokenizer
+    if (last_token->type == INDENT || last_token->type == DEDENT) {
+        RAISE_INDENTATION_ERROR(last_token->type == INDENT ? "unexpected indent" : "unexpected unindent");
+        return;
+    }
+    // Unknown error (generic case)
+
+    // Use the last token we found on the first pass to avoid reporting
+    // incorrect locations for generic syntax errors just because we reached
+    // further away when trying to find specific syntax errors in the second
+    // pass.
+    RAISE_SYNTAX_ERROR_KNOWN_LOCATION(last_token, "invalid syntax");
+    // _PyPegen_tokenize_full_source_to_check_for_errors will override the existing
+    // generic SyntaxError we just raised if errors are found.
+    _PyPegen_tokenize_full_source_to_check_for_errors(p);
+}
\ No newline at end of file
diff --git a/Parser/tokenizer.c b/Parser/tokenizer.c
index 76a22da..672fdb9 100644
--- a/Parser/tokenizer.c
+++ b/Parser/tokenizer.c
@@ -86,7 +86,7 @@ tok_new(void)
     tok->async_def_indent = 0;
     tok->async_def_nl = 0;
     tok->interactive_underflow = IUNDERFLOW_NORMAL;
-
+    tok->str = NULL;
     return tok;
 }
 
diff --git a/Parser/tokenizer.h b/Parser/tokenizer.h
index 677f9db..61f0a61 100644
--- a/Parser/tokenizer.h
+++ b/Parser/tokenizer.h
@@ -71,7 +71,7 @@ struct tok_state {
     PyObject *decoding_readline; /* open(...).readline */
     PyObject *decoding_buffer;
     const char* enc;        /* Encoding for the current str. */
-    char* str;
+    char* str;          /* Source string being tokenized (if tokenizing from a string)*/
     char* input;       /* Tokenizer's newline translated copy of the string. */
 
     int type_comments;      /* Whether to look for type comments */