#! /usr/bin/python3.8
'''
From gdb 7 onwards, gdb's build can be configured --with-python, allowing gdb
to be extended with Python code e.g. for library-specific data visualizations,
such as for the C++ STL types.  Documentation on this API can be seen at:
http://sourceware.org/gdb/current/onlinedocs/gdb/Python-API.html


This python module deals with the case when the process being debugged (the
"inferior process" in gdb parlance) is itself python, or more specifically,
linked against libpython.  In this situation, almost every item of data is a
(PyObject*), and having the debugger merely print their addresses is not very
enlightening.

This module embeds knowledge about the implementation details of libpython so
that we can emit useful visualizations e.g. a string, a list, a dict, a frame
giving file/line information and the state of local variables

In particular, given a gdb.Value corresponding to a PyObject* in the inferior
process, we can generate a "proxy value" within the gdb process.  For example,
given a PyObject* in the inferior process that is in fact a PyListObject*
holding three PyObject* that turn out to be PyBytesObject* instances, we can
generate a proxy value within the gdb process that is a list of bytes
instances:
  [b"foo", b"bar", b"baz"]

Doing so can be expensive for complicated graphs of objects, and could take
some time, so we also have a "write_repr" method that writes a representation
of the data to a file-like object.  This allows us to stop the traversal by
having the file-like object raise an exception if it gets too much data.

With both "proxyval" and "write_repr" we keep track of the set of all addresses
visited so far in the traversal, to avoid infinite recursion due to cycles in
the graph of object references.

We try to defer gdb.lookup_type() invocations for python types until as late as
possible: for a dynamically linked python binary, when the process starts in
the debugger, the libpython.so hasn't been dynamically loaded yet, so none of
the type names are known to the debugger

The module also extends gdb with some python-specific commands.
'''

# NOTE: some gdbs are linked with Python 3, so this file should be dual-syntax
# compatible (2.6+ and 3.0+).  See #19308.

from __future__ import print_function
import gdb
import os
import locale
import sys

if sys.version_info[0] >= 3:
    unichr = chr
    xrange = range
    long = int

# Look up the gdb.Type for some standard types:
# Those need to be refreshed as types (pointer sizes) may change when
# gdb loads different executables

def _type_char_ptr():
    return gdb.lookup_type('char').pointer()  # char*


def _type_unsigned_char_ptr():
    return gdb.lookup_type('unsigned char').pointer()  # unsigned char*


def _type_unsigned_short_ptr():
    return gdb.lookup_type('unsigned short').pointer()


def _type_unsigned_int_ptr():
    return gdb.lookup_type('unsigned int').pointer()


def _sizeof_void_p():
    return gdb.lookup_type('void').pointer().sizeof


# value computed later, see PyUnicodeObjectPtr.proxy()
_is_pep393 = None

Py_TPFLAGS_HEAPTYPE = (1 << 9)
Py_TPFLAGS_LONG_SUBCLASS     = (1 << 24)
Py_TPFLAGS_LIST_SUBCLASS     = (1 << 25)
Py_TPFLAGS_TUPLE_SUBCLASS    = (1 << 26)
Py_TPFLAGS_BYTES_SUBCLASS    = (1 << 27)
Py_TPFLAGS_UNICODE_SUBCLASS  = (1 << 28)
Py_TPFLAGS_DICT_SUBCLASS     = (1 << 29)
Py_TPFLAGS_BASE_EXC_SUBCLASS = (1 << 30)
Py_TPFLAGS_TYPE_SUBCLASS     = (1 << 31)


MAX_OUTPUT_LEN=1024

hexdigits = "0123456789abcdef"

ENCODING = locale.getpreferredencoding()

FRAME_INFO_OPTIMIZED_OUT = '(frame information optimized out)'
UNABLE_READ_INFO_PYTHON_FRAME = 'Unable to read information on python frame'
EVALFRAME = '_PyEval_EvalFrameDefault'

class NullPyObjectPtr(RuntimeError):
    pass


def safety_limit(val):
    # Given an integer value from the process being debugged, limit it to some
    # safety threshold so that arbitrary breakage within said process doesn't
    # break the gdb process too much (e.g. sizes of iterations, sizes of lists)
    return min(val, 1000)


def safe_range(val):
    # As per range, but don't trust the value too much: cap it to a safety
    # threshold in case the data was corrupted
    return xrange(safety_limit(int(val)))

if sys.version_info[0] >= 3:
    def write_unicode(file, text):
        file.write(text)
else:
    def write_unicode(file, text):
        # Write a byte or unicode string to file. Unicode strings are encoded to
        # ENCODING encoding with 'backslashreplace' error handler to avoid
        # UnicodeEncodeError.
        if isinstance(text, unicode):
            text = text.encode(ENCODING, 'backslashreplace')
        file.write(text)

try:
    os_fsencode = os.fsencode
except AttributeError:
    def os_fsencode(filename):
        if not isinstance(filename, unicode):
            return filename
        encoding = sys.getfilesystemencoding()
        if encoding == 'mbcs':
            # mbcs doesn't support surrogateescape
            return filename.encode(encoding)
        encoded = []
        for char in filename:
            # surrogateescape error handler
            if 0xDC80 <= ord(char) <= 0xDCFF:
                byte = chr(ord(char) - 0xDC00)
            else:
                byte = char.encode(encoding)
            encoded.append(byte)
        return ''.join(encoded)

class StringTruncated(RuntimeError):
    pass

class TruncatedStringIO(object):
    '''Similar to io.StringIO, but can truncate the output by raising a
    StringTruncated exception'''
    def __init__(self, maxlen=None):
        self._val = ''
        self.maxlen = maxlen

    def write(self, data):
        if self.maxlen:
            if len(data) + len(self._val) > self.maxlen:
                # Truncation:
                self._val += data[0:self.maxlen - len(self._val)]
                raise StringTruncated()

        self._val += data

    def getvalue(self):
        return self._val

class PyObjectPtr(object):
    """
    Class wrapping a gdb.Value that's either a (PyObject*) within the
    inferior process, or some subclass pointer e.g. (PyBytesObject*)

    There will be a subclass for every refined PyObject type that we care
    about.

    Note that at every stage the underlying pointer could be NULL, point
    to corrupt data, etc; this is the debugger, after all.
    """
    _typename = 'PyObject'

    def __init__(self, gdbval, cast_to=None):
        if cast_to:
            self._gdbval = gdbval.cast(cast_to)
        else:
            self._gdbval = gdbval

    def field(self, name):
        '''
        Get the gdb.Value for the given field within the PyObject, coping with
        some python 2 versus python 3 differences.

        Various libpython types are defined using the "PyObject_HEAD" and
        "PyObject_VAR_HEAD" macros.

        In Python 2, this these are defined so that "ob_type" and (for a var
        object) "ob_size" are fields of the type in question.

        In Python 3, this is defined as an embedded PyVarObject type thus:
           PyVarObject ob_base;
        so that the "ob_size" field is located insize the "ob_base" field, and
        the "ob_type" is most easily accessed by casting back to a (PyObject*).
        '''
        if self.is_null():
            raise NullPyObjectPtr(self)

        if name == 'ob_type':
            pyo_ptr = self._gdbval.cast(PyObjectPtr.get_gdb_type())
            return pyo_ptr.dereference()[name]

        if name == 'ob_size':
            pyo_ptr = self._gdbval.cast(PyVarObjectPtr.get_gdb_type())
            return pyo_ptr.dereference()[name]

        # General case: look it up inside the object:
        return self._gdbval.dereference()[name]

    def pyop_field(self, name):
        '''
        Get a PyObjectPtr for the given PyObject* field within this PyObject,
        coping with some python 2 versus python 3 differences.
        '''
        return PyObjectPtr.from_pyobject_ptr(self.field(name))

    def write_field_repr(self, name, out, visited):
        '''
        Extract the PyObject* field named "name", and write its representation
        to file-like object "out"
        '''
        field_obj = self.pyop_field(name)
        field_obj.write_repr(out, visited)

    def get_truncated_repr(self, maxlen):
        '''
        Get a repr-like string for the data, but truncate it at "maxlen" bytes
        (ending the object graph traversal as soon as you do)
        '''
        out = TruncatedStringIO(maxlen)
        try:
            self.write_repr(out, set())
        except StringTruncated:
            # Truncation occurred:
            return out.getvalue() + '...(truncated)'

        # No truncation occurred:
        return out.getvalue()

    def type(self):
        return PyTypeObjectPtr(self.field('ob_type'))

    def is_null(self):
        return 0 == long(self._gdbval)

    def is_optimized_out(self):
        '''
        Is the value of the underlying PyObject* visible to the debugger?

        This can vary with the precise version of the compiler used to build
        Python, and the precise version of gdb.

        See e.g. https://bugzilla.redhat.com/show_bug.cgi?id=556975 with
        PyEval_EvalFrameEx's "f"
        '''
        return self._gdbval.is_optimized_out

    def safe_tp_name(self):
        try:
            ob_type = self.type()
            tp_name = ob_type.field('tp_name')
            return tp_name.string()
        # NullPyObjectPtr: NULL tp_name?
        # RuntimeError: Can't even read the object at all?
        # UnicodeDecodeError: Failed to decode tp_name bytestring
        except (NullPyObjectPtr, RuntimeError, UnicodeDecodeError):
            return 'unknown'

    def proxyval(self, visited):
        '''
        Scrape a value from the inferior process, and try to represent it
        within the gdb process, whilst (hopefully) avoiding crashes when
        the remote data is corrupt.

        Derived classes will override this.

        For example, a PyIntObject* with ob_ival 42 in the inferior process
        should result in an int(42) in this process.

        visited: a set of all gdb.Value pyobject pointers already visited
        whilst generating this value (to guard against infinite recursion when
        visiting object graphs with loops).  Analogous to Py_ReprEnter and
        Py_ReprLeave
        '''

        class FakeRepr(object):
            """
            Class representing a non-descript PyObject* value in the inferior
            process for when we don't have a custom scraper, intended to have
            a sane repr().
            """

            def __init__(self, tp_name, address):
                self.tp_name = tp_name
                self.address = address

            def __repr__(self):
                # For the NULL pointer, we have no way of knowing a type, so
                # special-case it as per
                # http://bugs.python.org/issue8032#msg100882
                if self.address == 0:
                    return '0x0'
                return '<%s at remote 0x%x>' % (self.tp_name, self.address)

        return FakeRepr(self.safe_tp_name(),
                        long(self._gdbval))

    def write_repr(self, out, visited):
        '''
        Write a string representation of the value scraped from the inferior
        process to "out", a file-like object.
        '''
        # Default implementation: generate a proxy value and write its repr
        # However, this could involve a lot of work for complicated objects,
        # so for derived classes we specialize this
        return out.write(repr(self.proxyval(visited)))

    @classmethod
    def subclass_from_type(cls, t):
        '''
        Given a PyTypeObjectPtr instance wrapping a gdb.Value that's a
        (PyTypeObject*), determine the corresponding subclass of PyObjectPtr
        to use

        Ideally, we would look up the symbols for the global types, but that
        isn't working yet:
          (gdb) python print gdb.lookup_symbol('PyList_Type')[0].value
          Traceback (most recent call last):
            File "<string>", line 1, in <module>
          NotImplementedError: Symbol type not yet supported in Python scripts.