""" Parse a ppt (MS PowerPoint 97-2003) file Based on olefile, parse the ppt-specific info Code much influenced by olevba._extract_vba but much more object-oriented (possibly slightly excessively so) Currently quite narrowly focused on extracting VBA from ppt files, no slides or stuff, but built to be extended to parsing more/all of the file. For better "understanding" of ppt files, see module ppt_record_parser, which will probably replace this module some time soon. References: * https://msdn.microsoft.com/en-us/library/dd921564%28v=office.12%29.aspx and links there-in WARNING! Before thinking about understanding or even extending this module, please keep in mind that module ppt_record_parser has a better "understanding" of the ppt file structure and will replace this module some time soon! """ # === LICENSE ================================================================= # TODO #------------------------------------------------------------------------------ # TODO: # - make stream optional in PptUnexpectedData # - can speed-up by using less bigger struct.parse calls? # - license # - make buffered stream from output of iterative_decompress # - maybe can merge the 2 decorators into 1? (with_opened_main_stream) # - REPLACE THIS MODULE with ppt_record_parser # CHANGELOG: # 2016-05-04 v0.01 CH: - start parsing "Current User" stream # 2016-07-20 v0.50 SL: - added Python 3 support # 2016-09-13 PL: - fixed olefile import for Python 2+3 # - fixed format strings for Python 2.6 (issue #75) # 2017-04-23 v0.51 PL: - fixed absolute imports and issue #101 # 2018-09-11 v0.54 PL: - olefile is now a dependency __version__ = '0.54' # --- IMPORTS ------------------------------------------------------------------ import sys import logging import struct import traceback import os import zlib # IMPORTANT: it should be possible to run oletools directly as scripts # in any directory without installing them with pip or setup.py. # In that case, relative imports are NOT usable. # And to enable Python 2+3 compatibility, we need to use absolute imports, # so we add the oletools parent folder to sys.path (absolute+normalized path): _thismodule_dir = os.path.normpath(os.path.abspath(os.path.dirname(__file__))) # print('_thismodule_dir = %r' % _thismodule_dir) _parent_dir = os.path.normpath(os.path.join(_thismodule_dir, '..')) # print('_parent_dir = %r' % _thirdparty_dir) if not _parent_dir in sys.path: sys.path.insert(0, _parent_dir) import olefile # TODO: this is a temporary fix until all logging features are unified in oletools def get_logger(name, level=logging.CRITICAL+1): """ Create a suitable logger object for this module. The goal is not to change settings of the root logger, to avoid getting other modules' logs on the screen. If a logger exists with same name, reuse it. (Else it would have duplicate handlers and messages would be doubled.) The level is set to CRITICAL+1 by default, to avoid any logging. """ # First, test if there is already a logger with the same name, else it # will generate duplicate messages (due to duplicate handlers): if name in logging.Logger.manager.loggerDict: #NOTE: another less intrusive but more "hackish" solution would be to # use getLogger then test if its effective level is not default. logger = logging.getLogger(name) # make sure level is OK: logger.setLevel(level) return logger # get a new logger: logger = logging.getLogger(name) # only add a NullHandler for this logger, it is up to the application # to configure its own logging: logger.addHandler(logging.NullHandler()) logger.setLevel(level) return logger # a global logger object used for debugging: log = get_logger('ppt') def enable_logging(): """ Enable logging for this module (disabled by default). This will set the module-specific logger level to NOTSET, which means the main application controls the actual logging level. """ log.setLevel(logging.NOTSET) #--- CONSTANTS ---------------------------------------------------------------- # name of main stream MAIN_STREAM_NAME = 'PowerPoint Document' # URL and message to report issues: URL_OLEVBA_ISSUES = 'https://bitbucket.org/decalage/oletools/issues' MSG_OLEVBA_ISSUES = 'Please report this issue on %s' % URL_OLEVBA_ISSUES # === EXCEPTIONS ============================================================== class PptUnexpectedData(Exception): """ raise by PptParser if some field's value is not as expected """ def __init__(self, stream, field_name, found_value, expected_value): self.msg = \ 'In stream "{0}" for field "{1}" found value "{2}" but expected {3}!' \ .format(stream, field_name, found_value, expected_value) super(PptUnexpectedData, self).__init__(self.msg) # === HELPERS ================================================================= def read_1(stream): """ read 1 byte from stream """ return struct.unpack('<B', stream.read(1))[0] def read_2(stream): """ read 2 byte (short) from stream """ return struct.unpack('<H', stream.read(2))[0] def read_4(stream): """ read 4 byte (long) from stream """ return struct.unpack('<L', stream.read(4))[0] def check_value(name, value, expected): """ simplify verification of values in extract_from """ if isinstance(expected, (list, tuple)): if value not in expected: exp_str = '[' + ' OR '.join('{0:04X}'.format(val) for val in expected) + ']' raise PptUnexpectedData( 'Current User', name, '{0:04X}'.format(value), exp_str) elif expected != value: raise PptUnexpectedData( 'Current User', name, '{0:04X}'.format(value), '{0:04X}'.format(expected)) # === STRUCTS ================================================================= class RecordHeader(object): """ a record header, at start of many types found in ppt files https://msdn.microsoft.com/en-us/library/dd926377%28v=office.12%29.aspx https://msdn.microsoft.com/en-us/library/dd948895%28v=office.12%29.aspx """ def __init__(self): self.rec_ver = None self.rec_instance = None self.rec_type = None self.rec_len = None @classmethod def extract_from(clz, stream): """ reads 8 byte from stream """ #log.debug('parsing RecordHeader from stream') obj = clz() # first half byte is version, next 3 half bytes are instance version_instance, = struct.unpack('<H', stream.read(2)) obj.rec_instance, obj.rec_ver = divmod(version_instance, 2**4) obj.rec_type, = struct.unpack('<H', stream.read(2)) obj.rec_len, = struct.unpack('<L', stream.read(4)) #log.debug('type is {0:04X}, instance {1:04X}, version {2:04X},len {3}' # .format(obj.rec_type, obj.rec_instance, obj.rec_ver, # obj.rec_len)) return obj @classmethod def generate(clz, rec_type, rec_len=None, rec_instance=0, rec_ver=0): """ generate a record header string given values length of result depends on rec_len being given or not """ if rec_type is None: raise ValueError('RECORD_TYPE not set!') version_instance = rec_ver + 2**4 * rec_instance if rec_len is None: return struct.pack('<HH', version_instance, rec_type) else: return struct.pack('<HHL', version_instance, rec_type, rec_len) class PptType(object): """ base class of data types found in ppt ole files starts with a RecordHeader, has a extract_from and a check_validity method """ RECORD_TYPE = None # must be specified in subclasses RECORD_INSTANCE = 0x0 # can be overwritten in subclasses RECORD_VERSION = 0x000 # can be overwritten in subclasses @classmethod def extract_from(clz, stream): raise NotImplementedError('abstract base function!') def __init__(self, stream_name=MAIN_STREAM_NAME): self.stream_name = stream_name self.rec_head = None def read_rec_head(self, stream): self.rec_head = RecordHeader.extract_from(stream) def check_validity(self): """ check validity of data replaces 'raise PptUnexpectedData' so caller can get all the errors (not just the first) whenever she wishes. to be overwritten in subclasses :returns: list of PptUnexpectedData """ raise NotImplementedError('abstract base function!') def check_value(self, name, value, expected): """ simplify verification of values: check value equals/is in expected :returns: list of PptUnexpectedData exceptions """ if isinstance(expected, (list, tuple)): if value not in expected: clz_name = self.__class__.__name__ exp_str = '[' + ' OR '.join('{0:04X}'.format(val) for val in expected) + ']' return [PptUnexpectedData( self.stream_name, clz_name + '.' + name, '{0:04X}'.format(value), exp_str), ] elif expected != value: clz_name = self.__class__.__name__ return [PptUnexpectedData( self.stream_name, clz_name + '.' + name, '{0:04X}'.format(value), '{0:04X}'.format(expected)), ] return [] def check_range(self, name, value, expect_lower, expect_upper): """ simplify verification of values: check value is in given range expect_lower or expected_upper can be given as None to check only one boundary. If value equals one of the boundaries, that is also an error (boundaries form an open interval) :returns: list of PptUnexpectedData exceptions """ is_err = False if expect_upper is None and expect_lower is None: raise ValueError('need at least one non-None boundary!') if expect_lower is not None: if value <= expect_lower: is_err = True if expect_upper is not None: if value >= expect_upper: is_err = True if is_err: clz_name = self.__class__.__name__ if expect_lower is None: expect_str = '< {0:04X}'.format(expect_upper) elif expect_upper is None: expect_str = '> {0:04X}'.format(expect_lower) else: expect_str = 'within ({0:04X}, {1:04X})'.format(expect_lower, expect_upper) return [PptUnexpectedData(self.stream_name, clz_name + '.' + name, '{0:04X}'.format(value), expect_str), ] else: return [] def check_rec_head(self, length=None): """ to be called by check_validity to check the self.rec_head uses self.RECORD_... constants, (not quite that constant for DummyType) """ errs = [] errs.extend(self.check_value('rec_head.recVer', self.rec_head.rec_ver, self.RECORD_VERSION)) errs.extend(self.check_value('rec_head.recInstance', self.rec_head.rec_instance, self.RECORD_INSTANCE)) if self.RECORD_TYPE is None: raise NotImplementedError('RECORD_TYPE not specified!') errs.extend(self.check_value('rec_head.recType', self.rec_head.rec_type, self.RECORD_TYPE)) if length is not None: errs.extend(self.check_value('rec_head.recLen', self.rec_head.rec_len, length)) return errs @classmethod def generate_pattern(clz, rec_len=None): """ call RecordHeader.generate with values for this type """ return RecordHeader.generate(clz.RECORD_TYPE, rec_len, clz.RECORD_INSTANCE, clz.RECORD_VERSION) class CurrentUserAtom(PptType): """ An atom record that specifies information about the last user to modify the file and where the most recent user edit is located. This is the only record in the Current User Stream (section 2.1.1). https://msdn.microsoft.com/en-us/library/dd948895%28v=office.12%29.aspx """ # allowed values for header_token HEADER_TOKEN_ENCRYPT = 0xF3D1C4DF HEADER_TOKEN_NOCRYPT = 0xE391C05F # allowed values for rel_version REL_VERSION_CAN_USE = 0x00000008 REL_VERSION_NO_USE = 0x00000009 # required values RECORD_TYPE = 0x0FF6 SIZE = 0x14 DOC_FILE_VERSION = 0x03F4 MAJOR_VERSION = 0x03 MINOR_VERSION = 0x00 def __init__(self): super(CurrentUserAtom, self).__init__(stream_name='Current User') self.rec_head = None self.size = None self.header_token = None self.offset_to_current_edit = None self.len_user_name = None self.doc_file_version = None self.major_version = None self.minor_version = None self.ansi_user_name = None self.unicode_user_name = None self.rel_version = None def is_encrypted(self): return self.header_token == self.HEADER_TOKEN_ENCRYPT @classmethod def extract_from(clz, stream): """ create instance with info from stream """ obj = clz() # parse record header obj.rec_head = RecordHeader.extract_from(stream) obj.size, = struct.unpack('<L', stream.read(4)) obj.header_token, = struct.unpack('<L', stream.read(4)) obj.offset_to_current_edit, = struct.unpack('<L', stream.read(4)) obj.len_user_name, = struct.unpack('<H', stream.read(2)) obj.doc_file_version, = struct.unpack('<H', stream.read(2)) obj.major_version, = struct.unpack('<B', stream.read(1)) obj.minor_version, = struct.unpack('<B', stream.read(1)) stream.read(2) # unused obj.ansi_user_name = stream.read(obj.len_user_name) obj.rel_version, = struct.unpack('<L', stream.read(4)) obj.unicode_user_name = stream.read(2 * obj.len_user_name) return obj def check_validity(self): errs = self.check_rec_head() errs.extend(self.check_value('size', self.size, self.SIZE)) errs.extend(self.check_value('headerToken', self.header_token, [self.HEADER_TOKEN_ENCRYPT, self.HEADER_TOKEN_NOCRYPT])) errs.extend(self.check_range('lenUserName', self.len_user_name, None, 256)) errs.extend(self.check_value('docFileVersion', self.doc_file_version, self.DOC_FILE_VERSION)) errs.extend(self.check_value('majorVersion', self.major_version, self.MAJOR_VERSION)) errs.extend(self.check_value('minorVersion', self.minor_version, self.MINOR_VERSION)) errs.extend(self.check_value('relVersion', self.rel_version, [self.REL_VERSION_CAN_USE, self.REL_VERSION_NO_USE])) return errs class UserEditAtom(PptType): """ An atom record that specifies information about a user edit https://msdn.microsoft.com/en-us/library/dd945746%28v=office.12%29.aspx """ RECORD_TYPE = 0x0FF5 MINOR_VERSION = 0x00 MAJOR_VERSION = 0x03 def __init__(self): super(UserEditAtom, self).__init__() self.rec_head = None self.last_slide_id_ref = None self.version = None self.minor_version = None self.major_version = None self.offset_last_edit = None self.offset_persist_directory = None self.doc_persist_id_ref = None self.persist_id_seed = None self.last_view = None self.encrypt_session_persist_id_ref = None @classmethod def extract_from(clz, stream, is_encrypted): """ extract info from given stream (already positioned correctly!) """ log.debug('extract UserEditAtom from stream') obj = clz() # parse record header obj.rec_head = RecordHeader.extract_from(stream) obj.last_slide_id_ref, = struct.unpack('<L', stream.read(4)) obj.version, = struct.unpack('<H', stream.read(2)) obj.minor_version, = struct.unpack('<B', stream.read(1)) obj.major_version, = struct.unpack('<B', stream.read(1)) obj.offset_last_edit, = struct.unpack('<L', stream.read(4)) obj.offset_persist_directory, = struct.unpack('<L', stream.read(4)) obj.doc_persist_id_ref, = struct.unpack('<L', stream.read(4)) obj.persist_id_seed, = struct.unpack('<L', stream.read(4)) # (can only check once have the PersistDirectoryAtom) obj.last_view, = struct.unpack('<H', stream.read(2)) stream.read(2) # unused if is_encrypted: obj.encrypt_session_persist_id_ref, = \ struct.unpack('<L', stream.read(4)) else: # this entry may be there or may not obj.encrypt_session_persist_id_ref = None return obj def check_validity(self, offset=None): errs = self.check_rec_head() errs.extend(self.check_value('minorVersion', self.minor_version, self.MINOR_VERSION)) errs.extend(self.check_value('majorVersion', self.major_version, self.MAJOR_VERSION)) if offset is not None: if self.offset_last_edit >= offset: errs.append(PptUnexpectedData( 'PowerPoint Document', 'UserEditAtom.offsetLastEdit', self.offset_last_edit, '< {0}'.format(offset))) if self.offset_persist_directory >= offset or \ self.offset_persist_directory <= self.offset_last_edit: errs.append(PptUnexpectedData( 'PowerPoint Document', 'UserEditAtom.offsetPersistDirectory', self.offset_last_edit, 'in ({0}, {1})'.format(self.offset_last_edit, offset))) errs.extend(self.check_value('docPersistIdRef', self.doc_persist_id_ref, 1)) return errs # TODO: offer to check persist_id_seed given PersistDirectoryAtom) class DummyType(PptType): """ a type that is found in ppt documents we are not interested in instead of parsing many uninteresting types, we just read their RecordHeader and set the RECORD_... values on an instance- (instead of class-) level used to skip over uninteresting types in e.g. DocumentContainer """ def __init__(self, type_name, record_type, rec_ver=0, rec_instance=0, rec_len=None): super(DummyType, self).__init__() self.type_name = type_name self.RECORD_TYPE = record_type self.RECORD_VERSION = rec_ver self.RECORD_INSTANCE = rec_instance self.record_length = rec_len def extract_from(self, stream): """ extract record header and just skip as many bytes as header says Since this requires RECORD_... values set in constructor, this is NOT a classmethod like all the other extract_from! Otherwise this tries to be compatible with other extract_from methods (e.g. returns self) """ self.read_rec_head(stream) log.debug('skipping over {0} Byte for type {1}' .format(self.rec_head.rec_len, self.type_name)) log.debug('start at pos {0}'.format(stream.tell())) stream.seek(self.rec_head.rec_len, os.SEEK_CUR) log.debug('now at pos {0}'.format(stream.tell())) return self def check_validity(self): return self.check_rec_head(self.record_length) class PersistDirectoryAtom(PptType): """ one part of a persist object directory with unique persist object id contains PersistDirectoryEntry objects https://msdn.microsoft.com/en-us/library/dd952680%28v=office.12%29.aspx """ RECORD_TYPE = 0x1772 def __init__(self): super(PersistDirectoryAtom, self).__init__() self.rg_persist_dir_entry = None # actually, this will be an array self.stream_offset = None @classmethod def extract_from(clz, stream): """ create and return object with data from given stream """ log.debug("Extracting a PersistDirectoryAtom from stream") obj = clz() # remember own offset for checking validity obj.stream_offset = stream.tell() # parse record header obj.read_rec_head(stream) # read directory entries from list until reach size for this object curr_pos = stream.tell() stop_pos = curr_pos + obj.rec_head.rec_len log.debug('start reading at pos {0}, read until {1}' .format(curr_pos, stop_pos)) obj.rg_persist_dir_entry = [] while curr_pos < stop_pos: new_entry = PersistDirectoryEntry.extract_from(stream) obj.rg_persist_dir_entry.append(new_entry) curr_pos = stream.tell() log.debug('at pos {0}'.format(curr_pos)) return obj def check_validity(self, user_edit_last_offset=None): errs = self.check_rec_head() for entry in self.rg_persist_dir_entry: errs.extend(entry.check_validity(user_edit_last_offset, self.stream_offset)) return errs class PersistDirectoryEntry(object): """ an entry contained in a PersistDirectoryAtom.rg_persist_dir_entry A structure that specifies a compressed table of sequential persist object identifiers and stream offsets to associated persist objects. NOT a subclass of PptType because has no RecordHeader https://msdn.microsoft.com/en-us/library/dd947347%28v=office.12%29.aspx """ def __init__(self): self.persist_id = None self.c_persist = None self.rg_persist_offset = None @classmethod def extract_from(clz, stream): # take a 4-byte (=32bit) number, divide into 20bit and 12 bit) log.debug("Extracting a PersistDirectoryEntry from stream") obj = clz() # persistId (20 bits): An unsigned integer that specifies a starting # persist object identifier. It MUST be less than or equal to 0xFFFFE. # The first entry in rgPersistOffset is associated with persistId. The # next entry, if present, is associated with persistId plus 1. Each # entry in rgPersistOffset is associated with a persist object # identifier in this manner, with the final entry associated with # persistId + cPersist - 1. # cPersist (12 bits): An unsigned integer that specifies the count of # items in rgPersistOffset. It MUST be greater than or equal to 0x001. temp, = struct.unpack('<L', stream.read(4)) obj.c_persist, obj.persist_id = divmod(temp, 2**20) log.debug('temp is 0x{0:04X} --> id is {1}, reading {2} offsets' .format(temp, obj.persist_id, obj.c_persist)) # rgPersistOffset (variable): An array of PersistOffsetEntry (section # 2.3.6) that specifies stream offsets to persist objects. The count of # items in the array is specified by cPersist. The value of each item # MUST be greater than or equal to offsetLastEdit in the corresponding # user edit and MUST be less than the offset, in bytes, of the # corresponding persist object directory. # PersistOffsetEntry: An unsigned 4-byte integer that specifies an # offset, in bytes, from the beginning of the PowerPoint Document # Stream (section 2.1.2) to a persist object. obj.rg_persist_offset = [struct.unpack('<L', stream.read(4))[0] \ for _ in range(obj.c_persist)] log.debug('offsets are: {0}'.format(obj.rg_persist_offset)) return obj def check_validity(self, user_edit_last_offset=None, persist_obj_dir_offset=None): errs = [] if self.persist_id > 0xFFFFE: # (--> == 0xFFFFF since 20bit) errs.append(PptUnexpectedData( MAIN_STREAM_NAME, 'PersistDirectoryEntry.persist_id', self.persist_id, '< 0xFFFFE (dec: {0})'.format(0xFFFFE))) if self.c_persist == 0: errs.append(PptUnexpectedData( MAIN_STREAM_NAME, 'PersistDirectoryEntry.c_persist', self.c_persist, '> 0')) if user_edit_last_offset is not None \ and min(self.rg_persist_offset) < user_edit_last_offset: errs.append(PptUnexpectedData( MAIN_STREAM_NAME, 'PersistDirectoryEntry.rg_persist_offset', min(self.rg_persist_offset), '> UserEdit.offsetLastEdit ({0})' .format(user_edit_last_offset))) if persist_obj_dir_offset is not None \ and max(self.rg_persist_offset) > persist_obj_dir_offset: errs.append(PptUnexpectedData( MAIN_STREAM_NAME, 'PersistDirectoryEntry.rg_persist_offset', max(self.rg_persist_offset), '> PersistObjectDirectory offset ({0})' .format(persist_obj_dir_offset))) return errs class DocInfoListSubContainerOrAtom(PptType): """ one of various types found in a DocInfoListContainer https://msdn.microsoft.com/en-us/library/dd921705%28v=office.12%29.aspx actual type of this object is defined by the recVersion field in its Record Head Similar to DummyType, RECORD_TYPE varies from instance to instance for this type """ # RECORD_TYPE varies, is specified only in extract_from VALID_RECORD_TYPES = [0x1388, # self.RECORD_TYPE_PROG_TAGS, \ 0x0414, # self.RECORD_TYPE_NORMAL_VIEW_SET_INFO_9, \ 0x0413, # self.RECORD_TYPE_NOTES_TEXT_VIEW_INFO_9, \ 0x0407, # self.RECORD_TYPE_OUTLINE_VIEW_INFO, \ 0x03FA, # self.RECORD_TYPE_SLIDE_VIEW_INFO, \ 0x0408] # self.RECORD_TYPE_SORTER_VIEW_INFO def __init__(self): super(DocInfoListSubContainerOrAtom, self).__init__() @classmethod def extract_from(clz, stream): """ build instance with info read from stream """ log.debug('Parsing DocInfoListSubContainerOrAtom from stream') obj = clz() obj.read_rec_head(stream) if obj.rec_head.rec_type == VBAInfoContainer.RECORD_TYPE: obj = VBAInfoContainer.extract_from(stream, obj.rec_head) else: log.debug('skipping over {0} Byte in DocInfoListSubContainerOrAtom' .format(obj.rec_head.rec_len)) log.debug('start at pos {0}'.format(stream.tell())) stream.seek(obj.rec_head.rec_len, os.SEEK_CUR) log.debug('now at pos {0}'.format(stream.tell())) return obj def check_validity(self): """ can be any of multiple types """ self.check_value('rh.recType', self.rec_head.rec_type, self.VALID_RECORD_TYPES) class DocInfoListContainer(PptType): """ information about the document and document display settings https://msdn.microsoft.com/en-us/library/dd926767%28v=office.12%29.aspx """ RECORD_VERSION = 0xF RECORD_TYPE = 0x07D0 def __init__(self): super(DocInfoListContainer, self).__init__() self.rg_child_rec = None @classmethod def extract_from(clz, stream): """ build instance with info read from stream """ log.debug('Parsing DocInfoListContainer from stream') obj = clz() obj.read_rec_head(stream) # rgChildRec (variable): An array of DocInfoListSubContainerOrAtom # records (section 2.4.5) that specifies information about the document # or how the document is displayed. The size, in bytes, of the array is # specified by rh.recLen curr_pos = stream.tell() end_pos = curr_pos + obj.rec_head.rec_len log.debug('start reading at pos {0}, will read until {1}' .format(curr_pos, end_pos)) obj.rg_child_rec = [] while curr_pos < end_pos: new_obj = DocInfoListSubContainerOrAtom().extract_from(stream) obj.rg_child_rec.append(new_obj) curr_pos = stream.tell() log.debug('now at pos {0}'.format(curr_pos)) log.debug('reached end pos {0} ({1}). stop reading DocInfoListContainer' .format(end_pos, curr_pos)) def check_validity(self): errs = self.check_rec_head() for obj in self.rg_child_rec: errs.extend(obj.check_validity()) return errs class DocumentContainer(PptType): """ a DocumentContainer record https://msdn.microsoft.com/en-us/library/dd947357%28v=office.12%29.aspx """ RECORD_TYPE = 0x03E8 def __init__(self): super(DocumentContainer, self).__init__() self.document_atom = None self.ex_obj_list = None self.document_text_info = None self.sound_collection = None self.drawing_group = None self.master_list = None self.doc_info_list = None self.slide_hf = None self.notes_hf = None self.slide_list = None self.notes_list = None self.slide_show_doc_info = None self.named_shows = None self.summary = None self.doc_routing_slip = None self.print_options = None self.rt_custom_table_styles_1 = None self.end_document = None self.rt_custom_table_styles_2 = None @classmethod def extract_from(clz, stream): """ created object with values from given stream stream is assumed to be positioned correctly this container contains lots of data we are not interested in. """ log.debug('Parsing DocumentContainer from stream') obj = clz() # parse record header obj.read_rec_head(stream) log.info('validity: {0} errs'.format(len(obj.check_rec_head()))) # documentAtom (48 bytes): A DocumentAtom record (section 2.4.2) that # specifies size information for presentation slides and notes slides. obj.document_atom = DummyType('DocumentAtom', 0x03E9, rec_ver=0x1, rec_len=0x28).extract_from(stream) log.info('validity: {0} errs' .format(len(obj.document_atom.check_validity()))) # exObjList (variable): An optional ExObjListContainer record (section # 2.10.1) that specifies the list of external objects in the document. obj.ex_obj_list = DummyType('ExObjListContainer', 0x0409, rec_ver=0xF)\ .extract_from(stream) log.info('validity: {0} errs' .format(len(obj.ex_obj_list.check_validity()))) # documentTextInfo (variable): A DocumentTextInfoContainer record # (section 2.9.1) that specifies the default text styles for the # document. obj.document_text_info = DummyType('DocumentTextInfoContainer', 0x03F2, rec_ver=0xF).extract_from(stream) log.info('validity: {0} errs' .format(len(obj.document_text_info.check_validity()))) # soundCollection (variable): An optional SoundCollectionContainer # record (section 2.4.16.1) that specifies the list of sounds in the # file. obj.sound_collection = DummyType('SoundCollectionContainer', 0x07E4, rec_ver=0xF, rec_instance=0x005)\ .extract_from(stream) log.info('validity: {0} errs' .format(len(obj.sound_collection.check_validity()))) # drawingGroup (variable): A DrawingGroupContainer record (section # 2.4.3) that specifies drawing information for the document. obj.drawing_group = DummyType('DrawingGroupContainer', 0x040B, rec_ver=0xF).extract_from(stream) log.info('validity: {0} errs' .format(len(obj.drawing_group.check_validity()))) # masterList (variable): A MasterListWithTextContainer record (section # 2.4.14.1) that specifies the list of main master slides and title # master slides. obj.master_list = DummyType('MasterListWithContainer', 0x0FF0, rec_ver=0xF).extract_from(stream) log.info('validity: {0} errs' .format(len(obj.master_list.check_validity()))) # docInfoList (variable): An optional DocInfoListContainer record # (section 2.4.4) that specifies additional document information. # this is the variable we are interested in! obj.doc_info_list = DocInfoListContainer.extract_from(stream) # slideHF (variable): An optional SlideHeadersFootersContainer record # (section 2.4.15.1) that specifies the default header and footer # information for presentation slides. #obj.slide_hf = None # notesHF (variable): An optional NotesHeadersFootersContainer record # (section 2.4.15.6) that specifies the default header and footer # information for notes slides. #obj.notes_hf = None # slideList (variable): An optional SlideListWithTextContainer record # (section 2.4.14.3) that specifies the list of presentation slides. #obj.slide_list = None # notesList (variable): An optional NotesListWithTextContainer record # (section 2.4.14.6) that specifies the list of notes slides. #obj.notes_list = None # slideShowDocInfoAtom (88 bytes): An optional SlideShowDocInfoAtom # record (section 2.6.1) that specifies slide show information for the # document. #obj.slide_show_doc_info = None # namedShows (variable): An optional NamedShowsContainer record # (section 2.6.2) that specifies named shows in the document. #obj.named_shows = None # summary (variable): An optional SummaryContainer record (section # 2.4.22.3) that specifies bookmarks for the document. #obj.summary = None # docRoutingSlipAtom (variable): An optional DocRoutingSlipAtom record # (section 2.11.1) that specifies document routing information. #obj.doc_routing_slip = None # printOptionsAtom (13 bytes): An optional PrintOptionsAtom record # (section 2.4.12) that specifies default print options. #obj.print_options = None # rtCustomTableStylesAtom1 (variable): An optional # RoundTripCustomTableStyles12Atom record (section 2.11.13) that # specifies round-trip information for custom table styles. #obj.rt_custom_table_styles_1 = None # endDocumentAtom (8 bytes): An EndDocumentAtom record (section 2.4.13) # that specifies the end of the information for the document. #obj.end_document = None # rtCustomTableStylesAtom2 (variable): An optional # RoundTripCustomTableStyles12Atom record that specifies round-trip # information for custom table styles. It MUST NOT exist if # rtCustomTableStylesAtom1 exists. #obj.rt_custom_table_styles_2 = None return obj def check_validity(self): """ check all values in object for valid values """ errs = self.check_rec_head() errs.extend(self.document_atom.check_validity()) errs.extend(self.ex_obj_list.check_validity()) errs.extend(self.document_text_info.check_validity()) errs.extend(self.sound_collection.check_validity()) errs.extend(self.drawing_group.check_validity()) errs.extend(self.master_list.check_validity()) errs.extend(self.doc_info_list.check_validity()) return errs class VBAInfoContainer(PptType): """ A container record that specifies VBA information for the document. https://msdn.microsoft.com/en-us/library/dd952168%28v=office.12%29.aspx """ RECORD_TYPE = 0x03FF RECORD_VERSION = 0xF RECORD_INSTANCE = 0x001 RECORD_LENGTH = 0x14 def __init__(self): super(VBAInfoContainer, self).__init__() self.vba_info_atom = None @classmethod def extract_from(clz, stream, rec_head=None): """ since can determine this type only after reading header, it is arg """ log.debug('parsing VBAInfoContainer') obj = clz() if rec_head is None: obj.read_rec_head(stream) else: log.debug('skip parsing of RecordHeader') obj.rec_head = rec_head obj.vba_info_atom = VBAInfoAtom.extract_from(stream) return obj def check_validity(self): errs = self.check_rec_head(length=self.RECORD_LENGTH) errs.extend(self.vba_info_atom.check_validity()) return errs class VBAInfoAtom(PptType): """ An atom record that specifies a reference to the VBA project storage. https://msdn.microsoft.com/en-us/library/dd948874%28v=office.12%29.aspx """ RECORD_TYPE = 0x0400 RECORD_VERSION = 0x2 RECORD_LENGTH = 0x0C def __init__(self): super(VBAInfoAtom, self).__init__() self.persist_id_ref = None self.f_has_macros = None self.version = None @classmethod def extract_from(clz, stream): log.debug('parsing VBAInfoAtom') obj = clz() obj.read_rec_head(stream) # persistIdRef (4 bytes): A PersistIdRef (section 2.2.21) that # specifies the value to look up in the persist object directory to # find the offset of a VbaProjectStg record (section 2.10.40). obj.persist_id_ref = read_4(stream) # fHasMacros (4 bytes): An unsigned integer that specifies whether the # VBA project storage contains data. It MUST be 0 (empty vba storage) # or 1 (vba storage contains data) obj.f_has_macros = read_4(stream) # version (4 bytes): An unsigned integer that specifies the VBA runtime # version that generated the VBA project storage. It MUST be # 0x00000002. obj.version = read_4(stream) return obj def check_validity(self): errs = self.check_rec_head(length=self.RECORD_LENGTH) # must be 0 or 1: errs.extend(self.check_range('fHasMacros', self.f_has_macros, None, 2)) errs.extend(self.check_value('version', self.version, 2)) return errs class ExternalObjectStorage(PptType): """ storage for compressed/uncompressed OLE/VBA/ActiveX control data Matches types ExOleObjStgCompressedAtom, ExOleObjStgUncompressedAtom, VbaProjectStgCompressedAtom, VbaProjectStgUncompressedAtom, ExControlStgUncompressedAtom, ExControlStgCompressedAtom Difference between compressed and uncompressed: RecordHeader.rec_instance is 0 or 1, first variable after RecordHeader is decompressed_size Data is not read at first, only its offset in the stream and size is saved e.g. https://msdn.microsoft.com/en-us/library/dd952169%28v=office.12%29.aspx """ RECORD_TYPE = 0x1011 RECORD_INSTANCE_COMPRESSED = 1 RECORD_INSTANCE_UNCOMPRESSED = 0 def __init__(self, is_compressed=None): super(ExternalObjectStorage, self).__init__() if is_compressed is None: self.RECORD_INSTANCE = None # otherwise defaults to 0 elif is_compressed: self.RECORD_INSTANCE = self.RECORD_INSTANCE_COMPRESSED self.is_compressed = True else: self.RECORD_INSTANCE = self.RECORD_INSTANCE_UNCOMPRESSED self.is_compressed = False self.uncompressed_size = None self.data_offset = None self.data_size = None def extract_from(self, stream): """ not a classmethod because of is_compressed attrib see also: DummyType """ log.debug('Parsing ExternalObjectStorage (compressed={0}) from stream' .format(self.is_compressed)) self.read_rec_head(stream) self.data_size = self.rec_head.rec_len if self.is_compressed: self.uncompressed_size = read_4(stream) self.data_size -= 4 self.data_offset = stream.tell() def check_validity(self): return self.check_rec_head() class ExternalObjectStorageUncompressed(ExternalObjectStorage): """ subclass of ExternalObjectStorage for uncompressed objects """ RECORD_INSTANCE = ExternalObjectStorage.RECORD_INSTANCE_UNCOMPRESSED def __init__(self): super(ExternalObjectStorageUncompressed, self).__init__(False) @classmethod def extract_from(clz, stream): """ note the usage of super here: call instance method of super class! """ obj = clz() super(ExternalObjectStorageUncompressed, obj).extract_from(stream) return obj class ExternalObjectStorageCompressed(ExternalObjectStorage): """ subclass of ExternalObjectStorage for compressed objects """ RECORD_INSTANCE = ExternalObjectStorage.RECORD_INSTANCE_COMPRESSED def __init__(self): super(ExternalObjectStorageCompressed, self).__init__(True) @classmethod def extract_from(clz, stream): """ note the usage of super here: call instance method of super class! """ obj = clz() super(ExternalObjectStorageCompressed, obj).extract_from(stream) return obj # === PptParser =============================================================== def with_opened_main_stream(func): """ a decorator that can open and close the default stream for func to be applied only to functions in PptParser that read from default stream (:py:data:`MAIN_STREAM_NAME`) Decorated functions need to accept args (self, stream, ...) """ def wrapped(self, *args, **kwargs): # remember who opened the stream so that function also closes it stream_opened_by_me = False try: # open stream if required if self._open_main_stream is None: log.debug('opening stream {0!r} for {1}' .format(MAIN_STREAM_NAME, func.__name__)) self._open_main_stream = self.ole.openstream(MAIN_STREAM_NAME) stream_opened_by_me = True # run wrapped function return func(self, self._open_main_stream, *args, **kwargs) # error handling except Exception: if self.fast_fail: raise else: self._log_exception() finally: # ensure stream is closed by the one who opened it (even if error) if stream_opened_by_me: log.debug('closing stream {0!r} after {1}' .format(MAIN_STREAM_NAME, func.__name__)) self._open_main_stream.close() self._open_main_stream = None return wrapped def generator_with_opened_main_stream(func): """ same as with_opened_main_stream but with yield instead of return """ def wrapped(self, *args, **kwargs): # remember who opened the stream so that function also closes it stream_opened_by_me = False try: # open stream if required if self._open_main_stream is None: log.debug('opening stream {0!r} for {1}' .format(MAIN_STREAM_NAME, func.__name__)) self._open_main_stream = self.ole.openstream(MAIN_STREAM_NAME) stream_opened_by_me = True # run actual function for result in func(self, self._open_main_stream, *args, **kwargs): yield result # error handling except Exception: if self.fast_fail: raise else: self._log_exception() finally: # ensure stream is closed by the one who opened it (even if error) if stream_opened_by_me: log.debug('closing stream {0!r} after {1}' .format(MAIN_STREAM_NAME, func.__name__)) self._open_main_stream.close() self._open_main_stream = None return wrapped class PptParser(object): """ Parser for PowerPoint 97-2003 specific data structures requires an OleFileIO """ def __init__(self, ole, fast_fail=False): """ constructor :param ole: OleFileIO or anything that OleFileIO constructor accepts :param bool fast_fail: if True, all unexpected data will raise a PptUnexpectedData; if False will only log error """ if isinstance(ole, olefile.OleFileIO): log.debug('using open OleFileIO') self.ole = ole else: log.debug('Opening file {0}'.format(ole)) self.ole = olefile.OleFileIO(ole) self.fast_fail = fast_fail self.current_user_atom = None self.newest_user_edit = None self.document_persist_obj = None self.persist_object_directory = None # basic compatibility check: root directory structure is # [['\x05DocumentSummaryInformation'], # ['\x05SummaryInformation'], # ['Current User'], # ['PowerPoint Document']] root_streams = self.ole.listdir() #for stream in root_streams: # log.debug('found root stream {0!r}'.format(stream)) if any(len(stream) != 1 for stream in root_streams): self._fail('root', 'listdir', root_streams, 'len = 1') root_streams = [stream[0].lower() for stream in root_streams] if not 'current user' in root_streams: self._fail('root', 'listdir', root_streams, 'Current User') if not MAIN_STREAM_NAME.lower() in root_streams: self._fail('root', 'listdir', root_streams, MAIN_STREAM_NAME) self._open_main_stream = None def _log_exception(self, msg=None): """ log an exception instead of raising it call in one of 2 ways: try: if fail(): self._log_exception('this is the message') except: self._log_exception() # only possible in except clause """ if msg is not None: stack = traceback.extract_stack()[:-1] else: _, exc, trace = sys.exc_info() stack = traceback.extract_tb(trace) msg = str(exc) log.error(msg) for i_entry, entry in enumerate(traceback.format_list(stack)): for line in entry.splitlines(): log.debug('trace {0}: {1}'.format(i_entry, line)) def _fail(self, *args): """ depending on self.fast_fail raise PptUnexpectedData or just log err args as for PptUnexpectedData """ if self.fast_fail: raise PptUnexpectedData(*args) else: self._log_exception(PptUnexpectedData(*args).msg) def parse_current_user(self): """ parse the CurrentUserAtom record from stream 'Current User' Structure described in https://msdn.microsoft.com/en-us/library/dd948895%28v=office.12%29.aspx """ if self.current_user_atom is not None: log.warning('re-reading and overwriting ' 'previously read current_user_atom') log.debug('parsing "Current User"') stream = None try: log.debug('opening stream "Current User"') stream = self.ole.openstream('Current User') self.current_user_atom = CurrentUserAtom.extract_from(stream) except Exception: if self.fast_fail: raise else: self._log_exception() finally: if stream is not None: log.debug('closing stream "Current User"') stream.close() @with_opened_main_stream def parse_persist_object_directory(self, stream): """ Part 1: Construct the persist object directory """ if self.persist_object_directory is not None: log.warning('re-reading and overwriting ' 'previously read persist_object_directory') # Step 1: Read the CurrentUserAtom record (section 2.3.2) from the # Current User Stream (section 2.1.1). All seek operations in the steps # that follow this step are in the PowerPoint Document Stream. if self.current_user_atom is None: self.parse_current_user() offset = self.current_user_atom.offset_to_current_edit is_encrypted = self.current_user_atom.is_encrypted() self.persist_object_directory = {} self.newest_user_edit = None # Repeat steps 3 through 6 until offsetLastEdit is 0x00000000. while offset != 0: # Step 2: Seek, in the PowerPoint Document Stream, to the # offset specified by the offsetToCurrentEdit field of the # CurrentUserAtom record identified in step 1. stream.seek(offset, os.SEEK_SET) # Step 3: Read the UserEditAtom record at the current offset. # Let this record be a live record. user_edit = UserEditAtom.extract_from(stream, is_encrypted) if self.newest_user_edit is None: self.newest_user_edit = user_edit log.debug('checking validity') errs = user_edit.check_validity() if errs: log.warning('check_validity found {0} issues' .format(len(errs))) for err in errs: log.warning('UserEditAtom.check_validity: {0}'.format(err)) if errs and self.fast_fail: raise errs[0] # Step 4: Seek to the offset specified by the # offsetPersistDirectory field of the UserEditAtom record # identified in step 3. log.debug('seeking to pos {0}' .format(user_edit.offset_persist_directory)) stream.seek(user_edit.offset_persist_directory, os.SEEK_SET) # Step 5: Read the PersistDirectoryAtom record at the current # offset. Let this record be a live record. persist_dir_atom = PersistDirectoryAtom.extract_from(stream) log.debug('checking validity') errs = persist_dir_atom.check_validity(offset) if errs: log.warning('check_validity found {0} issues' .format(len(errs))) for err in errs: log.warning('PersistDirectoryAtom.check_validity: {0}' .format(err)) if errs and self.fast_fail: raise errs[0] # Construct the complete persist object directory for this file # as follows: # - For each PersistDirectoryAtom record previously identified # in step 5, add the persist object identifier and persist # object stream offset pairs to the persist object directory # starting with the PersistDirectoryAtom record last # identified, that is, the one closest to the beginning of the # stream. # - Continue adding these pairs to the persist object directory # for each PersistDirectoryAtom record in the reverse order # that they were identified in step 5; that is, the pairs from # the PersistDirectoryAtom record closest to the end of the # stream are added last. # - When adding a new pair to the persist object directory, if # the persist object identifier already exists in the persist # object directory, the persist object stream offset from the # new pair replaces the existing persist object stream offset # for that persist object identifier. for entry in persist_dir_atom.rg_persist_dir_entry: last_id = entry.persist_id+len(entry.rg_persist_offset)-1 log.debug('for persist IDs {0}-{1}, save offsets {2}' .format(entry.persist_id, last_id, entry.rg_persist_offset)) for count, offset in enumerate(entry.rg_persist_offset): self.persist_object_directory[entry.persist_id+count] \ = offset # check for more # Step 6: Seek to the offset specified by the offsetLastEdit # field in the UserEditAtom record identified in step 3. offset = user_edit.offset_last_edit @with_opened_main_stream def parse_document_persist_object(self, stream): """ Part 2: Identify the document persist object """ if self.document_persist_obj is not None: log.warning('re-reading and overwriting ' 'previously read document_persist_object') # Step 1: Read the docPersistIdRef field of the UserEditAtom record # first identified in step 3 of Part 1, that is, the UserEditAtom # record closest to the end of the stream. if self.persist_object_directory is None: self.parse_persist_object_directory() # pylint: disable=no-value-for-parameter # Step 2: Lookup the value of the docPersistIdRef field in the persist # object directory constructed in step 8 of Part 1 to find the stream # offset of a persist object. newest_ref = self.newest_user_edit.doc_persist_id_ref offset = self.persist_object_directory[newest_ref] log.debug('newest user edit ID is {0}, offset is {1}' .format(newest_ref, offset)) # Step 3: Seek to the stream offset specified in step 2. log.debug('seek to {0}'.format(offset)) stream.seek(offset, os.SEEK_SET) # Step 4: Read the DocumentContainer record at the current offset. # Let this record be a live record. self.document_persist_obj = DocumentContainer.extract_from(stream) log.debug('checking validity') errs = self.document_persist_obj.check_validity() if errs: log.warning('check_validity found {0} issues'.format(len(errs))) for err in errs: log.warning('check_validity(document_persist_obj): {0}' .format(err)) if errs and self.fast_fail: raise errs[0] #-------------------------------------------------------------------------- # 2nd attempt: do not parse whole structure but search through stream and # yield results as they become available # Keep in mind that after every yield the stream position may be anything! @generator_with_opened_main_stream def search_pattern(self, stream, pattern): """ search for pattern in stream, return indices """ BUF_SIZE = 1024 pattern_len = len(pattern) log.debug('pattern length is {0}'.format(pattern_len)) if pattern_len > BUF_SIZE: raise ValueError('need buf > pattern to search!') n_reads = 0 while True: start_pos = stream.tell() n_reads += 1 #log.debug('read {0} starting from {1}' # .format(BUF_SIZE, start_pos)) buf = stream.read(BUF_SIZE) idx = buf.find(pattern) while idx != -1: log.debug('found pattern at index {0}'.format(start_pos+idx)) yield start_pos + idx idx = buf.find(pattern, idx+1) if len(buf) == BUF_SIZE: # move back a bit to avoid splitting of pattern through buf stream.seek(start_pos + BUF_SIZE - pattern_len, os.SEEK_SET) else: log.debug('reached end of buf (read {0}<{1}) after {2} reads' .format(len(buf), BUF_SIZE, n_reads)) break @generator_with_opened_main_stream def search_vba_info(self, stream): """ search through stream for VBAInfoContainer, alternative to parse... quick-and-dirty: do not parse everything, just look for right bytes "quick" here means quick to program. Runtime now is linear is document size (--> for big documents the other method might be faster) .. seealso:: search_vba_storage """ log.debug('looking for VBA info containers') pattern = VBAInfoContainer.generate_pattern( rec_len=VBAInfoContainer.RECORD_LENGTH) \ + VBAInfoAtom.generate_pattern( rec_len=VBAInfoAtom.RECORD_LENGTH) # try parse for idx in self.search_pattern(pattern): # pylint: disable=no-value-for-parameter # assume that in stream at idx there is a VBAInfoContainer stream.seek(idx) log.debug('extracting at idx {0}'.format(idx)) try: container = VBAInfoContainer.extract_from(stream) except Exception: self._log_exception() continue errs = container.check_validity() if errs: log.warning('check_validity found {0} issues' .format(len(errs))) else: log.debug('container is ok') atom = container.vba_info_atom log.debug('persist id ref is {0}, has_macros {1}, version {2}' .format(atom.persist_id_ref, atom.f_has_macros, atom.version)) yield container for err in errs: log.warning('check_validity(VBAInfoContainer): {0}' .format(err)) if errs and self.fast_fail: raise errs[0] @generator_with_opened_main_stream def search_vba_storage(self, stream): """ search through stream for VBAProjectStg, alternative to parse... quick-and-dirty: do not parse everything, just look for right bytes "quick" here means quick to program. Runtime now is linear is document size (--> for big documents the other method might be faster) The storages found could also contain (instead of VBA data): ActiveX data or general OLE data yields results as it finds them .. seealso:: :py:meth:`search_vba_info` """ log.debug('looking for VBA storage objects') for obj_type in (ExternalObjectStorageUncompressed, ExternalObjectStorageCompressed): # re-position stream at start stream.seek(0, os.SEEK_SET) pattern = obj_type.generate_pattern() # try parse for idx in self.search_pattern(pattern): # pylint: disable=no-value-for-parameter # assume a ExternalObjectStorage in stream at idx stream.seek(idx) log.debug('extracting at idx {0}'.format(idx)) try: storage = obj_type.extract_from(stream) except Exception: self._log_exception() continue errs = storage.check_validity() if errs: log.warning('check_validity found {0} issues' .format(len(errs))) else: log.debug('storage is ok; compressed={0}, size={1}, ' 'size_decomp={2}' .format(storage.is_compressed, storage.rec_head.rec_len, storage.uncompressed_size)) yield storage for err in errs: log.warning('check_validity({0}): {1}' .format(obj_type.__name__, err)) if errs and self.fast_fail: raise errs[0] @with_opened_main_stream def decompress_vba_storage(self, stream, storage): """ return decompressed data from search_vba_storage """ log.debug('decompressing storage for VBA OLE data stream ') # decompress iteratively; a zlib.decompress of all data # failed with Error -5 (incomplete or truncated stream) stream.seek(storage.data_offset, os.SEEK_SET) decomp, n_read, err = \ iterative_decompress(stream, storage.data_size) log.debug('decompressed {0} to {1} bytes; found err: {2}' .format(n_read, len(decomp), err)) if err and self.fast_fail: raise err # otherwise try to continue with partial data return decomp ## create OleFileIO from decompressed data #ole = olefile.OleFileIO(decomp) #root_streams = [entry[0].lower() for entry in ole.listdir()] #for required in 'project', 'projectwm', 'vba': # if required not in root_streams: # raise ValueError('storage seems to not be a VBA storage ' # '({0} not found in root streams)' # .format(required)) #log.debug('tests succeeded') #return ole @with_opened_main_stream def read_vba_storage_data(self, stream, storage): """ return data pointed to by uncompressed storage """ log.debug('reading uncompressed VBA OLE data stream: ' '{0} bytes starting at {1}' .format(storage.data_size, storage.data_offset)) stream.seek(storage.data_offset, os.SEEK_SET) data = stream.read(storage.data_size) return data @generator_with_opened_main_stream def iter_vba_data(self, stream): """ search vba infos and storages, yield uncompressed storage data """ n_infos = 0 n_macros = 0 for info in self.search_vba_info(): # pylint: disable=no-value-for-parameter n_infos += 1 if info.vba_info_atom.f_has_macros > 0: n_macros += 1 # TODO: does it make sense at all to continue if n_macros == 0? # --> no vba-info, so all storages probably ActiveX or other OLE n_storages = 0 n_compressed = 0 for storage in self.search_vba_storage(): # pylint: disable=no-value-for-parameter n_storages += 1 if storage.is_compressed: n_compressed += 1 yield self.decompress_vba_storage(storage) # pylint: disable=no-value-for-parameter else: yield self.read_vba_storage_data(storage) # pylint: disable=no-value-for-parameter log.info('found {0} infos ({1} with macros) and {2} storages ' '({3} compressed)' .format(n_infos, n_macros, n_storages, n_compressed)) def iterative_decompress(stream, size, chunk_size=4096): """ decompress data from stream chunk-wise """ decompressor = zlib.decompressobj() n_read = 0 decomp = b'' return_err = None try: while n_read < size: n_new = min(size-n_read, chunk_size) decomp += decompressor.decompress(stream.read(n_new)) n_read += n_new except zlib.error as err: return_err = err return decomp, n_read, return_err if __name__ == '__main__': print('nothing here to run!')
Memory