277 lines
9.5 KiB
Python
277 lines
9.5 KiB
Python
import typing
|
|
|
|
import os
|
|
|
|
from . import dna_io, header, exceptions
|
|
|
|
# Either a simple path b'propname', or a tuple (b'parentprop', b'actualprop', arrayindex)
|
|
FieldPath = typing.Union[bytes, typing.Iterable[typing.Union[bytes, int]]]
|
|
|
|
|
|
class Name:
|
|
"""dna.Name is a C-type name stored in the DNA as bytes."""
|
|
|
|
def __init__(self, name_full: bytes):
|
|
self.name_full = name_full
|
|
self.name_only = self.calc_name_only()
|
|
self.is_pointer = self.calc_is_pointer()
|
|
self.is_method_pointer = self.calc_is_method_pointer()
|
|
self.array_size = self.calc_array_size()
|
|
|
|
def __repr__(self):
|
|
return '%s(%r)' % (type(self).__qualname__, self.name_full)
|
|
|
|
def as_reference(self, parent) -> bytes:
|
|
if not parent:
|
|
return self.name_only
|
|
return parent + b'.' + self.name_only
|
|
|
|
def calc_name_only(self) -> bytes:
|
|
result = self.name_full.strip(b'*()')
|
|
index = result.find(b'[')
|
|
if index == -1:
|
|
return result
|
|
return result[:index]
|
|
|
|
def calc_is_pointer(self) -> bool:
|
|
return b'*' in self.name_full
|
|
|
|
def calc_is_method_pointer(self):
|
|
return b'(*' in self.name_full
|
|
|
|
def calc_array_size(self):
|
|
result = 1
|
|
partial_name = self.name_full
|
|
|
|
while True:
|
|
idx_start = partial_name.find(b'[')
|
|
if idx_start < 0:
|
|
break
|
|
|
|
idx_stop = partial_name.find(b']')
|
|
result *= int(partial_name[idx_start + 1:idx_stop])
|
|
partial_name = partial_name[idx_stop + 1:]
|
|
|
|
return result
|
|
|
|
|
|
class Field:
|
|
"""dna.Field is a coupled dna.Struct and dna.Name.
|
|
|
|
It also contains the file offset in bytes.
|
|
|
|
:ivar name: the name of the field.
|
|
:ivar dna_type: the type of the field.
|
|
:ivar size: size of the field on disk, in bytes.
|
|
:ivar offset: cached offset of the field, in bytes.
|
|
"""
|
|
|
|
def __init__(self,
|
|
dna_type: 'Struct',
|
|
name: Name,
|
|
size: int,
|
|
offset: int):
|
|
self.dna_type = dna_type
|
|
self.name = name
|
|
self.size = size
|
|
self.offset = offset
|
|
|
|
def __repr__(self):
|
|
return '<%r %r (%s)>' % (type(self).__qualname__, self.name, self.dna_type)
|
|
|
|
|
|
class Struct:
|
|
"""dna.Struct is a C-type structure stored in the DNA."""
|
|
|
|
def __init__(self, dna_type_id: bytes, size: int = None):
|
|
"""
|
|
:param dna_type_id: name of the struct in C, like b'AlembicObjectPath'.
|
|
:param size: only for unit tests; typically set after construction by
|
|
BlendFile.decode_structs(). If not set, it is calculated on the fly
|
|
when struct.size is evaluated, based on the available fields.
|
|
"""
|
|
self.dna_type_id = dna_type_id
|
|
self._size = size
|
|
self._fields = []
|
|
self._fields_by_name = {}
|
|
|
|
def __repr__(self):
|
|
return '%s(%r)' % (type(self).__qualname__, self.dna_type_id)
|
|
|
|
@property
|
|
def size(self) -> int:
|
|
if self._size is None:
|
|
if not self._fields:
|
|
raise ValueError('Unable to determine size of fieldless %r' % self)
|
|
last_field = max(self._fields, key=lambda f: f.offset)
|
|
self._size = last_field.offset + last_field.size
|
|
return self._size
|
|
|
|
@size.setter
|
|
def size(self, new_size: int):
|
|
self._size = new_size
|
|
|
|
def append_field(self, field: Field):
|
|
self._fields.append(field)
|
|
self._fields_by_name[field.name.name_only] = field
|
|
|
|
@property
|
|
def fields(self) -> typing.List[Field]:
|
|
"""Return the fields of this Struct.
|
|
|
|
Do not modify the returned list; use append_field() instead.
|
|
"""
|
|
return self._fields
|
|
|
|
def field_from_path(self,
|
|
pointer_size: int,
|
|
path: FieldPath) \
|
|
-> typing.Tuple[Field, int]:
|
|
"""
|
|
Support lookups as bytes or a tuple of bytes and optional index.
|
|
|
|
C style 'id.name' --> (b'id', b'name')
|
|
C style 'array[4]' --> (b'array', 4)
|
|
|
|
:returns: the field itself, and its offset taking into account the
|
|
optional index. The offset is relative to the start of the struct,
|
|
i.e. relative to the BlendFileBlock containing the data.
|
|
:raises KeyError: if the field does not exist.
|
|
"""
|
|
if isinstance(path, tuple):
|
|
name = path[0]
|
|
if len(path) >= 2 and not isinstance(path[1], bytes):
|
|
name_tail = path[2:]
|
|
index = path[1]
|
|
assert isinstance(index, int)
|
|
else:
|
|
name_tail = path[1:]
|
|
index = 0
|
|
else:
|
|
name = path
|
|
name_tail = None
|
|
index = 0
|
|
|
|
if not isinstance(name, bytes):
|
|
raise TypeError('name should be bytes, but is %r' % type(name))
|
|
|
|
field = self._fields_by_name.get(name)
|
|
if not field:
|
|
raise KeyError('%r has no field %r, only %r' %
|
|
(self, name, sorted(self._fields_by_name.keys())))
|
|
|
|
offset = field.offset
|
|
if index:
|
|
if field.name.is_pointer:
|
|
index_offset = pointer_size * index
|
|
else:
|
|
index_offset = field.dna_type.size * index
|
|
if index_offset >= field.size:
|
|
raise OverflowError('path %r is out of bounds of its DNA type %s' %
|
|
(path, field.dna_type))
|
|
offset += index_offset
|
|
|
|
if name_tail:
|
|
subval, suboff = field.dna_type.field_from_path(pointer_size, name_tail)
|
|
return subval, suboff + offset
|
|
|
|
return field, offset
|
|
|
|
def field_get(self,
|
|
file_header: header.BlendFileHeader,
|
|
fileobj: typing.BinaryIO,
|
|
path: FieldPath,
|
|
default=...,
|
|
null_terminated=True,
|
|
as_str=True,
|
|
):
|
|
"""Read the value of the field from the blend file.
|
|
|
|
Assumes the file pointer of `fileobj` is seek()ed to the start of the
|
|
struct on disk (e.g. the start of the BlendFileBlock containing the
|
|
data).
|
|
|
|
:param file_header:
|
|
:param fileobj:
|
|
:param path:
|
|
:param default: The value to return when the field does not exist.
|
|
Use Ellipsis (the default value) to raise a KeyError instead.
|
|
:param null_terminated: Only used when reading bytes or strings. When
|
|
True, stops reading at the first zero byte. Be careful with this
|
|
default when reading binary data.
|
|
:param as_str: When True, automatically decode bytes to string
|
|
(assumes UTF-8 encoding).
|
|
"""
|
|
try:
|
|
field, offset = self.field_from_path(file_header.pointer_size, path)
|
|
except KeyError:
|
|
if default is ...:
|
|
raise
|
|
return default
|
|
|
|
fileobj.seek(offset, os.SEEK_CUR)
|
|
|
|
dna_type = field.dna_type
|
|
dna_name = field.name
|
|
types = file_header.endian
|
|
|
|
# Some special cases (pointers, strings/bytes)
|
|
if dna_name.is_pointer:
|
|
return types.read_pointer(fileobj, file_header.pointer_size)
|
|
if dna_type.dna_type_id == b'char':
|
|
if field.size == 1:
|
|
# Single char, assume it's bitflag or int value, and not a string/bytes data...
|
|
return types.read_char(fileobj)
|
|
if null_terminated or (null_terminated is None and as_str):
|
|
data = types.read_bytes0(fileobj, dna_name.array_size)
|
|
else:
|
|
data = fileobj.read(dna_name.array_size)
|
|
|
|
if as_str:
|
|
return data.decode('utf8')
|
|
return data
|
|
|
|
simple_readers = {
|
|
b'int': types.read_int,
|
|
b'short': types.read_short,
|
|
b'uint64_t': types.read_ulong,
|
|
b'float': types.read_float,
|
|
}
|
|
try:
|
|
simple_reader = simple_readers[dna_type.dna_type_id]
|
|
except KeyError:
|
|
raise exceptions.NoReaderImplemented(
|
|
"%r exists but not simple type (%r), can't resolve field %r" %
|
|
(path, dna_type.dna_type_id.decode(), dna_name.name_only),
|
|
dna_name, dna_type) from None
|
|
|
|
if isinstance(path, tuple) and len(path) > 1 and isinstance(path[-1], int):
|
|
# The caller wants to get a single item from an array. The offset we seeked to already
|
|
# points to this item. In this case we do not want to look at dna_name.array_size,
|
|
# because we want a single item from that array.
|
|
return simple_reader(fileobj)
|
|
|
|
if dna_name.array_size > 1:
|
|
return [simple_reader(fileobj) for _ in range(dna_name.array_size)]
|
|
return simple_reader(fileobj)
|
|
|
|
def field_set(self, header, handle, path, value):
|
|
assert (type(path) == bytes)
|
|
|
|
field = self.field_from_path(header, handle, path)
|
|
if field is None:
|
|
raise KeyError("%r not found in %r" %
|
|
(path, [f.dna_name.name_only for f in self.fields]))
|
|
|
|
dna_type = field.dna_type
|
|
dna_name = field.dna_name
|
|
|
|
if dna_type.dna_type_id == b'char':
|
|
if type(value) is str:
|
|
return DNA_IO.write_string(handle, value, dna_name.array_size)
|
|
else:
|
|
return DNA_IO.write_bytes(handle, value, dna_name.array_size)
|
|
else:
|
|
raise NotImplementedError("Setting %r is not yet supported for %r" %
|
|
(dna_type, dna_name), dna_name, dna_type)
|