2021-04-11 22:44:31 +02:00
|
|
|
import os
|
|
|
|
import struct
|
|
|
|
import sys
|
2021-04-12 05:09:57 +02:00
|
|
|
from typing import Any, Dict, List, Optional, Tuple
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
from .decompile import ByteCode
|
2021-04-11 22:44:31 +02:00
|
|
|
from .types import Matrix, Color, Point, Rectangle
|
2021-04-24 19:59:36 +02:00
|
|
|
from .types import (
|
|
|
|
AP2Action,
|
|
|
|
AP2Tag,
|
|
|
|
DefineFunction2Action,
|
|
|
|
InitRegisterAction,
|
|
|
|
StoreRegisterAction,
|
|
|
|
JumpAction,
|
|
|
|
WithAction,
|
|
|
|
PushAction,
|
|
|
|
AddNumVariableAction,
|
|
|
|
AddNumRegisterAction,
|
|
|
|
IfAction,
|
|
|
|
GetURL2Action,
|
|
|
|
StartDragAction,
|
|
|
|
GotoFrame2Action,
|
|
|
|
Register,
|
2021-04-24 20:05:58 +02:00
|
|
|
StringConstant,
|
2021-04-24 19:59:36 +02:00
|
|
|
NULL,
|
|
|
|
UNDEFINED,
|
|
|
|
THIS,
|
|
|
|
ROOT,
|
|
|
|
PARENT,
|
|
|
|
CLIP,
|
|
|
|
GLOBAL,
|
|
|
|
)
|
|
|
|
from .util import TrackedCoverage, VerboseOutput
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
class NamedTagReference:
|
|
|
|
def __init__(self, swf_name: str, tag_name: str) -> None:
|
|
|
|
self.swf = swf_name
|
|
|
|
self.tag = tag_name
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
2021-04-12 05:09:57 +02:00
|
|
|
return {
|
|
|
|
'swf': self.swf,
|
|
|
|
'tag': self.tag,
|
|
|
|
}
|
|
|
|
|
|
|
|
def __repr__(self) -> str:
|
|
|
|
return f"{self.swf}.{self.tag}"
|
|
|
|
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
class TagPointer:
|
|
|
|
# A pointer to a tag in this SWF by Tag ID and containing an optional initialization bytecode
|
|
|
|
# to run for this tag when it is placed/executed.
|
|
|
|
def __init__(self, id: Optional[int], init_bytecode: Optional[ByteCode] = None) -> None:
|
|
|
|
self.id = id
|
|
|
|
self.init_bytecode = init_bytecode
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
'id': self.id,
|
|
|
|
'init_bytecode': self.init_bytecode.as_dict(*args, **kwargs) if self.init_bytecode else None,
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class Frame:
|
|
|
|
def __init__(self, start_tag_offset: int, num_tags: int, imported_tags: List[TagPointer] = []) -> None:
|
|
|
|
# The start tag offset into the tag list where we should begin placing/executing tags for this frame.
|
|
|
|
self.start_tag_offset = start_tag_offset
|
|
|
|
|
|
|
|
# The number of tags to pace/execute during this frame.
|
|
|
|
self.num_tags = num_tags
|
|
|
|
|
|
|
|
# A list of any imported tags that are to be placed this frame.
|
2021-04-24 19:59:36 +02:00
|
|
|
self.imported_tags = imported_tags or []
|
2021-04-12 05:10:31 +02:00
|
|
|
|
2021-04-18 01:31:08 +02:00
|
|
|
# The current tag we're processing, if any.
|
|
|
|
self.current_tag = 0
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
'start_tag_offset': self.start_tag_offset,
|
|
|
|
'num_tags': self.num_tags,
|
|
|
|
'imported_tags': [i.as_dict(*args, **kwargs) for i in self.imported_tags],
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class Tag:
|
|
|
|
# Any tag that can appear in the SWF. All tags will subclass from this for their behavior.
|
|
|
|
def __init__(self, id: Optional[int]) -> None:
|
|
|
|
self.id = id
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
'id': self.id,
|
|
|
|
'type': self.__class__.__name__,
|
|
|
|
}
|
2021-04-16 01:18:33 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2ShapeTag(Tag):
|
|
|
|
def __init__(self, id: int, reference: str) -> None:
|
|
|
|
super().__init__(id)
|
|
|
|
|
|
|
|
# The reference is the name of a shape (geo structure) that defines this primitive or sprite.
|
|
|
|
self.reference = reference
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'reference': self.reference,
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2DefineFontTag(Tag):
|
|
|
|
def __init__(self, id: int, fontname: str, xml_prefix: str, heights: List[int]) -> None:
|
|
|
|
super().__init__(id)
|
|
|
|
|
|
|
|
# The font name is just the pretty name of the font.
|
|
|
|
self.fontname = fontname
|
|
|
|
|
|
|
|
# The XML prefix is the reference into any font XML to look up individual
|
|
|
|
# glyphs for a font in a texture map.
|
|
|
|
self.xml_prefix = xml_prefix
|
|
|
|
|
|
|
|
# The list of heights are concatenated with the above XML prefix and the
|
|
|
|
# unicode glyph you want to display, to find the corresponding location
|
|
|
|
# in the texture map.
|
|
|
|
self.heights = heights
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'fontname': self.fontname,
|
|
|
|
'xml_prefix': self.xml_prefix,
|
|
|
|
'heights': self.heights,
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2DoActionTag(Tag):
|
|
|
|
def __init__(self, bytecode: ByteCode) -> None:
|
|
|
|
# Do Action Tags are not identified by any tag ID.
|
|
|
|
super().__init__(None)
|
|
|
|
|
|
|
|
# The bytecode is the actual execution that we expect to perform once
|
|
|
|
# this tag is placed/executed.
|
|
|
|
self.bytecode = bytecode
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'bytecode': self.bytecode.as_dict(*args, **kwargs),
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2PlaceObjectTag(Tag):
|
|
|
|
def __init__(
|
|
|
|
self,
|
|
|
|
object_id: int,
|
|
|
|
depth: int,
|
|
|
|
src_tag_id: Optional[int],
|
|
|
|
name: Optional[str],
|
|
|
|
blend: Optional[int],
|
|
|
|
update: bool,
|
|
|
|
transform: Optional[Matrix],
|
2021-04-16 01:17:15 +02:00
|
|
|
rotation_offset: Optional[Point],
|
2021-04-16 01:16:45 +02:00
|
|
|
mult_color: Optional[Color],
|
|
|
|
add_color: Optional[Color],
|
2021-04-12 05:10:31 +02:00
|
|
|
triggers: Dict[int, List[ByteCode]],
|
|
|
|
) -> None:
|
|
|
|
# Place Object Tags are not identified by any tag ID.
|
|
|
|
super().__init__(None)
|
|
|
|
|
|
|
|
# The object ID that we should associate with this object, for removal
|
|
|
|
# and presumably update and other uses. Not the same as Tag ID.
|
|
|
|
self.object_id = object_id
|
|
|
|
|
|
|
|
# The depth (level) that we should remove objects from.
|
|
|
|
self.depth = depth
|
|
|
|
|
2021-04-18 01:31:08 +02:00
|
|
|
# The source tag ID (should point at an AP2ShapeTag or AP2SpriteTag by ID) if present.
|
2021-04-12 05:10:31 +02:00
|
|
|
self.source_tag_id = src_tag_id
|
|
|
|
|
|
|
|
# The name of this object, if present.
|
|
|
|
self.name = name
|
|
|
|
|
|
|
|
# The blend mode of this object, if present.
|
|
|
|
self.blend = blend
|
|
|
|
|
|
|
|
# Whether this is an object update (True) or a new object (False).
|
|
|
|
self.update = update
|
|
|
|
|
|
|
|
# Whether there is a transform matrix to apply before placing/updating this object or not.
|
|
|
|
self.transform = transform
|
2021-04-16 01:17:15 +02:00
|
|
|
self.rotation_offset = rotation_offset
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
# If there is a color to blend with the sprite/shape when drawing.
|
2021-04-16 01:16:45 +02:00
|
|
|
self.mult_color = mult_color
|
2021-04-12 05:10:31 +02:00
|
|
|
|
2021-04-16 01:16:45 +02:00
|
|
|
# If there is a color to add with the sprite/shape when drawing.
|
|
|
|
self.add_color = add_color
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
# List of triggers for this object, and their respective bytecodes to execute when the trigger
|
|
|
|
# fires.
|
|
|
|
self.triggers = triggers
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'object_id': self.object_id,
|
|
|
|
'depth': self.depth,
|
|
|
|
'source_tag_id': self.source_tag_id,
|
|
|
|
'name': self.name,
|
|
|
|
'blend': self.blend,
|
|
|
|
'update': self.update,
|
|
|
|
'transform': self.transform.as_dict(*args, **kwargs) if self.transform is not None else None,
|
|
|
|
'rotation_offset': self.rotation_offset.as_dict(*args, **kwargs) if self.rotation_offset is not None else None,
|
|
|
|
'mult_color': self.mult_color.as_dict(*args, **kwargs) if self.mult_color is not None else None,
|
|
|
|
'add_color': self.add_color.as_dict(*args, **kwargs) if self.add_color is not None else None,
|
|
|
|
'triggers': {i: [b.as_dict(*args, **kwargs) for b in t] for (i, t) in self.triggers.items()}
|
|
|
|
}
|
|
|
|
|
2021-04-20 23:41:28 +02:00
|
|
|
def __repr__(self) -> str:
|
|
|
|
return f"AP2PlaceObjectTag(object_id={self.object_id}, depth={self.depth})"
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2RemoveObjectTag(Tag):
|
|
|
|
def __init__(self, object_id: int, depth: int) -> None:
|
|
|
|
# Remove Object Tags are not identified by any tag ID.
|
|
|
|
super().__init__(None)
|
|
|
|
|
|
|
|
# The object ID that we should remove, or 0 if we should only remove by depth.
|
|
|
|
self.object_id = object_id
|
|
|
|
|
|
|
|
# The depth (level) that we should remove objects from.
|
|
|
|
self.depth = depth
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'object_id': self.object_id,
|
|
|
|
'depth': self.depth,
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2DefineSpriteTag(Tag):
|
|
|
|
def __init__(self, id: int, tags: List[Tag], frames: List[Frame]) -> None:
|
|
|
|
super().__init__(id)
|
|
|
|
|
|
|
|
# The list of tags that this sprite consists of. Sprites are, much like vanilla
|
|
|
|
# SWFs, basically entire SWF movies embedded in them.
|
|
|
|
self.tags = tags
|
|
|
|
|
|
|
|
# The list of frames this SWF occupies.
|
|
|
|
self.frames = frames
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'tags': [t.as_dict(*args, **kwargs) for t in self.tags],
|
|
|
|
'frames': [f.as_dict(*args, **kwargs) for f in self.frames],
|
|
|
|
}
|
2021-04-16 01:18:33 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
class AP2DefineEditTextTag(Tag):
|
|
|
|
def __init__(self, id: int, font_tag_id: int, font_height: int, rect: Rectangle, color: Color, default_text: Optional[str] = None) -> None:
|
|
|
|
super().__init__(id)
|
|
|
|
|
|
|
|
# The ID of the Ap2DefineFontTag that we want to use for the text.
|
|
|
|
self.font_tag_id = font_tag_id
|
|
|
|
|
|
|
|
# The height we want to select for the text (must be one of the heights in
|
|
|
|
# the referenced Ap2DefineFontTag tag).
|
|
|
|
self.font_height = font_height
|
|
|
|
|
|
|
|
# The bounding rectangle for this exit text control.
|
|
|
|
self.rect = rect
|
|
|
|
|
|
|
|
# The text color we want to use when displaying the text.
|
|
|
|
self.color = color
|
|
|
|
|
|
|
|
# The default text that should be present in the control when it is initially placed/executed.
|
|
|
|
self.default_text = default_text
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
|
|
|
return {
|
|
|
|
**super().as_dict(*args, **kwargs),
|
|
|
|
'font_tag_id': self.font_tag_id,
|
|
|
|
'font_height': self.font_height,
|
|
|
|
'rect': self.rect.as_dict(*args, **kwargs),
|
|
|
|
'color': self.color.as_dict(*args, **kwargs),
|
|
|
|
'default_text': self.default_text,
|
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
class SWF(TrackedCoverage, VerboseOutput):
|
2021-04-11 22:44:31 +02:00
|
|
|
def __init__(
|
|
|
|
self,
|
|
|
|
name: str,
|
|
|
|
data: bytes,
|
|
|
|
descramble_info: bytes = b"",
|
|
|
|
) -> None:
|
2021-04-11 22:44:55 +02:00
|
|
|
# First, init the coverage engine.
|
|
|
|
super().__init__()
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
# Name of this SWF, according to the container it was extracted from.
|
|
|
|
self.name: str = name
|
|
|
|
|
|
|
|
# Name of this SWF, as referenced by other SWFs that require imports from it.
|
|
|
|
self.exported_name: str = ""
|
|
|
|
|
|
|
|
# Full, unparsed data for this SWF, as well as the descrambling headers.
|
|
|
|
self.data: bytes = data
|
|
|
|
self.descramble_info: bytes = descramble_info
|
|
|
|
|
|
|
|
# Data version of this SWF.
|
|
|
|
self.data_version: int = 0
|
|
|
|
|
|
|
|
# Container version of this SWF.
|
|
|
|
self.container_version: int = 0
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
# The requested frames per second this SWF plays at.
|
|
|
|
self.fps: float = 0.0
|
|
|
|
|
|
|
|
# Background color of this SWF movie.
|
|
|
|
self.color: Optional[Color] = None
|
|
|
|
|
|
|
|
# Location of this SWF in screen space.
|
|
|
|
self.location: Rectangle = Rectangle.Empty()
|
|
|
|
|
|
|
|
# Exported tags, indexed by their name and pointing to the Tag ID that name identifies.
|
|
|
|
self.exported_tags: Dict[str, int] = {}
|
|
|
|
|
|
|
|
# Imported tags, indexed by their Tag ID, and pointing at the SWF asset and exported tag name.
|
|
|
|
self.imported_tags: Dict[int, NamedTagReference] = {}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
# Actual tags for this SWF, ordered by their appearance in the file.
|
|
|
|
self.tags: List[Tag] = []
|
|
|
|
|
|
|
|
# Frames of this SWF, with the tag offset in the above list and number of tags to
|
|
|
|
# "execute" that frame.
|
|
|
|
self.frames: List[Frame] = []
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
# SWF string table. This is used for faster lookup of strings as well as
|
|
|
|
# tracking which strings in the table have been parsed correctly.
|
|
|
|
self.__strings: Dict[int, Tuple[str, bool]] = {}
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-16 01:18:33 +02:00
|
|
|
# Whether this is parsed or not.
|
|
|
|
self.parsed = False
|
|
|
|
|
2021-04-11 22:44:31 +02:00
|
|
|
def print_coverage(self) -> None:
|
2021-04-11 22:44:55 +02:00
|
|
|
# First print uncovered bytes
|
|
|
|
super().print_coverage()
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Now, print uncovered strings
|
2021-04-12 05:09:57 +02:00
|
|
|
for offset, (string, covered) in self.__strings.items():
|
2021-04-11 22:44:31 +02:00
|
|
|
if covered:
|
|
|
|
continue
|
|
|
|
|
|
|
|
print(f"Uncovered string: {hex(offset)} - {string}", file=sys.stderr)
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
def as_dict(self, *args: Any, **kwargs: Any) -> Dict[str, Any]:
|
2021-04-11 22:44:31 +02:00
|
|
|
return {
|
|
|
|
'name': self.name,
|
2021-04-24 19:59:36 +02:00
|
|
|
'exported_name': self.exported_name,
|
|
|
|
'data_version': self.data_version,
|
|
|
|
'container_version': self.container_version,
|
|
|
|
'fps': self.fps,
|
|
|
|
'color': self.color.as_dict(*args, **kwargs) if self.color is not None else None,
|
|
|
|
'location': self.location.as_dict(*args, **kwargs),
|
|
|
|
'exported_tags': self.exported_tags,
|
|
|
|
'imported_tags': {i: self.imported_tags[i].as_dict(*args, **kwargs) for i in self.imported_tags},
|
|
|
|
'tags': [t.as_dict(*args, **kwargs) for t in self.tags],
|
|
|
|
'frames': [f.as_dict(*args, **kwargs) for f in self.frames],
|
2021-04-11 22:44:31 +02:00
|
|
|
}
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
def __parse_bytecode(self, datachunk: bytes, string_offsets: List[int] = [], prefix: str = "") -> ByteCode:
|
2021-04-11 22:44:31 +02:00
|
|
|
# First, we need to check if this is a SWF-style bytecode or an AP2 bytecode.
|
|
|
|
ap2_sentinel = struct.unpack("<B", datachunk[0:1])[0]
|
|
|
|
|
|
|
|
if ap2_sentinel != 0xFF:
|
|
|
|
raise Exception("Encountered SWF-style bytecode but we don't support this!")
|
|
|
|
|
|
|
|
# Now, we need to grab the flags byte which tells us how to find the actual bytecode.
|
|
|
|
flags = struct.unpack("<B", datachunk[1:2])[0]
|
|
|
|
|
|
|
|
if flags & 0x1:
|
|
|
|
# There is an offset pointer telling us where the data is as well as string offset tables.
|
|
|
|
string_offsets_count = struct.unpack("<H", datachunk[2:4])[0]
|
|
|
|
|
|
|
|
# We don't want to overwrite the global ones with our current ones.
|
|
|
|
if not string_offsets:
|
|
|
|
string_offsets = list(struct.unpack("<" + ("H" * string_offsets_count), datachunk[4:(4 + (2 * string_offsets_count))]))
|
|
|
|
|
|
|
|
offset_ptr = (string_offsets_count + 2) * 2
|
|
|
|
else:
|
|
|
|
# The data directly follows, no pointer.
|
|
|
|
offset_ptr = 2
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
self.vprint(f"{prefix} Flags: {hex(flags)}, ByteCode Actual Offset: {hex(offset_ptr)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Actually parse out the opcodes:
|
2021-04-21 05:58:53 +02:00
|
|
|
actions: List[AP2Action] = []
|
2021-04-11 22:44:31 +02:00
|
|
|
while offset_ptr < len(datachunk):
|
|
|
|
# We leave it up to the individual opcode handlers to increment the offset pointer. By default, parameterless
|
|
|
|
# opcodes increase by one. Everything else increases by its own amount. Opcode parsing here is done in big-endian
|
|
|
|
# as the game code seems to always parse big-endian values.
|
|
|
|
opcode = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
|
|
|
action_name = AP2Action.action_to_name(opcode)
|
2021-04-21 05:58:53 +02:00
|
|
|
lineno = offset_ptr
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if opcode in AP2Action.actions_without_params():
|
2021-04-21 05:58:53 +02:00
|
|
|
# Simple opcodes need no parsing, they can go directly onto the stack.
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(AP2Action(lineno, opcode))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.DEFINE_FUNCTION2:
|
|
|
|
function_flags, funcname_offset, bytecode_offset, _, bytecode_count = struct.unpack(
|
|
|
|
">HHHBH",
|
|
|
|
datachunk[(offset_ptr + 1):(offset_ptr + 10)],
|
|
|
|
)
|
|
|
|
|
|
|
|
if funcname_offset == 0:
|
2021-04-21 05:58:53 +02:00
|
|
|
funcname = None
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
|
|
|
funcname = self.__get_string(funcname_offset)
|
|
|
|
offset_ptr += 10 + (3 * bytecode_offset)
|
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Flags: {hex(function_flags)}, Name: {funcname or '<anonymous function>'}, ByteCode Offset: {hex(bytecode_offset)}, ByteCode Length: {hex(bytecode_count)}")
|
|
|
|
|
|
|
|
function = self.__parse_bytecode(datachunk[offset_ptr:(offset_ptr + bytecode_count)], string_offsets=string_offsets, prefix=prefix + " ")
|
2021-04-12 05:10:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} END_{action_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(DefineFunction2Action(lineno, funcname, function_flags, function))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += bytecode_count
|
|
|
|
elif opcode == AP2Action.PUSH:
|
|
|
|
obj_count = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
objects: List[Any] = []
|
|
|
|
|
2021-04-11 22:44:31 +02:00
|
|
|
while obj_count > 0:
|
|
|
|
obj_to_create = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
|
|
|
offset_ptr += 1
|
|
|
|
|
|
|
|
if obj_to_create == 0x0:
|
|
|
|
# Integer "0" object.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(0)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} INTEGER: 0")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x1:
|
|
|
|
# Float object, represented internally as a double.
|
|
|
|
fval = struct.unpack(">f", datachunk[offset_ptr:(offset_ptr + 4)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(fval)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 4
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} FLOAT: {fval}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x2:
|
|
|
|
# Null pointer object.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(NULL)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} NULL")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x3:
|
|
|
|
# Undefined constant.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(UNDEFINED)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} UNDEFINED")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x4:
|
|
|
|
# Register value.
|
|
|
|
regno = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(Register(regno))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} REGISTER NO: {regno}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x5:
|
|
|
|
# Boolean "TRUE" object.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(True)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} BOOLEAN: True")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x6:
|
|
|
|
# Boolean "FALSE" object.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(False)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} BOOLEAN: False")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x7:
|
|
|
|
# Integer object.
|
2021-04-16 01:19:56 +02:00
|
|
|
ival = struct.unpack(">i", datachunk[offset_ptr:(offset_ptr + 4)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(ival)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 4
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} INTEGER: {ival}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x8:
|
|
|
|
# String constant object.
|
|
|
|
const_offset = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
|
|
|
const = self.__get_string(string_offsets[const_offset])
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(const)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} STRING CONST: {const}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x9:
|
|
|
|
# String constant, but with 16 bits for the offset. Probably not used except
|
|
|
|
# on the largest files.
|
|
|
|
const_offset = struct.unpack(">H", datachunk[offset_ptr:(offset_ptr + 2)])[0]
|
|
|
|
const = self.__get_string(string_offsets[const_offset])
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(const)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
self.vprint(f"{prefix} STRING CONST: {const}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xa:
|
|
|
|
# NaN constant.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(float("nan"))
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} NAN")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xb:
|
|
|
|
# Infinity constant.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(float("inf"))
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} INFINITY")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xc:
|
|
|
|
# Pointer to "this" object, whatever currently is executing the bytecode.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(THIS)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} POINTER TO THIS")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xd:
|
|
|
|
# Pointer to "root" object, which is the movieclip this bytecode exists in.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(ROOT)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} POINTER TO ROOT")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xe:
|
|
|
|
# Pointer to "parent" object, whatever currently is executing the bytecode.
|
|
|
|
# This seems to be the parent of the movie clip, or the current movieclip
|
|
|
|
# if that isn't set.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(PARENT)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} POINTER TO PARENT")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0xf:
|
|
|
|
# Current movie clip.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(CLIP)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} POINTER TO CURRENT MOVIECLIP")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x10:
|
2021-04-14 03:07:33 +02:00
|
|
|
# Property constant with no alias.
|
2021-04-11 22:44:31 +02:00
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x100
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} PROPERTY CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x11:
|
|
|
|
# Property constant referencing a string table entry.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x100
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} PROPERTY CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x12:
|
|
|
|
# Same as above, but with allowance for a 16-bit constant offset.
|
|
|
|
propertyval, reference = struct.unpack(">BH", datachunk[offset_ptr:(offset_ptr + 3)])
|
|
|
|
propertyval += 0x100
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 3
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} PROPERTY CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x13:
|
|
|
|
# Class property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x300
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} CLASS CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x14:
|
|
|
|
# Class property constant with alias.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x300
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} CLASS CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# One would expect 0x15 to be identical to 0x12 but for class properties instead. However, it appears
|
|
|
|
# that this has been omitted from game binaries.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x16:
|
|
|
|
# Func property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x400
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} FUNC CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x17:
|
|
|
|
# Func property name referencing a string table entry.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x400
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} FUNC CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Same comment with 0x15 applies here with 0x18.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x19:
|
|
|
|
# Other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x200
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} OTHER CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x1a:
|
|
|
|
# Other property name referencing a string table entry.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x200
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} OTHER CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Same comment with 0x15 and 0x18 applies here with 0x1b.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x1c:
|
|
|
|
# Event property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x500
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} EVENT CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x1d:
|
|
|
|
# Event property name referencing a string table entry.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x500
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} EVENT CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Same comment with 0x15, 0x18 and 0x1b applies here with 0x1e.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x1f:
|
|
|
|
# Key constants.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x600
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} KEY CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x20:
|
|
|
|
# Key property name referencing a string table entry.
|
|
|
|
propertyval, reference = struct.unpack(">BB", datachunk[offset_ptr:(offset_ptr + 2)])
|
|
|
|
propertyval += 0x600
|
|
|
|
referenceval = self.__get_string(string_offsets[reference])
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval, referenceval))
|
2021-04-14 03:07:33 +02:00
|
|
|
|
|
|
|
offset_ptr += 2
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} KEY CONST NAME: {StringConstant.property_to_name(propertyval)}, ALIAS: {referenceval}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Same comment with 0x15, 0x18, 0x1b and 0x1e applies here with 0x21.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x22:
|
|
|
|
# Pointer to global object.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(GLOBAL)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} POINTER TO GLOBAL OBJECT")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x23:
|
|
|
|
# Negative infinity.
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(float("-inf"))
|
2021-04-14 03:07:33 +02:00
|
|
|
self.vprint(f"{prefix} -INFINITY")
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x24:
|
|
|
|
# Some other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x700
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} ETC2 CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x25 and 0x26 are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x24.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x27:
|
|
|
|
# Some other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x800
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} ORGFUNC2 CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x28 and 0x29 are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x27.
|
|
|
|
elif obj_to_create == 0x2a:
|
|
|
|
# Some other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0x900
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} ETCFUNC2 CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x2b and 0x2c are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x2a.
|
|
|
|
elif obj_to_create == 0x2d:
|
|
|
|
# Some other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0xa00
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} EVENT2 CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x2e and 0x2f are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x2d.
|
|
|
|
elif obj_to_create == 0x30:
|
|
|
|
# Some other property name.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0xb00
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} EVENT METHOD CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x31 and 0x32 are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x30.
|
|
|
|
elif obj_to_create == 0x33:
|
|
|
|
# Signed 64 bit integer init. Uses special "S64" type.
|
|
|
|
int64 = struct.unpack(">q", datachunk[offset_ptr:(offset_ptr + 8)])
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(int64)
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 8
|
|
|
|
|
|
|
|
self.vprint(f"{prefix} INTEGER: {int64}")
|
|
|
|
elif obj_to_create == 0x34:
|
|
|
|
# Some other property names.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0xc00
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} GENERIC CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x35 and 0x36 are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x34.
|
2021-04-11 22:44:31 +02:00
|
|
|
elif obj_to_create == 0x37:
|
|
|
|
# Integer object but one byte.
|
2021-04-16 01:19:56 +02:00
|
|
|
ival = struct.unpack(">b", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
objects.append(ival)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} INTEGER: {ival}")
|
2021-04-14 03:07:33 +02:00
|
|
|
elif obj_to_create == 0x38:
|
|
|
|
# Some other property names.
|
|
|
|
propertyval = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0] + 0xd00
|
2021-04-24 20:05:58 +02:00
|
|
|
objects.append(StringConstant(propertyval))
|
2021-04-14 03:07:33 +02:00
|
|
|
offset_ptr += 1
|
2021-04-24 20:05:58 +02:00
|
|
|
self.vprint(f"{prefix} GENERIC2 CONST NAME: {StringConstant.property_to_name(propertyval)}")
|
2021-04-14 03:07:33 +02:00
|
|
|
# Possibly in newer binaries, 0x39 and 0x3a are implemented as 8-bit and 16-bit alias pointer
|
|
|
|
# versions of 0x38.
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
|
|
|
raise Exception(f"Unsupported object {hex(obj_to_create)} to push!")
|
|
|
|
|
|
|
|
obj_count -= 1
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} END_{action_name}")
|
2021-04-21 05:58:53 +02:00
|
|
|
|
|
|
|
actions.append(PushAction(lineno, objects))
|
2021-04-14 03:07:01 +02:00
|
|
|
elif opcode == AP2Action.INIT_REGISTER:
|
|
|
|
obj_count = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name}")
|
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
init_registers: List[Register] = []
|
2021-04-14 03:07:01 +02:00
|
|
|
while obj_count > 0:
|
|
|
|
register_no = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
init_registers.append(Register(register_no))
|
2021-04-14 03:07:01 +02:00
|
|
|
offset_ptr += 1
|
|
|
|
obj_count -= 1
|
|
|
|
|
|
|
|
self.vprint(f"{prefix} REGISTER NO: {register_no}")
|
|
|
|
self.vprint(f"{prefix} END_{action_name}")
|
2021-04-21 05:58:53 +02:00
|
|
|
|
|
|
|
actions.append(InitRegisterAction(lineno, init_registers))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.STORE_REGISTER:
|
|
|
|
obj_count = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
store_registers: List[Register] = []
|
2021-04-11 22:44:31 +02:00
|
|
|
while obj_count > 0:
|
|
|
|
register_no = struct.unpack(">B", datachunk[offset_ptr:(offset_ptr + 1)])[0]
|
2021-04-21 05:58:53 +02:00
|
|
|
store_registers.append(Register(register_no))
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 1
|
|
|
|
obj_count -= 1
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} REGISTER NO: {register_no}")
|
|
|
|
self.vprint(f"{prefix} END_{action_name}")
|
2021-04-21 05:58:53 +02:00
|
|
|
|
2021-04-24 20:05:58 +02:00
|
|
|
actions.append(StoreRegisterAction(lineno, store_registers, preserve_stack=True))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.STORE_REGISTER2:
|
|
|
|
register_no = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name}")
|
|
|
|
self.vprint(f"{prefix} REGISTER NO: {register_no}")
|
|
|
|
self.vprint(f"{prefix} END_{action_name}")
|
2021-04-21 05:58:53 +02:00
|
|
|
|
2021-04-24 20:05:58 +02:00
|
|
|
actions.append(StoreRegisterAction(lineno, [Register(register_no)], preserve_stack=False))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.IF:
|
2021-04-21 05:58:53 +02:00
|
|
|
jump_if_true_offset = struct.unpack(">h", datachunk[(offset_ptr + 1):(offset_ptr + 3)])[0]
|
|
|
|
jump_if_true_offset += (lineno + 3)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 3
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: Offset If True: {jump_if_true_offset}")
|
2021-04-24 21:36:58 +02:00
|
|
|
actions.append(IfAction(lineno, IfAction.IS_TRUE, jump_if_true_offset))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.IF2:
|
2021-04-21 05:58:53 +02:00
|
|
|
if2_type, jump_if_true_offset = struct.unpack(">Bh", datachunk[(offset_ptr + 1):(offset_ptr + 4)])
|
|
|
|
jump_if_true_offset += (lineno + 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 4
|
|
|
|
|
2021-04-24 21:36:58 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} {IfAction.comparison_to_str(if2_type)}, Offset If True: {jump_if_true_offset}")
|
|
|
|
actions.append(IfAction(lineno, if2_type, jump_if_true_offset))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.JUMP:
|
2021-04-21 05:58:53 +02:00
|
|
|
jump_offset = struct.unpack(">h", datachunk[(offset_ptr + 1):(offset_ptr + 3)])[0]
|
|
|
|
jump_offset += (lineno + 3)
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 3
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Offset: {jump_offset}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(JumpAction(lineno, jump_offset))
|
2021-04-14 03:07:01 +02:00
|
|
|
elif opcode == AP2Action.WITH:
|
|
|
|
skip_offset = struct.unpack(">H", datachunk[(offset_ptr + 1):(offset_ptr + 3)])[0]
|
|
|
|
offset_ptr += 3
|
|
|
|
|
|
|
|
# TODO: I have absolutely no idea what the data which exists in the bytecode buffer at this point
|
|
|
|
# represents...
|
2021-04-21 05:58:53 +02:00
|
|
|
unknown_data = datachunk[offset_ptr:(offset_ptr + skip_offset)]
|
2021-04-14 03:07:01 +02:00
|
|
|
offset_ptr += skip_offset
|
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Unknown Data Length: {skip_offset}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(WithAction(lineno, unknown_data))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.ADD_NUM_VARIABLE:
|
2021-04-21 05:58:53 +02:00
|
|
|
amount_to_add = struct.unpack(">b", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Add Value: {amount_to_add}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(AddNumVariableAction(lineno, amount_to_add))
|
2021-04-14 03:07:01 +02:00
|
|
|
elif opcode == AP2Action.GET_URL2:
|
2021-04-21 05:58:53 +02:00
|
|
|
get_url_action = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
2021-04-14 03:07:01 +02:00
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-21 05:58:53 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} URL Action: {get_url_action >> 6}")
|
|
|
|
actions.append(GetURL2Action(lineno, get_url_action >> 6))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.START_DRAG:
|
|
|
|
constraint = struct.unpack(">b", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Constrain Mouse: {'yes' if constraint > 0 else ('no' if constraint == 0 else 'check stack')}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(StartDragAction(lineno, constrain=True if constraint > 0 else (False if constraint == 0 else None)))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.ADD_NUM_REGISTER:
|
2021-04-21 05:58:53 +02:00
|
|
|
register_no, amount_to_add = struct.unpack(">Bb", datachunk[(offset_ptr + 1):(offset_ptr + 3)])
|
2021-04-11 22:44:31 +02:00
|
|
|
offset_ptr += 3
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} Register No: {register_no}, Add Value: {amount_to_add}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(AddNumRegisterAction(lineno, Register(register_no), amount_to_add))
|
2021-04-11 22:44:31 +02:00
|
|
|
elif opcode == AP2Action.GOTO_FRAME2:
|
|
|
|
flags = struct.unpack(">B", datachunk[(offset_ptr + 1):(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
|
|
|
|
if flags & 0x1:
|
|
|
|
post = "STOP"
|
|
|
|
else:
|
|
|
|
post = "PLAY"
|
|
|
|
|
|
|
|
if flags & 0x2:
|
|
|
|
# Additional frames to add on top of stack value.
|
|
|
|
additional_frames = struct.unpack(">H", datachunk[offset_ptr:(offset_ptr + 2)])[0]
|
|
|
|
offset_ptr += 2
|
|
|
|
else:
|
|
|
|
additional_frames = 0
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} {lineno}: {action_name} AND {post} Additional Frames: {additional_frames}")
|
2021-04-21 05:58:53 +02:00
|
|
|
actions.append(GotoFrame2Action(lineno, additional_frames, stop=bool(flags & 0x1)))
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
|
|
|
raise Exception(f"Can't advance, no handler for opcode {opcode} ({hex(opcode)})!")
|
|
|
|
|
2021-04-24 19:59:36 +02:00
|
|
|
return ByteCode(actions, offset_ptr)
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
def __parse_tag(self, ap2_version: int, afp_version: int, ap2data: bytes, tagid: int, size: int, dataoffset: int, prefix: str = "") -> Tag:
|
2021-04-11 22:44:31 +02:00
|
|
|
if tagid == AP2Tag.AP2_SHAPE:
|
|
|
|
if size != 4:
|
|
|
|
raise Exception(f"Invalid shape size {size}")
|
|
|
|
|
2021-04-14 03:07:33 +02:00
|
|
|
unknown, shape_id = struct.unpack("<HH", ap2data[dataoffset:(dataoffset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, size)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-14 03:07:33 +02:00
|
|
|
# I'm not sure what the unknown value is. It doesn't seem to be parsed by either BishiBashi or Jubeat
|
|
|
|
# when I've looked, but it does appear to be non-zero sometimes in Pop'n Music animations.
|
2021-04-11 22:44:31 +02:00
|
|
|
shape_reference = f"{self.exported_name}_shape{shape_id}"
|
2021-04-14 03:08:13 +02:00
|
|
|
self.vprint(f"{prefix} Tag ID: {shape_id}, AFP Reference: {shape_reference}, Unknown: {unknown}")
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
return AP2ShapeTag(shape_id, shape_reference)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_DEFINE_SPRITE:
|
|
|
|
sprite_flags, sprite_id = struct.unpack("<HH", ap2data[dataoffset:(dataoffset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if sprite_flags & 1 == 0:
|
|
|
|
# This is an old-style tag, it has data directly following the header.
|
|
|
|
subtags_offset = dataoffset + 4
|
|
|
|
else:
|
|
|
|
# This is a new-style tag, it has a relative data pointer.
|
|
|
|
subtags_offset = struct.unpack("<I", ap2data[(dataoffset + 4):(dataoffset + 8)])[0] + dataoffset
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + 4, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Tag ID: {sprite_id}")
|
2021-04-12 05:10:31 +02:00
|
|
|
tags, frames = self.__parse_tags(ap2_version, afp_version, ap2data, subtags_offset, prefix=" " + prefix)
|
|
|
|
|
|
|
|
return AP2DefineSpriteTag(sprite_id, tags, frames)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_DEFINE_FONT:
|
|
|
|
unk, font_id, fontname_offset, xml_prefix_offset, data_offset, data_count = struct.unpack("<HHHHHH", ap2data[dataoffset:(dataoffset + 12)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, 12)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
fontname = self.__get_string(fontname_offset)
|
|
|
|
xml_prefix = self.__get_string(xml_prefix_offset)
|
|
|
|
|
2021-04-16 23:28:53 +02:00
|
|
|
self.vprint(f"{prefix} Tag ID: {font_id}, Unknown: {unk}, Font Name: {fontname}, XML Prefix: {xml_prefix}, Entries: {data_count}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
heights: List[int] = []
|
2021-04-11 22:44:31 +02:00
|
|
|
for i in range(data_count):
|
|
|
|
entry_offset = dataoffset + 12 + (data_offset * 2) + (i * 2)
|
|
|
|
entry_value = struct.unpack("<H", ap2data[entry_offset:(entry_offset + 2)])[0]
|
2021-04-12 05:10:31 +02:00
|
|
|
heights.append(entry_value)
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(entry_offset, 2)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Height: {entry_value}")
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
return AP2DefineFontTag(font_id, fontname, xml_prefix, heights)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_DO_ACTION:
|
|
|
|
datachunk = ap2data[dataoffset:(dataoffset + size)]
|
2021-04-12 05:10:31 +02:00
|
|
|
bytecode = self.__parse_bytecode(datachunk, prefix=prefix)
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, size)
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
return AP2DoActionTag(bytecode)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_PLACE_OBJECT:
|
|
|
|
# Allow us to keep track of what we've consumed.
|
|
|
|
datachunk = ap2data[dataoffset:(dataoffset + size)]
|
|
|
|
flags, depth, object_id = struct.unpack("<IHH", datachunk[0:8])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Flags: {hex(flags)}, Object ID: {object_id}, Depth: {depth}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
running_pointer = 8
|
|
|
|
unhandled_flags = flags
|
|
|
|
|
|
|
|
if flags & 0x2:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has a shape component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x2
|
|
|
|
src_tag_id = struct.unpack("<H", datachunk[running_pointer:(running_pointer + 2)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 2)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 2
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Source Tag ID: {src_tag_id}")
|
2021-04-12 05:10:31 +02:00
|
|
|
else:
|
|
|
|
src_tag_id = None
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x10:
|
|
|
|
unhandled_flags &= ~0x10
|
|
|
|
unk2 = struct.unpack("<H", datachunk[running_pointer:(running_pointer + 2)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 2)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 2
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unk2: {hex(unk2)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x20:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has name component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x20
|
|
|
|
nameoffset = struct.unpack("<H", datachunk[running_pointer:(running_pointer + 2)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 2)
|
2021-04-11 22:44:31 +02:00
|
|
|
name = self.__get_string(nameoffset)
|
|
|
|
running_pointer += 2
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Name: {name}")
|
2021-04-12 05:10:31 +02:00
|
|
|
else:
|
|
|
|
name = None
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x40:
|
|
|
|
unhandled_flags &= ~0x40
|
|
|
|
unk3 = struct.unpack("<H", datachunk[running_pointer:(running_pointer + 2)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 2)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 2
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unk3: {hex(unk3)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x20000:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has blend component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x20000
|
|
|
|
blend = struct.unpack("<B", datachunk[running_pointer:(running_pointer + 1)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 1)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 1
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Blend: {hex(blend)}")
|
2021-04-12 05:10:31 +02:00
|
|
|
else:
|
|
|
|
blend = None
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Due to possible misalignment, we need to realign.
|
|
|
|
misalignment = running_pointer & 3
|
|
|
|
if misalignment > 0:
|
|
|
|
catchup = 4 - misalignment
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, catchup)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += catchup
|
|
|
|
|
|
|
|
# Handle transformation matrix.
|
|
|
|
transform = Matrix.identity()
|
|
|
|
|
|
|
|
if flags & 0x100:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has scale component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x100
|
2021-04-16 01:17:45 +02:00
|
|
|
a_int, d_int = struct.unpack("<ii", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
|
|
|
transform.a = float(a_int) * 0.0009765625
|
|
|
|
transform.d = float(d_int) * 0.0009765625
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Transform Matrix A: {transform.a}, D: {transform.d}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x200:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has rotate component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x200
|
2021-04-16 01:17:45 +02:00
|
|
|
b_int, c_int = struct.unpack("<ii", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
|
|
|
transform.b = float(b_int) * 0.0009765625
|
|
|
|
transform.c = float(c_int) * 0.0009765625
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Transform Matrix B: {transform.b}, C: {transform.c}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x400:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Has translate component.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x400
|
2021-04-16 01:17:45 +02:00
|
|
|
tx_int, ty_int = struct.unpack("<ii", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
|
|
|
transform.tx = float(tx_int) / 20.0
|
2021-04-16 01:17:45 +02:00
|
|
|
transform.ty = float(ty_int) / 20.0
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Transform Matrix TX: {transform.tx}, TY: {transform.ty}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Handle object colors
|
2021-04-16 01:16:45 +02:00
|
|
|
multcolor = Color(1.0, 1.0, 1.0, 1.0)
|
2021-04-16 01:20:27 +02:00
|
|
|
addcolor = Color(0.0, 0.0, 0.0, 0.0)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x800:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Multiplicative color present.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x800
|
|
|
|
r, g, b, a = struct.unpack("<HHHH", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
2021-04-16 01:16:45 +02:00
|
|
|
multcolor.r = float(r) * 0.003921569
|
|
|
|
multcolor.g = float(g) * 0.003921569
|
|
|
|
multcolor.b = float(b) * 0.003921569
|
|
|
|
multcolor.a = float(a) * 0.003921569
|
|
|
|
self.vprint(f"{prefix} Mult Color: {multcolor}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x1000:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Additive color present.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x1000
|
|
|
|
r, g, b, a = struct.unpack("<HHHH", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
2021-04-16 01:16:45 +02:00
|
|
|
addcolor.r = float(r) * 0.003921569
|
|
|
|
addcolor.g = float(g) * 0.003921569
|
|
|
|
addcolor.b = float(b) * 0.003921569
|
|
|
|
addcolor.a = float(a) * 0.003921569
|
|
|
|
self.vprint(f"{prefix} Add Color: {addcolor}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x2000:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Multiplicative color present, smaller integers.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x2000
|
|
|
|
rgba = struct.unpack("<I", datachunk[running_pointer:(running_pointer + 4)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 4
|
|
|
|
|
2021-04-16 01:16:45 +02:00
|
|
|
multcolor.r = float((rgba >> 24) & 0xFF) * 0.003921569
|
|
|
|
multcolor.g = float((rgba >> 16) & 0xFF) * 0.003921569
|
|
|
|
multcolor.b = float((rgba >> 8) & 0xFF) * 0.003921569
|
|
|
|
multcolor.a = float(rgba & 0xFF) * 0.003921569
|
|
|
|
self.vprint(f"{prefix} Mult Color: {multcolor}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x4000:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Additive color present, smaller integers.
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x4000
|
|
|
|
rgba = struct.unpack("<I", datachunk[running_pointer:(running_pointer + 4)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 4
|
|
|
|
|
2021-04-16 01:16:45 +02:00
|
|
|
addcolor.r = float((rgba >> 24) & 0xFF) * 0.003921569
|
|
|
|
addcolor.g = float((rgba >> 16) & 0xFF) * 0.003921569
|
|
|
|
addcolor.b = float((rgba >> 8) & 0xFF) * 0.003921569
|
|
|
|
addcolor.a = float(rgba & 0xFF) * 0.003921569
|
|
|
|
self.vprint(f"{prefix} Add Color: {addcolor}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
bytecodes: Dict[int, List[ByteCode]] = {}
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x80:
|
|
|
|
# Object event triggers.
|
|
|
|
unhandled_flags &= ~0x80
|
|
|
|
event_flags, event_size = struct.unpack("<II", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if event_flags != 0:
|
|
|
|
_, count = struct.unpack("<HH", datachunk[(running_pointer + 8):(running_pointer + 12)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer + 8, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# The game does not seem to care about length here, but we do, so let's calculate
|
|
|
|
# offsets and use that for lengths.
|
|
|
|
bytecode_offsets: List[int] = []
|
|
|
|
for evt in range(count):
|
|
|
|
evt_offset = running_pointer + 12 + (evt * 8)
|
|
|
|
bytecode_offset = struct.unpack("<H", datachunk[(evt_offset + 6):(evt_offset + 8)])[0] + evt_offset
|
|
|
|
bytecode_offsets.append(bytecode_offset)
|
|
|
|
bytecode_offsets.append(event_size + running_pointer)
|
|
|
|
|
|
|
|
beginning_to_end: Dict[int, int] = {}
|
|
|
|
for i, bytecode_offset in enumerate(bytecode_offsets[:-1]):
|
|
|
|
beginning_to_end[bytecode_offset] = bytecode_offsets[i + 1]
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Event Triggers, Count: {count}")
|
2021-04-11 22:44:31 +02:00
|
|
|
for evt in range(count):
|
|
|
|
evt_offset = running_pointer + 12 + (evt * 8)
|
|
|
|
evt_flags, _, keycode, bytecode_offset = struct.unpack("<IBBH", datachunk[evt_offset:(evt_offset + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + evt_offset, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
events: List[str] = []
|
|
|
|
if evt_flags & 0x1:
|
|
|
|
events.append("ON_LOAD")
|
|
|
|
if evt_flags & 0x2:
|
|
|
|
events.append("ON_ENTER_FRAME")
|
|
|
|
if evt_flags & 0x4:
|
|
|
|
events.append("ON_UNLOAD")
|
|
|
|
if evt_flags & 0x8:
|
|
|
|
events.append("ON_MOUSE_MOVE")
|
|
|
|
if evt_flags & 0x10:
|
|
|
|
events.append("ON_MOUSE_DOWN")
|
|
|
|
if evt_flags & 0x20:
|
|
|
|
events.append("ON_MOUSE_UP")
|
|
|
|
if evt_flags & 0x40:
|
|
|
|
events.append("ON_KEY_DOWN")
|
|
|
|
if evt_flags & 0x80:
|
|
|
|
events.append("ON_KEY_UP")
|
|
|
|
if evt_flags & 0x100:
|
|
|
|
events.append("ON_DATA")
|
|
|
|
if evt_flags & 0x400:
|
|
|
|
events.append("ON_PRESS")
|
|
|
|
if evt_flags & 0x800:
|
|
|
|
events.append("ON_RELEASE")
|
|
|
|
if evt_flags & 0x1000:
|
|
|
|
events.append("ON_RELEASE_OUTSIDE")
|
|
|
|
if evt_flags & 0x2000:
|
|
|
|
events.append("ON_ROLL_OVER")
|
|
|
|
if evt_flags & 0x4000:
|
|
|
|
events.append("ON_ROLL_OUT")
|
|
|
|
|
|
|
|
bytecode_offset += evt_offset
|
|
|
|
bytecode_length = beginning_to_end[bytecode_offset] - bytecode_offset
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
self.vprint(f"{prefix} Flags: {hex(evt_flags)} ({', '.join(events)}), KeyCode: {hex(keycode)}, ByteCode Offset: {hex(dataoffset + bytecode_offset)}, Length: {bytecode_length}")
|
|
|
|
bytecode = self.__parse_bytecode(datachunk[bytecode_offset:(bytecode_offset + bytecode_length)], prefix=prefix + " ")
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + bytecode_offset, bytecode_length)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
bytecodes[evt_flags] = [*bytecodes.get(evt_flags, []), bytecode]
|
|
|
|
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += event_size
|
|
|
|
|
|
|
|
if flags & 0x10000:
|
|
|
|
# Some sort of filter data? Not sure what this is either. Needs more investigation
|
|
|
|
# if I encounter files with it.
|
|
|
|
unhandled_flags &= ~0x10000
|
|
|
|
count, filter_size = struct.unpack("<HH", datachunk[running_pointer:(running_pointer + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += filter_size
|
|
|
|
|
|
|
|
# TODO: This is not understood at all. I need to find data that uses it to continue.
|
|
|
|
# running_pointer + 4 starts a series of shorts (exactly count of them) which are
|
|
|
|
# all in the range of 0-7, corresponding to some sort of filter. They get sizes
|
|
|
|
# looked up and I presume there's data following this corresponding to those sizes.
|
|
|
|
# I don't know however as I've not encountered data with this bit.
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unknown Filter data Count: {count}, Size: {filter_size}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-16 01:17:15 +02:00
|
|
|
rotation_offset = None
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x1000000:
|
2021-04-16 01:16:45 +02:00
|
|
|
# Some sort of point, perhaps an x, y offset for the object or a center point for rotation?
|
2021-04-11 22:44:31 +02:00
|
|
|
unhandled_flags &= ~0x1000000
|
2021-04-16 01:16:45 +02:00
|
|
|
x, y = struct.unpack("<II", datachunk[running_pointer:(running_pointer + 8)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 8
|
|
|
|
|
2021-04-16 01:17:15 +02:00
|
|
|
rotation_offset = Point(float(x) / 20.0, float(y) / 20.0)
|
2021-04-16 01:17:45 +02:00
|
|
|
self.vprint(f"{prefix} Rotation Origin: {rotation_offset}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x2000000:
|
|
|
|
# Same as above, but initializing to 0, 0 instead of from data.
|
|
|
|
unhandled_flags &= ~0x2000000
|
2021-04-16 01:17:15 +02:00
|
|
|
rotation_offset = Point(0.0, 0.0)
|
2021-04-16 01:17:45 +02:00
|
|
|
self.vprint(f"{prefix} Rotation Origin: {rotation_offset}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x40000:
|
|
|
|
# Some pair of shorts, not sure, its in DDR PS3 data.
|
|
|
|
unhandled_flags &= ~0x40000
|
|
|
|
x, y = struct.unpack("<HH", datachunk[running_pointer:(running_pointer + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 4
|
|
|
|
|
|
|
|
# TODO: I have no idea what these are.
|
2021-04-16 01:16:45 +02:00
|
|
|
point = Point(float(x) * 3.051758e-05, float(y) * 3.051758e-05)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Point: {point}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x80000:
|
|
|
|
# Some pair of shorts, not sure, its in DDR PS3 data.
|
|
|
|
unhandled_flags &= ~0x80000
|
|
|
|
x, y = struct.unpack("<HH", datachunk[running_pointer:(running_pointer + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + running_pointer, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
running_pointer += 4
|
|
|
|
|
|
|
|
# TODO: I have no idea what these are.
|
2021-04-16 01:16:45 +02:00
|
|
|
point = Point(float(x) * 3.051758e-05, float(y) * 3.051758e-05)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Point: {point}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# This flag states whether we are creating a new object on this depth, or updating one.
|
|
|
|
unhandled_flags &= ~0xD
|
|
|
|
if flags & 0x1:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Update object request")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Create object request")
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x4:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Use transform matrix")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Ignore transform matrix")
|
2021-04-12 05:10:31 +02:00
|
|
|
if flags & 0x8:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Use color information")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Ignore color information")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if unhandled_flags != 0:
|
|
|
|
raise Exception(f"Did not handle {hex(unhandled_flags)} flag bits!")
|
|
|
|
if running_pointer < size:
|
|
|
|
raise Exception(f"Did not consume {size - running_pointer} bytes ({[hex(x) for x in datachunk[running_pointer:]]}) in object instantiation!")
|
|
|
|
if running_pointer != size:
|
|
|
|
raise Exception("Logic error!")
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
return AP2PlaceObjectTag(
|
|
|
|
object_id,
|
|
|
|
depth,
|
|
|
|
src_tag_id=src_tag_id,
|
|
|
|
name=name,
|
|
|
|
blend=blend,
|
|
|
|
update=True if (flags & 0x1) else False,
|
|
|
|
transform=transform if (flags & 0x4) else None,
|
2021-04-16 01:17:15 +02:00
|
|
|
rotation_offset=rotation_offset,
|
2021-04-16 01:16:45 +02:00
|
|
|
mult_color=multcolor if (flags & 0x8) else None,
|
|
|
|
add_color=addcolor if (flags & 0x8) else None,
|
2021-04-12 05:10:31 +02:00
|
|
|
triggers=bytecodes,
|
|
|
|
)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_REMOVE_OBJECT:
|
|
|
|
if size != 4:
|
|
|
|
raise Exception(f"Invalid shape size {size}")
|
|
|
|
|
|
|
|
object_id, depth = struct.unpack("<HH", ap2data[dataoffset:(dataoffset + 4)])
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Object ID: {object_id}, Depth: {depth}")
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, 4)
|
2021-04-12 05:10:31 +02:00
|
|
|
|
|
|
|
return AP2RemoveObjectTag(object_id, depth)
|
2021-04-11 22:44:31 +02:00
|
|
|
elif tagid == AP2Tag.AP2_DEFINE_EDIT_TEXT:
|
|
|
|
if size != 44:
|
|
|
|
raise Exception("Invalid size {size} to get data from AP2_DEFINE_EDIT_TEXT!")
|
|
|
|
|
|
|
|
flags, edit_text_id, defined_font_tag_id, font_height, unk_str2_offset = struct.unpack("<IHHHH", ap2data[dataoffset:(dataoffset + 12)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset, 12)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
unk1, unk2, unk3, unk4 = struct.unpack("<HHHH", ap2data[(dataoffset + 12):(dataoffset + 20)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + 12, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
rgba, f1, f2, f3, f4, variable_name_offset, default_text_offset = struct.unpack("<IiiiiHH", ap2data[(dataoffset + 20):(dataoffset + 44)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(dataoffset + 20, 24)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Tag ID: {edit_text_id}, Font Tag: {defined_font_tag_id}, Height Selection: {font_height}, Flags: {hex(flags)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
unk_string2 = self.__get_string(unk_str2_offset) or None
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unk String: {unk_string2}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
rect = Rectangle(f1 / 20.0, f2 / 20.0, f3 / 20.0, f4 / 20.0)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Rectangle: {rect}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
variable_name = self.__get_string(variable_name_offset) or None
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Variable Name: {variable_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
color = Color(
|
|
|
|
r=(rgba & 0xFF) / 255.0,
|
|
|
|
g=((rgba >> 8) & 0xFF) / 255.0,
|
|
|
|
b=((rgba >> 16) & 0xFF) / 255.0,
|
|
|
|
a=((rgba >> 24) & 0xFF) / 255.0,
|
|
|
|
)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Text Color: {color}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unk1: {unk1}, Unk2: {unk2}, Unk3: {unk3}, Unk4: {unk4}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# flags & 0x20 means something with offset 16-18.
|
|
|
|
# flags & 0x200 is unk str below is a HTML tag.
|
|
|
|
|
|
|
|
if flags & 0x80:
|
|
|
|
# Has some sort of string pointer.
|
|
|
|
default_text = self.__get_string(default_text_offset) or None
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Default Text: {default_text}")
|
2021-04-12 05:10:31 +02:00
|
|
|
else:
|
|
|
|
default_text = None
|
|
|
|
|
|
|
|
return AP2DefineEditTextTag(edit_text_id, defined_font_tag_id, font_height, rect, color, default_text=default_text)
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
|
|
|
raise Exception(f"Unimplemented tag {hex(tagid)}!")
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
def __parse_tags(self, ap2_version: int, afp_version: int, ap2data: bytes, tags_base_offset: int, prefix: str = "") -> Tuple[List[Tag], List[Frame]]:
|
2021-04-11 22:44:31 +02:00
|
|
|
unknown_tags_flags, unknown_tags_count, frame_count, tags_count, unknown_tags_offset, frame_offset, tags_offset = struct.unpack(
|
|
|
|
"<HHIIIII",
|
|
|
|
ap2data[tags_base_offset:(tags_base_offset + 24)]
|
|
|
|
)
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(tags_base_offset, 24)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Fix up pointers.
|
|
|
|
tags_offset += tags_base_offset
|
|
|
|
unknown_tags_offset += tags_base_offset
|
|
|
|
frame_offset += tags_base_offset
|
|
|
|
|
|
|
|
# First, parse regular tags.
|
2021-04-12 05:10:31 +02:00
|
|
|
tags: List[Tag] = []
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix}Number of Tags: {tags_count}")
|
2021-04-11 22:44:31 +02:00
|
|
|
for i in range(tags_count):
|
|
|
|
tag = struct.unpack("<I", ap2data[tags_offset:(tags_offset + 4)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(tags_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
tagid = (tag >> 22) & 0x3FF
|
|
|
|
size = tag & 0x3FFFFF
|
|
|
|
|
|
|
|
if size > 0x200000:
|
|
|
|
raise Exception(f"Invalid tag size {size} ({hex(size)})")
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Tag: {hex(tagid)} ({AP2Tag.tag_to_name(tagid)}), Size: {hex(size)}, Offset: {hex(tags_offset + 4)}")
|
2021-04-16 01:18:33 +02:00
|
|
|
tags.append(self.__parse_tag(ap2_version, afp_version, ap2data, tagid, size, tags_offset + 4, prefix=prefix))
|
2021-04-11 22:44:31 +02:00
|
|
|
tags_offset += ((size + 3) & 0xFFFFFFFC) + 4 # Skip past tag header and data, rounding to the nearest 4 bytes.
|
|
|
|
|
|
|
|
# Now, parse frames.
|
2021-04-12 05:10:31 +02:00
|
|
|
frames: List[Frame] = []
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix}Number of Frames: {frame_count}")
|
2021-04-11 22:44:31 +02:00
|
|
|
for i in range(frame_count):
|
|
|
|
frame_info = struct.unpack("<I", ap2data[frame_offset:(frame_offset + 4)])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(frame_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
start_tag_offset = frame_info & 0xFFFFF
|
2021-04-11 22:44:31 +02:00
|
|
|
num_tags_to_play = (frame_info >> 20) & 0xFFF
|
2021-04-12 05:10:31 +02:00
|
|
|
frames.append(Frame(start_tag_offset, num_tags_to_play))
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
self.vprint(f"{prefix} Frame Start Tag: {start_tag_offset}, Count: {num_tags_to_play}")
|
2021-04-11 22:44:31 +02:00
|
|
|
frame_offset += 4
|
|
|
|
|
|
|
|
# Now, parse unknown tags? I have no idea what these are, but they're referencing strings that
|
|
|
|
# are otherwise unused.
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix}Number of Unknown Tags: {unknown_tags_count}, Flags: {hex(unknown_tags_flags)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
for i in range(unknown_tags_count):
|
|
|
|
unk1, stringoffset = struct.unpack("<HH", ap2data[unknown_tags_offset:(unknown_tags_offset + 4)])
|
|
|
|
strval = self.__get_string(stringoffset)
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(unknown_tags_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{prefix} Unknown Tag: {hex(unk1)} Name: {strval}")
|
2021-04-11 22:44:31 +02:00
|
|
|
unknown_tags_offset += 4
|
|
|
|
|
2021-04-12 05:10:31 +02:00
|
|
|
return tags, frames
|
|
|
|
|
2021-04-11 22:44:31 +02:00
|
|
|
def __descramble(self, scrambled_data: bytes, descramble_info: bytes) -> bytes:
|
|
|
|
swap_len = {
|
|
|
|
1: 2,
|
|
|
|
2: 4,
|
|
|
|
3: 8,
|
|
|
|
}
|
|
|
|
|
|
|
|
data = bytearray(scrambled_data)
|
|
|
|
data_offset = 0
|
|
|
|
for i in range(0, len(descramble_info), 2):
|
|
|
|
swapword = struct.unpack("<H", descramble_info[i:(i + 2)])[0]
|
|
|
|
if swapword == 0:
|
|
|
|
break
|
|
|
|
|
|
|
|
offset = (swapword & 0x7F) * 2
|
|
|
|
swap_type = (swapword >> 13) & 0x7
|
|
|
|
loops = ((swapword >> 7) & 0x3F)
|
|
|
|
data_offset += offset
|
|
|
|
|
|
|
|
if swap_type == 0:
|
|
|
|
# Just jump forward based on loops
|
|
|
|
data_offset += 256 * loops
|
|
|
|
continue
|
|
|
|
|
|
|
|
if swap_type not in swap_len:
|
|
|
|
raise Exception(f"Unknown swap type {swap_type}!")
|
|
|
|
|
|
|
|
# Reverse the bytes
|
|
|
|
for _ in range(loops + 1):
|
|
|
|
data[data_offset:(data_offset + swap_len[swap_type])] = data[data_offset:(data_offset + swap_len[swap_type])][::-1]
|
|
|
|
data_offset += swap_len[swap_type]
|
|
|
|
|
|
|
|
return bytes(data)
|
|
|
|
|
|
|
|
def __descramble_stringtable(self, scrambled_data: bytes, stringtable_offset: int, stringtable_size: int) -> bytes:
|
|
|
|
data = bytearray(scrambled_data)
|
|
|
|
curstring: List[int] = []
|
|
|
|
curloc = stringtable_offset
|
|
|
|
|
|
|
|
addition = 128
|
|
|
|
for i in range(stringtable_size):
|
|
|
|
byte = (data[stringtable_offset + i] - addition) & 0xFF
|
|
|
|
data[stringtable_offset + i] = byte
|
|
|
|
addition += 1
|
|
|
|
|
|
|
|
if byte == 0:
|
|
|
|
if curstring:
|
|
|
|
# We found a string!
|
2021-04-12 05:09:57 +02:00
|
|
|
self.__strings[curloc - stringtable_offset] = (bytes(curstring).decode('utf8'), False)
|
2021-04-11 22:44:31 +02:00
|
|
|
curloc = stringtable_offset + i + 1
|
|
|
|
curstring = []
|
|
|
|
curloc = stringtable_offset + i + 1
|
|
|
|
else:
|
|
|
|
curstring.append(byte)
|
|
|
|
|
|
|
|
if curstring:
|
|
|
|
raise Exception("Logic error!")
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
if 0 in self.__strings:
|
2021-04-11 22:44:31 +02:00
|
|
|
raise Exception("Should not include null string!")
|
|
|
|
|
|
|
|
return bytes(data)
|
|
|
|
|
|
|
|
def __get_string(self, offset: int) -> str:
|
|
|
|
if offset == 0:
|
|
|
|
return ""
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
self.__strings[offset] = (self.__strings[offset][0], True)
|
|
|
|
return self.__strings[offset][0]
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
def parse(self, verbose: bool = False) -> None:
|
2021-04-11 22:44:55 +02:00
|
|
|
with self.covered(len(self.data), verbose):
|
2021-04-11 22:45:17 +02:00
|
|
|
with self.debugging(verbose):
|
|
|
|
self.__parse(verbose)
|
2021-04-11 22:44:55 +02:00
|
|
|
|
|
|
|
def __parse(self, verbose: bool) -> None:
|
2021-04-11 22:44:31 +02:00
|
|
|
# First, use the byteswap header to descramble the data.
|
|
|
|
data = self.__descramble(self.data, self.descramble_info)
|
|
|
|
|
|
|
|
# Start with the basic file header.
|
|
|
|
magic, length, version, nameoffset, flags, left, right, top, bottom = struct.unpack("<4sIHHIHHHH", data[0:24])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(0, 24)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
ap2_data_version = magic[0] & 0xFF
|
|
|
|
magic = bytes([magic[3] & 0x7F, magic[2] & 0x7F, magic[1] & 0x7F, 0x0])
|
|
|
|
if magic != b'AP2\x00':
|
|
|
|
raise Exception(f"Unrecognzied magic {magic}!")
|
|
|
|
if length != len(data):
|
|
|
|
raise Exception(f"Unexpected length in AFP header, {length} != {len(data)}!")
|
|
|
|
if ap2_data_version not in [8, 9, 10]:
|
|
|
|
raise Exception(f"Unsupported AP2 container version {ap2_data_version}!")
|
|
|
|
if version != 0x200:
|
|
|
|
raise Exception(f"Unsupported AP2 version {version}!")
|
|
|
|
|
2021-04-12 05:09:57 +02:00
|
|
|
# The container version is analogous to the SWF file version. I'm pretty sure it
|
|
|
|
# dictates certain things like what properties are available. These appear strictly
|
|
|
|
# additive so we don't concern ourselves with this.
|
|
|
|
self.container_version = ap2_data_version
|
|
|
|
|
|
|
|
# The data version is basically used for how to parse tags. There was an older data
|
|
|
|
# version 0x100 that used more SWF-like bit-packed tags and while lots of code exists
|
|
|
|
# to parse this, the AP2 libraries will reject SWF data with this version.
|
|
|
|
self.data_version = version
|
|
|
|
|
|
|
|
# As far as I can tell, most things only care about the width and height of this
|
|
|
|
# movie, and I think the Shapes are rendered based on the width/height. However, it
|
|
|
|
# can have a non-zero x/y offset and I think this is used when rendering multiple
|
|
|
|
# movie clips?
|
|
|
|
self.location = Rectangle(left=left, right=right, top=top, bottom=bottom)
|
|
|
|
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x1:
|
|
|
|
# This appears to be the animation background color.
|
|
|
|
rgba = struct.unpack("<I", data[28:32])[0]
|
2021-04-12 05:09:57 +02:00
|
|
|
self.color = Color(
|
2021-04-11 22:44:31 +02:00
|
|
|
r=(rgba & 0xFF) / 255.0,
|
|
|
|
g=((rgba >> 8) & 0xFF) / 255.0,
|
|
|
|
b=((rgba >> 16) & 0xFF) / 255.0,
|
|
|
|
a=((rgba >> 24) & 0xFF) / 255.0,
|
|
|
|
)
|
|
|
|
else:
|
2021-04-12 05:09:57 +02:00
|
|
|
self.color = None
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(28, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x2:
|
|
|
|
# FPS can be either an integer or a float.
|
2021-04-12 05:09:57 +02:00
|
|
|
self.fps = struct.unpack("<i", data[24:28])[0] * 0.0009765625
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-12 05:09:57 +02:00
|
|
|
self.fps = struct.unpack("<f", data[24:28])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(24, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if flags & 0x4:
|
|
|
|
# This seems related to imported tags.
|
|
|
|
imported_tag_initializers_offset = struct.unpack("<I", data[56:60])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(56, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
|
|
|
# Unknown offset is not present.
|
|
|
|
imported_tag_initializers_offset = None
|
|
|
|
|
|
|
|
# String table
|
|
|
|
stringtable_offset, stringtable_size = struct.unpack("<II", data[48:56])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(48, 8)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Descramble string table.
|
|
|
|
data = self.__descramble_stringtable(data, stringtable_offset, stringtable_size)
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(stringtable_offset, stringtable_size)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Get exported SWF name.
|
|
|
|
self.exported_name = self.__get_string(nameoffset)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"{os.linesep}AFP name: {self.name}")
|
2021-04-12 05:09:57 +02:00
|
|
|
self.vprint(f"Container Version: {hex(self.container_version)}")
|
|
|
|
self.vprint(f"Version: {hex(self.data_version)}")
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"Exported Name: {self.exported_name}")
|
|
|
|
self.vprint(f"SWF Flags: {hex(flags)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x1:
|
2021-04-12 05:09:57 +02:00
|
|
|
self.vprint(f" 0x1: Movie background color: {self.color}")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(" 0x2: No movie background color")
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x2:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(" 0x2: FPS is an integer")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(" 0x2: FPS is a float")
|
2021-04-11 22:44:31 +02:00
|
|
|
if flags & 0x4:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(" 0x4: Imported tag initializer section present")
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(" 0x4: Imported tag initializer section not present")
|
2021-04-12 05:09:57 +02:00
|
|
|
self.vprint(f"Dimensions: {self.location.width}x{self.location.height}")
|
|
|
|
self.vprint(f"Requested FPS: {self.fps}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Exported assets
|
|
|
|
num_exported_assets = struct.unpack("<H", data[32:34])[0]
|
|
|
|
asset_offset = struct.unpack("<I", data[40:44])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(32, 2)
|
|
|
|
self.add_coverage(40, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Parse exported asset tag names and their tag IDs.
|
2021-04-12 05:09:57 +02:00
|
|
|
self.exported_tags = {}
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"Number of Exported Tags: {num_exported_assets}")
|
2021-04-11 22:44:31 +02:00
|
|
|
for assetno in range(num_exported_assets):
|
2021-04-12 05:09:57 +02:00
|
|
|
asset_tag_id, asset_string_offset = struct.unpack("<HH", data[asset_offset:(asset_offset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(asset_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
asset_offset += 4
|
|
|
|
|
|
|
|
asset_name = self.__get_string(asset_string_offset)
|
2021-04-12 05:09:57 +02:00
|
|
|
self.exported_tags[asset_name] = asset_tag_id
|
|
|
|
|
|
|
|
self.vprint(f" {assetno}: Tag Name: {asset_name}, Tag ID: {asset_tag_id}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Tag sections
|
|
|
|
tags_offset = struct.unpack("<I", data[36:40])[0]
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(36, 4)
|
2021-04-12 05:10:31 +02:00
|
|
|
self.tags, self.frames = self.__parse_tags(ap2_data_version, version, data, tags_offset)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Imported tags sections
|
|
|
|
imported_tags_count = struct.unpack("<h", data[34:36])[0]
|
|
|
|
imported_tags_offset = struct.unpack("<I", data[44:48])[0]
|
|
|
|
imported_tags_data_offset = imported_tags_offset + 4 * imported_tags_count
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(34, 2)
|
|
|
|
self.add_coverage(44, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"Number of Imported Tags: {imported_tags_count}")
|
2021-04-12 05:09:57 +02:00
|
|
|
self.imported_tags = {}
|
2021-04-11 22:44:31 +02:00
|
|
|
for i in range(imported_tags_count):
|
|
|
|
# First grab the SWF this is importing from, and the number of assets being imported.
|
|
|
|
swf_name_offset, count = struct.unpack("<HH", data[imported_tags_offset:(imported_tags_offset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(imported_tags_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
swf_name = self.__get_string(swf_name_offset)
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f" Source SWF: {swf_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
# Now, grab the actual asset names being imported.
|
|
|
|
for j in range(count):
|
|
|
|
asset_id_no, asset_name_offset = struct.unpack("<HH", data[imported_tags_data_offset:(imported_tags_data_offset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(imported_tags_data_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
asset_name = self.__get_string(asset_name_offset)
|
2021-04-12 05:09:57 +02:00
|
|
|
self.imported_tags[asset_id_no] = NamedTagReference(swf_name=swf_name, tag_name=asset_name)
|
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f" Tag ID: {asset_id_no}, Requested Asset: {asset_name}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
imported_tags_data_offset += 4
|
|
|
|
|
|
|
|
imported_tags_offset += 4
|
|
|
|
|
|
|
|
# This appears to be bytecode to execute on a per-frame basis. We execute this every frame and
|
|
|
|
# only execute up to the point where we equal the current frame.
|
|
|
|
if imported_tag_initializers_offset is not None:
|
|
|
|
|
|
|
|
unk1, length = struct.unpack("<HH", data[imported_tag_initializers_offset:(imported_tag_initializers_offset + 4)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(imported_tag_initializers_offset, 4)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
2021-04-11 22:45:17 +02:00
|
|
|
self.vprint(f"Imported Tag Initializer Offset: {hex(imported_tag_initializers_offset)}, Length: {length}")
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
for i in range(length):
|
|
|
|
item_offset = imported_tag_initializers_offset + 4 + (i * 12)
|
|
|
|
tag_id, frame, action_bytecode_offset, action_bytecode_length = struct.unpack("<HHII", data[item_offset:(item_offset + 12)])
|
2021-04-11 22:44:55 +02:00
|
|
|
self.add_coverage(item_offset, 12)
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if action_bytecode_length != 0:
|
2021-04-12 05:10:31 +02:00
|
|
|
self.vprint(f" Tag ID: {tag_id}, Frame: {frame}, ByteCode Offset: {hex(action_bytecode_offset + imported_tag_initializers_offset)}")
|
2021-04-11 22:44:31 +02:00
|
|
|
bytecode_data = data[(action_bytecode_offset + imported_tag_initializers_offset):(action_bytecode_offset + imported_tag_initializers_offset + action_bytecode_length)]
|
2021-04-12 05:10:31 +02:00
|
|
|
bytecode = self.__parse_bytecode(bytecode_data)
|
2021-04-11 22:44:31 +02:00
|
|
|
else:
|
2021-04-12 05:10:31 +02:00
|
|
|
self.vprint(f" Tag ID: {tag_id}, Frame: {frame}, No ByteCode Present")
|
|
|
|
bytecode = None
|
|
|
|
|
|
|
|
# Add it to the frame's instructions
|
|
|
|
if frame >= len(self.frames):
|
|
|
|
raise Exception(f"Unexpected frame {frame}, we only have {len(self.frames)} frames in this movie!")
|
|
|
|
self.frames[frame].imported_tags.append(TagPointer(tag_id, bytecode))
|
2021-04-11 22:44:31 +02:00
|
|
|
|
|
|
|
if verbose:
|
|
|
|
self.print_coverage()
|
2021-04-16 01:18:33 +02:00
|
|
|
|
|
|
|
self.parsed = True
|