This is a (relatively) simple, pure-Python, no dependency library, aiming to simplify parsing and building binary data structures. It uses dataclasses
as its main container type, and struct
-compatible format specifiers for writing field definitions.
The way of composing structures is somewhat similar to (and inspired by) Construct. While probably not as powerful, it should give more flexibility and control over the data, as well as full IDE type hinting.
pip install py-datastruct
NOTE: pip install datastruct
installs a different package by the same name!
In DataStruct v2.0.0, the field type validation methods have been rewritten. They are now stricter, which means that the type hints will more closely represent the actual possible field values.
The new mechanism allows using union types (int | float | bytes
), as well as optional fields (MyStruct | None
) for fields, which wasn't previously possible. This is particularly useful for cond()
and switch()
fields.
Due to this new logic, there are a few breaking changes in v2.0.0:
`cond()` field default `if_not=` value is now `None` (breaking)
Previously, if the cond()
field evaluated to False
, its value was set to the wrapped field's default value (unless otherwise specified using the if_not=
argument). For subfield()
, the structure was created using default values.
Starting in v2.0.0, the field's value will be set to None
. You can still use if_not=
to change that (which you should do, if you rely on that field's default value in any way). This means, that the cond()
field's type specification must now include None
as one of its types.
If your structure was:
@dataclass
class MyStruct(DataStruct):
var: int = cond(lambda ctx: ctx.my_condition)(field("I"))
it must now be changed to either:
@dataclass
class MyStruct(DataStruct):
var: int | None = cond(lambda ctx: ctx.my_condition)(field("I"))
or, using if_not=
:
@dataclass
class MyStruct(DataStruct):
var: int = cond(lambda ctx: ctx.my_condition, if_not=0)(field("I"))
The same change applies to subfield()
wrapped in cond()
.
Note that you cannot use Any
for the cond()
field, unless it wraps a switch()
field (in which case the cond()
field's type is transparently proxied to the switch()
field).
`switch()` field's type must now account for all possible cases (possibly breaking)
Since union types are now usable with switch()
fields, it is required to include all possible cases in the union.
The following structure demonstrates various ways of using the switch()
field correctly:
@dataclass
class MyStruct(DataStruct):
var1: int = switch(False)(
false=(int, field("H")),
true=(int, field("I")),
)
var2: int | bool = switch(False)(
false=(int, field("H")),
true=(bool, field("B")),
)
var3: Any = switch(False)(
false=(int, field("H")),
true=(bool, field("B")),
)
var4: Any = switch(False)(
false=(..., padding(4)),
true=(int, field("I")),
)
var5: ... = switch(False)(
false=(..., padding(4)),
true=(int, field("I")),
)
Note that the usage of Ellipsis (...
) is restricted for switch()
fields that have at least one case using the ...
type.
By the examples above, if you have a switch()
field that uses union types, but doesn't list all possible cases, you should either add the missing types or change the type to Any
.
If your switch()
field uses subfield()
cases, and you don't want to use the Any
type, and you don't want to list all possible types, consider using a base class (this is now possible!), like this:
@dataclass
class MyBase(DataStruct):
# you can optionally add fields here - they will be *before* any subclass' fields
pass
@dataclass
class MyStruct1(MyBase): # note - no DataStruct here!
pass
@dataclass
class MyStruct2(MyBase):
pass
@dataclass
class MySwitchStruct(MyBase):
var1: MyBase = switch(False)(
false=(MyStruct1, subfield()),
true=(MyStruct2, subfield()),
)
The minimum required Python version is now 3.8
While it may still work on 3.7, it is recommended to use 3.10 at least. It should work on 3.8, but I can't reliably test everything on old versions to make sure it's fine.
Before you read this "documentation", be aware that it is by no means complete, and will probably be not enough for you to understand everything you need.
Here are a few projects that are using datastruct
:
- https://github.com/tuya-cloudcutter/cloudcutter-universal/blob/master/cloudcutter/modules/dhcp/structs.py
- https://github.com/tuya-cloudcutter/bk7231tools/blob/main/bk7231tools/analysis/kvstorage.py
- https://github.com/libretiny-eu/ltchiptool/blob/master/uf2tool/models/partition.py
- https://github.com/libretiny-eu/ltchiptool/blob/master/ltchiptool/soc/ambz2/util/models/images.py
If you want your project on this list, feel free to submit a PR.
This simple example illustrates creating a 24-byte long structure, consisting of a 32-bit integer, an 8-byte 0xFF-filled padding, and a 12-byte bytes
string.
from hexdump import hexdump
from dataclasses import dataclass
from datastruct import DataStruct
from datastruct.fields import field, padding
@dataclass
class MyStruct(DataStruct):
my_number: int = field("I", default=123)
_1: ... = padding(8)
my_binary: bytes = field("12s")
my_object = MyStruct(my_binary=b"Hello Python")
print(my_object)
# MyStruct(my_number=123, my_binary=b'Hello World!')
my_object = MyStruct(my_number=5, my_binary=b"Hello World!")
print(my_object)
# MyStruct(my_number=5, my_binary=b'Hello World!')
packed = my_object.pack()
hexdump(packed)
# 00000000: 05 00 00 00 FF FF FF FF FF FF FF FF 48 65 6C 6C ............Hell
# 00000010: 6F 20 57 6F 72 6C 64 21 o World!
unpacked = MyStruct.unpack(packed)
print(unpacked)
# MyStruct(my_number=5, my_binary=b'Hello World!')
print(my_object == unpacked)
# True
You might also pass a stream (file/BytesIO/etc.) to pack()
and unpack()
. Otherwise, pack()
will create a BytesIO stream and return its contents after packing; unpack()
will accept a bytes
object as its parameter.
pack()
and unpack()
also accept custom, keyword-only arguments, that are available in the Context, throughout the entire operation.
Upon starting a pack/unpack operation, a Context
object is created. The context is a container scoped to the currently processed structure. It's composed of the following main elements:
- all values of the current structure - when packing; during unpacking, it contains all values of fields that were already processes (the context "grows")
- all keyword arguments passed to
pack()
/unpack()
(for the root context only) - all keyword arguments passed to
subfield()
(for child contexts only) _: Context
- reference to the parent object's context (only when nestingDataStruct
s)self: Any
- the current datastruct - note that it's aDataStruct
subclass when packing, and aContainer
when unpackingG
- global context - general-purpose container that is not scoped to the current structure (it's identical for nested structs)io: IO[bytes]
- the stream being read from/written topacking: bool
- whether current operation is packingunpacking: bool
- whether current operation is unpackingroot: Context
- context of the topmost structuretell: () -> int
- function returning the current position in the streamseek: (offset: int, whence: int) -> int
- function allowing to seek to an absolute offset
P
- local context - general-purpose container that is different for each nested structconfig: Config
- current DataStruct's configtell: () -> int
- function returning the current position in the current structure (in bytes)seek: (offset: int, whence: int) -> int
- function allowing to seek to an offset within the current structureskip: (length: int) -> int
- function allowing to skiplength
bytesi: int
- (forrepeat()
fields only) index of the current item of the listitem: Any
- (forrepeat()
fields, inlast=
lambda only) item processed right before evaluationself: Any
- (packing only) value of the current field
The context is "general-purpose", meaning that the user can write custom values to it. All fields presented above can be accessed by lambda functions - see "Parameter evaluation".
Most field parameters support pack/unpack-time evaluation (which means they can e.g. depend on previously read fields). Lambda expressions are then given the current context, and expected to return a simple value, that would be statically valid in this parameter.
an_unpredictable_field: int = field(lambda ctx: "I" if randint(1, 10) % 2 == 0 else "H")
A special value of type Ellipsis
/...
is used in the library, to indicate something not having a type or a value. It's not the same as None
. built()
fields, for example, have ...
as value after creating the struct, but before packing it for the first time.
Special fields (like padding()
, which don't have any value) must have ...
as their type hint.
This is a simple example of using parameter evaluation to dynamically size a bytes
string. Binary strings use the <len>s
specifier, which can be omitted (simple int
can be used instead).
@dataclass
class MyStruct(DataStruct):
data_length: int = field("I")
data: bytes = field(lambda ctx: ctx.data_length)
The user is still responsible for adjusting data_length
after changing data
. The built()
field comes in handy here:
@dataclass
class MyStruct(DataStruct):
data_length: int = built("I", lambda ctx: len(ctx.data))
data: bytes = field(lambda ctx: ctx.data_length)
When unpacking, the data_length
field will be used to dynamically size the data
field. When packing, data_length
will always be recalculated based on what's in data
.
Lists are also iterables, like bytes
, but they store a number of items of a specific type. Thus, the repeat()
field wrapper has to be used.
Wrapper fields simply require calling them first with any used parameters, then with the "base" field.
@dataclass
class MyStruct(DataStruct):
item_count: int = built("H", lambda ctx: len(ctx.items))
# This creates a list of 16-bit integers.
# The list is empty by default.
items: List[int] = repeat(lambda ctx: ctx.item_count)(field("H"))
my_object = MyStruct()
my_object.items = [0x5555, 0x4444, 0x3333, 0x2222]
my_object.item_count = 1 # this doesn't matter, as the field is rebuilt
packed = my_object.pack()
hexdump(packed)
# 00000000: 04 00 55 55 44 44 33 33 22 22
They're also wrapper fields - if the condition is not met, they act like as if the field didn't exist at all.
@dataclass
class MyStruct(DataStruct):
has_text: bool = field("?")
text: str = cond(lambda ctx: ctx.has_text)(field("8s", default=""))
my_object = MyStruct.unpack(b"\x01HELOWRLD")
print(my_object)
# MyStruct(has_text=True, text='HELOWRLD')
my_object = MyStruct.unpack(b"\x00")
print(my_object)
# MyStruct(has_text=False, text='')
Switch fields are like more powerful conditional fields. The following example reads an 8/16/32-bit number, depending on the prefixing length byte. If the length is not supported, it reads the value as bytes
instead.
number_length: int = field("B", default=1)
number: Union[int, bytes] = switch(lambda ctx: ctx.number_length)(
_1=(int, field("B")),
_2=(int, field("H")),
_4=(int, field("I")),
default=(bytes, field(lambda ctx: ctx.number_length)),
)
The values on the left (_1
, _2
, _4
) are the keys. The key is picked depending on the key-lambda result (ctx.number_length
). The value on the right is a tuple of the expected field type, and a field()
specifier.
Since it's not possible to pass just 1
as a keyword argument, integers are looked up prefixed with an underscore as well. Enums are additionally looked up by their name and value, and booleans are looked up by lowercase true
/false
.
Note that you can pass (probably) any kind of field to the switch list.
MIT License
Copyright (c) 2023 Kuba Szczodrzyński
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.