Compiler: Construct-Dataclasses

[MatrixEditor]

Compiler

I wrote a "compiler" that creates Python files and writes struct definitions into construct-dataclasses types. In addition to the default construct compiler, this one also supports the generation of documentation comments and bit structures. The exported Python files contain a structure that is slightly different from the default Construct compiler:

# typing is used for type annotations on dataclass fields
import typing as t
import dataclasses as dc

# default imports
from construct import *
from construct.lib import *
from construct_dataclasses import *

# -- Enum section --
# ...

# Each dataclass is named with a trailing "_t"
@container
@dc.dataclass
class png_t:
    magic: bytes = csfield(Bytes(8))
    ihdr_len: int = csfield(Int32ub)
    ihdr_type: bytes = csfield(Bytes(4))
    ihdr: png__ihdr_chunk_t = csfield(LazyBound(lambda: png__ihdr_chunk))
    ihdr_crc: bytes = csfield(Bytes(4))
    chunks: t.List[png__chunk_t] = csfield(
        RepeatUntil(
            lambda obj_, list_, this: (
                # This PR also fixes an issue with the Construct compiler where a direct
                # reference to _io was made
                (obj_.type == "IEND") or (stream_iseof(this._io))
            ),
            LazyBound(lambda: png__chunk),
        )
    )


# The actual parser instance stores the original name
png = DataclassStruct(png_t)

# Schema definition and __all__ attribute for wildcard imports at the end
_schema = png
__all__ = [...]

The approach is used so that static code analyzers can utilize the constructor signature of the generated dataclasses (I know not very pretty, but using @dataclass_struct makes constructor arguments invisible):

Why?

1. Dataclass objects are more useful than default Container or ListContainer, and they can be easily converted to JSON for further processing.
2. Since this is not a "real" language, I suppose it can be integrated more easily.

Other Features

• This pull request also fixes an issue with the construct compiler that in certain situations writes only "_io" instead of using the this._io reference.
• Documentation comments are generated using a class taken from Generate docs in Construct compiler #243