OxidizedFinder Python Type

oxidized_importer.OxidizedFinder is a Python type that implements a custom meta path finder. Oxidized is in its name because it is implemented in Rust.

Unlike traditional meta path finders which have to dynamically discover resources (often by scanning the filesystem), OxidizedFinder instances maintain an index of known resources. When a resource is requested, OxidizedFinder can retrieve that resource by effectively performing 1 or 2 lookups in a Rust HashMap. This makes resource resolution extremely efficient.

Instances of OxidizedFinder are optionally bound to a binary blob holding packed resources data. This is a custom serialization format for expressing Python modules (source and bytecode), Python extension modules, resource files, shared libraries, etc. This data format along with a Rust library for interacting with it are defined by the python-packed-resources crate.

When an OxidizedFinder instance is created, the packed resources data is parsed into a Rust data structure. On a modern machine, parsing this resources data for the entirety of the Python standard library takes ~1 ms.

OxidizedFinder instances can index built-in extension modules and frozen modules, which are compiled into the Python interpreter. This allows OxidizedFinder to subsume functionality normally provided by the BuiltinImporter and FrozenImporter meta path finders, allowing you to potentially replace sys.meta_path with a single instance of OxidizedFinder.

OxidizedFinder in PyOxidizer Applications

When running from an application built with PyOxidizer (or using the pyembed crate directly), an OxidizedFinder instance will (likely) be automatically registered as the first element in sys.meta_path when starting a Python interpreter.

You can verify this inside a binary built with PyOxidizer:

>>> import sys
>>> sys.meta_path
[<OxidizedFinder object at 0x7f16bb6f93d0>]

Contrast with a typical Python environment:

>>> import sys
>>> sys.meta_path
    <class '_frozen_importlib.BuiltinImporter'>,
    <class '_frozen_importlib.FrozenImporter'>,
    <class '_frozen_importlib_external.PathFinder'>

The OxidizedFinder instance will (likely) be associated with resources data embedded in the binary.

This OxidizedFinder instance is constructed very early during Python interpreter initialization. It is registered on sys.meta_path before the first import requesting a .py/.pyc is performed, allowing it to service every import except those from the very few built-in extension modules that are compiled into the interpreter and loaded as part of Python initialization (e.g. the sys module).

Python API

OxidizedFinder instances implement the following interfaces:

  • importlib.abc.MetaPathFinder
  • importlib.abc.Loader
  • importlib.abc.InspectLoader
  • importlib.abc.ExecutionLoader

See the importlib.abc documentation for more on these interfaces.

In addition to the methods on the above interfaces, the following methods defined elsewhere in importlib are exposed:

  • get_resource_reader(fullname: str) -> importlib.abc.ResourceReader
  • find_distributions(context: Optional[DistributionFinder.Context]) -> [Distribution]

ResourceReader is documented alongside other importlib.abc interfaces. find_distribution() is documented in importlib.metadata.

Non-importlib API

OxidizedFinder instances have additional functionality beyond what is defined by importlib. This functionality allows you to construct, inspect, and manipulate instances.

__new__(cls, ...)

New instances of OxidizedFinder can be constructed like normal Python types:

finder = OxidizedFinder()

The constructor takes the following named arguments:

Bytes-like packed resources data to parse. A reference to the passed in value will be stored internally in the constructed instance, as the memory needs to live for the lifetime of the OxidizedFinder instance.
A path-like object defining the filesystem path to a file containing packed resources data. If provided, the file will be opened and memory mapped and resources data will be parsed from it.
A path-like object denoting the filesystem path that should be used as the origin value for relative path resources. Filesystem-based resources are stored as a relative path to an anchor value. This is that anchor value. If not specified, the directory of the current executable will be used.

See the python_packed_resources Rust crate for the specification of the binary data blob defining packed resources data.


The packed resources data format is still evolving. It is recommended to use the same version of the oxidized_importer extension to produce and consume this data structure to ensure compatibility.

indexed_resources(self) -> List[OxidizedResource]

This method returns a list of resources that are indexed by the OxidizedFinder instance. It allows Python code to inspect what the finder knows about.

See OxidizedResource for more on the returned type.

add_resource(self, resource: OxidizedResource)

This method registers an OxidizedResource instance with the finder, enabling the finder to use it to service lookups.

When an OxidizedResource is registered, its data is copied into the finder instance. So changes to the original OxidizedResource are not reflected on the finder. (This is because OxidizedFinder maintains an index and it is important for the data behind that index to not change out from under it.)

Resources are stored in an invisible hash map where they are indexed by the name attribute. When a resource is added, any existing resource under the same name has its data replaced by the incoming OxidizedResource instance.

If you have source code and want to produce bytecode, you can do something like the following:

def register_module(finder, module_name, source):
    code = compile(source, module_name, "exec")
    bytecode = marshal.dumps(code)

    resource = OxidizedResource()
    resource.name = module_name
    resource.flavor = "module"
    resource.in_memory_bytecode = bytecode
    resource.in_memory_source = source


add_resources(self, resources: List[OxidizedResource])

This method is syntactic sugar for calling add_resource() for every item in an iterable. It is exposed because function call overhead in Python can be non-trivial and it can be quicker to pass in an iterable of OxidizedResource than to call add_resource() potentially hundreds of times.

serialize_indexed_resources(self, ...) -> bytes

This method serializes all resources currently indexed by the instance into an opaque bytes instance. The returned data can be fed into a separate OxidizedFinder instance by passing it to __new__(cls, ...).


ignore_builtin (bool)

Whether to ignore builtin extension modules from the serialized data.

Default is True

ignore_frozen (bool)

Whether to ignore frozen extension modules from the serialized data.

Default is True.

Entries for built-in and frozen modules are ignored by default because they aren’t portable, as they are compiled into the interpreter and aren’t guaranteed to work from one Python interpreter to another. The serialized format does support expressing them. Use at your own risk.