Skip to content

Archivey developer guide

This guide explains how to extend archivey by creating custom archivey.archive_reader.ArchiveReader classes for new archive formats.

Overview

The library's modules are organized into these packages: * archivey.core and related modules – public API functions, types and classes * archivey.internal – base classes and helpers * archivey.formats – format-specific readers

The library exposes an archivey.archive_reader.ArchiveReader abstract base class for users with the public API. The actual format readers extend from the helper class archivey.internal.base_reader.BaseArchiveReader, which implements most of the public API by delegating to some simpler methods that the readers must implement. New readers will almost always want to inherit from BaseArchiveReader.

Registering Your Reader

For your reader to be called, you'll need to:

  • add the format(s) your reader handles to archivey.types.ArchiveFormat;
  • detect archives of these formats by signature and/or filename, in archivey.formats.format_detection;
  • create your reader class, see below;
  • modify archivey.core.open_archive to associate the archive format with your reader.

Creating a new format reader

It's probably easiest to understand what to do by looking at existing reader implementations. archivey.formats.zip_reader.ZipReader may be a good starting point.

Your reader should implement a few required methods, and may also implement some optional ones in special cases. These are the required methods:

Constructor (__init__):

Should accept the same parameters as other ArchiveReaders, so it can be called from archivey.core.open_archive:

    def __init__(
        self,
        archive_path: BinaryIO | str | bytes | os.PathLike,
        format: ArchiveFormat,
        *,
        pwd: bytes | str | None = None,
        streaming_only: bool = False,
    ):
  • archive_path: the path to the archive file, or a stream or file-like object.
  • format: an enum value from archivey.types.ArchiveFormat with the detected archive format. It will be one of the format your reader handles.
  • pwd: the password to use for encrypted members and/or archive headers.
  • streaming_only: set to True if the archive is being opened only for sequential, forward-only access. If True, methods like open() (for random access) and extract() will be disabled, and iteration will be a one-time operation.

The constructor should call archivey.internal.base_reader.BaseArchiveReader.__init__ with the same parameters, plus another one:

   super().__init__(format, archive_path, pwd, streaming_only, members_list_supported)
  • members_list_supported: set this to True if your reader can provide a complete list of archive members upfront (e.g., by reading a central directory like in ZIP files) without reading the entire archive. If False and if the file is opened in streaming-only mode, get_members_if_available() will not return a list of members until after the archive has been iterated through.

The constructor should initialize any internal state or open resources specific to the archive format you are supporting (e.g., open the archive file using a third-party library).

_close_archive(self) -> None

This method is called exactly once when the ArchiveReader is closed or destructed. It should release any resources held by the reader (e.g., close the archive object provided by the underlying library, cleanup temporary files).

iter_members_for_registration(self) -> Iterator[ArchiveMember]

This method is called by BaseArchiveReader to discover and register all members in the archive. It should be implemented to yield archivey.types.ArchiveMember objects one by one, from the members in the archive. It will be iterated through only once as the members are needed, so you can use an iterator from the inner library if available.

For each member in the archive, you need to convert the metadata (filename, size, modification time, type, permissions, etc.) read from the underlying archive library/format, and create an ArchiveMember instance with the appropriate field values. Typically you'll store the library-specific, original member object in the raw_info field of the ArchiveMember. Don't set _archive_id or _member_id, they'll be set by BaseArchiveReader.

Pay special attention to the modification time field. Some archive formats store member times as UTC timestamps, others using the local timezone, some may even contain both. You should set the mtime_with_tz field to a datetime.datetime object with the timezone set if known (likely UTC), or without a timezone if it's in local time.

get_archive_info(self) -> ArchiveInfo

Return an archivey.types.ArchiveInfo object with all archive-level metadata:

  • format: the ArchiveFormat enum for this archive.
  • is_solid: whether the archive is solid (True means that decompressing a specific member may require other members to be uncompressed first).
  • please look at the ArchiveInfo class for other fields
  • extra: A dictionary for any other format-specific information that doesn't have a dedicated field

_open_member(self, member: ArchiveMember, *, pwd: Optional[Union[bytes, str]] = None, for_iteration: bool = False) -> BinaryIO

This method is called by BaseArchiveReader to open a specific archive member, when the user calls open(), or during iteration or extraction. It should return a binary I/O stream (BinaryIO) to read the member's contents. Ideally the reader should avoid loading the whole contents to memory at once, and decompress the data as it's read by the client.

It is guaranteed to be called only for MemberType.FILE members (i.e. not directories or links). If the archive has been opened in streaming_only mode, it is guaranteed to be called only after the member has been yielded by iter_members_for_registration and before that iterator continues (but it is not guaranteed to be called for all files, and the stream is not guaranteed to be fully read), and is guaranteed to be closed before the iterator continues.

Arguments:

  • member: an ArchiveMember object previously yielded by iter_members_for_registration.
  • pwd: password for encrypted members if applicable. If the user does not specify a file-specific password, this will default to the password passed in the constructor, if any.
  • for_iteration: if True, the open request is part of a sequential iteration (e.g., via iter_members_with_streams). When streaming_only=True, this is guaranteed to be True.

Tips:

  • As some archive formats (e.g. tar) may contain multiple members with the same file name, you should pass member.raw_info object to the underlying library if possible instead of the filename, to avoid opening a different member with the same name.
  • Exceptions raised while reading the returned stream will be passed through your _translate_exception() method to ensure they become ArchiveError subclasses.

Optional Methods to Override

While the above are essential, you might override other methods from BaseArchiveReader for efficiency or specific behavior:

  • _prepare_member_for_open(self, member: ArchiveMember, *, pwd: bytes | str | None, for_iteration: bool) -> ArchiveMember:

    • This is a hook called by BaseArchiveReader just before it calls your _open_member method.
    • The base implementation simply returns the member unmodified; you can override this to perform tasks like fetching additional metadata required for opening, or decrypting member-specific headers, if not done during iter_members_for_registration.
    • This method receives the ArchiveMember as initially resolved from the filename or iteration, which may be a MemberType.LINK. _open_member will then be called with the target of this member if it's a link (after internal resolution), or the same member if not a link.

  • iter_members_with_streams(...):

    • The default implementation in BaseArchiveReader iterates using self.iter_members() (which relies on iter_members_for_registration) and then calls the internal open mechanism (which in turn uses your _prepare_member_for_open and _open_member methods with the for_iteration=True flag) for each member.
    • If your underlying library requires a different approach to iterate through members and get their I/O streams, you can override this method directly.
    • Important: If overridden, you are responsible for correctly applying filtering logic based on the members and filter arguments. BaseArchiveReader._build_filter can be a useful utility for this. Please look at existing implementations for details.

  • _extract_pending_files(self, path: str, extraction_helper: ExtractionHelper, pwd: bytes | str | None):

    • BaseArchiveReader.extractall() uses an ExtractionHelper. If the reader was opened for random access (streaming_only=False), the extraction first identifies all files to extract by applying any filters, handling directories and links, and then calls _extract_pending_files() with only the MemberType.FILE members that need to be extracted.
    • The default implementation of _extract_pending_files() simply calls the public self.open() method for each file and extracts it, sequentially. If your underlying library has a more optimized (e.g. multithreaded) way to extract multiple files at once (e.g., zipfile.ZipFile.extractall()), override this method to use that more efficient approach.
    • Your implementation should read pending extractions via extraction_helper.get_pending_extractions(), extract the files and then call extraction_helper.process_file_extracted(member, extracted_file_path). See the implementation in archivey.formats.sevenzip_reader.SevenZipReader for an example.
    • Note: The extractall method in BaseArchiveReader handles overall filtering. If you override _extract_pending_files specifically, it receives a list of already filtered members to extract from the extraction_helper. If you were to override extractall itself, you'd need to manage filtering and other details.

Exception handling

Libraries often have their own exception base classes, or raise builtin exceptions such as OSError when there's a problem with an archive. Archivey tries to guarantee that all exceptions raised due to archive issues are subclasses of archivey.exceptions.ArchiveError, and so readers need to translate all exceptions raised by the libraries into them.

When your reader returns a stream from _open_member(), BaseArchiveReader wraps it in an internal ArchiveStream. Any exceptions raised while reading from the stream will be passed to your _translate_exception() method so that they can be converted to ArchiveError subclasses.

To do so, implement a translation function that receives any exception raised by the underlying library and returns an ArchiveError (or None to propagate the original error). Example:

from archivey.exceptions import ArchiveCorruptedError, ArchiveIOError # Corrected path
# Import specific exceptions from your third-party library
from third_party_lib import ThirdPartyReadError, ThirdPartyCorruptError

class MyReader(BaseArchiveReader):
    ...

    def _translate_exception(exc: Exception) -> Optional[ArchiveError]:
        if isinstance(exc, ThirdPartyCorruptError):
            return ArchiveCorruptedError(f"Archive data seems corrupted: {exc}")
        elif isinstance(exc, ThirdPartyReadError):
            return ArchiveIOError(f"I/O error while reading member: {exc}")
        # Add more specific translations as needed
        return None # Let other exceptions (or already ArchiveErrors) pass through

    def _open_member(self, member: ArchiveMember, pwd: Optional[str | bytes] = None, for_iteration: bool = False):
        ...
        return underlying_library.open_member(member.raw_info)

When opening an archive in your reader __init__(), an exception may also be raised. You may use the same translation function by wrapping your archive opening with archivey.internal.io_helpers.run_with_exception_translation.

Be sure to add extensive tests of your format, especially of corrupted archives, wrong passwords and any other corner cases you can think of, so that you can be sure all possible exceptions are being caught.

Tips:

  • Specificity: Your exception_translator function should be as specific as possible. Catch known exceptions from the third-party library you are using and map them to appropriate ArchiveError subclasses (e.g., ArchiveCorruptedError, ArchiveIOError, ArchiveEncryptedError).
  • Avoid generic Exception: It's fine to catch the base exception class from your library and raise a base ArchiveError, but you should not add a catch-all to convert any Exception that may be received, as it can hide code bugs or unexpected behavior. If the exception is unknown, return None to raise the original exception. (yes, this will result in exceptions raised in client code, but then we'll know about it and can hopefully add them to the list)
  • Testing: It's highly recommended to write tests that specifically trigger various error conditions in the underlying library to ensure your translator handles them correctly. This might involve creating corrupted or specially crafted archive files, passing wrong passwords etc. See the Testing section below.

Linting

Ensure Hatch and uv are installed. The lint script runs Ruff and Pyright; the latter is executed via npx, so Node.js and npm must be available. Run the linters with:

hatch run lint

On the first run npx may ask to download Pyright – answer y to continue.

Testing

Run the tests through Hatch, which executes pytest via uv with the optional dependencies enabled:

hatch run test

You can run a subset of tests with the -k option, e.g. to run only ZIP related tests:

hatch run test -k .zip

Sample archives used by the tests are versioned in tests/test_archives. If you add new archives or change them, regenerate the files with:

uv run --extra optional python -m tests.archivey.create_archives [pattern]

Omit the optional pattern to rebuild all archives. RAR tests require the unrar tool. If it's missing those tests will fail and can be ignored.

Building the documentation

The API documentation is generated dynamically using mkdocstrings and assembled into a static site with MkDocs. Run Hatch to build everything:

hatch run docs
This copies the project README.md to docs/index.md and then builds the MkDocs site (into site/). Ensure the optional development dependencies are installed (e.g. pip install -e .[dev]).