|author||Kevin Cheng <firstname.lastname@example.org>||Thu May 02 18:31:40 2019 -0700|
|committer||android-build-merger <email@example.com>||Thu May 02 18:31:40 2019 -0700|
Merge commit '6060d29' into import am: b80da1501c Change-Id: I25a25945607fa45d3815357a30a233608a95f23f
A fast, pure Python library for parsing and serializing ASN.1 structures.
In addition to an ASN.1 BER/DER decoder and DER serializer, the project includes a bunch of ASN.1 structures for use with various common cryptography standards:
|CSR||RFC 2986, RFC 2985|
|PKCS#1 v2.1 (RSA keys)||RFC 3447|
|DSA keys||RFC 3279|
|Elliptic curve keys||SECG SEC1 V2|
|PKCS#3 v1.4||PKCS#3 v1.4|
|PKCS#5 v2.1||PKCS#5 v2.1|
|CMS (and PKCS#7)||RFC 5652, RFC 2315|
|PDF signatures||PDF 1.7|
Python has long had the pyasn1 and pyasn1_modules available for parsing and serializing ASN.1 structures. While the project does include a comprehensive set of tools for parsing and serializing, the performance of the library can be very poor, especially when dealing with bit fields and parsing large structures such as CRLs.
After spending extensive time using pyasn1, the following issues were identified:
The pyasn1 API is largely method driven, and uses extensive configuration objects and lowerCamelCase names. There were no consistent options for converting types of native Python data structures. Since the project supports out-dated versions of Python, many newer language features are unavailable for use.
Time was spent trying to profile issues with the performance, however the architecture made it hard to pin down the primary source of the poor performance. Attempts were made to improve performance by utilizing unreleased patches and delaying parsing using the
Any type. Even with such changes, the performance was still unacceptably slow.
Finally, a number of structures in the cryptographic space use universal data types such as
OctetString, but interpret the data as other types. For instance, signatures are really byte strings, but are encoded as
BitString. Elliptic curve keys use both
OctetString to represent integers. Parsing these structures as the base universal types and then re-interpreting them wastes computation.
asn1crypto uses the following techniques to improve performance, especially when extracting one or two fields from large, complex structures:
While there is no extensive performance test suite, the
CRLTests.test_parse_crl test case was used to parse a 21MB CRL file on a late 2013 rMBP. asn1crypto parsed the certificate serial numbers in just under 8 seconds. With pyasn1, using definitions from pyasn1-modules, the same parsing took over 4,100 seconds.
For smaller structures the performance difference can range from a few times faster to an order of magnitude of more.
asn1crypto is part of the modularcrypto family of Python packages:
0.24.0 - changelog
Python 2.6, 2.7, 3.2, 3.3, 3.4, 3.5, 3.6 or pypy. No third-party packages required.
pip install asn1crypto
asn1crypto is licensed under the terms of the MIT license. See the LICENSE file for the exact license text.
The documentation for asn1crypto is composed of tutorials on basic usage and links to the source for the various pre-defined type classes.
Tests are written using
unittest and require no third-party packages:
python run.py tests
To run only some tests, pass a regular expression as a parameter to
python run.py tests ocsp
To install the package used for linting, execute:
pip install --user -r requires/lint
The following command will run the linter:
python run.py lint
Support for code coverage can be installed via:
pip install --user -r requires/coverage
Coverage is measured by running:
python run.py coverage
To install the necessary packages for releasing a new version on PyPI, run:
pip install --user -r requires/release
Releases are created by:
Making a git tag in semver format
Running the command:
python run.py release
Existing releases can be found at https://pypi.python.org/pypi/asn1crypto.
A task named
deps exists to ensure a modern version of
pip is installed, along with all necessary testing dependencies.
ci task runs
lint (if flake8 is available for the version of Python) and
tests if coverage is not available for the version of Python). If the current directory is a clean git working copy, the coverage data is submitted to codecov.io.
python run.py deps python run.py ci