Contributing to crypto-condor#

Be it issues, pull requests, or suggestions, contributions are welcome!

Requirements for development#

The external dependencies are (not including Python packages):

To configure the Python dependencies and the repo:

  • Run make install to install the development dependencies.

  • Use a virtual environment with poetry shell.

  • Run make init to configure the repo.

  • When coding and committing, run make all to run the test suite.

make install installs all the dependencies, including the development and documentation dependencies, inside a virtual environment[1]. It uses a lock file (poetry.lock) to ensure that package versions are consistent between developers’ environments.

It also installs the project as an editable package inside that virtual environment, meaning that changes to the source code are immediately reflected in the package. The CLI is installed as crypto-condor-cli.

poetry shell activates the virtual environment inside a sub-shell that can be exited with ctrl+D or by typing exit. This is the recommended way, as otherwise all commands requiring the venv’s python would have to be preceded by poetry run.

make all is the go-to command for testing changes. It runs the linter, tests, and coverage. It also builds the docs, checking for errors and running the doctest examples, ensuring they stay up-to-date.

You can run commands normally inside this sub-shell:

# Display the CLI's help.
crypto-condor-cli --help

# Run the unit tests.
make test

Design#

The source code is inside the crypto_condor directory. It has three main modules: cli, primitives, and vectors.

The cli is divided in commands, some of which have their own module. The main app can be found in main.py and contains some generic commands (i.e. that do not depend on a given primitive) such as method and get-wrapper.

The primitives are separated by modules, each with their own functions to test implementations, protocols to describe the expected function signatures, and enums to define the parameters (mode of operation, elliptic curve, etc.).

The vectors contain subdirectories where the source files for test vectors are stored. These are parsed with the primitive’s _import.py script and then serialised with protobuf, making it easy to load them at runtime.

Finally, there is a fourth directory, resources. It contains the version of the method guides that is used by the method commands, as well as the wrapper templates and examples for each primitive.

Documentation#

The documentation is generated with Sphinx. Most documents are written in Markdown, thanks to MyST parser. The exception to this are the documents that make use of autodoc directives like autofunction, as the sphinx.ext.autodoc extension doesn’t support Markdown files.[2]

Protobuf#

We use protobuf to store test vectors. Protobuf uses .proto files that describe the message (in our case the vectors). These are then compiled with protoc to Python classes. For type-checking and adding docstrings to these classes, we use mypy-protobuf, which creates .pyi files when compiling with protoc.

You can use the Makefile target compile-proto to compile the protobufs. It finds the corresponding files, and only updates those that require it. It also shows the protoc version, which should preferably be included in the commit message.

Testing#

Testing is done with pytest and pytest-cov for code coverage. The structure of tests reflects that of crypto_condor: tests under primitives/ test the functions and implementations directly, as a library user would use them, and tests under cli/ test the CLI commands. This includes running the wrapper examples bundled with crypto-condor, which is especially useful as these examples cover a lot of code, from the CLI to the primitives and test vectors.

Adding new primitives#

Here are some guidelines on how to add a new primitive. To get started, the handy utils/add_primitive.py script creates templates of most of the necessary files:

python utils/add_primitive.py <primitive name>

From here on out, we’ll use AES as an example.

Test vectors#

TL;DR:

  • Create a protobuf descriptor:

    • Add a parameter to Vectors that characterises a set of tests.

    • Add the necessary fields to Test so any source of vectors is supported.

  • Create a parsing script.

First, there are the test vectors. It creates a directory named _AES to store the source files, protobuf descriptors, parsing script, and the serialized vectors. We mainly use test vectors from NIST CAVP and Project Wycheproof, though any reference (RFC, official implementation, etc.) source is welcome.

All test vectors should be serialized. To serialize test vectors we use Protocol Buffers or protobufs for short. You will need two files: a protobuf descriptor and a parsing script.

The protobuf descriptor is a .proto file that defines the messages and their attributes, similar to Python dataclasses. add_primitive.py creates it with two types of message needed by crypto-condor: Vectors and Test. These are already filled with a common set of values requiring minimal changes. Vectors should define a characterising parameter (mode of operation, elliptic curve, etc.), while Test should define fields that allow any source of vectors to be added easily. This is pretty vague, but the idea is to stop having separate types of vectors for each source, which may require a bit of additional logic in the parsing to conform to a “standard format”.

The descriptor is compiled using protoc to a Python module that provide the messages as classes, which can be imported and used by the primitive module.

The parsing script will use these classes, creating a new instance for each group of vectors, and parsing the text file to extract the values of each vector.

The add_primitive script also generates a JSON file, which should associate the chosen parameter to a list of protobufs with vectors for that parameter.

Primitive#

Second, it creates the primitive module, primitives/AES.py in this case, where the code to test implementations will lie.

As a rule of thumb, this module includes:

  • An enum which defines a parameter for the primitive (e.g. mode of operation, elliptic curve). This makes it easy to document (with the autoenum directive) and makes it clear which options are implemented. Also Typer uses enums to provide auto-completion.

  • A function that loads test vectors, usually based on that enum and the JSON file mentioned above.

  • One or more Protocol classes that describe the function signature that the implementation must have in order to be tested.

  • A test function for each operation that is supported, which runs with test vectors or user input files.

  • A function that runs a Python wrapper (or more if other languages are supported).

Additionally, internal implementations or wrappers of third-party implementations are considered private. The convention in Python is that the function name should start with an underscore. To improve its privacy, we do not include this function in the module’s __dir__() (see below). Python does not have a way of enforcing this “privacy”, users can still access these functions if they know they exist, but the idea is to convey the message that these are not meant to be used anywhere else, no guarantees are made.

A side-note on imports#

Currently the primitive modules are structured to be imported and used “directly”. For example:

from crypto_condor.primitives import AES

AES.test(...)

We use __dir__ to declare the public API, as it limits what is returned when using an IDE’s or interpreter’s auto-completion. This allows to remove names such as logging as well as avoid exposing functions meant to be only used internally, like our wrapper of the primitives.

__dir__ returns a list of strings. Objects like type aliases have to be referenced by name directly (e.g. "CiphertextAndTag"), while most other objects can be referenced by their __name__ attribute (e.g. verify.__name__). The advantage of the latter is that renaming the function/class/etc. using an IDE will change this reference automatically.

CLI commands#

Once this work on the primitive is done, add the integration to the CLI. This should mostly consist in adding a function for the primitive under the corresponding command, which parses the inputs with typer.Argument and typer.Option, and passes them to the corresponding function e.g. AES.verify(...).

When the corresponding functions are implemented, add a new entry to the SUPPORTED_MODES dictionary in constants.py and the necessary tests.

A few aspects to consider:

  • When adding wrappers, the tool checks that the get-wrapper command is supported for the given primitive, and then looks for a directory under resources/wrappers. This directory must be named as the primitive, in lower-case. Inside it the wrappers are organized by language, each with their own subdirectory named in lower-case. Examples are in subdirectories named <language>-example. Each example has its own sub-subdirectory inside it. These sub-subdirectories are numbered by an increasing counter that starts at 1.

  • Guides are first written for the documentation then copied with the utils/copy_guides.py script. The name matches the one for the documentation, namely the primitive name in upper-case.

Adding a new harness#

crypto-condor can test functions exposed by a shared library, similar to a fuzzing hook. To do so, the functions must follow the conventions described by the harness API. Internally, this means adding a test_lib function to the corresponding primitive. This function has a particular signature:

def test_lib(ffi: cffi.FFI, lib, functions: list[str]) -> ResultsDict:
    ...

Where:

  • ffi is the cffi.FFI instance.

  • lib is the library dlopen’d with ffi.

  • functions is a list of function names to test, which should correspond to the primitive called.

Each primitive is in charge of calling ffi.cdef() to define the signature of the exposed function, and to wrap it and test it. The crypto_condor.harness.test_harness function is in charge of determining the available functions, importing the corresponding primitives, and passing the list of function that each primitive should test.

The documentation for this mode can be found in docs/source/harness-api.

Documenting a new primitive#

The documentation can be found under docs/source. There, it is divided in several directories which correspond to different pages in the HTML render. As indicated above, most documents can be written in Markdown, but those that make use of autodoc must be written in rST as autodoc doesn’t support Markdown.

Building the documentation#

The packages required to build the documentation can be installed with poetry install --with=docs. Then you can either use make docs which builds the docs to docs/build/html or use make livedocs with uses sphinx-autobuild to build the docs, watch for changes, and reload open tabs after rebuilding changes. Both options ensure that the dependencies are installed before building.

For publishing, the docs are automatically built by the CI. It uses the pages-ci target which calls the all-versions target of docs/Makefile is used. This target uses a hard-coded list of Git refs (tags or branches), checks out each ref and builds its corresponding documentation under docs/build/public/[ref]. Then the pages-ci targets moves the resulting docs to the correct directory used by GitLab Pages.

Versioning#

As indicated in the README, this project currently adheres to CalVer. This version is shown in various parts of the project (--version option, the documentation, the git tags, etc.). For each release, the version must be updated in both the git tag and pyproject.toml, otherwise the CI pipeline will fail the publish step.

To avoid pushing a tagged version with an out-of-date pyproject.toml or vice versa, you can add a pre-push hook that runs the utils/check_tag_and_version.py script. Create .git/hooks/pre-push with the following content:

current=$(git branch --show-current)
if test "$current" = "main"
then
    .venv/bin/python utils/check_tag_and_version.py
fi

This checks that the hook only runs on the main branch, as others should not be tagged. It also assumes that we are using a virtual environment to run and test the tool, and said venv is inside the .venv directory.

Note: when using poetry, it might be necessary to run poetry install to refresh the package version, otherwise the hook will fail.

Contributing to CONTRIBUTING#

Modifications to CONTRIBUTING must be done to the version found in docs/source/development/CONTRIBUTING, as the one found in the root of the repo is a copy of that version (see the root Makefile’s copy-contributing target).