Submitting a bug report

Please report any bugs to the GitLab (internal) or GitHub (public) issue trackers. Your issue will be resolved more quickly if you are able to provide a minimal working example, including an example concrete data file if applicable.

Contributing code

  1. Ensure an issue has been created for your new feature/bugfix on GitLab (internal) or GitHub (public).

  2. If you are adding a new feature, create a stub (placeholder implementation) of the desired argument/function/class/etc.

  3. Write a test for your new feature/bugfix and run it, ensuring that it fails on the current implementation:

    py.test tests/

    NameErrors, ImportErrors, SyntaxErrors, etc. do not count (they indicate the API is wrong).

  4. Implement your new feature/bugfix.

  5. Run the test again, ensuring that it now passes.

  6. Run all tests and style checks, ensuring that they pass:


    Optionally, run integration tests (you must have Redis server version 2.8 or later in your path; do redis-server --version to check):

    tox integration-tests
  7. If you created a new module (file) or package (directory) in the library, please see “Adding new modules and packages” in the next section.

  8. Push your changes to a feature branch on GitLab/GitHub (e.g., called n-issue-abbrev where n is the issue number and issue-abbrev is a very short abbreviation of the issue title) and ensure that the build passes. The build is defined in .gitlab-ci.yml (.travis.yml and appveyor.yml for public builds); tox is configured in tox.ini. The build includes unit tests, integration tests, and style checks and runs on Python 2.7 and 3.5 across multiple platforms; if it fails, please find the error in the build log, fix it, and try again.

  9. Add a line to CHANGELOG under the current version-in-progress describing your changes simply and concisely. Add yourself to AUTHORS if you are not already listed.

  10. If you’ve made multiple commits, please squash them and git push -f to the feature branch.

  11. Create a merge/pull request for your feature branch into master, referencing the GitLab/GitHub issue.

For maintainers

Adding new modules and packages

If a new module or package is created, either by hand or in the auto-generated code from Thrift, a small amount of additional configuration must be performed.

In either case, the name of the package (if it is a package and not a module) should be added to the packages parameter in The name of the package or module should be added to the subpackage or submodule list in docs/concrete.rst, respectively. A new ReStructuredText file should also be created under docs/ for the package or module; follow the conventions set by the other packages and modules.

If the new module or package was written by hand, a guard should be added to autodoc_process_docstring in docs/ so that that module or package is not ignored by the documentation parser. If it is a package, a guard should also be added to generate.bash so that the corresponding directory is not deleted when the auto-generated code is copied into concrete/ from the Thrift build directory.

If a new package was generated by Thrift, a corresponding exclude should be added to the flake8 configuration in setup.cfg and the new package’s ttypes module should be added to the star imports in concrete/ If a new module (not package) was generated by thrift, no action is necessary.

Branches, versions, and releases

The master branch is kept stable at all times. Before a commit is pushed to master, it should be checked by CI on another branch. The recommended way of maintaining this is to do all work in feature branches that are kept up-to-date with master and pushed to GitLab, waiting for CI to finish before merging.

We use zest.releaser to manage versions, the CHANGELOG, and releases. (Making a new release is a many-step process that requires great care; doing so by hand is strongly discouraged.) Using zest.releaser, stable versions are released to PyPI and master is kept on a development version number (so that a stable version number never represents more than one snapshot of the code). To make a new release install zest.releaser (pip install zest.releaser) and run fullrelease.

Testing PyPI releases

To test how changes to concrete-python will show up on PyPI (for example, how the readme is rendered) you can use the PyPI testing site. To do so, set the following in ~/.pypirc:

repository =

You will also need to create a testpypi user account and you may need to request access to the concrete package on testpypi.

Testing documentation

The automated build checks for syntax errors in the documentation. When a push is made to the GitHub repository the online documentation is automatically re-generated. You can run the automatic validation and generate the HTML documentation locally by doing:

tox -e docs

The generated HTML documentation is stored it in .tox/docs/tmp/html (relative to the top of your repository). Open this file path in a web browser to check how your changes will look when published online.

(Re)generating code from concrete

The Python code generated by the thrift compiler on the schema defined in the concrete project is checked in to concrete-python manually after applying necessary patches. For trivial modifications to the schema this process is automated by generate.bash, which assumes concrete has been cloned alongside concrete-python (in the same parent directory):

bash generate.bash

After this succeeds, tests should be run and the changes should be manually inspected (git diff) for sanity. Note that this will not delete previously-generated files that are no longer produced by thrift (whose entries were removed from the schema).

Note: Often generate.bash is not sufficient: the patches (in patches/) document where it (thrift) falls short on the previously-compiled schema. Additionally, if new packages (namespaces) are added to the schema, they must be added to, setup.cfg, and concrete/

If generate.bash throws an error, the necessary changes should be performed manually and checked in to the index, at which point the generated code should be removed from the working tree, raw (unpatched) generated code should be generated, and new patches should be produced and stored in patches/ using git diff. See the arguments to generate.bash for generating the unpatched code.