Developer Guide

Girder is a platform-centric web application whose client and server are very loosely coupled. As such, development of Girder can be divided into the server (a CherryPy-based Python module) and the primary client (a Backbone-based) web client. This section is intended to get prospective contributors to understand the tools used to develop Girder.

Configuring Your Development Environment

Next, you should install the Python development dependencies with pip, to provide helpful development tools and to allow the test suite to run:

pip install -r requirements-dev.txt

Install front-end web client development dependencies and build the web client code:

girder build --dev

For more options for building the web client, run:

girder build --help

Finally, you’ll want to set your server into development mode. Add the following entry into your local config file (see Configuration for instructions):

[server]
mode="development"

Vagrant

A shortcut to going through the development environment configuration steps is to use Vagrant to setup the environment on a VirtualBox virtual machine. For more documentation on how to set this up, see Developer Installation

During Development

Once Girder is started via girder serve, the server will reload itself whenever a Python file is modified.

If you are doing front-end development, it’s much faster to use a watch process to perform automatic fast rebuilds of your code whenever you make changes to source files.

If you are making changes to Girder’s core web client, run the following watch command:

girder build --watch

If you are developing a web client of a plugin, run:

girder build --watch-plugin your_plugin_name

With watch option, sourcemaps will be generated, which helps debugging front-end code in browser. When you want to end the watch process, press Ctrl+C (or however you would normally terminate a process in your terminal).

Girder Shell

To test various functionality in a typical REPL (Python, IPython, etc) some bootstrapping is required to configure the Girder server. This sets up an “embedded” server, meaning no TCP ports are actually bound but requests can still be performed via Python. Bootstrapping the server involves running girder.utility.server.configureServer with the plugins to be enabled.

Girder provides a utility script for entering into a shell with the server preconfigured. Once Girder is installed the script can be run using girder shell which optionally takes a comma separated list of plugins to enable.

Utilities

Girder has a set of utility modules and classes that provide handy extractions for certain functionality. Detailed API documentation can be found here.

Configuration Loading

The Girder configuration loader allows for lazy-loading of configuration values in a CherryPy-agnostic manner. The recommended idiom for getting the config object is:

from girder.utility import config
cur_config = config.getConfig()

There is a configuration file for Girder located in girder/conf. The file girder.dist.cfg is the file distributed with the repository and containing the default configuration values. This file should not be edited when deploying Girder. Rather, create a custom girder.cfg file and place it in one of the supported locations (see Configuration). You only need to edit the values in the file that you wish to change from their default values; the system loads the dist file first, then the custom file, so your local settings will override the defaults.

Client Development

If you are writing a custom client application that communicates with the Girder REST API, you should look at the Swagger page that describes all of the available API endpoints. The Swagger page can be accessed by navigating a web browser to /api/v1. If you wish to consume the Swagger-compliant API specification programmatically, the JSON listing is served out of /api/v1/describe.

If you are working on the main Girder web client, either in core or extending it via plugins, there are a few conventions that should be followed. Namely, if you write code that instantiates new View descendant objects, you should pass a parentView property when constructing it. This will allow the child view to be cleaned up recursively when the parent view is destroyed. If you forget to set the parentView property when constructing the view, the view will still work as expected, but a warning message will appear in the console to remind you. Example:

import View from '@girder/core/views/View';

MySubView = View.extend({
   ...
});

new MySubView({
    el: ...,
    otherProperty: ...,
    parentView: this
});

If you use View in custom Backbone apps and need to create a new root view object, set the parentView to null. If you are using a Girder widget in a custom app that does not use the View as the base object for its views, you should pass parentView: null and make sure to call destroy() on the view manually when it should be cleaned up.

Server Side Testing

Most of Girder’s server tests are run via tox, which provides virtual environment isolation and automatic dependency installation for test environments. The tox Python package must be installed:

pip install tox

To run the basic test suite, ensure that a MongoDB instance is ready on localhost:27017, then run:

tox

To destroy and recreate all virtual environments used for testing, pass the -r flag to tox.

Static Analysis Tests

Girder’s static analysis (linting) tests are fast to execute, run on all code, and don’t require a running MongoDB. It’s recommended to run them locally before any Python code changes are committed. To execute them, run:

tox -e lint

pytest Tests

Girder’s modern automated tests are written with pytest. To execute them, ensure MongoDB is ready, then run:

tox -e pytest

Specific arguments can be passed through tox to pytest by adding them after a --.

For example, pytest uses -k to filter tests; to run only the testLoadModelDecorator test, run:

tox -e pytest -- -k testLoadModelDecorator

Legacy unittest Tests

Girder’s legacy automated tests are written with Python’s unittest framework and executed with CMake. All new tests should be written with pytest, but many important test cases are still covered only by unitest.

Note

Unless debugging code that is already coverered by a legacy test case, it may be more convenient to allow these tests to be run by Girder’s CI envionment, instead of configuring them locally.

To initialize the legacy tests, from the root girder repo, run:

mkdir ../girder-build
cd ../girder-build
cmake ../girder
make

You only need to do this once. From then on, whenever you want to run the tests, run:

cd girder-build
ctest

There are many ways to filter tests when running CTest or run the tests in parallel. For example, this command will run tests with name matches regex server_user with verbose output:

ctest -V -R server_user

Client Side Testing

Static Analysis Tests

To run static analysis tests on client side code, run from the top-level Girder directory:

npm i
npm run lint

Running the Tests with CTest

Using the same setup as above for the Server Side Tests, your environment will be set up. The client side tests and server side tests are both harnessed with CTest, so use the following commands to run both

cd girder-build
ctest

will run all of the tests, which include the client side tests. Our client tests use the Jasmine JS testing framework.

If you encounter errors, there is a chance you missed certain steps for setting up development dependencies. You could use ccmake to change CMake configuration. Or, it might be easier to recreate the environment from the beginning.

When running client side tests, if you try to SIGINT (ctrl+c) the CTest process, CTest won’t pass that signal down to the test processes for them to handle. This can result in orphaned python unittest processes and can prevent future runs of client tests. If you run a client side test and see an error message similar to IOError: Port 30015 not free on '0.0.0.0', then look for an existing process similar to /usr/bin/python3 -m unittest -v tests.web_client_test, kill the process, and then try your tests again.

Adding a New Client Side Test

To add a new client side test, add a new spec file in /girder/web_client/test/spec/, add a line referencing your spec file to /girder/tests/CMakeLists.txt using the add_web_client_test function, and then run in your build directory

cmake ../girder

before running your tests.

An example of a very simple client side test would be as follows

add_web_client_test(some_client_test "someSpec.js" PLUGIN "my_plugin")

The PLUGIN argument indicates that “my_plugin” is the owner of some_client_test, at the time of the test my_plugin and all of its dependencies will be loaded.

If additional plugins are needed for a specific test, that can be achieved using the ENABLEDPLUGINS argument

add_web_client_test(another_client_test "anotherSpec.js" PLUGIN "my_plugin" ENABLEDPLUGINS "my_plugin" "jobs")

Here ENABLEDPLUGINS ensures that my_plugin and the jobs plugin are loaded, along with their dependencies at the time of another_client_test.

Note

Core functionality shouldn’t depend on plugins being enabled, this test definition is more suitable for a plugin. Information for testing plugins can be found under Plugin Development.

You will find many useful methods for client side testing in the girderTest object defined at /girder/web_client/test/testUtils.js.

Test Coverage Reporting

When Girder’s full test suite is run in the CI environment, a test coverage report for both server and client code is generated and uploaded to Codecov. This may be viewed online at any time.

Initializing the Database for a Test

Note

This functionality has not yet been ported to our pytest tests.

When running tests in Girder, the database will initially be empty. Often times, you want to be able to start the test with the database in a particular state. To avoid repetitive initialization code, Girder provides a way to import a folder hierarchy from the file system using a simple initialization file. This file is in YAML (or JSON) format and provides a list of objects to insert into the database before executing your test. A typical example of this format is as follows

---
users:
  - login: 'admin'
    password: 'password'
    firstName: 'First'
    lastName: 'Last'
    email: 'admin@girder.test'
    admin: true
    import: 'files/user'

collections:
  - name: 'My collection'
    public: true
    creator: 'admin'
    import: 'files/collection'

This will create one admin user and a public collection owned by that user. Both the generated user and collection objects will contain folders imported from the file system. Relative paths provided by the import key will be resolved relative to the location of the YAML file on disk. You can also describe the full hierarchy in the YAML file itself for more complicated use cases. See the test spec in tests/cases/setup_database_test.yml for a more complete example.

Note

When importing from a local path into a user or collection, files directly under that path are ignored because items can be only inserted under folders.

To use the initialization mechanism, you should add the YAML file next to your test file. For example, if your test is defined in tests/cases/my_test.py, then the initialization spec should go in tests/cases/my_test.yml. This file will be automatically detected and loaded before executing your test code. This is true for both python and javascript tests added in core or inside plugins.

The python module setup_database.py that generates the database can also be run standalone to help in development. To use it, you should point girder to an empty database

GIRDER_MONGO_URI='mongodb://127.0.0.1:27017/mytest' python tests/setup_database.py tests/test_database/spec.yml

You can browse the result in Girder by running

GIRDER_MONGO_URI='mongodb://127.0.0.1:27017/mytest' girder serve

Note

The setup_database module is meant to provision fixures for tests only. If you want to provision a Girder instance for deployment, see the Girder ansible client.

Code Review

Contributions to Girder are done via pull requests with a core developer approving the PR with GitHub review system. At this point, the topic branch can be merged to master. This is meant to be a simple, low-friction process; however, code review is very important. It should be done carefully and not taken lightly. Thorough code review is a crucial part of developing quality software. When performing a code review, ask the following:

  1. Is the continuous integration server happy with this?
  2. Are there tests for this feature or bug fix?
  3. Is this documented (for users and/or developers)?
  4. Are the commits modular with good notes?
  5. Will this merge cleanly?
  6. Does this break backward compatibility? Is that okay?
  7. What are the security implications of this change? Does this open Girder up to any vulnerabilities (XSS, CSRF, DB Injection, etc)?

Third-Party Libraries

Girder’s standard procedure is to use a tool like piprot to check for out-of-date third-party library requirements on a quarterly basis (typically near the dates of the solstices and equinoxes). Library packages should generally be upgraded to the latest released version, except when:

  • Doing so would introduce any new unfixable bugs or regressions.
  • Other closely-affiliated projects (e.g. Romanesco, Minerva) use the same library and the other project cannot also feasibly be upgraded simultaneously.
  • The library has undergone a major API change, and development resources do not permit updating Girder accordingly or Girder exposes parts of the library as members of Girder’s API surface (e.g. CherryPy) and upgrading would cause incompatible API changes to be exposed. In this case, the library should still be upgraded to the highest non-breaking version that is available at the time.

Note

In the event that a security vulnerability is discovered in a third-party library used by Girder, the library must be upgraded to patch the vulnerability immediately and without regard to the aforementioned exceptions. However, attempts should still be made to maintain API compatibility via monkey patching, wrapper classes, etc.

Modifying core web client libraries

Web client libraries in Girder core are managed via npm. When a new npm package is required, or an existing package is upgraded, the following should be done:

  1. Ensure that you are using a development environment with version >=5.6 of npm installed:

    npm install -g 'npm@>=5.6'
    
  2. Update girder/web_client/package.json.template or girder/web_client/src/package.json to add a new abstract specifier for the package:

  • Packages that are bundled into the web client must be listed under the dependencies field of girder/web_client/src/package.json and should generally use the tilde range to specify versions.
  • Packages that are part of the build or testing process should be listed under either the dependencies or devDependencies fields of girder/web_client/package.json.template and should generally use the caret range to specify versions.

If updating npm libraries related to linting or documentation, you should instead modify the top-level package.json file, run npm update, then commit the modified files.

Creating a new release

Girder releases are uploaded to PyPI for easy installation via pip. Each time a pull request is merged to master, an incremental “dev” release is created during CI as a pre-release package and published to PyPI, making it easy for downstreams to install bleeding edge packages without needing to clone the Girder repository.

The major, minor, and patch version are inferred automatically using setuptools-scm based on the latest git tag. Hence, creating a new release is as simple as pushing a new git tag. For example, from the target commit, you could simply run:

git tag v4.5.6
git push --tags

That will trigger CircleCI to run, and if all tests pass, the 4.5.6 release will be uploaded to PyPI.