Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
8 contributors

Users who have contributed to this file

@ajdavis @bjori @kevinAlbs @chergert @TylerBrock @rcsanchez97 @derickr @fionaRowan
290 lines (199 sloc) 10.1 KB

Contributing to mongo-c-driver

Thanks for considering contributing to the mongo-c-driver!

This document intends to be a short guide to helping you contribute to the codebase. It expects a familiarity with the C programming language and writing portable software. Whenever in doubt, feel free to ask others that have contributed or look at the existing body of code.

Guidelines

The mongo-c-driver has a few guidelines that help direct the process.

Portability

mongo-c-driver is portable software. It needs to run on a multitude of operating systems and architectures.

  • Linux (RHEL 5 and newer)
  • FreeBSD (10 and newer)
  • Windows (Vista and newer)
  • macOS (10.8 and newer)
  • ARM/SPARC/x86/x86_64

Licensing

Some of the mongo-c-driver users embed the library statically in their products. Therefore, the driver and all contributions must be liberally licensed. As a policy, we have chosen Apache 2.0 as the license for the project.

Coding Style

We try not to be pedantic with taking contributions that are not properly formatted, but we will likely perform a followup commit that cleans things up. The basics are, in vim:

 : set ts=3 sw=3 et

3 space tabs, insert spaces instead of tabs.

For all the gory details, see .clang-format

Commit Style

Commit messages follow this pattern:

CDRIVER-#### lowercase commit message

There's no colon after the ticket number, and no period at the end of the subject line. If there's no related Jira ticket, then the message is simply all lowercase:

typos in NEWS

We follow the widely adopted "50/72" rule: the subject line is ideally only 50 characters, but definitely only 72 characters. This requires some thoughtful writing but it's worthwhile when you're scrolling through commits. The message body can be as large as you want, wrapped at 72 columns.

Adding a new error code or domain

When adding a new error code or domain, you must do the following. This is most applicable if you are adding a new symbol with a bson_error_t as a parameter, and the existing codes or domains are inappropriate.

  • Add the domain to mongoc_error_domain_t in src/mongoc/mongoc-error.h
  • Add the code to mongoc_error_code_t in src/mongoc/mongoc-error.h
  • Add documentation for the domain or code to the table in doc/mongoc_errors.rst

Adding a new symbol

This should be done rarely but there are several things that you need to do when adding a new symbol.

  • Add documentation for the new symbol in doc/mongoc_your_new_symbol_name.rst

Documentation

We strive to document all symbols. See doc/ for documentation examples. If you add a new public function, add a new .rst file describing the function so that we can generate man pages and HTML for it.

For complex internal functions, comment above the function definition with a block comment like the following:

/*--------------------------------------------------------------------------
 *
 * mongoc_cmd_parts_append_read_write --
 *
 *       Append user-supplied options to @parts->command_extra, taking the
 *       selected server's max wire version into account.
 *
 * Return:
 *       True if the options were successfully applied. If any options are
 *       invalid, returns false and fills out @error. In that case @parts is
 *       invalid and must not be used.
 *
 * Side effects:
 *       May partly apply options before returning an error.
 *
 *--------------------------------------------------------------------------
 */

Public functions do not need these comment blocks, since they are documented in the .rst files.

Testing

To run the entire test suite, including authentication tests, start mongod with auth enabled:

$ mongod --auth

In another terminal, use the mongo shell to create a user:

$ mongo --eval "db.createUser({user: 'admin', pwd: 'pass', roles: ['root']})" admin

Authentication in MongoDB 3.0 and later uses SCRAM-SHA-1, which in turn requires a driver built with SSL.

Set the user and password environment variables, then build and run the tests:

$ export MONGOC_TEST_USER=admin
$ export MONGOC_TEST_PASSWORD=pass
$ ./test-libmongoc

Additional environment variables:

  • MONGOC_TEST_HOST: default localhost, the host running MongoDB.
  • MONGOC_TEST_PORT: default 27017, MongoDB's listening port.
  • MONGOC_TEST_URI: override both host and port with a full connection string, like "mongodb://server1,server2".
  • MONGOC_TEST_SERVER_LOG: set to stdout or stderr for wire protocol logging from tests that use mock_server_t. Set to json to include these logs in the test framework's JSON output, in a format compatible with Evergreen.
  • MONGOC_TEST_MONITORING_VERBOSE: set to on for verbose output from Application Performance Monitoring tests.
  • MONGOC_TEST_COMPRESSORS=snappy,zlib: wire protocol compressors to use

If you start mongod with SSL, set these variables to configure how test-libmongoc connects to it:

  • MONGOC_TEST_SSL: set to on to connect to the server with SSL.
  • MONGOC_TEST_SSL_PEM_FILE: path to a client PEM file.
  • MONGOC_TEST_SSL_PEM_PWD: the PEM file's password.
  • MONGOC_TEST_SSL_CA_FILE: path to a certificate authority file.
  • MONGOC_TEST_SSL_CA_DIR: path to a certificate authority directory.
  • MONGOC_TEST_SSL_CRL_FILE: path to a certificate revocation list.
  • MONGOC_TEST_SSL_WEAK_CERT_VALIDATION: set to on to relax the client's validation of the server's certificate.

The SASL / GSSAPI / Kerberos tests are skipped by default. To run them, set up a separate mongod with Kerberos and set its host and Kerberos principal name as environment variables:

  • MONGOC_TEST_GSSAPI_HOST
  • MONGOC_TEST_GSSAPI_USER

URI-escape the username, for example write "user@realm" as "user%40realm". The user must be authorized to query kerberos.test.

MongoDB 3.2 adds support for readConcern, but does not enable support for read concern majority by default. mongod must be launched using --enableMajorityReadConcern. The test framework does not (and can't) automatically discover if this option was provided to MongoDB, so an additional variable must be set to enable these tests:

  • MONGOC_ENABLE_MAJORITY_READ_CONCERN

Set this environment variable to on if MongoDB has enabled majority read concern.

Some tests require Internet access, e.g. to check the error message when failing to open a MongoDB connection to example.com. Skip them with:

  • MONGOC_TEST_OFFLINE=on

Some tests require a running MongoDB server. Skip them with:

  • MONGOC_TEST_SKIP_LIVE=on

For quick checks during development, disable long-running tests:

  • MONGOC_TEST_SKIP_SLOW=on

Some tests run against a local mock server, these can be skipped with:

  • MONGOC_TEST_SKIP_MOCK=on

If you have started with MongoDB with --ipv6, you can test IPv6 with:

  • MONGOC_CHECK_IPV6=on

The tests for mongodb+srv:// connection strings require some setup, see the Initial DNS Seedlist Discovery Spec. By default these connection strings are NOT tested, enable them with:

  • MONGOC_TEST_DNS=on

The mock server timeout threshold for future functions can be set with:

  • MONGOC_TEST_FUTURE_TIMEOUT_MS=<int>

This is useful for debugging, so future calls don't timeout when stepping through code.

Tests of Client-Side Field Level Encryption require credentials to an AWS user which has list and read permissions to AWS KMS.

  • MONGOC_TEST_AWS_SECRET_ACCESS_KEY=<string>
  • MONGOC_TEST_AWS_ACCESS_KEY_ID=<string>

Tests of Client-Side Field Level Encryption spawn an extra process, "mongocryptd", by default. To bypass this spawning, start mongocryptd on port 27020 and set the following:

  • MONGOC_TEST_MONGOCRYPTD_BYPASS_SPAWN=on

All tests should pass before submitting a patch.

Configuring the test runner

The test runner can be configured with command-line options. Run test-libmongoc --help for details.

To run just a specific portion of the test suite use the -l option like so:

$ ./test-libmongoc -l "/server_selection/*"

The full list of tests is shown in the help.

Creating and checking a distribution tarball

The make distcheck command can be used to confirm that any modifications are able to be packaged into the distribution tarball and that the resulting distribution tarball can be used to successfully build the project.

A failure of the make distcheck target is an indicator of an oversight in the modification to the project. For example, if a new source file is added to the project but it is not added to the proper distribution list, it is possible that the distribution tarball will be created without that file. An attempt to build the project without the file is likely to fail.

When make distcheck is invoked, several things happen. The dist target is executed to create a distribution tarball. Then the tarball is unpacked, configured (with an invocation of cmake), built (by calling make), installed (by calling make install), and tested (by calling make check). Three environment variables can be used to modify these steps.

To adjust the options passed to make during the build step, set:

  • DISTCHECK_BUILD_OPTS

If this variable is not set, then make is called with a default of "-j 8".

To adjust the options passed to make install during the installation step, set:

  • DISTCHECK_INSTALL_OPTS

To adjust the options passed to make check during the test step, set:

  • DISTCHECK_CHECK_OPTS

Remember, if you want to modify the top-level make invocation, you will need to pass options on the command line as normal.

For example, the command make -j 6 distcheck DISTCHECK_BUILD_OPTS="-j 4" will call the standard sequence of targets depended upon by distcheck with a parallelism level of 6, while the build step that is later called by the distcheck target will be executed with a parallelism level of 4.

You can’t perform that action at this time.