13 KiB
Contributing to Catch2
Contents
Using Git(Hub)
Testing your changes
Writing documentation
Writing code
CoC
So you want to contribute something to Catch2? That's great! Whether it's a bug fix, a new feature, support for additional compilers - or just a fix to the documentation - all contributions are very welcome and very much appreciated. Of course so are bug reports, other comments, and questions, but generally it is a better idea to ask questions in our Discord, than in the issue tracker.
This page covers some guidelines and helpful tips for contributing to the codebase itself.
Using Git(Hub)
Ongoing development happens in the devel
branch for Catch2 v3, and in
v2.x
for maintenance updates to the v2 versions.
Commits should be small and atomic. A commit is atomic when, after it is applied, the codebase, tests and all, still works as expected. Small commits are also preferred, as they make later operations with git history, whether it is bisecting, reverting, or something else, easier.
When submitting a pull request please do not include changes to the amalgamated distribution files. This means do not include them in your git commits!
When addressing review comments in a MR, please do not rebase/squash the
commits immediately. Doing so makes it harder to review the new changes,
slowing down the process of merging a MR. Instead, when addressing review
comments, you should append new commits to the branch and only squash
them into other commits when the MR is ready to be merged. We recommend
creating new commits with git commit --fixup
(or --squash
) and then
later squashing them with git rebase --autosquash
to make things easier.
Testing your changes
Note: Running Catch2's tests requires Python3
Catch2 has multiple layers of tests that are then run as part of our CI.
The most obvious one are the unit tests compiled into the SelfTest
binary. These are then used in "Approval tests", which run (almost) all
tests from SelfTest
through a specific reporter and then compare the
generated output with a known good output ("Baseline"). By default, new
tests should be placed here.
However, not all tests can be written as plain unit tests. For example, checking that Catch2 orders tests randomly when asked to, and that this random ordering is subset-invariant, is better done as an integration test using an external check script. Catch2 integration tests are written using CTest, either as a direct command invocation + pass/fail regex, or by delegating the check to a Python script.
Catch2 is slowly gaining more and more types of tests, currently Catch2 project also has buildable examples, "ExtraTests", and CMake config tests. Examples present a small and self-contained snippets of code that use Catch2's facilities for specific purpose. Currently they are assumed passing if they compile.
ExtraTests then are expensive tests, that we do not want to run all the time. This can be either because they take a long time to run, or because they take a long time to compile, e.g. because they test compile time configuration and require separate compilation.
Finally, CMake config tests test that you set Catch2's compile-time configuration options through CMake, using CMake options of the same name.
None of these tests are enabled by default. To enable them, add
-DCATCH_BUILD_EXAMPLES=ON
, -DCATCH_BUILD_EXTRA_TESTS=ON
, and
-DCATCH_ENABLE_CONFIGURE_TESTS=ON
when configuration the CMake build.
Bringing this all together, the steps below should configure, build,
and run all tests in the Debug
compilation.
# 1. Regenerate the amalgamated distribution
./tools/scripts/generateAmalgamatedFiles.py
# 2. Configure the full test build
cmake -Bdebug-build -H. -DCMAKE_BUILD_TYPE=Debug -DCATCH_DEVELOPMENT_BUILD=ON -DCATCH_BUILD_EXAMPLES=ON -DCATCH_BUILD_EXTRA_TESTS=ON
# 3. Run the actual build
cmake --build debug-build
# 4. Run the tests using CTest
cd debug-build
ctest -j 4 --output-on-failure -C Debug
For convenience, the above commands are in the script tools/scripts/buildAndTest.sh
, and can be run like this:
cd Catch2
./tools/scripts/buildAndTest.sh
A Windows version of the script is available at tools\scripts\buildAndTest.cmd
.
If you added new tests, you will likely see ApprovalTests
failure.
After you check that the output difference is expected, you should
run tools/scripts/approve.py
to confirm them, and include these changes
in your commit.
Writing documentation
If you have added new feature to Catch2, it needs documentation, so that other people can use it as well. This section collects some technical information that you will need for updating Catch2's documentation, and possibly some generic advise as well.
Technicalities
First, the technicalities:
- If you have introduced a new document, there is a simple template you should use. It provides you with the top anchor mentioned to link to (more below), and also with a backlink to the top of the documentation:
<a id="top"></a>
# Cool feature
> [Introduced](https://github.com/catchorg/Catch2/pull/123456) in Catch2 X.Y.Z
Text that explains how to use the cool feature.
---
[Home](Readme.md#top)
-
Crosslinks to different pages should target the
top
anchor, like this[link to contributing](contributing.md#top)
. -
We introduced version tags to the documentation, which show users in which version a specific feature was introduced. This means that newly written documentation should be tagged with a placeholder, that will be replaced with the actual version upon release. There are 2 styles of placeholders used through the documentation, you should pick one that fits your text better (if in doubt, take a look at the existing version tags for other features).
> [Introduced](link-to-issue-or-PR) in Catch2 X.Y.Z
- this placeholder is usually used after a section heading> X (Y and Z) was [introduced](link-to-issue-or-PR) in Catch2 X.Y.Z
- this placeholder is used when you need to tag a subpart of something, e.g. a list
-
For pages with more than 4 subheadings, we provide a table of contents (ToC) at the top of the page. Because GitHub markdown does not support automatic generation of ToC, it has to be handled semi-manually. Thus, if you've added a new subheading to some page, you should add it to the ToC. This can be done either manually, or by running the
updateDocumentToC.py
script in thescripts/
folder.
Contents
Now, for some content tips:
-
Usage examples are good. However, having large code snippets inline can make the documentation less readable, and so the inline snippets should be kept reasonably short. To provide more complex compilable examples, consider adding new .cpp file to
examples/
. -
Don't be afraid to introduce new pages. The current documentation tends towards long pages, but a lot of that is caused by legacy, and we know that some of the pages are overly big and unfocused.
-
When adding information to an existing page, please try to keep your formatting, style and changes consistent with the rest of the page.
-
Any documentation has multiple different audiences, that desire different information from the text. The 3 basic user-types to try and cover are:
- A beginner to Catch2, who requires closer guidance for the usage of Catch2.
- Advanced user of Catch2, who want to customize their usage.
- Experts, looking for full reference of Catch2's capabilities.
Writing code
If want to contribute code, this section contains some simple rules and tips on things like code formatting, code constructions to avoid, and so on.
C++ standard version
Catch2 currently targets C++14 as the minimum supported C++ version. Features from higher language versions should be used only sparingly, when the benefits from using them outweigh the maintenance overhead.
Example of good use of polyfilling features is our use of conjunction
,
where if available we use std::conjunction
and otherwise provide our
own implementation. The reason it is good is that the surface area for
maintenance is quite small, and std::conjunction
can directly use
compiler built-ins, thus providing significant compilation benefits.
Example of bad use of polyfilling features would be to keep around two
sets of metaprogramming in the stringification implementation, once
using C++14 compliant TMP and once using C++17's if constexpr
. While
the C++17 would provide significant compilation speedups, the maintenance
cost would be too high.
Formatting
To make code formatting simpler for the contributors, Catch2 provides
its own config for clang-format
. However, because it is currently
impossible to replicate existing Catch2's formatting in clang-format,
using it to reformat a whole file would cause massive diffs. To keep
the size of your diffs reasonable, you should only use clang-format
on the newly changed code.
Code constructs to watch out for
This section is a (sadly incomplete) listing of various constructs that are problematic and are not always caught by our CI infrastructure.
Naked exceptions and exceptions-related function
If you are throwing an exception, it should be done via CATCH_ERROR
or CATCH_RUNTIME_ERROR
in internal/catch_enforce.hpp
. These macros will handle
the differences between compilation with or without exceptions for you.
However, some platforms (IAR) also have problems with exceptions-related
functions, such as std::current_exceptions
. We do not have IAR in our
CI, but luckily there should not be too many reasons to use these.
However, if you do, they should be kept behind a
CATCH_CONFIG_DISABLE_EXCEPTIONS
macro.
Avoid std::move
and std::forward
std::move
and std::forward
provide nice semantic name for a specific
static_cast
. However, being function templates they have surprisingly
high cost during compilation, and can also have a negative performance
impact for low-optimization builds.
You should be using CATCH_MOVE
and CATCH_FORWARD
macros from
internal/catch_move_and_forward.hpp
instead. They expand into the proper
static_cast
, and avoid the overhead of std::move
and std::forward
.
Unqualified usage of functions from C's stdlib
If you are using a function from C's stdlib, please include the header
as <cfoo>
and call the function qualified. The common knowledge that
there is no difference is wrong, QNX and VxWorks won't compile if you
include the header as <cfoo>
and call the function unqualified.
User-Defined Literals (UDL) for Catch2' types
Due to messy standardese and ... not great ... implementation of
-Wreserved-identifier
in Clang, avoid declaring UDLs as
Approx operator "" _a(long double);
and instead declare them as
Approx operator ""_a(long double);
Notice that the second version does not have a space between the ""
and
the literal suffix.
New source file template
If you are adding new source file, there is a template you should use. Specifically, every source file should start with the licence header:
// Copyright Catch2 Authors
// Distributed under the Boost Software License, Version 1.0.
// (See accompanying file LICENSE.txt or copy at
// https://www.boost.org/LICENSE_1_0.txt)
// SPDX-License-Identifier: BSL-1.0
The include guards for header files should follow the pattern {FILENAME}_INCLUDED
.
This means that for file catch_matchers_foo.hpp
, the include guard should
be CATCH_MATCHERS_FOO_HPP_INCLUDED
, for catch_generators_bar.hpp
, the include
guard should be CATCH_GENERATORS_BAR_HPP_INCLUDED
, and so on.
Adding new CATCH_CONFIG
option
When adding new CATCH_CONFIG
option, there are multiple places to edit:
CMake/CatchConfigOptions.cmake
- this is used to generate the configuration options in CMake, so that CMake frontends know about them.docs/configuration.md
- this is where the options are documentedsrc/catch2/catch_user_config.hpp.in
- this is template for generatingcatch_user_config.hpp
which contains the materialized configuration- other files as needed, e.g.
catch2/internal/catch_config_foo.hpp
for the logic that guards the configuration
CoC
This project has a CoC. Please adhere to it while contributing to Catch2.
This documentation will always be in-progress as new information comes up, but we are trying to keep it as up to date as possible.