diff options
Diffstat (limited to 'CONTRIBUTING.md')
| -rw-r--r-- | CONTRIBUTING.md | 494 |
1 files changed, 371 insertions, 123 deletions
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 6ee39eec..fb10c7da 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -17,15 +17,15 @@ The long-read part: ## General information For specific proposals, please provide them as -[pull requests](https://github.com/QMatrixClient/libQMatrixClient/pulls) +[pull requests](https://github.com/quotient-im/libQuotient/pulls) or -[issues](https://github.com/QMatrixClient/libqmatrxclient/issues) +[issues](https://github.com/quotient-im/libQuotient/issues) For general discussion, feel free to use our Matrix room: -[#quaternion:matrix.org](https://matrix.to/#/#quaternion:matrix.org). +[#quotient:matrix.org](https://matrix.to/#/#quotient:matrix.org). If you're new to the project (or FLOSS in general), -[issues tagged as simple](https://github.com/QMatrixClient/libQMatrixClient/labels/simple) -are smaller tasks that may typically take 1-3 days. +[issues tagged as easy](https://github.com/quotient-im/libQuotient/labels/easy) +are smaller tasks that don't require much knowledge about the project. You are welcome aboard! ### Pull requests and different branches recommended @@ -35,8 +35,8 @@ See the GitHub Help [articles about pull requests](https://help.github.com/artic to learn how to deal with them. We recommend creating different branches for different (logical) -changes, and creating a pull request when you're done into the master branch. -See the GitHub documentation on +changes, and creating a pull request when you're done; the development +integration branch is `dev`. See the GitHub documentation on [creating branches](https://help.github.com/articles/creating-and-deleting-branches-within-your-repository/) and [using pull requests](https://help.github.com/articles/using-pull-requests/). @@ -44,8 +44,8 @@ and ### How we handle proposals We use GitHub to track all changes via its -[issue tracker](https://github.com/QMatrixClient/libQMatrixClient/issues) and -[pull requests](https://github.com/QMatrixClient/libQMatrixClient/pulls). +[issue tracker](https://github.com/quotient-im/libQuotient/issues) and +[pull requests](https://github.com/quotient-im/libQuotient/pulls). Specific changes are proposed using those mechanisms. Issues are assigned to an individual who works on it and then marks it complete. If there are questions or objections, the conversation area of that @@ -86,33 +86,30 @@ a commit without a DCO is an accident and the DCO still applies. --> ### License -Unless a contributor explicitly specifies otherwise, we assume that all -contributed code is released under [the same license as libQMatrixClient itself](./COPYING), -which is LGPL v2.1 as of the time of this writing. +Unless a contributor explicitly specifies otherwise, we assume contributors +to agree that all contributed code is released either under *LGPL v2.1 or later*. +This is more than just [LGPL v2.1 libQuotient now uses](./COPYING) +because the project plans to switch to LGPL v3 for library code in the near future. <!-- The below is invalid yet! All new contributed material that is not executable, including all text when not executed, is also released under the [Creative Commons Attribution 4.0 International (CC BY 4.0) license](https://creativecommons.org/licenses/by/4.0/) or later. --> Any components proposed for reuse should have a license that permits releasing -a derivative work under LGPL v2.1. Moreover, the license of a proposed component -should be approved by OSI, no exceptions. +a derivative work under *LGPL v3 or later* (that includes licenses permitting +*LGPL v2.1 or later* but not *LGPL v2.1 only*). In any case, the component +should be redistributable under a license from +[the list approved by OSI](https://opensource.org/licenses), no exceptions. -## Vulnerability reporting (security issues) +We use [SPDX](https://spdx.dev) conventions for copyright statements. Please +follow them when making a sizable contribution: add your name and year to +the top of the file. New files should begin with the following preamble: +```cpp +// SPDX-FileCopyrightText: 2021 Your Name <your@email.address> +// SPDX-License-Identifier: LGPL-2.1-or-later +``` -If you find a significant vulnerability, or evidence of one, -use either of the following contacts: -* send an email to Kitsune Ral [Kitsune-Ral@users.sf.net](mailto:Kitsune-Ral@users.sf.net) -* reach out in Matrix to #kitsune:matrix.org (if you can, switch encryption **on**) - -In any of these two options, _indicate that you have such information_ -(do not share the information yet), and we'll tell you the next steps. - -By default, we will give credit to anyone who reports a vulnerability so that -we can fix it. If you want to remain anonymous or pseudonymous instead, please -let us know that; we will gladly respect your wishes. If you provide a fix as -a PR, you have no way to remain anonymous but we still accept contributors with -pseudonyms. +## Vulnerability reporting (security issues) - see [SECURITY.md](./SECURITY.md) ## Documentation changes @@ -121,7 +118,7 @@ filename extension. Any help on fixing/extending these is more than welcome. Where reasonable, limit yourself to Markdown that will be accepted by different markdown processors (e.g., what is specified by CommonMark or the original -Markdown). In practice, as long as libQMatrixClient is hosted at GitHub, +Markdown). In practice, as long as libQuotient is hosted at GitHub, [GFM (GitHub-flavoured Markdown)](https://help.github.com/articles/github-flavored-markdown/) is used to show those files in a browser, so it's fine to use its extensions. In particular, you can mark code snippets with the programming language used; @@ -136,10 +133,9 @@ unfortunately this other algorithm is *also* called GitHub-flavoured markdown. In your markdown, please don't use tab characters and avoid "bare" URLs. In a hyperlink, the link text and URL should be on the same line. While historically we didn't care about the line length in markdown texts -(and more often than not put the whole paragraph into one line), this is subject -to change anytime soon, with 80-character limit _recommendation_ -(which is softer than the limit for C/C++ code) imposed on everything -_except hyperlinks_ (because wrapping hyperlinks breaks the rendering). +(and more often than not put the whole paragraph into one line), this is no more +recommended; instead, try to use 80-character limit (similar to the limit for +C/C++ code) _except hyperlinks_ - wrapping breaks them. Do not use trailing two spaces for line breaks, since these cannot be seen and may be silently removed by some tools. If, for whatever reason, a blank line @@ -147,103 +143,293 @@ is not an option, use <tt><br /></tt> (an HTML break). ## End of TL;DR -If you don't plan/have substantial contributions, you can end reading here. Further sections are for those who's going to actively hack on the library code. +If you don't plan/have substantial contributions, you can stop reading here. +Further sections are for those who's going to actively hack on the library code. ## Code changes -The code should strive to be DRY (don't repeat yourself), clear, and obviously correct. Some technical debt is inevitable, just don't bankrupt us with it. Refactoring is welcome. - -### Generated C++ code for CS API -The code in lib/csapi, although it resides in Git, is actually generated from the official Matrix Swagger/OpenAPI definition files. If you're unhappy with something in that directory and want to improve the code, you have to understand the way these files are produced and setup some additional tooling. The shortest possible procedure resembling the below text can be found in .travis.yml (our Travis CI configuration actually regenerates those files upon every build). The generating sequence only works with CMake atm; patches to enable it with qmake are (you guessed it) very welcome. - -#### Why generate the code at all? -Because before both original authors of libQMatrixClient had to do monkey business of writing boilerplate code, with the same patterns, types etc., literally, for every single API endpoint, and one of the authors got fed up with it at some point in time. By then about 15 job classes were written; the entire API counts about 100 endpoints. Besides, the existing jobs had to be updated according to changes in CS API that have been, and will keep coming. Other considerations can be found in [this talk about API description languages that briefly touches on GTAD](https://youtu.be/W5TmRozH-rg). - -#### Prerequisites for CS API code generation -1. Get the source code of GTAD and its dependencies, e.g. using the command: `git clone --recursive https://github.com/KitsuneRal/gtad.git` -2. Build GTAD: in the source code directory, do `cmake . && cmake --build .` (you might need to pass `-DCMAKE_PREFIX_PATH=<path to Qt>`, similar to libQMatrixClient itself). -3. Get the Matrix CS API definitions that are included in the matrix-doc repo: `git clone https://github.com/QMatrixClient/matrix-doc.git` (QMatrixClient/matrix-doc is a fork that's known to produce working code; you may want to use your own fork if you wish to alter something in the API). - -#### Generating lib/csapi contents -1. Pass additional configuration to CMake when configuring libQMatrixClient, namely: `-DMATRIX_DOC_PATH=<path you your matrix-doc repo> -DGTAD_PATH=<path to gtad binary (not the repo!)>`. If everything's right, these two CMake variables will be mentioned in CMake output and will trigger configuration of an additional build target, see the next step. -2. Generate the code: `cmake --build <your build dir> --target update-api`; if you use CMake with GNU Make, you can just do `make update-api` instead. Building this target will create (overwriting without warning) .h and .cpp files in lib/csapi directory for all YAML files it can find in `matrix-doc/api/client-server`. -3. Once you've done that, you can build the library as usual; rerunning CMake is recommended if the list of generated files has changed. - -#### Changing things in lib/csapi -See the more detailed description of what GTAD is and how it works in the documentation on GTAD in its source repo. Only parts specific for libQMatrixClient are described here. - -GTAD uses the following three kinds of sources: -1. OpenAPI files. Each file is treated as a separate source (because this is how GTAD works now). -2. A configuration file, in our case it's lib/csapi/gtad.yaml - this one is common for the whole API. -3. Source code template files: lib/csapi/{{base}}.*.mustache - also common. - -The mustache files have a templated (not in C++ sense) definition of a network job, deriving from BaseJob; each job class is prepended, if necessary, with data structure definitions used by this job. The look of those files is hideous for a newcomer; the fact that there's no highlighter for the combination of Mustache (originally a web templating language) and C++ doesn't help things, either. To slightly simplify things some more or less generic constructs are defined in gtad.yaml (see its "mustache:" section). Adventurous souls that would like to figure what's going on in these files should speak up in the libQMatrixClient room - I (Kitsune) will be very glad to help you out. - -The types map in gtad.yaml is the central switchboard when it comes to matching OpenAPI types with C++ (and Qt) ones. It uses the following type attributes aside from pretty obvious "imports:": -* `avoidCopy` - this attribute defines whether a const ref should be used instead of a value. For basic types like int this is obviously unnecessary; but compound types like `QVector` should rather be taken by reference when possible. -* `moveOnly` - some types are not copyable at all and must be moved instead (an obvious example is anything "tainted" with a member of type `std::unique_ptr<>`). The template will use `T&&` instead of `T` or `const T&` to pass such types around. -* `useOmittable` - wrap types that have no value with "null" semantics (i.e. number types and custom-defined data structures) into a special `Omittable<>` template defined in `converters.h` - a substitute for `std::optional` from C++17 (we're still at C++14 yet). -* `omittedValue` - an alternative for `useOmittable`, just provide a value used for an omitted parameter. This is used for bool parameters which normally are considered false if omitted (or they have an explicit default value, passed in the "official" GTAD's `defaultValue` variable). -* `initializer` - this is a _partial_ (see GTAD and Mustache documentation for explanations but basically it's a variable that is a Mustache template itself) that specifies how exactly a default value should be passed to the parameter. E.g., the default value for a `QString` parameter is enclosed into `QStringLiteral`. - -Instead of relying on the event structure definition in the OpenAPI files, gtad.yaml uses pointers to libQMatrixClient's event structures: `EventPtr`, `RoomEventPtr` and `StateEventPtr`. Respectively, arrays of events, when encountered in OpenAPI definitions, are converted to `Events`, `RoomEvents` and `StateEvents` containers. When there's no way to figure the type from the definition, an opaque `QJsonObject` is used, leaving the conversion to the library and/or client code. - -### Library API and doc-comments - -Whenever you add a new call to the library API that you expect to be used from client code, you must supply a proper doc-comment along with the call. Doxygen (with backslashes) style is preferred. You can find that some parts of the code still use JavaDoc (with @'s) style; feel free to replace it with Doxygen backslashes if that bothers you. Some parts are not even documented; add doc-comments to them is highly encouraged. - -Calls, data structures and other symbols not intended for use by clients should _not_ be exposed in (public) .h files, unless they are necessary to declare other public symbols. In particular, this involves private members (functions, typedefs, or variables) in public classes; use pimpl idiom to hide implementation details as much as possible. - -Note: As of now, all header files of libQMatrixClient are considered public; this may change eventually. - -### Qt-flavoured C++ - -This is our primary language. We don't have a particular code style _as of yet_ but some rules-of-thumb are below: -* 4-space indents, no tabs, no trailing spaces, no last empty lines. If you spot the code abusing these - we'll thank you for fixing it. -* Prefer keeping lines within 80 characters. -* Braces after if's, while's, do's, function signatures etc. take a separate line. Keeping the opening brace on the same line is still ok. -* A historical deviation from the usual Qt code format conventions is an extra indent inside _classes_ (access specifiers go at +4 spaces to the base, members at +8 spaces) but not _structs_ (members at +4 spaces). This may change in the future for something more conventional. -* Please don't make "hypocritical structs" with protected or private members. Just make them classes instead. -* For newly created classes, keep to [the rule of 3/5/0](http://en.cppreference.com/w/cpp/language/rule_of_three) - make sure to read about the rule of zero if you haven't before, it's not what you might think it is. -* Qt containers are generally preferred to STL containers; however, there are notable exceptions, and libQMatrixClient already uses them: +The code should strive to be DRY (don't repeat yourself), clear, and obviously +correct (i.e. buildable). Some technical debt is inevitable, +just don't bankrupt us with it. Refactoring is welcome. + +### Code style and formatting + +As of Quotient 0.7, the C++ standard for newly written code is C++20. + +The code style is defined by `.clang-format`, and in general, all C++ files +should follow it. Files with minor deviations from the defined style are still +accepted in PRs; however, unless explicitly marked with `// clang-format off` +and `// clang-format on`, these deviations will be rectified any commit soon +after. + +Notable things from .clang-format: +* 4-space indents, no tabs, no trailing spaces, no last empty lines. If you + spot the code abusing these - thank you for fixing it. +* Prefer keeping lines within 80 characters. Slight overflows are ok only + if that helps readability. + +Additionally: +* Please don't make "hypocritical structs" with protected or private members. + In general, `struct` is used to denote a plain-old-data structure, rather + than data+behaviour. If you need access control or are adding yet another + non-trivial (construction, assignment) member function to a `struct`, + just make it a `class` instead. +* For newly created classes, keep to + [the rule of 3/5/0](http://en.cppreference.com/w/cpp/language/rule_of_three) - + make sure to read about the rule of zero if you haven't before, it's not + what you might think it is. +* Qt containers are generally preferred to STL containers; however, there are + notable exceptions, and libQuotient already uses them: * `std::array` and `std::deque` have no direct counterparts in Qt. - * Because of COW semantics, Qt containers cannot hold uncopyable classes. Classes without a default constructor are a problem too. Examples of that are `SyncRoomData` and `EventsArray<>`. Use STL containers for those but see the next point and also consider if you can supply a reasonable copy/default constructor. - * STL containers can be freely used in code internal to a translation unit (i.e., in a certain .cpp file) _as long as that is not exposed in the API_. It's ok to use, e.g., `std::vector` instead of `QVector` to tighten up code where you don't need COW, or when dealing with uncopyable data structures (see the previous point). However, exposing STL containers in the API is not encouraged (except where absolutely necessary, e.g. we use `std::deque` for a timeline). Exposing STL containers or iterators in API intended for usage by QML code (e.g. in `Q_PROPERTY`) is unlikely to work and therefore unlikely to be accepted into `master`. -* Use `QVector` instead of `QList` where possible - see a [great article by Marc Mutz on Qt containers](https://marcmutz.wordpress.com/effective-qt/containers/) for details. + * Because of COW semantics, Qt containers cannot hold uncopyable classes. + Classes without a default constructor are a problem too. Examples of that + are `SyncRoomData` and `EventsArray<>`. Use STL containers for structures + having those but see the next point and also consider if you can supply + a reasonable copy/default constructor. + * STL containers can be freely used in code internal to a translation unit + (i.e., in a certain .cpp file) _as long as that is not exposed in the API_. + It's ok to use, e.g., `std::vector` instead of `QVector` to tighten up + code where you don't need COW, or when dealing with uncopyable + data structures (see the previous point). However, exposing STL containers + in the API is not encouraged (except where absolutely necessary, e.g. we use + `std::deque` for a timeline). Especially when it comes to API intended + for usage from QML (e.g. `Q_PROPERTY`), STL containers or iterators are + unlikely to work and therefore unlikely to be accepted into `dev`. + * Notwithstanding the above (you're not going to use them with QML anyway), + prefer `std::unique_ptr<>` over `QScopedPointer<>` as it gives stronger + guarantees. +* Always use `QVector` instead of `QList` unless Qt's own API uses it - see the + [great article by Marc Mutz on Qt containers](https://marcmutz.wordpress.com/effective-qt/containers/) + for details. With Qt 6, these two become the same type matching what used + to be `QVector` in Qt 5. + +### API conventions + +Calls, data structures and other symbols not intended for use by clients +should _not_ be exposed in (public) .h files, unless they are necessary +to declare other public symbols. In particular, this involves private members +(functions, typedefs, or variables) in public classes; use pimpl idiom to hide +implementation details as much as possible. `_impl` namespace is reserved for +definitions that should not be used by clients and are not covered by +API guarantees. + +Note: As of now, all header files of libQuotient are considered public; +this may change eventually. + +### Comments + +Whenever you add a new call to the library API that you expect to be used +from client code, make sure to supply a proper doc-comment along with the call. +Quotient uses the Doxygen style; some legacy code may use Javadoc style but it +is not encouraged any more. Some parts are not documented at all; +adding doc-comments to them is highly encouraged and is a great first-time +contribution. + +Use `\brief` for the summary, and follow with details after +an empty doc-comment line, using `\param`, `\return` etc. as necessary. + +For in-code comments, the advice is as follows: +* Don't restate what's happening in the code unless it's not really obvious. + We assume the readers to have some command of C++ and Qt. If your code is + not obvious, consider making it clearer itself before commenting. +* That said, both C++ and Qt have their arcane features and dark corners, + and we don't want to limit anybody who feels they have a case for + variadic templates, raw literals, and so on. Use your experience to figure + what might be less well-known to readers and comment such cases: leave + references to web pages, Quotient wiki etc. +* Make sure to document not so much "what" but more "why" certain code is done + the way it is. In the worst case, the logic of the code can be + reverse-engineered; but you can almost never reverse-engineer the line of + reasoning and the pitfalls avoided. ### Automated tests -There's no testing framework as of now; either Catch or Qt Test or both will be used eventually. However, as a stopgap measure, qmc-example is used for automated end-to-end testing. - -Any significant addition to the library API should be accompanied by a respective test in qmc-example. To add a test you should: -- Add a new private slot to the `QMCTest` class. -- Add to the beginning of the slot the line `running.push_back("Test name");`. -- Add test logic to the slot, using `QMC_CHECK` macro to assert the test outcome. ALL (even failing) branches should conclude with a QMC_CHECK invocation, unless you intend to have a "DID NOT FINISH" message in the logs under certain conditions. -- Call the slot from `QMCTest::startTests()`. - -`QMCTest` sets up some basic test fixture to help you with testing; notably by the moment `startTests()` is invoked you can rely on having a working connection in `c` member variable and a test room in `targetRoom` member variable. PRs to introduce a proper testing framework are very welcome (make sure to migrate tests from qmc-example though); shifting qmc-example to use Qt Test seems to be a particularly low-hanging fruit. +We gradually introduce autotests based on a combination of CTest and Qt Test +frameworks - see `autotests/` directory. There are very few of those, as we +have just started adding those to the new code (you guessed it; adding more +tests to the old code is very welcome). + +Aside from that, libQuotient comes with a command-line end-to-end test suite +called Quotest. Any significant addition to the library API should be +accompanied by a respective test in `autotests/` and/or in Quotest. + +To add a test to autotests: +- In a new .cpp file in `autotests/`, define a test class derived from + QObject with `private Q_SLOTS:` section having the member functions called + for testing. If you feel more comfortable using a header file to define + the class, feel free to do so. If you're new to Qt Test framework, use + existing tests as a guidance. +- Add a `quotient_add_test` macro call with your test to + `autotests/CMakeLists.txt` + +To add a test to Quotest: +- In `quotest.cpp`, add a new test to the `TestSuite` class. Similar to Qt Test, + each test in Quotest is a private slot; unlike Qt Test, you should use + special macros, `TEST_DECL()` and `TEST_IMPL()`, to declare and define + the test (those macros conceal passing the testing handle in `thisTest` + variable to the test method). +- In the test function definition, add test logic using `FINISH_TEST` macro + to check for the test outcome and complete the test (be mindful that + `FINISH_TEST` always `return`s, not only in case of error). ALL (even failing) + branches should conclude with a `FINISH_TEST` (or `FAIL_TEST` that is + a shortcut for a failing `FINISH_TEST`) invocation, unless you intend to have + a "DID NOT FINISH" message in the logs in certain conditions. + +The `TestManager` class sets up some basic test fixture to help you with testing; +notably, the tests can rely on having an initialised `Room` object with loaded +state for the test room in `targetRoom` member variable. Note that it's normal +for tests to go async, which is not something Qt Test is easy with (and this +is why Quotest doesn't directly use Qt Test but rather fetches a few ideas +from it). ### Security, privacy, and performance -Pay attention to security, and work *with* (not against) the usual security hardening mechanisms (however few in C++). +Pay attention to security, and work *with*, not against, the usual security hardening practices (however few in C++). -`char *` and similar unchecked C-style read/write arrays are forbidden - use Qt containers or at the very least `std::array<>` instead. Where you see fit (usually with data structures), try to use smart pointers, especially `std::unique_ptr<>` or `QScopedPointer` instead of bare pointers. When dealing with `QObject`s, use the parent-child ownership semantics exercised by Qt (this is preferred to using smart pointers). Shared pointers are not used in the code so far; but if you find a particular use case where the strict semantic of unique pointers doesn't help and a shared pointer is necessary, feel free to step up with the working code and it will be considered for inclusion. +`char *` and similar unchecked C-style read/write arrays are forbidden - use +Qt containers or at the very least `std::array<>` instead. Where you see fit +(usually with data structures), try to use smart pointers, especially +`std::unique_ptr<>` or `QScopedPointer` instead of bare pointers. When dealing +with `QObject`s, use the parent-child ownership semantics exercised by Qt +(this is preferred to using smart pointers). If you find a particular use case +where the strict semantic of unique pointers doesn't help and a shared pointer +is necessary, feel free to step up with the working code and it will be +considered for inclusion. Exercise the [principle of least privilege](https://en.wikipedia.org/wiki/Principle_of_least_privilege) where reasonable and appropriate. Prefer less-coupled cohesive code. -Protect private information, in particular passwords and email addresses. Absolutely _don't_ spill around this information in logs - use `access_token` and similar opaque ids instead, and only display those in UI where needed. Do not forget about local access to data (in particular, be very careful when storing something in temporary files, let alone permanent configuration or state). Avoid mechanisms that could be used for tracking where possible (we do need to verify people are logged in but that's pretty much it), and ensure that third parties can't use interactions for tracking. Matrix protocols evolve towards decoupling the personally identifiable information from user activity entirely - follow this trend. +Protect private information, in particular passwords and email addresses. +Absolutely _don't_ spill around this information in logs, and only display +those in UI where really needed. Do not forget about local access to data +(in particular, be very careful when storing something in temporary files, +let alone permanent configuration or state). Avoid mechanisms that could be +used for tracking where possible (we do need to verify people are logged in +but that's pretty much it), and ensure that third parties can't use interactions +for tracking. Matrix protocols evolve towards decoupling +the personally identifiable information from user activity entirely - follow +this trend. + +We want the software to have decent performance for users even on weaker +machines. At the same time we keep libQuotient single-threaded as much +as possible, to keep the code simple. That means being cautious about operation +complexity (read about big-O notation if you need a kickstart on the subject). +This especially refers to operations on the whole timeline and the list of +users - each of these can have tens of thousands of elements so even operations +with linear complexity, if heavy enough (with I/O or complex processing), +can produce noticeable GUI freezing or stuttering. When you don't see a way +to reduce algorithmic complexity, either split processing into isolated +pieces that can be individually scheduled as queued events (see the end of +`Connection::consumeRoomData()` to get the idea) or uncouple the logic from +GUI and execute it outside of the main thread with `QtConcurrent` facilities. + +Having said that, there's always a trade-off between various attributes; +in particular, readability and maintainability of the code is more important +than squeezing every bit out of that clumsy algorithm. Beware of premature +optimization and profile the code before before diving into hardcore tweaking +that might not give the benefits you think it would. + +Speaking of profiling logs (see README.md on how to turn them on) - if you +expect some code to take considerable (more than 10k "simple operations") time +you might want to setup a `QElapsedTimer` and drop the elapsed time into logs +under `PROFILER` logging category. See the existing code for examples - +`room.cpp` has quite a few. In order to reduce small timespan logging spam, +`PROFILER` log lines are usually guarded by a check that the timer counted big +enough time (200 microseconds by default, 20 microseconds for tighter parts); +this threshold can be altered at compile-time by defining `PROFILER_LOG_USECS` +preprocessor symbol (i.e. passing `-DPROFILE_LOG_USECS=<usecs>` to the compiler +if you're on Linux/macOS). + +### Generated C++ code for CS API +The code in `lib/csapi`, `lib/identity` and `lib/application-service`, although +it resides in Git, is actually generated from the official Swagger/OpenAPI +definition files. If you're unhappy with something in there and want to improve +the code, you have to understand the way these files are produced and setup +some additional tooling. The shortest possible procedure resembling +the below text can be found in .github/workflows/ci.yml (our CI configuration +tests regeneration of those files). As described below, there is also a handy +build target for CMake. -We want the software to have decent performance for typical users. At the same time we keep libQMatrixClient single-threaded as much as possible, to keep the code simple. That means being cautious about operation complexity (read about big-O notation if you need a kickstart on the topic). This especially refers to operations on the whole timeline and the list of users - each of these can have tens of thousands of elements so even operations with linear complexity, if heavy enough, can produce noticeable GUI freezing. When you don't see a way to reduce algorithmic complexity, embed occasional `processEvents()` invocations in heavy loops (see `Connection::saveState()` to get the idea). +#### Why generate the code at all? +Because otherwise we have to do monkey business of writing boilerplate code, +with the same patterns, types etc., literally, for every single API endpoint, +and one of libQuotient authors got fed up with it at some point in time. +By then about 15 job classes were written; the entire API is about 100 endpoints +and counting. Besides, the existing jobs had to be updated according to changes +in CS API that have been, and will keep, coming. Other considerations can be +found in [this talk about API description languages](https://youtu.be/W5TmRozH-rg) +that also briefly touches on GTAD. -Having said that, there's always a trade-off between various attributes; in particular, readability and maintainability of the code is more important than squeezing every bit out of that clumsy algorithm. Beware of premature optimization and have profiling data around before going into some hardcore optimization. +#### Prerequisites for CS API code generation +1. Get the source code of GTAD and its dependencies. Recent libQuotient + includes GTAD as a submodule so you can get everything you need by updating + gtad/gtad submodule in libQuotient sources: + `git submodule update --init --recursive gtad/gtad`. + + You can also just clone GTAD sources to keep them separate from libQuotient: + `git clone --recursive https://github.com/quotient-im/gtad.git` +2. Configure and build GTAD: same as libQuotient, it uses CMake so this should + be quite straightforward (if not - you're probably not quite ready for this + stuff anyway). +3. Get Matrix CS API definitions from a matrix-spec repo. Although the official + repo is at https://github.com/matrix-org/matrix-spec.git` (formerly + https://github.com/matrix-org/matrix-doc.git), you may or may not be able + to generate working code from it because the way it evolves is not + necessarily in line with libQuotient needs. For that reason, a soft fork + of the official definitions is kept at + https://github.com/quotient-im/matrix-spec.git that guarantees buildability + of the generated code. This repo closely follows the official one (but maybe + not its freshest commit), applying a few adjustments on top. And of course + you can use your own repository if you need to change the API definition. +4. If you plan to submit a PR with the generated code to libQuotient or just + would like it to be properly formatted, you should either ensure you have + clang-format (version 10 at least) in your PATH or pass + `-DCLANG_FORMAT=<path>` to CMake, as mentioned in the next section. + +#### Generating CS API contents +1. Pass additional configuration to CMake when configuring libQuotient: + `-DMATRIX_SPEC_PATH=/path/to/matrix-spec/ -DGTAD_PATH=/path/to/gtad`. + Note that `MATRIX_SPEC_PATH` should lead to the repo while `GTAD_PATH` should + have the path to GTAD binary. If you need to specify where your clang-format + is (see the previous section) add `-DCLANG_FORMAT=/path/to/clang-format` to + the line above. If everything's right, the detected locations will be + mentioned in CMake output and will trigger configuration of an additional + build target called `update-api`. +2. Generate the code: `cmake --build <your build dir> --target update-api`. + Building this target will create (overwriting without warning) source files + in `lib/csapi`, `lib/identity`, `lib/application-service` for all YAML files + it can find in `/path/to/matrix-spec/data/api/client-server` and their + dependencies. + +#### Changing generated code +See the more detailed description of what GTAD is and how it works in the documentation on GTAD in its source repo. Only parts specific for libQuotient are described here. -Speaking of profiling logs (see README.md on how to turn them on) - in order -to reduce small timespan logging spam, there's a default limit of at least -200 microseconds to log most operations with the PROFILER -(aka libqmatrixclient.profile.debug) logging category. You can override this -limit by passing the new value (in microseconds) in PROFILER_LOG_USECS to -the compiler. In the future, this parameter will be made changeable at runtime -_if_ needed. +GTAD uses the following three kinds of sources: +1. OpenAPI files. Each file is treated as a separate source (unlike + swagger-codegen, you do _not_ need to have a single file for the whole API). +2. A configuration file, in Quotient case it's `gtad/gtad.yaml` - common for + all OpenAPI files GTAD is invoked on. +3. Source code template files: `gtad/*.mustache` - are also common. + +The Mustache files have a templated (not in C++ sense) definition of a network +job class derived from BaseJob; if necessary, data structure definitions used +by this job are put before the job class. Bigger Mustache files look a bit +hideous for a newcomer; and the only known highlighter that can handle +the combination of Mustache (originally a web templating language) and C++ can +be found in CLion IDE. Fortunately, all our Mustache files are reasonably +concise and well-formatted these days. +To simplify things some reusable Mustache blocks are defined in `gtad.yaml` - +see its `mustache:` section. Adventurous souls that would like to figure +what's going on in these files should speak up in the Quotient room - +I (Kitsune) will be very glad to help you out. + +The types map in `gtad.yaml` is the central switchboard when it comes to matching OpenAPI types with C++ (and Qt) ones. It uses the following type attributes aside from pretty obvious "imports:": +* `avoidCopy` - this attribute defines whether a const ref should be used instead of a value. For basic types like int this is obviously unnecessary; but compound types like `QVector` should rather be taken by reference when possible. +* `moveOnly` - some types are not copyable at all and must be moved instead (an obvious example is anything "tainted" with a member of type `std::unique_ptr<>`). +* `useOmittable` - wrap types that have no value with "null" semantics (i.e. number types and custom-defined data structures) into a special `Omittable<>` template defined in `converters.h`, a drop-in upgrade over `std::optional`. +* `omittedValue` - an alternative for `useOmittable`, just provide a value used for an omitted parameter. This is used for bool parameters which normally are considered false if omitted (or they have an explicit default value, passed in the "official" GTAD's `defaultValue` variable). +* `initializer` - this is a _partial_ (see GTAD and Mustache documentation for explanations but basically it's a variable that is a Mustache template itself) that specifies how exactly a default value should be passed to the parameter. E.g., the default value for a `QString` parameter is enclosed into `QStringLiteral`. + +Instead of relying on the event structure definition in the OpenAPI files, `gtad.yaml` uses pointers to libQuotient's event structures: `EventPtr`, `RoomEventPtr` and `StateEventPtr`. Respectively, arrays of events, when encountered in OpenAPI definitions, are converted to `Events`, `RoomEvents` and `StateEvents` containers. When there's no way to figure the type from the definition, an opaque `QJsonObject` is used, leaving the conversion to the library and/or client code. ## How to check proposed changes before submitting them @@ -253,15 +439,66 @@ your commit into it (with an explanation what it is about and why). ### Standard checks -The following warnings configuration is applied with GCC and Clang when using CMake: `-W -Wall -Wextra -pedantic -Werror=return-type -Wno-unused-parameter -Wno-gnu-zero-variadic-macro-arguments` (the last one is to mute a warning triggered by Qt code for debug logging). We don't turn most of the warnings to errors but please treat them as such. In Qt Creator, the following line can be used with the Clang code model (before Qt Creator 4.7 you should explicitly enable the Clang code model plugin): `-Weverything -Werror=return-type -Wno-c++98-compat -Wno-c++98-compat-pedantic -Wno-unused-macros -Wno-newline-eof -Wno-exit-time-destructors -Wno-global-constructors -Wno-gnu-zero-variadic-macro-arguments -Wno-documentation -Wno-missing-prototypes -Wno-shadow-field-in-constructor -Wno-padded -Wno-weak-vtables` +The warnings configuration applied when using CMake can be found in +`CMakeLists.txt`. Most warnings triggered by that configuration are not formally +considered errors (the compiler will keep going) but please treat them as such. +If you want to be cautious, you can use the following line for your IDE's Clang +analyzer code model to enable as many compiler warnings as reasonable (that +does not include `clang-tidy`/`clazy` warnings - see below on those): +`-Weverything -Werror=return-type -Wno-c++98-compat -Wno-c++98-compat-pedantic -Wno-unused-macros -Wno-newline-eof -Wno-exit-time-destructors -Wno-global-constructors -Wno-gnu-zero-variadic-macro-arguments -Wno-documentation -Wno-missing-prototypes -Wno-shadow-field-in-constructor -Wno-padded -Wno-weak-vtables -Wno-unknown-attributes -Wno-comma -Wno-string-conversion -Wno-return-std-move-in-c++11`. ### Continuous Integration -We use Travis CI to check buildability and smoke-testing on Linux (GCC, Clang) and MacOS (Clang), and AppVeyor CI to build on Windows (MSVC). Every PR will go through these, and you'll see the traffic lights from them on the PR page. Failure on any platform will most likely entail a request to you for a fix before merging a PR. +We use CI to check buildability and smoke-testing on Linux (GCC, Clang), +MacOS (Clang), and Windows (MSVC). Every PR will go through these, and you'll +see the traffic lights from them on the PR page. If your PR fails +on any platform double-check that it's not your code causing it - and fix +(or ask how to fix if you don't know) if it is. ### Other tools -If you know how to use clang-tidy, here's a list of checks we do and do not use (a leading hyphen means a disabled check, an asterisk is a wildcard): `*,cert-env33-c,-cppcoreguidelines-pro-bounds-array-to-pointer-decay,-cppcoreguidelines-pro-bounds-constant-array-index,-cppcoreguidelines-pro-bounds-pointer-arithmetic,-cppcoreguidelines-pro-type-const-cast,-cppcoreguidelines-pro-type-union-access,-cppcoreguidelines-special-member-functions,-google-build-using-namespace,-google-readability-braces-around-statements,-hicpp-*,-llvm-*,-misc-unused-parameters,-misc-noexcept-moveconstructor,-modernize-use-using,-readability-braces-around-statements,readability-identifier-naming,-readability-implicit-bool-cast,-clang-diagnostic-*,-clang-analyzer-*`. If you're on CLion, you can simply copy-paste the above list into the Clang-Tidy inspection configuration. In Qt Creator 4.6 and newer, one can enable clang-tidy and clazy (clazy level 1 eats away CPU but produces some very relevant and unobvious notices, such as possible unintended copying of a Qt container, or unguarded null pointers). +Recent versions of Qt Creator and CLion can automatically run your code through +clang-tidy. The source code contains `.clang-tidy` file with the recommended +set of checks that doesn't give too many false positives. + +Qt Creator in addition knows about clazy, a Qt-aware static analysis tool that +hunts for Qt-specific issues that are easy to overlook otherwise, such as +possible unintended copying of a Qt container. Most of clazy checks are relevant +to our code, except: +`fully-qualified-moc-types,overloaded-signal,qstring-comparison-to-implicit-char,foreach,non-pod-global-static,qstring-allocations,jni-signatures,qt4-qstring-from-array`. + +### Submitting API changes + +If you changed the API definitions, the path to upstream becomes somewhat +intricate, as you have to coordinate with two projects, making up to 4 PRs along +the way. The recommended sequence depends on whether or not you have to +[write a Matrix Spec Change aka MSC](https://matrix.org/docs/spec/proposals). +Usually you have to, unless your API changes keep API semantics intact. +In that case: +1. Submit an MSC before submitting changes to the API definition files and + libQuotient. +2. The MSC gets reviewed by the Spec Core Team. This can be a lengthy process + but it's necessary for the Matrix ecosystem integrity. +3. When your MSC has at least some approvals (not necessarily a complete + acceptance but at least some approvals should be there) submit a PR to + libQuotient, referring to your `matrix-spec` repo. Make sure that generated + files are committed separately from non-generated ones (no need to make two + PRs; just separate them in different commits). +4. If/when your libQuotient PR is approved and MSC is not there yet you'll + be asked to submit a PR with API definition files at + `https://github.com/quotient-im/matrix-spec`. Note that this is _not_ + an official repo; but you can refer to your libQuotient PR as + an _implementation_ of the MSC - a necessary step before making a so-called + "spec PR". +5. Once MSC is accepted, submit your `matrix-spec` changes as a PR to + `https://github.com/matrix-org/matrix-spec` (the "spec PR" mentioned above). + This will require that your submission meets the standards set by this + project (they are quite reasonable and not too hard to meet). + +If your changes don't need an MSC, it becomes a more straightforward combination +of 2 PRs: one to `https://github.com/matrix-org/matrix-spec` ("spec PR") and one +to libQuotient (with the same guidance about putting generated and non-generated +files in different commits). ## Git commit messages @@ -279,15 +516,26 @@ When writing git commit messages, try to follow the guidelines in ## Reuse (libraries, frameworks, etc.) -C++ is unfortunately not very coherent about SDK/package management, and we try to keep building the library as easy as possible. Because of that we are very conservative about adding dependencies to libQMatrixClient. That relates to additional Qt components and even more to other libraries. Fortunately, even the Qt components now in use (Qt Core and Network) are very feature-rich and provide plenty of ready-made stuff. - -Regardless of the above paragraph (and as mentioned earlier in the text), we're now looking at possible options for automated testing, so PRs onboarding a test framework will be considered with much gratitude. +C++ is unfortunately not very coherent about SDK/package management, and we try to keep building the library as easy as possible. Because of that we are very conservative about adding dependencies to libQuotient. That relates to additional Qt components and even more to other libraries. Fortunately, even the Qt components now in use (Qt Core and Network) are very feature-rich and provide plenty of ready-made stuff. Some cases need additional explanation: -* Before rolling out your own super-optimised container or algorithm written from scratch, take a good long look through documentation on Qt and C++ standard library. Please try to reuse the existing facilities as much as possible. -* You should have a good reason (or better several ones) to add a component from KDE Frameworks. We don't rule this out and there's no prejudice against KDE; it just so happened that KDE Frameworks is one of most obvious reuse candidates but so far none of these components survived as libQMatrixClient deps. So we are cautious. -* Never forget that libQMatrixClient is aimed to be a non-visual library; QtGui in dependencies is only driven by (entirely offscreen) dealing with QPixmaps. While there's a bunch of visual code (in C++ and QML) shared between libQMatrixClient-enabled _applications_, this is likely to end up in a separate (libQMatrixClient-enabled) library, rather than libQMatrixClient itself. +* Before rolling out your own super-optimised container or algorithm written + from scratch, take a good long look through documentation on Qt and + C++ standard library. Please try to reuse the existing facilities + as much as possible. +* libQuotient is a library to build Qt applications; for that reason, + components from KDE Frameworks should be really lightweight and useful + to be accepted as a dependency. If the intention is to better integrate + libQuotient into KDE environment there's nothing wrong in building another + library on top of libQuotient. Consider people who run LXDE and normally + don't have KDE frameworks installed (some even oppose installing those) - + libQuotient caters to them too. +* Never forget that libQuotient is aimed to be a non-visual library; + QtGui in dependencies is only driven by (entirely offscreen) dealing with + QImages. While there's a bunch of visual code (in C++ and QML) shared + between Quotient-enabled _applications_, this is likely to end up + in a separate (Quotient-backed) library, rather than libQuotient itself. ## Attribution -This text is largely based on CONTRIBUTING.md from CII Best Practices Badge project, which is a collective work of its contributors (many thanks!). The text itself is licensed under CC-BY-4.0. +This text is based on CONTRIBUTING.md from CII Best Practices Badge project, which is a collective work of its contributors (many thanks!). The text itself is licensed under CC-BY-4.0. |