Thank you for inquiring facil.io‘s contribution guide. It’s people like you and me, that are willing to share our efforts, who help make the world of open source development so inspiring and wonderful.
“Facil” comes from the Spanish word “easy”, and this is embedded in facil.io's DNA.
facil.io contributions should (ideally) be:
Easy to use: clear and concise API, with macros that emulate “named arguments” when appropriate.
Easy to maintain:
Modular: even at the price of performance and even (although less desired) at the price of keeping things DRY.
Developers should be able to simply remove the module from their implementation if they're not using it.
To clarify, a module should have as small a responsibility as possible without requiring non-core modules. This makes the module easier to maintain and minimizes code fragility and code entanglement.
Clearly commented: you might not always have time to maintain your contribution. It's important to pave the way for others to lend a hand.
Easy to port:
When possible, code should be portable. This is both true in regards to CPU architecture and in regards to OS and environment.
The project currently has the following limitation that might be addressed in the future:
The code requires kqueue or epoll services from the OS, which means Linux / BSD / macOS.
The code assumes a Unix environment (file naming etc').
Some of the code (namely some HTTP parts) uses unaligned memory access (requiring newer CPUs and possibly introducing undefined behavior).
Easy to compile:
The code uses GNU make and although we have CMake support, neither CMake nor configure should be required at any point.
Easy to manage:
See the License section below. Contributions must relinquish ownership of contributed code, so licensing and copyright can be managed without the need to reach out to every contributer.
As a child, I wasn‘t any good with people (I’m not sure I‘m any better now that I’m older)... which is how come I became good with computers and why we have facil.io and other open source projects ;-)
However, I promise to do my best to be a respectful communicator and I ask that you do your best as well.
No matter if discussing a PR (where we might find ourselves entering a heated discussion) or answering an Issue (where sometime we find ourselves wondering why people think we work for them)... we should all remember that a little compassion and respect goes a long way.
facil.io is comprised of the following module “families”:
The Core:
This module family comprises facil.io's core. Although it can (mostly) be used outside of facil.io, none of the modules in this family can be removed. This includes:
The IO reactor (evio), the event loop (defer), the socket abstraction layer (sock) and the “glue” (facil).
Core types available for all the rest of the modules: fio_llist.h (a simple doubly linked list), fio_hashmap.h (a simple Hash Table), and spnlock (simple atomic operations and a spin-lock).
I avoided atomic.h and some C11 features because they weren't available on all the platforms where I was compiling facil.io, which is partially why I wrote the spnlock and the core types.
Core dynamic types (FIOBJ) with native JSON support.
This module is used like “glue” and a soft type system layer. Even the core can‘t function without it (for example, it’s used in cluster communication).
Services:
The services folder refers to independent modules that might require Core integration for initialization and resource management.
Services should always be optional. A good example is the Pub/Sub module that can be removed without disrupting the core...
However, this is also a good example for what not to do, since the Websocket module (as of version 0.4.5) requires the Pub/Sub service. This should be fixed by adding weak stub versions of the pubsub_* functions to the Websocket module.
HTTP / Websocket:
Although this module family seems very entangled, I did my best to make it easy to maintain and extend with a minimum of entanglement.
HTTP request and response modules support virtual function tables for future HTTP/2 extensions. The actual request/response implementations might vary between protocol implementation, but their interface should be version agnostic.
The Websocket module depends on the Pub/Sub service.
Optional modules that attach to the HTTP / Websocket layer should be placed in a subfolder (which should be listed in the makefile's LIB_PUBLIC_SUBFOLDERS).
Redis:
The redis engine is in it‘s own folder, both because it’s clearly an “add-on” (even though it‘s a pub/sub add-on) and because it’s as optional as it gets.
This is also a good example for my preference for modular design. The RESP parser is a single file library. It can be easily ported to different projects and is totally separate from the network layer.
Before you start working on a feature, I recommend that you open a PR to edit this CONTRIBUTING file.
Add the feature you want to work on to the following list (or assign an existing feature to yourself). This will allow us to discuss, in the PR's thread, any questions you might have or any expectations that might effect the API or the feature.
Once you have all the information you need to implementing the feature, the discussion can move to the actual feature's PR.
These are the features that have been requested so far. Even if any of them are assigned, feel free to offer your help:
| Feature | assigned | remarks |
|---|---|---|
| Documentation | 🙏 Help 🙏 | Placed at docs/_SOURCE |
| Tests | Never enough | run through tests.c but implement in source files. |
| Early Hints HTTP/1.1 | ||
| SSL/TLS | 🙏 Help 🙏 | See fio_tls_missing.c for example. |
| Websocket Client | Missing cookie retention. | |
| HTTP Client | Missing SSL/TLS, cookie retention and auto-redirect(?) | |
| HTTP/2 | Bo (me), help me? | |
| HTTP Router | No RegEx. Example: /users/(:id) | |
| PostgreSQL | Wrap libpq.h for events + pub/sub engine (?) | |
| Gossip (?) | For Pub/Sub engine scaling |
The project requires that all the code is licensed under the MIT license. Please refrain from using or offering code that requires a change to the licensing scheme.
I discovered GitHub doesn't offer a default CLA (Copyright and Licensing Agreement), so I adopted the one used by BearSSL, meaning:
the resulting code uses the MIT license, listing me (and only me) as the author. You can take credit by stating that the code was written by yourself, but should attribute copyright and authorship to me (Boaz Segev).
I will list contributors in the CHANGELOG and special contributions will be listed in the README and/or here.
This allows me to circumvent any future licensing concerns and prevent contributors from revoking the MIT license attached to their code.
@area55git (Area55) contributed the logo under a Creative Commons Attribution 4.0 International License..
@cdkrot took the time to test some of the demo code using valgrind, detecting a shutdown issue with in core defer library and offering a quick fix.
@madsheep and @nilclass took the time to expose a very quite issue (#16) that involved a long processing on_open websocket callback and very short network roundtrips, exposing a weakness in the HTTP/1.x logic.
@64 took the time to test the pre-released 0.6.0 version and submit PR #25, fixing a silent error and some warnings.