💾 Archived View for skyjake.fi › gemlog › 2021-11_the-foundation.gmi captured on 2021-11-30 at 19:37:34. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
I've been meaning to write more about my C library that powers Lagrange and a few of my other projects. Here is the revised introduction I wrote for its source repository, hot off the press.
-----
An object-oriented C library whose API is designed for a particular coding style, taking cues from C++ STL and Qt.
I used to write a lot of C++. Over the years, code bases grew larger and more complex, and despite advances in hardware capabilities, everything got slower and more cumbersome. There is a real cost incurred by an ever-increasing cognitive load. C++ seems to be evolving toward being an all-purposes language, so it includes all the bells and whistles in addition to the kitchen sink, but personally, I like something that is smaller and optimized around a small set of core ideas.
These core ideas can be summarized as:
Yes, preprocessor macros are a key part of the library. They are used for all the boilerplate declarations needed to make the rest of the system work in a coherent manner. While one could write all of it out manually, that would a) take a lot more effort, b) be a maintenance nightmare, and c) be quite error-prone.
Here is an example of declaring a simple data type that uses a constructor and a destructor.
iDeclareType(Example) iDeclareTypeConstruction(Example)
These would give you an opaque type `iExample` and declare its basic initializer and deinitializer methods:
void init_Example (iExample *); void deinit_Example (iExample *);
One needs to manually write the implementation of these two methods. They initialize and deinitialize the members of the type as needed, allocate/release resources, etc. But `iDeclareTypeConstruction` also automatically defines the following methods as very small `static inline` functions:
iExample * new_Example (void); void delete_Example (iExample *); iExample * collect_Example (iExample *); iExample * collectNew_Example (void);
Using these, one can create and destroy instances of the type without having access to the actual members of `iExample`. The collector methods are important as well: one can use those at appropriate times, for example when receiving a return value from a function, to give ownership of an object to the garbage collector. Then the object can be used normally until the garbage collector is told to free all the previously collected memory. In practice, this could happen periodically in a program, for example immediately after rendering a frame.
Global identifiers and functions generally use the following naming convention:
C doesn't have namespaces, so some kind of a prefix is required to avoid conflicts. `i` was chosen to be visually as small as possible, so it doesn't interfere readability as much as a longer or large capital letter prefix.
Usually type names precede a method name in C APIs, but I find it makes reading the code more difficult. The beginnings of words are important as the eye is naturally drawn to them, so we want to have actually meaning information at the beginning of each line of code.
Having the type name be the last part of an identifier is also advantegous because thanks to its location, it becomes naturally associated with the instance pointer, i.e., the first parameter. This enables chaining method calls without losing track of types.
This naming convention is also why the library is called "the_Foundation": "Foundation" is the actual name, representing the library's role as a low-level programming framework upon which applications can be built.
Iterators are one of the most useful parts of C++ STL. The iterators in the_Foundation try to replicate some of STL's most useful properties:
The primary limitation for iterators in the library is that they cannot allocate memory on the heap. The iteration must be done with only local variables that do not need deinitialization when exiting the scope. This allows the use of a simple `for` loop:
iArrayIterator i;
for (init_ArrayIterator(&i, foo); i.value; next_ArrayIterator(&i)) {
/* do stuff... */
}
There is a macro for the code above:
iForEach(Array, i, foo) {
/* do stuff... */
}
Similarly, there are `iConstForEach`, `iReverseForEach`, and `iReverseConstForEach` macros for different types of iterators. These can be used with any type as long as the corresponding functions are available to be called; it all hinges on the naming convention.
There isn't a one-size-fits-all solution to memory ownership. Sometimes it's better to allocate lots of elements in a tightly packed array, and sometimes it's better if they're separate and shareable copy-on-write buffers. Therefore, the library supports a couple of different ownership models. It's important to choose the simplest, most suitable model for the problem at hand.
A key utility is the garbage collector. However, this isn't some autonomous memory-traversing monster, but the simplest possible implementation: you hand it a pointer and its destructor function, and later on when you have time, tell the collector to call all the pending destructors. It does this in a thread-safe manner and generally tries to get out of the way. Importantly, nothing is deleted until it is explicitly requested.
Consequently, this allows writing code in a manner where you can simply create instances as needed, perhaps during a chained method call, and not have to worry about cleaning up after yourself. This makes it more like C++ with its implicitly caller destructors. In special cases like long loops where lots of allocations would crop up before one gets a chance to recycle them, it is possible to call `iBeginCollect()`/`iEndCollect()` functions to partially clear out the garbage.
Reference counting is another key technique that is particularly useful when objects have shared ownership. And of course, it can be used together with the garbage collector. The `iClob()` macro ("collect object") gives ownership of a reference to any object (of any class) to the collector. Classes know how to destroy themselves, so in this case one doesn't need to specify the destructor.
Programming is about holding a complex mental model in your head and trying to reason about its behavior. The more complexity one's language and tools layer on, the more difficult the task becomes. This library tries to make it as straightforward as possible to write and read a program, while understanding how the code behaves from top to bottom at the level of individual bytes in memory.
📅 2021-11-01
🏷 Programming, the_Foundation
CC-BY-SA 4.0