Attention: Here be dragons
This is the latest
(unstable) version of this documentation, which may document features
not available in or compatible with released stable versions of Godot.
Checking the stable version of the documentation...
Core types
Godot has a rich set of classes and templates that compose its core, and everything is built upon them.
This reference will try to list them in order for their better understanding.
Definitions
Godot uses the standard C99 datatypes, such as uint8_t,
uint32_t, int64_t, etc. which are nowadays supported by every
compiler. Reinventing the wheel for those is not fun, as it makes code
more difficult to read.
In general, care is not taken to use the most efficient datatype for a
given task unless using large structures or arrays. int is used
through most of the code unless necessary. This is done because nowadays
every device has at least a 32 bits bus and can do such operations in
one cycle. It makes code more readable too.
For files or memory sizes, size_t is used, which is warranted to be
64 bits.
For Unicode characters, CharType instead of wchar_t is used, because many architectures have 4 bytes long wchar_t, where 2 bytes might be desired. However, by default, this has not been forced and CharType maps directly to wchar_t.
References:
Memory model
PC is a wonderful architecture. Computers often have gigabytes of RAM, terabytes of storage and gigahertz of CPU, and when an application needs more resources the OS will swap out the inactive ones. Other architectures (like mobile or consoles) are in general more limited.
The most common memory model is the heap, where an application will request a region of memory, and the underlying OS will try to fit it somewhere and return it. This often works best and is flexible, but over time and with abuse, this can lead to segmentation.
Segmentation slowly creates holes that are too small for most common allocations, so that memory is wasted. There is a lot of literature about heap and segmentation, so this topic will not be developed further here. Modern operating systems use paged memory, which helps mitigate the problem of segmentation but doesn't solve it.
However, in many studies and tests, it is shown that given enough memory, if the maximum allocation size is below a given threshold in proportion to the maximum heap size and proportion of memory intended to be unused, segmentation will not be a problem over time as it will remain constant. In other words, leave 10-20% of your memory free and perform all small allocations and you are fine.
Godot ensures that all objects that can be allocated dynamically are small (less than a few kb at most). But what happens if an allocation is too large (like an image or mesh geometry or large array)? In this case Godot has the option to use a dynamic memory pool. This memory needs to be locked to be accessed, and if an allocation runs out of memory, the pool will be rearranged and compacted on demand. Depending on the need of the game, the programmer can configure the dynamic memory pool size.
Allocating memory
Godot has many tools for tracking memory usage in a game, especially during debug. Because of this, the regular C and C++ library calls should not be used. Instead, a few other ones are provided.
For C-style allocation, Godot provides a few macros:
memalloc()
memrealloc()
memfree()
These are equivalent to the usual malloc, realloc, free of the standard C library.
For C++-style allocation, special macros are provided:
memnew( Class / Class(args) )
memdelete( instance )
memnew_arr( Class , amount )
memdelete_arr( pointer to array )
which are equivalent to new, delete, new[] and delete[].
memnew/memdelete also use a little C++ magic and notify Objects right after they are created, and right before they are deleted.
For dynamic memory, use one of Godot's sequence types such as Vector<>
or LocalVector<>. Vector<> behaves much like an STL std::vector<>,
but is simpler and uses Copy-On-Write (CoW) semantics. CoW copies of
Vector<> can safely access the same data from different threads, but
several threads cannot access the same Vector<> instance safely.
It can be safely passed via public API if it has a Packed alias.
The Packed*Array types are aliases
for specific Vector<*> types (e.g., PackedByteArray,
PackedInt32Array) that are accessible via GDScript. Outside of core,
prefer using the Packed*Array aliases for functions exposed to scripts,
and Vector<> for other occasions.
LocalVector<> is much more like std::vector than Vector<>.
It is non-CoW, with less overhead. It is intended for internal use where
the benefits of CoW are not needed. Note that neither LocalVector<>
nor Vector<> are drop-in replacements for each other. They are two
unrelated types with similar interfaces, both using a buffer as their
storage strategy.
List<> is another Godot sequence type, using a doubly-linked list as
its storage strategy. Prefer Vector<> (or LocalVector<>) over
List<> unless you're sure you need it, as cache locality and memory
fragmentation tend to be more important with small collections.
References:
Containers
Godot provides also a set of common containers:
벡터
List
설정
맵
They aim to be as minimal as possible, as templates in C++ are often inlined and make the binary size much fatter, both in debug symbols and code. List, Set and Map can be iterated using pointers, like this:
for(List<int>::Element *E=somelist.front();E;E=E->next()) {
print_line(E->get()); // print the element
}
The Vector<> class also has a few nice features:
It does copy on write, so making copies of it is cheap as long as they are not modified.
It supports multi-threading, by using atomic operations on the reference counter.
References:
String
Godot also provides a String class. This class has a huge amount of features, full Unicode support in all the functions (like case operations) and utf8 parsing/extracting, as well as helpers for conversion and visualization.
References:
StringName
StringNames are like a String, but they are unique. Creating a StringName from a string results in a unique internal pointer for all equal strings. StringNames are useful for using strings as identifier, as comparing them is basically comparing a pointer.
Creation of a StringName (especially a new one) is slow, but comparison is fast.
References:
Math types
There are several linear math types available in the core/math directory.
References:
NodePath
This is a special datatype used for storing paths in a scene tree and referencing them fast.
References:
RID
RIDs are resource IDs. Servers use these to reference data stored in them. RIDs are opaque, meaning that the data they reference can't be accessed directly. RIDs are unique, even for different types of referenced data.