Compatibility with C¶

Syntax¶

There are however, some minor syntactic changes. For example, struct type names should not be prefixed with the keyword ‘struct’ (except in their definition):

struct ExampleType {
        int v;
};

// Invalid.
struct ExampleType function();

// Valid.
ExampleType function();

This syntactic change applies similarly for enum and union types. The intention, clearly, is to reduce the verbosity of the code and therefore the burden on the programmer to write it.

Casts also use a different syntax:

void function(){
        float pi = 3.141592654;

        // Invalid.
        int intPi = (int) pi;

        // Valid.
        int intPi = cast<int>(pi);
}

The new syntax is much clearer, helping readers to understand exactly what value is being cast, since the precedence of the type cast may not be obvious (few programmers remember all, or close to all, of their language’s operator precedences). Consider this example in C++:

void function(Type * ptr, Type::*function()){
        // As one expects...
        int i = (int) ptr->member;

        // ...is equivalent to this.
        int i = ((int) ptr->member);

        // But this...
        int i = (int) ptr->member + 1;

        // ...is equivalent to this.
        int i = ((int) ptr->member) + 1;

        // And this...
        int i = (int) ptr->*function();

        // ...would be equivalent to this.
        int i = ((int) ptr)->*function();

        // (Which is of course broken.)
}

Semantics¶

There are also some minor semantic changes, in particular to the available implicit casts:

void function(){
        float pi = 3.141592654f;

        // Invalid - no implicit cast.
        int intPi = pi;

        // Valid - explicit cast is allowed.
        int intPi = cast<int>(pi);

        // Invalid - no implicit cast from 'void *' to 'int *'.
        int * intPtr = malloc(sizeof(int));

        // Valid.
        int * intPtr = reinterpret_cast<int*>(malloc(sizeof(int)));
}

In this case, the intention is to make it clear to a reader when values are being cast to fundamentally different types (such as an integer and a floating point value). The ‘cast’ operator supports legal implicit and explicit casts, whereas the ‘reinterpret_cast’ operator allows casts between pointers whose target types (i.e. the type being pointed-to) are different.

Type Syntax¶

Loci follows the following structure for all typed variable declarations:

varDecl ::= TYPE NAME

This differs significantly from C, in which types and names can overlap in confusing ways, such as in function pointer types:

void (*f)(); // C
(*)(void)() f; // Loci

int (*f)(); // C
(*)(int)() f; // Loci

int (*f)(int, int); // C
(*)(int)(int, int) f; // Loci

Similarly, typedefs are much clearer in Loci:

typedef ReturnType (*NewName)(ParamType param); // C
using NewName = (*)(ReturnType)(ParamType); // Loci

Some types (such as structs and datatypes) also support pattern matching, with the following (rough) structure:

pattern ::= '_' /* wildcard */
pattern ::= varDecl
pattern ::= NAME '(' patternList ')'

nonEmptyPatternList ::= pattern
nonEmptyPatternList ::= nonEmptyPatternList ',' pattern

patternList ::= /* empty */
patternList ::= nonEmptyPatternList

This facilities code like the following:

struct Example {
        int x;
        int y;
};

void function(Example value) {
        Example(int x, _) = value;
}

Structs¶

As shown above, Loci supports C’s struct types. For example:

struct Point {
        int x;
        int y;
}

Like everything else, structs are actually just objects, the practical impact being that structs have:

• A default constructor: auto point = Point(1, 2);
• A default implicit copy method: auto point = other_point;
• A default compare method: if (point == other_point) { ... }
• Other obvious requirements, such as having alignment/size.

struct OpaqueStruct;

Enums¶

Loci supports C’s enums. For example:

enum Color {
        RED,
        GREEN,
        BLUE
}

This actually effectively builds an object type (in Loci all values are essentially objects, with some internal state and a set of methods). You can construct values using the automatically generated constructors:

Color function() {
        return Color.RED();
        // Or: Color::Red() if you prefer that.
}

Unions¶

Loci supports C’s unions, though as in C care must be taken with this particular feature. Here’s an example:

union IntOrFloat {
        int intValue;
        float floatValue;
}

In this case there is a default (‘create’) constructor that zero-initialises the union.

IntOrFloat function() {
        auto value = IntOrFloat();
        value.intValue = 100;
        return value;
}

This feature exists for compatibility with C and it is strongly advised in the vast majority of cases to use Algebraic Datatypes as a safer alternative.

Calling to/from C¶

All Loci functions/methods are generated to use the standard C calling convention on the target platform. This means it’s trivial to call to/from C code. For example:

struct CStruct {
        int value;
};

void doSomeProcessingInC(struct CStruct* data) {
        data->value = 42;
}

If this is some C code, then you can call into it from Loci with essentially identical code:

struct CStruct {
        int value;
};

void doSomeProcessingInC(CStruct* data) noexcept;

Note that the noexcept qualifier has been added to the function in Loci; this is not required but clearly represents the fact that the C function won’t throw an exception and this aids static analysis of exception safety in Loci.

namespace Namespace {
        void function();
}

void cfunction() {
        F2N9NamespaceN8function();
}

namespace Namespace {
        class TestClass {
                void method();

                // etc.
        }

        TestClass function();
}

void cfunction() {
        const size_t size = MT2N9NamespaceN9TestClassF1N8__sizeof();

        // Not needed in the case of heap allocation, but queried for completeness.
        const size_t alignMask = MT2N9NamespaceN9TestClassF1N11__alignmask();

        // A heap allocation isn't necessary - Loci-generated code uses
        // stack allocations - but it's simpler to demonstrate here.
        void* objectPointer = malloc(size);

        // The function will write to the given pointer.
        F2N9NamespaceN8function(objectPointer);

        // Call a method on the class.
        MT2N9NamespaceN9TestClassF1N6method(objectPointer);

        // Must call class destructor!
        MT2N9NamespaceN9TestClassF1N9__destroy(objectPointer);

        free(objectPointer);
}

C Strings¶

Loci supports C strings, which essentially just involves manipulating pointers to ubyte (the type char is renamed to byte and Loci treats ASCII character bytes as unsigned). For example:

size_t get_cstring_length(const ubyte* ptr) {
        size_t length = 0u;
        while (*ptr != 0u) {
                length++;
                ptr++;
        }
        return length;
}

Needless to say, it’s recommended to use standard library Strings rather than trying to manipulate C strings (an extremely error-prone process).

Note that C string literals must use the ‘C’ prefix or suffix:

// Prefix:
const size_t length = get_cstring_length(C"Hello world!");

// Suffix:
const size_t length = get_cstring_length("Hello world!"C);

Without the prefix or suffix Loci will try to find a function called ‘string_literal’, which conveniently gives std.string a hook to provide a standard library string when no prefix/suffix is specified.