I have a Matrix class, and it would be really convenient to be able to initialize the matrix using an initialization list.

For example, the following would produce a 2x2 Matrix.

[source]Matrix M = { {1,2}, {3,4} };[/source]

It appears that a multidimensional initialization list can only be used on template types, because it relies on a subtype.

"as_compiler.cpp" around line 2540...

[source]void asCCompiler::CompileInitList

if( el->nodeType == snAssignment )

{

// Compile the assignment expression

CompileAssignment(el, &rctx);

}

else if( el->nodeType == snInitList )

{

int offset = AllocateVariable(var->dataType.GetSubType(), true)[/source]

I don't see any way to initialize my Matrix, which is registered as follows, with an initialization list.

[source]engine->RegisterObjectType("Matrix", 0, asOBJ_REF);[/source]

I saw the following source comment:

"as_compiler.cpp"

[source] // TODO: initlist: Should we have a special indexing operator for this? How can we support

// initialization lists with different types for different elements? Maybe

// by using the variable arguments the initialization can be done with one

// call, passing all the elements as arguments. The registered function can

// then traverse them however it wants.[/source]

Perhaps a nested initialization list could take advantage of a similar (or the same) solution. If some sort of generic, variable argument function could be called which passed in all the arguments at once, then a nested initialization list could appear as another variable argument list.

It is likely that you were already thinking of this, and I didn't check the entire forum to see if anybody else had expressed a desire for a similar feature. In any case, mark me down as wanting this feature

I think something like this was mentioned a while back, but it wasn't decided for or against. I'd also be interested in this feature, especially for initializing Dictionary types with key/value pairs.

I definitely plan to allow arbitrary objects to be initialized through initialization lists. It's not a trivial thing to implement though, and I currently have no estimate when I'll get to this.

I want to allow the application to register the different patterns that the initialization list must follow, so that the compiler can do most of the validation rather than the object doing it at run-time. E.g the Matrix class could be registered as taking only initialization lists with 4 float values, and the dictionary class can be registered to take lists with even number of values, and all even values must be of string type.

There are a lot of more things to think about. For example, the matrix type should preferably receive a pointer to the memory buffer rather than the index operator being called 4 times to set each element. The dictionary class on the other hand should receive each key/value pair together in the same call.

Do you have any thoughts on a registration string for an initialization list? If you have, feel free to disregard this post.

Perhaps something like this:

(This isn't perfect. I'm just thinking as I'm typing.)

reg ::= '{' pattern '}'

pattern ::= reg | def [, def]

def ::= type | type var | type '[' var ']' | expr '?' def ':' def

type ::= '?' | 'int' | 'double' | 'string' ... etc

var ::= (rules for C Variables or digits)

expr ::= (rules for C conditional expressions)

(the '...'var syntax could store the number of repeats encountered, and ensure the same count)

Examples:

{ ?[] } == list of variable types

{ int[] } == list of integers

{ int n, double[n] } == list of an integer (n) followed by 'n' doubles

{ {string, string}[] } == list of string pairs

{ {double[...n]}[] } == list of the same number of doubles

Or, instead of an array-like syntax, perhaps a variable-argument-like syntax:

{ ?, ... }

{ int, ... }

{ int n, double, n } <-- this one might be an integer followed by n+1 doubles

{ {string, string}, ... }

{ {double, ...n}, ... }

Obviously it could get complicated. I would be willing to help in any way I could.

I haven't decided exactly how the application would define the patterns. At first I'll probably just provide a few different options, as it is easier to control. But eventually I may have to fully generic if the need arises.