generic-arrays

[horasal]

Description

This pull request allow the following code work:

array := GenericClass<Int>[10] new()

In detail, the above code will be unwrapped to:

array := Array new(GenericClass*, 10)

Implementation Details

In current rock, only ArrayAccess is automatically unwrapped to creation. However, GenericClass<TypeArg>[Expression] will be parsed as a TypeAccess of ArrayType. It is clearly that calling methods on ArrayType can not get anything. Thus, before FunctionCall starts resolve, TypeAccess should be turned into an array creation.

Here the unwrap process happens before resolving expr. This is because that we only need to know the type of expr and inner, which can be judged by instanceOf?.

[horasal]

Test, please notice that ExampleClass<Int>[10] new() is just for creating the array, where its elements remain un-initialized:

ExampleClass: class <T> {
    member: T
    totalCount: static Int = 0
    count: Int
    init: func {
        This totalCount += 1
        this count = This totalCount
    }
    increase: func -> Int {
        this count += 1
        this count
    }
}

array := ExampleClass<Int>[10] new()
for (i in 0 .. 10) {
    array[i] = array[i] new()
    array[i] increase()
    raise(array[i] count != i + 2, "Element at %d had count=%d" format(i, array[i] count))
}
array free()

[thomasfanell]

I would expect this to work:

GenericClass: class <T> {
    value: T
    init: func (=value)
}

arrayOfGenericClass := GenericClass<Int>[5] new()
// Initialize array elements
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] = GenericClass<Int> new(i + 1)
for (i in 0 .. arrayOfGenericClass length)
    "%d" printfln(arrayOfGenericClass[i] value)
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] free()
arrayOfGenericClass free()

/*
Output:
  5
  -241166464
  -241166464
  -241166464
  -241166464
*/

If I instead print the value immediately after initializing the array item (in the first loop), it works as expected.

[horasal]

Compiling the following code with magic-lang/rock:master gives the same error. I think it is not a bug introduced by this pull request.

Anyway, I will take a look at it.

GenericClass: class <T> {
    value: T
    init: func (=value)
}

arrayOfGenericClass := [GenericClass<Int> new(1),GenericClass<Int> new(2),GenericClass<Int> new(3),GenericClass<Int> new(4),GenericClass<Int> new(5)]
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] = GenericClass<Int> new(i + 1)
for (i in 0 .. arrayOfGenericClass length){
    arrayOfGenericClass[i] value toString() println()
}
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] free()
arrayOfGenericClass free()

/*
Output:
  5
  -241166464
  -241166464
  -241166464
  -241166464
*/

[horasal]

Can also be reproduced by this

GenericClass: class <T> {
    value: T
    init: func (=value)
}

test1 := GenericClass<Int> new(1)
test2 := GenericClass<Int> new(2)

printf("%d\n", test1 value)
printf("%d\n", test2 value)

[alexnask]

I think this is because alloca is used to initialize the memory of the generic value (here named value) inside the class's defaults, so it just gets free'd immediately...

void test__GenericClass___defaults___impl(test__GenericClass* this) {
    types__Class___defaults___impl((types__Class*) this);
    this->value = Memory__alloca(this->T->size);
}

Probably should malloc and have some notion of a free function in the language, which would have calls to free for the generic values.

[alexnask]

I think a good solution would be to auto-generate free functions that do cleanup like generic values or a Closure's context etc. and then call the user defined __destroy__ function (which I think was originally going to be a finalizer but was never implemented).

[horasal]

@shamanas Good idea, It sounds good to add cleanup sections for every block (functiondecl, classdecl, etc.) to free the generic/closure. I will try to implement it.

@thomasfanell This reminds me that ooc-kean uses the __onheap__ property to allocate the memory on the heap. So I guess you may want to add a __onheap__ before T?

GenericClass: class <T> {
    value: __onheap__ T
    init: func (=value)
}

arrayOfGenericClass := GenericClass<Int>[5] new()
// Initialize array elements
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] = GenericClass<Int> new(i + 1)
for (i in 0 .. arrayOfGenericClass length)
    "%d" printfln(arrayOfGenericClass[i] value)
for (i in 0 .. arrayOfGenericClass length)
    arrayOfGenericClass[i] free()
arrayOfGenericClass free()

[alexnask]

@zhaihj While I do think it's the way to go, I think it should happen in another issue/PR and everyone needs to agree with the way it will be done, since it will be a breaking change.

The way I see it, every cover/class decl gets an autogenerated free function and the current free functions are just renamed to __destroy__ (still calling free on their children etc.)

EDIT: rock could also detect if a free function is already manually declared and not generate one itself, that way a lot of code is backwards compatible.

EDIT2: I may fork a version of rock with radical changes, perhaps give ownership and some form of light-RAII a shot, I always wondered what ooc would look like with some of those features and GC off.

[thomasfanell]

@zhaihj Conflicts

[horasal]

rebase to master

[horasal]

Oops, it seems that I merged an un-related branch. rebased to master again

[thomasfanell]

Merged via #61