Typechecking and code generation

TODO: Add notes here TODO: Talk about the queues that exist

Instructions

The process of code generation involves the production (creation) of instructions and consumption of them (consuming them and embedding them in other instructions). There are several types of instructions but the main important base ones are listed below.

The base Instruction

Every type of instruction that is produced during the code generation phase is a kind-of Instruction, it is the base class for all instructions and contains some common methods used by all of them:

1. setContext(Context)
   • Sets the Context object that is to be associated with this instruction.
   • This is normally done as a way to transfer the context from the respective parser-node to the corresponding instruction such that if such context is needed during further code generation (or even emit) it can then be accessed
2. Context getContext()
   • Returns this instruction’s associated context via its Context object
3. string produceToStrEnclose(string addInfo)
   • Returns a string containing the additional info provided through addInfo
   • The format of the returned string will be [Instruction: <className>: <addInfo>] where <className> is the name of the instruction type (kind-of) and <addInfo> as explained previously

Value-based instructions (Value)

TODO: Talk about the Value instruction base class here

A Value instruction is a kind-of Instruction of which represents code which generates some sort of value, think of literals, arithmetic operations, pointer dereferences, variable reads and so on. Every such instruction always has an associated Type object associated with it in order to know the intended type of the instruction. Below we show the API usage of the Value class:

1. Type getType()
   • Returns the type associated with this instruction
2. setType(Type)
   • Set the type to be associated with this instruction

There are many instructions which sub-type this Value class, these can be found in <TODO: Insert path here and put all Value-based instructions in their own module>.

Code generation

The method of code generation and type checking starts by being provided a so-called “action list” which is a linear array of dependency-nodes (or DNodes for code’s sake), this list is then iterated through by a for-loop, and each DNode is passed to a method called typeCheckThing(DNode):

foreach(DNode node; actionList)
{
    /* Type-check/code-gen this node */
    typeCheckThing(node);
}

The handling of every different instruction type and its associated typechecking requirements are handled in one huge if-statement within the typeCheckThing(DNode) method. This method will analyse a given dependency-node and perform the required typechecking by extracting the DNode’s emebedded parser-node, whilst doing so if a type check passes then code generation takes place by generating the corresponding instruction and adding this to some position in the code queue (discussed later).

Code queue

TODO: Add information on this

The code queue is used as a stack and a queue in order to facilitate instruction generation. Certain instructions are produced once off and then added to the back of the queue (“consuming” instructions) whilst other are produced and pushed onto the top of the queue (“producing” instructions) for consumption by other consuming instructions later.

An example of this would be the following T code which uses a binary operation with two operands (one being a LiteralValue instruction and the other being a FuncCall instruction):

                                1 + func()

This would result in a situation where we have the following production

Enforcement

Enforcement is the procedure of ensuring that a given Value-based instruction, \(instr_{i}\), conforms to the target type or “to-type”, \(type_{i}\). An optional flag can be passed such that if the \(typeof(instr_{i}) \neq type_{i}\) that it can then attempt coercion as to bring it to the equal type.

The method by which this is done is:

typeEnforce(Type toType,
            Value v2,
            ref Instruction coercedInstruction,
            bool allowCoercion = false)

We will discuss exact equality and exact equality through coercion in the next two sections.

Type equality

In order to check strict equality the type enforcer will initially check the following condition. We label the toType as \(t_{1}\) and the the type of v2 as \(typeof(v_{2})\) (otherwise referred to as \(t_{2}\)).

The method isSameType(Type t1, Type t2) provides exact quality checking between the two given types in the form of \(t_{1} = t_{2}\).

Coercion

In the case of coercion an application of \(coerce()\) is applied to the incoming instruction, as to produce an instruction \(coerceInstr_{i}\), a CastedValueInstruction, which wraps the original instruction inside of it but allows for a type cast/conversion to the target type, therefore making the statement, \(type_{i} = typeof(coerce(instr_{i}))\) (which is the same as \(type_{i} = typeof(coerceInstr_{i})\)), valid.

TODO: Document this now

Coercion has a set of rules (TODO: document them) in terms of what can be coerced. AT the end of the day if the coercion fails then a CoercionException is thrown, if, however it succeeds then a CastedValueInstruction will be placed into the memory location (the variable) pointed to by the ref parameter of typeEnforce(..., ..., ref Instruction coercedInstruction, true).

Example

Below we have an example of the code which processes variable declarations with assignments (think of byte i = 2):

Value assignmentInstr;
if(variablePNode.getAssignment())
{
  Instruction poppedInstr = popInstr();
  assert(poppedInstr);

  // Obtain the value instruction of the variable assignment
  // ... along with the assignment's type
  assignmentInstr = cast(Value)poppedInstr;
  assert(assignmentInstr);
  Type assignmentType = assignmentInstr.getInstrType();

  /** 
   * Here we can call the `typeEnforce` with the popped
   * `Value` instruction and the type to coerce to
   * (our variable's type)
   */
  typeEnforce(variableDeclarationType, assignmentInstr, assignmentInstr, true);
  assert(isSameType(variableDeclarationType, assignmentInstr.getInstrType())); // Sanity check
}

...

What the above code is doing is:

1. Firstly popping off an Instruction from the stack-queue and then down-casting it to Value (for Value-based instructions would be required as an expression is being assigned)
2. We then call typeEnforce() providing it with: * The variable’s type - the variableDeclarationType * The incoming Value-based instruction assignmentInstr * The third argument, is ref-based, meaning what we provide it is the variable which will have the result of the enforcement (if coercion is required) placed into * The last argument is true, meaning “Please attempt coercion if the types are not exactly equal, please”

The last line containing an assertion:

assert(isSameType(
        variableDeclarationType,
        assignmentInstr.getInstrType()
        ); // Sanity check

This is a sanity check, as if the type coercion failed then an exception would be thrown and the assertion would not be reached, however if the types were an exact match or if they were not but could be coerced as such then the two types should match.

Variable referencing counting

Firstly let me make it clear that this has nothing to do with runtime reference counting but rather a simple mechanism used to maintain a count or number of references to variables after their declaration.

Below is a method table of the methods of concern:

This aids us in implementing a single feature unused variable detection. It’s rather simple, reference counts are incremented by using a touch(Variable) method defined in the TypeChecker and this is called whilst doing dependency generation in the dependency generator.

The first time a variable is encountered, such as even its declaration, we will then touch(...)-it. At the end of type checking we then call the getUnusedVariables() method which returns a list of the undeclared variables. These are variables with a reference count higher than 1. We then print these out so the user can see which are unused.

Usage

Example usage below shows us touch-ing a variable when we process them in expressions such as a VariableExpression in the dependency module:

...

/* Get the entity as a Variable */
Variable variable = cast(Variable)namedEntity;

/* Variable reference count must increase */
tc.touch(variable);

...

We then, after typechecking, run the following in the type checker module’s doPostChecks() method:

/** 
 * Find the variables which were declared but never used
 */
if(this.config.hasConfig("typecheck:warnUnusedVars") & this.config.getConfig("typecheck:warnUnusedVars").getBoolean())
{
        Variable[] unusedVariables = getUnusedVariables();
        gprintln("There are "~to!(string)(unusedVariables.length)~" unused variables");
        if(unusedVariables.length)
        {
                foreach(Variable unusedVariable; unusedVariables)
                {
                        // TODO: Get a nicer name, full path-based
                        gprintln("Variable '"~to!(string)(unusedVariable.getName())~"' is declared but never");
                }
        }
}