Design of getargs()

[certik]

Let's create SymbolicGetArg(expr e, int n):

--- a/src/libasr/pass/intrinsic_function_registry.h
+++ b/src/libasr/pass/intrinsic_function_registry.h
@@ -84,6 +84,7 @@ enum class IntrinsicScalarFunctions : int64_t {
     SymbolicMulQ,
     SymbolicPowQ,
     SymbolicLogQ,
+    SymbolicGetArg,
     // ...
 };

Which will take the expression e, call get_args on it, and return the n-th argument as an expression.

The SymEngine's getargs are slow anyway, since it returns std::vector<Basic> and internally SymEngine does not store the arguments in a vector, and so it must construct it (slow). Consequently, the SymbolicGetArg will be slow as well. We can start by using basic_get_args, implemented in SymEngine as:

CWRAPPER_OUTPUT_TYPE basic_get_args(const basic self, CVecBasic *args)
{
    CWRAPPER_BEGIN
    args->m = self->m->get_args();
    CWRAPPER_END
}

and then extract the n-th argument using vecbasic_get, implemented as:

CWRAPPER_OUTPUT_TYPE vecbasic_get(CVecBasic *self, size_t n, basic result)
{                     
    CWRAPPER_BEGIN                                                         
                                    
    SYMENGINE_ASSERT(n < self->m.size());
    result->m = self->m[n];                            
                                                              
    CWRAPPER_END
}                 

[certik]

The other design is to introduce SymbolicGetArguments(expr e):

--- a/src/libasr/pass/intrinsic_function_registry.h
+++ b/src/libasr/pass/intrinsic_function_registry.h
@@ -84,6 +84,7 @@ enum class IntrinsicScalarFunctions : int64_t {
     SymbolicMulQ,
     SymbolicPowQ,
     SymbolicLogQ,
+    SymbolicGetArguments,
     // ...
 };
 

This function will return a list[S] of all the arguments at once. Then we use regular LPython mechanism for accessing a list.

This might be a better approach.

In fact this approach is equivalent to what we already designed in the past: #2332 (comment), so let's do that.

[anutosh491]

Closing as completed through #2399