Implement Set

[namannimmo10]

No description provided.

[certik]

We should start with extending our type system with Set and Dict, and then go from there. See above.

[namannimmo10]

ASR:

               test_Set [] [
               (=
                  (Var 2 a)
                  (ConstantSet [(ConstantInteger 1
                     (Integer 4 [])) (ConstantInteger 2
                     (Integer 4 [])) (ConstantInteger 3
                     (Integer 4 []))]
                     (Set
                        (Integer 4 []))) ())
               (=
                  (Var 2 a)
                  (ConstantSet [(ConstantInteger 2
                     (Integer 4 [])) (ConstantInteger 3
                     (Integer 4 [])) (ConstantInteger 4
                     (Integer 4 [])) (ConstantInteger 5
                     (Integer 4 [])) (ConstantInteger 5
                     (Integer 4 []))]
                     (Set
                        (Integer 4 []))) ())
               (=
                  (Var 2 a)
                  (ConstantSet [
                     (ConstantString "a"
                        (Character 1 1 () []))
                     (ConstantString "b"
                        (Character 1 1 () []))
                     (ConstantString "c"
                        (Character 1 1 () []))]
                     (Set
                        (Character 1 1 () []))) ())
               (=
                  (Var 2 b)
                  (ArrayRef 2 a [
                     (() (ConstantInteger 0
                        (Integer 4 [])) ())]
                     (Integer 4 []) ()) ())]

[namannimmo10]

However, the type of ConstantSet is not the type of the element (say i32), but rather a new type Set<i32>

Seems like this is true in the above-produced ASR.

[certik]

All these should give an error message that "Set[i32] (or character) cannot be assigned to i32".

Then let's add tests like:

a: Set[i32]
a = {1, 2, 3}

And we need to improve the assignment to do this check.

[certik]

Yes, this looks very good. The RHS seems correct in ASR. The Assignment is incorrect, as you cannot assign a set to an integer, see my comment above. You can only assign a set to a set.

[namannimmo10]

The Assignment is incorrect, as you cannot assign a set to an integer, see my comment above. You can only assign a set to a set.

I know, just I was planning to add the set annotation.

[certik]

Here are examples of typing: https://mypy.readthedocs.io/en/stable/cheat_sheet_py3.html#built-in-types, let's use the same syntax

[namannimmo10]

Let's move the improvement of the assignment to a separate PR?

I know many tests will fail once I apply this diff:

diff --git a/src/lpython/semantics/python_ast_to_asr.cpp b/src/lpython/semantics/python_ast_to_asr.cpp
index bc7bb209..72fe972d 100644
--- a/src/lpython/semantics/python_ast_to_asr.cpp
+++ b/src/lpython/semantics/python_ast_to_asr.cpp
@@ -515,7 +515,11 @@ public:
         }
         this->visit_expr(*x.m_value);
         ASR::expr_t *value = ASRUtils::EXPR(tmp);
-        ASR::stmt_t *overloaded=nullptr;
+        ASR::stmt_t *overloaded = nullptr;
+        if (!ASRUtils::check_equal_type(ASRUtils::expr_type(target), ASRUtils::expr_type(value))) {
+            throw SemanticError("Assignment types do not match",
+                x.base.base.loc);
+        }
         tmp = ASR::make_Assignment_t(al, x.base.base.loc, target, value,
                                 overloaded);
     }

[certik]

We can't merge this PR anyway until we run all tests in LFortran, so we have to do that.

[namannimmo10]

until we run all tests in LFortran

I did so -- no test failures.

[certik]

Awesome. Would you mind please sending a Merge Request against LFortran with this change in ASR? Let's get it merged there first. Then we'll come back and finish this PR.

[namannimmo10]

MR opened: !1650.

[certik]

The same comment as for the dictionary. I think this is ok to merge.

The ConstantSet creation is correct. But declaring a set and assignment to it does not generate a correct ASR yet, but we can fix that later.