nac3core/src/codegen/mod.rs

@@ -77,7 +77,8 @@ pub struct CodeGenContext<'ctx, 'a> {
     pub const_strings: HashMap<String, BasicValueEnum<'ctx>>,
     // stores the alloca for variables
     pub init_bb: BasicBlock<'ctx>,
-    // the first one is the test_bb, and the second one is bb after the loop
+    /// The header and exit basic blocks of a loop in this context. See
+    /// https://llvm.org/docs/LoopTerminology.html for explanation of these terminology.
     pub loop_target: Option<(BasicBlock<'ctx>, BasicBlock<'ctx>)>,
     // unwind target bb
     pub unwind_target: Option<BasicBlock<'ctx>>,

nac3core/src/codegen/stmt.rs

@@ -13,8 +13,8 @@ use inkwell::{
     attributes::{Attribute, AttributeLoc},
     basic_block::BasicBlock,
     types::BasicTypeEnum,
-    values::{BasicValue, BasicValueEnum, FunctionValue, PointerValue},
+    values::{BasicValue, BasicValueEnum, FunctionValue, IntValue, PointerValue},
-    IntPredicate::EQ,
+    IntPredicate,
 };
 use nac3parser::ast::{
     Constant, ExcepthandlerKind, Expr, ExprKind, Location, Stmt, StmtKind, StrRef,
@@ -107,7 +107,7 @@ pub fn gen_store_target<'ctx, 'a, G: CodeGenerator>(
             );
             // handle negative index
             let is_negative = ctx.builder.build_int_compare(
-                inkwell::IntPredicate::SLT,
+                IntPredicate::SLT,
                 raw_index,
                 generator.get_size_type(ctx.ctx).const_zero(),
                 "is_neg",
@@ -120,7 +120,7 @@ pub fn gen_store_target<'ctx, 'a, G: CodeGenerator>(
             // unsigned less than is enough, because negative index after adjustment is
             // bigger than the length (for unsigned cmp)
             let bound_check = ctx.builder.build_int_compare(
-                inkwell::IntPredicate::ULT,
+                IntPredicate::ULT,
                 index,
                 len,
                 "inbound",
@@ -193,7 +193,13 @@ pub fn gen_assign<'ctx, 'a, G: CodeGenerator>(
             }
         }
         _ => {
-            let ptr = generator.gen_store_target(ctx, target, Some("target.addr"))?;
+            let name = if let ExprKind::Name { id, .. } = &target.node {
+                format!("{}.addr", id.to_string())
+            } else {
+                String::from("target.addr")
+            };
+            let ptr = generator.gen_store_target(ctx, target, Some(name.as_str()))?;
+
             if let ExprKind::Name { id, .. } = &target.node {
                 let (_, static_value, counter) = ctx.var_assignment.get_mut(id).unwrap();
                 *counter += 1;
@@ -208,6 +214,26 @@ pub fn gen_assign<'ctx, 'a, G: CodeGenerator>(
     Ok(())
 }
 
+/// Generates a sequence of IR which checks whether [value] does not exceed the upper bound of the
+/// range as defined by [stop] and [step].
+///
+/// Note that the generated IR will **not** check whether value is part of the range or whether
+/// value exceeds the lower bound of the range (as evident by the missing `start` argument).
+///
+/// Returns an [IntValue] representing the result of whether the [value] is in the range.
+fn gen_in_range_check<'ctx, 'a>(
+    ctx: &CodeGenContext<'ctx, 'a>,
+    value: IntValue<'ctx>,
+    stop: IntValue<'ctx>,
+    step: IntValue<'ctx>,
+) -> IntValue<'ctx> {
+    let sign = ctx.builder.build_int_compare(IntPredicate::SGT, step, ctx.ctx.i32_type().const_zero(), "");
+    let lo = ctx.builder.build_select(sign, value, stop, "").into_int_value();
+    let hi = ctx.builder.build_select(sign, stop, value, "").into_int_value();
+
+    ctx.builder.build_int_compare(IntPredicate::SLT, lo, hi, "cmp")
+}
+
 pub fn gen_for<'ctx, 'a, G: CodeGenerator>(
     generator: &mut G,
     ctx: &mut CodeGenContext<'ctx, 'a>,
@@ -221,60 +247,82 @@ pub fn gen_for<'ctx, 'a, G: CodeGenerator>(
         let int32 = ctx.ctx.i32_type();
         let size_t = generator.get_size_type(ctx.ctx);
         let zero = int32.const_zero();
-        let current = ctx.builder.get_insert_block().unwrap().get_parent().unwrap();
+        let current = ctx.builder.get_insert_block().and_then(|bb| bb.get_parent()).unwrap();
-        let test_bb = ctx.ctx.append_basic_block(current, "test");
+        let body_bb = ctx.ctx.append_basic_block(current, "for.body");
-        let body_bb = ctx.ctx.append_basic_block(current, "body");
+        let cont_bb = ctx.ctx.append_basic_block(current, "for.end");
-        let cont_bb = ctx.ctx.append_basic_block(current, "cont");
         // if there is no orelse, we just go to cont_bb
         let orelse_bb =
-            if orelse.is_empty() { cont_bb } else { ctx.ctx.append_basic_block(current, "orelse") };
+            if orelse.is_empty() { cont_bb } else { ctx.ctx.append_basic_block(current, "for.orelse") };
+
+        // Whether the iterable is a range() expression
+        let is_iterable_range_expr = ctx.unifier.unioned(iter.custom.unwrap(), ctx.primitives.range);
+
+        // The target BB of the loop backedge
+        let backedge_bb_target = if is_iterable_range_expr {
+            body_bb
+        } else {
+            ctx.ctx.append_basic_block(current, "for.cond")
+        };
+
         // store loop bb information and restore it later
-        let loop_bb = ctx.loop_target.replace((test_bb, cont_bb));
+        let loop_bb = ctx.loop_target.replace((backedge_bb_target, cont_bb));
 
         let iter_val = generator.gen_expr(ctx, iter)?.unwrap().to_basic_value_enum(
             ctx,
             generator,
             iter.custom.unwrap(),
         )?;
-        if ctx.unifier.unioned(iter.custom.unwrap(), ctx.primitives.range) {
+        if is_iterable_range_expr {
-            // setup
             let iter_val = iter_val.into_pointer_value();
+            // Internal variable for loop; Cannot be assigned
             let i = generator.gen_var_alloc(ctx, int32.into(), Some("for.i.addr"))?;
-            let user_i = generator.gen_store_target(ctx, target, Some("for.user_i.addr"))?;
+            // Variable declared in "target" expression of the loop; Can be reassigned *or* shadowed
-            let (start, end, step) = destructure_range(ctx, iter_val);
+            let target_i = generator.gen_store_target(ctx, target, Some("for.target.addr"))?;
-            ctx.builder.build_store(i, ctx.builder.build_int_sub(start, step, "start_init"));
+            let (start, stop, step) = destructure_range(ctx, iter_val);
-            ctx.builder.build_unconditional_branch(test_bb);
+
-            ctx.builder.position_at_end(test_bb);
+            ctx.builder.build_store(i, start);
-            let sign = ctx.builder.build_int_compare(
+
-                inkwell::IntPredicate::SGT,
+            // Check "If step is zero, ValueError is raised."
-                step,
+            let rangenez = ctx.builder.build_int_compare(IntPredicate::NE, step, int32.const_zero(), "");
-                int32.const_zero(),
+            ctx.make_assert(
-                "sign",
+                generator,
+                rangenez,
+                "ValueError",
+                "range() arg 3 must not be zero",
+                [None, None, None],
+                ctx.current_loc
             );
-            // add and test
+
-            let tmp = ctx.builder.build_int_add(
-                ctx.builder.build_load(i, "i").into_int_value(),
-                step,
-                "start_loop",
-            );
-            ctx.builder.build_store(i, tmp);
-            ctx.builder.build_store(user_i, tmp);
-            // // if step > 0, continue when i < end
-            let cmp1 = ctx.builder.build_int_compare(inkwell::IntPredicate::SLT, tmp, end, "cmp1");
-            // if step < 0, continue when i > end
-            let cmp2 = ctx.builder.build_int_compare(inkwell::IntPredicate::SGT, tmp, end, "cmp2");
-            let pos = ctx.builder.build_and(sign, cmp1, "pos");
-            let neg = ctx.builder.build_and(ctx.builder.build_not(sign, "inv"), cmp2, "neg");
             ctx.builder.build_conditional_branch(
-                ctx.builder.build_or(pos, neg, "or"),
+                gen_in_range_check(ctx, start, stop, step),
                 body_bb,
                 orelse_bb,
             );
+
             ctx.builder.position_at_end(body_bb);
+            ctx.builder.build_store(target_i, ctx.builder.build_load(i, "").into_int_value());
+            gen_block(generator, ctx, body.iter())?;
+
+            // Test if next element is still in range
+            let next_i = ctx.builder.build_int_add(
+                ctx.builder.build_load(i, "").into_int_value(),
+                step,
+                "next_i",
+            );
+            let cond_cont_bb = ctx.ctx.append_basic_block(current, "for.cond.cont");
+            ctx.builder.build_conditional_branch(
+                gen_in_range_check(ctx, next_i, stop, step),
+                cond_cont_bb,
+                orelse_bb,
+            );
+
+            ctx.builder.position_at_end(cond_cont_bb);
+            ctx.builder.build_store(i, next_i);
         } else {
-            let counter = generator.gen_var_alloc(ctx, size_t.into(), Some("for.counter.addr"))?;
+            let test_bb = backedge_bb_target;
-            // counter = -1
+
-            ctx.builder.build_store(counter, size_t.const_int(u64::max_value(), true));
+            let index_addr = generator.gen_var_alloc(ctx, size_t.into(), Some("for.index.addr"))?;
+            ctx.builder.build_store(index_addr, size_t.const_zero());
             let len = ctx
                 .build_gep_and_load(
                     iter_val.into_pointer_value(),
@@ -282,30 +330,36 @@ pub fn gen_for<'ctx, 'a, G: CodeGenerator>(
                 )
                 .into_int_value();
             ctx.builder.build_unconditional_branch(test_bb);
+
             ctx.builder.position_at_end(test_bb);
-            let tmp = ctx.builder.build_load(counter, "i").into_int_value();
+            let index = ctx.builder.build_load(index_addr, "for.index").into_int_value();
-            let tmp = ctx.builder.build_int_add(tmp, size_t.const_int(1, false), "inc");
+            let cmp = ctx.builder.build_int_compare(IntPredicate::SLT, index, len, "cond");
-            ctx.builder.build_store(counter, tmp);
-            let cmp = ctx.builder.build_int_compare(inkwell::IntPredicate::SLT, tmp, len, "cmp");
             ctx.builder.build_conditional_branch(cmp, body_bb, orelse_bb);
+
             ctx.builder.position_at_end(body_bb);
             let arr_ptr = ctx
                 .build_gep_and_load(iter_val.into_pointer_value(), &[zero, zero])
                 .into_pointer_value();
-            let val = ctx.build_gep_and_load(arr_ptr, &[tmp]);
+            let val = ctx.build_gep_and_load(arr_ptr, &[index]);
             generator.gen_assign(ctx, target, val.into())?;
+            gen_block(generator, ctx, body.iter())?;
+
+            let index = ctx.builder.build_load(index_addr, "for.index").into_int_value();
+            let inc = ctx.builder.build_int_add(index, size_t.const_int(1, true), "");
+            ctx.builder.build_store(index_addr, inc);
         }
 
-        gen_block(generator, ctx, body.iter())?;
         for (k, (_, _, counter)) in var_assignment.iter() {
             let (_, static_val, counter2) = ctx.var_assignment.get_mut(k).unwrap();
             if counter != counter2 {
                 *static_val = None;
             }
         }
+
         if !ctx.is_terminated() {
-            ctx.builder.build_unconditional_branch(test_bb);
+            ctx.builder.build_unconditional_branch(backedge_bb_target);
         }
+
         if !orelse.is_empty() {
             ctx.builder.position_at_end(orelse_bb);
             gen_block(generator, ctx, orelse.iter())?;
@@ -313,12 +367,14 @@ pub fn gen_for<'ctx, 'a, G: CodeGenerator>(
                 ctx.builder.build_unconditional_branch(cont_bb);
             }
         }
+
         for (k, (_, _, counter)) in var_assignment.iter() {
             let (_, static_val, counter2) = ctx.var_assignment.get_mut(k).unwrap();
             if counter != counter2 {
                 *static_val = None;
             }
         }
+
         ctx.builder.position_at_end(cont_bb);
         ctx.loop_target = loop_bb;
     } else {
@@ -338,12 +394,12 @@ pub fn gen_while<'ctx, 'a, G: CodeGenerator>(
         let var_assignment = ctx.var_assignment.clone();
 
         let current = ctx.builder.get_insert_block().unwrap().get_parent().unwrap();
-        let test_bb = ctx.ctx.append_basic_block(current, "test");
+        let test_bb = ctx.ctx.append_basic_block(current, "while.test");
-        let body_bb = ctx.ctx.append_basic_block(current, "body");
+        let body_bb = ctx.ctx.append_basic_block(current, "while.body");
-        let cont_bb = ctx.ctx.append_basic_block(current, "cont");
+        let cont_bb = ctx.ctx.append_basic_block(current, "while.cont");
         // if there is no orelse, we just go to cont_bb
         let orelse_bb =
-            if orelse.is_empty() { cont_bb } else { ctx.ctx.append_basic_block(current, "orelse") };
+            if orelse.is_empty() { cont_bb } else { ctx.ctx.append_basic_block(current, "while.orelse") };
         // store loop bb information and restore it later
         let loop_bb = ctx.loop_target.replace((test_bb, cont_bb));
         ctx.builder.build_unconditional_branch(test_bb);
@@ -401,15 +457,15 @@ pub fn gen_if<'ctx, 'a, G: CodeGenerator>(
         let var_assignment = ctx.var_assignment.clone();
 
         let current = ctx.builder.get_insert_block().unwrap().get_parent().unwrap();
-        let test_bb = ctx.ctx.append_basic_block(current, "test");
+        let test_bb = ctx.ctx.append_basic_block(current, "if.test");
-        let body_bb = ctx.ctx.append_basic_block(current, "body");
+        let body_bb = ctx.ctx.append_basic_block(current, "if.body");
         let mut cont_bb = None;
         // if there is no orelse, we just go to cont_bb
         let orelse_bb = if orelse.is_empty() {
-            cont_bb = Some(ctx.ctx.append_basic_block(current, "cont"));
+            cont_bb = Some(ctx.ctx.append_basic_block(current, "if.cont"));
             cont_bb.unwrap()
         } else {
-            ctx.ctx.append_basic_block(current, "orelse")
+            ctx.ctx.append_basic_block(current, "if.orelse")
         };
         ctx.builder.build_unconditional_branch(test_bb);
         ctx.builder.position_at_end(test_bb);
@@ -844,7 +900,7 @@ pub fn gen_try<'ctx, 'a, G: CodeGenerator>(
                     .builder
                     .build_load(exn_type.into_pointer_value(), "expected_id")
                     .into_int_value();
-                let result = ctx.builder.build_int_compare(EQ, actual_id, expected_id, "exncheck");
+                let result = ctx.builder.build_int_compare(IntPredicate::EQ, actual_id, expected_id, "exncheck");
                 ctx.builder.build_conditional_branch(result, handler_bb, dispatcher_cont);
                 dispatcher_end = dispatcher_cont;
             } else {

nac3standalone/demo/src/loop.py

@@ -1,9 +1,12 @@
+# For Loop using an increasing range() expression as its iterable
+
 @extern
 def output_int32(x: int32):
     ...
 
 def run() -> int32:
-    for _ in range(10):
+    i = 0
-        output_int32(_)
+    for i in range(10):
-        _ = 0
+        output_int32(i)
+    output_int32(i)
     return 0

nac3standalone/demo/src/loop_decr.py

@@ -0,0 +1,12 @@
+# For Loop using a decreasing range() expression as its iterable
+
+@extern
+def output_int32(x: int32):
+    ...
+
+def run() -> int32:
+    i = 0
+    for i in range(10, 0, -1):
+        output_int32(i)
+    output_int32(i)
+    return 0

nac3standalone/demo/src/loop_iterable.py

@@ -0,0 +1,17 @@
+# For Loop using a list as its iterable
+
+@extern
+def output_int32(x: int32):
+    ...
+
+def run() -> int32:
+    l = [0, 1, 2, 3, 4]
+
+    # i: int32  # declaration-without-initializer not yet supported
+    i = 0       # i must be declared before the loop; this is not necessary in Python
+    for i in l:
+        output_int32(i)
+        i = 0
+        output_int32(i)
+    output_int32(i)
+    return 0

nac3standalone/demo/src/loop_mutate_var.py

@@ -0,0 +1,14 @@
+# For Loop using an range() expression as its iterable, additionally reassigning the target on each iteration
+
+@extern
+def output_int32(x: int32):
+    ...
+
+def run() -> int32:
+    i = 0
+    for i in range(10):
+        output_int32(i)
+        i = 0
+        output_int32(i)
+    output_int32(i)
+    return 0