diff --git a/Cargo.lock b/Cargo.lock
index fbdb3ba093..22a6dcae25 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -3918,9 +3918,9 @@ dependencies = [
 
 [[package]]
 name = "target-lexicon"
-version = "0.10.0"
+version = "0.12.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ab0e7238dcc7b40a7be719a25365910f6807bd864f4cce6b2e6b873658e2b19d"
+checksum = "d9bffcddbc2458fa3e6058414599e3c838a022abae82e5c67b4f7f80298d5bff"
 
 [[package]]
 name = "tempfile"
diff --git a/cli/Cargo.toml b/cli/Cargo.toml
index 8288d26d6f..4d8c6bc4b4 100644
--- a/cli/Cargo.toml
+++ b/cli/Cargo.toml
@@ -67,7 +67,7 @@ libc = "0.2"
 libloading = "0.6"
 
 inkwell = { path = "../vendor/inkwell", optional = true }
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 tempfile = "3.1.0"
 
 [dev-dependencies]
diff --git a/cli/src/repl/eval.rs b/cli/src/repl/eval.rs
index d97d8087c6..c3792a1153 100644
--- a/cli/src/repl/eval.rs
+++ b/cli/src/repl/eval.rs
@@ -222,7 +222,8 @@ fn jit_to_ast_help<'a>(
                     let tags_map: roc_collections::all::MutMap<_, _> =
                         tags_vec.iter().cloned().collect();
 
-                    let union_variant = union_sorted_tags_help(env.arena, tags_vec, None, env.subs);
+                    let union_variant =
+                        union_sorted_tags_help(env.arena, tags_vec, None, env.subs, env.ptr_bytes);
 
                     let size = layout.stack_size(env.ptr_bytes);
                     use roc_mono::layout::WrappedVariant::*;
@@ -886,7 +887,8 @@ fn byte_to_ast<'a>(env: &Env<'a, '_>, value: u8, content: &Content) -> Expr<'a>
                         .map(|(a, b)| (a.clone(), b.to_vec()))
                         .collect();
 
-                    let union_variant = union_sorted_tags_help(env.arena, tags_vec, None, env.subs);
+                    let union_variant =
+                        union_sorted_tags_help(env.arena, tags_vec, None, env.subs, env.ptr_bytes);
 
                     match union_variant {
                         UnionVariant::ByteUnion(tagnames) => {
diff --git a/cli/tests/cli_run.rs b/cli/tests/cli_run.rs
index 0fb00a9e48..c201bc009e 100644
--- a/cli/tests/cli_run.rs
+++ b/cli/tests/cli_run.rs
@@ -223,13 +223,13 @@ mod cli_run {
         //     expected_ending: "",
         //     use_valgrind: true,
         // },
-        // cli:"cli" => Example {
-        //     filename: "Echo.roc",
-        //     executable_filename: "echo",
-        //     stdin: &["Giovanni\n", "Giorgio\n"],
-        //     expected_ending: "Giovanni Giorgio!\n",
-        //     use_valgrind: true,
-        // },
+        cli:"cli" => Example {
+            filename: "Echo.roc",
+            executable_filename: "echo",
+            stdin: &["Giovanni\n", "Giorgio\n"],
+            expected_ending: "Hi, Giovanni Giorgio!\n",
+            use_valgrind: true,
+        },
         // custom_malloc:"custom-malloc" => Example {
         //     filename: "Main.roc",
         //     executable_filename: "custom-malloc-example",
diff --git a/compiler/build/Cargo.toml b/compiler/build/Cargo.toml
index dcddc88461..96e91511c3 100644
--- a/compiler/build/Cargo.toml
+++ b/compiler/build/Cargo.toml
@@ -30,7 +30,7 @@ libloading = "0.6"
 tempfile = "3.1.0"
 serde_json = "1.0"
 inkwell = { path = "../../vendor/inkwell", optional = true }
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 
 [dev-dependencies]
 pretty_assertions = "0.5.1"
diff --git a/compiler/gen_dev/Cargo.toml b/compiler/gen_dev/Cargo.toml
index 40f35682f7..07b85fd205 100644
--- a/compiler/gen_dev/Cargo.toml
+++ b/compiler/gen_dev/Cargo.toml
@@ -21,7 +21,7 @@ roc_mono = { path = "../mono" }
 im    = "14" # im and im-rc should always have the same version!
 im-rc = "14" # im and im-rc should always have the same version!
 bumpalo = { version = "3.6.1", features = ["collections"] }
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 libloading = "0.6"
 object = { version = "0.24", features = ["write"] }
 
diff --git a/compiler/gen_llvm/Cargo.toml b/compiler/gen_llvm/Cargo.toml
index 2706ee35eb..eaec8982bd 100644
--- a/compiler/gen_llvm/Cargo.toml
+++ b/compiler/gen_llvm/Cargo.toml
@@ -22,7 +22,7 @@ im    = "14" # im and im-rc should always have the same version!
 im-rc = "14" # im and im-rc should always have the same version!
 bumpalo = { version = "3.6.1", features = ["collections"] }
 inkwell = { path = "../../vendor/inkwell" }
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 
 [dev-dependencies]
 roc_can = { path = "../can" }
diff --git a/compiler/gen_llvm/src/llvm/build.rs b/compiler/gen_llvm/src/llvm/build.rs
index 9515e43130..a416cbb8a4 100644
--- a/compiler/gen_llvm/src/llvm/build.rs
+++ b/compiler/gen_llvm/src/llvm/build.rs
@@ -3935,6 +3935,8 @@ pub fn get_call_conventions(cc: target_lexicon::CallingConvention) -> u32 {
         SystemV => C_CALL_CONV,
         WasmBasicCAbi => C_CALL_CONV,
         WindowsFastcall => C_CALL_CONV,
+        AppleAarch64 => C_CALL_CONV,
+        _ => C_CALL_CONV,
     }
 }
 
@@ -5174,89 +5176,171 @@ fn run_low_level<'a, 'ctx, 'env>(
     }
 }
 
-fn build_foreign_symbol_return_result<'a, 'ctx, 'env>(
+/// A type that is valid according to the C ABI
+///
+/// As an example, structs that fit inside an integer type should
+/// (this does not currently happen here) be coerced to that integer type.
+fn to_cc_type<'a, 'ctx, 'env>(
     env: &Env<'a, 'ctx, 'env>,
-    scope: &mut Scope<'a, 'ctx>,
-    foreign: &roc_module::ident::ForeignSymbol,
-    arguments: &[Symbol],
-    return_type: BasicTypeEnum<'ctx>,
-) -> (FunctionValue<'ctx>, &'a [BasicValueEnum<'ctx>]) {
-    let mut arg_vals: Vec<BasicValueEnum> = Vec::with_capacity_in(arguments.len(), env.arena);
-    let mut arg_types = Vec::with_capacity_in(arguments.len() + 1, env.arena);
-
-    for arg in arguments.iter() {
-        let (value, layout) = load_symbol_and_layout(scope, arg);
-        arg_vals.push(value);
-        let arg_type = basic_type_from_layout(env, layout);
-        debug_assert_eq!(arg_type, value.get_type());
-        arg_types.push(arg_type);
+    layout: &Layout<'a>,
+) -> BasicTypeEnum<'ctx> {
+    match layout {
+        Layout::Builtin(builtin) => to_cc_type_builtin(env, builtin),
+        _ => {
+            // TODO this is almost certainly incorrect for bigger structs
+            basic_type_from_layout(env, layout)
+        }
     }
-
-    let function_type = return_type.fn_type(&arg_types, false);
-    let function = get_foreign_symbol(env, foreign.clone(), function_type);
-
-    (function, arg_vals.into_bump_slice())
 }
 
-fn build_foreign_symbol_write_result_into_ptr<'a, 'ctx, 'env>(
+fn to_cc_type_builtin<'a, 'ctx, 'env>(
     env: &Env<'a, 'ctx, 'env>,
-    scope: &mut Scope<'a, 'ctx>,
-    foreign: &roc_module::ident::ForeignSymbol,
-    arguments: &[Symbol],
-    return_pointer: PointerValue<'ctx>,
-) -> (FunctionValue<'ctx>, &'a [BasicValueEnum<'ctx>]) {
-    let mut arg_vals: Vec<BasicValueEnum> = Vec::with_capacity_in(arguments.len(), env.arena);
-    let mut arg_types = Vec::with_capacity_in(arguments.len() + 1, env.arena);
-
-    arg_vals.push(return_pointer.into());
-    arg_types.push(return_pointer.get_type().into());
-
-    for arg in arguments.iter() {
-        let (value, layout) = load_symbol_and_layout(scope, arg);
-        arg_vals.push(value);
-        let arg_type = basic_type_from_layout(env, layout);
-        debug_assert_eq!(arg_type, value.get_type());
-        arg_types.push(arg_type);
+    builtin: &Builtin<'a>,
+) -> BasicTypeEnum<'ctx> {
+    match builtin {
+        Builtin::Int128
+        | Builtin::Int64
+        | Builtin::Int32
+        | Builtin::Int16
+        | Builtin::Int8
+        | Builtin::Int1
+        | Builtin::Usize
+        | Builtin::Decimal
+        | Builtin::Float128
+        | Builtin::Float64
+        | Builtin::Float32
+        | Builtin::Float16 => basic_type_from_builtin(env, builtin),
+        Builtin::Str | Builtin::EmptyStr | Builtin::List(_) | Builtin::EmptyList => {
+            env.str_list_c_abi().into()
+        }
+        Builtin::Dict(_, _) | Builtin::Set(_) | Builtin::EmptyDict | Builtin::EmptySet => {
+            // TODO verify this is what actually happens
+            basic_type_from_builtin(env, builtin)
+        }
+    }
+}
+
+enum CCReturn {
+    /// Return as normal
+    Return,
+    /// require an extra argument, a pointer
+    /// where the result is written into
+    /// returns void
+    ByPointer,
+    /// The return type is zero-sized
+    Void,
+}
+
+/// According to the C ABI, how should we return a value with the given layout?
+fn to_cc_return<'a, 'ctx, 'env>(env: &Env<'a, 'ctx, 'env>, layout: &Layout<'a>) -> CCReturn {
+    let return_size = layout.stack_size(env.ptr_bytes);
+    let pass_result_by_pointer = return_size > 2 * env.ptr_bytes;
+
+    if return_size == 0 {
+        CCReturn::Void
+    } else if pass_result_by_pointer {
+        CCReturn::ByPointer
+    } else {
+        CCReturn::Return
     }
-
-    let function_type = env.context.void_type().fn_type(&arg_types, false);
-    let function = get_foreign_symbol(env, foreign.clone(), function_type);
-
-    (function, arg_vals.into_bump_slice())
 }
 
-#[allow(clippy::too_many_arguments)]
 fn build_foreign_symbol<'a, 'ctx, 'env>(
     env: &Env<'a, 'ctx, 'env>,
     scope: &mut Scope<'a, 'ctx>,
     foreign: &roc_module::ident::ForeignSymbol,
-    arguments: &[Symbol],
+    argument_symbols: &[Symbol],
     ret_layout: &Layout<'a>,
 ) -> BasicValueEnum<'ctx> {
-    let ret_type = basic_type_from_layout(env, ret_layout);
-    let return_pointer = env.builder.build_alloca(ret_type, "return_value");
+    let builder = env.builder;
+    let context = env.context;
 
-    // crude approximation of the C calling convention
-    let pass_result_by_pointer = ret_layout.stack_size(env.ptr_bytes) > 2 * env.ptr_bytes;
+    // Here we build two functions:
+    //
+    // - an C_CALL_CONV extern that will be provided by the host, e.g. `roc_fx_putLine`
+    //      This is just a type signature that we make available to the linker,
+    //      and can use in the wrapper
+    // - a FAST_CALL_CONV wrapper that we make here, e.g. `roc_fx_putLine_fastcc_wrapper`
 
-    let (function, arguments) = if pass_result_by_pointer {
-        build_foreign_symbol_write_result_into_ptr(env, scope, foreign, arguments, return_pointer)
-    } else {
-        build_foreign_symbol_return_result(env, scope, foreign, arguments, ret_type)
+    let return_type = basic_type_from_layout(env, ret_layout);
+    let cc_return = to_cc_return(env, ret_layout);
+
+    let mut cc_argument_types = Vec::with_capacity_in(argument_symbols.len() + 1, env.arena);
+    let mut fastcc_argument_types = Vec::with_capacity_in(argument_symbols.len(), env.arena);
+    let mut arguments = Vec::with_capacity_in(argument_symbols.len(), env.arena);
+
+    for symbol in argument_symbols {
+        let (value, layout) = load_symbol_and_layout(scope, symbol);
+
+        cc_argument_types.push(to_cc_type(env, layout));
+
+        let basic_type = basic_type_from_layout(env, layout);
+        fastcc_argument_types.push(basic_type);
+
+        arguments.push(value);
+    }
+
+    let cc_type = match cc_return {
+        CCReturn::Void => env.context.void_type().fn_type(&cc_argument_types, false),
+        CCReturn::ByPointer => {
+            cc_argument_types.push(return_type.ptr_type(AddressSpace::Generic).into());
+            env.context.void_type().fn_type(&cc_argument_types, false)
+        }
+        CCReturn::Return => return_type.fn_type(&cc_argument_types, false),
     };
 
-    let call = env.builder.build_call(function, arguments, "tmp");
+    let cc_function = get_foreign_symbol(env, foreign.clone(), cc_type);
 
-    // this is a foreign function, use c calling convention
-    call.set_call_convention(C_CALL_CONV);
+    let fastcc_type = return_type.fn_type(&fastcc_argument_types, false);
 
-    call.try_as_basic_value();
+    let fastcc_function = add_func(
+        env.module,
+        &format!("{}_fastcc_wrapper", foreign.as_str()),
+        fastcc_type,
+        Linkage::Private,
+        FAST_CALL_CONV,
+    );
 
-    if pass_result_by_pointer {
-        env.builder.build_load(return_pointer, "read_result")
-    } else {
-        call.try_as_basic_value().left().unwrap()
+    let old = builder.get_insert_block().unwrap();
+
+    let entry = context.append_basic_block(fastcc_function, "entry");
+    {
+        builder.position_at_end(entry);
+        let return_pointer = env.builder.build_alloca(return_type, "return_value");
+
+        let fastcc_parameters = fastcc_function.get_params();
+        let mut cc_arguments = Vec::with_capacity_in(fastcc_parameters.len() + 1, env.arena);
+
+        for (param, cc_type) in fastcc_parameters.into_iter().zip(cc_argument_types.iter()) {
+            if param.get_type() == *cc_type {
+                cc_arguments.push(param);
+            } else {
+                let as_cc_type = complex_bitcast(env.builder, param, *cc_type, "to_cc_type");
+                cc_arguments.push(as_cc_type);
+            }
+        }
+
+        if let CCReturn::ByPointer = cc_return {
+            cc_arguments.push(return_pointer.into());
+        }
+
+        let call = env.builder.build_call(cc_function, &cc_arguments, "tmp");
+        call.set_call_convention(C_CALL_CONV);
+
+        let return_value = match cc_return {
+            CCReturn::Return => call.try_as_basic_value().left().unwrap(),
+
+            CCReturn::ByPointer => env.builder.build_load(return_pointer, "read_result"),
+            CCReturn::Void => return_type.const_zero(),
+        };
+
+        builder.build_return(Some(&return_value));
     }
+
+    builder.position_at_end(old);
+    let call = env.builder.build_call(fastcc_function, &arguments, "tmp");
+    call.set_call_convention(FAST_CALL_CONV);
+    return call.try_as_basic_value().left().unwrap();
 }
 
 fn throw_on_overflow<'a, 'ctx, 'env>(
diff --git a/compiler/load/src/file.rs b/compiler/load/src/file.rs
index 639f3a3cd6..6b4867fe09 100644
--- a/compiler/load/src/file.rs
+++ b/compiler/load/src/file.rs
@@ -832,6 +832,7 @@ struct State<'a> {
     pub exposed_types: SubsByModule,
     pub output_path: Option<&'a str>,
     pub platform_path: PlatformPath<'a>,
+    pub ptr_bytes: u32,
 
     pub headers_parsed: MutSet<ModuleId>,
 
@@ -1467,6 +1468,7 @@ where
 
             let mut state = State {
                 root_id,
+                ptr_bytes,
                 platform_data: None,
                 goal_phase,
                 stdlib,
@@ -1978,7 +1980,10 @@ fn update<'a>(
                 );
 
                 if state.goal_phase > Phase::SolveTypes {
-                    let layout_cache = state.layout_caches.pop().unwrap_or_default();
+                    let layout_cache = state
+                        .layout_caches
+                        .pop()
+                        .unwrap_or_else(|| LayoutCache::new(state.ptr_bytes));
 
                     let typechecked = TypeCheckedModule {
                         module_id,
diff --git a/compiler/mono/src/ir.rs b/compiler/mono/src/ir.rs
index 6f032c46ff..daade748e6 100644
--- a/compiler/mono/src/ir.rs
+++ b/compiler/mono/src/ir.rs
@@ -3124,7 +3124,8 @@ pub fn with_hole<'a>(
             mut fields,
             ..
         } => {
-            let sorted_fields = crate::layout::sort_record_fields(env.arena, record_var, env.subs);
+            let sorted_fields =
+                crate::layout::sort_record_fields(env.arena, record_var, env.subs, env.ptr_bytes);
 
             let mut field_symbols = Vec::with_capacity_in(fields.len(), env.arena);
             let mut can_fields = Vec::with_capacity_in(fields.len(), env.arena);
@@ -3459,7 +3460,8 @@ pub fn with_hole<'a>(
             loc_expr,
             ..
         } => {
-            let sorted_fields = crate::layout::sort_record_fields(env.arena, record_var, env.subs);
+            let sorted_fields =
+                crate::layout::sort_record_fields(env.arena, record_var, env.subs, env.ptr_bytes);
 
             let mut index = None;
             let mut field_layouts = Vec::with_capacity_in(sorted_fields.len(), env.arena);
@@ -3601,7 +3603,8 @@ pub fn with_hole<'a>(
             // This has the benefit that we don't need to do anything special for reference
             // counting
 
-            let sorted_fields = crate::layout::sort_record_fields(env.arena, record_var, env.subs);
+            let sorted_fields =
+                crate::layout::sort_record_fields(env.arena, record_var, env.subs, env.ptr_bytes);
 
             let mut field_layouts = Vec::with_capacity_in(sorted_fields.len(), env.arena);
 
@@ -4205,7 +4208,8 @@ fn convert_tag_union<'a>(
     arena: &'a Bump,
 ) -> Stmt<'a> {
     use crate::layout::UnionVariant::*;
-    let res_variant = crate::layout::union_sorted_tags(env.arena, variant_var, env.subs);
+    let res_variant =
+        crate::layout::union_sorted_tags(env.arena, variant_var, env.subs, env.ptr_bytes);
     let variant = match res_variant {
         Ok(cached) => cached,
         Err(LayoutProblem::UnresolvedTypeVar(_)) => {
@@ -4541,7 +4545,7 @@ fn sorted_field_symbols<'a>(
             }
         };
 
-        let alignment = layout.alignment_bytes(8);
+        let alignment = layout.alignment_bytes(env.ptr_bytes);
 
         let symbol = possible_reuse_symbol(env, procs, &arg.value);
         field_symbols_temp.push((alignment, symbol, ((var, arg), &*env.arena.alloc(symbol))));
@@ -6978,7 +6982,8 @@ fn from_can_pattern_help<'a>(
             use crate::exhaustive::Union;
             use crate::layout::UnionVariant::*;
 
-            let res_variant = crate::layout::union_sorted_tags(env.arena, *whole_var, env.subs);
+            let res_variant =
+                crate::layout::union_sorted_tags(env.arena, *whole_var, env.subs, env.ptr_bytes);
 
             let variant = match res_variant {
                 Ok(cached) => cached,
@@ -7397,7 +7402,8 @@ fn from_can_pattern_help<'a>(
             ..
         } => {
             // sorted fields based on the type
-            let sorted_fields = crate::layout::sort_record_fields(env.arena, *whole_var, env.subs);
+            let sorted_fields =
+                crate::layout::sort_record_fields(env.arena, *whole_var, env.subs, env.ptr_bytes);
 
             // sorted fields based on the destruct
             let mut mono_destructs = Vec::with_capacity_in(destructs.len(), env.arena);
diff --git a/compiler/mono/src/layout.rs b/compiler/mono/src/layout.rs
index e0a96e6a32..141bd9877b 100644
--- a/compiler/mono/src/layout.rs
+++ b/compiler/mono/src/layout.rs
@@ -138,7 +138,8 @@ impl<'a> RawFunctionLayout<'a> {
                 let fn_args = fn_args.into_bump_slice();
                 let ret = arena.alloc(ret);
 
-                let lambda_set = LambdaSet::from_var(env.arena, env.subs, closure_var)?;
+                let lambda_set =
+                    LambdaSet::from_var(env.arena, env.subs, closure_var, env.ptr_bytes)?;
 
                 Ok(Self::Function(fn_args, lambda_set, ret))
             }
@@ -516,6 +517,7 @@ impl<'a> LambdaSet<'a> {
         arena: &'a Bump,
         subs: &Subs,
         closure_var: Variable,
+        ptr_bytes: u32,
     ) -> Result<Self, LayoutProblem> {
         let mut tags = std::vec::Vec::new();
         match roc_types::pretty_print::chase_ext_tag_union(subs, closure_var, &mut tags) {
@@ -529,6 +531,7 @@ impl<'a> LambdaSet<'a> {
                     arena,
                     subs,
                     seen: Vec::new_in(arena),
+                    ptr_bytes,
                 };
 
                 for (tag_name, variables) in tags.iter() {
@@ -545,7 +548,8 @@ impl<'a> LambdaSet<'a> {
                     }
                 }
 
-                let representation = arena.alloc(Self::make_representation(arena, subs, tags));
+                let representation =
+                    arena.alloc(Self::make_representation(arena, subs, tags, ptr_bytes));
 
                 Ok(LambdaSet {
                     set: set.into_bump_slice(),
@@ -568,9 +572,10 @@ impl<'a> LambdaSet<'a> {
         arena: &'a Bump,
         subs: &Subs,
         tags: std::vec::Vec<(TagName, std::vec::Vec<Variable>)>,
+        ptr_bytes: u32,
     ) -> Layout<'a> {
         // otherwise, this is a closure with a payload
-        let variant = union_sorted_tags_help(arena, tags, None, subs);
+        let variant = union_sorted_tags_help(arena, tags, None, subs, ptr_bytes);
 
         use UnionVariant::*;
         match variant {
@@ -648,6 +653,7 @@ pub enum Builtin<'a> {
 }
 
 pub struct Env<'a, 'b> {
+    ptr_bytes: u32,
     arena: &'a Bump,
     seen: Vec<'a, Variable>,
     subs: &'b Subs,
@@ -972,8 +978,9 @@ impl<'a> Layout<'a> {
 /// e.g. `identity : a -> a` could be specialized to `Bool -> Bool` or `Str -> Str`.
 /// Therefore in general it's invalid to store a map from variables to layouts
 /// But if we're careful when to invalidate certain keys, we still get some benefit
-#[derive(Default, Debug)]
+#[derive(Debug)]
 pub struct LayoutCache<'a> {
+    ptr_bytes: u32,
     _marker: std::marker::PhantomData<&'a u8>,
 }
 
@@ -985,6 +992,13 @@ pub enum CachedLayout<'a> {
 }
 
 impl<'a> LayoutCache<'a> {
+    pub fn new(ptr_bytes: u32) -> Self {
+        Self {
+            ptr_bytes,
+            _marker: Default::default(),
+        }
+    }
+
     pub fn from_var(
         &mut self,
         arena: &'a Bump,
@@ -998,6 +1012,7 @@ impl<'a> LayoutCache<'a> {
             arena,
             subs,
             seen: Vec::new_in(arena),
+            ptr_bytes: self.ptr_bytes,
         };
 
         Layout::from_var(&mut env, var)
@@ -1016,6 +1031,7 @@ impl<'a> LayoutCache<'a> {
             arena,
             subs,
             seen: Vec::new_in(arena),
+            ptr_bytes: self.ptr_bytes,
         };
 
         RawFunctionLayout::from_var(&mut env, var)
@@ -1182,6 +1198,7 @@ fn layout_from_flat_type<'a>(
 
     let arena = env.arena;
     let subs = env.subs;
+    let ptr_bytes = env.ptr_bytes;
 
     match flat_type {
         Apply(symbol, args) => {
@@ -1273,7 +1290,7 @@ fn layout_from_flat_type<'a>(
             }
         }
         Func(_, closure_var, _) => {
-            let lambda_set = LambdaSet::from_var(env.arena, env.subs, closure_var)?;
+            let lambda_set = LambdaSet::from_var(env.arena, env.subs, closure_var, env.ptr_bytes)?;
 
             Ok(lambda_set.runtime_representation())
         }
@@ -1299,8 +1316,6 @@ fn layout_from_flat_type<'a>(
             let mut pairs = Vec::from_iter_in(pairs_it, arena);
 
             pairs.sort_by(|(label1, layout1), (label2, layout2)| {
-                let ptr_bytes = 8;
-
                 let size1 = layout1.alignment_bytes(ptr_bytes);
                 let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -1320,14 +1335,14 @@ fn layout_from_flat_type<'a>(
         TagUnion(tags, ext_var) => {
             debug_assert!(ext_var_is_empty_tag_union(subs, ext_var));
 
-            Ok(layout_from_tag_union(arena, tags, subs))
+            Ok(layout_from_tag_union(arena, tags, subs, env.ptr_bytes))
         }
         FunctionOrTagUnion(tag_name, _, ext_var) => {
             debug_assert!(ext_var_is_empty_tag_union(subs, ext_var));
 
             let tags = UnionTags::from_tag_name_index(tag_name);
 
-            Ok(layout_from_tag_union(arena, tags, subs))
+            Ok(layout_from_tag_union(arena, tags, subs, env.ptr_bytes))
         }
         RecursiveTagUnion(rec_var, tags, ext_var) => {
             debug_assert!(ext_var_is_empty_tag_union(subs, ext_var));
@@ -1377,8 +1392,6 @@ fn layout_from_flat_type<'a>(
                 }
 
                 tag_layout.sort_by(|layout1, layout2| {
-                    let ptr_bytes = 8;
-
                     let size1 = layout1.alignment_bytes(ptr_bytes);
                     let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -1425,11 +1438,13 @@ pub fn sort_record_fields<'a>(
     arena: &'a Bump,
     var: Variable,
     subs: &Subs,
+    ptr_bytes: u32,
 ) -> Vec<'a, (Lowercase, Variable, Result<Layout<'a>, Layout<'a>>)> {
     let mut env = Env {
         arena,
         subs,
         seen: Vec::new_in(arena),
+        ptr_bytes,
     };
 
     let (it, _) = gather_fields_unsorted_iter(subs, RecordFields::empty(), var);
@@ -1445,6 +1460,8 @@ fn sort_record_fields_help<'a>(
     env: &mut Env<'a, '_>,
     fields_map: impl Iterator<Item = (Lowercase, RecordField<Variable>)>,
 ) -> Vec<'a, (Lowercase, Variable, Result<Layout<'a>, Layout<'a>>)> {
+    let ptr_bytes = env.ptr_bytes;
+
     // Sort the fields by label
     let mut sorted_fields = Vec::with_capacity_in(fields_map.size_hint().0, env.arena);
 
@@ -1468,8 +1485,6 @@ fn sort_record_fields_help<'a>(
         |(label1, _, res_layout1), (label2, _, res_layout2)| match res_layout1 {
             Ok(layout1) | Err(layout1) => match res_layout2 {
                 Ok(layout2) | Err(layout2) => {
-                    let ptr_bytes = 8;
-
                     let size1 = layout1.alignment_bytes(ptr_bytes);
                     let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -1605,6 +1620,7 @@ pub fn union_sorted_tags<'a>(
     arena: &'a Bump,
     var: Variable,
     subs: &Subs,
+    ptr_bytes: u32,
 ) -> Result<UnionVariant<'a>, LayoutProblem> {
     let var =
         if let Content::RecursionVar { structure, .. } = subs.get_content_without_compacting(var) {
@@ -1617,7 +1633,7 @@ pub fn union_sorted_tags<'a>(
     let result = match roc_types::pretty_print::chase_ext_tag_union(subs, var, &mut tags_vec) {
         Ok(()) | Err((_, Content::FlexVar(_))) | Err((_, Content::RecursionVar { .. })) => {
             let opt_rec_var = get_recursion_var(subs, var);
-            union_sorted_tags_help(arena, tags_vec, opt_rec_var, subs)
+            union_sorted_tags_help(arena, tags_vec, opt_rec_var, subs, ptr_bytes)
         }
         Err((_, Content::Error)) => return Err(LayoutProblem::Erroneous),
         Err(other) => panic!("invalid content in tag union variable: {:?}", other),
@@ -1651,6 +1667,7 @@ fn union_sorted_tags_help_new<'a>(
     mut tags_vec: Vec<(&'_ TagName, VariableSubsSlice)>,
     opt_rec_var: Option<Variable>,
     subs: &Subs,
+    ptr_bytes: u32,
 ) -> UnionVariant<'a> {
     // sort up front; make sure the ordering stays intact!
     tags_vec.sort_unstable_by(|(a, _), (b, _)| a.cmp(b));
@@ -1659,6 +1676,7 @@ fn union_sorted_tags_help_new<'a>(
         arena,
         subs,
         seen: Vec::new_in(arena),
+        ptr_bytes,
     };
 
     match tags_vec.len() {
@@ -1708,8 +1726,6 @@ fn union_sorted_tags_help_new<'a>(
             }
 
             layouts.sort_by(|layout1, layout2| {
-                let ptr_bytes = 8;
-
                 let size1 = layout1.alignment_bytes(ptr_bytes);
                 let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -1793,8 +1809,6 @@ fn union_sorted_tags_help_new<'a>(
                 }
 
                 arg_layouts.sort_by(|layout1, layout2| {
-                    let ptr_bytes = 8;
-
                     let size1 = layout1.alignment_bytes(ptr_bytes);
                     let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -1867,6 +1881,7 @@ pub fn union_sorted_tags_help<'a>(
     mut tags_vec: std::vec::Vec<(TagName, std::vec::Vec<Variable>)>,
     opt_rec_var: Option<Variable>,
     subs: &Subs,
+    ptr_bytes: u32,
 ) -> UnionVariant<'a> {
     // sort up front; make sure the ordering stays intact!
     tags_vec.sort_unstable_by(|(a, _), (b, _)| a.cmp(b));
@@ -1875,6 +1890,7 @@ pub fn union_sorted_tags_help<'a>(
         arena,
         subs,
         seen: Vec::new_in(arena),
+        ptr_bytes,
     };
 
     match tags_vec.len() {
@@ -1921,8 +1937,6 @@ pub fn union_sorted_tags_help<'a>(
             }
 
             layouts.sort_by(|layout1, layout2| {
-                let ptr_bytes = 8;
-
                 let size1 = layout1.alignment_bytes(ptr_bytes);
                 let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -2005,8 +2019,6 @@ pub fn union_sorted_tags_help<'a>(
                 }
 
                 arg_layouts.sort_by(|layout1, layout2| {
-                    let ptr_bytes = 8;
-
                     let size1 = layout1.alignment_bytes(ptr_bytes);
                     let size2 = layout2.alignment_bytes(ptr_bytes);
 
@@ -2091,7 +2103,12 @@ fn cheap_sort_tags<'a, 'b>(
     tags_vec
 }
 
-fn layout_from_newtype<'a>(arena: &'a Bump, tags: UnionTags, subs: &Subs) -> Layout<'a> {
+fn layout_from_newtype<'a>(
+    arena: &'a Bump,
+    tags: UnionTags,
+    subs: &Subs,
+    ptr_bytes: u32,
+) -> Layout<'a> {
     debug_assert!(tags.is_newtype_wrapper(subs));
 
     let slice_index = tags.variables().into_iter().next().unwrap();
@@ -2109,6 +2126,7 @@ fn layout_from_newtype<'a>(arena: &'a Bump, tags: UnionTags, subs: &Subs) -> Lay
             arena,
             subs,
             seen: Vec::new_in(arena),
+            ptr_bytes,
         };
 
         match Layout::from_var(&mut env, var) {
@@ -2128,11 +2146,16 @@ fn layout_from_newtype<'a>(arena: &'a Bump, tags: UnionTags, subs: &Subs) -> Lay
     }
 }
 
-fn layout_from_tag_union<'a>(arena: &'a Bump, tags: UnionTags, subs: &Subs) -> Layout<'a> {
+fn layout_from_tag_union<'a>(
+    arena: &'a Bump,
+    tags: UnionTags,
+    subs: &Subs,
+    ptr_bytes: u32,
+) -> Layout<'a> {
     use UnionVariant::*;
 
     if tags.is_newtype_wrapper(subs) {
-        return layout_from_newtype(arena, tags, subs);
+        return layout_from_newtype(arena, tags, subs, ptr_bytes);
     }
 
     let tags_vec = cheap_sort_tags(arena, tags, subs);
@@ -2148,7 +2171,7 @@ fn layout_from_tag_union<'a>(arena: &'a Bump, tags: UnionTags, subs: &Subs) -> L
         }
         _ => {
             let opt_rec_var = None;
-            let variant = union_sorted_tags_help_new(arena, tags_vec, opt_rec_var, subs);
+            let variant = union_sorted_tags_help_new(arena, tags_vec, opt_rec_var, subs, ptr_bytes);
 
             match variant {
                 Never => Layout::Union(UnionLayout::NonRecursive(&[])),
diff --git a/compiler/reporting/tests/test_reporting.rs b/compiler/reporting/tests/test_reporting.rs
index af5a4c3d36..5575e0a20f 100644
--- a/compiler/reporting/tests/test_reporting.rs
+++ b/compiler/reporting/tests/test_reporting.rs
@@ -92,14 +92,15 @@ mod test_reporting {
             let mut ident_ids = interns.all_ident_ids.remove(&home).unwrap();
 
             // Populate Procs and Subs, and get the low-level Expr from the canonical Expr
-            let mut layout_cache = LayoutCache::default();
+            let ptr_bytes = 8;
+            let mut layout_cache = LayoutCache::new(ptr_bytes);
             let mut mono_env = roc_mono::ir::Env {
                 arena: &arena,
                 subs: &mut subs,
                 problems: &mut mono_problems,
                 home,
                 ident_ids: &mut ident_ids,
-                ptr_bytes: 8,
+                ptr_bytes,
                 update_mode_counter: 0,
                 // call_specialization_counter=0 is reserved
                 call_specialization_counter: 1,
diff --git a/compiler/test_gen/Cargo.toml b/compiler/test_gen/Cargo.toml
index 30f3e1a646..06a457e089 100644
--- a/compiler/test_gen/Cargo.toml
+++ b/compiler/test_gen/Cargo.toml
@@ -30,7 +30,7 @@ either = "1.6.1"
 indoc = "0.3.3"
 libc = "0.2"
 inkwell = { path = "../../vendor/inkwell" }
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 libloading = "0.6"
 
 [dev-dependencies]
diff --git a/compiler/test_mono/Cargo.toml b/compiler/test_mono/Cargo.toml
index 7929679da6..3167bbc789 100644
--- a/compiler/test_mono/Cargo.toml
+++ b/compiler/test_mono/Cargo.toml
@@ -28,7 +28,7 @@ bumpalo = { version = "3.6.1", features = ["collections"] }
 either = "1.6.1"
 indoc = "0.3.3"
 libc = "0.2"
-target-lexicon = "0.10"
+target-lexicon = "0.12.2"
 libloading = "0.6"
 
 [dev-dependencies]
diff --git a/examples/.gitignore b/examples/.gitignore
index a7b3b64e43..874bbd86e5 100644
--- a/examples/.gitignore
+++ b/examples/.gitignore
@@ -4,4 +4,3 @@ app
 libhost.a
 roc_app.ll
 roc_app.bc
-effect-example
diff --git a/examples/benchmarks/platform/host.zig b/examples/benchmarks/platform/host.zig
index 945c4d515e..7d6bba7632 100644
--- a/examples/benchmarks/platform/host.zig
+++ b/examples/benchmarks/platform/host.zig
@@ -29,23 +29,33 @@ extern fn roc__mainForHost_1_Fx_caller(*const u8, [*]u8, [*]u8) void;
 extern fn roc__mainForHost_1_Fx_size() i64;
 extern fn roc__mainForHost_1_Fx_result_size() i64;
 
-extern fn malloc(size: usize) callconv(.C) ?*c_void;
-extern fn realloc(c_ptr: [*]align(@alignOf(u128)) u8, size: usize) callconv(.C) ?*c_void;
-extern fn free(c_ptr: [*]align(@alignOf(u128)) u8) callconv(.C) void;
+const Align = extern struct { a: usize, b: usize };
+extern fn malloc(size: usize) callconv(.C) ?*align(@alignOf(Align)) c_void;
+extern fn realloc(c_ptr: [*]align(@alignOf(Align)) u8, size: usize) callconv(.C) ?*c_void;
+extern fn free(c_ptr: [*]align(@alignOf(Align)) u8) callconv(.C) void;
 
 export fn roc_alloc(size: usize, alignment: u32) callconv(.C) ?*c_void {
+    _ = alignment;
+
     return malloc(size);
 }
 
 export fn roc_realloc(c_ptr: *c_void, new_size: usize, old_size: usize, alignment: u32) callconv(.C) ?*c_void {
-    return realloc(@alignCast(16, @ptrCast([*]u8, c_ptr)), new_size);
+    _ = old_size;
+    _ = alignment;
+
+    return realloc(@alignCast(@alignOf(Align), @ptrCast([*]u8, c_ptr)), new_size);
 }
 
 export fn roc_dealloc(c_ptr: *c_void, alignment: u32) callconv(.C) void {
-    free(@alignCast(16, @ptrCast([*]u8, c_ptr)));
+    _ = alignment;
+
+    free(@alignCast(@alignOf(Align), @ptrCast([*]u8, c_ptr)));
 }
 
 export fn roc_panic(c_ptr: *c_void, tag_id: u32) callconv(.C) void {
+    _ = tag_id;
+
     const stderr = std.io.getStdErr().writer();
     const msg = @ptrCast([*:0]const u8, c_ptr);
     stderr.print("Application crashed with message\n\n    {s}\n\nShutting down\n", .{msg}) catch unreachable;
@@ -54,12 +64,9 @@ export fn roc_panic(c_ptr: *c_void, tag_id: u32) callconv(.C) void {
 
 const Unit = extern struct {};
 
-pub export fn main() u8 {
-    const stdout = std.io.getStdOut().writer();
-    const stderr = std.io.getStdErr().writer();
-
+pub fn main() u8 {
     const size = @intCast(usize, roc__mainForHost_size());
-    const raw_output = std.heap.c_allocator.alloc(u8, size) catch unreachable;
+    const raw_output = std.heap.c_allocator.allocAdvanced(u8, @alignOf(u64), @intCast(usize, size), .at_least) catch unreachable;
     var output = @ptrCast([*]u8, raw_output);
 
     defer {
@@ -71,17 +78,17 @@ pub export fn main() u8 {
 
     roc__mainForHost_1_exposed(output);
 
-    const elements = @ptrCast([*]u64, @alignCast(8, output));
-
-    var flag = elements[0];
+    const flag = @ptrCast(*u64, @alignCast(@alignOf(u64), output)).*;
 
     if (flag == 0) {
         // all is well
-        const closure_data_pointer = @ptrCast([*]u8, output[8..size]);
+        const closure_data_pointer = @ptrCast([*]u8, output[@sizeOf(u64)..size]);
 
         call_the_closure(closure_data_pointer);
     } else {
-        const msg = @intToPtr([*:0]const u8, elements[1]);
+        const ptr = @ptrCast(*u32, output + @sizeOf(u64));
+        const msg = @intToPtr([*:0]const u8, ptr.*);
+        const stderr = std.io.getStdErr().writer();
         stderr.print("Application crashed with message\n\n    {s}\n\nShutting down\n", .{msg}) catch unreachable;
 
         return 0;
@@ -92,6 +99,7 @@ pub export fn main() u8 {
 
     const delta = to_seconds(ts2) - to_seconds(ts1);
 
+    const stderr = std.io.getStdErr().writer();
     stderr.print("runtime: {d:.3}ms\n", .{delta * 1000}) catch unreachable;
 
     return 0;
@@ -103,7 +111,7 @@ fn to_seconds(tms: std.os.timespec) f64 {
 
 fn call_the_closure(closure_data_pointer: [*]u8) void {
     const size = roc__mainForHost_1_Fx_result_size();
-    const raw_output = std.heap.c_allocator.alloc(u8, @intCast(usize, size)) catch unreachable;
+    const raw_output = std.heap.c_allocator.allocAdvanced(u8, @alignOf(u64), @intCast(usize, size), .at_least) catch unreachable;
     var output = @ptrCast([*]u8, raw_output);
 
     defer {
@@ -135,7 +143,7 @@ pub export fn roc_fx_putInt(int: i64) i64 {
     return 0;
 }
 
-pub export fn roc_fx_putLine(rocPath: str.RocStr) i64 {
+export fn roc_fx_putLine(rocPath: str.RocStr) callconv(.C) void {
     const stdout = std.io.getStdOut().writer();
 
     for (rocPath.asSlice()) |char| {
@@ -143,8 +151,6 @@ pub export fn roc_fx_putLine(rocPath: str.RocStr) i64 {
     }
 
     stdout.print("\n", .{}) catch unreachable;
-
-    return 0;
 }
 
 const GetInt = extern struct {
diff --git a/examples/cli/.gitignore b/examples/cli/.gitignore
new file mode 100644
index 0000000000..fa11a6a9c5
--- /dev/null
+++ b/examples/cli/.gitignore
@@ -0,0 +1 @@
+echo
diff --git a/examples/cli/Echo.roc b/examples/cli/Echo.roc
new file mode 100644
index 0000000000..1dbb742a07
--- /dev/null
+++ b/examples/cli/Echo.roc
@@ -0,0 +1,18 @@
+#!/usr/bin/env roc
+
+app "echo"
+    packages { base: "platform" }
+    imports [ base.Task.{ Task, await }, base.Stdout, base.Stdin ]
+    provides [ main ] to base
+
+main : Task {} *
+main =
+    {} <- await (Stdout.line "What's your first name?")
+
+    firstName <- await Stdin.line
+
+    {} <- await (Stdout.line "What's your last name?")
+
+    lastName <- await Stdin.line
+
+    Stdout.line "Hi, \(firstName) \(lastName)!"
diff --git a/examples/cli/cli-example b/examples/cli/cli-example
new file mode 100755
index 0000000000..28e07a40f7
Binary files /dev/null and b/examples/cli/cli-example differ
diff --git a/examples/cli/hello-world b/examples/cli/hello-world
new file mode 100755
index 0000000000..c3248ebe2c
Binary files /dev/null and b/examples/cli/hello-world differ
diff --git a/examples/cli/platform/Cargo.lock b/examples/cli/platform/Cargo.lock
new file mode 100644
index 0000000000..cfd1e1e09d
--- /dev/null
+++ b/examples/cli/platform/Cargo.lock
@@ -0,0 +1,21 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 3
+
+[[package]]
+name = "host"
+version = "0.1.0"
+dependencies = [
+ "libc",
+ "roc_std",
+]
+
+[[package]]
+name = "libc"
+version = "0.2.100"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a1fa8cddc8fbbee11227ef194b5317ed014b8acbf15139bd716a18ad3fe99ec5"
+
+[[package]]
+name = "roc_std"
+version = "0.1.0"
diff --git a/examples/cli/platform/Cargo.toml b/examples/cli/platform/Cargo.toml
new file mode 100644
index 0000000000..ad2bc7c449
--- /dev/null
+++ b/examples/cli/platform/Cargo.toml
@@ -0,0 +1,15 @@
+[package]
+name = "host"
+version = "0.1.0"
+authors = ["The Roc Contributors"]
+license = "UPL-1.0"
+edition = "2018"
+
+[lib]
+crate-type = ["staticlib"]
+
+[dependencies]
+roc_std = { path = "../../../roc_std" }
+libc = "0.2"
+
+[workspace]
diff --git a/examples/cli/platform/Package-Config.roc b/examples/cli/platform/Package-Config.roc
new file mode 100644
index 0000000000..1efb7e95fa
--- /dev/null
+++ b/examples/cli/platform/Package-Config.roc
@@ -0,0 +1,14 @@
+platform examples/cli
+    requires {}{ main : Task {} [] } # TODO FIXME
+    exposes []
+    packages {}
+    imports [ Task.{ Task } ]
+    provides [ mainForHost ]
+    effects fx.Effect
+        {
+            putLine : Str -> Effect {},
+            getLine : Effect Str
+        }
+
+mainForHost : Task {} [] as Fx
+mainForHost = main
diff --git a/examples/cli/platform/Stdin.roc b/examples/cli/platform/Stdin.roc
new file mode 100644
index 0000000000..b46406c6de
--- /dev/null
+++ b/examples/cli/platform/Stdin.roc
@@ -0,0 +1,6 @@
+interface Stdin
+    exposes [ line ]
+    imports [ fx.Effect, Task ]
+
+line : Task.Task Str *
+line = Effect.after Effect.getLine Task.succeed # TODO FIXME Effect.getLine should suffice
diff --git a/examples/cli/platform/Stdout.roc b/examples/cli/platform/Stdout.roc
new file mode 100644
index 0000000000..dbde579ffc
--- /dev/null
+++ b/examples/cli/platform/Stdout.roc
@@ -0,0 +1,9 @@
+interface Stdout
+    exposes [ line ]
+    imports [ fx.Effect, Task.{ Task } ]
+
+# line : Str -> Task.Task {} *
+# line = \line -> Effect.map (Effect.putLine line) (\_ -> Ok {})
+
+line : Str -> Task {} *
+line = \str -> Effect.map (Effect.putLine str) (\_ -> Ok {})
diff --git a/examples/cli/platform/Task.roc b/examples/cli/platform/Task.roc
new file mode 100644
index 0000000000..d3f996bc8a
--- /dev/null
+++ b/examples/cli/platform/Task.roc
@@ -0,0 +1,44 @@
+interface Task
+    exposes [ Task, succeed, fail, await, map, onFail, attempt ]
+    imports [ fx.Effect ]
+
+
+Task ok err : Effect.Effect (Result ok err)
+
+
+succeed : val -> Task val *
+succeed = \val ->
+    Effect.always (Ok val)
+
+
+fail : err -> Task * err
+fail = \val ->
+    Effect.always (Err val)
+
+attempt : Task a b, (Result a b -> Task c d) -> Task c d
+attempt = \effect, transform ->
+    Effect.after effect \result ->
+        when result is
+            Ok ok -> transform (Ok ok)
+            Err err -> transform (Err err)
+
+await : Task a err, (a -> Task b err) -> Task b err
+await = \effect, transform ->
+    Effect.after effect \result ->
+        when result is
+            Ok a -> transform a
+            Err err -> Task.fail err
+
+onFail : Task ok a, (a -> Task ok b) -> Task ok b
+onFail = \effect, transform ->
+    Effect.after effect \result ->
+        when result is
+            Ok a -> Task.succeed a
+            Err err -> transform err
+
+map : Task a err, (a -> b) -> Task b err
+map = \effect, transform ->
+    Effect.after effect \result ->
+        when result is
+            Ok a -> Task.succeed (transform a)
+            Err err -> Task.fail err
diff --git a/examples/cli/platform/host.c b/examples/cli/platform/host.c
new file mode 100644
index 0000000000..0378c69589
--- /dev/null
+++ b/examples/cli/platform/host.c
@@ -0,0 +1,7 @@
+#include <stdio.h>
+
+extern int rust_main();
+
+int main() {
+  return rust_main();
+}
diff --git a/examples/cli/platform/src/lib.rs b/examples/cli/platform/src/lib.rs
new file mode 100644
index 0000000000..2a98179f8e
--- /dev/null
+++ b/examples/cli/platform/src/lib.rs
@@ -0,0 +1,139 @@
+#![allow(non_snake_case)]
+
+use core::alloc::Layout;
+use core::ffi::c_void;
+use core::mem::MaybeUninit;
+use libc;
+use roc_std::{RocCallResult, RocStr};
+use std::ffi::CStr;
+use std::os::raw::c_char;
+
+extern "C" {
+    #[link_name = "roc__mainForHost_1_exposed"]
+    fn roc_main(output: *mut u8) -> ();
+
+    #[link_name = "roc__mainForHost_size"]
+    fn roc_main_size() -> i64;
+
+    #[link_name = "roc__mainForHost_1_Fx_caller"]
+    fn call_Fx(flags: *const u8, closure_data: *const u8, output: *mut u8) -> ();
+
+    #[allow(dead_code)]
+    #[link_name = "roc__mainForHost_1_Fx_size"]
+    fn size_Fx() -> i64;
+
+    #[link_name = "roc__mainForHost_1_Fx_result_size"]
+    fn size_Fx_result() -> i64;
+}
+
+#[no_mangle]
+pub unsafe fn roc_alloc(size: usize, _alignment: u32) -> *mut c_void {
+    libc::malloc(size)
+}
+
+#[no_mangle]
+pub unsafe fn roc_realloc(
+    c_ptr: *mut c_void,
+    new_size: usize,
+    _old_size: usize,
+    _alignment: u32,
+) -> *mut c_void {
+    libc::realloc(c_ptr, new_size)
+}
+
+#[no_mangle]
+pub unsafe fn roc_dealloc(c_ptr: *mut c_void, _alignment: u32) {
+    libc::free(c_ptr)
+}
+
+#[no_mangle]
+pub unsafe fn roc_panic(c_ptr: *mut c_void, tag_id: u32) {
+    match tag_id {
+        0 => {
+            let slice = CStr::from_ptr(c_ptr as *const c_char);
+            let string = slice.to_str().unwrap();
+            eprintln!("Roc hit a panic: {}", string);
+            std::process::exit(1);
+        }
+        _ => todo!(),
+    }
+}
+
+#[no_mangle]
+pub fn rust_main() -> isize {
+    let size = unsafe { roc_main_size() } as usize;
+    let layout = Layout::array::<u8>(size).unwrap();
+
+    unsafe {
+        // TODO allocate on the stack if it's under a certain size
+        let buffer = std::alloc::alloc(layout);
+
+        roc_main(buffer);
+
+        let output = buffer as *mut RocCallResult<()>;
+
+        match (&*output).into() {
+            Ok(()) => {
+                let closure_data_ptr = buffer.offset(8);
+                let result = call_the_closure(closure_data_ptr as *const u8);
+
+                std::alloc::dealloc(buffer, layout);
+
+                result
+            }
+            Err(msg) => {
+                std::alloc::dealloc(buffer, layout);
+
+                panic!("Roc failed with message: {}", msg);
+            }
+        }
+    };
+
+    // Exit code
+    0
+}
+
+unsafe fn call_the_closure(closure_data_ptr: *const u8) -> i64 {
+    let size = size_Fx_result() as usize;
+    let layout = Layout::array::<u8>(size).unwrap();
+    let buffer = std::alloc::alloc(layout) as *mut u8;
+
+    call_Fx(
+        // This flags pointer will never get dereferenced
+        MaybeUninit::uninit().as_ptr(),
+        closure_data_ptr as *const u8,
+        buffer as *mut u8,
+    );
+
+    let output = &*(buffer as *mut RocCallResult<()>);
+
+    match output.into() {
+        Ok(_) => {
+            std::alloc::dealloc(buffer, layout);
+            0
+        }
+        Err(e) => panic!("failed with {}", e),
+    }
+}
+
+#[no_mangle]
+pub fn roc_fx_getLine() -> RocStr {
+    use std::io::{self, BufRead};
+
+    let stdin = io::stdin();
+    let line1 = stdin.lock().lines().next().unwrap().unwrap();
+
+    RocStr::from_slice(line1.as_bytes())
+}
+
+#[no_mangle]
+pub fn roc_fx_putLine(line: RocStr) -> () {
+    let bytes = line.as_slice();
+    let string = unsafe { std::str::from_utf8_unchecked(bytes) };
+    println!("{}", string);
+
+    // don't mess with the refcount!
+    core::mem::forget(line);
+
+    ()
+}
diff --git a/examples/hello-zig/platform/host.zig b/examples/hello-zig/platform/host.zig
index 02b4d939a0..07a6a1d52f 100644
--- a/examples/hello-zig/platform/host.zig
+++ b/examples/hello-zig/platform/host.zig
@@ -19,23 +19,32 @@ comptime {
     }
 }
 
-extern fn malloc(size: usize) callconv(.C) ?*c_void;
-extern fn realloc(c_ptr: [*]align(@alignOf(u128)) u8, size: usize) callconv(.C) ?*c_void;
-extern fn free(c_ptr: [*]align(@alignOf(u128)) u8) callconv(.C) void;
+const Align = extern struct { a: usize, b: usize };
+extern fn malloc(size: usize) callconv(.C) ?*align(@alignOf(Align)) c_void;
+extern fn realloc(c_ptr: [*]align(@alignOf(Align)) u8, size: usize) callconv(.C) ?*c_void;
+extern fn free(c_ptr: [*]align(@alignOf(Align)) u8) callconv(.C) void;
 
 export fn roc_alloc(size: usize, alignment: u32) callconv(.C) ?*c_void {
+    _ = alignment;
+
     return malloc(size);
 }
 
-export fn roc_realloc(c_ptr: *c_void, old_size: usize, new_size: usize, alignment: u32) callconv(.C) ?*c_void {
-    return realloc(@alignCast(16, @ptrCast([*]u8, c_ptr)), new_size);
+export fn roc_realloc(c_ptr: *c_void, new_size: usize, old_size: usize, alignment: u32) callconv(.C) ?*c_void {
+    _ = old_size;
+    _ = alignment;
+
+    return realloc(@alignCast(@alignOf(Align), @ptrCast([*]u8, c_ptr)), new_size);
 }
 
 export fn roc_dealloc(c_ptr: *c_void, alignment: u32) callconv(.C) void {
-    free(@alignCast(16, @ptrCast([*]u8, c_ptr)));
+    _ = alignment;
+
+    free(@alignCast(@alignOf(Align), @ptrCast([*]u8, c_ptr)));
 }
 
 export fn roc_panic(c_ptr: *c_void, tag_id: u32) callconv(.C) void {
+    _ = tag_id;
     const stderr = std.io.getStdErr().writer();
     const msg = @ptrCast([*:0]const u8, c_ptr);
     stderr.print("Application crashed with message\n\n    {s}\n\nShutting down\n", .{msg}) catch unreachable;
@@ -47,11 +56,11 @@ const Allocator = mem.Allocator;
 
 extern fn roc__mainForHost_1_exposed(*RocCallResult) void;
 
-const RocCallResult = extern struct { flag: usize, content: RocStr };
+const RocCallResult = extern struct { flag: u64, content: RocStr };
 
 const Unit = extern struct {};
 
-pub export fn main() i32 {
+pub fn main() u8 {
     const stdout = std.io.getStdOut().writer();
     const stderr = std.io.getStdErr().writer();
 
diff --git a/roc_std/src/lib.rs b/roc_std/src/lib.rs
index f2880ed4ef..c5b1f3e87a 100644
--- a/roc_std/src/lib.rs
+++ b/roc_std/src/lib.rs
@@ -584,6 +584,10 @@ impl RocStr {
 
                 let raw_ptr = Self::get_element_ptr(raw_ptr as *mut u8);
 
+                // write the refcount
+                let refcount_ptr = raw_ptr as *mut isize;
+                *(refcount_ptr.offset(-1)) = isize::MIN;
+
                 {
                     // NOTE: using a memcpy here causes weird issues
                     let target_ptr = raw_ptr as *mut u8;