diff --git a/Makefile b/Makefile
index 1978ffb..f61f58d 100644
--- a/Makefile
+++ b/Makefile
@@ -1,5 +1,5 @@
 #CXX := clang++-19
-CXXFLAGS += -O0 -std=c++23 -Wall -Wextra -Wpedantic -ggdb -g3
+CXXFLAGS += -m32 -Os -std=c++23 -Wall -Wextra -Wpedantic -ggdb -g3
 
 all: main
 
diff --git a/forth.hpp b/forth.hpp
index 560cafd..48ebbc0 100644
--- a/forth.hpp
+++ b/forth.hpp
@@ -19,6 +19,7 @@
 
 #include <algorithm>
 #include <array>
+#include <bit>
 #include <charconv>
 #include <cstdint>
 #include <cstddef>
@@ -68,16 +69,16 @@ struct forth
         addr flags_len;
 
         auto name() const -> std::string_view {
-            return {reinterpret_cast<const char *>(this + 1)};
+            return {std::bit_cast<const char *>(this + 1)};
         }
 
         auto body() const -> const func * {
-            auto ptr = reinterpret_cast<const std::uint8_t *>(this + 1)
-                + (flags_len & 0xFF);
-            return reinterpret_cast<const func *>(ptr);
+            const auto ptr = std::bit_cast<const std::uint8_t *>(this + 1);
+            const auto fptr = ptr + (flags_len & 0xFF);
+            return std::bit_cast<const func *>(fptr);
         }
 
-        void make_immediate() {
+        constexpr void make_immediate() {
             flags_len |= immediate;
         }
     };
@@ -113,7 +114,7 @@ struct forth
 
     void rpush(func *v) {
         assert<error::return_stack_overflow>(rp != rstack.begin());
-        *--rp = reinterpret_cast<cell>(v);
+        *--rp = v;
     }
 
     cell& top() {
@@ -123,16 +124,12 @@ struct forth
 
     cell pop() {
         assert<error::stack_underflow>(sp != dstack.end());
-        auto ret = *sp;
-        *sp++ = -1;
-        return ret;//*sp++;
+        return *sp++;
     }
 
     auto rpop() -> func * {
         assert<error::return_stack_underflow>(rp != rstack.end());
-        auto ret = reinterpret_cast<func *>(*rp);
-        *rp++ = -1;
-        return ret;
+        return *rp++;
     }
 
     template<int N>
@@ -154,12 +151,11 @@ struct forth
         //assert<error::dictionary_overflow>(state->here + size < &dictionary.back());
 
         const auto h = std::exchange(here, here + size);
-        auto w = new (h) word_base (nullptr, namesz);
-        w->next = std::exchange(latest, w);
+        latest = new (h) word_base (latest, namesz);
         std::copy(name.begin(), name.end(),
-            reinterpret_cast<char *>(h) + sizeof(word_base));
+            std::bit_cast<char *>(h) + sizeof(word_base));
         if (entry)
-            *here++ = reinterpret_cast<cell>(entry);
+            *here++ = std::bit_cast<cell>(entry);
         return *this;
     }
 
@@ -185,7 +181,7 @@ struct forth
                 auto body = (*ent)->body();
 
                 if (compiling && ((*ent)->flags_len & word_base::immediate) == 0) {
-                    *here++ = reinterpret_cast<cell>(body);
+                    *here++ = std::bit_cast<cell>(body);
                 } else {
                     execute(body);
                 }
@@ -225,7 +221,7 @@ struct forth
         fth.rpush(fth.ip);
 
         for (fth.ip = body + 1; *fth.ip; fth.ip++)
-            fth.execute(reinterpret_cast<func *>(*fth.ip));
+            fth.execute(std::bit_cast<func *>(*fth.ip));
 
         fth.ip = fth.rpop();
     }
@@ -238,10 +234,10 @@ struct forth
         static auto& fth = **fthp;
 
         constexpr static func lit_impl = [](auto) {
-            auto ptr = reinterpret_cast<cell *>(++fth.ip);
+            auto ptr = std::bit_cast<cell *>(++fth.ip);
             fth.push(*ptr);
         };
-        auto f_dict   = [](auto) { fth.push(reinterpret_cast<cell>(&fth)); };
+        auto f_dict   = [](auto) { fth.push(std::bit_cast<cell>(&fth)); };
         auto f_add    = [](auto) { fth.top() += fth.pop(); };
         auto f_minus  = [](auto) { fth.top() -= fth.pop(); };
         auto f_times  = [](auto) { fth.top() *= fth.pop(); };
@@ -258,16 +254,16 @@ struct forth
             const_cast<word_base *>(fth.latest)->make_immediate(); };
         auto f_lit    = [](auto) {
             //assert<error::compile_only_word>(fth.compiling);
-            *fth.here++ = reinterpret_cast<cell>(&lit_impl);
+            *fth.here++ = std::bit_cast<cell>(&lit_impl);
             *fth.here++ = fth.pop(); };
-        auto f_peek = [](auto) { fth.push(*reinterpret_cast<cell *>(fth.pop())); };
+        auto f_peek = [](auto) { fth.push(*std::bit_cast<cell *>(fth.pop())); };
         auto f_poke = [](auto) {
             auto [p, v] = fth.pop<2>();
-            *reinterpret_cast<cell *>(p) = v; };
-        auto f_cpeek = [](auto) { fth.push(*reinterpret_cast<char *>(fth.pop())); };
+            *std::bit_cast<cell *>(p) = v; };
+        auto f_cpeek = [](auto) { fth.push(*std::bit_cast<char *>(fth.pop())); };
         auto f_cpoke = [](auto) {
             auto [p, v] = fth.pop<2>();
-            *reinterpret_cast<char *>(p) = v; };
+            *std::bit_cast<char *>(p) = v; };
         auto f_swap = [](auto) { auto [a, b] = fth.pop<2>(); fth.push(a, b); };
         auto f_drop = [](auto) { fth.pop(); };
         auto f_dup  = [](auto) { fth.push(fth.top()); };
@@ -278,26 +274,27 @@ struct forth
             auto w = fth.parse();
 
             if (auto g = fth.get(w); g)
-                fth.push(reinterpret_cast<cell>((*g)->body()));
+                fth.push(std::bit_cast<cell>((*g)->body()));
             else
                 fth.push(0); };
         auto f_colon = [](auto) {
+            const auto prologue = forth::prologue<fthp>;
             auto w = fth.parse();
             fth.add(w);
-            *fth.here++ = reinterpret_cast<cell>(forth::prologue<fthp>);
+            *fth.here++ = std::bit_cast<cell>(prologue);
             fth.compiling = true; };
         auto f_semic = [](auto) { *fth.here++ = 0; fth.compiling = false; };
         auto f_comm  = [](auto) { fth.sourcei = npos; };
         auto f_cell  = [](auto) { fth.push(sizeof(cell)); };
         auto f_jmp   = [](auto) {
             auto ptr = ++fth.ip;
-            fth.ip = *reinterpret_cast<func **>(ptr) - 1;
+            fth.ip = *std::bit_cast<func **>(ptr) - 1;
         };
         auto f_jmp0  = [](auto) {
             auto ptr = ++fth.ip;
 
             if (fth.pop() == 0)
-                fth.ip = *reinterpret_cast<func **>(ptr) - 1;
+                fth.ip = *std::bit_cast<func **>(ptr) - 1;
         };
         auto f_postpone = [](auto) {
             assert<error::compile_only_word>(fth.compiling);
@@ -307,7 +304,7 @@ struct forth
 
             assert<error::word_not_found>(g.has_value());
 
-            *fth.here++ = reinterpret_cast<cell>((*g)->body());
+            *fth.here++ = std::bit_cast<cell>((*g)->body());
         };
 
         constexpr static word w_dict {"_d", f_dict};
@@ -372,9 +369,9 @@ struct forth
     }
 
     cell *sp;
-    cell *rp;
+    func **rp;
     func *ip = nullptr;
-    cell *here = reinterpret_cast<cell *>(this + 1);
+    cell *here = std::bit_cast<cell *>(this + 1);
     const word_base *latest = nullptr;
     const char *source = nullptr;
     std::size_t sourcei = npos;
@@ -382,7 +379,7 @@ struct forth
     cell *end = nullptr;
     cell base = 10;
     std::array<cell, data_size> dstack;
-    std::array<cell, return_size> rstack;
+    std::array<func *, return_size> rstack;
 };
 
 static_assert(offsetof(forth::word_base, flags_len) == 1 * sizeof(forth::cell));
diff --git a/main.cpp b/main.cpp
index 16de440..d3bbe42 100644
--- a/main.cpp
+++ b/main.cpp
@@ -38,6 +38,7 @@ int main(int argc, const char *argv[])
         std::cout << buf << ' ';
     });
     fth->add("emit", [](auto) { std::cout << static_cast<char>(fth->pop()); });
+    fth->add("dictsize", [](auto) { fth->push(dict.size() * sizeof(forth::cell)); });
 
     for (auto arg : args) {
         if (std::ifstream file {arg}; parse_stream(fth, file))