diff --git a/WL.c b/WL.c
index 7adedd2..cda680b 100644
--- a/WL.c
+++ b/WL.c
@@ -689,12 +689,6 @@ Token next_token(Parser *p)
         if (streq(kword, S("false"))) return (Token) { .type=TOKEN_KWORD_FALSE };
         if (streq(kword, S("include"))) return (Token) { .type=TOKEN_KWORD_INCLUDE };
 
-        kword = copystr(kword, p->a);
-        if (kword.len == 0) {
-            parser_report(p, "Out of memory");
-            return (Token) { .type=TOKEN_ERROR };
-        }
-
         return (Token) { .type=TOKEN_IDENT, .sval=kword };
     }
 
@@ -4040,6 +4034,28 @@ char *print_instruction(char *p, char *data)
         case OPCODE_PUSHFL:
         printf("PUSHFL");
         break;
+
+        case OPCODE_FOR:
+        {
+            uint8_t a;
+            memcpy(&a, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint8_t b;
+            memcpy(&b, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint8_t c;
+            memcpy(&c, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint32_t d;
+            memcpy(&d, p, sizeof(uint32_t));
+            p += sizeof(uint32_t);
+
+            printf("FOR %u %u %u %u", a, b, c, d);
+        }
+        break;
     }
 
     return p;
@@ -4572,7 +4588,7 @@ static void value_to_string_inner(Value v, ToStringContext *tostr)
             int cursor = 0;
             while (batch) {
 
-                int num = ITEMS_PER_MAP_BATCH;
+                int num = ITEMS_PER_ARRAY_BATCH;
                 if (batch->next == NULL)
                     num = a->tail_count;
 
@@ -4644,6 +4660,8 @@ struct WL_State {
     String data;
     int off;
 
+    bool trace;
+
     WL_Arena *a;
 
     char *errbuf;
@@ -4693,6 +4711,8 @@ void eval_report(WL_State *state, char *fmt, ...)
 
 static uint8_t read_u8(WL_State *state)
 {
+    assert(state->off >= 0);
+    assert(state->off < state->code.len);
     return state->code.ptr[state->off++];
 }
 
@@ -4726,11 +4746,12 @@ static double read_f64(WL_State *state)
 int step(WL_State *state)
 {
     uint8_t opcode = read_u8(state);
-/*
-    printf("%-3d: ", e->off);
-    print_instruction(state->code.ptr + e->off, e->data.ptr);
-    printf("\n");
-*/
+
+    if (state->trace) {
+        printf("%-3d: ", state->off-1);
+        print_instruction(state->code.ptr + state->off - 1, state->data.ptr);
+        printf("\n");
+    }
 
     switch (opcode) {
 
@@ -5505,6 +5526,8 @@ int step(WL_State *state)
             uint8_t  var_2 = read_u8(state);
             uint32_t end   = read_u32(state);
 
+            printf("for end %u\n", end); // TODO
+
             int base;
             {
                 int group = state->frames[state->num_frames-1].group;
@@ -5578,6 +5601,7 @@ WL_State *WL_State_init(WL_Arena *a, WL_Program p, char *err, int errmax)
         .code=code,
         .data=data,
         .off=0,
+        .trace=false,
         .a=a,
         .errbuf=err,
         .errmax=errmax,
@@ -5601,6 +5625,11 @@ void WL_State_free(WL_State *state)
     // TODO
 }
 
+void WL_State_trace(WL_State *state, int trace)
+{
+    state->trace = (trace != 0);
+}
+
 WL_Result WL_eval(WL_State *state)
 {
     if (state->sysvar.len > 0) {
@@ -6105,3 +6134,8 @@ WL_CompileResult WL_compile(WL_Compiler *compiler, WL_String file, WL_String con
 
     return (WL_CompileResult) { .type=WL_COMPILE_RESULT_DONE, .program=ares.program };
 }
+
+void WL_dump_program(WL_Program program)
+{
+    print_program(program);
+}
diff --git a/WL.h b/WL.h
index d806438..f0118bb 100644
--- a/WL.h
+++ b/WL.h
@@ -64,8 +64,11 @@ void             WL_Compiler_free (WL_Compiler *compiler);
 WL_CompileResult WL_compile       (WL_Compiler *compiler, WL_String file, WL_String content);
 WL_State*        WL_State_init    (WL_Arena *a, WL_Program p, char *err, int errmax);
 void             WL_State_free    (WL_State *state);
+void             WL_State_trace   (WL_State *state, int trace);
 WL_Result        WL_eval          (WL_State *state);
 
+void             WL_dump_program(WL_Program program);
+
 int       WL_streq      (WL_String a, char *b, int blen);
 int       WL_peeknone   (WL_State *state, int off);
 int       WL_peekint    (WL_State *state, int off, long long *x);
diff --git a/main.c b/main.c
index fcac6f7..79d5571 100644
--- a/main.c
+++ b/main.c
@@ -23,6 +23,9 @@ int main(int argc, char **argv)
         assert(0); // TODO
     }
 
+    int num_loaded_files = 0;
+    char *loaded_files[128];
+
     WL_CompileResult result;
     WL_String path = { file, strlen(file) };
     for (int i = 0;; i++) {
@@ -51,7 +54,7 @@ int main(int argc, char **argv)
 
         result = WL_compile(compiler, path, (WL_String) { file_data, file_size });
 
-        free(file_data);
+        loaded_files[num_loaded_files++] = file_data;
 
         if (result.type == WL_COMPILE_RESULT_ERROR) {
             printf("Compilation of '%.*s' failed\n", path.len, path.ptr);
@@ -63,7 +66,10 @@ int main(int argc, char **argv)
 
         assert(result.type == WL_COMPILE_RESULT_FILE);
         path = result.path;
-   }
+    }
+
+    for (int i = 0; i < num_loaded_files; i++)
+        free(loaded_files[i]);
 
     WL_Compiler_free(compiler);
 
@@ -75,6 +81,8 @@ int main(int argc, char **argv)
 
     WL_State *state = WL_State_init(&a, program, err, (int) sizeof(err));
 
+    WL_State_trace(state, 0);
+
     for (bool done = false; !done; ) {
 
         WL_Result result = WL_eval(state);
diff --git a/src/assemble.c b/src/assemble.c
index d8cbedd..6660606 100644
--- a/src/assemble.c
+++ b/src/assemble.c
@@ -1594,6 +1594,28 @@ char *print_instruction(char *p, char *data)
         case OPCODE_PUSHFL:
         printf("PUSHFL");
         break;
+
+        case OPCODE_FOR:
+        {
+            uint8_t a;
+            memcpy(&a, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint8_t b;
+            memcpy(&b, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint8_t c;
+            memcpy(&c, p, sizeof(uint8_t));
+            p += sizeof(uint8_t);
+
+            uint32_t d;
+            memcpy(&d, p, sizeof(uint32_t));
+            p += sizeof(uint32_t);
+
+            printf("FOR %u %u %u %u", a, b, c, d);
+        }
+        break;
     }
 
     return p;
diff --git a/src/compile.c b/src/compile.c
index 82006fe..8f71e3a 100644
--- a/src/compile.c
+++ b/src/compile.c
@@ -116,3 +116,8 @@ WL_CompileResult WL_compile(WL_Compiler *compiler, WL_String file, WL_String con
 
     return (WL_CompileResult) { .type=WL_COMPILE_RESULT_DONE, .program=ares.program };
 }
+
+void WL_dump_program(WL_Program program)
+{
+    print_program(program);
+}
\ No newline at end of file
diff --git a/src/compile.h b/src/compile.h
index e5def46..0022ceb 100644
--- a/src/compile.h
+++ b/src/compile.h
@@ -55,8 +55,11 @@ void             WL_Compiler_free (WL_Compiler *compiler);
 WL_CompileResult WL_compile       (WL_Compiler *compiler, WL_String file, WL_String content);
 WL_State*        WL_State_init    (WL_Arena *a, WL_Program p, char *err, int errmax);
 void             WL_State_free    (WL_State *state);
+void             WL_State_trace   (WL_State *state, int trace);
 WL_Result        WL_eval          (WL_State *state);
 
+void             WL_dump_program(WL_Program program);
+
 int       WL_streq      (WL_String a, char *b, int blen);
 int       WL_peeknone   (WL_State *state, int off);
 int       WL_peekint    (WL_State *state, int off, long long *x);
diff --git a/src/eval.c b/src/eval.c
index f719f99..2c4f3fe 100644
--- a/src/eval.c
+++ b/src/eval.c
@@ -20,6 +20,8 @@ struct WL_State {
     String data;
     int off;
 
+    bool trace;
+
     WL_Arena *a;
 
     char *errbuf;
@@ -69,6 +71,8 @@ void eval_report(WL_State *state, char *fmt, ...)
 
 static uint8_t read_u8(WL_State *state)
 {
+    assert(state->off >= 0);
+    assert(state->off < state->code.len);
     return state->code.ptr[state->off++];
 }
 
@@ -102,11 +106,12 @@ static double read_f64(WL_State *state)
 int step(WL_State *state)
 {
     uint8_t opcode = read_u8(state);
-/*
-    printf("%-3d: ", e->off);
-    print_instruction(state->code.ptr + e->off, e->data.ptr);
-    printf("\n");
-*/
+
+    if (state->trace) {
+        printf("%-3d: ", state->off-1);
+        print_instruction(state->code.ptr + state->off - 1, state->data.ptr);
+        printf("\n");
+    }
 
     switch (opcode) {
 
@@ -881,6 +886,8 @@ int step(WL_State *state)
             uint8_t  var_2 = read_u8(state);
             uint32_t end   = read_u32(state);
 
+            printf("for end %u\n", end); // TODO
+
             int base;
             {
                 int group = state->frames[state->num_frames-1].group;
@@ -954,6 +961,7 @@ WL_State *WL_State_init(WL_Arena *a, WL_Program p, char *err, int errmax)
         .code=code,
         .data=data,
         .off=0,
+        .trace=false,
         .a=a,
         .errbuf=err,
         .errmax=errmax,
@@ -977,6 +985,11 @@ void WL_State_free(WL_State *state)
     // TODO
 }
 
+void WL_State_trace(WL_State *state, int trace)
+{
+    state->trace = (trace != 0);
+}
+
 WL_Result WL_eval(WL_State *state)
 {
     if (state->sysvar.len > 0) {
diff --git a/src/parse.c b/src/parse.c
index 241da09..91f3281 100644
--- a/src/parse.c
+++ b/src/parse.c
@@ -255,12 +255,6 @@ Token next_token(Parser *p)
         if (streq(kword, S("false"))) return (Token) { .type=TOKEN_KWORD_FALSE };
         if (streq(kword, S("include"))) return (Token) { .type=TOKEN_KWORD_INCLUDE };
 
-        kword = copystr(kword, p->a);
-        if (kword.len == 0) {
-            parser_report(p, "Out of memory");
-            return (Token) { .type=TOKEN_ERROR };
-        }
-
         return (Token) { .type=TOKEN_IDENT, .sval=kword };
     }
 
diff --git a/src/value.c b/src/value.c
index 3eb0d48..5fbd922 100644
--- a/src/value.c
+++ b/src/value.c
@@ -470,7 +470,7 @@ static void value_to_string_inner(Value v, ToStringContext *tostr)
             int cursor = 0;
             while (batch) {
 
-                int num = ITEMS_PER_MAP_BATCH;
+                int num = ITEMS_PER_ARRAY_BATCH;
                 if (batch->next == NULL)
                     num = a->tail_count;