Add lexical closures
diff --git a/.vscode/launch.json b/.vscode/launch.json
index b438376..c265161 100644
--- a/.vscode/launch.json
+++ b/.vscode/launch.json
@@ -9,7 +9,7 @@
"type": "cppdbg",
"request": "launch",
"program": "${workspaceFolder}/src/lisp/lisp",
- "args": ["test-macros.lisp"],
+ "args": ["test-closures.lisp"],
"stopAtEntry": false,
"cwd": "${workspaceFolder}/src/lisp",
"environment": [],
diff --git a/src/lisp/call_list.s b/src/lisp/call_list.s
index da0c00c..72d6277 100644
--- a/src/lisp/call_list.s
+++ b/src/lisp/call_list.s
@@ -7,12 +7,16 @@
;;; This function should call it's first argument with the arguments from
;;; the cons-list passed as its second argument.
- ;;; _call_list(function pointer, cons list)
+ ;;; _call_list(function pointer, cons list, edi)
_call_list:
+ ;; esi and edi are callee-saved on x86, these are the only registers
+ ;; we clobber.
+ push esi
+ push edi
push ebp
mov ebp, esp
- mov edi, [ebp + 12] ; Cons list
+ mov edi, [ebp + 20] ; Cons list
push edi
call length ; Length of cons list in eax
@@ -49,9 +53,12 @@
jmp .loop
.done:
- mov ebx, [ebp + 8] ; Function pointer
+ mov ebx, [ebp + 16] ; Function pointer
+ mov edi, [ebp + 24] ; Closure data pointer
call ebx
mov esp, ebp
pop ebp
+ pop edi
+ pop esi
ret
diff --git a/src/lisp/compiler.dasc b/src/lisp/compiler.dasc
index 8c876d3..8d96a52 100644
--- a/src/lisp/compiler.dasc
+++ b/src/lisp/compiler.dasc
@@ -3,6 +3,7 @@
#include "compiler.h"
#include "lib/std.h"
#include "plat/plat.h"
+#include "gc.h"
#include <dasm_proto.h>
#include <dasm_x86.h>
@@ -32,8 +33,6 @@
dasm_State *d;
unsigned int npc = 8;
-extern void _do_gc(unsigned int ebp, unsigned int esp);
-
|.macro run_gc;
| mov eax, esp;
| push ebp;
@@ -88,8 +87,8 @@
}
struct dasm_State *compile_function(value_t args, enum namespace namespace,
- struct environment *env,
- struct local *local_out, int *nargs)
+ struct environment *env, struct local *local_out,
+ struct local *local_parent, int *nargs)
{
dasm_State *d;
dasm_State **Dst = &d;
@@ -114,20 +113,17 @@
memset(local.stack_slots, 0, sizeof(bool) * 4);
local.num_stack_slots = 4;
local.num_stack_entries = 0;
+ local.num_closure_slots = 0;
+ local.parent = local_parent;
dasm_growpc(&d, local.npc);
// Generate code
// TODO: first pass, extract bound and free variables
- value_t name = car(args);
- args = cdr(args);
value_t arglist = car(args);
value_t body = cdr(args);
- if ((name & HEAP_MASK) != SYMBOL_TAG)
- err("function name must be a symbol");
-
value_t a = arglist;
for (int i = 0; !nilp(a); a = cdr(a), i++)
{
@@ -163,6 +159,9 @@
*nargs = length(arglist);
return d;
+
+ // TODO: local leaks memory! free variables too, not just stack slots (in
+ // two places). Add a free_local() function that does this.
}
void compile_tl(value_t val, struct environment *env)
@@ -173,8 +172,6 @@
value_t form = car(val);
value_t args = cdr(val);
- printf("Compiling function %s in %s\n", (char *)(car(args) ^ SYMBOL_TAG), (char *)(form ^ SYMBOL_TAG));
-
if (symstreq(form, "defun") || symstreq(form, "defmacro"))
{
enum namespace namespace = NS_FUNCTION;
@@ -182,10 +179,9 @@
if (symstreq(form, "defmacro"))
namespace = NS_MACRO;
-
struct local local;
int nargs;
- dasm_State *d = compile_function(args, namespace, env, &local, &nargs);
+ dasm_State *d = compile_function(cdr(args), namespace, env, &local, NULL, &nargs);
add_function(env, (char *)(car(args) ^ SYMBOL_TAG), link(&d),
nargs, namespace);
@@ -209,7 +205,7 @@
value_t expr = cdr(args);
for (; !nilp(expr); expr = cdr(expr))
{
- walk_and_alloc(local, expr);
+ walk_and_alloc(local, car(expr));
}
local_free(local, slot);
@@ -241,7 +237,6 @@
while (read1(is, &val))
{
- printval(val, 0);
compile_tl(val, &env);
}
@@ -278,6 +273,25 @@
}
}
+void compile_variable(struct variable *v, dasm_State *Dst)
+{
+ switch (v->type)
+ {
+ case V_ARGUMENT:
+ | mov eax, dword [ebp + (value_size * (v->number + 2))];
+ break;
+ case V_BOUND:
+ | mov eax, dword [ebp - ((v->number + 1) * value_size)];
+ break;
+ case V_FREE:
+ // edi is the closure context pointer
+ | mov eax, dword [edi + (v->number * value_size)];
+ break;
+ default:
+ err("Sorry, can only access V_ARGUMENT, V_FREE and V_BOUND variables for now :(");
+ }
+}
+
void compile_expression(struct environment *env, struct local *local,
value_t val, dasm_State **Dst)
{
@@ -401,6 +415,55 @@
| pop eax;
}
+ else if (symstreq(fsym, "lambda"))
+ {
+ // Compile the function with this as the parent scope
+ struct local new_local;
+ int nargs_out;
+ dasm_State *d = compile_function(args, NS_ANONYMOUS, env, &new_local, local, &nargs_out);
+
+ // Link the function
+ void *func_ptr = link(&d);
+
+ // Create a closure object with the correct number of captures at
+ // runtime
+ | mov ebx, (create_closure);
+ | push (new_local.num_closure_slots);
+ | push (nargs_out);
+ | push (func_ptr);
+ | call ebx;
+ | add esp, 12;
+
+ // Walk the generated local scope for V_FREE variables, since each
+ // of these exists in our scope (or higher), evaluate it and set it
+ // as a member of the lambda capture.
+
+ for (struct variable *var = new_local.first; var; var = var->prev)
+ {
+ if (var->type == V_FREE)
+ {
+ // Closure in eax
+ | push eax;
+ // Variable now in eax
+ compile_variable(find_variable(local, var->name), Dst);
+ | push eax;
+
+ | mov ebx, (set_closure_capture_variable);
+ // The capture offset
+ | push (var->number);
+ | call ebx;
+ // Skip the value and index
+ | add esp, 8;
+ // Pop the closure back in to eax
+ | pop eax;
+ }
+ }
+
+ // Closure is still in eax
+
+ dasm_free(&d);
+ free(new_local.stack_slots);
+ }
else
{
struct function *func =
@@ -432,16 +495,12 @@
{
value_t expanded_to = call_list(func, args);
- printf("Macro expanded to:\n");
- printval(expanded_to, 2);
-
compile_expression(env, local, expanded_to, Dst);
}
}
}
else if (symbolp(val))
{
- // For now ignore global variables, only search locally
struct variable *v = find_variable(local, (char *)(val ^ SYMBOL_TAG));
if (!v)
@@ -450,17 +509,7 @@
err("Variable unbound");
}
- switch (v->type)
- {
- case V_ARGUMENT:
- | mov eax, dword [ebp + (value_size * (v->number + 2))];
- break;
- case V_BOUND:
- | mov eax, dword [ebp - ((v->number + 1) * value_size)];
- break;
- default:
- err("Sorry, can only access V_ARGUMENT and V_BOUND variables for now :(");
- }
+ compile_variable(v, Dst);
}
}
@@ -508,12 +557,30 @@
for (; v && strcmp(v->name, name) != 0; v = v->prev)
{}
+ if (!v)
+ {
+ if (local->parent)
+ {
+ v = find_variable(local->parent, name);
+
+ if (v)
+ {
+ // We found this in a parent scope, add it as a V_FREE variable to skip the search.
+ v = add_variable(local, V_FREE, name, local->num_closure_slots++);
+ }
+ }
+ }
return v;
}
-extern value_t _call_list(void *addr, value_t list);
+extern value_t _call_list(void *addr, value_t list, value_t *edi);
value_t call_list(struct function *func, value_t list)
{
- return _call_list(func->code_ptr, list);
+ return _call_list(func->code_ptr, list, NULL);
+}
+
+value_t call_list_closure(struct closure *c, value_t list)
+{
+ return _call_list(c->function, list, c->data);
}
diff --git a/src/lisp/compiler.h b/src/lisp/compiler.h
index 0b61c14..07928d2 100644
--- a/src/lisp/compiler.h
+++ b/src/lisp/compiler.h
@@ -7,8 +7,12 @@
enum namespace
{
+ /// A function
NS_FUNCTION,
+ /// A macro
NS_MACRO,
+ /// An anonymous function (a lambda/closure)
+ NS_ANONYMOUS,
};
struct function
@@ -63,13 +67,39 @@
int npc;
int nextpc;
bool *stack_slots;
- int num_stack_slots, num_stack_entries;
+ /// Number of slots allocated in `stack_slots`
+ int num_stack_slots;
+ /// Number of entries used in `stack_slots`
+ int num_stack_entries;
+ /// Number of closure slots total (allocated as V_FREE variables)
+ int num_closure_slots;
};
void compile_expression(struct environment *env, struct local *local,
value_t val, dasm_State **Dst);
/**
+ * Compile a function
+ * @param args The function args and body, e.g. `((b c) d)`
+ * @param namespace The function namespace.
+ * @param env The environment.
+ * @param local_out The local environment generated for this function will be
+ * returned here. NULL if you do not care about it being returned (you probably
+ * should since you need to free the stack slot allocation map).
+ * @param local_parent Parent local environment, only needed for closures. NULL
+ * if no parent.
+ * @param nargs The number of arguments for this function will be returned here.
+ * NULL if you don't care about it.
+ * @returns The compiled function state. You should probably give this to
+ * `add_function` or something similar.
+ */
+struct dasm_State *compile_function(value_t args, enum namespace namespace,
+ struct environment *env, struct local *local_out,
+ struct local *local_parent, int *nargs);
+
+void compile_variable(struct variable *v, dasm_State *Dst);
+
+/**
* Compile a backquoted expression
*/
void compile_backquote(struct environment *env, struct local *local,
@@ -94,11 +124,17 @@
struct function *find_function(struct environment *env, char *name);
struct variable *add_variable(struct local *local, enum var_type type,
char *name, int number);
-// Might return null
+
+/**
+ * Find a variable in `local` with name `name`.
+ * @returns The variable, NULL if not found.
+ */
struct variable *find_variable(struct local *local, char *name);
+
void destroy_local(struct local *local);
/**
* Like `apply` in lisp, calls func with list args and returns the result.
*/
value_t call_list(struct function *func, value_t list);
+value_t call_list_closure(struct closure *c, value_t list);
diff --git a/src/lisp/gc.c b/src/lisp/gc.c
index 16ec471..8820ef5 100644
--- a/src/lisp/gc.c
+++ b/src/lisp/gc.c
@@ -39,6 +39,16 @@
_mark(cons->cons.cdr, marked);
break;
}
+ case CLOSURE_TAG: {
+ struct closure_alloc *closure = (void *)alloc;
+
+ for (int i = 0; i < closure->closure.num_captured; i++)
+ {
+ _mark(closure->closure.data[i], marked);
+ }
+
+ break;
+ }
}
}
}
diff --git a/src/lisp/lib/std.c b/src/lisp/lib/std.c
index 485bf49..028dd3d 100644
--- a/src/lisp/lib/std.c
+++ b/src/lisp/lib/std.c
@@ -39,6 +39,17 @@
return nil;
}
+value_t l_apply(value_t func, value_t args)
+{
+ if (!closurep(func))
+ return nil;
+
+ if (!listp(args))
+ return nil;
+
+ return call_list_closure((struct closure *)(func ^ CLOSURE_TAG), args);
+}
+
void add_function(struct environment *env, char *name, void *func, int nargs, enum namespace ns)
{
struct function *last, *new = malloc(sizeof(struct function));
@@ -65,4 +76,6 @@
add_function(env, "cons", cons, 2, NS_FUNCTION);
add_function(env, "print", l_printval, 1, NS_FUNCTION);
+
+ add_function(env, "apply", l_apply, 2, NS_FUNCTION);
}
diff --git a/src/lisp/lib/std.h b/src/lisp/lib/std.h
index 5162bab..9070de8 100644
--- a/src/lisp/lib/std.h
+++ b/src/lisp/lib/std.h
@@ -4,6 +4,7 @@
#include "../lisp.h"
value_t l_plus(value_t a, value_t b);
+value_t l_printval(value_t val);
void add_function(struct environment *env, char *name, void *func, int nargs, enum namespace ns);
void load_std(struct environment *env);
diff --git a/src/lisp/lisp.c b/src/lisp/lisp.c
index 1056db7..b619033 100644
--- a/src/lisp/lisp.c
+++ b/src/lisp/lisp.c
@@ -73,7 +73,8 @@
if (is->peek(is) == ';')
{
while (is->get(is) != '\n')
- {}
+ {
+ }
// Only time I ever use labels is for stuff like this. Compiler would
// probably optimize this if I used recursion but I don't want to
@@ -234,6 +235,12 @@
{
printf("nil\n");
}
+ else if (closurep(v))
+ {
+ struct closure *c = (void *)(v ^ CLOSURE_TAG);
+ printf("closure %p taking %d argument(s) and capturing %d value(s)\n",
+ c->function, c->num_args, c->num_captured);
+ }
else
{
printf("<unknown %d>\n", v);
@@ -452,7 +459,12 @@
bool heapp(value_t v)
{
- return consp(v) || stringp(v) || symbolp(v);
+ return consp(v) || stringp(v) || symbolp(v) || closurep(v);
+}
+
+bool closurep(value_t v)
+{
+ return (v & HEAP_MASK) == CLOSURE_TAG;
}
bool listp(value_t v)
@@ -575,3 +587,27 @@
return c->name;
}
+
+value_t create_closure(void *code, int nargs, int ncaptures)
+{
+ struct closure_alloc *ca = malloc_aligned(sizeof(struct closure_alloc) +
+ ncaptures * sizeof(value_t));
+
+ ca->closure.function = code;
+ ca->closure.num_args = nargs;
+ ca->closure.num_captured = ncaptures;
+
+ add_this_alloc(&ca->alloc, CLOSURE_TAG);
+
+ return (value_t)(&ca->closure) | CLOSURE_TAG;
+}
+
+void set_closure_capture_variable(int index, value_t value, value_t closure)
+{
+ if (!closurep(closure))
+ return;
+
+ struct closure *c = (void *)(closure ^ CLOSURE_TAG);
+
+ c->data[index] = value;
+}
diff --git a/src/lisp/lisp.h b/src/lisp/lisp.h
index 675cc96..7c0e571 100644
--- a/src/lisp/lisp.h
+++ b/src/lisp/lisp.h
@@ -41,7 +41,11 @@
struct closure
{
+ /// How many arguments does this closure take
int num_args;
+ /// How many free variables does it capture (i.e. length of `data`)
+ int num_captured;
+ /// The function pointer itself
void *function;
/// This will be passed in edi.
@@ -153,6 +157,7 @@
bool listp(value_t v);
bool nilp(value_t v);
bool heapp(value_t v);
+bool closurep(value_t v);
int length(value_t v);
value_t elt(value_t v, int index);
@@ -162,5 +167,13 @@
bool symstreq(value_t sym, char *str);
+value_t create_closure(void *code, int nargs, int ncaptures);
+
+/**
+ * Set the `index`th capture variable of `closure`. This should really only be
+ * called when creating a new closure.
+ */
+void set_closure_capture_variable(int index, value_t value, value_t closure);
+
extern value_t nil;
extern value_t t;
diff --git a/src/lisp/test-closures.lisp b/src/lisp/test-closures.lisp
new file mode 100644
index 0000000..1311a98
--- /dev/null
+++ b/src/lisp/test-closures.lisp
@@ -0,0 +1,5 @@
+(defun main ()
+ (let1 (number 3)
+ (let1 (adds-3 (lambda (n)
+ (+ n number)))
+ (print (apply adds-3 '(4))))))