+ int size = ft->function.local_size;
+ struct variable *next = ft->function.scope;
+ struct variable *done = NULL;
+
+ while (next) {
+ struct variable *v = next;
+ struct type *t = v->type;
+ int pos;
+ next = v->in_scope;
+ if (v->merged != v)
+ continue;
+ if (!t)
+ continue; // NOTEST
+ if (v->frame_pos >= 0)
+ continue;
+ while (done && done->scope_end < v->scope_start)
+ done = done->in_scope;
+ if (done)
+ pos = done->frame_pos + done->type->size;
+ else
+ pos = ft->function.local_size;
+ if (pos & (t->align - 1))
+ pos = (pos + t->align) & ~(t->align-1);
+ v->frame_pos = pos;
+ if (size < pos + v->type->size)
+ size = pos + v->type->size;
+ v->in_scope = done;
+ done = v;
+ }
+ c->out_scope = NULL;
+ ft->function.local_size = size;
+ }
+
+###### free context storage
+ free(context.global);
+
+#### Variables as executables
+
+Just as we used a `val` to wrap a value into an `exec`, we similarly
+need a `var` to wrap a `variable` into an exec. While each `val`
+contained a copy of the value, each `var` holds a link to the variable
+because it really is the same variable no matter where it appears.
+When a variable is used, we need to remember to follow the `->merged`
+link to find the primary instance.
+
+When a variable is declared, it may or may not be given an explicit
+type. We need to record which so that we can report the parsed code
+correctly.
+
+###### exec type
+ Xvar,
+
+###### ast
+ struct var {
+ struct exec;
+ struct variable *var;
+ };
+
+###### variable fields
+ int explicit_type;
+
+###### Grammar
+
+ $TERM : ::
+
+ $*var
+ VariableDecl -> IDENTIFIER : ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v)
+ v->where_decl = $0;
+ else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER :: ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->constant = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER : Type ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->where_set = $0;
+ v->type = $<Type;
+ v->explicit_type = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER :: Type ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->where_set = $0;
+ v->type = $<Type;
+ v->constant = 1;
+ v->explicit_type = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+
+ $*exec
+ Variable -> IDENTIFIER ${ {
+ struct variable *v = var_ref(c, $1.txt);
+ $0 = new_pos(var, $1);
+ if (v == NULL) {
+ /* This might be a global const or a label
+ * Allocate a var with impossible type Tnone,
+ * which will be adjusted when we find out what it is,
+ * or will trigger an error.
+ */
+ v = var_decl(c, $1.txt);
+ if (v) {
+ v->type = Tnone;
+ v->where_decl = $0;
+ v->where_set = $0;
+ }
+ }
+ cast(var, $0)->var = v;
+ } }$
+
+###### print exec cases
+ case Xvar:
+ {
+ struct var *v = cast(var, e);
+ if (v->var) {
+ struct binding *b = v->var->name;
+ printf("%.*s", b->name.len, b->name.txt);
+ }
+ break;
+ }
+
+###### format cases
+ case 'v':
+ if (loc && loc->type == Xvar) {
+ struct var *v = cast(var, loc);
+ if (v->var) {
+ struct binding *b = v->var->name;
+ fprintf(stderr, "%.*s", b->name.len, b->name.txt);
+ } else
+ fputs("???", stderr); // NOTEST
+ } else
+ fputs("NOTVAR", stderr); // NOTEST
+ break;
+
+###### propagate exec cases
+
+ case Xvar:
+ {
+ struct var *var = cast(var, prog);
+ struct variable *v = var->var;
+ if (!v) {
+ type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
+ return Tnone; // NOTEST
+ }
+ v = v->merged;
+ if (v->type == Tnone && v->where_decl == prog)
+ type_err(c, "error: variable used but not declared: %v",
+ prog, NULL, 0, NULL);
+ if (v->type == NULL) {
+ if (type && !(*perr & Efail)) {
+ v->type = type;
+ v->where_set = prog;
+ *perr |= Eretry;
+ }
+ } else if (!type_compat(type, v->type, rules)) {
+ type_err(c, "error: expected %1 but variable '%v' is %2", prog,
+ type, rules, v->type);
+ type_err(c, "info: this is where '%v' was set to %1", v->where_set,
+ v->type, rules, NULL);
+ }
+ if (!v->global || v->frame_pos < 0)
+ *perr |= Eruntime;
+ if (v->constant)
+ *perr |= Econst;
+ return v->type;
+ }
+
+###### interp exec cases
+ case Xvar:
+ {
+ struct var *var = cast(var, e);
+ struct variable *v = var->var;
+
+ v = v->merged;
+ lrv = var_value(c, v);
+ rvtype = v->type;
+ break;
+ }
+
+###### ast functions
+
+ static void free_var(struct var *v)
+ {
+ free(v);
+ }
+
+###### free exec cases
+ case Xvar: free_var(cast(var, e)); break;
+
+### Complex types
+
+Now that we have the shape of the interpreter in place we can add some
+complex types and connected them in to the data structures and the
+different phases of parse, analyse, print, interpret.
+
+Being "complex" the language will naturally have syntax to access
+specifics of objects of these types. These will fit into the grammar as
+"Terms" which are the things that are combined with various operators to
+form an "Expression". Where a Term is formed by some operation on another
+Term, the subordinate Term will always come first, so for example a
+member of an array will be expressed as the Term for the array followed
+by an index in square brackets. The strict rule of using postfix
+operations makes precedence irrelevant within terms. To provide a place
+to put the grammar for terms of each type, we will start out by
+introducing the "Term" grammar production, with contains at least a
+simple "Value" (to be explained later).
+
+We also take this opportunity to introduce the "ExpressionsList" which
+is a simple comma-separated list of expressions - it may be used in
+various places.
+
+###### declare terminals
+ $TERM ,
+
+###### Grammar
+ $*exec
+ Term -> Value ${ $0 = $<1; }$
+ | Variable ${ $0 = $<1; }$
+ ## term grammar
+
+ $*binode
+ ExpressionList -> ExpressionList , Expression ${
+ $0 = new(binode);
+ $0->op = List;
+ $0->left = $<1;
+ $0->right = $<3;
+ }$
+ | Expression ${
+ $0 = new(binode);
+ $0->op = List;
+ $0->left = NULL;
+ $0->right = $<1;
+ }$
+
+Thus far the complex types we have are arrays, structs, functions and
+references.
+
+#### Arrays
+
+Arrays can be declared by giving a size and a type, as `[size]type' so
+`freq:[26]number` declares `freq` to be an array of 26 numbers. The
+size can be either a literal number, or a named constant. Some day an
+arbitrary expression will be supported.
+
+As a formal parameter to a function, the array can be declared with
+unknown size `name:[]string`. This is currently only supported for the
+"argv" parameter to "main" but will be extended more generally in a
+later version of the language. The length of this array - or any array
+- can be found with the "[]" postfix operator.
+
+Arrays cannot be assigned. When reference are extend to allow array
+slices which can refer to part or all of an array the assignment
+syntax will create a slice. For now, an array can only ever be
+referenced by the name it is declared with. It is likely that a
+"`copy`" primitive will eventually be defined which can be used to make a
+copy of an array with controllable recursive depth.
+
+For now we have two sorts of array, those with fixed size either because
+it is given as a literal number or because it is a struct member (which
+cannot have a runtime-changing size), and those with a size that is
+determined at runtime - local variables with a const size. The former
+have their size calculated at parse time, the latter at run time.
+
+For the latter type, the `size` field of the type is the size of a
+pointer, and the array is reallocated every time it comes into scope.
+
+We differentiate struct fields with a const size from local variables
+with a const size by whether they are prepared at parse time or not.
+
+###### type union fields
+
+ struct {
+ int unspec; // size is unspecified - vsize must be set.
+ short size;
+ short static_size;
+ struct variable *vsize;
+ struct type *member;
+ } array;
+
+###### value union fields
+ void *array; // used if not static_size
+
+###### value functions
+
+ static int array_prepare_type(struct parse_context *c, struct type *type,
+ int parse_time)
+ {
+ struct value *vsize;
+ mpz_t q;
+ if (type->array.static_size)
+ return 1; // NOTEST - guard against reentry
+ if (type->array.unspec && parse_time)
+ return 1; // NOTEST - unspec is still incomplete
+ if (parse_time && type->array.vsize && !type->array.vsize->global)
+ return 1; // NOTEST - should be impossible
+
+ if (type->array.vsize) {
+ vsize = var_value(c, type->array.vsize);
+ if (!vsize)
+ return 1; // NOTEST - should be impossible
+ mpz_init(q);
+ mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
+ type->array.size = mpz_get_si(q);
+ mpz_clear(q);
+ }
+ if (!parse_time)
+ return 1;
+ if (type->array.member->size <= 0)
+ return 0; // NOTEST - error caught before here
+
+ type->array.static_size = 1;
+ type->size = type->array.size * type->array.member->size;
+ type->align = type->array.member->align;
+
+ return 1;
+ }
+
+ static void array_init(struct type *type, struct value *val)
+ {
+ int i;
+ void *ptr = val->ptr;
+
+ if (!val)
+ return; // NOTEST
+ if (!type->array.static_size) {
+ val->array = calloc(type->array.size,
+ type->array.member->size);
+ ptr = val->array;
+ }
+ for (i = 0; i < type->array.size; i++) {
+ struct value *v;
+ v = (void*)ptr + i * type->array.member->size;
+ val_init(type->array.member, v);
+ }
+ }
+
+ static void array_free(struct type *type, struct value *val)
+ {
+ int i;
+ void *ptr = val->ptr;
+
+ if (!type->array.static_size)
+ ptr = val->array;
+ for (i = 0; i < type->array.size; i++) {
+ struct value *v;
+ v = (void*)ptr + i * type->array.member->size;
+ free_value(type->array.member, v);
+ }
+ if (!type->array.static_size)
+ free(ptr);
+ }
+
+ static int array_compat(struct type *require, struct type *have,
+ enum val_rules rules)
+ {
+ if (have->compat != require->compat)
+ return 0;
+ /* Both are arrays, so we can look at details */
+ if (!type_compat(require->array.member, have->array.member, 0))
+ return 0;
+ if (have->array.unspec && require->array.unspec &&
+ have->array.size != require->array.size)
+ return 0; // NOTEST
+ if (have->array.unspec || require->array.unspec)
+ return 1;
+ if (require->array.vsize == NULL && have->array.vsize == NULL)
+ return require->array.size == have->array.size;
+
+ return require->array.vsize == have->array.vsize;
+ }
+
+ static void array_print_type(struct type *type, FILE *f)
+ {
+ fputs("[", f);
+ if (type->array.vsize) {
+ struct binding *b = type->array.vsize->name;
+ fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
+ type->array.unspec ? "::" : "");
+ } else if (type->array.size)
+ fprintf(f, "%d]", type->array.size);
+ else
+ fprintf(f, "]");
+ type_print(type->array.member, f);
+ }
+
+ static struct type array_prototype = {
+ .init = array_init,
+ .prepare_type = array_prepare_type,
+ .print_type = array_print_type,
+ .compat = array_compat,
+ .free = array_free,
+ .size = sizeof(void*),
+ .align = sizeof(void*),
+ };
+
+###### declare terminals
+ $TERM [ ]
+
+###### type grammar
+
+ | [ NUMBER ] Type ${ {
+ char tail[3];
+ mpq_t num;
+ struct type *t;
+ int elements = 0;
+
+ if (number_parse(num, tail, $2.txt) == 0)
+ tok_err(c, "error: unrecognised number", &$2);
+ else if (tail[0]) {
+ tok_err(c, "error: unsupported number suffix", &$2);
+ mpq_clear(num);
+ } else {
+ elements = mpz_get_ui(mpq_numref(num));
+ if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
+ tok_err(c, "error: array size must be an integer",
+ &$2);
+ } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
+ tok_err(c, "error: array size is too large",
+ &$2);
+ mpq_clear(num);
+ }
+
+ $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
+ t->array.size = elements;
+ t->array.member = $<4;
+ t->array.vsize = NULL;
+ } }$
+
+ | [ IDENTIFIER ] Type ${ {
+ struct variable *v = var_ref(c, $2.txt);
+
+ if (!v)
+ tok_err(c, "error: name undeclared", &$2);
+ else if (!v->constant)
+ tok_err(c, "error: array size must be a constant", &$2);
+
+ $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
+ $0->array.member = $<4;
+ $0->array.size = 0;
+ $0->array.vsize = v;
+ } }$
+
+###### formal type grammar
+
+ | [ ] Type ${ {
+ $0 = add_anon_type(c, &array_prototype, "array[]");
+ $0->array.member = $<Type;
+ $0->array.size = 0;
+ $0->array.unspec = 1;
+ $0->array.vsize = NULL;
+ } }$
+
+###### Binode types
+ Index, Length,
+
+###### term grammar
+
+ | Term [ Expression ] ${ {
+ struct binode *b = new(binode);
+ b->op = Index;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+
+ | Term [ ] ${ {
+ struct binode *b = new(binode);
+ b->op = Length;
+ b->left = $<Term;
+ $0 = b;
+ } }$
+
+###### print binode cases
+ case Index:
+ print_exec(b->left, -1, bracket);
+ printf("[");
+ print_exec(b->right, -1, bracket);
+ printf("]");
+ break;
+
+ case Length:
+ print_exec(b->left, -1, bracket);
+ printf("[]");
+ break;
+
+###### propagate binode cases
+ case Index:
+ /* left must be an array, right must be a number,
+ * result is the member type of the array
+ */
+ propagate_types(b->right, c, perr_local, Tnum, 0);
+ t = propagate_types(b->left, c, perr, NULL, 0);
+ if (!t || t->compat != array_compat) {
+ type_err(c, "error: %1 cannot be indexed", prog, t, 0,
+ NULL);
+ return NULL;
+ } else {
+ if (!type_compat(type, t->array.member, rules)) {
+ type_err(c, "error: have %1 but need %2", prog,
+ t->array.member, rules, type);
+ }
+ return t->array.member;
+ }
+ break;
+
+ case Length:
+ /* left must be an array, result is a number
+ */
+ t = propagate_types(b->left, c, perr, NULL, 0);
+ if (!t || t->compat != array_compat) {
+ type_err(c, "error: %1 cannot provide length", prog, t,
+ 0, NULL);
+ return NULL;
+ }
+ if (!type_compat(type, Tnum, rules))
+ type_err(c, "error: have %1 but need %2", prog,
+ Tnum, rules, type);
+ return Tnum;
+ break;
+
+###### interp binode cases
+ case Index: {
+ mpz_t q;
+ long i;
+ void *ptr;
+
+ lleft = linterp_exec(c, b->left, <ype);
+ right = interp_exec(c, b->right, &rtype);
+ mpz_init(q);
+ mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
+ i = mpz_get_si(q);
+ mpz_clear(q);
+
+ if (ltype->array.static_size)
+ ptr = lleft;
+ else
+ ptr = *(void**)lleft;
+ rvtype = ltype->array.member;
+ if (i >= 0 && i < ltype->array.size)
+ lrv = ptr + i * rvtype->size;
+ else
+ val_init(ltype->array.member, &rv); // UNSAFE
+ ltype = NULL;
+ break;
+ }
+ case Length: {
+ lleft = linterp_exec(c, b->left, <ype);
+ mpq_set_ui(rv.num, ltype->array.size, 1);
+ ltype = NULL;
+ rvtype = Tnum;
+ break;
+ }
+
+#### Structs
+
+A `struct` is a data-type that contains one or more other data-types.
+It differs from an array in that each member can be of a different
+type, and they are accessed by name rather than by number. Thus you
+cannot choose an element by calculation, you need to know what you
+want up-front.
+
+The language makes no promises about how a given structure will be
+stored in memory - it is free to rearrange fields to suit whatever
+criteria seems important.
+
+Structs are declared separately from program code - they cannot be
+declared in-line in a variable declaration like arrays can. A struct
+is given a name and this name is used to identify the type - the name
+is not prefixed by the word `struct` as it would be in C.
+
+Structs are only treated as the same if they have the same name.
+Simply having the same fields in the same order is not enough. This
+might change once we can create structure initializers from a list of
+values.
+
+Each component datum is identified much like a variable is declared,
+with a name, one or two colons, and a type. The type cannot be omitted
+as there is no opportunity to deduce the type from usage. An initial
+value can be given following an equals sign, so
+
+##### Example: a struct type
+
+ struct complex:
+ x:number = 0
+ y:number = 0
+
+would declare a type called "complex" which has two number fields,
+each initialised to zero.
+
+Struct will need to be declared separately from the code that uses
+them, so we will need to be able to print out the declaration of a
+struct when reprinting the whole program. So a `print_type_decl` type
+function will be needed.
+
+###### type union fields
+
+ struct {
+ int nfields;
+ struct field {
+ struct text name;
+ struct type *type;
+ struct value *init;
+ int offset;
+ } *fields; // This is created when field_list is analysed.
+ struct fieldlist {
+ struct fieldlist *prev;
+ struct field f;
+ struct exec *init;
+ } *field_list; // This is created during parsing
+ } structure;
+
+###### type functions
+ void (*print_type_decl)(struct type *type, FILE *f);
+ struct type *(*fieldref)(struct type *t, struct parse_context *c,
+ struct fieldref *f, struct value **vp);
+
+###### value functions
+
+ static void structure_init(struct type *type, struct value *val)
+ {
+ int i;
+
+ for (i = 0; i < type->structure.nfields; i++) {
+ struct value *v;
+ v = (void*) val->ptr + type->structure.fields[i].offset;
+ if (type->structure.fields[i].init)
+ dup_value(type->structure.fields[i].type,
+ type->structure.fields[i].init,
+ v);
+ else
+ val_init(type->structure.fields[i].type, v);
+ }
+ }
+
+ static void structure_free(struct type *type, struct value *val)
+ {
+ int i;
+
+ for (i = 0; i < type->structure.nfields; i++) {
+ struct value *v;
+ v = (void*)val->ptr + type->structure.fields[i].offset;
+ free_value(type->structure.fields[i].type, v);
+ }
+ }
+
+ static void free_fieldlist(struct fieldlist *f)
+ {
+ if (!f)
+ return;
+ free_fieldlist(f->prev);
+ free_exec(f->init);
+ free(f);
+ }
+
+ static void structure_free_type(struct type *t)
+ {
+ int i;
+ for (i = 0; i < t->structure.nfields; i++)
+ if (t->structure.fields[i].init) {
+ free_value(t->structure.fields[i].type,
+ t->structure.fields[i].init);
+ }
+ free(t->structure.fields);
+ free_fieldlist(t->structure.field_list);
+ }
+
+ static int structure_prepare_type(struct parse_context *c,
+ struct type *t, int parse_time)
+ {
+ int cnt = 0;
+ struct fieldlist *f;
+
+ if (!parse_time || t->structure.fields)
+ return 1;
+
+ for (f = t->structure.field_list; f; f=f->prev) {
+ enum prop_err perr;
+ cnt += 1;
+
+ if (f->f.type->size <= 0)
+ return 0;
+ if (f->f.type->prepare_type)
+ f->f.type->prepare_type(c, f->f.type, parse_time);
+
+ if (f->init == NULL)
+ continue;
+ do {
+ perr = 0;
+ propagate_types(f->init, c, &perr, f->f.type, 0);
+ } while (perr & Eretry);
+ if (perr & Efail)
+ c->parse_error += 1; // NOTEST
+ }
+
+ t->structure.nfields = cnt;
+ t->structure.fields = calloc(cnt, sizeof(struct field));
+ f = t->structure.field_list;
+ while (cnt > 0) {
+ int a = f->f.type->align;
+ cnt -= 1;
+ t->structure.fields[cnt] = f->f;
+ if (t->size & (a-1))
+ t->size = (t->size | (a-1)) + 1;
+ t->structure.fields[cnt].offset = t->size;
+ t->size += ((f->f.type->size - 1) | (a-1)) + 1;
+ if (a > t->align)
+ t->align = a;
+
+ if (f->init && !c->parse_error) {
+ struct value vl = interp_exec(c, f->init, NULL);
+ t->structure.fields[cnt].init =
+ global_alloc(c, f->f.type, NULL, &vl);
+ }
+
+ f = f->prev;
+ }
+ return 1;
+ }
+
+ static int find_struct_index(struct type *type, struct text field)
+ {
+ int i;
+ for (i = 0; i < type->structure.nfields; i++)
+ if (text_cmp(type->structure.fields[i].name, field) == 0)
+ return i;
+ return IndexInvalid;
+ }
+
+ static struct type *structure_fieldref(struct type *t, struct parse_context *c,
+ struct fieldref *f, struct value **vp)
+ {
+ if (f->index == IndexUnknown) {
+ f->index = find_struct_index(t, f->name);
+ if (f->index < 0)
+ type_err(c, "error: cannot find requested field in %1",
+ f->left, t, 0, NULL);
+ }
+ if (f->index < 0)
+ return NULL;
+ if (vp) {
+ struct value *v = *vp;
+ v = (void*)v->ptr + t->structure.fields[f->index].offset;
+ *vp = v;
+ }
+ return t->structure.fields[f->index].type;
+ }
+
+ static struct type structure_prototype = {
+ .init = structure_init,
+ .free = structure_free,
+ .free_type = structure_free_type,
+ .print_type_decl = structure_print_type,
+ .prepare_type = structure_prepare_type,
+ .fieldref = structure_fieldref,
+ };
+
+###### exec type
+ Xfieldref,
+
+###### ast
+ struct fieldref {
+ struct exec;
+ struct exec *left;
+ int index;
+ struct text name;
+ };
+ enum { IndexUnknown = -1, IndexInvalid = -2 };
+
+###### free exec cases
+ case Xfieldref:
+ free_exec(cast(fieldref, e)->left);
+ free(e);
+ break;
+
+###### declare terminals
+ $TERM struct
+
+###### term grammar
+
+ | Term . IDENTIFIER ${ {
+ struct fieldref *fr = new_pos(fieldref, $2);
+ fr->left = $<1;
+ fr->name = $3.txt;
+ fr->index = IndexUnknown;
+ $0 = fr;
+ } }$
+
+###### print exec cases
+
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, e);
+ print_exec(f->left, -1, bracket);
+ printf(".%.*s", f->name.len, f->name.txt);
+ break;
+ }
+
+###### propagate exec cases
+
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, prog);
+ struct type *st = propagate_types(f->left, c, perr, NULL, 0);
+
+ if (!st || !st->fieldref)
+ type_err(c, "error: field reference on %1 is not supported",
+ f->left, st, 0, NULL);
+ else {
+ t = st->fieldref(st, c, f, NULL);
+ if (t && !type_compat(type, t, rules))
+ type_err(c, "error: have %1 but need %2", prog,
+ t, rules, type);
+ return t;
+ }
+ break;
+ }
+
+###### interp exec cases
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, e);
+ struct type *ltype;
+ struct value *lleft = linterp_exec(c, f->left, <ype);
+ lrv = lleft;
+ rvtype = ltype->fieldref(ltype, c, f, &lrv);
+ break;
+ }
+
+###### top level grammar
+ $*type
+ StructName -> IDENTIFIER ${ {
+ struct type *t = find_type(c, $ID.txt);
+
+ if (t && t->size >= 0) {
+ tok_err(c, "error: type already declared", &$ID);
+ tok_err(c, "info: this is location of declaration",
+ &t->first_use);
+ t = NULL;
+ }
+ if (!t)
+ t = add_type(c, $ID.txt, NULL);
+ t->first_use = $ID;
+ $0 = t;
+ } }$
+ $void
+ DeclareStruct -> struct StructName FieldBlock Newlines ${ {
+ struct type *t = $<SN;
+ struct type tmp = *t;
+
+ *t = structure_prototype;
+ t->name = tmp.name;
+ t->next = tmp.next;
+ t->first_use = tmp.first_use;
+
+ t->structure.field_list = $<FB;
+ } }$
+
+ $*fieldlist
+ FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
+ | { SimpleFieldList } ${ $0 = $<SFL; }$
+ | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
+ | SimpleFieldList EOL ${ $0 = $<SFL; }$
+
+ FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
+ | FieldLines SimpleFieldList Newlines ${ {
+ struct fieldlist *f = $<SFL;
+
+ if (f) {
+ $0 = f;
+ while (f->prev)
+ f = f->prev;
+ f->prev = $<FL;
+ } else
+ $0 = $<FL;
+ } }$
+
+ SimpleFieldList -> Field ${ $0 = $<F; }$
+ | SimpleFieldList ; Field ${
+ $F->prev = $<SFL;
+ $0 = $<F;
+ }$
+ | SimpleFieldList ; ${
+ $0 = $<SFL;
+ }$
+ | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
+
+ Field -> IDENTIFIER : Type = Expression ${ {
+ $0 = calloc(1, sizeof(struct fieldlist));
+ $0->f.name = $ID.txt;
+ $0->f.type = $<Type;
+ $0->f.init = NULL;
+ $0->init = $<Expr;
+ } }$
+ | IDENTIFIER : Type ${
+ $0 = calloc(1, sizeof(struct fieldlist));
+ $0->f.name = $ID.txt;
+ $0->f.type = $<Type;
+ }$
+
+###### forward decls
+ static void structure_print_type(struct type *t, FILE *f);
+
+###### value functions
+ static void structure_print_type(struct type *t, FILE *f)
+ {
+ int i;
+
+ fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
+
+ for (i = 0; i < t->structure.nfields; i++) {
+ struct field *fl = t->structure.fields + i;
+ fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
+ type_print(fl->type, f);
+ if (fl->type->print && fl->init) {
+ fprintf(f, " = ");
+ if (fl->type == Tstr)
+ fprintf(f, "\"");
+ print_value(fl->type, fl->init, f);
+ if (fl->type == Tstr)
+ fprintf(f, "\"");
+ }
+ fprintf(f, "\n");
+ }
+ }
+
+###### print type decls
+ {
+ struct type *t;
+ int target = -1;
+
+ while (target != 0) {
+ int i = 0;
+ for (t = context.typelist; t ; t=t->next)
+ if (!t->anon && t->print_type_decl &&
+ !t->check_args) {
+ i += 1;
+ if (i == target)
+ break;
+ }
+
+ if (target == -1) {
+ target = i;
+ } else {
+ t->print_type_decl(t, stdout);
+ target -= 1;
+ }
+ }
+ }
+
+#### References
+
+References, or pointers, are values that refer to another value. They
+can only refer to a type that is named, which excludes arrays or other
+references. As these can be included in a struct which is named, it is
+still possible to reference an array or reference - though indirectly.
+
+References are potentially dangerous as they might refer to some
+variable which no longer exists - either because a stack frame
+containing it has been discarded or because the value was allocated on
+the heap and has now been free. Ocean does not yet provide any
+protection against these problems. It will in due course.
+
+With references comes the opportunity and the need to explicitly
+allocate values on the "heap" and to free them. We currently provide
+fairly basic support for this.
+
+Reference make use of the `@` symbol in various ways. A type that starts
+with `@` is a reference to whatever follows. A reference value
+followed by an `@` acts as the referred value, though the `@` is often
+not needed. Finally, an expression that starts with `@` is a special
+reference related expression. Some examples might help.
+
+##### Example: Reference examples
+
+ struct foo
+ a: number
+ b: string
+ ref: @foo
+ bar: foo
+ bar.number = 23; bar.string = "hello"
+ baz: foo
+ ref = bar
+ baz = @ref
+ baz.a = ref.a * 2
+
+ ref = @new()
+ ref@ = baz
+ @free = ref
+ ref = @nil
+
+Obviously this is very contrived. `ref` is a reference to a `foo` which
+is initially set to refer to the value stored in `bar` - no extra syntax
+is needed to "Take the address of" `bar` - the fact that `ref` is a
+reference means that only the address make sense.
+
+When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
+The same syntax is used for accessing fields both in structs and in
+references to structs. It would be correct to use `ref@.a`, but not
+necessary.
+
+`@new()` creates an object of whatever type is needed for the program to
+by type-correct. In future iterations of Ocean, a constructor will
+access arguments, so the the syntax now looks like a function call.
+`@free` can be assigned any reference that was returned by `@new()`, and
+it will be freed. `@nil` is a value of whatever reference type is
+appropriate, and is stable and never the address of anything in the heap
+or on the stack. A reference can be assigned `nil` or compared against
+that value.
+
+###### declare terminals
+ $TERM @
+
+###### type union fields
+
+ struct {
+ struct type *referent;
+ } reference;
+
+###### value union fields
+ struct value *ref;
+
+###### value functions
+
+ static void reference_print_type(struct type *t, FILE *f)
+ {
+ fprintf(f, "@");
+ type_print(t->reference.referent, f);
+ }
+
+ static int reference_cmp(struct type *tl, struct type *tr,
+ struct value *left, struct value *right)
+ {
+ return left->ref == right->ref ? 0 : 1;
+ }
+
+ static void reference_dup(struct type *t,
+ struct value *vold, struct value *vnew)
+ {
+ vnew->ref = vold->ref;
+ }
+
+ static void reference_free(struct type *t, struct value *v)
+ {
+ /* Nothing to do here */
+ }
+
+ static int reference_compat(struct type *require, struct type *have,
+ enum val_rules rules)
+ {
+ if (rules & Rrefok)
+ if (require->reference.referent == have)
+ return 1;
+ if (have->compat != require->compat)
+ return 0;
+ if (have->reference.referent != require->reference.referent)
+ return 0;
+ return 1;
+ }
+
+ static int reference_test(struct type *type, struct value *val)
+ {
+ return val->ref != NULL;
+ }
+
+ static struct type *reference_fieldref(
+ struct type *t, struct parse_context *c, struct fieldref *f,
+ struct value **vp)
+ {
+ struct type *rt = t->reference.referent;
+
+ if (rt->fieldref) {
+ if (vp)
+ *vp = (*vp)->ref;
+ return rt->fieldref(rt, c, f, vp);
+ }
+ type_err(c, "error: field reference on %1 is not supported",
+ f->left, rt, 0, NULL);
+ return Tnone;