1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
114 struct parse_context {
115 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s = NULL, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
240 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
241 goto cleanup; // NOTEST
244 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
246 resolve_consts(&context);
247 prepare_types(&context);
248 if (!context.parse_error && !analyse_funcs(&context)) {
249 fprintf(stderr, "oceani: type error in program - not running.\n");
250 context.parse_error += 1;
258 if (doexec && !context.parse_error)
259 interp_main(&context, argc - optind, argv + optind);
262 struct section *t = s->next;
267 // FIXME parser should pop scope even on error
268 while (context.scope_depth > 0)
272 ## free context types
273 ## free context storage
274 exit(context.parse_error ? 1 : 0);
279 The four requirements of parse, analyse, print, interpret apply to
280 each language element individually so that is how most of the code
283 Three of the four are fairly self explanatory. The one that requires
284 a little explanation is the analysis step.
286 The current language design does not require the types of variables to
287 be declared, but they must still have a single type. Different
288 operations impose different requirements on the variables, for example
289 addition requires both arguments to be numeric, and assignment
290 requires the variable on the left to have the same type as the
291 expression on the right.
293 Analysis involves propagating these type requirements around and
294 consequently setting the type of each variable. If any requirements
295 are violated (e.g. a string is compared with a number) or if a
296 variable needs to have two different types, then an error is raised
297 and the program will not run.
299 If the same variable is declared in both branchs of an 'if/else', or
300 in all cases of a 'switch' then the multiple instances may be merged
301 into just one variable if the variable is referenced after the
302 conditional statement. When this happens, the types must naturally be
303 consistent across all the branches. When the variable is not used
304 outside the if, the variables in the different branches are distinct
305 and can be of different types.
307 Undeclared names may only appear in "use" statements and "case" expressions.
308 These names are given a type of "label" and a unique value.
309 This allows them to fill the role of a name in an enumerated type, which
310 is useful for testing the `switch` statement.
312 As we will see, the condition part of a `while` statement can return
313 either a Boolean or some other type. This requires that the expected
314 type that gets passed around comprises a type and a flag to indicate
315 that `Tbool` is also permitted.
317 As there are, as yet, no distinct types that are compatible, there
318 isn't much subtlety in the analysis. When we have distinct number
319 types, this will become more interesting.
323 When analysis discovers an inconsistency it needs to report an error;
324 just refusing to run the code ensures that the error doesn't cascade,
325 but by itself it isn't very useful. A clear understanding of the sort
326 of error message that are useful will help guide the process of
329 At a simplistic level, the only sort of error that type analysis can
330 report is that the type of some construct doesn't match a contextual
331 requirement. For example, in `4 + "hello"` the addition provides a
332 contextual requirement for numbers, but `"hello"` is not a number. In
333 this particular example no further information is needed as the types
334 are obvious from local information. When a variable is involved that
335 isn't the case. It may be helpful to explain why the variable has a
336 particular type, by indicating the location where the type was set,
337 whether by declaration or usage.
339 Using a recursive-descent analysis we can easily detect a problem at
340 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
341 will detect that one argument is not a number and the usage of `hello`
342 will detect that a number was wanted, but not provided. In this
343 (early) version of the language, we will generate error reports at
344 multiple locations, so the use of `hello` will report an error and
345 explain were the value was set, and the addition will report an error
346 and say why numbers are needed. To be able to report locations for
347 errors, each language element will need to record a file location
348 (line and column) and each variable will need to record the language
349 element where its type was set. For now we will assume that each line
350 of an error message indicates one location in the file, and up to 2
351 types. So we provide a `printf`-like function which takes a format, a
352 location (a `struct exec` which has not yet been introduced), and 2
353 types. "`%1`" reports the first type, "`%2`" reports the second. We
354 will need a function to print the location, once we know how that is
355 stored. e As will be explained later, there are sometimes extra rules for
356 type matching and they might affect error messages, we need to pass those
359 As well as type errors, we sometimes need to report problems with
360 tokens, which might be unexpected or might name a type that has not
361 been defined. For these we have `tok_err()` which reports an error
362 with a given token. Each of the error functions sets the flag in the
363 context so indicate that parsing failed.
367 static void fput_loc(struct exec *loc, FILE *f);
368 static void type_err(struct parse_context *c,
369 char *fmt, struct exec *loc,
370 struct type *t1, int rules, struct type *t2);
371 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
373 ###### core functions
375 static void type_err(struct parse_context *c,
376 char *fmt, struct exec *loc,
377 struct type *t1, int rules, struct type *t2)
379 fprintf(stderr, "%s:", c->file_name);
380 fput_loc(loc, stderr);
381 for (; *fmt ; fmt++) {
388 case '%': fputc(*fmt, stderr); break; // NOTEST
389 default: fputc('?', stderr); break; // NOTEST
391 type_print(t1, stderr);
394 type_print(t2, stderr);
403 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
405 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
406 t->txt.len, t->txt.txt);
410 ## Entities: declared and predeclared.
412 There are various "things" that the language and/or the interpreter
413 needs to know about to parse and execute a program. These include
414 types, variables, values, and executable code. These are all lumped
415 together under the term "entities" (calling them "objects" would be
416 confusing) and introduced here. The following section will present the
417 different specific code elements which comprise or manipulate these
422 Executables can be lots of different things. In many cases an
423 executable is just an operation combined with one or two other
424 executables. This allows for expressions and lists etc. Other times an
425 executable is something quite specific like a constant or variable name.
426 So we define a `struct exec` to be a general executable with a type, and
427 a `struct binode` which is a subclass of `exec`, forms a node in a
428 binary tree, and holds an operation. There will be other subclasses,
429 and to access these we need to be able to `cast` the `exec` into the
430 various other types. The first field in any `struct exec` is the type
431 from the `exec_types` enum.
434 #define cast(structname, pointer) ({ \
435 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
436 if (__mptr && *__mptr != X##structname) abort(); \
437 (struct structname *)( (char *)__mptr);})
439 #define new(structname) ({ \
440 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
441 __ptr->type = X##structname; \
442 __ptr->line = -1; __ptr->column = -1; \
445 #define new_pos(structname, token) ({ \
446 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
447 __ptr->type = X##structname; \
448 __ptr->line = token.line; __ptr->column = token.col; \
457 enum exec_types type;
466 struct exec *left, *right;
471 static int __fput_loc(struct exec *loc, FILE *f)
475 if (loc->line >= 0) {
476 fprintf(f, "%d:%d: ", loc->line, loc->column);
479 if (loc->type == Xbinode)
480 return __fput_loc(cast(binode,loc)->left, f) ||
481 __fput_loc(cast(binode,loc)->right, f); // NOTEST
484 static void fput_loc(struct exec *loc, FILE *f)
486 if (!__fput_loc(loc, f))
487 fprintf(f, "??:??: "); // NOTEST
490 Each different type of `exec` node needs a number of functions defined,
491 a bit like methods. We must be able to free it, print it, analyse it
492 and execute it. Once we have specific `exec` types we will need to
493 parse them too. Let's take this a bit more slowly.
497 The parser generator requires a `free_foo` function for each struct
498 that stores attributes and they will often be `exec`s and subtypes
499 there-of. So we need `free_exec` which can handle all the subtypes,
500 and we need `free_binode`.
504 static void free_binode(struct binode *b)
513 ###### core functions
514 static void free_exec(struct exec *e)
525 static void free_exec(struct exec *e);
527 ###### free exec cases
528 case Xbinode: free_binode(cast(binode, e)); break;
532 Printing an `exec` requires that we know the current indent level for
533 printing line-oriented components. As will become clear later, we
534 also want to know what sort of bracketing to use.
538 static void do_indent(int i, char *str)
545 ###### core functions
546 static void print_binode(struct binode *b, int indent, int bracket)
550 ## print binode cases
554 static void print_exec(struct exec *e, int indent, int bracket)
560 print_binode(cast(binode, e), indent, bracket); break;
565 do_indent(indent, "/* FREE");
566 for (v = e->to_free; v; v = v->next_free) {
567 printf(" %.*s", v->name->name.len, v->name->name.txt);
568 printf("[%d,%d]", v->scope_start, v->scope_end);
569 if (v->frame_pos >= 0)
570 printf("(%d+%d)", v->frame_pos,
571 v->type ? v->type->size:0);
579 static void print_exec(struct exec *e, int indent, int bracket);
583 As discussed, analysis involves propagating type requirements around the
584 program and looking for errors.
586 So `propagate_types` is passed an expected type (being a `struct type`
587 pointer together with some `val_rules` flags) that the `exec` is
588 expected to return, and returns the type that it does return, either of
589 which can be `NULL` signifying "unknown". A `prop_err` flag set is
590 passed by reference. It has `Efail` set when an error is found, and
591 `Eretry` when the type for some element is set via propagation. If
592 any expression cannot be evaluated immediately, `Enoconst` is set.
593 If the expression can be copied, `Emaycopy` is set.
595 If it remains unchanged at `0`, then no more propagation is needed.
599 enum val_rules {Rboolok = 1<<1, Rnoconstant = 1<<2};
600 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Enoconst = 1<<2,
604 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
605 struct type *type, int rules);
606 ###### core functions
608 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
609 struct type *type, int rules)
616 switch (prog->type) {
619 struct binode *b = cast(binode, prog);
621 ## propagate binode cases
625 ## propagate exec cases
630 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
631 struct type *type, int rules)
633 int pre_err = c->parse_error;
634 struct type *ret = __propagate_types(prog, c, perr, type, rules);
636 if (c->parse_error > pre_err)
643 Interpreting an `exec` doesn't require anything but the `exec`. State
644 is stored in variables and each variable will be directly linked from
645 within the `exec` tree. The exception to this is the `main` function
646 which needs to look at command line arguments. This function will be
647 interpreted separately.
649 Each `exec` can return a value combined with a type in `struct lrval`.
650 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
651 the location of a value, which can be updated, in `lval`. Others will
652 set `lval` to NULL indicating that there is a value of appropriate type
656 static struct value interp_exec(struct parse_context *c, struct exec *e,
657 struct type **typeret);
658 ###### core functions
662 struct value rval, *lval;
665 /* If dest is passed, dtype must give the expected type, and
666 * result can go there, in which case type is returned as NULL.
668 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
669 struct value *dest, struct type *dtype);
671 static struct value interp_exec(struct parse_context *c, struct exec *e,
672 struct type **typeret)
674 struct lrval ret = _interp_exec(c, e, NULL, NULL);
676 if (!ret.type) abort();
680 dup_value(ret.type, ret.lval, &ret.rval);
684 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
685 struct type **typeret)
687 struct lrval ret = _interp_exec(c, e, NULL, NULL);
689 if (!ret.type) abort();
693 free_value(ret.type, &ret.rval);
697 /* dinterp_exec is used when the destination type is certain and
698 * the value has a place to go.
700 static void dinterp_exec(struct parse_context *c, struct exec *e,
701 struct value *dest, struct type *dtype,
704 struct lrval ret = _interp_exec(c, e, dest, dtype);
708 free_value(dtype, dest);
710 dup_value(dtype, ret.lval, dest);
712 memcpy(dest, &ret.rval, dtype->size);
715 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
716 struct value *dest, struct type *dtype)
718 /* If the result is copied to dest, ret.type is set to NULL */
720 struct value rv = {}, *lrv = NULL;
723 rvtype = ret.type = Tnone;
733 struct binode *b = cast(binode, e);
734 struct value left, right, *lleft;
735 struct type *ltype, *rtype;
736 ltype = rtype = Tnone;
738 ## interp binode cases
740 free_value(ltype, &left);
741 free_value(rtype, &right);
751 ## interp exec cleanup
757 Values come in a wide range of types, with more likely to be added.
758 Each type needs to be able to print its own values (for convenience at
759 least) as well as to compare two values, at least for equality and
760 possibly for order. For now, values might need to be duplicated and
761 freed, though eventually such manipulations will be better integrated
764 Rather than requiring every numeric type to support all numeric
765 operations (add, multiply, etc), we allow types to be able to present
766 as one of a few standard types: integer, float, and fraction. The
767 existence of these conversion functions eventually enable types to
768 determine if they are compatible with other types, though such types
769 have not yet been implemented.
771 Named type are stored in a simple linked list. Objects of each type are
772 "values" which are often passed around by value.
774 There are both explicitly named types, and anonymous types. Anonymous
775 cannot be accessed by name, but are used internally and have a name
776 which might be reported in error messages.
783 ## value union fields
791 struct token first_use;
794 void (*init)(struct type *type, struct value *val);
795 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
796 void (*print)(struct type *type, struct value *val, FILE *f);
797 void (*print_type)(struct type *type, FILE *f);
798 int (*cmp_order)(struct type *t1, struct type *t2,
799 struct value *v1, struct value *v2);
800 int (*cmp_eq)(struct type *t1, struct type *t2,
801 struct value *v1, struct value *v2);
802 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
803 int (*test)(struct type *type, struct value *val);
804 void (*free)(struct type *type, struct value *val);
805 void (*free_type)(struct type *t);
806 long long (*to_int)(struct value *v);
807 double (*to_float)(struct value *v);
808 int (*to_mpq)(mpq_t *q, struct value *v);
817 struct type *typelist;
824 static struct type *find_type(struct parse_context *c, struct text s)
826 struct type *t = c->typelist;
828 while (t && (t->anon ||
829 text_cmp(t->name, s) != 0))
834 static struct type *_add_type(struct parse_context *c, struct text s,
835 struct type *proto, int anon)
839 n = calloc(1, sizeof(*n));
846 n->next = c->typelist;
851 static struct type *add_type(struct parse_context *c, struct text s,
854 return _add_type(c, s, proto, 0);
857 static struct type *add_anon_type(struct parse_context *c,
858 struct type *proto, char *name, ...)
864 vasprintf(&t.txt, name, ap);
866 t.len = strlen(t.txt);
867 return _add_type(c, t, proto, 1);
870 static struct type *find_anon_type(struct parse_context *c,
871 struct type *proto, char *name, ...)
873 struct type *t = c->typelist;
878 vasprintf(&nm.txt, name, ap);
880 nm.len = strlen(name);
882 while (t && (!t->anon ||
883 text_cmp(t->name, nm) != 0))
889 return _add_type(c, nm, proto, 1);
892 static void free_type(struct type *t)
894 /* The type is always a reference to something in the
895 * context, so we don't need to free anything.
899 static void free_value(struct type *type, struct value *v)
903 memset(v, 0x5a, type->size);
907 static void type_print(struct type *type, FILE *f)
910 fputs("*unknown*type*", f); // NOTEST
911 else if (type->name.len && !type->anon)
912 fprintf(f, "%.*s", type->name.len, type->name.txt);
913 else if (type->print_type)
914 type->print_type(type, f);
915 else if (type->name.len && type->anon)
916 fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
918 fputs("*invalid*type*", f); // NOTEST
921 static void val_init(struct type *type, struct value *val)
923 if (type && type->init)
924 type->init(type, val);
927 static void dup_value(struct type *type,
928 struct value *vold, struct value *vnew)
930 if (type && type->dup)
931 type->dup(type, vold, vnew);
934 static int value_cmp(struct type *tl, struct type *tr,
935 struct value *left, struct value *right)
937 if (tl && tl->cmp_order)
938 return tl->cmp_order(tl, tr, left, right);
939 if (tl && tl->cmp_eq)
940 return tl->cmp_eq(tl, tr, left, right);
944 static void print_value(struct type *type, struct value *v, FILE *f)
946 if (type && type->print)
947 type->print(type, v, f);
949 fprintf(f, "*Unknown*"); // NOTEST
952 static void prepare_types(struct parse_context *c)
956 enum { none, some, cannot } progress = none;
961 for (t = c->typelist; t; t = t->next) {
963 tok_err(c, "error: type used but not declared",
965 if (t->size == 0 && t->prepare_type) {
966 if (t->prepare_type(c, t, 1))
968 else if (progress == cannot)
969 tok_err(c, "error: type has recursive definition",
979 progress = cannot; break;
981 progress = none; break;
988 static void free_value(struct type *type, struct value *v);
989 static int type_compat(struct type *require, struct type *have, int rules);
990 static void type_print(struct type *type, FILE *f);
991 static void val_init(struct type *type, struct value *v);
992 static void dup_value(struct type *type,
993 struct value *vold, struct value *vnew);
994 static int value_cmp(struct type *tl, struct type *tr,
995 struct value *left, struct value *right);
996 static void print_value(struct type *type, struct value *v, FILE *f);
998 ###### free context types
1000 while (context.typelist) {
1001 struct type *t = context.typelist;
1003 context.typelist = t->next;
1011 Type can be specified for local variables, for fields in a structure,
1012 for formal parameters to functions, and possibly elsewhere. Different
1013 rules may apply in different contexts. As a minimum, a named type may
1014 always be used. Currently the type of a formal parameter can be
1015 different from types in other contexts, so we have a separate grammar
1021 Type -> IDENTIFIER ${
1022 $0 = find_type(c, $ID.txt);
1024 $0 = add_type(c, $ID.txt, NULL);
1025 $0->first_use = $ID;
1030 FormalType -> Type ${ $0 = $<1; }$
1031 ## formal type grammar
1035 Values of the base types can be numbers, which we represent as
1036 multi-precision fractions, strings, Booleans and labels. When
1037 analysing the program we also need to allow for places where no value
1038 is meaningful (type `Tnone`) and where we don't know what type to
1039 expect yet (type is `NULL`).
1041 Values are never shared, they are always copied when used, and freed
1042 when no longer needed.
1044 When propagating type information around the program, we need to
1045 determine if two types are compatible, where type `NULL` is compatible
1046 with anything. There are two special cases with type compatibility,
1047 both related to the Conditional Statement which will be described
1048 later. In some cases a Boolean can be accepted as well as some other
1049 primary type, and in others any type is acceptable except a label (`Vlabel`).
1050 A separate function encoding these cases will simplify some code later.
1052 ###### type functions
1054 int (*compat)(struct type *this, struct type *other);
1056 ###### ast functions
1058 static int type_compat(struct type *require, struct type *have, int rules)
1060 if ((rules & Rboolok) && have == Tbool)
1062 if (!require || !have)
1065 if (require->compat)
1066 return require->compat(require, have);
1068 return require == have;
1073 #include "parse_string.h"
1074 #include "parse_number.h"
1077 myLDLIBS := libnumber.o libstring.o -lgmp
1078 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1080 ###### type union fields
1081 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1083 ###### value union fields
1089 ###### ast functions
1090 static void _free_value(struct type *type, struct value *v)
1094 switch (type->vtype) {
1096 case Vstr: free(v->str.txt); break;
1097 case Vnum: mpq_clear(v->num); break;
1103 ###### value functions
1105 static void _val_init(struct type *type, struct value *val)
1107 switch(type->vtype) {
1108 case Vnone: // NOTEST
1111 mpq_init(val->num); break;
1113 val->str.txt = malloc(1);
1120 val->label = 0; // NOTEST
1125 static void _dup_value(struct type *type,
1126 struct value *vold, struct value *vnew)
1128 switch (type->vtype) {
1129 case Vnone: // NOTEST
1132 vnew->label = vold->label; // NOTEST
1135 vnew->bool = vold->bool;
1138 mpq_init(vnew->num);
1139 mpq_set(vnew->num, vold->num);
1142 vnew->str.len = vold->str.len;
1143 vnew->str.txt = malloc(vnew->str.len);
1144 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1149 static int _value_cmp(struct type *tl, struct type *tr,
1150 struct value *left, struct value *right)
1154 return tl - tr; // NOTEST
1155 switch (tl->vtype) {
1156 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1157 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1158 case Vstr: cmp = text_cmp(left->str, right->str); break;
1159 case Vbool: cmp = left->bool - right->bool; break;
1160 case Vnone: cmp = 0; // NOTEST
1165 static void _print_value(struct type *type, struct value *v, FILE *f)
1167 switch (type->vtype) {
1168 case Vnone: // NOTEST
1169 fprintf(f, "*no-value*"); break; // NOTEST
1170 case Vlabel: // NOTEST
1171 fprintf(f, "*label-%d*", v->label); break; // NOTEST
1173 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1175 fprintf(f, "%s", v->bool ? "True":"False"); break;
1180 mpf_set_q(fl, v->num);
1181 gmp_fprintf(f, "%.10Fg", fl);
1188 static void _free_value(struct type *type, struct value *v);
1190 static int bool_test(struct type *type, struct value *v)
1195 static struct type base_prototype = {
1197 .print = _print_value,
1198 .cmp_order = _value_cmp,
1199 .cmp_eq = _value_cmp,
1201 .free = _free_value,
1204 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1206 ###### ast functions
1207 static struct type *add_base_type(struct parse_context *c, char *n,
1208 enum vtype vt, int size)
1210 struct text txt = { n, strlen(n) };
1213 t = add_type(c, txt, &base_prototype);
1216 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1217 if (t->size & (t->align - 1))
1218 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1222 ###### context initialization
1224 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1225 Tbool->test = bool_test;
1226 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1227 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1228 Tnone = add_base_type(&context, "none", Vnone, 0);
1229 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1233 We have already met values as separate objects. When manifest constants
1234 appear in the program text, that must result in an executable which has
1235 a constant value. So the `val` structure embeds a value in an
1248 ###### ast functions
1249 struct val *new_val(struct type *T, struct token tk)
1251 struct val *v = new_pos(val, tk);
1256 ###### declare terminals
1263 $0 = new_val(Tbool, $1);
1267 $0 = new_val(Tbool, $1);
1272 $0 = new_val(Tnum, $1);
1273 if (number_parse($0->val.num, tail, $1.txt) == 0)
1274 mpq_init($0->val.num); // UNTESTED
1276 tok_err(c, "error: unsupported number suffix",
1281 $0 = new_val(Tstr, $1);
1282 string_parse(&$1, '\\', &$0->val.str, tail);
1284 tok_err(c, "error: unsupported string suffix",
1289 $0 = new_val(Tstr, $1);
1290 string_parse(&$1, '\\', &$0->val.str, tail);
1292 tok_err(c, "error: unsupported string suffix",
1296 ###### print exec cases
1299 struct val *v = cast(val, e);
1300 if (v->vtype == Tstr)
1302 // FIXME how to ensure numbers have same precision.
1303 print_value(v->vtype, &v->val, stdout);
1304 if (v->vtype == Tstr)
1309 ###### propagate exec cases
1312 struct val *val = cast(val, prog);
1313 if (!type_compat(type, val->vtype, rules))
1314 type_err(c, "error: expected %1 found %2",
1315 prog, type, rules, val->vtype);
1319 ###### interp exec cases
1321 rvtype = cast(val, e)->vtype;
1322 dup_value(rvtype, &cast(val, e)->val, &rv);
1325 ###### ast functions
1326 static void free_val(struct val *v)
1329 free_value(v->vtype, &v->val);
1333 ###### free exec cases
1334 case Xval: free_val(cast(val, e)); break;
1336 ###### ast functions
1337 // Move all nodes from 'b' to 'rv', reversing their order.
1338 // In 'b' 'left' is a list, and 'right' is the last node.
1339 // In 'rv', left' is the first node and 'right' is a list.
1340 static struct binode *reorder_bilist(struct binode *b)
1342 struct binode *rv = NULL;
1345 struct exec *t = b->right;
1349 b = cast(binode, b->left);
1359 Labels are a temporary concept until I implement enums. There are an
1360 anonymous enum which is declared by usage. Thet are only allowed in
1361 `use` statements and corresponding `case` entries. They appear as a
1362 period followed by an identifier. All identifiers that are "used" must
1365 For now, we have a global list of labels, and don't check that all "use"
1377 ###### free exec cases
1381 ###### print exec cases
1383 struct label *l = cast(label, e);
1384 printf(".%.*s", l->name.len, l->name.txt);
1390 struct labels *next;
1394 ###### parse context
1395 struct labels *labels;
1397 ###### ast functions
1398 static int label_lookup(struct parse_context *c, struct text name)
1400 struct labels *l, **lp = &c->labels;
1401 while (*lp && text_cmp((*lp)->name, name) < 0)
1403 if (*lp && text_cmp((*lp)->name, name) == 0)
1404 return (*lp)->value;
1405 l = calloc(1, sizeof(*l));
1408 if (c->next_label == 0)
1410 l->value = c->next_label;
1416 ###### free context storage
1417 while (context.labels) {
1418 struct labels *l = context.labels;
1419 context.labels = l->next;
1423 ###### declare terminals
1427 struct label *l = new_pos(label, $ID);
1431 ###### propagate exec cases
1433 struct label *l = cast(label, prog);
1434 l->value = label_lookup(c, l->name);
1435 if (!type_compat(type, Tlabel, rules))
1436 type_err(c, "error: expected %1 found %2",
1437 prog, type, rules, Tlabel);
1440 ###### interp exec cases
1442 struct label *l = cast(label, e);
1443 rv.label = l->value;
1451 Variables are scoped named values. We store the names in a linked list
1452 of "bindings" sorted in lexical order, and use sequential search and
1459 struct binding *next; // in lexical order
1463 This linked list is stored in the parse context so that "reduce"
1464 functions can find or add variables, and so the analysis phase can
1465 ensure that every variable gets a type.
1467 ###### parse context
1469 struct binding *varlist; // In lexical order
1471 ###### ast functions
1473 static struct binding *find_binding(struct parse_context *c, struct text s)
1475 struct binding **l = &c->varlist;
1480 (cmp = text_cmp((*l)->name, s)) < 0)
1484 n = calloc(1, sizeof(*n));
1491 Each name can be linked to multiple variables defined in different
1492 scopes. Each scope starts where the name is declared and continues
1493 until the end of the containing code block. Scopes of a given name
1494 cannot nest, so a declaration while a name is in-scope is an error.
1496 ###### binding fields
1497 struct variable *var;
1501 struct variable *previous;
1503 struct binding *name;
1504 struct exec *where_decl;// where name was declared
1505 struct exec *where_set; // where type was set
1509 When a scope closes, the values of the variables might need to be freed.
1510 This happens in the context of some `struct exec` and each `exec` will
1511 need to know which variables need to be freed when it completes.
1514 struct variable *to_free;
1516 ####### variable fields
1517 struct exec *cleanup_exec;
1518 struct variable *next_free;
1520 ####### interp exec cleanup
1523 for (v = e->to_free; v; v = v->next_free) {
1524 struct value *val = var_value(c, v);
1525 free_value(v->type, val);
1529 ###### ast functions
1530 static void variable_unlink_exec(struct variable *v)
1532 struct variable **vp;
1533 if (!v->cleanup_exec)
1535 for (vp = &v->cleanup_exec->to_free;
1536 *vp; vp = &(*vp)->next_free) {
1540 v->cleanup_exec = NULL;
1545 While the naming seems strange, we include local constants in the
1546 definition of variables. A name declared `var := value` can
1547 subsequently be changed, but a name declared `var ::= value` cannot -
1550 ###### variable fields
1553 Scopes in parallel branches can be partially merged. More
1554 specifically, if a given name is declared in both branches of an
1555 if/else then its scope is a candidate for merging. Similarly if
1556 every branch of an exhaustive switch (e.g. has an "else" clause)
1557 declares a given name, then the scopes from the branches are
1558 candidates for merging.
1560 Note that names declared inside a loop (which is only parallel to
1561 itself) are never visible after the loop. Similarly names defined in
1562 scopes which are not parallel, such as those started by `for` and
1563 `switch`, are never visible after the scope. Only variables defined in
1564 both `then` and `else` (including the implicit then after an `if`, and
1565 excluding `then` used with `for`) and in all `case`s and `else` of a
1566 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1568 Labels, which are a bit like variables, follow different rules.
1569 Labels are not explicitly declared, but if an undeclared name appears
1570 in a context where a label is legal, that effectively declares the
1571 name as a label. The declaration remains in force (or in scope) at
1572 least to the end of the immediately containing block and conditionally
1573 in any larger containing block which does not declare the name in some
1574 other way. Importantly, the conditional scope extension happens even
1575 if the label is only used in one parallel branch of a conditional --
1576 when used in one branch it is treated as having been declared in all
1579 Merge candidates are tentatively visible beyond the end of the
1580 branching statement which creates them. If the name is used, the
1581 merge is affirmed and they become a single variable visible at the
1582 outer layer. If not - if it is redeclared first - the merge lapses.
1584 To track scopes we have an extra stack, implemented as a linked list,
1585 which roughly parallels the parse stack and which is used exclusively
1586 for scoping. When a new scope is opened, a new frame is pushed and
1587 the child-count of the parent frame is incremented. This child-count
1588 is used to distinguish between the first of a set of parallel scopes,
1589 in which declared variables must not be in scope, and subsequent
1590 branches, whether they may already be conditionally scoped.
1592 We need a total ordering of scopes so we can easily compare to variables
1593 to see if they are concurrently in scope. To achieve this we record a
1594 `scope_count` which is actually a count of both beginnings and endings
1595 of scopes. Then each variable has a record of the scope count where it
1596 enters scope, and where it leaves.
1598 To push a new frame *before* any code in the frame is parsed, we need a
1599 grammar reduction. This is most easily achieved with a grammar
1600 element which derives the empty string, and creates the new scope when
1601 it is recognised. This can be placed, for example, between a keyword
1602 like "if" and the code following it.
1606 struct scope *parent;
1610 ###### parse context
1613 struct scope *scope_stack;
1615 ###### variable fields
1616 int scope_start, scope_end;
1618 ###### ast functions
1619 static void scope_pop(struct parse_context *c)
1621 struct scope *s = c->scope_stack;
1623 c->scope_stack = s->parent;
1625 c->scope_depth -= 1;
1626 c->scope_count += 1;
1629 static void scope_push(struct parse_context *c)
1631 struct scope *s = calloc(1, sizeof(*s));
1633 c->scope_stack->child_count += 1;
1634 s->parent = c->scope_stack;
1636 c->scope_depth += 1;
1637 c->scope_count += 1;
1643 OpenScope -> ${ scope_push(c); }$
1645 Each variable records a scope depth and is in one of four states:
1647 - "in scope". This is the case between the declaration of the
1648 variable and the end of the containing block, and also between
1649 the usage with affirms a merge and the end of that block.
1651 The scope depth is not greater than the current parse context scope
1652 nest depth. When the block of that depth closes, the state will
1653 change. To achieve this, all "in scope" variables are linked
1654 together as a stack in nesting order.
1656 - "pending". The "in scope" block has closed, but other parallel
1657 scopes are still being processed. So far, every parallel block at
1658 the same level that has closed has declared the name.
1660 The scope depth is the depth of the last parallel block that
1661 enclosed the declaration, and that has closed.
1663 - "conditionally in scope". The "in scope" block and all parallel
1664 scopes have closed, and no further mention of the name has been seen.
1665 This state includes a secondary nest depth (`min_depth`) which records
1666 the outermost scope seen since the variable became conditionally in
1667 scope. If a use of the name is found, the variable becomes "in scope"
1668 and that secondary depth becomes the recorded scope depth. If the
1669 name is declared as a new variable, the old variable becomes "out of
1670 scope" and the recorded scope depth stays unchanged.
1672 - "out of scope". The variable is neither in scope nor conditionally
1673 in scope. It is permanently out of scope now and can be removed from
1674 the "in scope" stack. When a variable becomes out-of-scope it is
1675 moved to a separate list (`out_scope`) of variables which have fully
1676 known scope. This will be used at the end of each function to assign
1677 each variable a place in the stack frame.
1679 ###### variable fields
1680 int depth, min_depth;
1681 enum { OutScope, PendingScope, CondScope, InScope } scope;
1682 struct variable *in_scope;
1684 ###### parse context
1686 struct variable *in_scope;
1687 struct variable *out_scope;
1689 All variables with the same name are linked together using the
1690 'previous' link. Those variable that have been affirmatively merged all
1691 have a 'merged' pointer that points to one primary variable - the most
1692 recently declared instance. When merging variables, we need to also
1693 adjust the 'merged' pointer on any other variables that had previously
1694 been merged with the one that will no longer be primary.
1696 A variable that is no longer the most recent instance of a name may
1697 still have "pending" scope, if it might still be merged with most
1698 recent instance. These variables don't really belong in the
1699 "in_scope" list, but are not immediately removed when a new instance
1700 is found. Instead, they are detected and ignored when considering the
1701 list of in_scope names.
1703 The storage of the value of a variable will be described later. For now
1704 we just need to know that when a variable goes out of scope, it might
1705 need to be freed. For this we need to be able to find it, so assume that
1706 `var_value()` will provide that.
1708 ###### variable fields
1709 struct variable *merged;
1711 ###### ast functions
1713 static void variable_merge(struct variable *primary, struct variable *secondary)
1717 primary = primary->merged;
1719 for (v = primary->previous; v; v=v->previous)
1720 if (v == secondary || v == secondary->merged ||
1721 v->merged == secondary ||
1722 v->merged == secondary->merged) {
1723 v->scope = OutScope;
1724 v->merged = primary;
1725 if (v->scope_start < primary->scope_start)
1726 primary->scope_start = v->scope_start;
1727 if (v->scope_end > primary->scope_end)
1728 primary->scope_end = v->scope_end; // NOTEST
1729 variable_unlink_exec(v);
1733 ###### forward decls
1734 static struct value *var_value(struct parse_context *c, struct variable *v);
1736 ###### free global vars
1738 while (context.varlist) {
1739 struct binding *b = context.varlist;
1740 struct variable *v = b->var;
1741 context.varlist = b->next;
1744 struct variable *next = v->previous;
1746 if (v->global && v->frame_pos >= 0) {
1747 free_value(v->type, var_value(&context, v));
1748 if (v->depth == 0 && v->type->free == function_free)
1749 // This is a function constant
1750 free_exec(v->where_decl);
1757 #### Manipulating Bindings
1759 When a name is conditionally visible, a new declaration discards the old
1760 binding - the condition lapses. Similarly when we reach the end of a
1761 function (outermost non-global scope) any conditional scope must lapse.
1762 Conversely a usage of the name affirms the visibility and extends it to
1763 the end of the containing block - i.e. the block that contains both the
1764 original declaration and the latest usage. This is determined from
1765 `min_depth`. When a conditionally visible variable gets affirmed like
1766 this, it is also merged with other conditionally visible variables with
1769 When we parse a variable declaration we either report an error if the
1770 name is currently bound, or create a new variable at the current nest
1771 depth if the name is unbound or bound to a conditionally scoped or
1772 pending-scope variable. If the previous variable was conditionally
1773 scoped, it and its homonyms becomes out-of-scope.
1775 When we parse a variable reference (including non-declarative assignment
1776 "foo = bar") we report an error if the name is not bound or is bound to
1777 a pending-scope variable; update the scope if the name is bound to a
1778 conditionally scoped variable; or just proceed normally if the named
1779 variable is in scope.
1781 When we exit a scope, any variables bound at this level are either
1782 marked out of scope or pending-scoped, depending on whether the scope
1783 was sequential or parallel. Here a "parallel" scope means the "then"
1784 or "else" part of a conditional, or any "case" or "else" branch of a
1785 switch. Other scopes are "sequential".
1787 When exiting a parallel scope we check if there are any variables that
1788 were previously pending and are still visible. If there are, then
1789 they weren't redeclared in the most recent scope, so they cannot be
1790 merged and must become out-of-scope. If it is not the first of
1791 parallel scopes (based on `child_count`), we check that there was a
1792 previous binding that is still pending-scope. If there isn't, the new
1793 variable must now be out-of-scope.
1795 When exiting a sequential scope that immediately enclosed parallel
1796 scopes, we need to resolve any pending-scope variables. If there was
1797 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1798 we need to mark all pending-scope variable as out-of-scope. Otherwise
1799 all pending-scope variables become conditionally scoped.
1802 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1804 ###### ast functions
1806 static struct variable *var_decl(struct parse_context *c, struct text s)
1808 struct binding *b = find_binding(c, s);
1809 struct variable *v = b->var;
1811 switch (v ? v->scope : OutScope) {
1813 /* Caller will report the error */
1817 v && v->scope == CondScope;
1819 v->scope = OutScope;
1823 v = calloc(1, sizeof(*v));
1824 v->previous = b->var;
1828 v->min_depth = v->depth = c->scope_depth;
1830 v->in_scope = c->in_scope;
1831 v->scope_start = c->scope_count;
1837 static struct variable *var_ref(struct parse_context *c, struct text s)
1839 struct binding *b = find_binding(c, s);
1840 struct variable *v = b->var;
1841 struct variable *v2;
1843 switch (v ? v->scope : OutScope) {
1846 /* Caller will report the error */
1849 /* All CondScope variables of this name need to be merged
1850 * and become InScope
1852 v->depth = v->min_depth;
1854 for (v2 = v->previous;
1855 v2 && v2->scope == CondScope;
1857 variable_merge(v, v2);
1865 static int var_refile(struct parse_context *c, struct variable *v)
1867 /* Variable just went out of scope. Add it to the out_scope
1868 * list, sorted by ->scope_start
1870 struct variable **vp = &c->out_scope;
1871 while ((*vp) && (*vp)->scope_start < v->scope_start)
1872 vp = &(*vp)->in_scope;
1878 static void var_block_close(struct parse_context *c, enum closetype ct,
1881 /* Close off all variables that are in_scope.
1882 * Some variables in c->scope may already be not-in-scope,
1883 * such as when a PendingScope variable is hidden by a new
1884 * variable with the same name.
1885 * So we check for v->name->var != v and drop them.
1886 * If we choose to make a variable OutScope, we drop it
1889 struct variable *v, **vp, *v2;
1892 for (vp = &c->in_scope;
1893 (v = *vp) && v->min_depth > c->scope_depth;
1894 (v->scope == OutScope || v->name->var != v)
1895 ? (*vp = v->in_scope, var_refile(c, v))
1896 : ( vp = &v->in_scope, 0)) {
1897 v->min_depth = c->scope_depth;
1898 if (v->name->var != v)
1899 /* This is still in scope, but we haven't just
1903 v->min_depth = c->scope_depth;
1904 if (v->scope == InScope)
1905 v->scope_end = c->scope_count;
1906 if (v->scope == InScope && e && !v->global) {
1907 /* This variable gets cleaned up when 'e' finishes */
1908 variable_unlink_exec(v);
1909 v->cleanup_exec = e;
1910 v->next_free = e->to_free;
1915 case CloseParallel: /* handle PendingScope */
1919 if (c->scope_stack->child_count == 1)
1920 /* first among parallel branches */
1921 v->scope = PendingScope;
1922 else if (v->previous &&
1923 v->previous->scope == PendingScope)
1924 /* all previous branches used name */
1925 v->scope = PendingScope;
1927 v->scope = OutScope;
1928 if (ct == CloseElse) {
1929 /* All Pending variables with this name
1930 * are now Conditional */
1932 v2 && v2->scope == PendingScope;
1934 v2->scope = CondScope;
1938 /* Not possible as it would require
1939 * parallel scope to be nested immediately
1940 * in a parallel scope, and that never
1944 /* Not possible as we already tested for
1951 if (v->scope == CondScope)
1952 /* Condition cannot continue past end of function */
1955 case CloseSequential:
1958 v->scope = OutScope;
1961 /* There was no 'else', so we can only become
1962 * conditional if we know the cases were exhaustive,
1963 * and that doesn't mean anything yet.
1964 * So only labels become conditional..
1967 v2 && v2->scope == PendingScope;
1969 v2->scope = OutScope;
1972 case OutScope: break;
1981 The value of a variable is store separately from the variable, on an
1982 analogue of a stack frame. There are (currently) two frames that can be
1983 active. A global frame which currently only stores constants, and a
1984 stacked frame which stores local variables. Each variable knows if it
1985 is global or not, and what its index into the frame is.
1987 Values in the global frame are known immediately they are relevant, so
1988 the frame needs to be reallocated as it grows so it can store those
1989 values. The local frame doesn't get values until the interpreted phase
1990 is started, so there is no need to allocate until the size is known.
1992 We initialize the `frame_pos` to an impossible value, so that we can
1993 tell if it was set or not later.
1995 ###### variable fields
1999 ###### variable init
2002 ###### parse context
2004 short global_size, global_alloc;
2006 void *global, *local;
2008 ###### forward decls
2009 static struct value *global_alloc(struct parse_context *c, struct type *t,
2010 struct variable *v, struct value *init);
2012 ###### ast functions
2014 static struct value *var_value(struct parse_context *c, struct variable *v)
2017 if (!c->local || !v->type)
2018 return NULL; // UNTESTED
2019 if (v->frame_pos + v->type->size > c->local_size) {
2020 printf("INVALID frame_pos\n"); // NOTEST
2023 return c->local + v->frame_pos;
2025 if (c->global_size > c->global_alloc) {
2026 int old = c->global_alloc;
2027 c->global_alloc = (c->global_size | 1023) + 1024;
2028 c->global = realloc(c->global, c->global_alloc);
2029 memset(c->global + old, 0, c->global_alloc - old);
2031 return c->global + v->frame_pos;
2034 static struct value *global_alloc(struct parse_context *c, struct type *t,
2035 struct variable *v, struct value *init)
2038 struct variable scratch;
2040 if (t->prepare_type)
2041 t->prepare_type(c, t, 1); // NOTEST
2043 if (c->global_size & (t->align - 1))
2044 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
2049 v->frame_pos = c->global_size;
2051 c->global_size += v->type->size;
2052 ret = var_value(c, v);
2054 memcpy(ret, init, t->size);
2056 val_init(t, ret); // NOTEST
2060 As global values are found -- struct field initializers, labels etc --
2061 `global_alloc()` is called to record the value in the global frame.
2063 When the program is fully parsed, each function is analysed, we need to
2064 walk the list of variables local to that function and assign them an
2065 offset in the stack frame. For this we have `scope_finalize()`.
2067 We keep the stack from dense by re-using space for between variables
2068 that are not in scope at the same time. The `out_scope` list is sorted
2069 by `scope_start` and as we process a varible, we move it to an FIFO
2070 stack. For each variable we consider, we first discard any from the
2071 stack anything that went out of scope before the new variable came in.
2072 Then we place the new variable just after the one at the top of the
2075 ###### ast functions
2077 static void scope_finalize(struct parse_context *c, struct type *ft)
2079 int size = ft->function.local_size;
2080 struct variable *next = ft->function.scope;
2081 struct variable *done = NULL;
2084 struct variable *v = next;
2085 struct type *t = v->type;
2092 if (v->frame_pos >= 0)
2094 while (done && done->scope_end < v->scope_start)
2095 done = done->in_scope;
2097 pos = done->frame_pos + done->type->size;
2099 pos = ft->function.local_size;
2100 if (pos & (t->align - 1))
2101 pos = (pos + t->align) & ~(t->align-1);
2103 if (size < pos + v->type->size)
2104 size = pos + v->type->size;
2108 c->out_scope = NULL;
2109 ft->function.local_size = size;
2112 ###### free context storage
2113 free(context.global);
2115 #### Variables as executables
2117 Just as we used a `val` to wrap a value into an `exec`, we similarly
2118 need a `var` to wrap a `variable` into an exec. While each `val`
2119 contained a copy of the value, each `var` holds a link to the variable
2120 because it really is the same variable no matter where it appears.
2121 When a variable is used, we need to remember to follow the `->merged`
2122 link to find the primary instance.
2124 When a variable is declared, it may or may not be given an explicit
2125 type. We need to record which so that we can report the parsed code
2134 struct variable *var;
2137 ###### variable fields
2145 VariableDecl -> IDENTIFIER : ${ {
2146 struct variable *v = var_decl(c, $1.txt);
2147 $0 = new_pos(var, $1);
2152 v = var_ref(c, $1.txt);
2154 type_err(c, "error: variable '%v' redeclared",
2156 type_err(c, "info: this is where '%v' was first declared",
2157 v->where_decl, NULL, 0, NULL);
2160 | IDENTIFIER :: ${ {
2161 struct variable *v = var_decl(c, $1.txt);
2162 $0 = new_pos(var, $1);
2168 v = var_ref(c, $1.txt);
2170 type_err(c, "error: variable '%v' redeclared",
2172 type_err(c, "info: this is where '%v' was first declared",
2173 v->where_decl, NULL, 0, NULL);
2176 | IDENTIFIER : Type ${ {
2177 struct variable *v = var_decl(c, $1.txt);
2178 $0 = new_pos(var, $1);
2184 v->explicit_type = 1;
2186 v = var_ref(c, $1.txt);
2188 type_err(c, "error: variable '%v' redeclared",
2190 type_err(c, "info: this is where '%v' was first declared",
2191 v->where_decl, NULL, 0, NULL);
2194 | IDENTIFIER :: Type ${ {
2195 struct variable *v = var_decl(c, $1.txt);
2196 $0 = new_pos(var, $1);
2203 v->explicit_type = 1;
2205 v = var_ref(c, $1.txt);
2207 type_err(c, "error: variable '%v' redeclared",
2209 type_err(c, "info: this is where '%v' was first declared",
2210 v->where_decl, NULL, 0, NULL);
2215 Variable -> IDENTIFIER ${ {
2216 struct variable *v = var_ref(c, $1.txt);
2217 $0 = new_pos(var, $1);
2219 /* This might be a global const or a label
2220 * Allocate a var with impossible type Tnone,
2221 * which will be adjusted when we find out what it is,
2222 * or will trigger an error.
2224 v = var_decl(c, $1.txt);
2231 cast(var, $0)->var = v;
2234 ###### print exec cases
2237 struct var *v = cast(var, e);
2239 struct binding *b = v->var->name;
2240 printf("%.*s", b->name.len, b->name.txt);
2247 if (loc && loc->type == Xvar) {
2248 struct var *v = cast(var, loc);
2250 struct binding *b = v->var->name;
2251 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2253 fputs("???", stderr); // NOTEST
2255 fputs("NOTVAR", stderr); // NOTEST
2258 ###### propagate exec cases
2262 struct var *var = cast(var, prog);
2263 struct variable *v = var->var;
2265 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2266 return Tnone; // NOTEST
2269 if (v->constant && (rules & Rnoconstant)) {
2270 type_err(c, "error: Cannot assign to a constant: %v",
2271 prog, NULL, 0, NULL);
2272 type_err(c, "info: name was defined as a constant here",
2273 v->where_decl, NULL, 0, NULL);
2276 if (v->type == Tnone && v->where_decl == prog)
2277 type_err(c, "error: variable used but not declared: %v",
2278 prog, NULL, 0, NULL);
2279 if (v->type == NULL) {
2280 if (type && !(*perr & Efail)) {
2282 v->where_set = prog;
2285 } else if (!type_compat(type, v->type, rules)) {
2286 type_err(c, "error: expected %1 but variable '%v' is %2", prog,
2287 type, rules, v->type);
2288 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2289 v->type, rules, NULL);
2291 if (!v->global || v->frame_pos < 0)
2298 ###### interp exec cases
2301 struct var *var = cast(var, e);
2302 struct variable *v = var->var;
2305 lrv = var_value(c, v);
2310 ###### ast functions
2312 static void free_var(struct var *v)
2317 ###### free exec cases
2318 case Xvar: free_var(cast(var, e)); break;
2323 Now that we have the shape of the interpreter in place we can add some
2324 complex types and connected them in to the data structures and the
2325 different phases of parse, analyse, print, interpret.
2327 Being "complex" the language will naturally have syntax to access
2328 specifics of objects of these types. These will fit into the grammar as
2329 "Terms" which are the things that are combined with various operators to
2330 form "Expression". Where a Term is formed by some operation on another
2331 Term, the subordinate Term will always come first, so for example a
2332 member of an array will be expressed as the Term for the array followed
2333 by an index in square brackets. The strict rule of using postfix
2334 operations makes precedence irrelevant within terms. To provide a place
2335 to put the grammar for each terms of each type, we will start out by
2336 introducing the "Term" grammar production, with contains at least a
2337 simple "Value" (to be explained later).
2341 Term -> Value ${ $0 = $<1; }$
2342 | Variable ${ $0 = $<1; }$
2345 Thus far the complex types we have are arrays and structs.
2349 Arrays can be declared by giving a size and a type, as `[size]type' so
2350 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2351 size can be either a literal number, or a named constant. Some day an
2352 arbitrary expression will be supported.
2354 As a formal parameter to a function, the array can be declared with a
2355 new variable as the size: `name:[size::number]string`. The `size`
2356 variable is set to the size of the array and must be a constant. As
2357 `number` is the only supported type, it can be left out:
2358 `name:[size::]string`.
2360 Arrays cannot be assigned. When pointers are introduced we will also
2361 introduce array slices which can refer to part or all of an array -
2362 the assignment syntax will create a slice. For now, an array can only
2363 ever be referenced by the name it is declared with. It is likely that
2364 a "`copy`" primitive will eventually be define which can be used to
2365 make a copy of an array with controllable recursive depth.
2367 For now we have two sorts of array, those with fixed size either because
2368 it is given as a literal number or because it is a struct member (which
2369 cannot have a runtime-changing size), and those with a size that is
2370 determined at runtime - local variables with a const size. The former
2371 have their size calculated at parse time, the latter at run time.
2373 For the latter type, the `size` field of the type is the size of a
2374 pointer, and the array is reallocated every time it comes into scope.
2376 We differentiate struct fields with a const size from local variables
2377 with a const size by whether they are prepared at parse time or not.
2379 ###### type union fields
2382 int unspec; // size is unspecified - vsize must be set.
2385 struct variable *vsize;
2386 struct type *member;
2389 ###### value union fields
2390 void *array; // used if not static_size
2392 ###### value functions
2394 static int array_prepare_type(struct parse_context *c, struct type *type,
2397 struct value *vsize;
2399 if (type->array.static_size)
2400 return 1; // UNTESTED
2401 if (type->array.unspec && parse_time)
2402 return 1; // UNTESTED
2403 if (parse_time && type->array.vsize && !type->array.vsize->global)
2404 return 1; // UNTESTED
2406 if (type->array.vsize) {
2407 vsize = var_value(c, type->array.vsize);
2409 return 1; // UNTESTED
2411 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2412 type->array.size = mpz_get_si(q);
2417 if (type->array.member->size <= 0)
2418 return 0; // UNTESTED
2420 type->array.static_size = 1;
2421 type->size = type->array.size * type->array.member->size;
2422 type->align = type->array.member->align;
2427 static void array_init(struct type *type, struct value *val)
2430 void *ptr = val->ptr;
2434 if (!type->array.static_size) {
2435 val->array = calloc(type->array.size,
2436 type->array.member->size);
2439 for (i = 0; i < type->array.size; i++) {
2441 v = (void*)ptr + i * type->array.member->size;
2442 val_init(type->array.member, v);
2446 static void array_free(struct type *type, struct value *val)
2449 void *ptr = val->ptr;
2451 if (!type->array.static_size)
2453 for (i = 0; i < type->array.size; i++) {
2455 v = (void*)ptr + i * type->array.member->size;
2456 free_value(type->array.member, v);
2458 if (!type->array.static_size)
2462 static int array_compat(struct type *require, struct type *have)
2464 if (have->compat != require->compat)
2466 /* Both are arrays, so we can look at details */
2467 if (!type_compat(require->array.member, have->array.member, 0))
2469 if (have->array.unspec && require->array.unspec) {
2470 if (have->array.vsize && require->array.vsize &&
2471 have->array.vsize != require->array.vsize) // UNTESTED
2472 /* sizes might not be the same */
2473 return 0; // UNTESTED
2476 if (have->array.unspec || require->array.unspec)
2477 return 1; // UNTESTED
2478 if (require->array.vsize == NULL && have->array.vsize == NULL)
2479 return require->array.size == have->array.size;
2481 return require->array.vsize == have->array.vsize; // UNTESTED
2484 static void array_print_type(struct type *type, FILE *f)
2487 if (type->array.vsize) {
2488 struct binding *b = type->array.vsize->name;
2489 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2490 type->array.unspec ? "::" : "");
2491 } else if (type->array.size)
2492 fprintf(f, "%d]", type->array.size);
2495 type_print(type->array.member, f);
2498 static struct type array_prototype = {
2500 .prepare_type = array_prepare_type,
2501 .print_type = array_print_type,
2502 .compat = array_compat,
2504 .size = sizeof(void*),
2505 .align = sizeof(void*),
2508 ###### declare terminals
2513 | [ NUMBER ] Type ${ {
2519 if (number_parse(num, tail, $2.txt) == 0)
2520 tok_err(c, "error: unrecognised number", &$2);
2522 tok_err(c, "error: unsupported number suffix", &$2);
2525 elements = mpz_get_ui(mpq_numref(num));
2526 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2527 tok_err(c, "error: array size must be an integer",
2529 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2530 tok_err(c, "error: array size is too large",
2535 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2536 t->array.size = elements;
2537 t->array.member = $<4;
2538 t->array.vsize = NULL;
2541 | [ IDENTIFIER ] Type ${ {
2542 struct variable *v = var_ref(c, $2.txt);
2545 tok_err(c, "error: name undeclared", &$2);
2546 else if (!v->constant)
2547 tok_err(c, "error: array size must be a constant", &$2);
2549 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2550 $0->array.member = $<4;
2552 $0->array.vsize = v;
2557 OptType -> Type ${ $0 = $<1; }$
2560 ###### formal type grammar
2562 | [ IDENTIFIER :: OptType ] Type ${ {
2563 struct variable *v = var_decl(c, $ID.txt);
2569 $0 = add_anon_type(c, &array_prototype, "array[var]");
2570 $0->array.member = $<6;
2572 $0->array.unspec = 1;
2573 $0->array.vsize = v;
2581 | Term [ Expression ] ${ {
2582 struct binode *b = new(binode);
2589 ###### print binode cases
2591 print_exec(b->left, -1, bracket);
2593 print_exec(b->right, -1, bracket);
2597 ###### propagate binode cases
2599 /* left must be an array, right must be a number,
2600 * result is the member type of the array
2602 propagate_types(b->right, c, perr, Tnum, 0);
2603 t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
2604 if (!t || t->compat != array_compat) {
2605 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2608 if (!type_compat(type, t->array.member, rules)) {
2609 type_err(c, "error: have %1 but need %2", prog,
2610 t->array.member, rules, type);
2612 return t->array.member;
2616 ###### interp binode cases
2622 lleft = linterp_exec(c, b->left, <ype);
2623 right = interp_exec(c, b->right, &rtype);
2625 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2629 if (ltype->array.static_size)
2632 ptr = *(void**)lleft;
2633 rvtype = ltype->array.member;
2634 if (i >= 0 && i < ltype->array.size)
2635 lrv = ptr + i * rvtype->size;
2637 val_init(ltype->array.member, &rv); // UNSAFE
2644 A `struct` is a data-type that contains one or more other data-types.
2645 It differs from an array in that each member can be of a different
2646 type, and they are accessed by name rather than by number. Thus you
2647 cannot choose an element by calculation, you need to know what you
2650 The language makes no promises about how a given structure will be
2651 stored in memory - it is free to rearrange fields to suit whatever
2652 criteria seems important.
2654 Structs are declared separately from program code - they cannot be
2655 declared in-line in a variable declaration like arrays can. A struct
2656 is given a name and this name is used to identify the type - the name
2657 is not prefixed by the word `struct` as it would be in C.
2659 Structs are only treated as the same if they have the same name.
2660 Simply having the same fields in the same order is not enough. This
2661 might change once we can create structure initializers from a list of
2664 Each component datum is identified much like a variable is declared,
2665 with a name, one or two colons, and a type. The type cannot be omitted
2666 as there is no opportunity to deduce the type from usage. An initial
2667 value can be given following an equals sign, so
2669 ##### Example: a struct type
2675 would declare a type called "complex" which has two number fields,
2676 each initialised to zero.
2678 Struct will need to be declared separately from the code that uses
2679 them, so we will need to be able to print out the declaration of a
2680 struct when reprinting the whole program. So a `print_type_decl` type
2681 function will be needed.
2683 ###### type union fields
2692 } *fields; // This is created when field_list is analysed.
2694 struct fieldlist *prev;
2697 } *field_list; // This is created during parsing
2700 ###### type functions
2701 void (*print_type_decl)(struct type *type, FILE *f);
2702 struct type *(*fieldref)(struct type *t, struct parse_context *c,
2703 struct fieldref *f, struct value **vp);
2705 ###### value functions
2707 static void structure_init(struct type *type, struct value *val)
2711 for (i = 0; i < type->structure.nfields; i++) {
2713 v = (void*) val->ptr + type->structure.fields[i].offset;
2714 if (type->structure.fields[i].init)
2715 dup_value(type->structure.fields[i].type,
2716 type->structure.fields[i].init,
2719 val_init(type->structure.fields[i].type, v);
2723 static void structure_free(struct type *type, struct value *val)
2727 for (i = 0; i < type->structure.nfields; i++) {
2729 v = (void*)val->ptr + type->structure.fields[i].offset;
2730 free_value(type->structure.fields[i].type, v);
2734 static void free_fieldlist(struct fieldlist *f)
2738 free_fieldlist(f->prev);
2743 static void structure_free_type(struct type *t)
2746 for (i = 0; i < t->structure.nfields; i++)
2747 if (t->structure.fields[i].init) {
2748 free_value(t->structure.fields[i].type,
2749 t->structure.fields[i].init);
2751 free(t->structure.fields);
2752 free_fieldlist(t->structure.field_list);
2755 static int structure_prepare_type(struct parse_context *c,
2756 struct type *t, int parse_time)
2759 struct fieldlist *f;
2761 if (!parse_time || t->structure.fields)
2764 for (f = t->structure.field_list; f; f=f->prev) {
2768 if (f->f.type->size <= 0)
2770 if (f->f.type->prepare_type)
2771 f->f.type->prepare_type(c, f->f.type, parse_time);
2773 if (f->init == NULL)
2777 propagate_types(f->init, c, &perr, f->f.type, 0);
2778 } while (perr & Eretry);
2780 c->parse_error += 1; // NOTEST
2783 t->structure.nfields = cnt;
2784 t->structure.fields = calloc(cnt, sizeof(struct field));
2785 f = t->structure.field_list;
2787 int a = f->f.type->align;
2789 t->structure.fields[cnt] = f->f;
2790 if (t->size & (a-1))
2791 t->size = (t->size | (a-1)) + 1;
2792 t->structure.fields[cnt].offset = t->size;
2793 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2797 if (f->init && !c->parse_error) {
2798 struct value vl = interp_exec(c, f->init, NULL);
2799 t->structure.fields[cnt].init =
2800 global_alloc(c, f->f.type, NULL, &vl);
2808 static int find_struct_index(struct type *type, struct text field)
2811 for (i = 0; i < type->structure.nfields; i++)
2812 if (text_cmp(type->structure.fields[i].name, field) == 0)
2814 return IndexInvalid;
2817 static struct type *structure_fieldref(struct type *t, struct parse_context *c,
2818 struct fieldref *f, struct value **vp)
2820 if (f->index == IndexUnknown) {
2821 f->index = find_struct_index(t, f->name);
2823 type_err(c, "error: cannot find requested field in %1",
2824 f->left, t, 0, NULL);
2829 struct value *v = *vp;
2830 v = (void*)v->ptr + t->structure.fields[f->index].offset;
2833 return t->structure.fields[f->index].type;
2836 static struct type structure_prototype = {
2837 .init = structure_init,
2838 .free = structure_free,
2839 .free_type = structure_free_type,
2840 .print_type_decl = structure_print_type,
2841 .prepare_type = structure_prepare_type,
2842 .fieldref = structure_fieldref,
2855 enum { IndexUnknown = -1, IndexInvalid = -2 };
2857 ###### free exec cases
2859 free_exec(cast(fieldref, e)->left);
2863 ###### declare terminals
2868 | Term . IDENTIFIER ${ {
2869 struct fieldref *fr = new_pos(fieldref, $2);
2872 fr->index = IndexUnknown;
2876 ###### print exec cases
2880 struct fieldref *f = cast(fieldref, e);
2881 print_exec(f->left, -1, bracket);
2882 printf(".%.*s", f->name.len, f->name.txt);
2886 ###### propagate exec cases
2890 struct fieldref *f = cast(fieldref, prog);
2891 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2893 if (!st || !st->fieldref)
2894 type_err(c, "error: field reference on %1 is not supported",
2895 f->left, st, 0, NULL);
2897 t = st->fieldref(st, c, f, NULL);
2898 if (t && !type_compat(type, t, rules))
2899 type_err(c, "error: have %1 but need %2", prog,
2906 ###### interp exec cases
2909 struct fieldref *f = cast(fieldref, e);
2911 struct value *lleft = linterp_exec(c, f->left, <ype);
2913 rvtype = ltype->fieldref(ltype, c, f, &lrv);
2917 ###### top level grammar
2918 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2920 t = find_type(c, $ID.txt);
2922 t = add_type(c, $ID.txt, &structure_prototype);
2923 else if (t->size >= 0) {
2924 tok_err(c, "error: type already declared", &$ID);
2925 tok_err(c, "info: this is location of declartion", &t->first_use);
2926 /* Create a new one - duplicate */
2927 t = add_type(c, $ID.txt, &structure_prototype);
2929 struct type tmp = *t;
2930 *t = structure_prototype;
2934 t->structure.field_list = $<FB;
2939 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2940 | { SimpleFieldList } ${ $0 = $<SFL; }$
2941 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2942 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2944 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2945 | FieldLines SimpleFieldList Newlines ${
2950 SimpleFieldList -> Field ${ $0 = $<F; }$
2951 | SimpleFieldList ; Field ${
2955 | SimpleFieldList ; ${
2958 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2960 Field -> IDENTIFIER : Type = Expression ${ {
2961 $0 = calloc(1, sizeof(struct fieldlist));
2962 $0->f.name = $ID.txt;
2963 $0->f.type = $<Type;
2967 | IDENTIFIER : Type ${
2968 $0 = calloc(1, sizeof(struct fieldlist));
2969 $0->f.name = $ID.txt;
2970 $0->f.type = $<Type;
2973 ###### forward decls
2974 static void structure_print_type(struct type *t, FILE *f);
2976 ###### value functions
2977 static void structure_print_type(struct type *t, FILE *f)
2981 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2983 for (i = 0; i < t->structure.nfields; i++) {
2984 struct field *fl = t->structure.fields + i;
2985 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2986 type_print(fl->type, f);
2987 if (fl->type->print && fl->init) {
2989 if (fl->type == Tstr)
2990 fprintf(f, "\""); // UNTESTED
2991 print_value(fl->type, fl->init, f);
2992 if (fl->type == Tstr)
2993 fprintf(f, "\""); // UNTESTED
2999 ###### print type decls
3004 while (target != 0) {
3006 for (t = context.typelist; t ; t=t->next)
3007 if (!t->anon && t->print_type_decl &&
3017 t->print_type_decl(t, stdout);
3025 References, or pointers, are values that refer to another value. They
3026 can only refer to a `struct`, though as a struct can embed anything they
3027 can effectively refer to anything.
3029 References are potentially dangerous as they might refer to some
3030 variable which no longer exists - either because a stack frame
3031 containing it has been discarded or because the value was allocated on
3032 the heap and has now been free. Ocean does not yet provide any
3033 protection against these problems. It will in due course.
3035 With references comes the opportunity and the need to explicitly
3036 allocate values on the "heap" and to free them. We currently provide
3037 fairly basic support for this.
3039 Reference make use of the `@` symbol in various ways. A type that starts
3040 with `@` is a reference to whatever follows. A reference value
3041 followed by an `@` acts as the referred value, though the `@` is often
3042 not needed. Finally, an expression that starts with `@` is a special
3043 reference related expression. Some examples might help.
3045 ##### Example: Reference examples
3052 bar.number = 23; bar.string = "hello"
3063 Obviously this is very contrived. `ref` is a reference to a `foo` which
3064 is initially set to refer to the value stored in `bar` - no extra syntax
3065 is needed to "Take the address of" `bar` - the fact that `ref` is a
3066 reference means that only the address make sense.
3068 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
3069 The same syntax is used for accessing fields both in structs and in
3070 references to structs. It would be correct to use `ref@.a`, but not
3073 `@new()` creates an object of whatever type is needed for the program
3074 to by type-correct. In future iterations of Ocean, arguments a
3075 constructor will access arguments, so the the syntax now looks like a
3076 function call. `@free` can be assigned any reference that was returned
3077 by `@new()`, and it will be freed. `@nil` is a value of whatever
3078 reference type is appropriate, and is stable and never the address of
3079 anything in the heap or on the stack. A reference can be assigned
3080 `@nil` or compared against that value.
3082 ###### declare terminals
3085 ###### type union fields
3088 struct type *referent;
3091 ###### value union fields
3094 ###### value functions
3096 static void reference_print_type(struct type *t, FILE *f)
3099 type_print(t->reference.referent, f);
3102 static int reference_cmp(struct type *tl, struct type *tr,
3103 struct value *left, struct value *right)
3105 return left->ref == right->ref ? 0 : 1;
3108 static void reference_dup(struct type *t,
3109 struct value *vold, struct value *vnew)
3111 vnew->ref = vold->ref;
3114 static void reference_free(struct type *t, struct value *v)
3116 /* Nothing to do here */
3119 static int reference_compat(struct type *require, struct type *have)
3121 if (have->compat != require->compat)
3123 if (have->reference.referent != require->reference.referent)
3128 static int reference_test(struct type *type, struct value *val)
3130 return val->ref != NULL;
3133 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3134 struct fieldref *f, struct value **vp)
3136 struct type *rt = t->reference.referent;
3141 return rt->fieldref(rt, c, f, vp);
3143 type_err(c, "error: field reference on %1 is not supported",
3144 f->left, rt, 0, NULL);
3149 static struct type reference_prototype = {
3150 .print_type = reference_print_type,
3151 .cmp_eq = reference_cmp,
3152 .dup = reference_dup,
3153 .test = reference_test,
3154 .free = reference_free,
3155 .compat = reference_compat,
3156 .fieldref = reference_fieldref,
3157 .size = sizeof(void*),
3158 .align = sizeof(void*),
3164 struct type *t = find_type(c, $ID.txt);
3166 t = add_type(c, $ID.txt, NULL);
3169 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3170 $ID.txt.len, $ID.txt.txt);
3171 $0->reference.referent = t;
3174 ###### core functions
3175 static int text_is(struct text t, char *s)
3177 return (strlen(s) == t.len &&
3178 strncmp(s, t.txt, t.len) == 0);
3187 enum ref_func { RefNew, RefFree, RefNil } action;
3188 struct type *reftype;
3192 ###### SimpleStatement Grammar
3194 | @ IDENTIFIER = Expression ${ {
3195 struct ref *r = new_pos(ref, $ID);
3197 if (!text_is($ID.txt, "free"))
3198 tok_err(c, "error: only \"@free\" makes sense here",
3202 r->action = RefFree;
3206 ###### expression grammar
3207 | @ IDENTIFIER ( ) ${
3208 // Only 'new' valid here
3209 if (!text_is($ID.txt, "new")) {
3210 tok_err(c, "error: Only reference function is \"@new()\"",
3213 struct ref *r = new_pos(ref,$ID);
3219 // Only 'nil' valid here
3220 if (!text_is($ID.txt, "nil")) {
3221 tok_err(c, "error: Only reference value is \"@nil\"",
3224 struct ref *r = new_pos(ref,$ID);
3230 ###### print exec cases
3232 struct ref *r = cast(ref, e);
3233 switch (r->action) {
3235 printf("@new()"); break;
3237 printf("@nil"); break;
3239 do_indent(indent, "@free = ");
3240 print_exec(r->right, indent, bracket);
3246 ###### propagate exec cases
3248 struct ref *r = cast(ref, prog);
3249 switch (r->action) {
3251 if (type && type->free != reference_free) {
3252 type_err(c, "error: @new() can only be used with references, not %1",
3253 prog, type, 0, NULL);
3256 if (type && !r->reftype) {
3262 if (type && type->free != reference_free)
3263 type_err(c, "error: @nil can only be used with reference, not %1",
3264 prog, type, 0, NULL);
3265 if (type && !r->reftype) {
3271 t = propagate_types(r->right, c, perr, NULL, 0);
3272 if (t && t->free != reference_free)
3273 type_err(c, "error: @free can only be assigned a reference, not %1",
3282 ###### interp exec cases
3284 struct ref *r = cast(ref, e);
3285 switch (r->action) {
3288 rv.ref = calloc(1, r->reftype->reference.referent->size);
3289 rvtype = r->reftype;
3293 rvtype = r->reftype;
3296 rv = interp_exec(c, r->right, &rvtype);
3297 free_value(rvtype->reference.referent, rv.ref);
3305 ###### free exec cases
3307 struct ref *r = cast(ref, e);
3308 free_exec(r->right);
3313 ###### Expressions: dereference
3321 struct binode *b = new(binode);
3327 ###### print binode cases
3329 print_exec(b->left, -1, bracket);
3333 ###### propagate binode cases
3335 /* left must be a reference, and we return what it refers to */
3336 /* FIXME how can I pass the expected type down? */
3337 t = propagate_types(b->left, c, perr, NULL, 0);
3338 if (!t || t->free != reference_free)
3339 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3341 return t->reference.referent;
3344 ###### interp binode cases
3346 left = interp_exec(c, b->left, <ype);
3348 rvtype = ltype->reference.referent;
3355 A function is a chunk of code which can be passed parameters and can
3356 return results. Each function has a type which includes the set of
3357 parameters and the return value. As yet these types cannot be declared
3358 separately from the function itself.
3360 The parameters can be specified either in parentheses as a ';' separated
3363 ##### Example: function 1
3365 func main(av:[ac::number]string; env:[envc::number]string)
3368 or as an indented list of one parameter per line (though each line can
3369 be a ';' separated list)
3371 ##### Example: function 2
3374 argv:[argc::number]string
3375 env:[envc::number]string
3379 In the first case a return type can follow the parentheses after a colon,
3380 in the second it is given on a line starting with the word `return`.
3382 ##### Example: functions that return
3384 func add(a:number; b:number): number
3394 Rather than returning a type, the function can specify a set of local
3395 variables to return as a struct. The values of these variables when the
3396 function exits will be provided to the caller. For this the return type
3397 is replaced with a block of result declarations, either in parentheses
3398 or bracketed by `return` and `do`.
3400 ##### Example: functions returning multiple variables
3402 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3415 For constructing the lists we use a `List` binode, which will be
3416 further detailed when Expression Lists are introduced.
3418 ###### type union fields
3421 struct binode *params;
3422 struct type *return_type;
3423 struct variable *scope;
3424 int inline_result; // return value is at start of 'local'
3428 ###### value union fields
3429 struct exec *function;
3431 ###### type functions
3432 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3433 struct type *require, struct exec *args);
3435 ###### value functions
3437 static void function_free(struct type *type, struct value *val)
3439 free_exec(val->function);
3440 val->function = NULL;
3443 static int function_compat(struct type *require, struct type *have)
3445 // FIXME can I do anything here yet?
3449 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3450 struct type *require, struct exec *args)
3452 /* This should be 'compat', but we don't have a 'tuple' type to
3453 * hold the type of 'args'
3455 struct binode *arg = cast(binode, args);
3456 struct binode *param = require->function.params;
3459 struct var *pv = cast(var, param->left);
3461 type_err(c, "error: insufficient arguments to function.",
3462 args, NULL, 0, NULL);
3466 propagate_types(arg->left, c, perr, pv->var->type, 0);
3467 param = cast(binode, param->right);
3468 arg = cast(binode, arg->right);
3471 type_err(c, "error: too many arguments to function.",
3472 args, NULL, 0, NULL);
3475 static void function_print(struct type *type, struct value *val, FILE *f)
3477 print_exec(val->function, 1, 0);
3480 static void function_print_type_decl(struct type *type, FILE *f)
3484 for (b = type->function.params; b; b = cast(binode, b->right)) {
3485 struct variable *v = cast(var, b->left)->var;
3486 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3487 v->constant ? "::" : ":");
3488 type_print(v->type, f);
3493 if (type->function.return_type != Tnone) {
3495 if (type->function.inline_result) {
3497 struct type *t = type->function.return_type;
3499 for (i = 0; i < t->structure.nfields; i++) {
3500 struct field *fl = t->structure.fields + i;
3503 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3504 type_print(fl->type, f);
3508 type_print(type->function.return_type, f);
3513 static void function_free_type(struct type *t)
3515 free_exec(t->function.params);
3518 static struct type function_prototype = {
3519 .size = sizeof(void*),
3520 .align = sizeof(void*),
3521 .free = function_free,
3522 .compat = function_compat,
3523 .check_args = function_check_args,
3524 .print = function_print,
3525 .print_type_decl = function_print_type_decl,
3526 .free_type = function_free_type,
3529 ###### declare terminals
3539 FuncName -> IDENTIFIER ${ {
3540 struct variable *v = var_decl(c, $1.txt);
3541 struct var *e = new_pos(var, $1);
3548 v = var_ref(c, $1.txt);
3550 type_err(c, "error: function '%v' redeclared",
3552 type_err(c, "info: this is where '%v' was first declared",
3553 v->where_decl, NULL, 0, NULL);
3559 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3560 | Args ArgsLine NEWLINE ${ {
3561 struct binode *b = $<AL;
3562 struct binode **bp = &b;
3564 bp = (struct binode **)&(*bp)->left;
3569 ArgsLine -> ${ $0 = NULL; }$
3570 | Varlist ${ $0 = $<1; }$
3571 | Varlist ; ${ $0 = $<1; }$
3573 Varlist -> Varlist ; ArgDecl ${
3574 $0 = new_pos(binode, $2);
3587 ArgDecl -> IDENTIFIER : FormalType ${ {
3588 struct variable *v = var_decl(c, $ID.txt);
3589 $0 = new_pos(var, $ID);
3596 ##### Function calls
3598 A function call can appear either as an expression or as a statement.
3599 We use a new 'Funcall' binode type to link the function with a list of
3600 arguments, form with the 'List' nodes.
3602 We have already seen the "Term" which is how a function call can appear
3603 in an expression. To parse a function call into a statement we include
3604 it in the "SimpleStatement Grammar" which will be described later.
3610 | Term ( ExpressionList ) ${ {
3611 struct binode *b = new(binode);
3614 b->right = reorder_bilist($<EL);
3618 struct binode *b = new(binode);
3625 ###### SimpleStatement Grammar
3627 | Term ( ExpressionList ) ${ {
3628 struct binode *b = new(binode);
3631 b->right = reorder_bilist($<EL);
3635 ###### print binode cases
3638 do_indent(indent, "");
3639 print_exec(b->left, -1, bracket);
3641 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3644 print_exec(b->left, -1, bracket);
3654 ###### propagate binode cases
3657 /* Every arg must match formal parameter, and result
3658 * is return type of function
3660 struct binode *args = cast(binode, b->right);
3661 struct var *v = cast(var, b->left);
3663 if (!v->var->type || v->var->type->check_args == NULL) {
3664 type_err(c, "error: attempt to call a non-function.",
3665 prog, NULL, 0, NULL);
3669 v->var->type->check_args(c, perr, v->var->type, args);
3670 if (v->var->type->function.inline_result)
3672 return v->var->type->function.return_type;
3675 ###### interp binode cases
3678 struct var *v = cast(var, b->left);
3679 struct type *t = v->var->type;
3680 void *oldlocal = c->local;
3681 int old_size = c->local_size;
3682 void *local = calloc(1, t->function.local_size);
3683 struct value *fbody = var_value(c, v->var);
3684 struct binode *arg = cast(binode, b->right);
3685 struct binode *param = t->function.params;
3688 struct var *pv = cast(var, param->left);
3689 struct type *vtype = NULL;
3690 struct value val = interp_exec(c, arg->left, &vtype);
3692 c->local = local; c->local_size = t->function.local_size;
3693 lval = var_value(c, pv->var);
3694 c->local = oldlocal; c->local_size = old_size;
3695 memcpy(lval, &val, vtype->size);
3696 param = cast(binode, param->right);
3697 arg = cast(binode, arg->right);
3699 c->local = local; c->local_size = t->function.local_size;
3700 if (t->function.inline_result && dtype) {
3701 _interp_exec(c, fbody->function, NULL, NULL);
3702 memcpy(dest, local, dtype->size);
3703 rvtype = ret.type = NULL;
3705 rv = interp_exec(c, fbody->function, &rvtype);
3706 c->local = oldlocal; c->local_size = old_size;
3711 ## Complex executables: statements and expressions
3713 Now that we have types and values and variables and most of the basic
3714 Terms which provide access to these, we can explore the more complex
3715 code that combine all of these to get useful work done. Specifically
3716 statements and expressions.
3718 Expressions are various combinations of Terms. We will use operator
3719 precedence to ensure correct parsing. The simplest Expression is just a
3720 Term - others will follow.
3725 Expression -> Term ${ $0 = $<Term; }$
3726 ## expression grammar
3728 ### Expressions: Conditional
3730 Our first user of the `binode` will be conditional expressions, which
3731 is a bit odd as they actually have three components. That will be
3732 handled by having 2 binodes for each expression. The conditional
3733 expression is the lowest precedence operator which is why we define it
3734 first - to start the precedence list.
3736 Conditional expressions are of the form "value `if` condition `else`
3737 other_value". They associate to the right, so everything to the right
3738 of `else` is part of an else value, while only a higher-precedence to
3739 the left of `if` is the if values. Between `if` and `else` there is no
3740 room for ambiguity, so a full conditional expression is allowed in
3746 ###### declare terminals
3750 ###### expression grammar
3752 | Expression if Expression else Expression $$ifelse ${ {
3753 struct binode *b1 = new(binode);
3754 struct binode *b2 = new(binode);
3764 ###### print binode cases
3767 b2 = cast(binode, b->right);
3768 if (bracket) printf("(");
3769 print_exec(b2->left, -1, bracket);
3771 print_exec(b->left, -1, bracket);
3773 print_exec(b2->right, -1, bracket);
3774 if (bracket) printf(")");
3777 ###### propagate binode cases
3780 /* cond must be Tbool, others must match */
3781 struct binode *b2 = cast(binode, b->right);
3784 propagate_types(b->left, c, perr, Tbool, 0);
3785 t = propagate_types(b2->left, c, perr, type, 0);
3786 t2 = propagate_types(b2->right, c, perr, type ?: t, 0);
3790 ###### interp binode cases
3793 struct binode *b2 = cast(binode, b->right);
3794 left = interp_exec(c, b->left, <ype);
3796 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3798 rv = interp_exec(c, b2->right, &rvtype);
3804 We take a brief detour, now that we have expressions, to describe lists
3805 of expressions. These will be needed for function parameters and
3806 possibly other situations. They seem generic enough to introduce here
3807 to be used elsewhere.
3809 And ExpressionList will use the `List` type of `binode`, building up at
3810 the end. And place where they are used will probably call
3811 `reorder_bilist()` to get a more normal first/next arrangement.
3813 ###### declare terminals
3816 `List` execs have no implicit semantics, so they are never propagated or
3817 interpreted. The can be printed as a comma separate list, which is how
3818 they are parsed. Note they are also used for function formal parameter
3819 lists. In that case a separate function is used to print them.
3821 ###### print binode cases
3825 print_exec(b->left, -1, bracket);
3828 b = cast(binode, b->right);
3832 ###### propagate binode cases
3833 case List: abort(); // NOTEST
3834 ###### interp binode cases
3835 case List: abort(); // NOTEST
3840 ExpressionList -> ExpressionList , Expression ${
3853 ### Expressions: Boolean
3855 The next class of expressions to use the `binode` will be Boolean
3856 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3857 have same corresponding precendence. The difference is that they don't
3858 evaluate the second expression if not necessary.
3867 ###### declare terminals
3872 ###### expression grammar
3873 | Expression or Expression ${ {
3874 struct binode *b = new(binode);
3880 | Expression or else Expression ${ {
3881 struct binode *b = new(binode);
3888 | Expression and Expression ${ {
3889 struct binode *b = new(binode);
3895 | Expression and then Expression ${ {
3896 struct binode *b = new(binode);
3903 | not Expression ${ {
3904 struct binode *b = new(binode);
3910 ###### print binode cases
3912 if (bracket) printf("(");
3913 print_exec(b->left, -1, bracket);
3915 print_exec(b->right, -1, bracket);
3916 if (bracket) printf(")");
3919 if (bracket) printf("(");
3920 print_exec(b->left, -1, bracket);
3921 printf(" and then ");
3922 print_exec(b->right, -1, bracket);
3923 if (bracket) printf(")");
3926 if (bracket) printf("(");
3927 print_exec(b->left, -1, bracket);
3929 print_exec(b->right, -1, bracket);
3930 if (bracket) printf(")");
3933 if (bracket) printf("(");
3934 print_exec(b->left, -1, bracket);
3935 printf(" or else ");
3936 print_exec(b->right, -1, bracket);
3937 if (bracket) printf(")");
3940 if (bracket) printf("(");
3942 print_exec(b->right, -1, bracket);
3943 if (bracket) printf(")");
3946 ###### propagate binode cases
3952 /* both must be Tbool, result is Tbool */
3953 propagate_types(b->left, c, perr, Tbool, 0);
3954 propagate_types(b->right, c, perr, Tbool, 0);
3955 if (type && type != Tbool)
3956 type_err(c, "error: %1 operation found where %2 expected", prog,
3960 ###### interp binode cases
3962 rv = interp_exec(c, b->left, &rvtype);
3963 right = interp_exec(c, b->right, &rtype);
3964 rv.bool = rv.bool && right.bool;
3967 rv = interp_exec(c, b->left, &rvtype);
3969 rv = interp_exec(c, b->right, NULL);
3972 rv = interp_exec(c, b->left, &rvtype);
3973 right = interp_exec(c, b->right, &rtype);
3974 rv.bool = rv.bool || right.bool;
3977 rv = interp_exec(c, b->left, &rvtype);
3979 rv = interp_exec(c, b->right, NULL);
3982 rv = interp_exec(c, b->right, &rvtype);
3986 ### Expressions: Comparison
3988 Of slightly higher precedence that Boolean expressions are Comparisons.
3989 A comparison takes arguments of any comparable type, but the two types
3992 To simplify the parsing we introduce an `eop` which can record an
3993 expression operator, and the `CMPop` non-terminal will match one of them.
4000 ###### ast functions
4001 static void free_eop(struct eop *e)
4015 ###### declare terminals
4016 $LEFT < > <= >= == != CMPop
4018 ###### expression grammar
4019 | Expression CMPop Expression ${ {
4020 struct binode *b = new(binode);
4030 CMPop -> < ${ $0.op = Less; }$
4031 | > ${ $0.op = Gtr; }$
4032 | <= ${ $0.op = LessEq; }$
4033 | >= ${ $0.op = GtrEq; }$
4034 | == ${ $0.op = Eql; }$
4035 | != ${ $0.op = NEql; }$
4037 ###### print binode cases
4045 if (bracket) printf("(");
4046 print_exec(b->left, -1, bracket);
4048 case Less: printf(" < "); break;
4049 case LessEq: printf(" <= "); break;
4050 case Gtr: printf(" > "); break;
4051 case GtrEq: printf(" >= "); break;
4052 case Eql: printf(" == "); break;
4053 case NEql: printf(" != "); break;
4054 default: abort(); // NOTEST
4056 print_exec(b->right, -1, bracket);
4057 if (bracket) printf(")");
4060 ###### propagate binode cases
4067 /* Both must match but not be labels, result is Tbool */
4068 t = propagate_types(b->left, c, perr, NULL, 0);
4070 propagate_types(b->right, c, perr, t, 0);
4072 t = propagate_types(b->right, c, perr, NULL, 0); // UNTESTED
4074 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
4076 if (!type_compat(type, Tbool, 0))
4077 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4078 Tbool, rules, type);
4081 ###### interp binode cases
4090 left = interp_exec(c, b->left, <ype);
4091 right = interp_exec(c, b->right, &rtype);
4092 cmp = value_cmp(ltype, rtype, &left, &right);
4095 case Less: rv.bool = cmp < 0; break;
4096 case LessEq: rv.bool = cmp <= 0; break;
4097 case Gtr: rv.bool = cmp > 0; break;
4098 case GtrEq: rv.bool = cmp >= 0; break;
4099 case Eql: rv.bool = cmp == 0; break;
4100 case NEql: rv.bool = cmp != 0; break;
4101 default: rv.bool = 0; break; // NOTEST
4106 ### Expressions: Arithmetic etc.
4108 The remaining expressions with the highest precedence are arithmetic,
4109 string concatenation, string conversion, and testing. String concatenation
4110 (`++`) has the same precedence as multiplication and division, but lower
4113 Testing comes in two forms. A single question mark (`?`) is a uniary
4114 operator which converts come types into Boolean. The general meaning is
4115 "is this a value value" and there will be more uses as the language
4116 develops. A double questionmark (`??`) is a binary operator (Choose),
4117 with same precedence as multiplication, which returns the LHS if it
4118 tests successfully, else returns the RHS.
4120 String conversion is a temporary feature until I get a better type
4121 system. `$` is a prefix operator which expects a string and returns
4124 `+` and `-` are both infix and prefix operations (where they are
4125 absolute value and negation). These have different operator names.
4127 We also have a 'Bracket' operator which records where parentheses were
4128 found. This makes it easy to reproduce these when printing. Possibly I
4129 should only insert brackets were needed for precedence. Putting
4130 parentheses around an expression converts it into a Term,
4136 Absolute, Negate, Test,
4140 ###### declare terminals
4142 $LEFT * / % ++ ?? Top
4146 ###### expression grammar
4147 | Expression Eop Expression ${ {
4148 struct binode *b = new(binode);
4155 | Expression Top Expression ${ {
4156 struct binode *b = new(binode);
4163 | Uop Expression ${ {
4164 struct binode *b = new(binode);
4172 | ( Expression ) ${ {
4173 struct binode *b = new_pos(binode, $1);
4182 Eop -> + ${ $0.op = Plus; }$
4183 | - ${ $0.op = Minus; }$
4185 Uop -> + ${ $0.op = Absolute; }$
4186 | - ${ $0.op = Negate; }$
4187 | $ ${ $0.op = StringConv; }$
4188 | ? ${ $0.op = Test; }$
4190 Top -> * ${ $0.op = Times; }$
4191 | / ${ $0.op = Divide; }$
4192 | % ${ $0.op = Rem; }$
4193 | ++ ${ $0.op = Concat; }$
4194 | ?? ${ $0.op = Choose; }$
4196 ###### print binode cases
4204 if (bracket) printf("(");
4205 print_exec(b->left, indent, bracket);
4207 case Plus: fputs(" + ", stdout); break;
4208 case Minus: fputs(" - ", stdout); break;
4209 case Times: fputs(" * ", stdout); break;
4210 case Divide: fputs(" / ", stdout); break;
4211 case Rem: fputs(" % ", stdout); break;
4212 case Concat: fputs(" ++ ", stdout); break;
4213 case Choose: fputs(" ?? ", stdout); break;
4214 default: abort(); // NOTEST
4216 print_exec(b->right, indent, bracket);
4217 if (bracket) printf(")");
4223 if (bracket) printf("(");
4225 case Absolute: fputs("+", stdout); break;
4226 case Negate: fputs("-", stdout); break;
4227 case StringConv: fputs("$", stdout); break;
4228 case Test: fputs("?", stdout); break;
4229 default: abort(); // NOTEST
4231 print_exec(b->right, indent, bracket);
4232 if (bracket) printf(")");
4236 print_exec(b->right, indent, bracket);
4240 ###### propagate binode cases
4246 /* both must be numbers, result is Tnum */
4249 /* as propagate_types ignores a NULL,
4250 * unary ops fit here too */
4251 propagate_types(b->left, c, perr, Tnum, 0);
4252 propagate_types(b->right, c, perr, Tnum, 0);
4253 if (!type_compat(type, Tnum, 0))
4254 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4259 /* both must be Tstr, result is Tstr */
4260 propagate_types(b->left, c, perr, Tstr, 0);
4261 propagate_types(b->right, c, perr, Tstr, 0);
4262 if (!type_compat(type, Tstr, 0))
4263 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4268 /* op must be string, result is number */
4269 propagate_types(b->left, c, perr, Tstr, 0);
4270 if (!type_compat(type, Tnum, 0))
4271 type_err(c, // UNTESTED
4272 "error: Can only convert string to number, not %1",
4273 prog, type, 0, NULL);
4277 /* LHS must support ->test, result is Tbool */
4278 t = propagate_types(b->right, c, perr, NULL, 0);
4280 type_err(c, "error: '?' requires a testable value, not %1",
4285 /* LHS and RHS must match and are returned. Must support
4288 t = propagate_types(b->left, c, perr, type, rules);
4289 t = propagate_types(b->right, c, perr, t, rules);
4290 if (t && t->test == NULL)
4291 type_err(c, "error: \"??\" requires a testable value, not %1",
4296 return propagate_types(b->right, c, perr, type, 0);
4298 ###### interp binode cases
4301 rv = interp_exec(c, b->left, &rvtype);
4302 right = interp_exec(c, b->right, &rtype);
4303 mpq_add(rv.num, rv.num, right.num);
4306 rv = interp_exec(c, b->left, &rvtype);
4307 right = interp_exec(c, b->right, &rtype);
4308 mpq_sub(rv.num, rv.num, right.num);
4311 rv = interp_exec(c, b->left, &rvtype);
4312 right = interp_exec(c, b->right, &rtype);
4313 mpq_mul(rv.num, rv.num, right.num);
4316 rv = interp_exec(c, b->left, &rvtype);
4317 right = interp_exec(c, b->right, &rtype);
4318 mpq_div(rv.num, rv.num, right.num);
4323 left = interp_exec(c, b->left, <ype);
4324 right = interp_exec(c, b->right, &rtype);
4325 mpz_init(l); mpz_init(r); mpz_init(rem);
4326 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4327 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4328 mpz_tdiv_r(rem, l, r);
4329 val_init(Tnum, &rv);
4330 mpq_set_z(rv.num, rem);
4331 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4336 rv = interp_exec(c, b->right, &rvtype);
4337 mpq_neg(rv.num, rv.num);
4340 rv = interp_exec(c, b->right, &rvtype);
4341 mpq_abs(rv.num, rv.num);
4344 rv = interp_exec(c, b->right, &rvtype);
4347 left = interp_exec(c, b->left, <ype);
4348 right = interp_exec(c, b->right, &rtype);
4350 rv.str = text_join(left.str, right.str);
4353 right = interp_exec(c, b->right, &rvtype);
4357 struct text tx = right.str;
4360 if (tx.txt[0] == '-') {
4361 neg = 1; // UNTESTED
4362 tx.txt++; // UNTESTED
4363 tx.len--; // UNTESTED
4365 if (number_parse(rv.num, tail, tx) == 0)
4366 mpq_init(rv.num); // UNTESTED
4368 mpq_neg(rv.num, rv.num); // UNTESTED
4370 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4374 right = interp_exec(c, b->right, &rtype);
4376 rv.bool = !!rtype->test(rtype, &right);
4379 left = interp_exec(c, b->left, <ype);
4380 if (ltype->test(ltype, &left)) {
4385 rv = interp_exec(c, b->right, &rvtype);
4388 ###### value functions
4390 static struct text text_join(struct text a, struct text b)
4393 rv.len = a.len + b.len;
4394 rv.txt = malloc(rv.len);
4395 memcpy(rv.txt, a.txt, a.len);
4396 memcpy(rv.txt+a.len, b.txt, b.len);
4400 ### Blocks, Statements, and Statement lists.
4402 Now that we have expressions out of the way we need to turn to
4403 statements. There are simple statements and more complex statements.
4404 Simple statements do not contain (syntactic) newlines, complex statements do.
4406 Statements often come in sequences and we have corresponding simple
4407 statement lists and complex statement lists.
4408 The former comprise only simple statements separated by semicolons.
4409 The later comprise complex statements and simple statement lists. They are
4410 separated by newlines. Thus the semicolon is only used to separate
4411 simple statements on the one line. This may be overly restrictive,
4412 but I'm not sure I ever want a complex statement to share a line with
4415 Note that a simple statement list can still use multiple lines if
4416 subsequent lines are indented, so
4418 ###### Example: wrapped simple statement list
4423 is a single simple statement list. This might allow room for
4424 confusion, so I'm not set on it yet.
4426 A simple statement list needs no extra syntax. A complex statement
4427 list has two syntactic forms. It can be enclosed in braces (much like
4428 C blocks), or it can be introduced by an indent and continue until an
4429 unindented newline (much like Python blocks). With this extra syntax
4430 it is referred to as a block.
4432 Note that a block does not have to include any newlines if it only
4433 contains simple statements. So both of:
4435 if condition: a=b; d=f
4437 if condition { a=b; print f }
4441 In either case the list is constructed from a `binode` list with
4442 `Block` as the operator. When parsing the list it is most convenient
4443 to append to the end, so a list is a list and a statement. When using
4444 the list it is more convenient to consider a list to be a statement
4445 and a list. So we need a function to re-order a list.
4446 `reorder_bilist` serves this purpose.
4448 The only stand-alone statement we introduce at this stage is `pass`
4449 which does nothing and is represented as a `NULL` pointer in a `Block`
4450 list. Other stand-alone statements will follow once the infrastructure
4453 As many statements will use binodes, we declare a binode pointer 'b' in
4454 the common header for all reductions to use.
4456 ###### Parser: reduce
4467 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4468 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4469 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4470 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4471 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4473 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4474 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4475 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4476 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4477 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4479 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4480 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4481 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4483 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4484 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4485 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4486 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4487 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4489 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4491 ComplexStatements -> ComplexStatements ComplexStatement ${
4501 | ComplexStatement ${
4513 ComplexStatement -> SimpleStatements Newlines ${
4514 $0 = reorder_bilist($<SS);
4516 | SimpleStatements ; Newlines ${
4517 $0 = reorder_bilist($<SS);
4519 ## ComplexStatement Grammar
4522 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4528 | SimpleStatement ${
4537 SimpleStatement -> pass ${ $0 = NULL; }$
4538 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4539 ## SimpleStatement Grammar
4541 ###### print binode cases
4545 if (b->left == NULL) // UNTESTED
4546 printf("pass"); // UNTESTED
4548 print_exec(b->left, indent, bracket); // UNTESTED
4549 if (b->right) { // UNTESTED
4550 printf("; "); // UNTESTED
4551 print_exec(b->right, indent, bracket); // UNTESTED
4554 // block, one per line
4555 if (b->left == NULL)
4556 do_indent(indent, "pass\n");
4558 print_exec(b->left, indent, bracket);
4560 print_exec(b->right, indent, bracket);
4564 ###### propagate binode cases
4567 /* If any statement returns something other than Tnone
4568 * or Tbool then all such must return same type.
4569 * As each statement may be Tnone or something else,
4570 * we must always pass NULL (unknown) down, otherwise an incorrect
4571 * error might occur. We never return Tnone unless it is
4576 for (e = b; e; e = cast(binode, e->right)) {
4577 t = propagate_types(e->left, c, perr, NULL, rules);
4578 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4580 if (t == Tnone && e->right)
4581 /* Only the final statement *must* return a value
4589 type_err(c, "error: expected %1, found %2",
4590 e->left, type, rules, t);
4596 ###### interp binode cases
4598 while (rvtype == Tnone &&
4601 rv = interp_exec(c, b->left, &rvtype);
4602 b = cast(binode, b->right);
4606 ### The Print statement
4608 `print` is a simple statement that takes a comma-separated list of
4609 expressions and prints the values separated by spaces and terminated
4610 by a newline. No control of formatting is possible.
4612 `print` uses `ExpressionList` to collect the expressions and stores them
4613 on the left side of a `Print` binode unlessthere is a trailing comma
4614 when the list is stored on the `right` side and no trailing newline is
4620 ##### declare terminals
4623 ###### SimpleStatement Grammar
4625 | print ExpressionList ${
4626 $0 = b = new_pos(binode, $1);
4629 b->left = reorder_bilist($<EL);
4631 | print ExpressionList , ${ {
4632 $0 = b = new_pos(binode, $1);
4634 b->right = reorder_bilist($<EL);
4638 $0 = b = new_pos(binode, $1);
4644 ###### print binode cases
4647 do_indent(indent, "print");
4649 print_exec(b->right, -1, bracket);
4652 print_exec(b->left, -1, bracket);
4657 ###### propagate binode cases
4660 /* don't care but all must be consistent */
4662 b = cast(binode, b->left);
4664 b = cast(binode, b->right);
4666 propagate_types(b->left, c, perr, NULL, 0);
4667 b = cast(binode, b->right);
4671 ###### interp binode cases
4675 struct binode *b2 = cast(binode, b->left);
4677 b2 = cast(binode, b->right);
4678 for (; b2; b2 = cast(binode, b2->right)) {
4679 left = interp_exec(c, b2->left, <ype);
4680 print_value(ltype, &left, stdout);
4681 free_value(ltype, &left);
4685 if (b->right == NULL)
4691 ###### Assignment statement
4693 An assignment will assign a value to a variable, providing it hasn't
4694 been declared as a constant. The analysis phase ensures that the type
4695 will be correct so the interpreter just needs to perform the
4696 calculation. There is a form of assignment which declares a new
4697 variable as well as assigning a value. If a name is used before
4698 it is declared, it is assumed to be a global constant which are allowed to
4699 be declared at any time.
4705 ###### declare terminals
4708 ###### SimpleStatement Grammar
4709 | Term = Expression ${
4710 $0 = b= new(binode);
4715 | VariableDecl = Expression ${
4716 $0 = b= new(binode);
4723 if ($1->var->where_set == NULL) {
4725 "Variable declared with no type or value: %v",
4729 $0 = b = new(binode);
4736 ###### print binode cases
4739 do_indent(indent, "");
4740 print_exec(b->left, -1, bracket);
4742 print_exec(b->right, -1, bracket);
4749 struct variable *v = cast(var, b->left)->var;
4750 do_indent(indent, "");
4751 print_exec(b->left, -1, bracket);
4752 if (cast(var, b->left)->var->constant) {
4754 if (v->explicit_type) {
4755 type_print(v->type, stdout);
4760 if (v->explicit_type) {
4761 type_print(v->type, stdout);
4767 print_exec(b->right, -1, bracket);
4774 ###### propagate binode cases
4778 /* Both must match and not be labels,
4779 * Type must support 'dup',
4780 * For Assign, left must not be constant.
4783 t = propagate_types(b->left, c, perr, NULL,
4784 (b->op == Assign ? Rnoconstant : 0));
4789 if (propagate_types(b->right, c, perr, t, 0) != t)
4790 if (b->left->type == Xvar)
4791 type_err(c, "info: variable '%v' was set as %1 here.",
4792 cast(var, b->left)->var->where_set, t, rules, NULL);
4794 t = propagate_types(b->right, c, perr, NULL, 0);
4796 propagate_types(b->left, c, perr, t,
4797 (b->op == Assign ? Rnoconstant : 0));
4799 if (t && t->dup == NULL && !(*perr & Emaycopy))
4800 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4805 ###### interp binode cases
4808 lleft = linterp_exec(c, b->left, <ype);
4810 dinterp_exec(c, b->right, lleft, ltype, 1);
4816 struct variable *v = cast(var, b->left)->var;
4819 val = var_value(c, v);
4820 if (v->type->prepare_type)
4821 v->type->prepare_type(c, v->type, 0);
4823 dinterp_exec(c, b->right, val, v->type, 0);
4825 val_init(v->type, val);
4829 ### The `use` statement
4831 The `use` statement is the last "simple" statement. It is needed when a
4832 statement block can return a value. This includes the body of a
4833 function which has a return type, and the "condition" code blocks in
4834 `if`, `while`, and `switch` statements.
4839 ###### declare terminals
4842 ###### SimpleStatement Grammar
4844 $0 = b = new_pos(binode, $1);
4849 ###### print binode cases
4852 do_indent(indent, "use ");
4853 print_exec(b->right, -1, bracket);
4858 ###### propagate binode cases
4861 /* result matches value */
4862 return propagate_types(b->right, c, perr, type, 0);
4864 ###### interp binode cases
4867 rv = interp_exec(c, b->right, &rvtype);
4870 ### The Conditional Statement
4872 This is the biggy and currently the only complex statement. This
4873 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4874 It is comprised of a number of parts, all of which are optional though
4875 set combinations apply. Each part is (usually) a key word (`then` is
4876 sometimes optional) followed by either an expression or a code block,
4877 except the `casepart` which is a "key word and an expression" followed
4878 by a code block. The code-block option is valid for all parts and,
4879 where an expression is also allowed, the code block can use the `use`
4880 statement to report a value. If the code block does not report a value
4881 the effect is similar to reporting `True`.
4883 The `else` and `case` parts, as well as `then` when combined with
4884 `if`, can contain a `use` statement which will apply to some
4885 containing conditional statement. `for` parts, `do` parts and `then`
4886 parts used with `for` can never contain a `use`, except in some
4887 subordinate conditional statement.
4889 If there is a `forpart`, it is executed first, only once.
4890 If there is a `dopart`, then it is executed repeatedly providing
4891 always that the `condpart` or `cond`, if present, does not return a non-True
4892 value. `condpart` can fail to return any value if it simply executes
4893 to completion. This is treated the same as returning `True`.
4895 If there is a `thenpart` it will be executed whenever the `condpart`
4896 or `cond` returns True (or does not return any value), but this will happen
4897 *after* `dopart` (when present).
4899 If `elsepart` is present it will be executed at most once when the
4900 condition returns `False` or some value that isn't `True` and isn't
4901 matched by any `casepart`. If there are any `casepart`s, they will be
4902 executed when the condition returns a matching value.
4904 The particular sorts of values allowed in case parts has not yet been
4905 determined in the language design, so nothing is prohibited.
4907 The various blocks in this complex statement potentially provide scope
4908 for variables as described earlier. Each such block must include the
4909 "OpenScope" nonterminal before parsing the block, and must call
4910 `var_block_close()` when closing the block.
4912 The code following "`if`", "`switch`" and "`for`" does not get its own
4913 scope, but is in a scope covering the whole statement, so names
4914 declared there cannot be redeclared elsewhere. Similarly the
4915 condition following "`while`" is in a scope the covers the body
4916 ("`do`" part) of the loop, and which does not allow conditional scope
4917 extension. Code following "`then`" (both looping and non-looping),
4918 "`else`" and "`case`" each get their own local scope.
4920 The type requirements on the code block in a `whilepart` are quite
4921 unusal. It is allowed to return a value of some identifiable type, in
4922 which case the loop aborts and an appropriate `casepart` is run, or it
4923 can return a Boolean, in which case the loop either continues to the
4924 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4925 This is different both from the `ifpart` code block which is expected to
4926 return a Boolean, or the `switchpart` code block which is expected to
4927 return the same type as the casepart values. The correct analysis of
4928 the type of the `whilepart` code block is the reason for the
4929 `Rboolok` flag which is passed to `propagate_types()`.
4931 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4932 defined. As there are two scopes which cover multiple parts - one for
4933 the whole statement and one for "while" and "do" - and as we will use
4934 the 'struct exec' to track scopes, we actually need two new types of
4935 exec. One is a `binode` for the looping part, the rest is the
4936 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4937 casepart` to track a list of case parts.
4948 struct exec *action;
4949 struct casepart *next;
4951 struct cond_statement {
4953 struct exec *forpart, *condpart, *thenpart, *elsepart;
4954 struct binode *looppart;
4955 struct casepart *casepart;
4958 ###### ast functions
4960 static void free_casepart(struct casepart *cp)
4964 free_exec(cp->value);
4965 free_exec(cp->action);
4972 static void free_cond_statement(struct cond_statement *s)
4976 free_exec(s->forpart);
4977 free_exec(s->condpart);
4978 free_exec(s->looppart);
4979 free_exec(s->thenpart);
4980 free_exec(s->elsepart);
4981 free_casepart(s->casepart);
4985 ###### free exec cases
4986 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4988 ###### ComplexStatement Grammar
4989 | CondStatement ${ $0 = $<1; }$
4991 ###### declare terminals
4992 $TERM for then while do
4999 // A CondStatement must end with EOL, as does CondSuffix and
5001 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
5002 // may or may not end with EOL
5003 // WhilePart and IfPart include an appropriate Suffix
5005 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
5006 // them. WhilePart opens and closes its own scope.
5007 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
5010 $0->thenpart = $<TP;
5011 $0->looppart = $<WP;
5012 var_block_close(c, CloseSequential, $0);
5014 | ForPart OptNL WhilePart CondSuffix ${
5017 $0->looppart = $<WP;
5018 var_block_close(c, CloseSequential, $0);
5020 | WhilePart CondSuffix ${
5022 $0->looppart = $<WP;
5024 | SwitchPart OptNL CasePart CondSuffix ${
5026 $0->condpart = $<SP;
5027 $CP->next = $0->casepart;
5028 $0->casepart = $<CP;
5029 var_block_close(c, CloseSequential, $0);
5031 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
5033 $0->condpart = $<SP;
5034 $CP->next = $0->casepart;
5035 $0->casepart = $<CP;
5036 var_block_close(c, CloseSequential, $0);
5038 | IfPart IfSuffix ${
5040 $0->condpart = $IP.condpart; $IP.condpart = NULL;
5041 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
5042 // This is where we close an "if" statement
5043 var_block_close(c, CloseSequential, $0);
5046 CondSuffix -> IfSuffix ${
5049 | Newlines CasePart CondSuffix ${
5051 $CP->next = $0->casepart;
5052 $0->casepart = $<CP;
5054 | CasePart CondSuffix ${
5056 $CP->next = $0->casepart;
5057 $0->casepart = $<CP;
5060 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
5061 | Newlines ElsePart ${ $0 = $<EP; }$
5062 | ElsePart ${$0 = $<EP; }$
5064 ElsePart -> else OpenBlock Newlines ${
5065 $0 = new(cond_statement);
5066 $0->elsepart = $<OB;
5067 var_block_close(c, CloseElse, $0->elsepart);
5069 | else OpenScope CondStatement ${
5070 $0 = new(cond_statement);
5071 $0->elsepart = $<CS;
5072 var_block_close(c, CloseElse, $0->elsepart);
5076 CasePart -> case Expression OpenScope ColonBlock ${
5077 $0 = calloc(1,sizeof(struct casepart));
5080 var_block_close(c, CloseParallel, $0->action);
5084 // These scopes are closed in CondStatement
5085 ForPart -> for OpenBlock ${
5089 ThenPart -> then OpenBlock ${
5091 var_block_close(c, CloseSequential, $0);
5095 // This scope is closed in CondStatement
5096 WhilePart -> while UseBlock OptNL do OpenBlock ${
5101 var_block_close(c, CloseSequential, $0->right);
5102 var_block_close(c, CloseSequential, $0);
5104 | while OpenScope Expression OpenScope ColonBlock ${
5109 var_block_close(c, CloseSequential, $0->right);
5110 var_block_close(c, CloseSequential, $0);
5114 IfPart -> if UseBlock OptNL then OpenBlock ${
5117 var_block_close(c, CloseParallel, $0.thenpart);
5119 | if OpenScope Expression OpenScope ColonBlock ${
5122 var_block_close(c, CloseParallel, $0.thenpart);
5124 | if OpenScope Expression OpenScope OptNL then Block ${
5127 var_block_close(c, CloseParallel, $0.thenpart);
5131 // This scope is closed in CondStatement
5132 SwitchPart -> switch OpenScope Expression ${
5135 | switch UseBlock ${
5139 ###### print binode cases
5141 if (b->left && b->left->type == Xbinode &&
5142 cast(binode, b->left)->op == Block) {
5144 do_indent(indent, "while {\n");
5146 do_indent(indent, "while\n");
5147 print_exec(b->left, indent+1, bracket);
5149 do_indent(indent, "} do {\n");
5151 do_indent(indent, "do\n");
5152 print_exec(b->right, indent+1, bracket);
5154 do_indent(indent, "}\n");
5156 do_indent(indent, "while ");
5157 print_exec(b->left, 0, bracket);
5162 print_exec(b->right, indent+1, bracket);
5164 do_indent(indent, "}\n");
5168 ###### print exec cases
5170 case Xcond_statement:
5172 struct cond_statement *cs = cast(cond_statement, e);
5173 struct casepart *cp;
5175 do_indent(indent, "for");
5176 if (bracket) printf(" {\n"); else printf("\n");
5177 print_exec(cs->forpart, indent+1, bracket);
5180 do_indent(indent, "} then {\n");
5182 do_indent(indent, "then\n");
5183 print_exec(cs->thenpart, indent+1, bracket);
5185 if (bracket) do_indent(indent, "}\n");
5188 print_exec(cs->looppart, indent, bracket);
5192 do_indent(indent, "switch");
5194 do_indent(indent, "if");
5195 if (cs->condpart && cs->condpart->type == Xbinode &&
5196 cast(binode, cs->condpart)->op == Block) {
5201 print_exec(cs->condpart, indent+1, bracket);
5203 do_indent(indent, "}\n");
5205 do_indent(indent, "then\n");
5206 print_exec(cs->thenpart, indent+1, bracket);
5210 print_exec(cs->condpart, 0, bracket);
5216 print_exec(cs->thenpart, indent+1, bracket);
5218 do_indent(indent, "}\n");
5223 for (cp = cs->casepart; cp; cp = cp->next) {
5224 do_indent(indent, "case ");
5225 print_exec(cp->value, -1, 0);
5230 print_exec(cp->action, indent+1, bracket);
5232 do_indent(indent, "}\n");
5235 do_indent(indent, "else");
5240 print_exec(cs->elsepart, indent+1, bracket);
5242 do_indent(indent, "}\n");
5247 ###### propagate binode cases
5249 t = propagate_types(b->right, c, perr, Tnone, 0);
5250 if (!type_compat(Tnone, t, 0))
5251 *perr |= Efail; // UNTESTED
5252 return propagate_types(b->left, c, perr, type, rules);
5254 ###### propagate exec cases
5255 case Xcond_statement:
5257 // forpart and looppart->right must return Tnone
5258 // thenpart must return Tnone if there is a loopart,
5259 // otherwise it is like elsepart.
5261 // be bool if there is no casepart
5262 // match casepart->values if there is a switchpart
5263 // either be bool or match casepart->value if there
5265 // elsepart and casepart->action must match the return type
5266 // expected of this statement.
5267 struct cond_statement *cs = cast(cond_statement, prog);
5268 struct casepart *cp;
5270 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5271 if (!type_compat(Tnone, t, 0))
5272 *perr |= Efail; // UNTESTED
5275 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5276 if (!type_compat(Tnone, t, 0))
5277 *perr |= Efail; // UNTESTED
5279 if (cs->casepart == NULL) {
5280 propagate_types(cs->condpart, c, perr, Tbool, 0);
5281 propagate_types(cs->looppart, c, perr, Tbool, 0);
5283 /* Condpart must match case values, with bool permitted */
5285 for (cp = cs->casepart;
5286 cp && !t; cp = cp->next)
5287 t = propagate_types(cp->value, c, perr, NULL, 0);
5288 if (!t && cs->condpart)
5289 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5290 if (!t && cs->looppart)
5291 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5292 // Now we have a type (I hope) push it down
5294 for (cp = cs->casepart; cp; cp = cp->next)
5295 propagate_types(cp->value, c, perr, t, 0);
5296 propagate_types(cs->condpart, c, perr, t, Rboolok);
5297 propagate_types(cs->looppart, c, perr, t, Rboolok);
5300 // (if)then, else, and case parts must return expected type.
5301 if (!cs->looppart && !type)
5302 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5304 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5305 for (cp = cs->casepart;
5307 cp = cp->next) // UNTESTED
5308 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5311 propagate_types(cs->thenpart, c, perr, type, rules);
5312 propagate_types(cs->elsepart, c, perr, type, rules);
5313 for (cp = cs->casepart; cp ; cp = cp->next)
5314 propagate_types(cp->action, c, perr, type, rules);
5320 ###### interp binode cases
5322 // This just performs one iterration of the loop
5323 rv = interp_exec(c, b->left, &rvtype);
5324 if (rvtype == Tnone ||
5325 (rvtype == Tbool && rv.bool != 0))
5326 // rvtype is Tnone or Tbool, doesn't need to be freed
5327 interp_exec(c, b->right, NULL);
5330 ###### interp exec cases
5331 case Xcond_statement:
5333 struct value v, cnd;
5334 struct type *vtype, *cndtype;
5335 struct casepart *cp;
5336 struct cond_statement *cs = cast(cond_statement, e);
5339 interp_exec(c, cs->forpart, NULL);
5341 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5342 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5343 interp_exec(c, cs->thenpart, NULL);
5345 cnd = interp_exec(c, cs->condpart, &cndtype);
5346 if ((cndtype == Tnone ||
5347 (cndtype == Tbool && cnd.bool != 0))) {
5348 // cnd is Tnone or Tbool, doesn't need to be freed
5349 rv = interp_exec(c, cs->thenpart, &rvtype);
5350 // skip else (and cases)
5354 for (cp = cs->casepart; cp; cp = cp->next) {
5355 v = interp_exec(c, cp->value, &vtype);
5356 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5357 free_value(vtype, &v);
5358 free_value(cndtype, &cnd);
5359 rv = interp_exec(c, cp->action, &rvtype);
5362 free_value(vtype, &v);
5364 free_value(cndtype, &cnd);
5366 rv = interp_exec(c, cs->elsepart, &rvtype);
5373 ### Top level structure
5375 All the language elements so far can be used in various places. Now
5376 it is time to clarify what those places are.
5378 At the top level of a file there will be a number of declarations.
5379 Many of the things that can be declared haven't been described yet,
5380 such as functions, procedures, imports, and probably more.
5381 For now there are two sorts of things that can appear at the top
5382 level. They are predefined constants, `struct` types, and the `main`
5383 function. While the syntax will allow the `main` function to appear
5384 multiple times, that will trigger an error if it is actually attempted.
5386 The various declarations do not return anything. They store the
5387 various declarations in the parse context.
5389 ###### Parser: grammar
5392 Ocean -> OptNL DeclarationList
5394 ## declare terminals
5402 DeclarationList -> Declaration
5403 | DeclarationList Declaration
5405 Declaration -> ERROR Newlines ${
5406 tok_err(c, // UNTESTED
5407 "error: unhandled parse error", &$1);
5413 ## top level grammar
5417 ### The `const` section
5419 As well as being defined in with the code that uses them, constants can
5420 be declared at the top level. These have full-file scope, so they are
5421 always `InScope`, even before(!) they have been declared. The value of
5422 a top level constant can be given as an expression, and this is
5423 evaluated after parsing and before execution.
5425 A function call can be used to evaluate a constant, but it will not have
5426 access to any program state, once such statement becomes meaningful.
5427 e.g. arguments and filesystem will not be visible.
5429 Constants are defined in a section that starts with the reserved word
5430 `const` and then has a block with a list of assignment statements.
5431 For syntactic consistency, these must use the double-colon syntax to
5432 make it clear that they are constants. Type can also be given: if
5433 not, the type will be determined during analysis, as with other
5436 ###### parse context
5437 struct binode *constlist;
5439 ###### top level grammar
5443 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5444 | const { SimpleConstList } Newlines
5445 | const IN OptNL ConstList OUT Newlines
5446 | const SimpleConstList Newlines
5448 ConstList -> ConstList SimpleConstLine
5451 SimpleConstList -> SimpleConstList ; Const
5455 SimpleConstLine -> SimpleConstList Newlines
5456 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5459 CType -> Type ${ $0 = $<1; }$
5463 Const -> IDENTIFIER :: CType = Expression ${ {
5465 struct binode *bl, *bv;
5466 struct var *var = new_pos(var, $ID);
5468 v = var_decl(c, $ID.txt);
5470 v->where_decl = var;
5476 v = var_ref(c, $1.txt);
5477 if (v->type == Tnone) {
5478 v->where_decl = var;
5484 tok_err(c, "error: name already declared", &$1);
5485 type_err(c, "info: this is where '%v' was first declared",
5486 v->where_decl, NULL, 0, NULL);
5498 bl->left = c->constlist;
5503 ###### core functions
5504 static void resolve_consts(struct parse_context *c)
5508 enum { none, some, cannot } progress = none;
5510 c->constlist = reorder_bilist(c->constlist);
5513 for (b = cast(binode, c->constlist); b;
5514 b = cast(binode, b->right)) {
5516 struct binode *vb = cast(binode, b->left);
5517 struct var *v = cast(var, vb->left);
5518 if (v->var->frame_pos >= 0)
5522 propagate_types(vb->right, c, &perr,
5524 } while (perr & Eretry);
5526 c->parse_error += 1;
5527 else if (!(perr & Enoconst)) {
5529 struct value res = interp_exec(
5530 c, vb->right, &v->var->type);
5531 global_alloc(c, v->var->type, v->var, &res);
5533 if (progress == cannot)
5534 type_err(c, "error: const %v cannot be resolved.",
5544 progress = cannot; break;
5546 progress = none; break;
5551 ###### print const decls
5556 for (b = cast(binode, context.constlist); b;
5557 b = cast(binode, b->right)) {
5558 struct binode *vb = cast(binode, b->left);
5559 struct var *vr = cast(var, vb->left);
5560 struct variable *v = vr->var;
5566 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5567 type_print(v->type, stdout);
5569 print_exec(vb->right, -1, 0);
5574 ###### free const decls
5575 free_binode(context.constlist);
5577 ### Function declarations
5579 The code in an Ocean program is all stored in function declarations.
5580 One of the functions must be named `main` and it must accept an array of
5581 strings as a parameter - the command line arguments.
5583 As this is the top level, several things are handled a bit differently.
5584 The function is not interpreted by `interp_exec` as that isn't passed
5585 the argument list which the program requires. Similarly type analysis
5586 is a bit more interesting at this level.
5588 ###### ast functions
5590 static struct type *handle_results(struct parse_context *c,
5591 struct binode *results)
5593 /* Create a 'struct' type from the results list, which
5594 * is a list for 'struct var'
5596 struct type *t = add_anon_type(c, &structure_prototype,
5601 for (b = results; b; b = cast(binode, b->right))
5603 t->structure.nfields = cnt;
5604 t->structure.fields = calloc(cnt, sizeof(struct field));
5606 for (b = results; b; b = cast(binode, b->right)) {
5607 struct var *v = cast(var, b->left);
5608 struct field *f = &t->structure.fields[cnt++];
5609 int a = v->var->type->align;
5610 f->name = v->var->name->name;
5611 f->type = v->var->type;
5613 f->offset = t->size;
5614 v->var->frame_pos = f->offset;
5615 t->size += ((f->type->size - 1) | (a-1)) + 1;
5618 variable_unlink_exec(v->var);
5620 free_binode(results);
5624 static struct variable *declare_function(struct parse_context *c,
5625 struct variable *name,
5626 struct binode *args,
5628 struct binode *results,
5632 struct value fn = {.function = code};
5634 var_block_close(c, CloseFunction, code);
5635 t = add_anon_type(c, &function_prototype,
5636 "func %.*s", name->name->name.len,
5637 name->name->name.txt);
5639 t->function.params = reorder_bilist(args);
5641 ret = handle_results(c, reorder_bilist(results));
5642 t->function.inline_result = 1;
5643 t->function.local_size = ret->size;
5645 t->function.return_type = ret;
5646 global_alloc(c, t, name, &fn);
5647 name->type->function.scope = c->out_scope;
5652 var_block_close(c, CloseFunction, NULL);
5654 c->out_scope = NULL;
5658 ###### declare terminals
5661 ###### top level grammar
5664 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5665 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5667 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5668 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5670 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5671 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5673 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5674 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5676 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5677 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5679 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5680 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5682 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5683 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5685 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5686 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5688 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5689 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5692 ###### print func decls
5697 while (target != 0) {
5699 for (v = context.in_scope; v; v=v->in_scope)
5700 if (v->depth == 0 && v->type && v->type->check_args) {
5709 struct value *val = var_value(&context, v);
5710 printf("func %.*s", v->name->name.len, v->name->name.txt);
5711 v->type->print_type_decl(v->type, stdout);
5713 print_exec(val->function, 0, brackets);
5715 print_value(v->type, val, stdout);
5716 printf("/* frame size %d */\n", v->type->function.local_size);
5722 ###### core functions
5724 static int analyse_funcs(struct parse_context *c)
5728 for (v = c->in_scope; v; v = v->in_scope) {
5732 if (v->depth != 0 || !v->type || !v->type->check_args)
5734 ret = v->type->function.inline_result ?
5735 Tnone : v->type->function.return_type;
5736 val = var_value(c, v);
5739 propagate_types(val->function, c, &perr, ret, 0);
5740 } while (!(perr & Efail) && (perr & Eretry));
5741 if (!(perr & Efail))
5742 /* Make sure everything is still consistent */
5743 propagate_types(val->function, c, &perr, ret, 0);
5746 if (!v->type->function.inline_result &&
5747 !v->type->function.return_type->dup) {
5748 type_err(c, "error: function cannot return value of type %1",
5749 v->where_decl, v->type->function.return_type, 0, NULL);
5752 scope_finalize(c, v->type);
5757 static int analyse_main(struct type *type, struct parse_context *c)
5759 struct binode *bp = type->function.params;
5763 struct type *argv_type;
5765 argv_type = add_anon_type(c, &array_prototype, "argv");
5766 argv_type->array.member = Tstr;
5767 argv_type->array.unspec = 1;
5769 for (b = bp; b; b = cast(binode, b->right)) {
5773 propagate_types(b->left, c, &perr, argv_type, 0);
5775 default: /* invalid */ // NOTEST
5776 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5779 c->parse_error += 1;
5782 return !c->parse_error;
5785 static void interp_main(struct parse_context *c, int argc, char **argv)
5787 struct value *progp = NULL;
5788 struct text main_name = { "main", 4 };
5789 struct variable *mainv;
5795 mainv = var_ref(c, main_name);
5797 progp = var_value(c, mainv);
5798 if (!progp || !progp->function) {
5799 fprintf(stderr, "oceani: no main function found.\n");
5800 c->parse_error += 1;
5803 if (!analyse_main(mainv->type, c)) {
5804 fprintf(stderr, "oceani: main has wrong type.\n");
5805 c->parse_error += 1;
5808 al = mainv->type->function.params;
5810 c->local_size = mainv->type->function.local_size;
5811 c->local = calloc(1, c->local_size);
5813 struct var *v = cast(var, al->left);
5814 struct value *vl = var_value(c, v->var);
5824 mpq_set_ui(argcq, argc, 1);
5825 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5826 t->prepare_type(c, t, 0);
5827 array_init(v->var->type, vl);
5828 for (i = 0; i < argc; i++) {
5829 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5831 arg.str.txt = argv[i];
5832 arg.str.len = strlen(argv[i]);
5833 free_value(Tstr, vl2);
5834 dup_value(Tstr, &arg, vl2);
5838 al = cast(binode, al->right);
5840 v = interp_exec(c, progp->function, &vtype);
5841 free_value(vtype, &v);
5846 ###### ast functions
5847 void free_variable(struct variable *v)
5851 ## And now to test it out.
5853 Having a language requires having a "hello world" program. I'll
5854 provide a little more than that: a program that prints "Hello world"
5855 finds the GCD of two numbers, prints the first few elements of
5856 Fibonacci, performs a binary search for a number, and a few other
5857 things which will likely grow as the languages grows.
5859 ###### File: oceani.mk
5862 @echo "===== DEMO ====="
5863 ./oceani --section "demo: hello" oceani.mdc 55 33
5869 four ::= 2 + 2 ; five ::= 10/2
5870 const pie ::= "I like Pie";
5871 cake ::= "The cake is"
5879 func main(argv:[argc::]string)
5880 print "Hello World, what lovely oceans you have!"
5881 print "Are there", five, "?"
5882 print pi, pie, "but", cake
5884 A := $argv[1]; B := $argv[2]
5886 /* When a variable is defined in both branches of an 'if',
5887 * and used afterwards, the variables are merged.
5893 print "Is", A, "bigger than", B,"? ", bigger
5894 /* If a variable is not used after the 'if', no
5895 * merge happens, so types can be different
5898 double:string = "yes"
5899 print A, "is more than twice", B, "?", double
5902 print "double", B, "is", double
5907 if a > 0 and then b > 0:
5913 print "GCD of", A, "and", B,"is", a
5915 print a, "is not positive, cannot calculate GCD"
5917 print b, "is not positive, cannot calculate GCD"
5922 print "Fibonacci:", f1,f2,
5923 then togo = togo - 1
5931 /* Binary search... */
5936 mid := (lo + hi) / 2
5949 print "Yay, I found", target
5951 print "Closest I found was", lo
5956 // "middle square" PRNG. Not particularly good, but one my
5957 // Dad taught me - the first one I ever heard of.
5958 for i:=1; then i = i + 1; while i < size:
5959 n := list[i-1] * list[i-1]
5960 list[i] = (n / 100) % 10 000
5962 print "Before sort:",
5963 for i:=0; then i = i + 1; while i < size:
5967 for i := 1; then i=i+1; while i < size:
5968 for j:=i-1; then j=j-1; while j >= 0:
5969 if list[j] > list[j+1]:
5973 print " After sort:",
5974 for i:=0; then i = i + 1; while i < size:
5978 if 1 == 2 then print "yes"; else print "no"
5982 bob.alive = (bob.name == "Hello")
5983 print "bob", "is" if bob.alive else "isn't", "alive"