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, enum val_rules 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, enum val_rules 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 a compile time, `Eruntime` is set.
593 If the expression can be copied, `Emaycopy` is set.
595 If `Erval` is set, then the value cannot be assigned to because it is
596 a temporary result. If `Erval` is clear but `Econst` is set, then
597 the value can only be assigned once, when the variable is declared.
601 enum val_rules {Rboolok = 1<<0,};
602 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Eruntime = 1<<2,
603 Emaycopy = 1<<3, Erval = 1<<4, Econst = 1<<5};
606 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
607 struct type *type, enum val_rules rules);
608 ###### core functions
610 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
611 enum prop_err *perr_local,
612 struct type *type, enum val_rules rules)
619 switch (prog->type) {
622 struct binode *b = cast(binode, prog);
624 ## propagate binode cases
628 ## propagate exec cases
633 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
634 struct type *type, enum val_rules rules)
636 int pre_err = c->parse_error;
637 enum prop_err perr_local = 0;
638 struct type *ret = __propagate_types(prog, c, perr, &perr_local, type, rules);
640 *perr |= perr_local & (Efail | Eretry);
641 if (c->parse_error > pre_err)
648 Interpreting an `exec` doesn't require anything but the `exec`. State
649 is stored in variables and each variable will be directly linked from
650 within the `exec` tree. The exception to this is the `main` function
651 which needs to look at command line arguments. This function will be
652 interpreted separately.
654 Each `exec` can return a value combined with a type in `struct lrval`.
655 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
656 the location of a value, which can be updated, in `lval`. Others will
657 set `lval` to NULL indicating that there is a value of appropriate type
661 static struct value interp_exec(struct parse_context *c, struct exec *e,
662 struct type **typeret);
663 ###### core functions
667 struct value rval, *lval;
670 /* If dest is passed, dtype must give the expected type, and
671 * result can go there, in which case type is returned as NULL.
673 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
674 struct value *dest, struct type *dtype);
676 static struct value interp_exec(struct parse_context *c, struct exec *e,
677 struct type **typeret)
679 struct lrval ret = _interp_exec(c, e, NULL, NULL);
681 if (!ret.type) abort();
685 dup_value(ret.type, ret.lval, &ret.rval);
689 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
690 struct type **typeret)
692 struct lrval ret = _interp_exec(c, e, NULL, NULL);
694 if (!ret.type) abort();
698 free_value(ret.type, &ret.rval);
702 /* dinterp_exec is used when the destination type is certain and
703 * the value has a place to go.
705 static void dinterp_exec(struct parse_context *c, struct exec *e,
706 struct value *dest, struct type *dtype,
709 struct lrval ret = _interp_exec(c, e, dest, dtype);
713 free_value(dtype, dest);
715 dup_value(dtype, ret.lval, dest);
717 memcpy(dest, &ret.rval, dtype->size);
720 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
721 struct value *dest, struct type *dtype)
723 /* If the result is copied to dest, ret.type is set to NULL */
725 struct value rv = {}, *lrv = NULL;
728 rvtype = ret.type = Tnone;
738 struct binode *b = cast(binode, e);
739 struct value left, right, *lleft;
740 struct type *ltype, *rtype;
741 ltype = rtype = Tnone;
743 ## interp binode cases
745 free_value(ltype, &left);
746 free_value(rtype, &right);
756 ## interp exec cleanup
762 Values come in a wide range of types, with more likely to be added.
763 Each type needs to be able to print its own values (for convenience at
764 least) as well as to compare two values, at least for equality and
765 possibly for order. For now, values might need to be duplicated and
766 freed, though eventually such manipulations will be better integrated
769 Rather than requiring every numeric type to support all numeric
770 operations (add, multiply, etc), we allow types to be able to present
771 as one of a few standard types: integer, float, and fraction. The
772 existence of these conversion functions eventually enable types to
773 determine if they are compatible with other types, though such types
774 have not yet been implemented.
776 Named type are stored in a simple linked list. Objects of each type are
777 "values" which are often passed around by value.
779 There are both explicitly named types, and anonymous types. Anonymous
780 cannot be accessed by name, but are used internally and have a name
781 which might be reported in error messages.
788 ## value union fields
796 struct token first_use;
799 void (*init)(struct type *type, struct value *val);
800 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
801 void (*print)(struct type *type, struct value *val, FILE *f);
802 void (*print_type)(struct type *type, FILE *f);
803 int (*cmp_order)(struct type *t1, struct type *t2,
804 struct value *v1, struct value *v2);
805 int (*cmp_eq)(struct type *t1, struct type *t2,
806 struct value *v1, struct value *v2);
807 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
808 int (*test)(struct type *type, struct value *val);
809 void (*free)(struct type *type, struct value *val);
810 void (*free_type)(struct type *t);
811 long long (*to_int)(struct value *v);
812 double (*to_float)(struct value *v);
813 int (*to_mpq)(mpq_t *q, struct value *v);
822 struct type *typelist;
829 static struct type *find_type(struct parse_context *c, struct text s)
831 struct type *t = c->typelist;
833 while (t && (t->anon ||
834 text_cmp(t->name, s) != 0))
839 static struct type *_add_type(struct parse_context *c, struct text s,
840 struct type *proto, int anon)
844 n = calloc(1, sizeof(*n));
851 n->next = c->typelist;
856 static struct type *add_type(struct parse_context *c, struct text s,
859 return _add_type(c, s, proto, 0);
862 static struct type *add_anon_type(struct parse_context *c,
863 struct type *proto, char *name, ...)
869 vasprintf(&t.txt, name, ap);
871 t.len = strlen(t.txt);
872 return _add_type(c, t, proto, 1);
875 static struct type *find_anon_type(struct parse_context *c,
876 struct type *proto, char *name, ...)
878 struct type *t = c->typelist;
883 vasprintf(&nm.txt, name, ap);
885 nm.len = strlen(name);
887 while (t && (!t->anon ||
888 text_cmp(t->name, nm) != 0))
894 return _add_type(c, nm, proto, 1);
897 static void free_type(struct type *t)
899 /* The type is always a reference to something in the
900 * context, so we don't need to free anything.
904 static void free_value(struct type *type, struct value *v)
908 memset(v, 0x5a, type->size);
912 static void type_print(struct type *type, FILE *f)
915 fputs("*unknown*type*", f); // NOTEST
916 else if (type->name.len && !type->anon)
917 fprintf(f, "%.*s", type->name.len, type->name.txt);
918 else if (type->print_type)
919 type->print_type(type, f);
920 else if (type->name.len && type->anon)
921 fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
923 fputs("*invalid*type*", f); // NOTEST
926 static void val_init(struct type *type, struct value *val)
928 if (type && type->init)
929 type->init(type, val);
932 static void dup_value(struct type *type,
933 struct value *vold, struct value *vnew)
935 if (type && type->dup)
936 type->dup(type, vold, vnew);
939 static int value_cmp(struct type *tl, struct type *tr,
940 struct value *left, struct value *right)
942 if (tl && tl->cmp_order)
943 return tl->cmp_order(tl, tr, left, right);
944 if (tl && tl->cmp_eq)
945 return tl->cmp_eq(tl, tr, left, right);
949 static void print_value(struct type *type, struct value *v, FILE *f)
951 if (type && type->print)
952 type->print(type, v, f);
954 fprintf(f, "*Unknown*"); // NOTEST
957 static void prepare_types(struct parse_context *c)
961 enum { none, some, cannot } progress = none;
966 for (t = c->typelist; t; t = t->next) {
968 tok_err(c, "error: type used but not declared",
970 if (t->size == 0 && t->prepare_type) {
971 if (t->prepare_type(c, t, 1))
973 else if (progress == cannot)
974 tok_err(c, "error: type has recursive definition",
984 progress = cannot; break;
986 progress = none; break;
993 static void free_value(struct type *type, struct value *v);
994 static int type_compat(struct type *require, struct type *have, enum val_rules rules);
995 static void type_print(struct type *type, FILE *f);
996 static void val_init(struct type *type, struct value *v);
997 static void dup_value(struct type *type,
998 struct value *vold, struct value *vnew);
999 static int value_cmp(struct type *tl, struct type *tr,
1000 struct value *left, struct value *right);
1001 static void print_value(struct type *type, struct value *v, FILE *f);
1003 ###### free context types
1005 while (context.typelist) {
1006 struct type *t = context.typelist;
1008 context.typelist = t->next;
1016 Type can be specified for local variables, for fields in a structure,
1017 for formal parameters to functions, and possibly elsewhere. Different
1018 rules may apply in different contexts. As a minimum, a named type may
1019 always be used. Currently the type of a formal parameter can be
1020 different from types in other contexts, so we have a separate grammar
1026 Type -> IDENTIFIER ${
1027 $0 = find_type(c, $ID.txt);
1029 $0 = add_type(c, $ID.txt, NULL);
1030 $0->first_use = $ID;
1035 FormalType -> Type ${ $0 = $<1; }$
1036 ## formal type grammar
1040 Values of the base types can be numbers, which we represent as
1041 multi-precision fractions, strings, Booleans and labels. When
1042 analysing the program we also need to allow for places where no value
1043 is meaningful (type `Tnone`) and where we don't know what type to
1044 expect yet (type is `NULL`).
1046 Values are never shared, they are always copied when used, and freed
1047 when no longer needed.
1049 When propagating type information around the program, we need to
1050 determine if two types are compatible, where type `NULL` is compatible
1051 with anything. There are two special cases with type compatibility,
1052 both related to the Conditional Statement which will be described
1053 later. In some cases a Boolean can be accepted as well as some other
1054 primary type, and in others any type is acceptable except a label (`Vlabel`).
1055 A separate function encoding these cases will simplify some code later.
1057 ###### type functions
1059 int (*compat)(struct type *this, struct type *other, enum val_rules rules);
1061 ###### ast functions
1063 static int type_compat(struct type *require, struct type *have,
1064 enum val_rules rules)
1066 if ((rules & Rboolok) && have == Tbool)
1068 if (!require || !have)
1071 if (require->compat)
1072 return require->compat(require, have, rules);
1074 return require == have;
1079 #include "parse_string.h"
1080 #include "parse_number.h"
1083 myLDLIBS := libnumber.o libstring.o -lgmp
1084 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1086 ###### type union fields
1087 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1089 ###### value union fields
1095 ###### ast functions
1096 static void _free_value(struct type *type, struct value *v)
1100 switch (type->vtype) {
1102 case Vstr: free(v->str.txt); break;
1103 case Vnum: mpq_clear(v->num); break;
1109 ###### value functions
1111 static void _val_init(struct type *type, struct value *val)
1113 switch(type->vtype) {
1114 case Vnone: // NOTEST
1117 mpq_init(val->num); break;
1119 val->str.txt = malloc(1);
1126 val->label = 0; // NOTEST
1131 static void _dup_value(struct type *type,
1132 struct value *vold, struct value *vnew)
1134 switch (type->vtype) {
1135 case Vnone: // NOTEST
1138 vnew->label = vold->label; // NOTEST
1141 vnew->bool = vold->bool;
1144 mpq_init(vnew->num);
1145 mpq_set(vnew->num, vold->num);
1148 vnew->str.len = vold->str.len;
1149 vnew->str.txt = malloc(vnew->str.len);
1150 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1155 static int _value_cmp(struct type *tl, struct type *tr,
1156 struct value *left, struct value *right)
1160 return tl - tr; // NOTEST
1161 switch (tl->vtype) {
1162 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1163 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1164 case Vstr: cmp = text_cmp(left->str, right->str); break;
1165 case Vbool: cmp = left->bool - right->bool; break;
1166 case Vnone: cmp = 0; // NOTEST
1171 static void _print_value(struct type *type, struct value *v, FILE *f)
1173 switch (type->vtype) {
1174 case Vnone: // NOTEST
1175 fprintf(f, "*no-value*"); break; // NOTEST
1176 case Vlabel: // NOTEST
1177 fprintf(f, "*label-%d*", v->label); break; // NOTEST
1179 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1181 fprintf(f, "%s", v->bool ? "True":"False"); break;
1186 mpf_set_q(fl, v->num);
1187 gmp_fprintf(f, "%.10Fg", fl);
1194 static void _free_value(struct type *type, struct value *v);
1196 static int bool_test(struct type *type, struct value *v)
1201 static struct type base_prototype = {
1203 .print = _print_value,
1204 .cmp_order = _value_cmp,
1205 .cmp_eq = _value_cmp,
1207 .free = _free_value,
1210 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1212 ###### ast functions
1213 static struct type *add_base_type(struct parse_context *c, char *n,
1214 enum vtype vt, int size)
1216 struct text txt = { n, strlen(n) };
1219 t = add_type(c, txt, &base_prototype);
1222 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1223 if (t->size & (t->align - 1))
1224 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1228 ###### context initialization
1230 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1231 Tbool->test = bool_test;
1232 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1233 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1234 Tnone = add_base_type(&context, "none", Vnone, 0);
1235 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1239 We have already met values as separate objects. When manifest constants
1240 appear in the program text, that must result in an executable which has
1241 a constant value. So the `val` structure embeds a value in an
1254 ###### ast functions
1255 struct val *new_val(struct type *T, struct token tk)
1257 struct val *v = new_pos(val, tk);
1262 ###### declare terminals
1269 $0 = new_val(Tbool, $1);
1273 $0 = new_val(Tbool, $1);
1278 $0 = new_val(Tnum, $1);
1279 if (number_parse($0->val.num, tail, $1.txt) == 0)
1280 mpq_init($0->val.num); // UNTESTED
1282 tok_err(c, "error: unsupported number suffix",
1287 $0 = new_val(Tstr, $1);
1288 string_parse(&$1, '\\', &$0->val.str, tail);
1290 tok_err(c, "error: unsupported string suffix",
1295 $0 = new_val(Tstr, $1);
1296 string_parse(&$1, '\\', &$0->val.str, tail);
1298 tok_err(c, "error: unsupported string suffix",
1302 ###### print exec cases
1305 struct val *v = cast(val, e);
1306 if (v->vtype == Tstr)
1308 // FIXME how to ensure numbers have same precision.
1309 print_value(v->vtype, &v->val, stdout);
1310 if (v->vtype == Tstr)
1315 ###### propagate exec cases
1318 struct val *val = cast(val, prog);
1319 if (!type_compat(type, val->vtype, rules))
1320 type_err(c, "error: expected %1 found %2",
1321 prog, type, rules, val->vtype);
1326 ###### interp exec cases
1328 rvtype = cast(val, e)->vtype;
1329 dup_value(rvtype, &cast(val, e)->val, &rv);
1332 ###### ast functions
1333 static void free_val(struct val *v)
1336 free_value(v->vtype, &v->val);
1340 ###### free exec cases
1341 case Xval: free_val(cast(val, e)); break;
1343 ###### ast functions
1344 // Move all nodes from 'b' to 'rv', reversing their order.
1345 // In 'b' 'left' is a list, and 'right' is the last node.
1346 // In 'rv', left' is the first node and 'right' is a list.
1347 static struct binode *reorder_bilist(struct binode *b)
1349 struct binode *rv = NULL;
1352 struct exec *t = b->right;
1356 b = cast(binode, b->left);
1366 Labels are a temporary concept until I implement enums. There are an
1367 anonymous enum which is declared by usage. Thet are only allowed in
1368 `use` statements and corresponding `case` entries. They appear as a
1369 period followed by an identifier. All identifiers that are "used" must
1372 For now, we have a global list of labels, and don't check that all "use"
1384 ###### free exec cases
1388 ###### print exec cases
1390 struct label *l = cast(label, e);
1391 printf(".%.*s", l->name.len, l->name.txt);
1397 struct labels *next;
1401 ###### parse context
1402 struct labels *labels;
1404 ###### ast functions
1405 static int label_lookup(struct parse_context *c, struct text name)
1407 struct labels *l, **lp = &c->labels;
1408 while (*lp && text_cmp((*lp)->name, name) < 0)
1410 if (*lp && text_cmp((*lp)->name, name) == 0)
1411 return (*lp)->value;
1412 l = calloc(1, sizeof(*l));
1415 if (c->next_label == 0)
1417 l->value = c->next_label;
1423 ###### free context storage
1424 while (context.labels) {
1425 struct labels *l = context.labels;
1426 context.labels = l->next;
1430 ###### declare terminals
1434 struct label *l = new_pos(label, $ID);
1438 ###### propagate exec cases
1440 struct label *l = cast(label, prog);
1441 l->value = label_lookup(c, l->name);
1442 if (!type_compat(type, Tlabel, rules))
1443 type_err(c, "error: expected %1 found %2",
1444 prog, type, rules, Tlabel);
1448 ###### interp exec cases
1450 struct label *l = cast(label, e);
1451 rv.label = l->value;
1459 Variables are scoped named values. We store the names in a linked list
1460 of "bindings" sorted in lexical order, and use sequential search and
1467 struct binding *next; // in lexical order
1471 This linked list is stored in the parse context so that "reduce"
1472 functions can find or add variables, and so the analysis phase can
1473 ensure that every variable gets a type.
1475 ###### parse context
1477 struct binding *varlist; // In lexical order
1479 ###### ast functions
1481 static struct binding *find_binding(struct parse_context *c, struct text s)
1483 struct binding **l = &c->varlist;
1488 (cmp = text_cmp((*l)->name, s)) < 0)
1492 n = calloc(1, sizeof(*n));
1499 Each name can be linked to multiple variables defined in different
1500 scopes. Each scope starts where the name is declared and continues
1501 until the end of the containing code block. Scopes of a given name
1502 cannot nest, so a declaration while a name is in-scope is an error.
1504 ###### binding fields
1505 struct variable *var;
1509 struct variable *previous;
1511 struct binding *name;
1512 struct exec *where_decl;// where name was declared
1513 struct exec *where_set; // where type was set
1517 When a scope closes, the values of the variables might need to be freed.
1518 This happens in the context of some `struct exec` and each `exec` will
1519 need to know which variables need to be freed when it completes.
1522 struct variable *to_free;
1524 ####### variable fields
1525 struct exec *cleanup_exec;
1526 struct variable *next_free;
1528 ####### interp exec cleanup
1531 for (v = e->to_free; v; v = v->next_free) {
1532 struct value *val = var_value(c, v);
1533 free_value(v->type, val);
1537 ###### ast functions
1538 static void variable_unlink_exec(struct variable *v)
1540 struct variable **vp;
1541 if (!v->cleanup_exec)
1543 for (vp = &v->cleanup_exec->to_free;
1544 *vp; vp = &(*vp)->next_free) {
1548 v->cleanup_exec = NULL;
1553 While the naming seems strange, we include local constants in the
1554 definition of variables. A name declared `var := value` can
1555 subsequently be changed, but a name declared `var ::= value` cannot -
1558 ###### variable fields
1561 Scopes in parallel branches can be partially merged. More
1562 specifically, if a given name is declared in both branches of an
1563 if/else then its scope is a candidate for merging. Similarly if
1564 every branch of an exhaustive switch (e.g. has an "else" clause)
1565 declares a given name, then the scopes from the branches are
1566 candidates for merging.
1568 Note that names declared inside a loop (which is only parallel to
1569 itself) are never visible after the loop. Similarly names defined in
1570 scopes which are not parallel, such as those started by `for` and
1571 `switch`, are never visible after the scope. Only variables defined in
1572 both `then` and `else` (including the implicit then after an `if`, and
1573 excluding `then` used with `for`) and in all `case`s and `else` of a
1574 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1576 Labels, which are a bit like variables, follow different rules.
1577 Labels are not explicitly declared, but if an undeclared name appears
1578 in a context where a label is legal, that effectively declares the
1579 name as a label. The declaration remains in force (or in scope) at
1580 least to the end of the immediately containing block and conditionally
1581 in any larger containing block which does not declare the name in some
1582 other way. Importantly, the conditional scope extension happens even
1583 if the label is only used in one parallel branch of a conditional --
1584 when used in one branch it is treated as having been declared in all
1587 Merge candidates are tentatively visible beyond the end of the
1588 branching statement which creates them. If the name is used, the
1589 merge is affirmed and they become a single variable visible at the
1590 outer layer. If not - if it is redeclared first - the merge lapses.
1592 To track scopes we have an extra stack, implemented as a linked list,
1593 which roughly parallels the parse stack and which is used exclusively
1594 for scoping. When a new scope is opened, a new frame is pushed and
1595 the child-count of the parent frame is incremented. This child-count
1596 is used to distinguish between the first of a set of parallel scopes,
1597 in which declared variables must not be in scope, and subsequent
1598 branches, whether they may already be conditionally scoped.
1600 We need a total ordering of scopes so we can easily compare to variables
1601 to see if they are concurrently in scope. To achieve this we record a
1602 `scope_count` which is actually a count of both beginnings and endings
1603 of scopes. Then each variable has a record of the scope count where it
1604 enters scope, and where it leaves.
1606 To push a new frame *before* any code in the frame is parsed, we need a
1607 grammar reduction. This is most easily achieved with a grammar
1608 element which derives the empty string, and creates the new scope when
1609 it is recognised. This can be placed, for example, between a keyword
1610 like "if" and the code following it.
1614 struct scope *parent;
1618 ###### parse context
1621 struct scope *scope_stack;
1623 ###### variable fields
1624 int scope_start, scope_end;
1626 ###### ast functions
1627 static void scope_pop(struct parse_context *c)
1629 struct scope *s = c->scope_stack;
1631 c->scope_stack = s->parent;
1633 c->scope_depth -= 1;
1634 c->scope_count += 1;
1637 static void scope_push(struct parse_context *c)
1639 struct scope *s = calloc(1, sizeof(*s));
1641 c->scope_stack->child_count += 1;
1642 s->parent = c->scope_stack;
1644 c->scope_depth += 1;
1645 c->scope_count += 1;
1651 OpenScope -> ${ scope_push(c); }$
1653 Each variable records a scope depth and is in one of four states:
1655 - "in scope". This is the case between the declaration of the
1656 variable and the end of the containing block, and also between
1657 the usage with affirms a merge and the end of that block.
1659 The scope depth is not greater than the current parse context scope
1660 nest depth. When the block of that depth closes, the state will
1661 change. To achieve this, all "in scope" variables are linked
1662 together as a stack in nesting order.
1664 - "pending". The "in scope" block has closed, but other parallel
1665 scopes are still being processed. So far, every parallel block at
1666 the same level that has closed has declared the name.
1668 The scope depth is the depth of the last parallel block that
1669 enclosed the declaration, and that has closed.
1671 - "conditionally in scope". The "in scope" block and all parallel
1672 scopes have closed, and no further mention of the name has been seen.
1673 This state includes a secondary nest depth (`min_depth`) which records
1674 the outermost scope seen since the variable became conditionally in
1675 scope. If a use of the name is found, the variable becomes "in scope"
1676 and that secondary depth becomes the recorded scope depth. If the
1677 name is declared as a new variable, the old variable becomes "out of
1678 scope" and the recorded scope depth stays unchanged.
1680 - "out of scope". The variable is neither in scope nor conditionally
1681 in scope. It is permanently out of scope now and can be removed from
1682 the "in scope" stack. When a variable becomes out-of-scope it is
1683 moved to a separate list (`out_scope`) of variables which have fully
1684 known scope. This will be used at the end of each function to assign
1685 each variable a place in the stack frame.
1687 ###### variable fields
1688 int depth, min_depth;
1689 enum { OutScope, PendingScope, CondScope, InScope } scope;
1690 struct variable *in_scope;
1692 ###### parse context
1694 struct variable *in_scope;
1695 struct variable *out_scope;
1697 All variables with the same name are linked together using the
1698 'previous' link. Those variable that have been affirmatively merged all
1699 have a 'merged' pointer that points to one primary variable - the most
1700 recently declared instance. When merging variables, we need to also
1701 adjust the 'merged' pointer on any other variables that had previously
1702 been merged with the one that will no longer be primary.
1704 A variable that is no longer the most recent instance of a name may
1705 still have "pending" scope, if it might still be merged with most
1706 recent instance. These variables don't really belong in the
1707 "in_scope" list, but are not immediately removed when a new instance
1708 is found. Instead, they are detected and ignored when considering the
1709 list of in_scope names.
1711 The storage of the value of a variable will be described later. For now
1712 we just need to know that when a variable goes out of scope, it might
1713 need to be freed. For this we need to be able to find it, so assume that
1714 `var_value()` will provide that.
1716 ###### variable fields
1717 struct variable *merged;
1719 ###### ast functions
1721 static void variable_merge(struct variable *primary, struct variable *secondary)
1725 primary = primary->merged;
1727 for (v = primary->previous; v; v=v->previous)
1728 if (v == secondary || v == secondary->merged ||
1729 v->merged == secondary ||
1730 v->merged == secondary->merged) {
1731 v->scope = OutScope;
1732 v->merged = primary;
1733 if (v->scope_start < primary->scope_start)
1734 primary->scope_start = v->scope_start;
1735 if (v->scope_end > primary->scope_end)
1736 primary->scope_end = v->scope_end; // NOTEST
1737 variable_unlink_exec(v);
1741 ###### forward decls
1742 static struct value *var_value(struct parse_context *c, struct variable *v);
1744 ###### free global vars
1746 while (context.varlist) {
1747 struct binding *b = context.varlist;
1748 struct variable *v = b->var;
1749 context.varlist = b->next;
1752 struct variable *next = v->previous;
1754 if (v->global && v->frame_pos >= 0) {
1755 free_value(v->type, var_value(&context, v));
1756 if (v->depth == 0 && v->type->free == function_free)
1757 // This is a function constant
1758 free_exec(v->where_decl);
1765 #### Manipulating Bindings
1767 When a name is conditionally visible, a new declaration discards the old
1768 binding - the condition lapses. Similarly when we reach the end of a
1769 function (outermost non-global scope) any conditional scope must lapse.
1770 Conversely a usage of the name affirms the visibility and extends it to
1771 the end of the containing block - i.e. the block that contains both the
1772 original declaration and the latest usage. This is determined from
1773 `min_depth`. When a conditionally visible variable gets affirmed like
1774 this, it is also merged with other conditionally visible variables with
1777 When we parse a variable declaration we either report an error if the
1778 name is currently bound, or create a new variable at the current nest
1779 depth if the name is unbound or bound to a conditionally scoped or
1780 pending-scope variable. If the previous variable was conditionally
1781 scoped, it and its homonyms becomes out-of-scope.
1783 When we parse a variable reference (including non-declarative assignment
1784 "foo = bar") we report an error if the name is not bound or is bound to
1785 a pending-scope variable; update the scope if the name is bound to a
1786 conditionally scoped variable; or just proceed normally if the named
1787 variable is in scope.
1789 When we exit a scope, any variables bound at this level are either
1790 marked out of scope or pending-scoped, depending on whether the scope
1791 was sequential or parallel. Here a "parallel" scope means the "then"
1792 or "else" part of a conditional, or any "case" or "else" branch of a
1793 switch. Other scopes are "sequential".
1795 When exiting a parallel scope we check if there are any variables that
1796 were previously pending and are still visible. If there are, then
1797 they weren't redeclared in the most recent scope, so they cannot be
1798 merged and must become out-of-scope. If it is not the first of
1799 parallel scopes (based on `child_count`), we check that there was a
1800 previous binding that is still pending-scope. If there isn't, the new
1801 variable must now be out-of-scope.
1803 When exiting a sequential scope that immediately enclosed parallel
1804 scopes, we need to resolve any pending-scope variables. If there was
1805 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1806 we need to mark all pending-scope variable as out-of-scope. Otherwise
1807 all pending-scope variables become conditionally scoped.
1810 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1812 ###### ast functions
1814 static struct variable *var_decl(struct parse_context *c, struct text s)
1816 struct binding *b = find_binding(c, s);
1817 struct variable *v = b->var;
1819 switch (v ? v->scope : OutScope) {
1821 /* Caller will report the error */
1825 v && v->scope == CondScope;
1827 v->scope = OutScope;
1831 v = calloc(1, sizeof(*v));
1832 v->previous = b->var;
1836 v->min_depth = v->depth = c->scope_depth;
1838 v->in_scope = c->in_scope;
1839 v->scope_start = c->scope_count;
1845 static struct variable *var_ref(struct parse_context *c, struct text s)
1847 struct binding *b = find_binding(c, s);
1848 struct variable *v = b->var;
1849 struct variable *v2;
1851 switch (v ? v->scope : OutScope) {
1854 /* Caller will report the error */
1857 /* All CondScope variables of this name need to be merged
1858 * and become InScope
1860 v->depth = v->min_depth;
1862 for (v2 = v->previous;
1863 v2 && v2->scope == CondScope;
1865 variable_merge(v, v2);
1873 static int var_refile(struct parse_context *c, struct variable *v)
1875 /* Variable just went out of scope. Add it to the out_scope
1876 * list, sorted by ->scope_start
1878 struct variable **vp = &c->out_scope;
1879 while ((*vp) && (*vp)->scope_start < v->scope_start)
1880 vp = &(*vp)->in_scope;
1886 static void var_block_close(struct parse_context *c, enum closetype ct,
1889 /* Close off all variables that are in_scope.
1890 * Some variables in c->scope may already be not-in-scope,
1891 * such as when a PendingScope variable is hidden by a new
1892 * variable with the same name.
1893 * So we check for v->name->var != v and drop them.
1894 * If we choose to make a variable OutScope, we drop it
1897 struct variable *v, **vp, *v2;
1900 for (vp = &c->in_scope;
1901 (v = *vp) && v->min_depth > c->scope_depth;
1902 (v->scope == OutScope || v->name->var != v)
1903 ? (*vp = v->in_scope, var_refile(c, v))
1904 : ( vp = &v->in_scope, 0)) {
1905 v->min_depth = c->scope_depth;
1906 if (v->name->var != v)
1907 /* This is still in scope, but we haven't just
1911 v->min_depth = c->scope_depth;
1912 if (v->scope == InScope)
1913 v->scope_end = c->scope_count;
1914 if (v->scope == InScope && e && !v->global) {
1915 /* This variable gets cleaned up when 'e' finishes */
1916 variable_unlink_exec(v);
1917 v->cleanup_exec = e;
1918 v->next_free = e->to_free;
1923 case CloseParallel: /* handle PendingScope */
1927 if (c->scope_stack->child_count == 1)
1928 /* first among parallel branches */
1929 v->scope = PendingScope;
1930 else if (v->previous &&
1931 v->previous->scope == PendingScope)
1932 /* all previous branches used name */
1933 v->scope = PendingScope;
1935 v->scope = OutScope;
1936 if (ct == CloseElse) {
1937 /* All Pending variables with this name
1938 * are now Conditional */
1940 v2 && v2->scope == PendingScope;
1942 v2->scope = CondScope;
1946 /* Not possible as it would require
1947 * parallel scope to be nested immediately
1948 * in a parallel scope, and that never
1952 /* Not possible as we already tested for
1959 if (v->scope == CondScope)
1960 /* Condition cannot continue past end of function */
1963 case CloseSequential:
1966 v->scope = OutScope;
1969 /* There was no 'else', so we can only become
1970 * conditional if we know the cases were exhaustive,
1971 * and that doesn't mean anything yet.
1972 * So only labels become conditional..
1975 v2 && v2->scope == PendingScope;
1977 v2->scope = OutScope;
1980 case OutScope: break;
1989 The value of a variable is store separately from the variable, on an
1990 analogue of a stack frame. There are (currently) two frames that can be
1991 active. A global frame which currently only stores constants, and a
1992 stacked frame which stores local variables. Each variable knows if it
1993 is global or not, and what its index into the frame is.
1995 Values in the global frame are known immediately they are relevant, so
1996 the frame needs to be reallocated as it grows so it can store those
1997 values. The local frame doesn't get values until the interpreted phase
1998 is started, so there is no need to allocate until the size is known.
2000 We initialize the `frame_pos` to an impossible value, so that we can
2001 tell if it was set or not later.
2003 ###### variable fields
2007 ###### variable init
2010 ###### parse context
2012 short global_size, global_alloc;
2014 void *global, *local;
2016 ###### forward decls
2017 static struct value *global_alloc(struct parse_context *c, struct type *t,
2018 struct variable *v, struct value *init);
2020 ###### ast functions
2022 static struct value *var_value(struct parse_context *c, struct variable *v)
2025 if (!c->local || !v->type)
2026 return NULL; // UNTESTED
2027 if (v->frame_pos + v->type->size > c->local_size) {
2028 printf("INVALID frame_pos\n"); // NOTEST
2031 return c->local + v->frame_pos;
2033 if (c->global_size > c->global_alloc) {
2034 int old = c->global_alloc;
2035 c->global_alloc = (c->global_size | 1023) + 1024;
2036 c->global = realloc(c->global, c->global_alloc);
2037 memset(c->global + old, 0, c->global_alloc - old);
2039 return c->global + v->frame_pos;
2042 static struct value *global_alloc(struct parse_context *c, struct type *t,
2043 struct variable *v, struct value *init)
2046 struct variable scratch;
2048 if (t->prepare_type)
2049 t->prepare_type(c, t, 1); // NOTEST
2051 if (c->global_size & (t->align - 1))
2052 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
2057 v->frame_pos = c->global_size;
2059 c->global_size += v->type->size;
2060 ret = var_value(c, v);
2062 memcpy(ret, init, t->size);
2064 val_init(t, ret); // NOTEST
2068 As global values are found -- struct field initializers, labels etc --
2069 `global_alloc()` is called to record the value in the global frame.
2071 When the program is fully parsed, each function is analysed, we need to
2072 walk the list of variables local to that function and assign them an
2073 offset in the stack frame. For this we have `scope_finalize()`.
2075 We keep the stack from dense by re-using space for between variables
2076 that are not in scope at the same time. The `out_scope` list is sorted
2077 by `scope_start` and as we process a varible, we move it to an FIFO
2078 stack. For each variable we consider, we first discard any from the
2079 stack anything that went out of scope before the new variable came in.
2080 Then we place the new variable just after the one at the top of the
2083 ###### ast functions
2085 static void scope_finalize(struct parse_context *c, struct type *ft)
2087 int size = ft->function.local_size;
2088 struct variable *next = ft->function.scope;
2089 struct variable *done = NULL;
2092 struct variable *v = next;
2093 struct type *t = v->type;
2100 if (v->frame_pos >= 0)
2102 while (done && done->scope_end < v->scope_start)
2103 done = done->in_scope;
2105 pos = done->frame_pos + done->type->size;
2107 pos = ft->function.local_size;
2108 if (pos & (t->align - 1))
2109 pos = (pos + t->align) & ~(t->align-1);
2111 if (size < pos + v->type->size)
2112 size = pos + v->type->size;
2116 c->out_scope = NULL;
2117 ft->function.local_size = size;
2120 ###### free context storage
2121 free(context.global);
2123 #### Variables as executables
2125 Just as we used a `val` to wrap a value into an `exec`, we similarly
2126 need a `var` to wrap a `variable` into an exec. While each `val`
2127 contained a copy of the value, each `var` holds a link to the variable
2128 because it really is the same variable no matter where it appears.
2129 When a variable is used, we need to remember to follow the `->merged`
2130 link to find the primary instance.
2132 When a variable is declared, it may or may not be given an explicit
2133 type. We need to record which so that we can report the parsed code
2142 struct variable *var;
2145 ###### variable fields
2153 VariableDecl -> IDENTIFIER : ${ {
2154 struct variable *v = var_decl(c, $1.txt);
2155 $0 = new_pos(var, $1);
2160 v = var_ref(c, $1.txt);
2162 type_err(c, "error: variable '%v' redeclared",
2164 type_err(c, "info: this is where '%v' was first declared",
2165 v->where_decl, NULL, 0, NULL);
2168 | IDENTIFIER :: ${ {
2169 struct variable *v = var_decl(c, $1.txt);
2170 $0 = new_pos(var, $1);
2176 v = var_ref(c, $1.txt);
2178 type_err(c, "error: variable '%v' redeclared",
2180 type_err(c, "info: this is where '%v' was first declared",
2181 v->where_decl, NULL, 0, NULL);
2184 | IDENTIFIER : Type ${ {
2185 struct variable *v = var_decl(c, $1.txt);
2186 $0 = new_pos(var, $1);
2192 v->explicit_type = 1;
2194 v = var_ref(c, $1.txt);
2196 type_err(c, "error: variable '%v' redeclared",
2198 type_err(c, "info: this is where '%v' was first declared",
2199 v->where_decl, NULL, 0, NULL);
2202 | IDENTIFIER :: Type ${ {
2203 struct variable *v = var_decl(c, $1.txt);
2204 $0 = new_pos(var, $1);
2211 v->explicit_type = 1;
2213 v = var_ref(c, $1.txt);
2215 type_err(c, "error: variable '%v' redeclared",
2217 type_err(c, "info: this is where '%v' was first declared",
2218 v->where_decl, NULL, 0, NULL);
2223 Variable -> IDENTIFIER ${ {
2224 struct variable *v = var_ref(c, $1.txt);
2225 $0 = new_pos(var, $1);
2227 /* This might be a global const or a label
2228 * Allocate a var with impossible type Tnone,
2229 * which will be adjusted when we find out what it is,
2230 * or will trigger an error.
2232 v = var_decl(c, $1.txt);
2239 cast(var, $0)->var = v;
2242 ###### print exec cases
2245 struct var *v = cast(var, e);
2247 struct binding *b = v->var->name;
2248 printf("%.*s", b->name.len, b->name.txt);
2255 if (loc && loc->type == Xvar) {
2256 struct var *v = cast(var, loc);
2258 struct binding *b = v->var->name;
2259 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2261 fputs("???", stderr); // NOTEST
2263 fputs("NOTVAR", stderr); // NOTEST
2266 ###### propagate exec cases
2270 struct var *var = cast(var, prog);
2271 struct variable *v = var->var;
2273 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2274 return Tnone; // NOTEST
2277 if (v->type == Tnone && v->where_decl == prog)
2278 type_err(c, "error: variable used but not declared: %v",
2279 prog, NULL, 0, NULL);
2280 if (v->type == NULL) {
2281 if (type && !(*perr & Efail)) {
2283 v->where_set = prog;
2286 } else if (!type_compat(type, v->type, rules)) {
2287 type_err(c, "error: expected %1 but variable '%v' is %2", prog,
2288 type, rules, v->type);
2289 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2290 v->type, rules, NULL);
2292 if (!v->global || v->frame_pos < 0)
2299 ###### interp exec cases
2302 struct var *var = cast(var, e);
2303 struct variable *v = var->var;
2306 lrv = var_value(c, v);
2311 ###### ast functions
2313 static void free_var(struct var *v)
2318 ###### free exec cases
2319 case Xvar: free_var(cast(var, e)); break;
2324 Now that we have the shape of the interpreter in place we can add some
2325 complex types and connected them in to the data structures and the
2326 different phases of parse, analyse, print, interpret.
2328 Being "complex" the language will naturally have syntax to access
2329 specifics of objects of these types. These will fit into the grammar as
2330 "Terms" which are the things that are combined with various operators to
2331 form "Expression". Where a Term is formed by some operation on another
2332 Term, the subordinate Term will always come first, so for example a
2333 member of an array will be expressed as the Term for the array followed
2334 by an index in square brackets. The strict rule of using postfix
2335 operations makes precedence irrelevant within terms. To provide a place
2336 to put the grammar for each terms of each type, we will start out by
2337 introducing the "Term" grammar production, with contains at least a
2338 simple "Value" (to be explained later).
2342 Term -> Value ${ $0 = $<1; }$
2343 | Variable ${ $0 = $<1; }$
2346 Thus far the complex types we have are arrays and structs.
2350 Arrays can be declared by giving a size and a type, as `[size]type' so
2351 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2352 size can be either a literal number, or a named constant. Some day an
2353 arbitrary expression will be supported.
2355 As a formal parameter to a function, the array can be declared with a
2356 new variable as the size: `name:[size::number]string`. The `size`
2357 variable is set to the size of the array and must be a constant. As
2358 `number` is the only supported type, it can be left out:
2359 `name:[size::]string`.
2361 Arrays cannot be assigned. When pointers are introduced we will also
2362 introduce array slices which can refer to part or all of an array -
2363 the assignment syntax will create a slice. For now, an array can only
2364 ever be referenced by the name it is declared with. It is likely that
2365 a "`copy`" primitive will eventually be define which can be used to
2366 make a copy of an array with controllable recursive depth.
2368 For now we have two sorts of array, those with fixed size either because
2369 it is given as a literal number or because it is a struct member (which
2370 cannot have a runtime-changing size), and those with a size that is
2371 determined at runtime - local variables with a const size. The former
2372 have their size calculated at parse time, the latter at run time.
2374 For the latter type, the `size` field of the type is the size of a
2375 pointer, and the array is reallocated every time it comes into scope.
2377 We differentiate struct fields with a const size from local variables
2378 with a const size by whether they are prepared at parse time or not.
2380 ###### type union fields
2383 int unspec; // size is unspecified - vsize must be set.
2386 struct variable *vsize;
2387 struct type *member;
2390 ###### value union fields
2391 void *array; // used if not static_size
2393 ###### value functions
2395 static int array_prepare_type(struct parse_context *c, struct type *type,
2398 struct value *vsize;
2400 if (type->array.static_size)
2401 return 1; // UNTESTED
2402 if (type->array.unspec && parse_time)
2403 return 1; // UNTESTED
2404 if (parse_time && type->array.vsize && !type->array.vsize->global)
2405 return 1; // UNTESTED
2407 if (type->array.vsize) {
2408 vsize = var_value(c, type->array.vsize);
2410 return 1; // UNTESTED
2412 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2413 type->array.size = mpz_get_si(q);
2418 if (type->array.member->size <= 0)
2419 return 0; // UNTESTED
2421 type->array.static_size = 1;
2422 type->size = type->array.size * type->array.member->size;
2423 type->align = type->array.member->align;
2428 static void array_init(struct type *type, struct value *val)
2431 void *ptr = val->ptr;
2435 if (!type->array.static_size) {
2436 val->array = calloc(type->array.size,
2437 type->array.member->size);
2440 for (i = 0; i < type->array.size; i++) {
2442 v = (void*)ptr + i * type->array.member->size;
2443 val_init(type->array.member, v);
2447 static void array_free(struct type *type, struct value *val)
2450 void *ptr = val->ptr;
2452 if (!type->array.static_size)
2454 for (i = 0; i < type->array.size; i++) {
2456 v = (void*)ptr + i * type->array.member->size;
2457 free_value(type->array.member, v);
2459 if (!type->array.static_size)
2463 static int array_compat(struct type *require, struct type *have,
2464 enum val_rules rules)
2466 if (have->compat != require->compat)
2468 /* Both are arrays, so we can look at details */
2469 if (!type_compat(require->array.member, have->array.member, 0))
2471 if (have->array.unspec && require->array.unspec) {
2472 if (have->array.vsize && require->array.vsize &&
2473 have->array.vsize != require->array.vsize) // UNTESTED
2474 /* sizes might not be the same */
2475 return 0; // UNTESTED
2478 if (have->array.unspec || require->array.unspec)
2479 return 1; // UNTESTED
2480 if (require->array.vsize == NULL && have->array.vsize == NULL)
2481 return require->array.size == have->array.size;
2483 return require->array.vsize == have->array.vsize; // UNTESTED
2486 static void array_print_type(struct type *type, FILE *f)
2489 if (type->array.vsize) {
2490 struct binding *b = type->array.vsize->name;
2491 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2492 type->array.unspec ? "::" : "");
2493 } else if (type->array.size)
2494 fprintf(f, "%d]", type->array.size);
2497 type_print(type->array.member, f);
2500 static struct type array_prototype = {
2502 .prepare_type = array_prepare_type,
2503 .print_type = array_print_type,
2504 .compat = array_compat,
2506 .size = sizeof(void*),
2507 .align = sizeof(void*),
2510 ###### declare terminals
2515 | [ NUMBER ] Type ${ {
2521 if (number_parse(num, tail, $2.txt) == 0)
2522 tok_err(c, "error: unrecognised number", &$2);
2524 tok_err(c, "error: unsupported number suffix", &$2);
2527 elements = mpz_get_ui(mpq_numref(num));
2528 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2529 tok_err(c, "error: array size must be an integer",
2531 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2532 tok_err(c, "error: array size is too large",
2537 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2538 t->array.size = elements;
2539 t->array.member = $<4;
2540 t->array.vsize = NULL;
2543 | [ IDENTIFIER ] Type ${ {
2544 struct variable *v = var_ref(c, $2.txt);
2547 tok_err(c, "error: name undeclared", &$2);
2548 else if (!v->constant)
2549 tok_err(c, "error: array size must be a constant", &$2);
2551 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2552 $0->array.member = $<4;
2554 $0->array.vsize = v;
2559 OptType -> Type ${ $0 = $<1; }$
2562 ###### formal type grammar
2564 | [ IDENTIFIER :: OptType ] Type ${ {
2565 struct variable *v = var_decl(c, $ID.txt);
2571 $0 = add_anon_type(c, &array_prototype, "array[var]");
2572 $0->array.member = $<6;
2574 $0->array.unspec = 1;
2575 $0->array.vsize = v;
2583 | Term [ Expression ] ${ {
2584 struct binode *b = new(binode);
2591 ###### print binode cases
2593 print_exec(b->left, -1, bracket);
2595 print_exec(b->right, -1, bracket);
2599 ###### propagate binode cases
2601 /* left must be an array, right must be a number,
2602 * result is the member type of the array
2604 propagate_types(b->right, c, perr_local, Tnum, 0);
2605 t = propagate_types(b->left, c, perr, NULL, 0);
2606 if (!t || t->compat != array_compat) {
2607 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2610 if (!type_compat(type, t->array.member, rules)) {
2611 type_err(c, "error: have %1 but need %2", prog,
2612 t->array.member, rules, type);
2614 return t->array.member;
2618 ###### interp binode cases
2624 lleft = linterp_exec(c, b->left, <ype);
2625 right = interp_exec(c, b->right, &rtype);
2627 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2631 if (ltype->array.static_size)
2634 ptr = *(void**)lleft;
2635 rvtype = ltype->array.member;
2636 if (i >= 0 && i < ltype->array.size)
2637 lrv = ptr + i * rvtype->size;
2639 val_init(ltype->array.member, &rv); // UNSAFE
2646 A `struct` is a data-type that contains one or more other data-types.
2647 It differs from an array in that each member can be of a different
2648 type, and they are accessed by name rather than by number. Thus you
2649 cannot choose an element by calculation, you need to know what you
2652 The language makes no promises about how a given structure will be
2653 stored in memory - it is free to rearrange fields to suit whatever
2654 criteria seems important.
2656 Structs are declared separately from program code - they cannot be
2657 declared in-line in a variable declaration like arrays can. A struct
2658 is given a name and this name is used to identify the type - the name
2659 is not prefixed by the word `struct` as it would be in C.
2661 Structs are only treated as the same if they have the same name.
2662 Simply having the same fields in the same order is not enough. This
2663 might change once we can create structure initializers from a list of
2666 Each component datum is identified much like a variable is declared,
2667 with a name, one or two colons, and a type. The type cannot be omitted
2668 as there is no opportunity to deduce the type from usage. An initial
2669 value can be given following an equals sign, so
2671 ##### Example: a struct type
2677 would declare a type called "complex" which has two number fields,
2678 each initialised to zero.
2680 Struct will need to be declared separately from the code that uses
2681 them, so we will need to be able to print out the declaration of a
2682 struct when reprinting the whole program. So a `print_type_decl` type
2683 function will be needed.
2685 ###### type union fields
2694 } *fields; // This is created when field_list is analysed.
2696 struct fieldlist *prev;
2699 } *field_list; // This is created during parsing
2702 ###### type functions
2703 void (*print_type_decl)(struct type *type, FILE *f);
2704 struct type *(*fieldref)(struct type *t, struct parse_context *c,
2705 struct fieldref *f, struct value **vp);
2707 ###### value functions
2709 static void structure_init(struct type *type, struct value *val)
2713 for (i = 0; i < type->structure.nfields; i++) {
2715 v = (void*) val->ptr + type->structure.fields[i].offset;
2716 if (type->structure.fields[i].init)
2717 dup_value(type->structure.fields[i].type,
2718 type->structure.fields[i].init,
2721 val_init(type->structure.fields[i].type, v);
2725 static void structure_free(struct type *type, struct value *val)
2729 for (i = 0; i < type->structure.nfields; i++) {
2731 v = (void*)val->ptr + type->structure.fields[i].offset;
2732 free_value(type->structure.fields[i].type, v);
2736 static void free_fieldlist(struct fieldlist *f)
2740 free_fieldlist(f->prev);
2745 static void structure_free_type(struct type *t)
2748 for (i = 0; i < t->structure.nfields; i++)
2749 if (t->structure.fields[i].init) {
2750 free_value(t->structure.fields[i].type,
2751 t->structure.fields[i].init);
2753 free(t->structure.fields);
2754 free_fieldlist(t->structure.field_list);
2757 static int structure_prepare_type(struct parse_context *c,
2758 struct type *t, int parse_time)
2761 struct fieldlist *f;
2763 if (!parse_time || t->structure.fields)
2766 for (f = t->structure.field_list; f; f=f->prev) {
2770 if (f->f.type->size <= 0)
2772 if (f->f.type->prepare_type)
2773 f->f.type->prepare_type(c, f->f.type, parse_time);
2775 if (f->init == NULL)
2779 propagate_types(f->init, c, &perr, f->f.type, 0);
2780 } while (perr & Eretry);
2782 c->parse_error += 1; // NOTEST
2785 t->structure.nfields = cnt;
2786 t->structure.fields = calloc(cnt, sizeof(struct field));
2787 f = t->structure.field_list;
2789 int a = f->f.type->align;
2791 t->structure.fields[cnt] = f->f;
2792 if (t->size & (a-1))
2793 t->size = (t->size | (a-1)) + 1;
2794 t->structure.fields[cnt].offset = t->size;
2795 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2799 if (f->init && !c->parse_error) {
2800 struct value vl = interp_exec(c, f->init, NULL);
2801 t->structure.fields[cnt].init =
2802 global_alloc(c, f->f.type, NULL, &vl);
2810 static int find_struct_index(struct type *type, struct text field)
2813 for (i = 0; i < type->structure.nfields; i++)
2814 if (text_cmp(type->structure.fields[i].name, field) == 0)
2816 return IndexInvalid;
2819 static struct type *structure_fieldref(struct type *t, struct parse_context *c,
2820 struct fieldref *f, struct value **vp)
2822 if (f->index == IndexUnknown) {
2823 f->index = find_struct_index(t, f->name);
2825 type_err(c, "error: cannot find requested field in %1",
2826 f->left, t, 0, NULL);
2831 struct value *v = *vp;
2832 v = (void*)v->ptr + t->structure.fields[f->index].offset;
2835 return t->structure.fields[f->index].type;
2838 static struct type structure_prototype = {
2839 .init = structure_init,
2840 .free = structure_free,
2841 .free_type = structure_free_type,
2842 .print_type_decl = structure_print_type,
2843 .prepare_type = structure_prepare_type,
2844 .fieldref = structure_fieldref,
2857 enum { IndexUnknown = -1, IndexInvalid = -2 };
2859 ###### free exec cases
2861 free_exec(cast(fieldref, e)->left);
2865 ###### declare terminals
2870 | Term . IDENTIFIER ${ {
2871 struct fieldref *fr = new_pos(fieldref, $2);
2874 fr->index = IndexUnknown;
2878 ###### print exec cases
2882 struct fieldref *f = cast(fieldref, e);
2883 print_exec(f->left, -1, bracket);
2884 printf(".%.*s", f->name.len, f->name.txt);
2888 ###### propagate exec cases
2892 struct fieldref *f = cast(fieldref, prog);
2893 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2895 if (!st || !st->fieldref)
2896 type_err(c, "error: field reference on %1 is not supported",
2897 f->left, st, 0, NULL);
2899 t = st->fieldref(st, c, f, NULL);
2900 if (t && !type_compat(type, t, rules))
2901 type_err(c, "error: have %1 but need %2", prog,
2908 ###### interp exec cases
2911 struct fieldref *f = cast(fieldref, e);
2913 struct value *lleft = linterp_exec(c, f->left, <ype);
2915 rvtype = ltype->fieldref(ltype, c, f, &lrv);
2919 ###### top level grammar
2920 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2922 t = find_type(c, $ID.txt);
2924 t = add_type(c, $ID.txt, &structure_prototype);
2925 else if (t->size >= 0) {
2926 tok_err(c, "error: type already declared", &$ID);
2927 tok_err(c, "info: this is location of declartion", &t->first_use);
2928 /* Create a new one - duplicate */
2929 t = add_type(c, $ID.txt, &structure_prototype);
2931 struct type tmp = *t;
2932 *t = structure_prototype;
2936 t->structure.field_list = $<FB;
2941 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2942 | { SimpleFieldList } ${ $0 = $<SFL; }$
2943 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2944 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2946 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2947 | FieldLines SimpleFieldList Newlines ${
2952 SimpleFieldList -> Field ${ $0 = $<F; }$
2953 | SimpleFieldList ; Field ${
2957 | SimpleFieldList ; ${
2960 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2962 Field -> IDENTIFIER : Type = Expression ${ {
2963 $0 = calloc(1, sizeof(struct fieldlist));
2964 $0->f.name = $ID.txt;
2965 $0->f.type = $<Type;
2969 | IDENTIFIER : Type ${
2970 $0 = calloc(1, sizeof(struct fieldlist));
2971 $0->f.name = $ID.txt;
2972 $0->f.type = $<Type;
2975 ###### forward decls
2976 static void structure_print_type(struct type *t, FILE *f);
2978 ###### value functions
2979 static void structure_print_type(struct type *t, FILE *f)
2983 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2985 for (i = 0; i < t->structure.nfields; i++) {
2986 struct field *fl = t->structure.fields + i;
2987 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2988 type_print(fl->type, f);
2989 if (fl->type->print && fl->init) {
2991 if (fl->type == Tstr)
2992 fprintf(f, "\""); // UNTESTED
2993 print_value(fl->type, fl->init, f);
2994 if (fl->type == Tstr)
2995 fprintf(f, "\""); // UNTESTED
3001 ###### print type decls
3006 while (target != 0) {
3008 for (t = context.typelist; t ; t=t->next)
3009 if (!t->anon && t->print_type_decl &&
3019 t->print_type_decl(t, stdout);
3027 References, or pointers, are values that refer to another value. They
3028 can only refer to a `struct`, though as a struct can embed anything they
3029 can effectively refer to anything.
3031 References are potentially dangerous as they might refer to some
3032 variable which no longer exists - either because a stack frame
3033 containing it has been discarded or because the value was allocated on
3034 the heap and has now been free. Ocean does not yet provide any
3035 protection against these problems. It will in due course.
3037 With references comes the opportunity and the need to explicitly
3038 allocate values on the "heap" and to free them. We currently provide
3039 fairly basic support for this.
3041 Reference make use of the `@` symbol in various ways. A type that starts
3042 with `@` is a reference to whatever follows. A reference value
3043 followed by an `@` acts as the referred value, though the `@` is often
3044 not needed. Finally, an expression that starts with `@` is a special
3045 reference related expression. Some examples might help.
3047 ##### Example: Reference examples
3054 bar.number = 23; bar.string = "hello"
3065 Obviously this is very contrived. `ref` is a reference to a `foo` which
3066 is initially set to refer to the value stored in `bar` - no extra syntax
3067 is needed to "Take the address of" `bar` - the fact that `ref` is a
3068 reference means that only the address make sense.
3070 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
3071 The same syntax is used for accessing fields both in structs and in
3072 references to structs. It would be correct to use `ref@.a`, but not
3075 `@new()` creates an object of whatever type is needed for the program
3076 to by type-correct. In future iterations of Ocean, arguments a
3077 constructor will access arguments, so the the syntax now looks like a
3078 function call. `@free` can be assigned any reference that was returned
3079 by `@new()`, and it will be freed. `@nil` is a value of whatever
3080 reference type is appropriate, and is stable and never the address of
3081 anything in the heap or on the stack. A reference can be assigned
3082 `@nil` or compared against that value.
3084 ###### declare terminals
3087 ###### type union fields
3090 struct type *referent;
3093 ###### value union fields
3096 ###### value functions
3098 static void reference_print_type(struct type *t, FILE *f)
3101 type_print(t->reference.referent, f);
3104 static int reference_cmp(struct type *tl, struct type *tr,
3105 struct value *left, struct value *right)
3107 return left->ref == right->ref ? 0 : 1;
3110 static void reference_dup(struct type *t,
3111 struct value *vold, struct value *vnew)
3113 vnew->ref = vold->ref;
3116 static void reference_free(struct type *t, struct value *v)
3118 /* Nothing to do here */
3121 static int reference_compat(struct type *require, struct type *have,
3122 enum val_rules rules)
3124 if (have->compat != require->compat)
3126 if (have->reference.referent != require->reference.referent)
3131 static int reference_test(struct type *type, struct value *val)
3133 return val->ref != NULL;
3136 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3137 struct fieldref *f, struct value **vp)
3139 struct type *rt = t->reference.referent;
3144 return rt->fieldref(rt, c, f, vp);
3146 type_err(c, "error: field reference on %1 is not supported",
3147 f->left, rt, 0, NULL);
3152 static struct type reference_prototype = {
3153 .print_type = reference_print_type,
3154 .cmp_eq = reference_cmp,
3155 .dup = reference_dup,
3156 .test = reference_test,
3157 .free = reference_free,
3158 .compat = reference_compat,
3159 .fieldref = reference_fieldref,
3160 .size = sizeof(void*),
3161 .align = sizeof(void*),
3167 struct type *t = find_type(c, $ID.txt);
3169 t = add_type(c, $ID.txt, NULL);
3172 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3173 $ID.txt.len, $ID.txt.txt);
3174 $0->reference.referent = t;
3177 ###### core functions
3178 static int text_is(struct text t, char *s)
3180 return (strlen(s) == t.len &&
3181 strncmp(s, t.txt, t.len) == 0);
3190 enum ref_func { RefNew, RefFree, RefNil } action;
3191 struct type *reftype;
3195 ###### SimpleStatement Grammar
3197 | @ IDENTIFIER = Expression ${ {
3198 struct ref *r = new_pos(ref, $ID);
3200 if (!text_is($ID.txt, "free"))
3201 tok_err(c, "error: only \"@free\" makes sense here",
3205 r->action = RefFree;
3209 ###### expression grammar
3210 | @ IDENTIFIER ( ) ${
3211 // Only 'new' valid here
3212 if (!text_is($ID.txt, "new")) {
3213 tok_err(c, "error: Only reference function is \"@new()\"",
3216 struct ref *r = new_pos(ref,$ID);
3222 // Only 'nil' valid here
3223 if (!text_is($ID.txt, "nil")) {
3224 tok_err(c, "error: Only reference value is \"@nil\"",
3227 struct ref *r = new_pos(ref,$ID);
3233 ###### print exec cases
3235 struct ref *r = cast(ref, e);
3236 switch (r->action) {
3238 printf("@new()"); break;
3240 printf("@nil"); break;
3242 do_indent(indent, "@free = ");
3243 print_exec(r->right, indent, bracket);
3249 ###### propagate exec cases
3251 struct ref *r = cast(ref, prog);
3252 switch (r->action) {
3254 if (type && type->free != reference_free) {
3255 type_err(c, "error: @new() can only be used with references, not %1",
3256 prog, type, 0, NULL);
3259 if (type && !r->reftype) {
3266 if (type && type->free != reference_free)
3267 type_err(c, "error: @nil can only be used with reference, not %1",
3268 prog, type, 0, NULL);
3269 if (type && !r->reftype) {
3276 t = propagate_types(r->right, c, perr_local, NULL, 0);
3277 if (t && t->free != reference_free)
3278 type_err(c, "error: @free can only be assigned a reference, not %1",
3287 ###### interp exec cases
3289 struct ref *r = cast(ref, e);
3290 switch (r->action) {
3293 rv.ref = calloc(1, r->reftype->reference.referent->size);
3294 rvtype = r->reftype;
3298 rvtype = r->reftype;
3301 rv = interp_exec(c, r->right, &rvtype);
3302 free_value(rvtype->reference.referent, rv.ref);
3310 ###### free exec cases
3312 struct ref *r = cast(ref, e);
3313 free_exec(r->right);
3318 ###### Expressions: dereference
3326 struct binode *b = new(binode);
3332 ###### print binode cases
3334 print_exec(b->left, -1, bracket);
3338 ###### propagate binode cases
3340 /* left must be a reference, and we return what it refers to */
3341 /* FIXME how can I pass the expected type down? */
3342 t = propagate_types(b->left, c, perr, NULL, 0);
3344 if (!t || t->free != reference_free)
3345 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3347 return t->reference.referent;
3350 ###### interp binode cases
3352 left = interp_exec(c, b->left, <ype);
3354 rvtype = ltype->reference.referent;
3361 A function is a chunk of code which can be passed parameters and can
3362 return results. Each function has a type which includes the set of
3363 parameters and the return value. As yet these types cannot be declared
3364 separately from the function itself.
3366 The parameters can be specified either in parentheses as a ';' separated
3369 ##### Example: function 1
3371 func main(av:[ac::number]string; env:[envc::number]string)
3374 or as an indented list of one parameter per line (though each line can
3375 be a ';' separated list)
3377 ##### Example: function 2
3380 argv:[argc::number]string
3381 env:[envc::number]string
3385 In the first case a return type can follow the parentheses after a colon,
3386 in the second it is given on a line starting with the word `return`.
3388 ##### Example: functions that return
3390 func add(a:number; b:number): number
3400 Rather than returning a type, the function can specify a set of local
3401 variables to return as a struct. The values of these variables when the
3402 function exits will be provided to the caller. For this the return type
3403 is replaced with a block of result declarations, either in parentheses
3404 or bracketed by `return` and `do`.
3406 ##### Example: functions returning multiple variables
3408 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3421 For constructing the lists we use a `List` binode, which will be
3422 further detailed when Expression Lists are introduced.
3424 ###### type union fields
3427 struct binode *params;
3428 struct type *return_type;
3429 struct variable *scope;
3430 int inline_result; // return value is at start of 'local'
3434 ###### value union fields
3435 struct exec *function;
3437 ###### type functions
3438 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3439 struct type *require, struct exec *args);
3441 ###### value functions
3443 static void function_free(struct type *type, struct value *val)
3445 free_exec(val->function);
3446 val->function = NULL;
3449 static int function_compat(struct type *require, struct type *have,
3450 enum val_rules rules)
3452 // FIXME can I do anything here yet?
3456 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3457 struct type *require, struct exec *args)
3459 /* This should be 'compat', but we don't have a 'tuple' type to
3460 * hold the type of 'args'
3462 struct binode *arg = cast(binode, args);
3463 struct binode *param = require->function.params;
3466 struct var *pv = cast(var, param->left);
3468 type_err(c, "error: insufficient arguments to function.",
3469 args, NULL, 0, NULL);
3473 propagate_types(arg->left, c, perr, pv->var->type, 0);
3474 param = cast(binode, param->right);
3475 arg = cast(binode, arg->right);
3478 type_err(c, "error: too many arguments to function.",
3479 args, NULL, 0, NULL);
3482 static void function_print(struct type *type, struct value *val, FILE *f)
3484 print_exec(val->function, 1, 0);
3487 static void function_print_type_decl(struct type *type, FILE *f)
3491 for (b = type->function.params; b; b = cast(binode, b->right)) {
3492 struct variable *v = cast(var, b->left)->var;
3493 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3494 v->constant ? "::" : ":");
3495 type_print(v->type, f);
3500 if (type->function.return_type != Tnone) {
3502 if (type->function.inline_result) {
3504 struct type *t = type->function.return_type;
3506 for (i = 0; i < t->structure.nfields; i++) {
3507 struct field *fl = t->structure.fields + i;
3510 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3511 type_print(fl->type, f);
3515 type_print(type->function.return_type, f);
3520 static void function_free_type(struct type *t)
3522 free_exec(t->function.params);
3525 static struct type function_prototype = {
3526 .size = sizeof(void*),
3527 .align = sizeof(void*),
3528 .free = function_free,
3529 .compat = function_compat,
3530 .check_args = function_check_args,
3531 .print = function_print,
3532 .print_type_decl = function_print_type_decl,
3533 .free_type = function_free_type,
3536 ###### declare terminals
3546 FuncName -> IDENTIFIER ${ {
3547 struct variable *v = var_decl(c, $1.txt);
3548 struct var *e = new_pos(var, $1);
3555 v = var_ref(c, $1.txt);
3557 type_err(c, "error: function '%v' redeclared",
3559 type_err(c, "info: this is where '%v' was first declared",
3560 v->where_decl, NULL, 0, NULL);
3566 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3567 | Args ArgsLine NEWLINE ${ {
3568 struct binode *b = $<AL;
3569 struct binode **bp = &b;
3571 bp = (struct binode **)&(*bp)->left;
3576 ArgsLine -> ${ $0 = NULL; }$
3577 | Varlist ${ $0 = $<1; }$
3578 | Varlist ; ${ $0 = $<1; }$
3580 Varlist -> Varlist ; ArgDecl ${
3581 $0 = new_pos(binode, $2);
3594 ArgDecl -> IDENTIFIER : FormalType ${ {
3595 struct variable *v = var_decl(c, $ID.txt);
3596 $0 = new_pos(var, $ID);
3603 ##### Function calls
3605 A function call can appear either as an expression or as a statement.
3606 We use a new 'Funcall' binode type to link the function with a list of
3607 arguments, form with the 'List' nodes.
3609 We have already seen the "Term" which is how a function call can appear
3610 in an expression. To parse a function call into a statement we include
3611 it in the "SimpleStatement Grammar" which will be described later.
3617 | Term ( ExpressionList ) ${ {
3618 struct binode *b = new(binode);
3621 b->right = reorder_bilist($<EL);
3625 struct binode *b = new(binode);
3632 ###### SimpleStatement Grammar
3634 | Term ( ExpressionList ) ${ {
3635 struct binode *b = new(binode);
3638 b->right = reorder_bilist($<EL);
3642 ###### print binode cases
3645 do_indent(indent, "");
3646 print_exec(b->left, -1, bracket);
3648 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3651 print_exec(b->left, -1, bracket);
3661 ###### propagate binode cases
3664 /* Every arg must match formal parameter, and result
3665 * is return type of function
3667 struct binode *args = cast(binode, b->right);
3668 struct var *v = cast(var, b->left);
3670 if (!v->var->type || v->var->type->check_args == NULL) {
3671 type_err(c, "error: attempt to call a non-function.",
3672 prog, NULL, 0, NULL);
3676 v->var->type->check_args(c, perr_local, v->var->type, args);
3677 if (v->var->type->function.inline_result)
3680 return v->var->type->function.return_type;
3683 ###### interp binode cases
3686 struct var *v = cast(var, b->left);
3687 struct type *t = v->var->type;
3688 void *oldlocal = c->local;
3689 int old_size = c->local_size;
3690 void *local = calloc(1, t->function.local_size);
3691 struct value *fbody = var_value(c, v->var);
3692 struct binode *arg = cast(binode, b->right);
3693 struct binode *param = t->function.params;
3696 struct var *pv = cast(var, param->left);
3697 struct type *vtype = NULL;
3698 struct value val = interp_exec(c, arg->left, &vtype);
3700 c->local = local; c->local_size = t->function.local_size;
3701 lval = var_value(c, pv->var);
3702 c->local = oldlocal; c->local_size = old_size;
3703 memcpy(lval, &val, vtype->size);
3704 param = cast(binode, param->right);
3705 arg = cast(binode, arg->right);
3707 c->local = local; c->local_size = t->function.local_size;
3708 if (t->function.inline_result && dtype) {
3709 _interp_exec(c, fbody->function, NULL, NULL);
3710 memcpy(dest, local, dtype->size);
3711 rvtype = ret.type = NULL;
3713 rv = interp_exec(c, fbody->function, &rvtype);
3714 c->local = oldlocal; c->local_size = old_size;
3719 ## Complex executables: statements and expressions
3721 Now that we have types and values and variables and most of the basic
3722 Terms which provide access to these, we can explore the more complex
3723 code that combine all of these to get useful work done. Specifically
3724 statements and expressions.
3726 Expressions are various combinations of Terms. We will use operator
3727 precedence to ensure correct parsing. The simplest Expression is just a
3728 Term - others will follow.
3733 Expression -> Term ${ $0 = $<Term; }$
3734 ## expression grammar
3736 ### Expressions: Conditional
3738 Our first user of the `binode` will be conditional expressions, which
3739 is a bit odd as they actually have three components. That will be
3740 handled by having 2 binodes for each expression. The conditional
3741 expression is the lowest precedence operator which is why we define it
3742 first - to start the precedence list.
3744 Conditional expressions are of the form "value `if` condition `else`
3745 other_value". They associate to the right, so everything to the right
3746 of `else` is part of an else value, while only a higher-precedence to
3747 the left of `if` is the if values. Between `if` and `else` there is no
3748 room for ambiguity, so a full conditional expression is allowed in
3754 ###### declare terminals
3758 ###### expression grammar
3760 | Expression if Expression else Expression $$ifelse ${ {
3761 struct binode *b1 = new(binode);
3762 struct binode *b2 = new(binode);
3772 ###### print binode cases
3775 b2 = cast(binode, b->right);
3776 if (bracket) printf("(");
3777 print_exec(b2->left, -1, bracket);
3779 print_exec(b->left, -1, bracket);
3781 print_exec(b2->right, -1, bracket);
3782 if (bracket) printf(")");
3785 ###### propagate binode cases
3788 /* cond must be Tbool, others must match */
3789 struct binode *b2 = cast(binode, b->right);
3792 propagate_types(b->left, c, perr_local, Tbool, 0);
3793 t = propagate_types(b2->left, c, perr, type, 0);
3794 t2 = propagate_types(b2->right, c, perr, type ?: t, 0);
3798 ###### interp binode cases
3801 struct binode *b2 = cast(binode, b->right);
3802 left = interp_exec(c, b->left, <ype);
3804 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3806 rv = interp_exec(c, b2->right, &rvtype);
3812 We take a brief detour, now that we have expressions, to describe lists
3813 of expressions. These will be needed for function parameters and
3814 possibly other situations. They seem generic enough to introduce here
3815 to be used elsewhere.
3817 And ExpressionList will use the `List` type of `binode`, building up at
3818 the end. And place where they are used will probably call
3819 `reorder_bilist()` to get a more normal first/next arrangement.
3821 ###### declare terminals
3824 `List` execs have no implicit semantics, so they are never propagated or
3825 interpreted. The can be printed as a comma separate list, which is how
3826 they are parsed. Note they are also used for function formal parameter
3827 lists. In that case a separate function is used to print them.
3829 ###### print binode cases
3833 print_exec(b->left, -1, bracket);
3836 b = cast(binode, b->right);
3840 ###### propagate binode cases
3841 case List: abort(); // NOTEST
3842 ###### interp binode cases
3843 case List: abort(); // NOTEST
3848 ExpressionList -> ExpressionList , Expression ${
3861 ### Expressions: Boolean
3863 The next class of expressions to use the `binode` will be Boolean
3864 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3865 have same corresponding precendence. The difference is that they don't
3866 evaluate the second expression if not necessary.
3875 ###### declare terminals
3880 ###### expression grammar
3881 | Expression or Expression ${ {
3882 struct binode *b = new(binode);
3888 | Expression or else Expression ${ {
3889 struct binode *b = new(binode);
3896 | Expression and Expression ${ {
3897 struct binode *b = new(binode);
3903 | Expression and then Expression ${ {
3904 struct binode *b = new(binode);
3911 | not Expression ${ {
3912 struct binode *b = new(binode);
3918 ###### print binode cases
3920 if (bracket) printf("(");
3921 print_exec(b->left, -1, bracket);
3923 print_exec(b->right, -1, bracket);
3924 if (bracket) printf(")");
3927 if (bracket) printf("(");
3928 print_exec(b->left, -1, bracket);
3929 printf(" and then ");
3930 print_exec(b->right, -1, bracket);
3931 if (bracket) printf(")");
3934 if (bracket) printf("(");
3935 print_exec(b->left, -1, bracket);
3937 print_exec(b->right, -1, bracket);
3938 if (bracket) printf(")");
3941 if (bracket) printf("(");
3942 print_exec(b->left, -1, bracket);
3943 printf(" or else ");
3944 print_exec(b->right, -1, bracket);
3945 if (bracket) printf(")");
3948 if (bracket) printf("(");
3950 print_exec(b->right, -1, bracket);
3951 if (bracket) printf(")");
3954 ###### propagate binode cases
3960 /* both must be Tbool, result is Tbool */
3961 propagate_types(b->left, c, perr, Tbool, 0);
3962 propagate_types(b->right, c, perr, Tbool, 0);
3963 if (type && type != Tbool)
3964 type_err(c, "error: %1 operation found where %2 expected", prog,
3969 ###### interp binode cases
3971 rv = interp_exec(c, b->left, &rvtype);
3972 right = interp_exec(c, b->right, &rtype);
3973 rv.bool = rv.bool && right.bool;
3976 rv = interp_exec(c, b->left, &rvtype);
3978 rv = interp_exec(c, b->right, NULL);
3981 rv = interp_exec(c, b->left, &rvtype);
3982 right = interp_exec(c, b->right, &rtype);
3983 rv.bool = rv.bool || right.bool;
3986 rv = interp_exec(c, b->left, &rvtype);
3988 rv = interp_exec(c, b->right, NULL);
3991 rv = interp_exec(c, b->right, &rvtype);
3995 ### Expressions: Comparison
3997 Of slightly higher precedence that Boolean expressions are Comparisons.
3998 A comparison takes arguments of any comparable type, but the two types
4001 To simplify the parsing we introduce an `eop` which can record an
4002 expression operator, and the `CMPop` non-terminal will match one of them.
4009 ###### ast functions
4010 static void free_eop(struct eop *e)
4024 ###### declare terminals
4025 $LEFT < > <= >= == != CMPop
4027 ###### expression grammar
4028 | Expression CMPop Expression ${ {
4029 struct binode *b = new(binode);
4039 CMPop -> < ${ $0.op = Less; }$
4040 | > ${ $0.op = Gtr; }$
4041 | <= ${ $0.op = LessEq; }$
4042 | >= ${ $0.op = GtrEq; }$
4043 | == ${ $0.op = Eql; }$
4044 | != ${ $0.op = NEql; }$
4046 ###### print binode cases
4054 if (bracket) printf("(");
4055 print_exec(b->left, -1, bracket);
4057 case Less: printf(" < "); break;
4058 case LessEq: printf(" <= "); break;
4059 case Gtr: printf(" > "); break;
4060 case GtrEq: printf(" >= "); break;
4061 case Eql: printf(" == "); break;
4062 case NEql: printf(" != "); break;
4063 default: abort(); // NOTEST
4065 print_exec(b->right, -1, bracket);
4066 if (bracket) printf(")");
4069 ###### propagate binode cases
4076 /* Both must match but not be labels, result is Tbool */
4077 t = propagate_types(b->left, c, perr, NULL, 0);
4079 propagate_types(b->right, c, perr, t, 0);
4081 t = propagate_types(b->right, c, perr, NULL, 0); // UNTESTED
4083 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
4085 if (!type_compat(type, Tbool, 0))
4086 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4087 Tbool, rules, type);
4091 ###### interp binode cases
4100 left = interp_exec(c, b->left, <ype);
4101 right = interp_exec(c, b->right, &rtype);
4102 cmp = value_cmp(ltype, rtype, &left, &right);
4105 case Less: rv.bool = cmp < 0; break;
4106 case LessEq: rv.bool = cmp <= 0; break;
4107 case Gtr: rv.bool = cmp > 0; break;
4108 case GtrEq: rv.bool = cmp >= 0; break;
4109 case Eql: rv.bool = cmp == 0; break;
4110 case NEql: rv.bool = cmp != 0; break;
4111 default: rv.bool = 0; break; // NOTEST
4116 ### Expressions: Arithmetic etc.
4118 The remaining expressions with the highest precedence are arithmetic,
4119 string concatenation, string conversion, and testing. String concatenation
4120 (`++`) has the same precedence as multiplication and division, but lower
4123 Testing comes in two forms. A single question mark (`?`) is a uniary
4124 operator which converts come types into Boolean. The general meaning is
4125 "is this a value value" and there will be more uses as the language
4126 develops. A double questionmark (`??`) is a binary operator (Choose),
4127 with same precedence as multiplication, which returns the LHS if it
4128 tests successfully, else returns the RHS.
4130 String conversion is a temporary feature until I get a better type
4131 system. `$` is a prefix operator which expects a string and returns
4134 `+` and `-` are both infix and prefix operations (where they are
4135 absolute value and negation). These have different operator names.
4137 We also have a 'Bracket' operator which records where parentheses were
4138 found. This makes it easy to reproduce these when printing. Possibly I
4139 should only insert brackets were needed for precedence. Putting
4140 parentheses around an expression converts it into a Term,
4146 Absolute, Negate, Test,
4150 ###### declare terminals
4152 $LEFT * / % ++ ?? Top
4156 ###### expression grammar
4157 | Expression Eop Expression ${ {
4158 struct binode *b = new(binode);
4165 | Expression Top Expression ${ {
4166 struct binode *b = new(binode);
4173 | Uop Expression ${ {
4174 struct binode *b = new(binode);
4182 | ( Expression ) ${ {
4183 struct binode *b = new_pos(binode, $1);
4192 Eop -> + ${ $0.op = Plus; }$
4193 | - ${ $0.op = Minus; }$
4195 Uop -> + ${ $0.op = Absolute; }$
4196 | - ${ $0.op = Negate; }$
4197 | $ ${ $0.op = StringConv; }$
4198 | ? ${ $0.op = Test; }$
4200 Top -> * ${ $0.op = Times; }$
4201 | / ${ $0.op = Divide; }$
4202 | % ${ $0.op = Rem; }$
4203 | ++ ${ $0.op = Concat; }$
4204 | ?? ${ $0.op = Choose; }$
4206 ###### print binode cases
4214 if (bracket) printf("(");
4215 print_exec(b->left, indent, bracket);
4217 case Plus: fputs(" + ", stdout); break;
4218 case Minus: fputs(" - ", stdout); break;
4219 case Times: fputs(" * ", stdout); break;
4220 case Divide: fputs(" / ", stdout); break;
4221 case Rem: fputs(" % ", stdout); break;
4222 case Concat: fputs(" ++ ", stdout); break;
4223 case Choose: fputs(" ?? ", stdout); break;
4224 default: abort(); // NOTEST
4226 print_exec(b->right, indent, bracket);
4227 if (bracket) printf(")");
4233 if (bracket) printf("(");
4235 case Absolute: fputs("+", stdout); break;
4236 case Negate: fputs("-", stdout); break;
4237 case StringConv: fputs("$", stdout); break;
4238 case Test: fputs("?", stdout); break;
4239 default: abort(); // NOTEST
4241 print_exec(b->right, indent, bracket);
4242 if (bracket) printf(")");
4246 print_exec(b->right, indent, bracket);
4250 ###### propagate binode cases
4256 /* both must be numbers, result is Tnum */
4259 /* as propagate_types ignores a NULL,
4260 * unary ops fit here too */
4261 propagate_types(b->left, c, perr, Tnum, 0);
4262 propagate_types(b->right, c, perr, Tnum, 0);
4263 if (!type_compat(type, Tnum, 0))
4264 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4270 /* both must be Tstr, result is Tstr */
4271 propagate_types(b->left, c, perr, Tstr, 0);
4272 propagate_types(b->right, c, perr, Tstr, 0);
4273 if (!type_compat(type, Tstr, 0))
4274 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4280 /* op must be string, result is number */
4281 propagate_types(b->left, c, perr, Tstr, 0);
4282 if (!type_compat(type, Tnum, 0))
4283 type_err(c, // UNTESTED
4284 "error: Can only convert string to number, not %1",
4285 prog, type, 0, NULL);
4290 /* LHS must support ->test, result is Tbool */
4291 t = propagate_types(b->right, c, perr, NULL, 0);
4293 type_err(c, "error: '?' requires a testable value, not %1",
4299 /* LHS and RHS must match and are returned. Must support
4302 t = propagate_types(b->left, c, perr, type, rules);
4303 t = propagate_types(b->right, c, perr, t, rules);
4304 if (t && t->test == NULL)
4305 type_err(c, "error: \"??\" requires a testable value, not %1",
4311 return propagate_types(b->right, c, perr, type, rules);
4313 ###### interp binode cases
4316 rv = interp_exec(c, b->left, &rvtype);
4317 right = interp_exec(c, b->right, &rtype);
4318 mpq_add(rv.num, rv.num, right.num);
4321 rv = interp_exec(c, b->left, &rvtype);
4322 right = interp_exec(c, b->right, &rtype);
4323 mpq_sub(rv.num, rv.num, right.num);
4326 rv = interp_exec(c, b->left, &rvtype);
4327 right = interp_exec(c, b->right, &rtype);
4328 mpq_mul(rv.num, rv.num, right.num);
4331 rv = interp_exec(c, b->left, &rvtype);
4332 right = interp_exec(c, b->right, &rtype);
4333 mpq_div(rv.num, rv.num, right.num);
4338 left = interp_exec(c, b->left, <ype);
4339 right = interp_exec(c, b->right, &rtype);
4340 mpz_init(l); mpz_init(r); mpz_init(rem);
4341 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4342 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4343 mpz_tdiv_r(rem, l, r);
4344 val_init(Tnum, &rv);
4345 mpq_set_z(rv.num, rem);
4346 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4351 rv = interp_exec(c, b->right, &rvtype);
4352 mpq_neg(rv.num, rv.num);
4355 rv = interp_exec(c, b->right, &rvtype);
4356 mpq_abs(rv.num, rv.num);
4359 rv = interp_exec(c, b->right, &rvtype);
4362 left = interp_exec(c, b->left, <ype);
4363 right = interp_exec(c, b->right, &rtype);
4365 rv.str = text_join(left.str, right.str);
4368 right = interp_exec(c, b->right, &rvtype);
4372 struct text tx = right.str;
4375 if (tx.txt[0] == '-') {
4376 neg = 1; // UNTESTED
4377 tx.txt++; // UNTESTED
4378 tx.len--; // UNTESTED
4380 if (number_parse(rv.num, tail, tx) == 0)
4381 mpq_init(rv.num); // UNTESTED
4383 mpq_neg(rv.num, rv.num); // UNTESTED
4385 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4389 right = interp_exec(c, b->right, &rtype);
4391 rv.bool = !!rtype->test(rtype, &right);
4394 left = interp_exec(c, b->left, <ype);
4395 if (ltype->test(ltype, &left)) {
4400 rv = interp_exec(c, b->right, &rvtype);
4403 ###### value functions
4405 static struct text text_join(struct text a, struct text b)
4408 rv.len = a.len + b.len;
4409 rv.txt = malloc(rv.len);
4410 memcpy(rv.txt, a.txt, a.len);
4411 memcpy(rv.txt+a.len, b.txt, b.len);
4415 ### Blocks, Statements, and Statement lists.
4417 Now that we have expressions out of the way we need to turn to
4418 statements. There are simple statements and more complex statements.
4419 Simple statements do not contain (syntactic) newlines, complex statements do.
4421 Statements often come in sequences and we have corresponding simple
4422 statement lists and complex statement lists.
4423 The former comprise only simple statements separated by semicolons.
4424 The later comprise complex statements and simple statement lists. They are
4425 separated by newlines. Thus the semicolon is only used to separate
4426 simple statements on the one line. This may be overly restrictive,
4427 but I'm not sure I ever want a complex statement to share a line with
4430 Note that a simple statement list can still use multiple lines if
4431 subsequent lines are indented, so
4433 ###### Example: wrapped simple statement list
4438 is a single simple statement list. This might allow room for
4439 confusion, so I'm not set on it yet.
4441 A simple statement list needs no extra syntax. A complex statement
4442 list has two syntactic forms. It can be enclosed in braces (much like
4443 C blocks), or it can be introduced by an indent and continue until an
4444 unindented newline (much like Python blocks). With this extra syntax
4445 it is referred to as a block.
4447 Note that a block does not have to include any newlines if it only
4448 contains simple statements. So both of:
4450 if condition: a=b; d=f
4452 if condition { a=b; print f }
4456 In either case the list is constructed from a `binode` list with
4457 `Block` as the operator. When parsing the list it is most convenient
4458 to append to the end, so a list is a list and a statement. When using
4459 the list it is more convenient to consider a list to be a statement
4460 and a list. So we need a function to re-order a list.
4461 `reorder_bilist` serves this purpose.
4463 The only stand-alone statement we introduce at this stage is `pass`
4464 which does nothing and is represented as a `NULL` pointer in a `Block`
4465 list. Other stand-alone statements will follow once the infrastructure
4468 As many statements will use binodes, we declare a binode pointer 'b' in
4469 the common header for all reductions to use.
4471 ###### Parser: reduce
4482 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4483 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4484 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4485 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4486 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4488 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4489 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4490 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4491 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4492 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4494 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4495 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4496 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4498 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4499 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4500 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4501 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4502 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4504 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4506 ComplexStatements -> ComplexStatements ComplexStatement ${
4516 | ComplexStatement ${
4528 ComplexStatement -> SimpleStatements Newlines ${
4529 $0 = reorder_bilist($<SS);
4531 | SimpleStatements ; Newlines ${
4532 $0 = reorder_bilist($<SS);
4534 ## ComplexStatement Grammar
4537 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4543 | SimpleStatement ${
4552 SimpleStatement -> pass ${ $0 = NULL; }$
4553 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4554 ## SimpleStatement Grammar
4556 ###### print binode cases
4560 if (b->left == NULL) // UNTESTED
4561 printf("pass"); // UNTESTED
4563 print_exec(b->left, indent, bracket); // UNTESTED
4564 if (b->right) { // UNTESTED
4565 printf("; "); // UNTESTED
4566 print_exec(b->right, indent, bracket); // UNTESTED
4569 // block, one per line
4570 if (b->left == NULL)
4571 do_indent(indent, "pass\n");
4573 print_exec(b->left, indent, bracket);
4575 print_exec(b->right, indent, bracket);
4579 ###### propagate binode cases
4582 /* If any statement returns something other than Tnone
4583 * or Tbool then all such must return same type.
4584 * As each statement may be Tnone or something else,
4585 * we must always pass NULL (unknown) down, otherwise an incorrect
4586 * error might occur. We never return Tnone unless it is
4591 for (e = b; e; e = cast(binode, e->right)) {
4592 t = propagate_types(e->left, c, perr, NULL, rules);
4593 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4595 if (t == Tnone && e->right)
4596 /* Only the final statement *must* return a value
4604 type_err(c, "error: expected %1, found %2",
4605 e->left, type, rules, t);
4611 ###### interp binode cases
4613 while (rvtype == Tnone &&
4616 rv = interp_exec(c, b->left, &rvtype);
4617 b = cast(binode, b->right);
4621 ### The Print statement
4623 `print` is a simple statement that takes a comma-separated list of
4624 expressions and prints the values separated by spaces and terminated
4625 by a newline. No control of formatting is possible.
4627 `print` uses `ExpressionList` to collect the expressions and stores them
4628 on the left side of a `Print` binode unlessthere is a trailing comma
4629 when the list is stored on the `right` side and no trailing newline is
4635 ##### declare terminals
4638 ###### SimpleStatement Grammar
4640 | print ExpressionList ${
4641 $0 = b = new_pos(binode, $1);
4644 b->left = reorder_bilist($<EL);
4646 | print ExpressionList , ${ {
4647 $0 = b = new_pos(binode, $1);
4649 b->right = reorder_bilist($<EL);
4653 $0 = b = new_pos(binode, $1);
4659 ###### print binode cases
4662 do_indent(indent, "print");
4664 print_exec(b->right, -1, bracket);
4667 print_exec(b->left, -1, bracket);
4672 ###### propagate binode cases
4675 /* don't care but all must be consistent */
4677 b = cast(binode, b->left);
4679 b = cast(binode, b->right);
4681 propagate_types(b->left, c, perr_local, NULL, 0);
4682 b = cast(binode, b->right);
4686 ###### interp binode cases
4690 struct binode *b2 = cast(binode, b->left);
4692 b2 = cast(binode, b->right);
4693 for (; b2; b2 = cast(binode, b2->right)) {
4694 left = interp_exec(c, b2->left, <ype);
4695 print_value(ltype, &left, stdout);
4696 free_value(ltype, &left);
4700 if (b->right == NULL)
4706 ###### Assignment statement
4708 An assignment will assign a value to a variable, providing it hasn't
4709 been declared as a constant. The analysis phase ensures that the type
4710 will be correct so the interpreter just needs to perform the
4711 calculation. There is a form of assignment which declares a new
4712 variable as well as assigning a value. If a name is used before
4713 it is declared, it is assumed to be a global constant which are allowed to
4714 be declared at any time.
4720 ###### declare terminals
4723 ###### SimpleStatement Grammar
4724 | Term = Expression ${
4725 $0 = b= new(binode);
4730 | VariableDecl = Expression ${
4731 $0 = b= new(binode);
4738 if ($1->var->where_set == NULL) {
4740 "Variable declared with no type or value: %v",
4744 $0 = b = new(binode);
4751 ###### print binode cases
4754 do_indent(indent, "");
4755 print_exec(b->left, -1, bracket);
4757 print_exec(b->right, -1, bracket);
4764 struct variable *v = cast(var, b->left)->var;
4765 do_indent(indent, "");
4766 print_exec(b->left, -1, bracket);
4767 if (cast(var, b->left)->var->constant) {
4769 if (v->explicit_type) {
4770 type_print(v->type, stdout);
4775 if (v->explicit_type) {
4776 type_print(v->type, stdout);
4782 print_exec(b->right, -1, bracket);
4789 ###### propagate binode cases
4793 /* Both must match and not be labels,
4794 * Type must support 'dup',
4795 * For Assign, left must not be constant.
4798 *perr &= ~(Erval | Econst);
4799 t = propagate_types(b->left, c, perr, NULL, 0);
4804 if (propagate_types(b->right, c, perr_local, t, 0) != t &&
4805 *perr_local & Efail)
4806 if (b->left->type == Xvar)
4807 type_err(c, "info: variable '%v' was set as %1 here.",
4808 cast(var, b->left)->var->where_set, t, rules, NULL);
4810 t = propagate_types(b->right, c, perr_local, NULL, 0);
4812 propagate_types(b->left, c, perr, t, 0);
4815 type_err(c, "error: cannot assign to an rval", b,
4817 else if (b->op == Assign && (*perr & Econst)) {
4818 type_err(c, "error: Cannot assign to a constant: %v",
4819 b->left, NULL, 0, NULL);
4820 if (b->left->type == Xvar) {
4821 struct var *var = cast(var, b->left);
4822 struct variable *v = var->var;
4823 type_err(c, "info: name was defined as a constant here",
4824 v->where_decl, NULL, 0, NULL);
4827 if (t && t->dup == NULL && !(*perr_local & Emaycopy))
4828 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4833 ###### interp binode cases
4836 lleft = linterp_exec(c, b->left, <ype);
4838 dinterp_exec(c, b->right, lleft, ltype, 1);
4844 struct variable *v = cast(var, b->left)->var;
4847 val = var_value(c, v);
4848 if (v->type->prepare_type)
4849 v->type->prepare_type(c, v->type, 0);
4851 dinterp_exec(c, b->right, val, v->type, 0);
4853 val_init(v->type, val);
4857 ### The `use` statement
4859 The `use` statement is the last "simple" statement. It is needed when a
4860 statement block can return a value. This includes the body of a
4861 function which has a return type, and the "condition" code blocks in
4862 `if`, `while`, and `switch` statements.
4867 ###### declare terminals
4870 ###### SimpleStatement Grammar
4872 $0 = b = new_pos(binode, $1);
4877 ###### print binode cases
4880 do_indent(indent, "use ");
4881 print_exec(b->right, -1, bracket);
4886 ###### propagate binode cases
4889 /* result matches value */
4890 return propagate_types(b->right, c, perr, type, 0);
4892 ###### interp binode cases
4895 rv = interp_exec(c, b->right, &rvtype);
4898 ### The Conditional Statement
4900 This is the biggy and currently the only complex statement. This
4901 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4902 It is comprised of a number of parts, all of which are optional though
4903 set combinations apply. Each part is (usually) a key word (`then` is
4904 sometimes optional) followed by either an expression or a code block,
4905 except the `casepart` which is a "key word and an expression" followed
4906 by a code block. The code-block option is valid for all parts and,
4907 where an expression is also allowed, the code block can use the `use`
4908 statement to report a value. If the code block does not report a value
4909 the effect is similar to reporting `True`.
4911 The `else` and `case` parts, as well as `then` when combined with
4912 `if`, can contain a `use` statement which will apply to some
4913 containing conditional statement. `for` parts, `do` parts and `then`
4914 parts used with `for` can never contain a `use`, except in some
4915 subordinate conditional statement.
4917 If there is a `forpart`, it is executed first, only once.
4918 If there is a `dopart`, then it is executed repeatedly providing
4919 always that the `condpart` or `cond`, if present, does not return a non-True
4920 value. `condpart` can fail to return any value if it simply executes
4921 to completion. This is treated the same as returning `True`.
4923 If there is a `thenpart` it will be executed whenever the `condpart`
4924 or `cond` returns True (or does not return any value), but this will happen
4925 *after* `dopart` (when present).
4927 If `elsepart` is present it will be executed at most once when the
4928 condition returns `False` or some value that isn't `True` and isn't
4929 matched by any `casepart`. If there are any `casepart`s, they will be
4930 executed when the condition returns a matching value.
4932 The particular sorts of values allowed in case parts has not yet been
4933 determined in the language design, so nothing is prohibited.
4935 The various blocks in this complex statement potentially provide scope
4936 for variables as described earlier. Each such block must include the
4937 "OpenScope" nonterminal before parsing the block, and must call
4938 `var_block_close()` when closing the block.
4940 The code following "`if`", "`switch`" and "`for`" does not get its own
4941 scope, but is in a scope covering the whole statement, so names
4942 declared there cannot be redeclared elsewhere. Similarly the
4943 condition following "`while`" is in a scope the covers the body
4944 ("`do`" part) of the loop, and which does not allow conditional scope
4945 extension. Code following "`then`" (both looping and non-looping),
4946 "`else`" and "`case`" each get their own local scope.
4948 The type requirements on the code block in a `whilepart` are quite
4949 unusal. It is allowed to return a value of some identifiable type, in
4950 which case the loop aborts and an appropriate `casepart` is run, or it
4951 can return a Boolean, in which case the loop either continues to the
4952 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4953 This is different both from the `ifpart` code block which is expected to
4954 return a Boolean, or the `switchpart` code block which is expected to
4955 return the same type as the casepart values. The correct analysis of
4956 the type of the `whilepart` code block is the reason for the
4957 `Rboolok` flag which is passed to `propagate_types()`.
4959 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4960 defined. As there are two scopes which cover multiple parts - one for
4961 the whole statement and one for "while" and "do" - and as we will use
4962 the 'struct exec' to track scopes, we actually need two new types of
4963 exec. One is a `binode` for the looping part, the rest is the
4964 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4965 casepart` to track a list of case parts.
4976 struct exec *action;
4977 struct casepart *next;
4979 struct cond_statement {
4981 struct exec *forpart, *condpart, *thenpart, *elsepart;
4982 struct binode *looppart;
4983 struct casepart *casepart;
4986 ###### ast functions
4988 static void free_casepart(struct casepart *cp)
4992 free_exec(cp->value);
4993 free_exec(cp->action);
5000 static void free_cond_statement(struct cond_statement *s)
5004 free_exec(s->forpart);
5005 free_exec(s->condpart);
5006 free_exec(s->looppart);
5007 free_exec(s->thenpart);
5008 free_exec(s->elsepart);
5009 free_casepart(s->casepart);
5013 ###### free exec cases
5014 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
5016 ###### ComplexStatement Grammar
5017 | CondStatement ${ $0 = $<1; }$
5019 ###### declare terminals
5020 $TERM for then while do
5027 // A CondStatement must end with EOL, as does CondSuffix and
5029 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
5030 // may or may not end with EOL
5031 // WhilePart and IfPart include an appropriate Suffix
5033 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
5034 // them. WhilePart opens and closes its own scope.
5035 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
5038 $0->thenpart = $<TP;
5039 $0->looppart = $<WP;
5040 var_block_close(c, CloseSequential, $0);
5042 | ForPart OptNL WhilePart CondSuffix ${
5045 $0->looppart = $<WP;
5046 var_block_close(c, CloseSequential, $0);
5048 | WhilePart CondSuffix ${
5050 $0->looppart = $<WP;
5052 | SwitchPart OptNL CasePart CondSuffix ${
5054 $0->condpart = $<SP;
5055 $CP->next = $0->casepart;
5056 $0->casepart = $<CP;
5057 var_block_close(c, CloseSequential, $0);
5059 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
5061 $0->condpart = $<SP;
5062 $CP->next = $0->casepart;
5063 $0->casepart = $<CP;
5064 var_block_close(c, CloseSequential, $0);
5066 | IfPart IfSuffix ${
5068 $0->condpart = $IP.condpart; $IP.condpart = NULL;
5069 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
5070 // This is where we close an "if" statement
5071 var_block_close(c, CloseSequential, $0);
5074 CondSuffix -> IfSuffix ${
5077 | Newlines CasePart CondSuffix ${
5079 $CP->next = $0->casepart;
5080 $0->casepart = $<CP;
5082 | CasePart CondSuffix ${
5084 $CP->next = $0->casepart;
5085 $0->casepart = $<CP;
5088 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
5089 | Newlines ElsePart ${ $0 = $<EP; }$
5090 | ElsePart ${$0 = $<EP; }$
5092 ElsePart -> else OpenBlock Newlines ${
5093 $0 = new(cond_statement);
5094 $0->elsepart = $<OB;
5095 var_block_close(c, CloseElse, $0->elsepart);
5097 | else OpenScope CondStatement ${
5098 $0 = new(cond_statement);
5099 $0->elsepart = $<CS;
5100 var_block_close(c, CloseElse, $0->elsepart);
5104 CasePart -> case Expression OpenScope ColonBlock ${
5105 $0 = calloc(1,sizeof(struct casepart));
5108 var_block_close(c, CloseParallel, $0->action);
5112 // These scopes are closed in CondStatement
5113 ForPart -> for OpenBlock ${
5117 ThenPart -> then OpenBlock ${
5119 var_block_close(c, CloseSequential, $0);
5123 // This scope is closed in CondStatement
5124 WhilePart -> while UseBlock OptNL do OpenBlock ${
5129 var_block_close(c, CloseSequential, $0->right);
5130 var_block_close(c, CloseSequential, $0);
5132 | while OpenScope Expression OpenScope ColonBlock ${
5137 var_block_close(c, CloseSequential, $0->right);
5138 var_block_close(c, CloseSequential, $0);
5142 IfPart -> if UseBlock OptNL then OpenBlock ${
5145 var_block_close(c, CloseParallel, $0.thenpart);
5147 | if OpenScope Expression OpenScope ColonBlock ${
5150 var_block_close(c, CloseParallel, $0.thenpart);
5152 | if OpenScope Expression OpenScope OptNL then Block ${
5155 var_block_close(c, CloseParallel, $0.thenpart);
5159 // This scope is closed in CondStatement
5160 SwitchPart -> switch OpenScope Expression ${
5163 | switch UseBlock ${
5167 ###### print binode cases
5169 if (b->left && b->left->type == Xbinode &&
5170 cast(binode, b->left)->op == Block) {
5172 do_indent(indent, "while {\n");
5174 do_indent(indent, "while\n");
5175 print_exec(b->left, indent+1, bracket);
5177 do_indent(indent, "} do {\n");
5179 do_indent(indent, "do\n");
5180 print_exec(b->right, indent+1, bracket);
5182 do_indent(indent, "}\n");
5184 do_indent(indent, "while ");
5185 print_exec(b->left, 0, bracket);
5190 print_exec(b->right, indent+1, bracket);
5192 do_indent(indent, "}\n");
5196 ###### print exec cases
5198 case Xcond_statement:
5200 struct cond_statement *cs = cast(cond_statement, e);
5201 struct casepart *cp;
5203 do_indent(indent, "for");
5204 if (bracket) printf(" {\n"); else printf("\n");
5205 print_exec(cs->forpart, indent+1, bracket);
5208 do_indent(indent, "} then {\n");
5210 do_indent(indent, "then\n");
5211 print_exec(cs->thenpart, indent+1, bracket);
5213 if (bracket) do_indent(indent, "}\n");
5216 print_exec(cs->looppart, indent, bracket);
5220 do_indent(indent, "switch");
5222 do_indent(indent, "if");
5223 if (cs->condpart && cs->condpart->type == Xbinode &&
5224 cast(binode, cs->condpart)->op == Block) {
5229 print_exec(cs->condpart, indent+1, bracket);
5231 do_indent(indent, "}\n");
5233 do_indent(indent, "then\n");
5234 print_exec(cs->thenpart, indent+1, bracket);
5238 print_exec(cs->condpart, 0, bracket);
5244 print_exec(cs->thenpart, indent+1, bracket);
5246 do_indent(indent, "}\n");
5251 for (cp = cs->casepart; cp; cp = cp->next) {
5252 do_indent(indent, "case ");
5253 print_exec(cp->value, -1, 0);
5258 print_exec(cp->action, indent+1, bracket);
5260 do_indent(indent, "}\n");
5263 do_indent(indent, "else");
5268 print_exec(cs->elsepart, indent+1, bracket);
5270 do_indent(indent, "}\n");
5275 ###### propagate binode cases
5277 t = propagate_types(b->right, c, perr_local, Tnone, 0);
5278 if (!type_compat(Tnone, t, 0))
5279 *perr |= Efail; // UNTESTED
5280 return propagate_types(b->left, c, perr, type, rules);
5282 ###### propagate exec cases
5283 case Xcond_statement:
5285 // forpart and looppart->right must return Tnone
5286 // thenpart must return Tnone if there is a loopart,
5287 // otherwise it is like elsepart.
5289 // be bool if there is no casepart
5290 // match casepart->values if there is a switchpart
5291 // either be bool or match casepart->value if there
5293 // elsepart and casepart->action must match the return type
5294 // expected of this statement.
5295 struct cond_statement *cs = cast(cond_statement, prog);
5296 struct casepart *cp;
5298 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5299 if (!type_compat(Tnone, t, 0))
5300 *perr |= Efail; // UNTESTED
5303 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5304 if (!type_compat(Tnone, t, 0))
5305 *perr |= Efail; // UNTESTED
5307 if (cs->casepart == NULL) {
5308 propagate_types(cs->condpart, c, perr, Tbool, 0);
5309 propagate_types(cs->looppart, c, perr, Tbool, 0);
5311 /* Condpart must match case values, with bool permitted */
5313 for (cp = cs->casepart;
5314 cp && !t; cp = cp->next)
5315 t = propagate_types(cp->value, c, perr, NULL, 0);
5316 if (!t && cs->condpart)
5317 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5318 if (!t && cs->looppart)
5319 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5320 // Now we have a type (I hope) push it down
5322 for (cp = cs->casepart; cp; cp = cp->next)
5323 propagate_types(cp->value, c, perr, t, 0);
5324 propagate_types(cs->condpart, c, perr, t, Rboolok);
5325 propagate_types(cs->looppart, c, perr, t, Rboolok);
5328 // (if)then, else, and case parts must return expected type.
5329 if (!cs->looppart && !type)
5330 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5332 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5333 for (cp = cs->casepart;
5335 cp = cp->next) // UNTESTED
5336 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5339 propagate_types(cs->thenpart, c, perr, type, rules);
5340 propagate_types(cs->elsepart, c, perr, type, rules);
5341 for (cp = cs->casepart; cp ; cp = cp->next)
5342 propagate_types(cp->action, c, perr, type, rules);
5348 ###### interp binode cases
5350 // This just performs one iterration of the loop
5351 rv = interp_exec(c, b->left, &rvtype);
5352 if (rvtype == Tnone ||
5353 (rvtype == Tbool && rv.bool != 0))
5354 // rvtype is Tnone or Tbool, doesn't need to be freed
5355 interp_exec(c, b->right, NULL);
5358 ###### interp exec cases
5359 case Xcond_statement:
5361 struct value v, cnd;
5362 struct type *vtype, *cndtype;
5363 struct casepart *cp;
5364 struct cond_statement *cs = cast(cond_statement, e);
5367 interp_exec(c, cs->forpart, NULL);
5369 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5370 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5371 interp_exec(c, cs->thenpart, NULL);
5373 cnd = interp_exec(c, cs->condpart, &cndtype);
5374 if ((cndtype == Tnone ||
5375 (cndtype == Tbool && cnd.bool != 0))) {
5376 // cnd is Tnone or Tbool, doesn't need to be freed
5377 rv = interp_exec(c, cs->thenpart, &rvtype);
5378 // skip else (and cases)
5382 for (cp = cs->casepart; cp; cp = cp->next) {
5383 v = interp_exec(c, cp->value, &vtype);
5384 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5385 free_value(vtype, &v);
5386 free_value(cndtype, &cnd);
5387 rv = interp_exec(c, cp->action, &rvtype);
5390 free_value(vtype, &v);
5392 free_value(cndtype, &cnd);
5394 rv = interp_exec(c, cs->elsepart, &rvtype);
5401 ### Top level structure
5403 All the language elements so far can be used in various places. Now
5404 it is time to clarify what those places are.
5406 At the top level of a file there will be a number of declarations.
5407 Many of the things that can be declared haven't been described yet,
5408 such as functions, procedures, imports, and probably more.
5409 For now there are two sorts of things that can appear at the top
5410 level. They are predefined constants, `struct` types, and the `main`
5411 function. While the syntax will allow the `main` function to appear
5412 multiple times, that will trigger an error if it is actually attempted.
5414 The various declarations do not return anything. They store the
5415 various declarations in the parse context.
5417 ###### Parser: grammar
5420 Ocean -> OptNL DeclarationList
5422 ## declare terminals
5430 DeclarationList -> Declaration
5431 | DeclarationList Declaration
5433 Declaration -> ERROR Newlines ${
5434 tok_err(c, // UNTESTED
5435 "error: unhandled parse error", &$1);
5441 ## top level grammar
5445 ### The `const` section
5447 As well as being defined in with the code that uses them, constants can
5448 be declared at the top level. These have full-file scope, so they are
5449 always `InScope`, even before(!) they have been declared. The value of
5450 a top level constant can be given as an expression, and this is
5451 evaluated after parsing and before execution.
5453 A function call can be used to evaluate a constant, but it will not have
5454 access to any program state, once such statement becomes meaningful.
5455 e.g. arguments and filesystem will not be visible.
5457 Constants are defined in a section that starts with the reserved word
5458 `const` and then has a block with a list of assignment statements.
5459 For syntactic consistency, these must use the double-colon syntax to
5460 make it clear that they are constants. Type can also be given: if
5461 not, the type will be determined during analysis, as with other
5464 ###### parse context
5465 struct binode *constlist;
5467 ###### top level grammar
5471 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5472 | const { SimpleConstList } Newlines
5473 | const IN OptNL ConstList OUT Newlines
5474 | const SimpleConstList Newlines
5476 ConstList -> ConstList SimpleConstLine
5479 SimpleConstList -> SimpleConstList ; Const
5483 SimpleConstLine -> SimpleConstList Newlines
5484 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5487 CType -> Type ${ $0 = $<1; }$
5491 Const -> IDENTIFIER :: CType = Expression ${ {
5493 struct binode *bl, *bv;
5494 struct var *var = new_pos(var, $ID);
5496 v = var_decl(c, $ID.txt);
5498 v->where_decl = var;
5504 v = var_ref(c, $1.txt);
5505 if (v->type == Tnone) {
5506 v->where_decl = var;
5512 tok_err(c, "error: name already declared", &$1);
5513 type_err(c, "info: this is where '%v' was first declared",
5514 v->where_decl, NULL, 0, NULL);
5526 bl->left = c->constlist;
5531 ###### core functions
5532 static void resolve_consts(struct parse_context *c)
5536 enum { none, some, cannot } progress = none;
5538 c->constlist = reorder_bilist(c->constlist);
5541 for (b = cast(binode, c->constlist); b;
5542 b = cast(binode, b->right)) {
5544 struct binode *vb = cast(binode, b->left);
5545 struct var *v = cast(var, vb->left);
5546 if (v->var->frame_pos >= 0)
5550 propagate_types(vb->right, c, &perr,
5552 } while (perr & Eretry);
5554 c->parse_error += 1;
5555 else if (!(perr & Eruntime)) {
5557 struct value res = interp_exec(
5558 c, vb->right, &v->var->type);
5559 global_alloc(c, v->var->type, v->var, &res);
5561 if (progress == cannot)
5562 type_err(c, "error: const %v cannot be resolved.",
5572 progress = cannot; break;
5574 progress = none; break;
5579 ###### print const decls
5584 for (b = cast(binode, context.constlist); b;
5585 b = cast(binode, b->right)) {
5586 struct binode *vb = cast(binode, b->left);
5587 struct var *vr = cast(var, vb->left);
5588 struct variable *v = vr->var;
5594 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5595 type_print(v->type, stdout);
5597 print_exec(vb->right, -1, 0);
5602 ###### free const decls
5603 free_binode(context.constlist);
5605 ### Function declarations
5607 The code in an Ocean program is all stored in function declarations.
5608 One of the functions must be named `main` and it must accept an array of
5609 strings as a parameter - the command line arguments.
5611 As this is the top level, several things are handled a bit differently.
5612 The function is not interpreted by `interp_exec` as that isn't passed
5613 the argument list which the program requires. Similarly type analysis
5614 is a bit more interesting at this level.
5616 ###### ast functions
5618 static struct type *handle_results(struct parse_context *c,
5619 struct binode *results)
5621 /* Create a 'struct' type from the results list, which
5622 * is a list for 'struct var'
5624 struct type *t = add_anon_type(c, &structure_prototype,
5629 for (b = results; b; b = cast(binode, b->right))
5631 t->structure.nfields = cnt;
5632 t->structure.fields = calloc(cnt, sizeof(struct field));
5634 for (b = results; b; b = cast(binode, b->right)) {
5635 struct var *v = cast(var, b->left);
5636 struct field *f = &t->structure.fields[cnt++];
5637 int a = v->var->type->align;
5638 f->name = v->var->name->name;
5639 f->type = v->var->type;
5641 f->offset = t->size;
5642 v->var->frame_pos = f->offset;
5643 t->size += ((f->type->size - 1) | (a-1)) + 1;
5646 variable_unlink_exec(v->var);
5648 free_binode(results);
5652 static struct variable *declare_function(struct parse_context *c,
5653 struct variable *name,
5654 struct binode *args,
5656 struct binode *results,
5660 struct value fn = {.function = code};
5662 var_block_close(c, CloseFunction, code);
5663 t = add_anon_type(c, &function_prototype,
5664 "func %.*s", name->name->name.len,
5665 name->name->name.txt);
5667 t->function.params = reorder_bilist(args);
5669 ret = handle_results(c, reorder_bilist(results));
5670 t->function.inline_result = 1;
5671 t->function.local_size = ret->size;
5673 t->function.return_type = ret;
5674 global_alloc(c, t, name, &fn);
5675 name->type->function.scope = c->out_scope;
5680 var_block_close(c, CloseFunction, NULL);
5682 c->out_scope = NULL;
5686 ###### declare terminals
5689 ###### top level grammar
5692 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5693 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5695 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5696 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5698 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5699 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5701 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5702 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5704 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5705 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5707 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5708 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5710 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5711 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5713 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5714 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5716 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5717 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5720 ###### print func decls
5725 while (target != 0) {
5727 for (v = context.in_scope; v; v=v->in_scope)
5728 if (v->depth == 0 && v->type && v->type->check_args) {
5737 struct value *val = var_value(&context, v);
5738 printf("func %.*s", v->name->name.len, v->name->name.txt);
5739 v->type->print_type_decl(v->type, stdout);
5741 print_exec(val->function, 0, brackets);
5743 print_value(v->type, val, stdout);
5744 printf("/* frame size %d */\n", v->type->function.local_size);
5750 ###### core functions
5752 static int analyse_funcs(struct parse_context *c)
5756 for (v = c->in_scope; v; v = v->in_scope) {
5760 if (v->depth != 0 || !v->type || !v->type->check_args)
5762 ret = v->type->function.inline_result ?
5763 Tnone : v->type->function.return_type;
5764 val = var_value(c, v);
5767 propagate_types(val->function, c, &perr, ret, 0);
5768 } while (!(perr & Efail) && (perr & Eretry));
5769 if (!(perr & Efail))
5770 /* Make sure everything is still consistent */
5771 propagate_types(val->function, c, &perr, ret, 0);
5774 if (!v->type->function.inline_result &&
5775 !v->type->function.return_type->dup) {
5776 type_err(c, "error: function cannot return value of type %1",
5777 v->where_decl, v->type->function.return_type, 0, NULL);
5780 scope_finalize(c, v->type);
5785 static int analyse_main(struct type *type, struct parse_context *c)
5787 struct binode *bp = type->function.params;
5791 struct type *argv_type;
5793 argv_type = add_anon_type(c, &array_prototype, "argv");
5794 argv_type->array.member = Tstr;
5795 argv_type->array.unspec = 1;
5797 for (b = bp; b; b = cast(binode, b->right)) {
5801 propagate_types(b->left, c, &perr, argv_type, 0);
5803 default: /* invalid */ // NOTEST
5804 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5807 c->parse_error += 1;
5810 return !c->parse_error;
5813 static void interp_main(struct parse_context *c, int argc, char **argv)
5815 struct value *progp = NULL;
5816 struct text main_name = { "main", 4 };
5817 struct variable *mainv;
5823 mainv = var_ref(c, main_name);
5825 progp = var_value(c, mainv);
5826 if (!progp || !progp->function) {
5827 fprintf(stderr, "oceani: no main function found.\n");
5828 c->parse_error += 1;
5831 if (!analyse_main(mainv->type, c)) {
5832 fprintf(stderr, "oceani: main has wrong type.\n");
5833 c->parse_error += 1;
5836 al = mainv->type->function.params;
5838 c->local_size = mainv->type->function.local_size;
5839 c->local = calloc(1, c->local_size);
5841 struct var *v = cast(var, al->left);
5842 struct value *vl = var_value(c, v->var);
5852 mpq_set_ui(argcq, argc, 1);
5853 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5854 t->prepare_type(c, t, 0);
5855 array_init(v->var->type, vl);
5856 for (i = 0; i < argc; i++) {
5857 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5859 arg.str.txt = argv[i];
5860 arg.str.len = strlen(argv[i]);
5861 free_value(Tstr, vl2);
5862 dup_value(Tstr, &arg, vl2);
5866 al = cast(binode, al->right);
5868 v = interp_exec(c, progp->function, &vtype);
5869 free_value(vtype, &v);
5874 ###### ast functions
5875 void free_variable(struct variable *v)
5879 ## And now to test it out.
5881 Having a language requires having a "hello world" program. I'll
5882 provide a little more than that: a program that prints "Hello world"
5883 finds the GCD of two numbers, prints the first few elements of
5884 Fibonacci, performs a binary search for a number, and a few other
5885 things which will likely grow as the languages grows.
5887 ###### File: oceani.mk
5890 @echo "===== DEMO ====="
5891 ./oceani --section "demo: hello" oceani.mdc 55 33
5897 four ::= 2 + 2 ; five ::= 10/2
5898 const pie ::= "I like Pie";
5899 cake ::= "The cake is"
5907 func main(argv:[argc::]string)
5908 print "Hello World, what lovely oceans you have!"
5909 print "Are there", five, "?"
5910 print pi, pie, "but", cake
5912 A := $argv[1]; B := $argv[2]
5914 /* When a variable is defined in both branches of an 'if',
5915 * and used afterwards, the variables are merged.
5921 print "Is", A, "bigger than", B,"? ", bigger
5922 /* If a variable is not used after the 'if', no
5923 * merge happens, so types can be different
5926 double:string = "yes"
5927 print A, "is more than twice", B, "?", double
5930 print "double", B, "is", double
5935 if a > 0 and then b > 0:
5941 print "GCD of", A, "and", B,"is", a
5943 print a, "is not positive, cannot calculate GCD"
5945 print b, "is not positive, cannot calculate GCD"
5950 print "Fibonacci:", f1,f2,
5951 then togo = togo - 1
5959 /* Binary search... */
5964 mid := (lo + hi) / 2
5977 print "Yay, I found", target
5979 print "Closest I found was", lo
5984 // "middle square" PRNG. Not particularly good, but one my
5985 // Dad taught me - the first one I ever heard of.
5986 for i:=1; then i = i + 1; while i < size:
5987 n := list[i-1] * list[i-1]
5988 list[i] = (n / 100) % 10 000
5990 print "Before sort:",
5991 for i:=0; then i = i + 1; while i < size:
5995 for i := 1; then i=i+1; while i < size:
5996 for j:=i-1; then j=j-1; while j >= 0:
5997 if list[j] > list[j+1]:
6001 print " After sort:",
6002 for i:=0; then i = i + 1; while i < size:
6006 if 1 == 2 then print "yes"; else print "no"
6010 bob.alive = (bob.name == "Hello")
6011 print "bob", "is" if bob.alive else "isn't", "alive"