1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
114 struct parse_context {
115 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s = NULL, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
240 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
241 goto cleanup; // NOTEST
244 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
246 resolve_consts(&context);
247 prepare_types(&context);
248 if (!context.parse_error && !analyse_funcs(&context)) {
249 fprintf(stderr, "oceani: type error in program - not running.\n");
250 context.parse_error += 1;
258 if (doexec && !context.parse_error)
259 interp_main(&context, argc - optind, argv + optind);
262 struct section *t = s->next;
267 // FIXME parser should pop scope even on error
268 while (context.scope_depth > 0)
272 ## free context types
273 ## free context storage
274 exit(context.parse_error ? 1 : 0);
279 The four requirements of parse, analyse, print, interpret apply to
280 each language element individually so that is how most of the code
283 Three of the four are fairly self explanatory. The one that requires
284 a little explanation is the analysis step.
286 The current language design does not require the types of variables to
287 be declared, but they must still have a single type. Different
288 operations impose different requirements on the variables, for example
289 addition requires both arguments to be numeric, and assignment
290 requires the variable on the left to have the same type as the
291 expression on the right.
293 Analysis involves propagating these type requirements around and
294 consequently setting the type of each variable. If any requirements
295 are violated (e.g. a string is compared with a number) or if a
296 variable needs to have two different types, then an error is raised
297 and the program will not run.
299 If the same variable is declared in both branchs of an 'if/else', or
300 in all cases of a 'switch' then the multiple instances may be merged
301 into just one variable if the variable is referenced after the
302 conditional statement. When this happens, the types must naturally be
303 consistent across all the branches. When the variable is not used
304 outside the if, the variables in the different branches are distinct
305 and can be of different types.
307 Undeclared names may only appear in "use" statements and "case" expressions.
308 These names are given a type of "label" and a unique value.
309 This allows them to fill the role of a name in an enumerated type, which
310 is useful for testing the `switch` statement.
312 As we will see, the condition part of a `while` statement can return
313 either a Boolean or some other type. This requires that the expected
314 type that gets passed around comprises a type and a flag to indicate
315 that `Tbool` is also permitted.
317 As there are, as yet, no distinct types that are compatible, there
318 isn't much subtlety in the analysis. When we have distinct number
319 types, this will become more interesting.
323 When analysis discovers an inconsistency it needs to report an error;
324 just refusing to run the code ensures that the error doesn't cascade,
325 but by itself it isn't very useful. A clear understanding of the sort
326 of error message that are useful will help guide the process of
329 At a simplistic level, the only sort of error that type analysis can
330 report is that the type of some construct doesn't match a contextual
331 requirement. For example, in `4 + "hello"` the addition provides a
332 contextual requirement for numbers, but `"hello"` is not a number. In
333 this particular example no further information is needed as the types
334 are obvious from local information. When a variable is involved that
335 isn't the case. It may be helpful to explain why the variable has a
336 particular type, by indicating the location where the type was set,
337 whether by declaration or usage.
339 Using a recursive-descent analysis we can easily detect a problem at
340 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
341 will detect that one argument is not a number and the usage of `hello`
342 will detect that a number was wanted, but not provided. In this
343 (early) version of the language, we will generate error reports at
344 multiple locations, so the use of `hello` will report an error and
345 explain were the value was set, and the addition will report an error
346 and say why numbers are needed. To be able to report locations for
347 errors, each language element will need to record a file location
348 (line and column) and each variable will need to record the language
349 element where its type was set. For now we will assume that each line
350 of an error message indicates one location in the file, and up to 2
351 types. So we provide a `printf`-like function which takes a format, a
352 location (a `struct exec` which has not yet been introduced), and 2
353 types. "`%1`" reports the first type, "`%2`" reports the second. We
354 will need a function to print the location, once we know how that is
355 stored. e As will be explained later, there are sometimes extra rules for
356 type matching and they might affect error messages, we need to pass those
359 As well as type errors, we sometimes need to report problems with
360 tokens, which might be unexpected or might name a type that has not
361 been defined. For these we have `tok_err()` which reports an error
362 with a given token. Each of the error functions sets the flag in the
363 context so indicate that parsing failed.
367 static void fput_loc(struct exec *loc, FILE *f);
368 static void type_err(struct parse_context *c,
369 char *fmt, struct exec *loc,
370 struct type *t1, int rules, struct type *t2);
371 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
373 ###### core functions
375 static void type_err(struct parse_context *c,
376 char *fmt, struct exec *loc,
377 struct type *t1, int rules, struct type *t2)
379 fprintf(stderr, "%s:", c->file_name);
380 fput_loc(loc, stderr);
381 for (; *fmt ; fmt++) {
388 case '%': fputc(*fmt, stderr); break; // NOTEST
389 default: fputc('?', stderr); break; // NOTEST
391 type_print(t1, stderr);
394 type_print(t2, stderr);
403 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
405 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
406 t->txt.len, t->txt.txt);
410 ## Entities: declared and predeclared.
412 There are various "things" that the language and/or the interpreter
413 needs to know about to parse and execute a program. These include
414 types, variables, values, and executable code. These are all lumped
415 together under the term "entities" (calling them "objects" would be
416 confusing) and introduced here. The following section will present the
417 different specific code elements which comprise or manipulate these
422 Executables can be lots of different things. In many cases an
423 executable is just an operation combined with one or two other
424 executables. This allows for expressions and lists etc. Other times an
425 executable is something quite specific like a constant or variable name.
426 So we define a `struct exec` to be a general executable with a type, and
427 a `struct binode` which is a subclass of `exec`, forms a node in a
428 binary tree, and holds an operation. There will be other subclasses,
429 and to access these we need to be able to `cast` the `exec` into the
430 various other types. The first field in any `struct exec` is the type
431 from the `exec_types` enum.
434 #define cast(structname, pointer) ({ \
435 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
436 if (__mptr && *__mptr != X##structname) abort(); \
437 (struct structname *)( (char *)__mptr);})
439 #define new(structname) ({ \
440 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
441 __ptr->type = X##structname; \
442 __ptr->line = -1; __ptr->column = -1; \
445 #define new_pos(structname, token) ({ \
446 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
447 __ptr->type = X##structname; \
448 __ptr->line = token.line; __ptr->column = token.col; \
457 enum exec_types type;
466 struct exec *left, *right;
471 static int __fput_loc(struct exec *loc, FILE *f)
475 if (loc->line >= 0) {
476 fprintf(f, "%d:%d: ", loc->line, loc->column);
479 if (loc->type == Xbinode)
480 return __fput_loc(cast(binode,loc)->left, f) ||
481 __fput_loc(cast(binode,loc)->right, f); // NOTEST
484 static void fput_loc(struct exec *loc, FILE *f)
486 if (!__fput_loc(loc, f))
487 fprintf(f, "??:??: "); // NOTEST
490 Each different type of `exec` node needs a number of functions defined,
491 a bit like methods. We must be able to free it, print it, analyse it
492 and execute it. Once we have specific `exec` types we will need to
493 parse them too. Let's take this a bit more slowly.
497 The parser generator requires a `free_foo` function for each struct
498 that stores attributes and they will often be `exec`s and subtypes
499 there-of. So we need `free_exec` which can handle all the subtypes,
500 and we need `free_binode`.
504 static void free_binode(struct binode *b)
513 ###### core functions
514 static void free_exec(struct exec *e)
525 static void free_exec(struct exec *e);
527 ###### free exec cases
528 case Xbinode: free_binode(cast(binode, e)); break;
532 Printing an `exec` requires that we know the current indent level for
533 printing line-oriented components. As will become clear later, we
534 also want to know what sort of bracketing to use.
538 static void do_indent(int i, char *str)
545 ###### core functions
546 static void print_binode(struct binode *b, int indent, int bracket)
550 ## print binode cases
554 static void print_exec(struct exec *e, int indent, int bracket)
560 print_binode(cast(binode, e), indent, bracket); break;
565 do_indent(indent, "/* FREE");
566 for (v = e->to_free; v; v = v->next_free) {
567 printf(" %.*s", v->name->name.len, v->name->name.txt);
568 printf("[%d,%d]", v->scope_start, v->scope_end);
569 if (v->frame_pos >= 0)
570 printf("(%d+%d)", v->frame_pos,
571 v->type ? v->type->size:0);
579 static void print_exec(struct exec *e, int indent, int bracket);
583 As discussed, analysis involves propagating type requirements around the
584 program and looking for errors.
586 So `propagate_types` is passed an expected type (being a `struct type`
587 pointer together with some `val_rules` flags) that the `exec` is
588 expected to return, and returns the type that it does return, either of
589 which can be `NULL` signifying "unknown". A `prop_err` flag set is
590 passed by reference. It has `Efail` set when an error is found, and
591 `Eretry` when the type for some element is set via propagation. If
592 any expression cannot be evaluated 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, int 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, int 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, int 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, int 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);
1061 ###### ast functions
1063 static int type_compat(struct type *require, struct type *have, int rules)
1065 if ((rules & Rboolok) && have == Tbool)
1067 if (!require || !have)
1070 if (require->compat)
1071 return require->compat(require, have);
1073 return require == have;
1078 #include "parse_string.h"
1079 #include "parse_number.h"
1082 myLDLIBS := libnumber.o libstring.o -lgmp
1083 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1085 ###### type union fields
1086 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1088 ###### value union fields
1094 ###### ast functions
1095 static void _free_value(struct type *type, struct value *v)
1099 switch (type->vtype) {
1101 case Vstr: free(v->str.txt); break;
1102 case Vnum: mpq_clear(v->num); break;
1108 ###### value functions
1110 static void _val_init(struct type *type, struct value *val)
1112 switch(type->vtype) {
1113 case Vnone: // NOTEST
1116 mpq_init(val->num); break;
1118 val->str.txt = malloc(1);
1125 val->label = 0; // NOTEST
1130 static void _dup_value(struct type *type,
1131 struct value *vold, struct value *vnew)
1133 switch (type->vtype) {
1134 case Vnone: // NOTEST
1137 vnew->label = vold->label; // NOTEST
1140 vnew->bool = vold->bool;
1143 mpq_init(vnew->num);
1144 mpq_set(vnew->num, vold->num);
1147 vnew->str.len = vold->str.len;
1148 vnew->str.txt = malloc(vnew->str.len);
1149 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1154 static int _value_cmp(struct type *tl, struct type *tr,
1155 struct value *left, struct value *right)
1159 return tl - tr; // NOTEST
1160 switch (tl->vtype) {
1161 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1162 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1163 case Vstr: cmp = text_cmp(left->str, right->str); break;
1164 case Vbool: cmp = left->bool - right->bool; break;
1165 case Vnone: cmp = 0; // NOTEST
1170 static void _print_value(struct type *type, struct value *v, FILE *f)
1172 switch (type->vtype) {
1173 case Vnone: // NOTEST
1174 fprintf(f, "*no-value*"); break; // NOTEST
1175 case Vlabel: // NOTEST
1176 fprintf(f, "*label-%d*", v->label); break; // NOTEST
1178 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1180 fprintf(f, "%s", v->bool ? "True":"False"); break;
1185 mpf_set_q(fl, v->num);
1186 gmp_fprintf(f, "%.10Fg", fl);
1193 static void _free_value(struct type *type, struct value *v);
1195 static int bool_test(struct type *type, struct value *v)
1200 static struct type base_prototype = {
1202 .print = _print_value,
1203 .cmp_order = _value_cmp,
1204 .cmp_eq = _value_cmp,
1206 .free = _free_value,
1209 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1211 ###### ast functions
1212 static struct type *add_base_type(struct parse_context *c, char *n,
1213 enum vtype vt, int size)
1215 struct text txt = { n, strlen(n) };
1218 t = add_type(c, txt, &base_prototype);
1221 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1222 if (t->size & (t->align - 1))
1223 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1227 ###### context initialization
1229 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1230 Tbool->test = bool_test;
1231 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1232 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1233 Tnone = add_base_type(&context, "none", Vnone, 0);
1234 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1238 We have already met values as separate objects. When manifest constants
1239 appear in the program text, that must result in an executable which has
1240 a constant value. So the `val` structure embeds a value in an
1253 ###### ast functions
1254 struct val *new_val(struct type *T, struct token tk)
1256 struct val *v = new_pos(val, tk);
1261 ###### declare terminals
1268 $0 = new_val(Tbool, $1);
1272 $0 = new_val(Tbool, $1);
1277 $0 = new_val(Tnum, $1);
1278 if (number_parse($0->val.num, tail, $1.txt) == 0)
1279 mpq_init($0->val.num); // UNTESTED
1281 tok_err(c, "error: unsupported number suffix",
1286 $0 = new_val(Tstr, $1);
1287 string_parse(&$1, '\\', &$0->val.str, tail);
1289 tok_err(c, "error: unsupported string suffix",
1294 $0 = new_val(Tstr, $1);
1295 string_parse(&$1, '\\', &$0->val.str, tail);
1297 tok_err(c, "error: unsupported string suffix",
1301 ###### print exec cases
1304 struct val *v = cast(val, e);
1305 if (v->vtype == Tstr)
1307 // FIXME how to ensure numbers have same precision.
1308 print_value(v->vtype, &v->val, stdout);
1309 if (v->vtype == Tstr)
1314 ###### propagate exec cases
1317 struct val *val = cast(val, prog);
1318 if (!type_compat(type, val->vtype, rules))
1319 type_err(c, "error: expected %1 found %2",
1320 prog, type, rules, val->vtype);
1325 ###### interp exec cases
1327 rvtype = cast(val, e)->vtype;
1328 dup_value(rvtype, &cast(val, e)->val, &rv);
1331 ###### ast functions
1332 static void free_val(struct val *v)
1335 free_value(v->vtype, &v->val);
1339 ###### free exec cases
1340 case Xval: free_val(cast(val, e)); break;
1342 ###### ast functions
1343 // Move all nodes from 'b' to 'rv', reversing their order.
1344 // In 'b' 'left' is a list, and 'right' is the last node.
1345 // In 'rv', left' is the first node and 'right' is a list.
1346 static struct binode *reorder_bilist(struct binode *b)
1348 struct binode *rv = NULL;
1351 struct exec *t = b->right;
1355 b = cast(binode, b->left);
1365 Labels are a temporary concept until I implement enums. There are an
1366 anonymous enum which is declared by usage. Thet are only allowed in
1367 `use` statements and corresponding `case` entries. They appear as a
1368 period followed by an identifier. All identifiers that are "used" must
1371 For now, we have a global list of labels, and don't check that all "use"
1383 ###### free exec cases
1387 ###### print exec cases
1389 struct label *l = cast(label, e);
1390 printf(".%.*s", l->name.len, l->name.txt);
1396 struct labels *next;
1400 ###### parse context
1401 struct labels *labels;
1403 ###### ast functions
1404 static int label_lookup(struct parse_context *c, struct text name)
1406 struct labels *l, **lp = &c->labels;
1407 while (*lp && text_cmp((*lp)->name, name) < 0)
1409 if (*lp && text_cmp((*lp)->name, name) == 0)
1410 return (*lp)->value;
1411 l = calloc(1, sizeof(*l));
1414 if (c->next_label == 0)
1416 l->value = c->next_label;
1422 ###### free context storage
1423 while (context.labels) {
1424 struct labels *l = context.labels;
1425 context.labels = l->next;
1429 ###### declare terminals
1433 struct label *l = new_pos(label, $ID);
1437 ###### propagate exec cases
1439 struct label *l = cast(label, prog);
1440 l->value = label_lookup(c, l->name);
1441 if (!type_compat(type, Tlabel, rules))
1442 type_err(c, "error: expected %1 found %2",
1443 prog, type, rules, Tlabel);
1447 ###### interp exec cases
1449 struct label *l = cast(label, e);
1450 rv.label = l->value;
1458 Variables are scoped named values. We store the names in a linked list
1459 of "bindings" sorted in lexical order, and use sequential search and
1466 struct binding *next; // in lexical order
1470 This linked list is stored in the parse context so that "reduce"
1471 functions can find or add variables, and so the analysis phase can
1472 ensure that every variable gets a type.
1474 ###### parse context
1476 struct binding *varlist; // In lexical order
1478 ###### ast functions
1480 static struct binding *find_binding(struct parse_context *c, struct text s)
1482 struct binding **l = &c->varlist;
1487 (cmp = text_cmp((*l)->name, s)) < 0)
1491 n = calloc(1, sizeof(*n));
1498 Each name can be linked to multiple variables defined in different
1499 scopes. Each scope starts where the name is declared and continues
1500 until the end of the containing code block. Scopes of a given name
1501 cannot nest, so a declaration while a name is in-scope is an error.
1503 ###### binding fields
1504 struct variable *var;
1508 struct variable *previous;
1510 struct binding *name;
1511 struct exec *where_decl;// where name was declared
1512 struct exec *where_set; // where type was set
1516 When a scope closes, the values of the variables might need to be freed.
1517 This happens in the context of some `struct exec` and each `exec` will
1518 need to know which variables need to be freed when it completes.
1521 struct variable *to_free;
1523 ####### variable fields
1524 struct exec *cleanup_exec;
1525 struct variable *next_free;
1527 ####### interp exec cleanup
1530 for (v = e->to_free; v; v = v->next_free) {
1531 struct value *val = var_value(c, v);
1532 free_value(v->type, val);
1536 ###### ast functions
1537 static void variable_unlink_exec(struct variable *v)
1539 struct variable **vp;
1540 if (!v->cleanup_exec)
1542 for (vp = &v->cleanup_exec->to_free;
1543 *vp; vp = &(*vp)->next_free) {
1547 v->cleanup_exec = NULL;
1552 While the naming seems strange, we include local constants in the
1553 definition of variables. A name declared `var := value` can
1554 subsequently be changed, but a name declared `var ::= value` cannot -
1557 ###### variable fields
1560 Scopes in parallel branches can be partially merged. More
1561 specifically, if a given name is declared in both branches of an
1562 if/else then its scope is a candidate for merging. Similarly if
1563 every branch of an exhaustive switch (e.g. has an "else" clause)
1564 declares a given name, then the scopes from the branches are
1565 candidates for merging.
1567 Note that names declared inside a loop (which is only parallel to
1568 itself) are never visible after the loop. Similarly names defined in
1569 scopes which are not parallel, such as those started by `for` and
1570 `switch`, are never visible after the scope. Only variables defined in
1571 both `then` and `else` (including the implicit then after an `if`, and
1572 excluding `then` used with `for`) and in all `case`s and `else` of a
1573 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1575 Labels, which are a bit like variables, follow different rules.
1576 Labels are not explicitly declared, but if an undeclared name appears
1577 in a context where a label is legal, that effectively declares the
1578 name as a label. The declaration remains in force (or in scope) at
1579 least to the end of the immediately containing block and conditionally
1580 in any larger containing block which does not declare the name in some
1581 other way. Importantly, the conditional scope extension happens even
1582 if the label is only used in one parallel branch of a conditional --
1583 when used in one branch it is treated as having been declared in all
1586 Merge candidates are tentatively visible beyond the end of the
1587 branching statement which creates them. If the name is used, the
1588 merge is affirmed and they become a single variable visible at the
1589 outer layer. If not - if it is redeclared first - the merge lapses.
1591 To track scopes we have an extra stack, implemented as a linked list,
1592 which roughly parallels the parse stack and which is used exclusively
1593 for scoping. When a new scope is opened, a new frame is pushed and
1594 the child-count of the parent frame is incremented. This child-count
1595 is used to distinguish between the first of a set of parallel scopes,
1596 in which declared variables must not be in scope, and subsequent
1597 branches, whether they may already be conditionally scoped.
1599 We need a total ordering of scopes so we can easily compare to variables
1600 to see if they are concurrently in scope. To achieve this we record a
1601 `scope_count` which is actually a count of both beginnings and endings
1602 of scopes. Then each variable has a record of the scope count where it
1603 enters scope, and where it leaves.
1605 To push a new frame *before* any code in the frame is parsed, we need a
1606 grammar reduction. This is most easily achieved with a grammar
1607 element which derives the empty string, and creates the new scope when
1608 it is recognised. This can be placed, for example, between a keyword
1609 like "if" and the code following it.
1613 struct scope *parent;
1617 ###### parse context
1620 struct scope *scope_stack;
1622 ###### variable fields
1623 int scope_start, scope_end;
1625 ###### ast functions
1626 static void scope_pop(struct parse_context *c)
1628 struct scope *s = c->scope_stack;
1630 c->scope_stack = s->parent;
1632 c->scope_depth -= 1;
1633 c->scope_count += 1;
1636 static void scope_push(struct parse_context *c)
1638 struct scope *s = calloc(1, sizeof(*s));
1640 c->scope_stack->child_count += 1;
1641 s->parent = c->scope_stack;
1643 c->scope_depth += 1;
1644 c->scope_count += 1;
1650 OpenScope -> ${ scope_push(c); }$
1652 Each variable records a scope depth and is in one of four states:
1654 - "in scope". This is the case between the declaration of the
1655 variable and the end of the containing block, and also between
1656 the usage with affirms a merge and the end of that block.
1658 The scope depth is not greater than the current parse context scope
1659 nest depth. When the block of that depth closes, the state will
1660 change. To achieve this, all "in scope" variables are linked
1661 together as a stack in nesting order.
1663 - "pending". The "in scope" block has closed, but other parallel
1664 scopes are still being processed. So far, every parallel block at
1665 the same level that has closed has declared the name.
1667 The scope depth is the depth of the last parallel block that
1668 enclosed the declaration, and that has closed.
1670 - "conditionally in scope". The "in scope" block and all parallel
1671 scopes have closed, and no further mention of the name has been seen.
1672 This state includes a secondary nest depth (`min_depth`) which records
1673 the outermost scope seen since the variable became conditionally in
1674 scope. If a use of the name is found, the variable becomes "in scope"
1675 and that secondary depth becomes the recorded scope depth. If the
1676 name is declared as a new variable, the old variable becomes "out of
1677 scope" and the recorded scope depth stays unchanged.
1679 - "out of scope". The variable is neither in scope nor conditionally
1680 in scope. It is permanently out of scope now and can be removed from
1681 the "in scope" stack. When a variable becomes out-of-scope it is
1682 moved to a separate list (`out_scope`) of variables which have fully
1683 known scope. This will be used at the end of each function to assign
1684 each variable a place in the stack frame.
1686 ###### variable fields
1687 int depth, min_depth;
1688 enum { OutScope, PendingScope, CondScope, InScope } scope;
1689 struct variable *in_scope;
1691 ###### parse context
1693 struct variable *in_scope;
1694 struct variable *out_scope;
1696 All variables with the same name are linked together using the
1697 'previous' link. Those variable that have been affirmatively merged all
1698 have a 'merged' pointer that points to one primary variable - the most
1699 recently declared instance. When merging variables, we need to also
1700 adjust the 'merged' pointer on any other variables that had previously
1701 been merged with the one that will no longer be primary.
1703 A variable that is no longer the most recent instance of a name may
1704 still have "pending" scope, if it might still be merged with most
1705 recent instance. These variables don't really belong in the
1706 "in_scope" list, but are not immediately removed when a new instance
1707 is found. Instead, they are detected and ignored when considering the
1708 list of in_scope names.
1710 The storage of the value of a variable will be described later. For now
1711 we just need to know that when a variable goes out of scope, it might
1712 need to be freed. For this we need to be able to find it, so assume that
1713 `var_value()` will provide that.
1715 ###### variable fields
1716 struct variable *merged;
1718 ###### ast functions
1720 static void variable_merge(struct variable *primary, struct variable *secondary)
1724 primary = primary->merged;
1726 for (v = primary->previous; v; v=v->previous)
1727 if (v == secondary || v == secondary->merged ||
1728 v->merged == secondary ||
1729 v->merged == secondary->merged) {
1730 v->scope = OutScope;
1731 v->merged = primary;
1732 if (v->scope_start < primary->scope_start)
1733 primary->scope_start = v->scope_start;
1734 if (v->scope_end > primary->scope_end)
1735 primary->scope_end = v->scope_end; // NOTEST
1736 variable_unlink_exec(v);
1740 ###### forward decls
1741 static struct value *var_value(struct parse_context *c, struct variable *v);
1743 ###### free global vars
1745 while (context.varlist) {
1746 struct binding *b = context.varlist;
1747 struct variable *v = b->var;
1748 context.varlist = b->next;
1751 struct variable *next = v->previous;
1753 if (v->global && v->frame_pos >= 0) {
1754 free_value(v->type, var_value(&context, v));
1755 if (v->depth == 0 && v->type->free == function_free)
1756 // This is a function constant
1757 free_exec(v->where_decl);
1764 #### Manipulating Bindings
1766 When a name is conditionally visible, a new declaration discards the old
1767 binding - the condition lapses. Similarly when we reach the end of a
1768 function (outermost non-global scope) any conditional scope must lapse.
1769 Conversely a usage of the name affirms the visibility and extends it to
1770 the end of the containing block - i.e. the block that contains both the
1771 original declaration and the latest usage. This is determined from
1772 `min_depth`. When a conditionally visible variable gets affirmed like
1773 this, it is also merged with other conditionally visible variables with
1776 When we parse a variable declaration we either report an error if the
1777 name is currently bound, or create a new variable at the current nest
1778 depth if the name is unbound or bound to a conditionally scoped or
1779 pending-scope variable. If the previous variable was conditionally
1780 scoped, it and its homonyms becomes out-of-scope.
1782 When we parse a variable reference (including non-declarative assignment
1783 "foo = bar") we report an error if the name is not bound or is bound to
1784 a pending-scope variable; update the scope if the name is bound to a
1785 conditionally scoped variable; or just proceed normally if the named
1786 variable is in scope.
1788 When we exit a scope, any variables bound at this level are either
1789 marked out of scope or pending-scoped, depending on whether the scope
1790 was sequential or parallel. Here a "parallel" scope means the "then"
1791 or "else" part of a conditional, or any "case" or "else" branch of a
1792 switch. Other scopes are "sequential".
1794 When exiting a parallel scope we check if there are any variables that
1795 were previously pending and are still visible. If there are, then
1796 they weren't redeclared in the most recent scope, so they cannot be
1797 merged and must become out-of-scope. If it is not the first of
1798 parallel scopes (based on `child_count`), we check that there was a
1799 previous binding that is still pending-scope. If there isn't, the new
1800 variable must now be out-of-scope.
1802 When exiting a sequential scope that immediately enclosed parallel
1803 scopes, we need to resolve any pending-scope variables. If there was
1804 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1805 we need to mark all pending-scope variable as out-of-scope. Otherwise
1806 all pending-scope variables become conditionally scoped.
1809 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1811 ###### ast functions
1813 static struct variable *var_decl(struct parse_context *c, struct text s)
1815 struct binding *b = find_binding(c, s);
1816 struct variable *v = b->var;
1818 switch (v ? v->scope : OutScope) {
1820 /* Caller will report the error */
1824 v && v->scope == CondScope;
1826 v->scope = OutScope;
1830 v = calloc(1, sizeof(*v));
1831 v->previous = b->var;
1835 v->min_depth = v->depth = c->scope_depth;
1837 v->in_scope = c->in_scope;
1838 v->scope_start = c->scope_count;
1844 static struct variable *var_ref(struct parse_context *c, struct text s)
1846 struct binding *b = find_binding(c, s);
1847 struct variable *v = b->var;
1848 struct variable *v2;
1850 switch (v ? v->scope : OutScope) {
1853 /* Caller will report the error */
1856 /* All CondScope variables of this name need to be merged
1857 * and become InScope
1859 v->depth = v->min_depth;
1861 for (v2 = v->previous;
1862 v2 && v2->scope == CondScope;
1864 variable_merge(v, v2);
1872 static int var_refile(struct parse_context *c, struct variable *v)
1874 /* Variable just went out of scope. Add it to the out_scope
1875 * list, sorted by ->scope_start
1877 struct variable **vp = &c->out_scope;
1878 while ((*vp) && (*vp)->scope_start < v->scope_start)
1879 vp = &(*vp)->in_scope;
1885 static void var_block_close(struct parse_context *c, enum closetype ct,
1888 /* Close off all variables that are in_scope.
1889 * Some variables in c->scope may already be not-in-scope,
1890 * such as when a PendingScope variable is hidden by a new
1891 * variable with the same name.
1892 * So we check for v->name->var != v and drop them.
1893 * If we choose to make a variable OutScope, we drop it
1896 struct variable *v, **vp, *v2;
1899 for (vp = &c->in_scope;
1900 (v = *vp) && v->min_depth > c->scope_depth;
1901 (v->scope == OutScope || v->name->var != v)
1902 ? (*vp = v->in_scope, var_refile(c, v))
1903 : ( vp = &v->in_scope, 0)) {
1904 v->min_depth = c->scope_depth;
1905 if (v->name->var != v)
1906 /* This is still in scope, but we haven't just
1910 v->min_depth = c->scope_depth;
1911 if (v->scope == InScope)
1912 v->scope_end = c->scope_count;
1913 if (v->scope == InScope && e && !v->global) {
1914 /* This variable gets cleaned up when 'e' finishes */
1915 variable_unlink_exec(v);
1916 v->cleanup_exec = e;
1917 v->next_free = e->to_free;
1922 case CloseParallel: /* handle PendingScope */
1926 if (c->scope_stack->child_count == 1)
1927 /* first among parallel branches */
1928 v->scope = PendingScope;
1929 else if (v->previous &&
1930 v->previous->scope == PendingScope)
1931 /* all previous branches used name */
1932 v->scope = PendingScope;
1934 v->scope = OutScope;
1935 if (ct == CloseElse) {
1936 /* All Pending variables with this name
1937 * are now Conditional */
1939 v2 && v2->scope == PendingScope;
1941 v2->scope = CondScope;
1945 /* Not possible as it would require
1946 * parallel scope to be nested immediately
1947 * in a parallel scope, and that never
1951 /* Not possible as we already tested for
1958 if (v->scope == CondScope)
1959 /* Condition cannot continue past end of function */
1962 case CloseSequential:
1965 v->scope = OutScope;
1968 /* There was no 'else', so we can only become
1969 * conditional if we know the cases were exhaustive,
1970 * and that doesn't mean anything yet.
1971 * So only labels become conditional..
1974 v2 && v2->scope == PendingScope;
1976 v2->scope = OutScope;
1979 case OutScope: break;
1988 The value of a variable is store separately from the variable, on an
1989 analogue of a stack frame. There are (currently) two frames that can be
1990 active. A global frame which currently only stores constants, and a
1991 stacked frame which stores local variables. Each variable knows if it
1992 is global or not, and what its index into the frame is.
1994 Values in the global frame are known immediately they are relevant, so
1995 the frame needs to be reallocated as it grows so it can store those
1996 values. The local frame doesn't get values until the interpreted phase
1997 is started, so there is no need to allocate until the size is known.
1999 We initialize the `frame_pos` to an impossible value, so that we can
2000 tell if it was set or not later.
2002 ###### variable fields
2006 ###### variable init
2009 ###### parse context
2011 short global_size, global_alloc;
2013 void *global, *local;
2015 ###### forward decls
2016 static struct value *global_alloc(struct parse_context *c, struct type *t,
2017 struct variable *v, struct value *init);
2019 ###### ast functions
2021 static struct value *var_value(struct parse_context *c, struct variable *v)
2024 if (!c->local || !v->type)
2025 return NULL; // UNTESTED
2026 if (v->frame_pos + v->type->size > c->local_size) {
2027 printf("INVALID frame_pos\n"); // NOTEST
2030 return c->local + v->frame_pos;
2032 if (c->global_size > c->global_alloc) {
2033 int old = c->global_alloc;
2034 c->global_alloc = (c->global_size | 1023) + 1024;
2035 c->global = realloc(c->global, c->global_alloc);
2036 memset(c->global + old, 0, c->global_alloc - old);
2038 return c->global + v->frame_pos;
2041 static struct value *global_alloc(struct parse_context *c, struct type *t,
2042 struct variable *v, struct value *init)
2045 struct variable scratch;
2047 if (t->prepare_type)
2048 t->prepare_type(c, t, 1); // NOTEST
2050 if (c->global_size & (t->align - 1))
2051 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
2056 v->frame_pos = c->global_size;
2058 c->global_size += v->type->size;
2059 ret = var_value(c, v);
2061 memcpy(ret, init, t->size);
2063 val_init(t, ret); // NOTEST
2067 As global values are found -- struct field initializers, labels etc --
2068 `global_alloc()` is called to record the value in the global frame.
2070 When the program is fully parsed, each function is analysed, we need to
2071 walk the list of variables local to that function and assign them an
2072 offset in the stack frame. For this we have `scope_finalize()`.
2074 We keep the stack from dense by re-using space for between variables
2075 that are not in scope at the same time. The `out_scope` list is sorted
2076 by `scope_start` and as we process a varible, we move it to an FIFO
2077 stack. For each variable we consider, we first discard any from the
2078 stack anything that went out of scope before the new variable came in.
2079 Then we place the new variable just after the one at the top of the
2082 ###### ast functions
2084 static void scope_finalize(struct parse_context *c, struct type *ft)
2086 int size = ft->function.local_size;
2087 struct variable *next = ft->function.scope;
2088 struct variable *done = NULL;
2091 struct variable *v = next;
2092 struct type *t = v->type;
2099 if (v->frame_pos >= 0)
2101 while (done && done->scope_end < v->scope_start)
2102 done = done->in_scope;
2104 pos = done->frame_pos + done->type->size;
2106 pos = ft->function.local_size;
2107 if (pos & (t->align - 1))
2108 pos = (pos + t->align) & ~(t->align-1);
2110 if (size < pos + v->type->size)
2111 size = pos + v->type->size;
2115 c->out_scope = NULL;
2116 ft->function.local_size = size;
2119 ###### free context storage
2120 free(context.global);
2122 #### Variables as executables
2124 Just as we used a `val` to wrap a value into an `exec`, we similarly
2125 need a `var` to wrap a `variable` into an exec. While each `val`
2126 contained a copy of the value, each `var` holds a link to the variable
2127 because it really is the same variable no matter where it appears.
2128 When a variable is used, we need to remember to follow the `->merged`
2129 link to find the primary instance.
2131 When a variable is declared, it may or may not be given an explicit
2132 type. We need to record which so that we can report the parsed code
2141 struct variable *var;
2144 ###### variable fields
2152 VariableDecl -> IDENTIFIER : ${ {
2153 struct variable *v = var_decl(c, $1.txt);
2154 $0 = new_pos(var, $1);
2159 v = var_ref(c, $1.txt);
2161 type_err(c, "error: variable '%v' redeclared",
2163 type_err(c, "info: this is where '%v' was first declared",
2164 v->where_decl, NULL, 0, NULL);
2167 | IDENTIFIER :: ${ {
2168 struct variable *v = var_decl(c, $1.txt);
2169 $0 = new_pos(var, $1);
2175 v = var_ref(c, $1.txt);
2177 type_err(c, "error: variable '%v' redeclared",
2179 type_err(c, "info: this is where '%v' was first declared",
2180 v->where_decl, NULL, 0, NULL);
2183 | IDENTIFIER : Type ${ {
2184 struct variable *v = var_decl(c, $1.txt);
2185 $0 = new_pos(var, $1);
2191 v->explicit_type = 1;
2193 v = var_ref(c, $1.txt);
2195 type_err(c, "error: variable '%v' redeclared",
2197 type_err(c, "info: this is where '%v' was first declared",
2198 v->where_decl, NULL, 0, NULL);
2201 | IDENTIFIER :: Type ${ {
2202 struct variable *v = var_decl(c, $1.txt);
2203 $0 = new_pos(var, $1);
2210 v->explicit_type = 1;
2212 v = var_ref(c, $1.txt);
2214 type_err(c, "error: variable '%v' redeclared",
2216 type_err(c, "info: this is where '%v' was first declared",
2217 v->where_decl, NULL, 0, NULL);
2222 Variable -> IDENTIFIER ${ {
2223 struct variable *v = var_ref(c, $1.txt);
2224 $0 = new_pos(var, $1);
2226 /* This might be a global const or a label
2227 * Allocate a var with impossible type Tnone,
2228 * which will be adjusted when we find out what it is,
2229 * or will trigger an error.
2231 v = var_decl(c, $1.txt);
2238 cast(var, $0)->var = v;
2241 ###### print exec cases
2244 struct var *v = cast(var, e);
2246 struct binding *b = v->var->name;
2247 printf("%.*s", b->name.len, b->name.txt);
2254 if (loc && loc->type == Xvar) {
2255 struct var *v = cast(var, loc);
2257 struct binding *b = v->var->name;
2258 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2260 fputs("???", stderr); // NOTEST
2262 fputs("NOTVAR", stderr); // NOTEST
2265 ###### propagate exec cases
2269 struct var *var = cast(var, prog);
2270 struct variable *v = var->var;
2272 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2273 return Tnone; // NOTEST
2276 if (v->type == Tnone && v->where_decl == prog)
2277 type_err(c, "error: variable used but not declared: %v",
2278 prog, NULL, 0, NULL);
2279 if (v->type == NULL) {
2280 if (type && !(*perr & Efail)) {
2282 v->where_set = prog;
2285 } else if (!type_compat(type, v->type, rules)) {
2286 type_err(c, "error: expected %1 but variable '%v' is %2", prog,
2287 type, rules, v->type);
2288 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2289 v->type, rules, NULL);
2291 if (!v->global || v->frame_pos < 0)
2300 ###### interp exec cases
2303 struct var *var = cast(var, e);
2304 struct variable *v = var->var;
2307 lrv = var_value(c, v);
2312 ###### ast functions
2314 static void free_var(struct var *v)
2319 ###### free exec cases
2320 case Xvar: free_var(cast(var, e)); break;
2325 Now that we have the shape of the interpreter in place we can add some
2326 complex types and connected them in to the data structures and the
2327 different phases of parse, analyse, print, interpret.
2329 Being "complex" the language will naturally have syntax to access
2330 specifics of objects of these types. These will fit into the grammar as
2331 "Terms" which are the things that are combined with various operators to
2332 form "Expression". Where a Term is formed by some operation on another
2333 Term, the subordinate Term will always come first, so for example a
2334 member of an array will be expressed as the Term for the array followed
2335 by an index in square brackets. The strict rule of using postfix
2336 operations makes precedence irrelevant within terms. To provide a place
2337 to put the grammar for each terms of each type, we will start out by
2338 introducing the "Term" grammar production, with contains at least a
2339 simple "Value" (to be explained later).
2343 Term -> Value ${ $0 = $<1; }$
2344 | Variable ${ $0 = $<1; }$
2347 Thus far the complex types we have are arrays and structs.
2351 Arrays can be declared by giving a size and a type, as `[size]type' so
2352 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2353 size can be either a literal number, or a named constant. Some day an
2354 arbitrary expression will be supported.
2356 As a formal parameter to a function, the array can be declared with a
2357 new variable as the size: `name:[size::number]string`. The `size`
2358 variable is set to the size of the array and must be a constant. As
2359 `number` is the only supported type, it can be left out:
2360 `name:[size::]string`.
2362 Arrays cannot be assigned. When pointers are introduced we will also
2363 introduce array slices which can refer to part or all of an array -
2364 the assignment syntax will create a slice. For now, an array can only
2365 ever be referenced by the name it is declared with. It is likely that
2366 a "`copy`" primitive will eventually be define which can be used to
2367 make a copy of an array with controllable recursive depth.
2369 For now we have two sorts of array, those with fixed size either because
2370 it is given as a literal number or because it is a struct member (which
2371 cannot have a runtime-changing size), and those with a size that is
2372 determined at runtime - local variables with a const size. The former
2373 have their size calculated at parse time, the latter at run time.
2375 For the latter type, the `size` field of the type is the size of a
2376 pointer, and the array is reallocated every time it comes into scope.
2378 We differentiate struct fields with a const size from local variables
2379 with a const size by whether they are prepared at parse time or not.
2381 ###### type union fields
2384 int unspec; // size is unspecified - vsize must be set.
2387 struct variable *vsize;
2388 struct type *member;
2391 ###### value union fields
2392 void *array; // used if not static_size
2394 ###### value functions
2396 static int array_prepare_type(struct parse_context *c, struct type *type,
2399 struct value *vsize;
2401 if (type->array.static_size)
2402 return 1; // UNTESTED
2403 if (type->array.unspec && parse_time)
2404 return 1; // UNTESTED
2405 if (parse_time && type->array.vsize && !type->array.vsize->global)
2406 return 1; // UNTESTED
2408 if (type->array.vsize) {
2409 vsize = var_value(c, type->array.vsize);
2411 return 1; // UNTESTED
2413 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2414 type->array.size = mpz_get_si(q);
2419 if (type->array.member->size <= 0)
2420 return 0; // UNTESTED
2422 type->array.static_size = 1;
2423 type->size = type->array.size * type->array.member->size;
2424 type->align = type->array.member->align;
2429 static void array_init(struct type *type, struct value *val)
2432 void *ptr = val->ptr;
2436 if (!type->array.static_size) {
2437 val->array = calloc(type->array.size,
2438 type->array.member->size);
2441 for (i = 0; i < type->array.size; i++) {
2443 v = (void*)ptr + i * type->array.member->size;
2444 val_init(type->array.member, v);
2448 static void array_free(struct type *type, struct value *val)
2451 void *ptr = val->ptr;
2453 if (!type->array.static_size)
2455 for (i = 0; i < type->array.size; i++) {
2457 v = (void*)ptr + i * type->array.member->size;
2458 free_value(type->array.member, v);
2460 if (!type->array.static_size)
2464 static int array_compat(struct type *require, struct type *have)
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)
3123 if (have->compat != require->compat)
3125 if (have->reference.referent != require->reference.referent)
3130 static int reference_test(struct type *type, struct value *val)
3132 return val->ref != NULL;
3135 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3136 struct fieldref *f, struct value **vp)
3138 struct type *rt = t->reference.referent;
3143 return rt->fieldref(rt, c, f, vp);
3145 type_err(c, "error: field reference on %1 is not supported",
3146 f->left, rt, 0, NULL);
3151 static struct type reference_prototype = {
3152 .print_type = reference_print_type,
3153 .cmp_eq = reference_cmp,
3154 .dup = reference_dup,
3155 .test = reference_test,
3156 .free = reference_free,
3157 .compat = reference_compat,
3158 .fieldref = reference_fieldref,
3159 .size = sizeof(void*),
3160 .align = sizeof(void*),
3166 struct type *t = find_type(c, $ID.txt);
3168 t = add_type(c, $ID.txt, NULL);
3171 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3172 $ID.txt.len, $ID.txt.txt);
3173 $0->reference.referent = t;
3176 ###### core functions
3177 static int text_is(struct text t, char *s)
3179 return (strlen(s) == t.len &&
3180 strncmp(s, t.txt, t.len) == 0);
3189 enum ref_func { RefNew, RefFree, RefNil } action;
3190 struct type *reftype;
3194 ###### SimpleStatement Grammar
3196 | @ IDENTIFIER = Expression ${ {
3197 struct ref *r = new_pos(ref, $ID);
3199 if (!text_is($ID.txt, "free"))
3200 tok_err(c, "error: only \"@free\" makes sense here",
3204 r->action = RefFree;
3208 ###### expression grammar
3209 | @ IDENTIFIER ( ) ${
3210 // Only 'new' valid here
3211 if (!text_is($ID.txt, "new")) {
3212 tok_err(c, "error: Only reference function is \"@new()\"",
3215 struct ref *r = new_pos(ref,$ID);
3221 // Only 'nil' valid here
3222 if (!text_is($ID.txt, "nil")) {
3223 tok_err(c, "error: Only reference value is \"@nil\"",
3226 struct ref *r = new_pos(ref,$ID);
3232 ###### print exec cases
3234 struct ref *r = cast(ref, e);
3235 switch (r->action) {
3237 printf("@new()"); break;
3239 printf("@nil"); break;
3241 do_indent(indent, "@free = ");
3242 print_exec(r->right, indent, bracket);
3248 ###### propagate exec cases
3250 struct ref *r = cast(ref, prog);
3251 switch (r->action) {
3253 if (type && type->free != reference_free) {
3254 type_err(c, "error: @new() can only be used with references, not %1",
3255 prog, type, 0, NULL);
3258 if (type && !r->reftype) {
3265 if (type && type->free != reference_free)
3266 type_err(c, "error: @nil can only be used with reference, not %1",
3267 prog, type, 0, NULL);
3268 if (type && !r->reftype) {
3275 t = propagate_types(r->right, c, perr_local, NULL, 0);
3276 if (t && t->free != reference_free)
3277 type_err(c, "error: @free can only be assigned a reference, not %1",
3286 ###### interp exec cases
3288 struct ref *r = cast(ref, e);
3289 switch (r->action) {
3292 rv.ref = calloc(1, r->reftype->reference.referent->size);
3293 rvtype = r->reftype;
3297 rvtype = r->reftype;
3300 rv = interp_exec(c, r->right, &rvtype);
3301 free_value(rvtype->reference.referent, rv.ref);
3309 ###### free exec cases
3311 struct ref *r = cast(ref, e);
3312 free_exec(r->right);
3317 ###### Expressions: dereference
3325 struct binode *b = new(binode);
3331 ###### print binode cases
3333 print_exec(b->left, -1, bracket);
3337 ###### propagate binode cases
3339 /* left must be a reference, and we return what it refers to */
3340 /* FIXME how can I pass the expected type down? */
3341 t = propagate_types(b->left, c, perr, NULL, 0);
3343 if (!t || t->free != reference_free)
3344 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3346 return t->reference.referent;
3349 ###### interp binode cases
3351 left = interp_exec(c, b->left, <ype);
3353 rvtype = ltype->reference.referent;
3360 A function is a chunk of code which can be passed parameters and can
3361 return results. Each function has a type which includes the set of
3362 parameters and the return value. As yet these types cannot be declared
3363 separately from the function itself.
3365 The parameters can be specified either in parentheses as a ';' separated
3368 ##### Example: function 1
3370 func main(av:[ac::number]string; env:[envc::number]string)
3373 or as an indented list of one parameter per line (though each line can
3374 be a ';' separated list)
3376 ##### Example: function 2
3379 argv:[argc::number]string
3380 env:[envc::number]string
3384 In the first case a return type can follow the parentheses after a colon,
3385 in the second it is given on a line starting with the word `return`.
3387 ##### Example: functions that return
3389 func add(a:number; b:number): number
3399 Rather than returning a type, the function can specify a set of local
3400 variables to return as a struct. The values of these variables when the
3401 function exits will be provided to the caller. For this the return type
3402 is replaced with a block of result declarations, either in parentheses
3403 or bracketed by `return` and `do`.
3405 ##### Example: functions returning multiple variables
3407 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3420 For constructing the lists we use a `List` binode, which will be
3421 further detailed when Expression Lists are introduced.
3423 ###### type union fields
3426 struct binode *params;
3427 struct type *return_type;
3428 struct variable *scope;
3429 int inline_result; // return value is at start of 'local'
3433 ###### value union fields
3434 struct exec *function;
3436 ###### type functions
3437 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3438 struct type *require, struct exec *args);
3440 ###### value functions
3442 static void function_free(struct type *type, struct value *val)
3444 free_exec(val->function);
3445 val->function = NULL;
3448 static int function_compat(struct type *require, struct type *have)
3450 // FIXME can I do anything here yet?
3454 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3455 struct type *require, struct exec *args)
3457 /* This should be 'compat', but we don't have a 'tuple' type to
3458 * hold the type of 'args'
3460 struct binode *arg = cast(binode, args);
3461 struct binode *param = require->function.params;
3464 struct var *pv = cast(var, param->left);
3466 type_err(c, "error: insufficient arguments to function.",
3467 args, NULL, 0, NULL);
3471 propagate_types(arg->left, c, perr, pv->var->type, 0);
3472 param = cast(binode, param->right);
3473 arg = cast(binode, arg->right);
3476 type_err(c, "error: too many arguments to function.",
3477 args, NULL, 0, NULL);
3480 static void function_print(struct type *type, struct value *val, FILE *f)
3482 print_exec(val->function, 1, 0);
3485 static void function_print_type_decl(struct type *type, FILE *f)
3489 for (b = type->function.params; b; b = cast(binode, b->right)) {
3490 struct variable *v = cast(var, b->left)->var;
3491 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3492 v->constant ? "::" : ":");
3493 type_print(v->type, f);
3498 if (type->function.return_type != Tnone) {
3500 if (type->function.inline_result) {
3502 struct type *t = type->function.return_type;
3504 for (i = 0; i < t->structure.nfields; i++) {
3505 struct field *fl = t->structure.fields + i;
3508 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3509 type_print(fl->type, f);
3513 type_print(type->function.return_type, f);
3518 static void function_free_type(struct type *t)
3520 free_exec(t->function.params);
3523 static struct type function_prototype = {
3524 .size = sizeof(void*),
3525 .align = sizeof(void*),
3526 .free = function_free,
3527 .compat = function_compat,
3528 .check_args = function_check_args,
3529 .print = function_print,
3530 .print_type_decl = function_print_type_decl,
3531 .free_type = function_free_type,
3534 ###### declare terminals
3544 FuncName -> IDENTIFIER ${ {
3545 struct variable *v = var_decl(c, $1.txt);
3546 struct var *e = new_pos(var, $1);
3553 v = var_ref(c, $1.txt);
3555 type_err(c, "error: function '%v' redeclared",
3557 type_err(c, "info: this is where '%v' was first declared",
3558 v->where_decl, NULL, 0, NULL);
3564 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3565 | Args ArgsLine NEWLINE ${ {
3566 struct binode *b = $<AL;
3567 struct binode **bp = &b;
3569 bp = (struct binode **)&(*bp)->left;
3574 ArgsLine -> ${ $0 = NULL; }$
3575 | Varlist ${ $0 = $<1; }$
3576 | Varlist ; ${ $0 = $<1; }$
3578 Varlist -> Varlist ; ArgDecl ${
3579 $0 = new_pos(binode, $2);
3592 ArgDecl -> IDENTIFIER : FormalType ${ {
3593 struct variable *v = var_decl(c, $ID.txt);
3594 $0 = new_pos(var, $ID);
3601 ##### Function calls
3603 A function call can appear either as an expression or as a statement.
3604 We use a new 'Funcall' binode type to link the function with a list of
3605 arguments, form with the 'List' nodes.
3607 We have already seen the "Term" which is how a function call can appear
3608 in an expression. To parse a function call into a statement we include
3609 it in the "SimpleStatement Grammar" which will be described later.
3615 | Term ( ExpressionList ) ${ {
3616 struct binode *b = new(binode);
3619 b->right = reorder_bilist($<EL);
3623 struct binode *b = new(binode);
3630 ###### SimpleStatement Grammar
3632 | Term ( ExpressionList ) ${ {
3633 struct binode *b = new(binode);
3636 b->right = reorder_bilist($<EL);
3640 ###### print binode cases
3643 do_indent(indent, "");
3644 print_exec(b->left, -1, bracket);
3646 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3649 print_exec(b->left, -1, bracket);
3659 ###### propagate binode cases
3662 /* Every arg must match formal parameter, and result
3663 * is return type of function
3665 struct binode *args = cast(binode, b->right);
3666 struct var *v = cast(var, b->left);
3668 if (!v->var->type || v->var->type->check_args == NULL) {
3669 type_err(c, "error: attempt to call a non-function.",
3670 prog, NULL, 0, NULL);
3674 v->var->type->check_args(c, perr_local, v->var->type, args);
3675 if (v->var->type->function.inline_result)
3678 return v->var->type->function.return_type;
3681 ###### interp binode cases
3684 struct var *v = cast(var, b->left);
3685 struct type *t = v->var->type;
3686 void *oldlocal = c->local;
3687 int old_size = c->local_size;
3688 void *local = calloc(1, t->function.local_size);
3689 struct value *fbody = var_value(c, v->var);
3690 struct binode *arg = cast(binode, b->right);
3691 struct binode *param = t->function.params;
3694 struct var *pv = cast(var, param->left);
3695 struct type *vtype = NULL;
3696 struct value val = interp_exec(c, arg->left, &vtype);
3698 c->local = local; c->local_size = t->function.local_size;
3699 lval = var_value(c, pv->var);
3700 c->local = oldlocal; c->local_size = old_size;
3701 memcpy(lval, &val, vtype->size);
3702 param = cast(binode, param->right);
3703 arg = cast(binode, arg->right);
3705 c->local = local; c->local_size = t->function.local_size;
3706 if (t->function.inline_result && dtype) {
3707 _interp_exec(c, fbody->function, NULL, NULL);
3708 memcpy(dest, local, dtype->size);
3709 rvtype = ret.type = NULL;
3711 rv = interp_exec(c, fbody->function, &rvtype);
3712 c->local = oldlocal; c->local_size = old_size;
3717 ## Complex executables: statements and expressions
3719 Now that we have types and values and variables and most of the basic
3720 Terms which provide access to these, we can explore the more complex
3721 code that combine all of these to get useful work done. Specifically
3722 statements and expressions.
3724 Expressions are various combinations of Terms. We will use operator
3725 precedence to ensure correct parsing. The simplest Expression is just a
3726 Term - others will follow.
3731 Expression -> Term ${ $0 = $<Term; }$
3732 ## expression grammar
3734 ### Expressions: Conditional
3736 Our first user of the `binode` will be conditional expressions, which
3737 is a bit odd as they actually have three components. That will be
3738 handled by having 2 binodes for each expression. The conditional
3739 expression is the lowest precedence operator which is why we define it
3740 first - to start the precedence list.
3742 Conditional expressions are of the form "value `if` condition `else`
3743 other_value". They associate to the right, so everything to the right
3744 of `else` is part of an else value, while only a higher-precedence to
3745 the left of `if` is the if values. Between `if` and `else` there is no
3746 room for ambiguity, so a full conditional expression is allowed in
3752 ###### declare terminals
3756 ###### expression grammar
3758 | Expression if Expression else Expression $$ifelse ${ {
3759 struct binode *b1 = new(binode);
3760 struct binode *b2 = new(binode);
3770 ###### print binode cases
3773 b2 = cast(binode, b->right);
3774 if (bracket) printf("(");
3775 print_exec(b2->left, -1, bracket);
3777 print_exec(b->left, -1, bracket);
3779 print_exec(b2->right, -1, bracket);
3780 if (bracket) printf(")");
3783 ###### propagate binode cases
3786 /* cond must be Tbool, others must match */
3787 struct binode *b2 = cast(binode, b->right);
3790 propagate_types(b->left, c, perr_local, Tbool, 0);
3791 t = propagate_types(b2->left, c, perr, type, 0);
3792 t2 = propagate_types(b2->right, c, perr, type ?: t, 0);
3796 ###### interp binode cases
3799 struct binode *b2 = cast(binode, b->right);
3800 left = interp_exec(c, b->left, <ype);
3802 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3804 rv = interp_exec(c, b2->right, &rvtype);
3810 We take a brief detour, now that we have expressions, to describe lists
3811 of expressions. These will be needed for function parameters and
3812 possibly other situations. They seem generic enough to introduce here
3813 to be used elsewhere.
3815 And ExpressionList will use the `List` type of `binode`, building up at
3816 the end. And place where they are used will probably call
3817 `reorder_bilist()` to get a more normal first/next arrangement.
3819 ###### declare terminals
3822 `List` execs have no implicit semantics, so they are never propagated or
3823 interpreted. The can be printed as a comma separate list, which is how
3824 they are parsed. Note they are also used for function formal parameter
3825 lists. In that case a separate function is used to print them.
3827 ###### print binode cases
3831 print_exec(b->left, -1, bracket);
3834 b = cast(binode, b->right);
3838 ###### propagate binode cases
3839 case List: abort(); // NOTEST
3840 ###### interp binode cases
3841 case List: abort(); // NOTEST
3846 ExpressionList -> ExpressionList , Expression ${
3859 ### Expressions: Boolean
3861 The next class of expressions to use the `binode` will be Boolean
3862 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3863 have same corresponding precendence. The difference is that they don't
3864 evaluate the second expression if not necessary.
3873 ###### declare terminals
3878 ###### expression grammar
3879 | Expression or Expression ${ {
3880 struct binode *b = new(binode);
3886 | Expression or else Expression ${ {
3887 struct binode *b = new(binode);
3894 | Expression and Expression ${ {
3895 struct binode *b = new(binode);
3901 | Expression and then Expression ${ {
3902 struct binode *b = new(binode);
3909 | not Expression ${ {
3910 struct binode *b = new(binode);
3916 ###### print binode cases
3918 if (bracket) printf("(");
3919 print_exec(b->left, -1, bracket);
3921 print_exec(b->right, -1, bracket);
3922 if (bracket) printf(")");
3925 if (bracket) printf("(");
3926 print_exec(b->left, -1, bracket);
3927 printf(" and then ");
3928 print_exec(b->right, -1, bracket);
3929 if (bracket) printf(")");
3932 if (bracket) printf("(");
3933 print_exec(b->left, -1, bracket);
3935 print_exec(b->right, -1, bracket);
3936 if (bracket) printf(")");
3939 if (bracket) printf("(");
3940 print_exec(b->left, -1, bracket);
3941 printf(" or else ");
3942 print_exec(b->right, -1, bracket);
3943 if (bracket) printf(")");
3946 if (bracket) printf("(");
3948 print_exec(b->right, -1, bracket);
3949 if (bracket) printf(")");
3952 ###### propagate binode cases
3958 /* both must be Tbool, result is Tbool */
3959 propagate_types(b->left, c, perr, Tbool, 0);
3960 propagate_types(b->right, c, perr, Tbool, 0);
3961 if (type && type != Tbool)
3962 type_err(c, "error: %1 operation found where %2 expected", prog,
3967 ###### interp binode cases
3969 rv = interp_exec(c, b->left, &rvtype);
3970 right = interp_exec(c, b->right, &rtype);
3971 rv.bool = rv.bool && right.bool;
3974 rv = interp_exec(c, b->left, &rvtype);
3976 rv = interp_exec(c, b->right, NULL);
3979 rv = interp_exec(c, b->left, &rvtype);
3980 right = interp_exec(c, b->right, &rtype);
3981 rv.bool = rv.bool || right.bool;
3984 rv = interp_exec(c, b->left, &rvtype);
3986 rv = interp_exec(c, b->right, NULL);
3989 rv = interp_exec(c, b->right, &rvtype);
3993 ### Expressions: Comparison
3995 Of slightly higher precedence that Boolean expressions are Comparisons.
3996 A comparison takes arguments of any comparable type, but the two types
3999 To simplify the parsing we introduce an `eop` which can record an
4000 expression operator, and the `CMPop` non-terminal will match one of them.
4007 ###### ast functions
4008 static void free_eop(struct eop *e)
4022 ###### declare terminals
4023 $LEFT < > <= >= == != CMPop
4025 ###### expression grammar
4026 | Expression CMPop Expression ${ {
4027 struct binode *b = new(binode);
4037 CMPop -> < ${ $0.op = Less; }$
4038 | > ${ $0.op = Gtr; }$
4039 | <= ${ $0.op = LessEq; }$
4040 | >= ${ $0.op = GtrEq; }$
4041 | == ${ $0.op = Eql; }$
4042 | != ${ $0.op = NEql; }$
4044 ###### print binode cases
4052 if (bracket) printf("(");
4053 print_exec(b->left, -1, bracket);
4055 case Less: printf(" < "); break;
4056 case LessEq: printf(" <= "); break;
4057 case Gtr: printf(" > "); break;
4058 case GtrEq: printf(" >= "); break;
4059 case Eql: printf(" == "); break;
4060 case NEql: printf(" != "); break;
4061 default: abort(); // NOTEST
4063 print_exec(b->right, -1, bracket);
4064 if (bracket) printf(")");
4067 ###### propagate binode cases
4074 /* Both must match but not be labels, result is Tbool */
4075 t = propagate_types(b->left, c, perr, NULL, 0);
4077 propagate_types(b->right, c, perr, t, 0);
4079 t = propagate_types(b->right, c, perr, NULL, 0); // UNTESTED
4081 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
4083 if (!type_compat(type, Tbool, 0))
4084 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4085 Tbool, rules, type);
4089 ###### interp binode cases
4098 left = interp_exec(c, b->left, <ype);
4099 right = interp_exec(c, b->right, &rtype);
4100 cmp = value_cmp(ltype, rtype, &left, &right);
4103 case Less: rv.bool = cmp < 0; break;
4104 case LessEq: rv.bool = cmp <= 0; break;
4105 case Gtr: rv.bool = cmp > 0; break;
4106 case GtrEq: rv.bool = cmp >= 0; break;
4107 case Eql: rv.bool = cmp == 0; break;
4108 case NEql: rv.bool = cmp != 0; break;
4109 default: rv.bool = 0; break; // NOTEST
4114 ### Expressions: Arithmetic etc.
4116 The remaining expressions with the highest precedence are arithmetic,
4117 string concatenation, string conversion, and testing. String concatenation
4118 (`++`) has the same precedence as multiplication and division, but lower
4121 Testing comes in two forms. A single question mark (`?`) is a uniary
4122 operator which converts come types into Boolean. The general meaning is
4123 "is this a value value" and there will be more uses as the language
4124 develops. A double questionmark (`??`) is a binary operator (Choose),
4125 with same precedence as multiplication, which returns the LHS if it
4126 tests successfully, else returns the RHS.
4128 String conversion is a temporary feature until I get a better type
4129 system. `$` is a prefix operator which expects a string and returns
4132 `+` and `-` are both infix and prefix operations (where they are
4133 absolute value and negation). These have different operator names.
4135 We also have a 'Bracket' operator which records where parentheses were
4136 found. This makes it easy to reproduce these when printing. Possibly I
4137 should only insert brackets were needed for precedence. Putting
4138 parentheses around an expression converts it into a Term,
4144 Absolute, Negate, Test,
4148 ###### declare terminals
4150 $LEFT * / % ++ ?? Top
4154 ###### expression grammar
4155 | Expression Eop Expression ${ {
4156 struct binode *b = new(binode);
4163 | Expression Top Expression ${ {
4164 struct binode *b = new(binode);
4171 | Uop Expression ${ {
4172 struct binode *b = new(binode);
4180 | ( Expression ) ${ {
4181 struct binode *b = new_pos(binode, $1);
4190 Eop -> + ${ $0.op = Plus; }$
4191 | - ${ $0.op = Minus; }$
4193 Uop -> + ${ $0.op = Absolute; }$
4194 | - ${ $0.op = Negate; }$
4195 | $ ${ $0.op = StringConv; }$
4196 | ? ${ $0.op = Test; }$
4198 Top -> * ${ $0.op = Times; }$
4199 | / ${ $0.op = Divide; }$
4200 | % ${ $0.op = Rem; }$
4201 | ++ ${ $0.op = Concat; }$
4202 | ?? ${ $0.op = Choose; }$
4204 ###### print binode cases
4212 if (bracket) printf("(");
4213 print_exec(b->left, indent, bracket);
4215 case Plus: fputs(" + ", stdout); break;
4216 case Minus: fputs(" - ", stdout); break;
4217 case Times: fputs(" * ", stdout); break;
4218 case Divide: fputs(" / ", stdout); break;
4219 case Rem: fputs(" % ", stdout); break;
4220 case Concat: fputs(" ++ ", stdout); break;
4221 case Choose: fputs(" ?? ", stdout); break;
4222 default: abort(); // NOTEST
4224 print_exec(b->right, indent, bracket);
4225 if (bracket) printf(")");
4231 if (bracket) printf("(");
4233 case Absolute: fputs("+", stdout); break;
4234 case Negate: fputs("-", stdout); break;
4235 case StringConv: fputs("$", stdout); break;
4236 case Test: fputs("?", stdout); break;
4237 default: abort(); // NOTEST
4239 print_exec(b->right, indent, bracket);
4240 if (bracket) printf(")");
4244 print_exec(b->right, indent, bracket);
4248 ###### propagate binode cases
4254 /* both must be numbers, result is Tnum */
4257 /* as propagate_types ignores a NULL,
4258 * unary ops fit here too */
4259 propagate_types(b->left, c, perr, Tnum, 0);
4260 propagate_types(b->right, c, perr, Tnum, 0);
4261 if (!type_compat(type, Tnum, 0))
4262 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4268 /* both must be Tstr, result is Tstr */
4269 propagate_types(b->left, c, perr, Tstr, 0);
4270 propagate_types(b->right, c, perr, Tstr, 0);
4271 if (!type_compat(type, Tstr, 0))
4272 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4278 /* op must be string, result is number */
4279 propagate_types(b->left, c, perr, Tstr, 0);
4280 if (!type_compat(type, Tnum, 0))
4281 type_err(c, // UNTESTED
4282 "error: Can only convert string to number, not %1",
4283 prog, type, 0, NULL);
4288 /* LHS must support ->test, result is Tbool */
4289 t = propagate_types(b->right, c, perr, NULL, 0);
4291 type_err(c, "error: '?' requires a testable value, not %1",
4297 /* LHS and RHS must match and are returned. Must support
4300 t = propagate_types(b->left, c, perr, type, rules);
4301 t = propagate_types(b->right, c, perr, t, rules);
4302 if (t && t->test == NULL)
4303 type_err(c, "error: \"??\" requires a testable value, not %1",
4309 return propagate_types(b->right, c, perr, type, rules);
4311 ###### interp binode cases
4314 rv = interp_exec(c, b->left, &rvtype);
4315 right = interp_exec(c, b->right, &rtype);
4316 mpq_add(rv.num, rv.num, right.num);
4319 rv = interp_exec(c, b->left, &rvtype);
4320 right = interp_exec(c, b->right, &rtype);
4321 mpq_sub(rv.num, rv.num, right.num);
4324 rv = interp_exec(c, b->left, &rvtype);
4325 right = interp_exec(c, b->right, &rtype);
4326 mpq_mul(rv.num, rv.num, right.num);
4329 rv = interp_exec(c, b->left, &rvtype);
4330 right = interp_exec(c, b->right, &rtype);
4331 mpq_div(rv.num, rv.num, right.num);
4336 left = interp_exec(c, b->left, <ype);
4337 right = interp_exec(c, b->right, &rtype);
4338 mpz_init(l); mpz_init(r); mpz_init(rem);
4339 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4340 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4341 mpz_tdiv_r(rem, l, r);
4342 val_init(Tnum, &rv);
4343 mpq_set_z(rv.num, rem);
4344 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4349 rv = interp_exec(c, b->right, &rvtype);
4350 mpq_neg(rv.num, rv.num);
4353 rv = interp_exec(c, b->right, &rvtype);
4354 mpq_abs(rv.num, rv.num);
4357 rv = interp_exec(c, b->right, &rvtype);
4360 left = interp_exec(c, b->left, <ype);
4361 right = interp_exec(c, b->right, &rtype);
4363 rv.str = text_join(left.str, right.str);
4366 right = interp_exec(c, b->right, &rvtype);
4370 struct text tx = right.str;
4373 if (tx.txt[0] == '-') {
4374 neg = 1; // UNTESTED
4375 tx.txt++; // UNTESTED
4376 tx.len--; // UNTESTED
4378 if (number_parse(rv.num, tail, tx) == 0)
4379 mpq_init(rv.num); // UNTESTED
4381 mpq_neg(rv.num, rv.num); // UNTESTED
4383 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4387 right = interp_exec(c, b->right, &rtype);
4389 rv.bool = !!rtype->test(rtype, &right);
4392 left = interp_exec(c, b->left, <ype);
4393 if (ltype->test(ltype, &left)) {
4398 rv = interp_exec(c, b->right, &rvtype);
4401 ###### value functions
4403 static struct text text_join(struct text a, struct text b)
4406 rv.len = a.len + b.len;
4407 rv.txt = malloc(rv.len);
4408 memcpy(rv.txt, a.txt, a.len);
4409 memcpy(rv.txt+a.len, b.txt, b.len);
4413 ### Blocks, Statements, and Statement lists.
4415 Now that we have expressions out of the way we need to turn to
4416 statements. There are simple statements and more complex statements.
4417 Simple statements do not contain (syntactic) newlines, complex statements do.
4419 Statements often come in sequences and we have corresponding simple
4420 statement lists and complex statement lists.
4421 The former comprise only simple statements separated by semicolons.
4422 The later comprise complex statements and simple statement lists. They are
4423 separated by newlines. Thus the semicolon is only used to separate
4424 simple statements on the one line. This may be overly restrictive,
4425 but I'm not sure I ever want a complex statement to share a line with
4428 Note that a simple statement list can still use multiple lines if
4429 subsequent lines are indented, so
4431 ###### Example: wrapped simple statement list
4436 is a single simple statement list. This might allow room for
4437 confusion, so I'm not set on it yet.
4439 A simple statement list needs no extra syntax. A complex statement
4440 list has two syntactic forms. It can be enclosed in braces (much like
4441 C blocks), or it can be introduced by an indent and continue until an
4442 unindented newline (much like Python blocks). With this extra syntax
4443 it is referred to as a block.
4445 Note that a block does not have to include any newlines if it only
4446 contains simple statements. So both of:
4448 if condition: a=b; d=f
4450 if condition { a=b; print f }
4454 In either case the list is constructed from a `binode` list with
4455 `Block` as the operator. When parsing the list it is most convenient
4456 to append to the end, so a list is a list and a statement. When using
4457 the list it is more convenient to consider a list to be a statement
4458 and a list. So we need a function to re-order a list.
4459 `reorder_bilist` serves this purpose.
4461 The only stand-alone statement we introduce at this stage is `pass`
4462 which does nothing and is represented as a `NULL` pointer in a `Block`
4463 list. Other stand-alone statements will follow once the infrastructure
4466 As many statements will use binodes, we declare a binode pointer 'b' in
4467 the common header for all reductions to use.
4469 ###### Parser: reduce
4480 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4481 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4482 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4483 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4484 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4486 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4487 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4488 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4489 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4490 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4492 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4493 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4494 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4496 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4497 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4498 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4499 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4500 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4502 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4504 ComplexStatements -> ComplexStatements ComplexStatement ${
4514 | ComplexStatement ${
4526 ComplexStatement -> SimpleStatements Newlines ${
4527 $0 = reorder_bilist($<SS);
4529 | SimpleStatements ; Newlines ${
4530 $0 = reorder_bilist($<SS);
4532 ## ComplexStatement Grammar
4535 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4541 | SimpleStatement ${
4550 SimpleStatement -> pass ${ $0 = NULL; }$
4551 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4552 ## SimpleStatement Grammar
4554 ###### print binode cases
4558 if (b->left == NULL) // UNTESTED
4559 printf("pass"); // UNTESTED
4561 print_exec(b->left, indent, bracket); // UNTESTED
4562 if (b->right) { // UNTESTED
4563 printf("; "); // UNTESTED
4564 print_exec(b->right, indent, bracket); // UNTESTED
4567 // block, one per line
4568 if (b->left == NULL)
4569 do_indent(indent, "pass\n");
4571 print_exec(b->left, indent, bracket);
4573 print_exec(b->right, indent, bracket);
4577 ###### propagate binode cases
4580 /* If any statement returns something other than Tnone
4581 * or Tbool then all such must return same type.
4582 * As each statement may be Tnone or something else,
4583 * we must always pass NULL (unknown) down, otherwise an incorrect
4584 * error might occur. We never return Tnone unless it is
4589 for (e = b; e; e = cast(binode, e->right)) {
4590 t = propagate_types(e->left, c, perr, NULL, rules);
4591 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4593 if (t == Tnone && e->right)
4594 /* Only the final statement *must* return a value
4602 type_err(c, "error: expected %1, found %2",
4603 e->left, type, rules, t);
4609 ###### interp binode cases
4611 while (rvtype == Tnone &&
4614 rv = interp_exec(c, b->left, &rvtype);
4615 b = cast(binode, b->right);
4619 ### The Print statement
4621 `print` is a simple statement that takes a comma-separated list of
4622 expressions and prints the values separated by spaces and terminated
4623 by a newline. No control of formatting is possible.
4625 `print` uses `ExpressionList` to collect the expressions and stores them
4626 on the left side of a `Print` binode unlessthere is a trailing comma
4627 when the list is stored on the `right` side and no trailing newline is
4633 ##### declare terminals
4636 ###### SimpleStatement Grammar
4638 | print ExpressionList ${
4639 $0 = b = new_pos(binode, $1);
4642 b->left = reorder_bilist($<EL);
4644 | print ExpressionList , ${ {
4645 $0 = b = new_pos(binode, $1);
4647 b->right = reorder_bilist($<EL);
4651 $0 = b = new_pos(binode, $1);
4657 ###### print binode cases
4660 do_indent(indent, "print");
4662 print_exec(b->right, -1, bracket);
4665 print_exec(b->left, -1, bracket);
4670 ###### propagate binode cases
4673 /* don't care but all must be consistent */
4675 b = cast(binode, b->left);
4677 b = cast(binode, b->right);
4679 propagate_types(b->left, c, perr_local, NULL, 0);
4680 b = cast(binode, b->right);
4684 ###### interp binode cases
4688 struct binode *b2 = cast(binode, b->left);
4690 b2 = cast(binode, b->right);
4691 for (; b2; b2 = cast(binode, b2->right)) {
4692 left = interp_exec(c, b2->left, <ype);
4693 print_value(ltype, &left, stdout);
4694 free_value(ltype, &left);
4698 if (b->right == NULL)
4704 ###### Assignment statement
4706 An assignment will assign a value to a variable, providing it hasn't
4707 been declared as a constant. The analysis phase ensures that the type
4708 will be correct so the interpreter just needs to perform the
4709 calculation. There is a form of assignment which declares a new
4710 variable as well as assigning a value. If a name is used before
4711 it is declared, it is assumed to be a global constant which are allowed to
4712 be declared at any time.
4718 ###### declare terminals
4721 ###### SimpleStatement Grammar
4722 | Term = Expression ${
4723 $0 = b= new(binode);
4728 | VariableDecl = Expression ${
4729 $0 = b= new(binode);
4736 if ($1->var->where_set == NULL) {
4738 "Variable declared with no type or value: %v",
4742 $0 = b = new(binode);
4749 ###### print binode cases
4752 do_indent(indent, "");
4753 print_exec(b->left, -1, bracket);
4755 print_exec(b->right, -1, bracket);
4762 struct variable *v = cast(var, b->left)->var;
4763 do_indent(indent, "");
4764 print_exec(b->left, -1, bracket);
4765 if (cast(var, b->left)->var->constant) {
4767 if (v->explicit_type) {
4768 type_print(v->type, stdout);
4773 if (v->explicit_type) {
4774 type_print(v->type, stdout);
4780 print_exec(b->right, -1, bracket);
4787 ###### propagate binode cases
4791 /* Both must match and not be labels,
4792 * Type must support 'dup',
4793 * For Assign, left must not be constant.
4796 *perr &= ~(Erval | Econst);
4797 t = propagate_types(b->left, c, perr, NULL, 0);
4802 if (propagate_types(b->right, c, perr_local, t, 0) != t)
4803 if (b->left->type == Xvar)
4804 type_err(c, "info: variable '%v' was set as %1 here.",
4805 cast(var, b->left)->var->where_set, t, rules, NULL);
4807 t = propagate_types(b->right, c, perr_local, NULL, 0);
4809 propagate_types(b->left, c, perr, t, 0);
4812 type_err(c, "error: cannot assign to an rval", b,
4814 else if (b->op == Assign && (*perr & Econst)) {
4815 type_err(c, "error: Cannot assign to a constant: %v",
4816 b->left, NULL, 0, NULL);
4817 if (b->left->type == Xvar) {
4818 struct var *var = cast(var, b->left);
4819 struct variable *v = var->var;
4820 type_err(c, "info: name was defined as a constant here",
4821 v->where_decl, NULL, 0, NULL);
4824 if (t && t->dup == NULL && !(*perr_local & Emaycopy))
4825 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4830 ###### interp binode cases
4833 lleft = linterp_exec(c, b->left, <ype);
4835 dinterp_exec(c, b->right, lleft, ltype, 1);
4841 struct variable *v = cast(var, b->left)->var;
4844 val = var_value(c, v);
4845 if (v->type->prepare_type)
4846 v->type->prepare_type(c, v->type, 0);
4848 dinterp_exec(c, b->right, val, v->type, 0);
4850 val_init(v->type, val);
4854 ### The `use` statement
4856 The `use` statement is the last "simple" statement. It is needed when a
4857 statement block can return a value. This includes the body of a
4858 function which has a return type, and the "condition" code blocks in
4859 `if`, `while`, and `switch` statements.
4864 ###### declare terminals
4867 ###### SimpleStatement Grammar
4869 $0 = b = new_pos(binode, $1);
4874 ###### print binode cases
4877 do_indent(indent, "use ");
4878 print_exec(b->right, -1, bracket);
4883 ###### propagate binode cases
4886 /* result matches value */
4887 return propagate_types(b->right, c, perr, type, 0);
4889 ###### interp binode cases
4892 rv = interp_exec(c, b->right, &rvtype);
4895 ### The Conditional Statement
4897 This is the biggy and currently the only complex statement. This
4898 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4899 It is comprised of a number of parts, all of which are optional though
4900 set combinations apply. Each part is (usually) a key word (`then` is
4901 sometimes optional) followed by either an expression or a code block,
4902 except the `casepart` which is a "key word and an expression" followed
4903 by a code block. The code-block option is valid for all parts and,
4904 where an expression is also allowed, the code block can use the `use`
4905 statement to report a value. If the code block does not report a value
4906 the effect is similar to reporting `True`.
4908 The `else` and `case` parts, as well as `then` when combined with
4909 `if`, can contain a `use` statement which will apply to some
4910 containing conditional statement. `for` parts, `do` parts and `then`
4911 parts used with `for` can never contain a `use`, except in some
4912 subordinate conditional statement.
4914 If there is a `forpart`, it is executed first, only once.
4915 If there is a `dopart`, then it is executed repeatedly providing
4916 always that the `condpart` or `cond`, if present, does not return a non-True
4917 value. `condpart` can fail to return any value if it simply executes
4918 to completion. This is treated the same as returning `True`.
4920 If there is a `thenpart` it will be executed whenever the `condpart`
4921 or `cond` returns True (or does not return any value), but this will happen
4922 *after* `dopart` (when present).
4924 If `elsepart` is present it will be executed at most once when the
4925 condition returns `False` or some value that isn't `True` and isn't
4926 matched by any `casepart`. If there are any `casepart`s, they will be
4927 executed when the condition returns a matching value.
4929 The particular sorts of values allowed in case parts has not yet been
4930 determined in the language design, so nothing is prohibited.
4932 The various blocks in this complex statement potentially provide scope
4933 for variables as described earlier. Each such block must include the
4934 "OpenScope" nonterminal before parsing the block, and must call
4935 `var_block_close()` when closing the block.
4937 The code following "`if`", "`switch`" and "`for`" does not get its own
4938 scope, but is in a scope covering the whole statement, so names
4939 declared there cannot be redeclared elsewhere. Similarly the
4940 condition following "`while`" is in a scope the covers the body
4941 ("`do`" part) of the loop, and which does not allow conditional scope
4942 extension. Code following "`then`" (both looping and non-looping),
4943 "`else`" and "`case`" each get their own local scope.
4945 The type requirements on the code block in a `whilepart` are quite
4946 unusal. It is allowed to return a value of some identifiable type, in
4947 which case the loop aborts and an appropriate `casepart` is run, or it
4948 can return a Boolean, in which case the loop either continues to the
4949 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4950 This is different both from the `ifpart` code block which is expected to
4951 return a Boolean, or the `switchpart` code block which is expected to
4952 return the same type as the casepart values. The correct analysis of
4953 the type of the `whilepart` code block is the reason for the
4954 `Rboolok` flag which is passed to `propagate_types()`.
4956 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4957 defined. As there are two scopes which cover multiple parts - one for
4958 the whole statement and one for "while" and "do" - and as we will use
4959 the 'struct exec' to track scopes, we actually need two new types of
4960 exec. One is a `binode` for the looping part, the rest is the
4961 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4962 casepart` to track a list of case parts.
4973 struct exec *action;
4974 struct casepart *next;
4976 struct cond_statement {
4978 struct exec *forpart, *condpart, *thenpart, *elsepart;
4979 struct binode *looppart;
4980 struct casepart *casepart;
4983 ###### ast functions
4985 static void free_casepart(struct casepart *cp)
4989 free_exec(cp->value);
4990 free_exec(cp->action);
4997 static void free_cond_statement(struct cond_statement *s)
5001 free_exec(s->forpart);
5002 free_exec(s->condpart);
5003 free_exec(s->looppart);
5004 free_exec(s->thenpart);
5005 free_exec(s->elsepart);
5006 free_casepart(s->casepart);
5010 ###### free exec cases
5011 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
5013 ###### ComplexStatement Grammar
5014 | CondStatement ${ $0 = $<1; }$
5016 ###### declare terminals
5017 $TERM for then while do
5024 // A CondStatement must end with EOL, as does CondSuffix and
5026 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
5027 // may or may not end with EOL
5028 // WhilePart and IfPart include an appropriate Suffix
5030 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
5031 // them. WhilePart opens and closes its own scope.
5032 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
5035 $0->thenpart = $<TP;
5036 $0->looppart = $<WP;
5037 var_block_close(c, CloseSequential, $0);
5039 | ForPart OptNL WhilePart CondSuffix ${
5042 $0->looppart = $<WP;
5043 var_block_close(c, CloseSequential, $0);
5045 | WhilePart CondSuffix ${
5047 $0->looppart = $<WP;
5049 | SwitchPart OptNL CasePart CondSuffix ${
5051 $0->condpart = $<SP;
5052 $CP->next = $0->casepart;
5053 $0->casepart = $<CP;
5054 var_block_close(c, CloseSequential, $0);
5056 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
5058 $0->condpart = $<SP;
5059 $CP->next = $0->casepart;
5060 $0->casepart = $<CP;
5061 var_block_close(c, CloseSequential, $0);
5063 | IfPart IfSuffix ${
5065 $0->condpart = $IP.condpart; $IP.condpart = NULL;
5066 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
5067 // This is where we close an "if" statement
5068 var_block_close(c, CloseSequential, $0);
5071 CondSuffix -> IfSuffix ${
5074 | Newlines CasePart CondSuffix ${
5076 $CP->next = $0->casepart;
5077 $0->casepart = $<CP;
5079 | CasePart CondSuffix ${
5081 $CP->next = $0->casepart;
5082 $0->casepart = $<CP;
5085 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
5086 | Newlines ElsePart ${ $0 = $<EP; }$
5087 | ElsePart ${$0 = $<EP; }$
5089 ElsePart -> else OpenBlock Newlines ${
5090 $0 = new(cond_statement);
5091 $0->elsepart = $<OB;
5092 var_block_close(c, CloseElse, $0->elsepart);
5094 | else OpenScope CondStatement ${
5095 $0 = new(cond_statement);
5096 $0->elsepart = $<CS;
5097 var_block_close(c, CloseElse, $0->elsepart);
5101 CasePart -> case Expression OpenScope ColonBlock ${
5102 $0 = calloc(1,sizeof(struct casepart));
5105 var_block_close(c, CloseParallel, $0->action);
5109 // These scopes are closed in CondStatement
5110 ForPart -> for OpenBlock ${
5114 ThenPart -> then OpenBlock ${
5116 var_block_close(c, CloseSequential, $0);
5120 // This scope is closed in CondStatement
5121 WhilePart -> while UseBlock OptNL do OpenBlock ${
5126 var_block_close(c, CloseSequential, $0->right);
5127 var_block_close(c, CloseSequential, $0);
5129 | while OpenScope Expression OpenScope ColonBlock ${
5134 var_block_close(c, CloseSequential, $0->right);
5135 var_block_close(c, CloseSequential, $0);
5139 IfPart -> if UseBlock OptNL then OpenBlock ${
5142 var_block_close(c, CloseParallel, $0.thenpart);
5144 | if OpenScope Expression OpenScope ColonBlock ${
5147 var_block_close(c, CloseParallel, $0.thenpart);
5149 | if OpenScope Expression OpenScope OptNL then Block ${
5152 var_block_close(c, CloseParallel, $0.thenpart);
5156 // This scope is closed in CondStatement
5157 SwitchPart -> switch OpenScope Expression ${
5160 | switch UseBlock ${
5164 ###### print binode cases
5166 if (b->left && b->left->type == Xbinode &&
5167 cast(binode, b->left)->op == Block) {
5169 do_indent(indent, "while {\n");
5171 do_indent(indent, "while\n");
5172 print_exec(b->left, indent+1, bracket);
5174 do_indent(indent, "} do {\n");
5176 do_indent(indent, "do\n");
5177 print_exec(b->right, indent+1, bracket);
5179 do_indent(indent, "}\n");
5181 do_indent(indent, "while ");
5182 print_exec(b->left, 0, bracket);
5187 print_exec(b->right, indent+1, bracket);
5189 do_indent(indent, "}\n");
5193 ###### print exec cases
5195 case Xcond_statement:
5197 struct cond_statement *cs = cast(cond_statement, e);
5198 struct casepart *cp;
5200 do_indent(indent, "for");
5201 if (bracket) printf(" {\n"); else printf("\n");
5202 print_exec(cs->forpart, indent+1, bracket);
5205 do_indent(indent, "} then {\n");
5207 do_indent(indent, "then\n");
5208 print_exec(cs->thenpart, indent+1, bracket);
5210 if (bracket) do_indent(indent, "}\n");
5213 print_exec(cs->looppart, indent, bracket);
5217 do_indent(indent, "switch");
5219 do_indent(indent, "if");
5220 if (cs->condpart && cs->condpart->type == Xbinode &&
5221 cast(binode, cs->condpart)->op == Block) {
5226 print_exec(cs->condpart, indent+1, bracket);
5228 do_indent(indent, "}\n");
5230 do_indent(indent, "then\n");
5231 print_exec(cs->thenpart, indent+1, bracket);
5235 print_exec(cs->condpart, 0, bracket);
5241 print_exec(cs->thenpart, indent+1, bracket);
5243 do_indent(indent, "}\n");
5248 for (cp = cs->casepart; cp; cp = cp->next) {
5249 do_indent(indent, "case ");
5250 print_exec(cp->value, -1, 0);
5255 print_exec(cp->action, indent+1, bracket);
5257 do_indent(indent, "}\n");
5260 do_indent(indent, "else");
5265 print_exec(cs->elsepart, indent+1, bracket);
5267 do_indent(indent, "}\n");
5272 ###### propagate binode cases
5274 t = propagate_types(b->right, c, perr_local, Tnone, 0);
5275 if (!type_compat(Tnone, t, 0))
5276 *perr |= Efail; // UNTESTED
5277 return propagate_types(b->left, c, perr, type, rules);
5279 ###### propagate exec cases
5280 case Xcond_statement:
5282 // forpart and looppart->right must return Tnone
5283 // thenpart must return Tnone if there is a loopart,
5284 // otherwise it is like elsepart.
5286 // be bool if there is no casepart
5287 // match casepart->values if there is a switchpart
5288 // either be bool or match casepart->value if there
5290 // elsepart and casepart->action must match the return type
5291 // expected of this statement.
5292 struct cond_statement *cs = cast(cond_statement, prog);
5293 struct casepart *cp;
5295 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5296 if (!type_compat(Tnone, t, 0))
5297 *perr |= Efail; // UNTESTED
5300 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5301 if (!type_compat(Tnone, t, 0))
5302 *perr |= Efail; // UNTESTED
5304 if (cs->casepart == NULL) {
5305 propagate_types(cs->condpart, c, perr, Tbool, 0);
5306 propagate_types(cs->looppart, c, perr, Tbool, 0);
5308 /* Condpart must match case values, with bool permitted */
5310 for (cp = cs->casepart;
5311 cp && !t; cp = cp->next)
5312 t = propagate_types(cp->value, c, perr, NULL, 0);
5313 if (!t && cs->condpart)
5314 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5315 if (!t && cs->looppart)
5316 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5317 // Now we have a type (I hope) push it down
5319 for (cp = cs->casepart; cp; cp = cp->next)
5320 propagate_types(cp->value, c, perr, t, 0);
5321 propagate_types(cs->condpart, c, perr, t, Rboolok);
5322 propagate_types(cs->looppart, c, perr, t, Rboolok);
5325 // (if)then, else, and case parts must return expected type.
5326 if (!cs->looppart && !type)
5327 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5329 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5330 for (cp = cs->casepart;
5332 cp = cp->next) // UNTESTED
5333 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5336 propagate_types(cs->thenpart, c, perr, type, rules);
5337 propagate_types(cs->elsepart, c, perr, type, rules);
5338 for (cp = cs->casepart; cp ; cp = cp->next)
5339 propagate_types(cp->action, c, perr, type, rules);
5345 ###### interp binode cases
5347 // This just performs one iterration of the loop
5348 rv = interp_exec(c, b->left, &rvtype);
5349 if (rvtype == Tnone ||
5350 (rvtype == Tbool && rv.bool != 0))
5351 // rvtype is Tnone or Tbool, doesn't need to be freed
5352 interp_exec(c, b->right, NULL);
5355 ###### interp exec cases
5356 case Xcond_statement:
5358 struct value v, cnd;
5359 struct type *vtype, *cndtype;
5360 struct casepart *cp;
5361 struct cond_statement *cs = cast(cond_statement, e);
5364 interp_exec(c, cs->forpart, NULL);
5366 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5367 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5368 interp_exec(c, cs->thenpart, NULL);
5370 cnd = interp_exec(c, cs->condpart, &cndtype);
5371 if ((cndtype == Tnone ||
5372 (cndtype == Tbool && cnd.bool != 0))) {
5373 // cnd is Tnone or Tbool, doesn't need to be freed
5374 rv = interp_exec(c, cs->thenpart, &rvtype);
5375 // skip else (and cases)
5379 for (cp = cs->casepart; cp; cp = cp->next) {
5380 v = interp_exec(c, cp->value, &vtype);
5381 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5382 free_value(vtype, &v);
5383 free_value(cndtype, &cnd);
5384 rv = interp_exec(c, cp->action, &rvtype);
5387 free_value(vtype, &v);
5389 free_value(cndtype, &cnd);
5391 rv = interp_exec(c, cs->elsepart, &rvtype);
5398 ### Top level structure
5400 All the language elements so far can be used in various places. Now
5401 it is time to clarify what those places are.
5403 At the top level of a file there will be a number of declarations.
5404 Many of the things that can be declared haven't been described yet,
5405 such as functions, procedures, imports, and probably more.
5406 For now there are two sorts of things that can appear at the top
5407 level. They are predefined constants, `struct` types, and the `main`
5408 function. While the syntax will allow the `main` function to appear
5409 multiple times, that will trigger an error if it is actually attempted.
5411 The various declarations do not return anything. They store the
5412 various declarations in the parse context.
5414 ###### Parser: grammar
5417 Ocean -> OptNL DeclarationList
5419 ## declare terminals
5427 DeclarationList -> Declaration
5428 | DeclarationList Declaration
5430 Declaration -> ERROR Newlines ${
5431 tok_err(c, // UNTESTED
5432 "error: unhandled parse error", &$1);
5438 ## top level grammar
5442 ### The `const` section
5444 As well as being defined in with the code that uses them, constants can
5445 be declared at the top level. These have full-file scope, so they are
5446 always `InScope`, even before(!) they have been declared. The value of
5447 a top level constant can be given as an expression, and this is
5448 evaluated after parsing and before execution.
5450 A function call can be used to evaluate a constant, but it will not have
5451 access to any program state, once such statement becomes meaningful.
5452 e.g. arguments and filesystem will not be visible.
5454 Constants are defined in a section that starts with the reserved word
5455 `const` and then has a block with a list of assignment statements.
5456 For syntactic consistency, these must use the double-colon syntax to
5457 make it clear that they are constants. Type can also be given: if
5458 not, the type will be determined during analysis, as with other
5461 ###### parse context
5462 struct binode *constlist;
5464 ###### top level grammar
5468 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5469 | const { SimpleConstList } Newlines
5470 | const IN OptNL ConstList OUT Newlines
5471 | const SimpleConstList Newlines
5473 ConstList -> ConstList SimpleConstLine
5476 SimpleConstList -> SimpleConstList ; Const
5480 SimpleConstLine -> SimpleConstList Newlines
5481 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5484 CType -> Type ${ $0 = $<1; }$
5488 Const -> IDENTIFIER :: CType = Expression ${ {
5490 struct binode *bl, *bv;
5491 struct var *var = new_pos(var, $ID);
5493 v = var_decl(c, $ID.txt);
5495 v->where_decl = var;
5501 v = var_ref(c, $1.txt);
5502 if (v->type == Tnone) {
5503 v->where_decl = var;
5509 tok_err(c, "error: name already declared", &$1);
5510 type_err(c, "info: this is where '%v' was first declared",
5511 v->where_decl, NULL, 0, NULL);
5523 bl->left = c->constlist;
5528 ###### core functions
5529 static void resolve_consts(struct parse_context *c)
5533 enum { none, some, cannot } progress = none;
5535 c->constlist = reorder_bilist(c->constlist);
5538 for (b = cast(binode, c->constlist); b;
5539 b = cast(binode, b->right)) {
5541 struct binode *vb = cast(binode, b->left);
5542 struct var *v = cast(var, vb->left);
5543 if (v->var->frame_pos >= 0)
5547 propagate_types(vb->right, c, &perr,
5549 } while (perr & Eretry);
5551 c->parse_error += 1;
5552 else if (!(perr & Eruntime)) {
5554 struct value res = interp_exec(
5555 c, vb->right, &v->var->type);
5556 global_alloc(c, v->var->type, v->var, &res);
5558 if (progress == cannot)
5559 type_err(c, "error: const %v cannot be resolved.",
5569 progress = cannot; break;
5571 progress = none; break;
5576 ###### print const decls
5581 for (b = cast(binode, context.constlist); b;
5582 b = cast(binode, b->right)) {
5583 struct binode *vb = cast(binode, b->left);
5584 struct var *vr = cast(var, vb->left);
5585 struct variable *v = vr->var;
5591 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5592 type_print(v->type, stdout);
5594 print_exec(vb->right, -1, 0);
5599 ###### free const decls
5600 free_binode(context.constlist);
5602 ### Function declarations
5604 The code in an Ocean program is all stored in function declarations.
5605 One of the functions must be named `main` and it must accept an array of
5606 strings as a parameter - the command line arguments.
5608 As this is the top level, several things are handled a bit differently.
5609 The function is not interpreted by `interp_exec` as that isn't passed
5610 the argument list which the program requires. Similarly type analysis
5611 is a bit more interesting at this level.
5613 ###### ast functions
5615 static struct type *handle_results(struct parse_context *c,
5616 struct binode *results)
5618 /* Create a 'struct' type from the results list, which
5619 * is a list for 'struct var'
5621 struct type *t = add_anon_type(c, &structure_prototype,
5626 for (b = results; b; b = cast(binode, b->right))
5628 t->structure.nfields = cnt;
5629 t->structure.fields = calloc(cnt, sizeof(struct field));
5631 for (b = results; b; b = cast(binode, b->right)) {
5632 struct var *v = cast(var, b->left);
5633 struct field *f = &t->structure.fields[cnt++];
5634 int a = v->var->type->align;
5635 f->name = v->var->name->name;
5636 f->type = v->var->type;
5638 f->offset = t->size;
5639 v->var->frame_pos = f->offset;
5640 t->size += ((f->type->size - 1) | (a-1)) + 1;
5643 variable_unlink_exec(v->var);
5645 free_binode(results);
5649 static struct variable *declare_function(struct parse_context *c,
5650 struct variable *name,
5651 struct binode *args,
5653 struct binode *results,
5657 struct value fn = {.function = code};
5659 var_block_close(c, CloseFunction, code);
5660 t = add_anon_type(c, &function_prototype,
5661 "func %.*s", name->name->name.len,
5662 name->name->name.txt);
5664 t->function.params = reorder_bilist(args);
5666 ret = handle_results(c, reorder_bilist(results));
5667 t->function.inline_result = 1;
5668 t->function.local_size = ret->size;
5670 t->function.return_type = ret;
5671 global_alloc(c, t, name, &fn);
5672 name->type->function.scope = c->out_scope;
5677 var_block_close(c, CloseFunction, NULL);
5679 c->out_scope = NULL;
5683 ###### declare terminals
5686 ###### top level grammar
5689 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5690 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5692 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5693 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5695 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5696 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5698 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5699 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5701 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5702 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5704 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5705 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5707 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5708 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5710 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5711 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5713 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5714 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5717 ###### print func decls
5722 while (target != 0) {
5724 for (v = context.in_scope; v; v=v->in_scope)
5725 if (v->depth == 0 && v->type && v->type->check_args) {
5734 struct value *val = var_value(&context, v);
5735 printf("func %.*s", v->name->name.len, v->name->name.txt);
5736 v->type->print_type_decl(v->type, stdout);
5738 print_exec(val->function, 0, brackets);
5740 print_value(v->type, val, stdout);
5741 printf("/* frame size %d */\n", v->type->function.local_size);
5747 ###### core functions
5749 static int analyse_funcs(struct parse_context *c)
5753 for (v = c->in_scope; v; v = v->in_scope) {
5757 if (v->depth != 0 || !v->type || !v->type->check_args)
5759 ret = v->type->function.inline_result ?
5760 Tnone : v->type->function.return_type;
5761 val = var_value(c, v);
5764 propagate_types(val->function, c, &perr, ret, 0);
5765 } while (!(perr & Efail) && (perr & Eretry));
5766 if (!(perr & Efail))
5767 /* Make sure everything is still consistent */
5768 propagate_types(val->function, c, &perr, ret, 0);
5771 if (!v->type->function.inline_result &&
5772 !v->type->function.return_type->dup) {
5773 type_err(c, "error: function cannot return value of type %1",
5774 v->where_decl, v->type->function.return_type, 0, NULL);
5777 scope_finalize(c, v->type);
5782 static int analyse_main(struct type *type, struct parse_context *c)
5784 struct binode *bp = type->function.params;
5788 struct type *argv_type;
5790 argv_type = add_anon_type(c, &array_prototype, "argv");
5791 argv_type->array.member = Tstr;
5792 argv_type->array.unspec = 1;
5794 for (b = bp; b; b = cast(binode, b->right)) {
5798 propagate_types(b->left, c, &perr, argv_type, 0);
5800 default: /* invalid */ // NOTEST
5801 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5804 c->parse_error += 1;
5807 return !c->parse_error;
5810 static void interp_main(struct parse_context *c, int argc, char **argv)
5812 struct value *progp = NULL;
5813 struct text main_name = { "main", 4 };
5814 struct variable *mainv;
5820 mainv = var_ref(c, main_name);
5822 progp = var_value(c, mainv);
5823 if (!progp || !progp->function) {
5824 fprintf(stderr, "oceani: no main function found.\n");
5825 c->parse_error += 1;
5828 if (!analyse_main(mainv->type, c)) {
5829 fprintf(stderr, "oceani: main has wrong type.\n");
5830 c->parse_error += 1;
5833 al = mainv->type->function.params;
5835 c->local_size = mainv->type->function.local_size;
5836 c->local = calloc(1, c->local_size);
5838 struct var *v = cast(var, al->left);
5839 struct value *vl = var_value(c, v->var);
5849 mpq_set_ui(argcq, argc, 1);
5850 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5851 t->prepare_type(c, t, 0);
5852 array_init(v->var->type, vl);
5853 for (i = 0; i < argc; i++) {
5854 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5856 arg.str.txt = argv[i];
5857 arg.str.len = strlen(argv[i]);
5858 free_value(Tstr, vl2);
5859 dup_value(Tstr, &arg, vl2);
5863 al = cast(binode, al->right);
5865 v = interp_exec(c, progp->function, &vtype);
5866 free_value(vtype, &v);
5871 ###### ast functions
5872 void free_variable(struct variable *v)
5876 ## And now to test it out.
5878 Having a language requires having a "hello world" program. I'll
5879 provide a little more than that: a program that prints "Hello world"
5880 finds the GCD of two numbers, prints the first few elements of
5881 Fibonacci, performs a binary search for a number, and a few other
5882 things which will likely grow as the languages grows.
5884 ###### File: oceani.mk
5887 @echo "===== DEMO ====="
5888 ./oceani --section "demo: hello" oceani.mdc 55 33
5894 four ::= 2 + 2 ; five ::= 10/2
5895 const pie ::= "I like Pie";
5896 cake ::= "The cake is"
5904 func main(argv:[argc::]string)
5905 print "Hello World, what lovely oceans you have!"
5906 print "Are there", five, "?"
5907 print pi, pie, "but", cake
5909 A := $argv[1]; B := $argv[2]
5911 /* When a variable is defined in both branches of an 'if',
5912 * and used afterwards, the variables are merged.
5918 print "Is", A, "bigger than", B,"? ", bigger
5919 /* If a variable is not used after the 'if', no
5920 * merge happens, so types can be different
5923 double:string = "yes"
5924 print A, "is more than twice", B, "?", double
5927 print "double", B, "is", double
5932 if a > 0 and then b > 0:
5938 print "GCD of", A, "and", B,"is", a
5940 print a, "is not positive, cannot calculate GCD"
5942 print b, "is not positive, cannot calculate GCD"
5947 print "Fibonacci:", f1,f2,
5948 then togo = togo - 1
5956 /* Binary search... */
5961 mid := (lo + hi) / 2
5974 print "Yay, I found", target
5976 print "Closest I found was", lo
5981 // "middle square" PRNG. Not particularly good, but one my
5982 // Dad taught me - the first one I ever heard of.
5983 for i:=1; then i = i + 1; while i < size:
5984 n := list[i-1] * list[i-1]
5985 list[i] = (n / 100) % 10 000
5987 print "Before sort:",
5988 for i:=0; then i = i + 1; while i < size:
5992 for i := 1; then i=i+1; while i < size:
5993 for j:=i-1; then j=j-1; while j >= 0:
5994 if list[j] > list[j+1]:
5998 print " After sort:",
5999 for i:=0; then i = i + 1; while i < size:
6003 if 1 == 2 then print "yes"; else print "no"
6007 bob.alive = (bob.name == "Hello")
6008 print "bob", "is" if bob.alive else "isn't", "alive"