1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
114 struct parse_context {
115 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s = NULL, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
240 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
241 goto cleanup; // NOTEST
244 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
246 resolve_consts(&context);
247 prepare_types(&context);
248 if (!context.parse_error && !analyse_funcs(&context)) {
249 fprintf(stderr, "oceani: type error in program - not running.\n");
250 context.parse_error += 1;
258 if (doexec && !context.parse_error)
259 interp_main(&context, argc - optind, argv + optind);
262 struct section *t = s->next;
267 // FIXME parser should pop scope even on error
268 while (context.scope_depth > 0)
272 ## free context types
273 ## free context storage
274 exit(context.parse_error ? 1 : 0);
279 The four requirements of parse, analyse, print, interpret apply to
280 each language element individually so that is how most of the code
283 Three of the four are fairly self explanatory. The one that requires
284 a little explanation is the analysis step.
286 The current language design does not require the types of variables to
287 be declared, but they must still have a single type. Different
288 operations impose different requirements on the variables, for example
289 addition requires both arguments to be numeric, and assignment
290 requires the variable on the left to have the same type as the
291 expression on the right.
293 Analysis involves propagating these type requirements around and
294 consequently setting the type of each variable. If any requirements
295 are violated (e.g. a string is compared with a number) or if a
296 variable needs to have two different types, then an error is raised
297 and the program will not run.
299 If the same variable is declared in both branchs of an 'if/else', or
300 in all cases of a 'switch' then the multiple instances may be merged
301 into just one variable if the variable is referenced after the
302 conditional statement. When this happens, the types must naturally be
303 consistent across all the branches. When the variable is not used
304 outside the if, the variables in the different branches are distinct
305 and can be of different types.
307 Undeclared names may only appear in "use" statements and "case" expressions.
308 These names are given a type of "label" and a unique value.
309 This allows them to fill the role of a name in an enumerated type, which
310 is useful for testing the `switch` statement.
312 As we will see, the condition part of a `while` statement can return
313 either a Boolean or some other type. This requires that the expected
314 type that gets passed around comprises a type and a flag to indicate
315 that `Tbool` is also permitted.
317 As there are, as yet, no distinct types that are compatible, there
318 isn't much subtlety in the analysis. When we have distinct number
319 types, this will become more interesting.
323 When analysis discovers an inconsistency it needs to report an error;
324 just refusing to run the code ensures that the error doesn't cascade,
325 but by itself it isn't very useful. A clear understanding of the sort
326 of error message that are useful will help guide the process of
329 At a simplistic level, the only sort of error that type analysis can
330 report is that the type of some construct doesn't match a contextual
331 requirement. For example, in `4 + "hello"` the addition provides a
332 contextual requirement for numbers, but `"hello"` is not a number. In
333 this particular example no further information is needed as the types
334 are obvious from local information. When a variable is involved that
335 isn't the case. It may be helpful to explain why the variable has a
336 particular type, by indicating the location where the type was set,
337 whether by declaration or usage.
339 Using a recursive-descent analysis we can easily detect a problem at
340 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
341 will detect that one argument is not a number and the usage of `hello`
342 will detect that a number was wanted, but not provided. In this
343 (early) version of the language, we will generate error reports at
344 multiple locations, so the use of `hello` will report an error and
345 explain were the value was set, and the addition will report an error
346 and say why numbers are needed. To be able to report locations for
347 errors, each language element will need to record a file location
348 (line and column) and each variable will need to record the language
349 element where its type was set. For now we will assume that each line
350 of an error message indicates one location in the file, and up to 2
351 types. So we provide a `printf`-like function which takes a format, a
352 location (a `struct exec` which has not yet been introduced), and 2
353 types. "`%1`" reports the first type, "`%2`" reports the second. We
354 will need a function to print the location, once we know how that is
355 stored. e As will be explained later, there are sometimes extra rules for
356 type matching and they might affect error messages, we need to pass those
359 As well as type errors, we sometimes need to report problems with
360 tokens, which might be unexpected or might name a type that has not
361 been defined. For these we have `tok_err()` which reports an error
362 with a given token. Each of the error functions sets the flag in the
363 context so indicate that parsing failed.
367 static void fput_loc(struct exec *loc, FILE *f);
368 static void type_err(struct parse_context *c,
369 char *fmt, struct exec *loc,
370 struct type *t1, int rules, struct type *t2);
371 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
373 ###### core functions
375 static void type_err(struct parse_context *c,
376 char *fmt, struct exec *loc,
377 struct type *t1, int rules, struct type *t2)
379 fprintf(stderr, "%s:", c->file_name);
380 fput_loc(loc, stderr);
381 for (; *fmt ; fmt++) {
388 case '%': fputc(*fmt, stderr); break; // NOTEST
389 default: fputc('?', stderr); break; // NOTEST
391 type_print(t1, stderr);
394 type_print(t2, stderr);
403 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
405 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
406 t->txt.len, t->txt.txt);
410 ## Entities: declared and predeclared.
412 There are various "things" that the language and/or the interpreter
413 needs to know about to parse and execute a program. These include
414 types, variables, values, and executable code. These are all lumped
415 together under the term "entities" (calling them "objects" would be
416 confusing) and introduced here. The following section will present the
417 different specific code elements which comprise or manipulate these
422 Executables can be lots of different things. In many cases an
423 executable is just an operation combined with one or two other
424 executables. This allows for expressions and lists etc. Other times an
425 executable is something quite specific like a constant or variable name.
426 So we define a `struct exec` to be a general executable with a type, and
427 a `struct binode` which is a subclass of `exec`, forms a node in a
428 binary tree, and holds an operation. There will be other subclasses,
429 and to access these we need to be able to `cast` the `exec` into the
430 various other types. The first field in any `struct exec` is the type
431 from the `exec_types` enum.
434 #define cast(structname, pointer) ({ \
435 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
436 if (__mptr && *__mptr != X##structname) abort(); \
437 (struct structname *)( (char *)__mptr);})
439 #define new(structname) ({ \
440 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
441 __ptr->type = X##structname; \
442 __ptr->line = -1; __ptr->column = -1; \
445 #define new_pos(structname, token) ({ \
446 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
447 __ptr->type = X##structname; \
448 __ptr->line = token.line; __ptr->column = token.col; \
457 enum exec_types type;
466 struct exec *left, *right;
471 static int __fput_loc(struct exec *loc, FILE *f)
475 if (loc->line >= 0) {
476 fprintf(f, "%d:%d: ", loc->line, loc->column);
479 if (loc->type == Xbinode)
480 return __fput_loc(cast(binode,loc)->left, f) ||
481 __fput_loc(cast(binode,loc)->right, f); // NOTEST
484 static void fput_loc(struct exec *loc, FILE *f)
486 if (!__fput_loc(loc, f))
487 fprintf(f, "??:??: "); // NOTEST
490 Each different type of `exec` node needs a number of functions defined,
491 a bit like methods. We must be able to free it, print it, analyse it
492 and execute it. Once we have specific `exec` types we will need to
493 parse them too. Let's take this a bit more slowly.
497 The parser generator requires a `free_foo` function for each struct
498 that stores attributes and they will often be `exec`s and subtypes
499 there-of. So we need `free_exec` which can handle all the subtypes,
500 and we need `free_binode`.
504 static void free_binode(struct binode *b)
513 ###### core functions
514 static void free_exec(struct exec *e)
525 static void free_exec(struct exec *e);
527 ###### free exec cases
528 case Xbinode: free_binode(cast(binode, e)); break;
532 Printing an `exec` requires that we know the current indent level for
533 printing line-oriented components. As will become clear later, we
534 also want to know what sort of bracketing to use.
538 static void do_indent(int i, char *str)
545 ###### core functions
546 static void print_binode(struct binode *b, int indent, int bracket)
550 ## print binode cases
554 static void print_exec(struct exec *e, int indent, int bracket)
560 print_binode(cast(binode, e), indent, bracket); break;
565 do_indent(indent, "/* FREE");
566 for (v = e->to_free; v; v = v->next_free) {
567 printf(" %.*s", v->name->name.len, v->name->name.txt);
568 printf("[%d,%d]", v->scope_start, v->scope_end);
569 if (v->frame_pos >= 0)
570 printf("(%d+%d)", v->frame_pos,
571 v->type ? v->type->size:0);
579 static void print_exec(struct exec *e, int indent, int bracket);
583 As discussed, analysis involves propagating type requirements around the
584 program and looking for errors.
586 So `propagate_types` is passed an expected type (being a `struct type`
587 pointer together with some `val_rules` flags) that the `exec` is
588 expected to return, and returns the type that it does return, either of
589 which can be `NULL` signifying "unknown". A `prop_err` flag set is
590 passed by reference. It has `Efail` set when an error is found, and
591 `Eretry` when the type for some element is set via propagation. If
592 any expression cannot be evaluated immediately, `Enoconst` is set.
593 If the expression can be copied, `Emaycopy` is set.
595 If it remains unchanged at `0`, then no more propagation is needed.
599 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 1<<2};
600 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Enoconst = 1<<2,
605 if (rules & Rnolabel)
606 fputs(" (labels not permitted)", stderr);
610 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
611 struct type *type, int rules);
612 ###### core functions
614 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
615 struct type *type, int rules)
622 switch (prog->type) {
625 struct binode *b = cast(binode, prog);
627 ## propagate binode cases
631 ## propagate exec cases
636 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
637 struct type *type, int rules)
639 int pre_err = c->parse_error;
640 struct type *ret = __propagate_types(prog, c, perr, type, rules);
642 if (c->parse_error > pre_err)
649 Interpreting an `exec` doesn't require anything but the `exec`. State
650 is stored in variables and each variable will be directly linked from
651 within the `exec` tree. The exception to this is the `main` function
652 which needs to look at command line arguments. This function will be
653 interpreted separately.
655 Each `exec` can return a value combined with a type in `struct lrval`.
656 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
657 the location of a value, which can be updated, in `lval`. Others will
658 set `lval` to NULL indicating that there is a value of appropriate type
662 static struct value interp_exec(struct parse_context *c, struct exec *e,
663 struct type **typeret);
664 ###### core functions
668 struct value rval, *lval;
671 /* If dest is passed, dtype must give the expected type, and
672 * result can go there, in which case type is returned as NULL.
674 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
675 struct value *dest, struct type *dtype);
677 static struct value interp_exec(struct parse_context *c, struct exec *e,
678 struct type **typeret)
680 struct lrval ret = _interp_exec(c, e, NULL, NULL);
682 if (!ret.type) abort();
686 dup_value(ret.type, ret.lval, &ret.rval);
690 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
691 struct type **typeret)
693 struct lrval ret = _interp_exec(c, e, NULL, NULL);
695 if (!ret.type) abort();
699 free_value(ret.type, &ret.rval);
703 /* dinterp_exec is used when the destination type is certain and
704 * the value has a place to go.
706 static void dinterp_exec(struct parse_context *c, struct exec *e,
707 struct value *dest, struct type *dtype,
710 struct lrval ret = _interp_exec(c, e, dest, dtype);
714 free_value(dtype, dest);
716 dup_value(dtype, ret.lval, dest);
718 memcpy(dest, &ret.rval, dtype->size);
721 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
722 struct value *dest, struct type *dtype)
724 /* If the result is copied to dest, ret.type is set to NULL */
726 struct value rv = {}, *lrv = NULL;
729 rvtype = ret.type = Tnone;
739 struct binode *b = cast(binode, e);
740 struct value left, right, *lleft;
741 struct type *ltype, *rtype;
742 ltype = rtype = Tnone;
744 ## interp binode cases
746 free_value(ltype, &left);
747 free_value(rtype, &right);
757 ## interp exec cleanup
763 Values come in a wide range of types, with more likely to be added.
764 Each type needs to be able to print its own values (for convenience at
765 least) as well as to compare two values, at least for equality and
766 possibly for order. For now, values might need to be duplicated and
767 freed, though eventually such manipulations will be better integrated
770 Rather than requiring every numeric type to support all numeric
771 operations (add, multiply, etc), we allow types to be able to present
772 as one of a few standard types: integer, float, and fraction. The
773 existence of these conversion functions eventually enable types to
774 determine if they are compatible with other types, though such types
775 have not yet been implemented.
777 Named type are stored in a simple linked list. Objects of each type are
778 "values" which are often passed around by value.
780 There are both explicitly named types, and anonymous types. Anonymous
781 cannot be accessed by name, but are used internally and have a name
782 which might be reported in error messages.
789 ## value union fields
797 struct token first_use;
800 void (*init)(struct type *type, struct value *val);
801 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
802 void (*print)(struct type *type, struct value *val, FILE *f);
803 void (*print_type)(struct type *type, FILE *f);
804 int (*cmp_order)(struct type *t1, struct type *t2,
805 struct value *v1, struct value *v2);
806 int (*cmp_eq)(struct type *t1, struct type *t2,
807 struct value *v1, struct value *v2);
808 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
809 int (*test)(struct type *type, struct value *val);
810 void (*free)(struct type *type, struct value *val);
811 void (*free_type)(struct type *t);
812 long long (*to_int)(struct value *v);
813 double (*to_float)(struct value *v);
814 int (*to_mpq)(mpq_t *q, struct value *v);
823 struct type *typelist;
830 static struct type *find_type(struct parse_context *c, struct text s)
832 struct type *t = c->typelist;
834 while (t && (t->anon ||
835 text_cmp(t->name, s) != 0))
840 static struct type *_add_type(struct parse_context *c, struct text s,
841 struct type *proto, int anon)
845 n = calloc(1, sizeof(*n));
852 n->next = c->typelist;
857 static struct type *add_type(struct parse_context *c, struct text s,
860 return _add_type(c, s, proto, 0);
863 static struct type *add_anon_type(struct parse_context *c,
864 struct type *proto, char *name, ...)
870 vasprintf(&t.txt, name, ap);
872 t.len = strlen(t.txt);
873 return _add_type(c, t, proto, 1);
876 static struct type *find_anon_type(struct parse_context *c,
877 struct type *proto, char *name, ...)
879 struct type *t = c->typelist;
884 vasprintf(&nm.txt, name, ap);
886 nm.len = strlen(name);
888 while (t && (!t->anon ||
889 text_cmp(t->name, nm) != 0))
895 return _add_type(c, nm, proto, 1);
898 static void free_type(struct type *t)
900 /* The type is always a reference to something in the
901 * context, so we don't need to free anything.
905 static void free_value(struct type *type, struct value *v)
909 memset(v, 0x5a, type->size);
913 static void type_print(struct type *type, FILE *f)
916 fputs("*unknown*type*", f); // NOTEST
917 else if (type->name.len && !type->anon)
918 fprintf(f, "%.*s", type->name.len, type->name.txt);
919 else if (type->print_type)
920 type->print_type(type, f);
921 else if (type->name.len && type->anon)
922 fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
924 fputs("*invalid*type*", f); // NOTEST
927 static void val_init(struct type *type, struct value *val)
929 if (type && type->init)
930 type->init(type, val);
933 static void dup_value(struct type *type,
934 struct value *vold, struct value *vnew)
936 if (type && type->dup)
937 type->dup(type, vold, vnew);
940 static int value_cmp(struct type *tl, struct type *tr,
941 struct value *left, struct value *right)
943 if (tl && tl->cmp_order)
944 return tl->cmp_order(tl, tr, left, right);
945 if (tl && tl->cmp_eq)
946 return tl->cmp_eq(tl, tr, left, right);
950 static void print_value(struct type *type, struct value *v, FILE *f)
952 if (type && type->print)
953 type->print(type, v, f);
955 fprintf(f, "*Unknown*"); // NOTEST
958 static void prepare_types(struct parse_context *c)
962 enum { none, some, cannot } progress = none;
967 for (t = c->typelist; t; t = t->next) {
969 tok_err(c, "error: type used but not declared",
971 if (t->size == 0 && t->prepare_type) {
972 if (t->prepare_type(c, t, 1))
974 else if (progress == cannot)
975 tok_err(c, "error: type has recursive definition",
985 progress = cannot; break;
987 progress = none; break;
994 static void free_value(struct type *type, struct value *v);
995 static int type_compat(struct type *require, struct type *have, int rules);
996 static void type_print(struct type *type, FILE *f);
997 static void val_init(struct type *type, struct value *v);
998 static void dup_value(struct type *type,
999 struct value *vold, struct value *vnew);
1000 static int value_cmp(struct type *tl, struct type *tr,
1001 struct value *left, struct value *right);
1002 static void print_value(struct type *type, struct value *v, FILE *f);
1004 ###### free context types
1006 while (context.typelist) {
1007 struct type *t = context.typelist;
1009 context.typelist = t->next;
1017 Type can be specified for local variables, for fields in a structure,
1018 for formal parameters to functions, and possibly elsewhere. Different
1019 rules may apply in different contexts. As a minimum, a named type may
1020 always be used. Currently the type of a formal parameter can be
1021 different from types in other contexts, so we have a separate grammar
1027 Type -> IDENTIFIER ${
1028 $0 = find_type(c, $ID.txt);
1030 $0 = add_type(c, $ID.txt, NULL);
1031 $0->first_use = $ID;
1036 FormalType -> Type ${ $0 = $<1; }$
1037 ## formal type grammar
1041 Values of the base types can be numbers, which we represent as
1042 multi-precision fractions, strings, Booleans and labels. When
1043 analysing the program we also need to allow for places where no value
1044 is meaningful (type `Tnone`) and where we don't know what type to
1045 expect yet (type is `NULL`).
1047 Values are never shared, they are always copied when used, and freed
1048 when no longer needed.
1050 When propagating type information around the program, we need to
1051 determine if two types are compatible, where type `NULL` is compatible
1052 with anything. There are two special cases with type compatibility,
1053 both related to the Conditional Statement which will be described
1054 later. In some cases a Boolean can be accepted as well as some other
1055 primary type, and in others any type is acceptable except a label (`Vlabel`).
1056 A separate function encoding these cases will simplify some code later.
1058 ###### type functions
1060 int (*compat)(struct type *this, struct type *other);
1062 ###### ast functions
1064 static int type_compat(struct type *require, struct type *have, int rules)
1066 if ((rules & Rboolok) && have == Tbool)
1068 if ((rules & Rnolabel) && have == Tlabel)
1070 if (!require || !have)
1073 if (require->compat)
1074 return require->compat(require, have);
1076 return require == have;
1081 #include "parse_string.h"
1082 #include "parse_number.h"
1085 myLDLIBS := libnumber.o libstring.o -lgmp
1086 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1088 ###### type union fields
1089 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1091 ###### value union fields
1097 ###### ast functions
1098 static void _free_value(struct type *type, struct value *v)
1102 switch (type->vtype) {
1104 case Vstr: free(v->str.txt); break;
1105 case Vnum: mpq_clear(v->num); break;
1111 ###### value functions
1113 static void _val_init(struct type *type, struct value *val)
1115 switch(type->vtype) {
1116 case Vnone: // NOTEST
1119 mpq_init(val->num); break;
1121 val->str.txt = malloc(1);
1128 val->label = 0; // NOTEST
1133 static void _dup_value(struct type *type,
1134 struct value *vold, struct value *vnew)
1136 switch (type->vtype) {
1137 case Vnone: // NOTEST
1140 vnew->label = vold->label; // NOTEST
1143 vnew->bool = vold->bool;
1146 mpq_init(vnew->num);
1147 mpq_set(vnew->num, vold->num);
1150 vnew->str.len = vold->str.len;
1151 vnew->str.txt = malloc(vnew->str.len);
1152 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1157 static int _value_cmp(struct type *tl, struct type *tr,
1158 struct value *left, struct value *right)
1162 return tl - tr; // NOTEST
1163 switch (tl->vtype) {
1164 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1165 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1166 case Vstr: cmp = text_cmp(left->str, right->str); break;
1167 case Vbool: cmp = left->bool - right->bool; break;
1168 case Vnone: cmp = 0; // NOTEST
1173 static void _print_value(struct type *type, struct value *v, FILE *f)
1175 switch (type->vtype) {
1176 case Vnone: // NOTEST
1177 fprintf(f, "*no-value*"); break; // NOTEST
1178 case Vlabel: // NOTEST
1179 fprintf(f, "*label-%d*", v->label); break; // NOTEST
1181 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1183 fprintf(f, "%s", v->bool ? "True":"False"); break;
1188 mpf_set_q(fl, v->num);
1189 gmp_fprintf(f, "%.10Fg", fl);
1196 static void _free_value(struct type *type, struct value *v);
1198 static int bool_test(struct type *type, struct value *v)
1203 static struct type base_prototype = {
1205 .print = _print_value,
1206 .cmp_order = _value_cmp,
1207 .cmp_eq = _value_cmp,
1209 .free = _free_value,
1212 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1214 ###### ast functions
1215 static struct type *add_base_type(struct parse_context *c, char *n,
1216 enum vtype vt, int size)
1218 struct text txt = { n, strlen(n) };
1221 t = add_type(c, txt, &base_prototype);
1224 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1225 if (t->size & (t->align - 1))
1226 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1230 ###### context initialization
1232 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1233 Tbool->test = bool_test;
1234 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1235 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1236 Tnone = add_base_type(&context, "none", Vnone, 0);
1237 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1241 We have already met values as separate objects. When manifest constants
1242 appear in the program text, that must result in an executable which has
1243 a constant value. So the `val` structure embeds a value in an
1256 ###### ast functions
1257 struct val *new_val(struct type *T, struct token tk)
1259 struct val *v = new_pos(val, tk);
1264 ###### declare terminals
1271 $0 = new_val(Tbool, $1);
1275 $0 = new_val(Tbool, $1);
1280 $0 = new_val(Tnum, $1);
1281 if (number_parse($0->val.num, tail, $1.txt) == 0)
1282 mpq_init($0->val.num); // UNTESTED
1284 tok_err(c, "error: unsupported number suffix",
1289 $0 = new_val(Tstr, $1);
1290 string_parse(&$1, '\\', &$0->val.str, tail);
1292 tok_err(c, "error: unsupported string suffix",
1297 $0 = new_val(Tstr, $1);
1298 string_parse(&$1, '\\', &$0->val.str, tail);
1300 tok_err(c, "error: unsupported string suffix",
1304 ###### print exec cases
1307 struct val *v = cast(val, e);
1308 if (v->vtype == Tstr)
1310 // FIXME how to ensure numbers have same precision.
1311 print_value(v->vtype, &v->val, stdout);
1312 if (v->vtype == Tstr)
1317 ###### propagate exec cases
1320 struct val *val = cast(val, prog);
1321 if (!type_compat(type, val->vtype, rules))
1322 type_err(c, "error: expected %1%r found %2",
1323 prog, type, rules, val->vtype);
1327 ###### interp exec cases
1329 rvtype = cast(val, e)->vtype;
1330 dup_value(rvtype, &cast(val, e)->val, &rv);
1333 ###### ast functions
1334 static void free_val(struct val *v)
1337 free_value(v->vtype, &v->val);
1341 ###### free exec cases
1342 case Xval: free_val(cast(val, e)); break;
1344 ###### ast functions
1345 // Move all nodes from 'b' to 'rv', reversing their order.
1346 // In 'b' 'left' is a list, and 'right' is the last node.
1347 // In 'rv', left' is the first node and 'right' is a list.
1348 static struct binode *reorder_bilist(struct binode *b)
1350 struct binode *rv = NULL;
1353 struct exec *t = b->right;
1357 b = cast(binode, b->left);
1367 Labels are a temporary concept until I implement enums. There are an
1368 anonymous enum which is declared by usage. Thet are only allowed in
1369 `use` statements and corresponding `case` entries. They appear as a
1370 period followed by an identifier. All identifiers that are "used" must
1373 For now, we have a global list of labels, and don't check that all "use"
1385 ###### free exec cases
1389 ###### print exec cases
1391 struct label *l = cast(label, e);
1392 printf(".%.*s", l->name.len, l->name.txt);
1398 struct labels *next;
1402 ###### parse context
1403 struct labels *labels;
1405 ###### ast functions
1406 static int label_lookup(struct parse_context *c, struct text name)
1408 struct labels *l, **lp = &c->labels;
1409 while (*lp && text_cmp((*lp)->name, name) < 0)
1411 if (*lp && text_cmp((*lp)->name, name) == 0)
1412 return (*lp)->value;
1413 l = calloc(1, sizeof(*l));
1416 if (c->next_label == 0)
1418 l->value = c->next_label;
1424 ###### free context storage
1425 while (context.labels) {
1426 struct labels *l = context.labels;
1427 context.labels = l->next;
1431 ###### declare terminals
1435 struct label *l = new_pos(label, $ID);
1439 ###### propagate exec cases
1441 struct label *l = cast(label, prog);
1442 l->value = label_lookup(c, l->name);
1443 if (!type_compat(type, Tlabel, rules))
1444 type_err(c, "error: expected %1%r found %2",
1445 prog, type, rules, Tlabel);
1448 ###### interp exec cases
1450 struct label *l = cast(label, e);
1451 rv.label = l->value;
1459 Variables are scoped named values. We store the names in a linked list
1460 of "bindings" sorted in lexical order, and use sequential search and
1467 struct binding *next; // in lexical order
1471 This linked list is stored in the parse context so that "reduce"
1472 functions can find or add variables, and so the analysis phase can
1473 ensure that every variable gets a type.
1475 ###### parse context
1477 struct binding *varlist; // In lexical order
1479 ###### ast functions
1481 static struct binding *find_binding(struct parse_context *c, struct text s)
1483 struct binding **l = &c->varlist;
1488 (cmp = text_cmp((*l)->name, s)) < 0)
1492 n = calloc(1, sizeof(*n));
1499 Each name can be linked to multiple variables defined in different
1500 scopes. Each scope starts where the name is declared and continues
1501 until the end of the containing code block. Scopes of a given name
1502 cannot nest, so a declaration while a name is in-scope is an error.
1504 ###### binding fields
1505 struct variable *var;
1509 struct variable *previous;
1511 struct binding *name;
1512 struct exec *where_decl;// where name was declared
1513 struct exec *where_set; // where type was set
1517 When a scope closes, the values of the variables might need to be freed.
1518 This happens in the context of some `struct exec` and each `exec` will
1519 need to know which variables need to be freed when it completes.
1522 struct variable *to_free;
1524 ####### variable fields
1525 struct exec *cleanup_exec;
1526 struct variable *next_free;
1528 ####### interp exec cleanup
1531 for (v = e->to_free; v; v = v->next_free) {
1532 struct value *val = var_value(c, v);
1533 free_value(v->type, val);
1537 ###### ast functions
1538 static void variable_unlink_exec(struct variable *v)
1540 struct variable **vp;
1541 if (!v->cleanup_exec)
1543 for (vp = &v->cleanup_exec->to_free;
1544 *vp; vp = &(*vp)->next_free) {
1548 v->cleanup_exec = NULL;
1553 While the naming seems strange, we include local constants in the
1554 definition of variables. A name declared `var := value` can
1555 subsequently be changed, but a name declared `var ::= value` cannot -
1558 ###### variable fields
1561 Scopes in parallel branches can be partially merged. More
1562 specifically, if a given name is declared in both branches of an
1563 if/else then its scope is a candidate for merging. Similarly if
1564 every branch of an exhaustive switch (e.g. has an "else" clause)
1565 declares a given name, then the scopes from the branches are
1566 candidates for merging.
1568 Note that names declared inside a loop (which is only parallel to
1569 itself) are never visible after the loop. Similarly names defined in
1570 scopes which are not parallel, such as those started by `for` and
1571 `switch`, are never visible after the scope. Only variables defined in
1572 both `then` and `else` (including the implicit then after an `if`, and
1573 excluding `then` used with `for`) and in all `case`s and `else` of a
1574 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1576 Labels, which are a bit like variables, follow different rules.
1577 Labels are not explicitly declared, but if an undeclared name appears
1578 in a context where a label is legal, that effectively declares the
1579 name as a label. The declaration remains in force (or in scope) at
1580 least to the end of the immediately containing block and conditionally
1581 in any larger containing block which does not declare the name in some
1582 other way. Importantly, the conditional scope extension happens even
1583 if the label is only used in one parallel branch of a conditional --
1584 when used in one branch it is treated as having been declared in all
1587 Merge candidates are tentatively visible beyond the end of the
1588 branching statement which creates them. If the name is used, the
1589 merge is affirmed and they become a single variable visible at the
1590 outer layer. If not - if it is redeclared first - the merge lapses.
1592 To track scopes we have an extra stack, implemented as a linked list,
1593 which roughly parallels the parse stack and which is used exclusively
1594 for scoping. When a new scope is opened, a new frame is pushed and
1595 the child-count of the parent frame is incremented. This child-count
1596 is used to distinguish between the first of a set of parallel scopes,
1597 in which declared variables must not be in scope, and subsequent
1598 branches, whether they may already be conditionally scoped.
1600 We need a total ordering of scopes so we can easily compare to variables
1601 to see if they are concurrently in scope. To achieve this we record a
1602 `scope_count` which is actually a count of both beginnings and endings
1603 of scopes. Then each variable has a record of the scope count where it
1604 enters scope, and where it leaves.
1606 To push a new frame *before* any code in the frame is parsed, we need a
1607 grammar reduction. This is most easily achieved with a grammar
1608 element which derives the empty string, and creates the new scope when
1609 it is recognised. This can be placed, for example, between a keyword
1610 like "if" and the code following it.
1614 struct scope *parent;
1618 ###### parse context
1621 struct scope *scope_stack;
1623 ###### variable fields
1624 int scope_start, scope_end;
1626 ###### ast functions
1627 static void scope_pop(struct parse_context *c)
1629 struct scope *s = c->scope_stack;
1631 c->scope_stack = s->parent;
1633 c->scope_depth -= 1;
1634 c->scope_count += 1;
1637 static void scope_push(struct parse_context *c)
1639 struct scope *s = calloc(1, sizeof(*s));
1641 c->scope_stack->child_count += 1;
1642 s->parent = c->scope_stack;
1644 c->scope_depth += 1;
1645 c->scope_count += 1;
1651 OpenScope -> ${ scope_push(c); }$
1653 Each variable records a scope depth and is in one of four states:
1655 - "in scope". This is the case between the declaration of the
1656 variable and the end of the containing block, and also between
1657 the usage with affirms a merge and the end of that block.
1659 The scope depth is not greater than the current parse context scope
1660 nest depth. When the block of that depth closes, the state will
1661 change. To achieve this, all "in scope" variables are linked
1662 together as a stack in nesting order.
1664 - "pending". The "in scope" block has closed, but other parallel
1665 scopes are still being processed. So far, every parallel block at
1666 the same level that has closed has declared the name.
1668 The scope depth is the depth of the last parallel block that
1669 enclosed the declaration, and that has closed.
1671 - "conditionally in scope". The "in scope" block and all parallel
1672 scopes have closed, and no further mention of the name has been seen.
1673 This state includes a secondary nest depth (`min_depth`) which records
1674 the outermost scope seen since the variable became conditionally in
1675 scope. If a use of the name is found, the variable becomes "in scope"
1676 and that secondary depth becomes the recorded scope depth. If the
1677 name is declared as a new variable, the old variable becomes "out of
1678 scope" and the recorded scope depth stays unchanged.
1680 - "out of scope". The variable is neither in scope nor conditionally
1681 in scope. It is permanently out of scope now and can be removed from
1682 the "in scope" stack. When a variable becomes out-of-scope it is
1683 moved to a separate list (`out_scope`) of variables which have fully
1684 known scope. This will be used at the end of each function to assign
1685 each variable a place in the stack frame.
1687 ###### variable fields
1688 int depth, min_depth;
1689 enum { OutScope, PendingScope, CondScope, InScope } scope;
1690 struct variable *in_scope;
1692 ###### parse context
1694 struct variable *in_scope;
1695 struct variable *out_scope;
1697 All variables with the same name are linked together using the
1698 'previous' link. Those variable that have been affirmatively merged all
1699 have a 'merged' pointer that points to one primary variable - the most
1700 recently declared instance. When merging variables, we need to also
1701 adjust the 'merged' pointer on any other variables that had previously
1702 been merged with the one that will no longer be primary.
1704 A variable that is no longer the most recent instance of a name may
1705 still have "pending" scope, if it might still be merged with most
1706 recent instance. These variables don't really belong in the
1707 "in_scope" list, but are not immediately removed when a new instance
1708 is found. Instead, they are detected and ignored when considering the
1709 list of in_scope names.
1711 The storage of the value of a variable will be described later. For now
1712 we just need to know that when a variable goes out of scope, it might
1713 need to be freed. For this we need to be able to find it, so assume that
1714 `var_value()` will provide that.
1716 ###### variable fields
1717 struct variable *merged;
1719 ###### ast functions
1721 static void variable_merge(struct variable *primary, struct variable *secondary)
1725 primary = primary->merged;
1727 for (v = primary->previous; v; v=v->previous)
1728 if (v == secondary || v == secondary->merged ||
1729 v->merged == secondary ||
1730 v->merged == secondary->merged) {
1731 v->scope = OutScope;
1732 v->merged = primary;
1733 if (v->scope_start < primary->scope_start)
1734 primary->scope_start = v->scope_start;
1735 if (v->scope_end > primary->scope_end)
1736 primary->scope_end = v->scope_end; // NOTEST
1737 variable_unlink_exec(v);
1741 ###### forward decls
1742 static struct value *var_value(struct parse_context *c, struct variable *v);
1744 ###### free global vars
1746 while (context.varlist) {
1747 struct binding *b = context.varlist;
1748 struct variable *v = b->var;
1749 context.varlist = b->next;
1752 struct variable *next = v->previous;
1754 if (v->global && v->frame_pos >= 0) {
1755 free_value(v->type, var_value(&context, v));
1756 if (v->depth == 0 && v->type->free == function_free)
1757 // This is a function constant
1758 free_exec(v->where_decl);
1765 #### Manipulating Bindings
1767 When a name is conditionally visible, a new declaration discards the old
1768 binding - the condition lapses. Similarly when we reach the end of a
1769 function (outermost non-global scope) any conditional scope must lapse.
1770 Conversely a usage of the name affirms the visibility and extends it to
1771 the end of the containing block - i.e. the block that contains both the
1772 original declaration and the latest usage. This is determined from
1773 `min_depth`. When a conditionally visible variable gets affirmed like
1774 this, it is also merged with other conditionally visible variables with
1777 When we parse a variable declaration we either report an error if the
1778 name is currently bound, or create a new variable at the current nest
1779 depth if the name is unbound or bound to a conditionally scoped or
1780 pending-scope variable. If the previous variable was conditionally
1781 scoped, it and its homonyms becomes out-of-scope.
1783 When we parse a variable reference (including non-declarative assignment
1784 "foo = bar") we report an error if the name is not bound or is bound to
1785 a pending-scope variable; update the scope if the name is bound to a
1786 conditionally scoped variable; or just proceed normally if the named
1787 variable is in scope.
1789 When we exit a scope, any variables bound at this level are either
1790 marked out of scope or pending-scoped, depending on whether the scope
1791 was sequential or parallel. Here a "parallel" scope means the "then"
1792 or "else" part of a conditional, or any "case" or "else" branch of a
1793 switch. Other scopes are "sequential".
1795 When exiting a parallel scope we check if there are any variables that
1796 were previously pending and are still visible. If there are, then
1797 they weren't redeclared in the most recent scope, so they cannot be
1798 merged and must become out-of-scope. If it is not the first of
1799 parallel scopes (based on `child_count`), we check that there was a
1800 previous binding that is still pending-scope. If there isn't, the new
1801 variable must now be out-of-scope.
1803 When exiting a sequential scope that immediately enclosed parallel
1804 scopes, we need to resolve any pending-scope variables. If there was
1805 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1806 we need to mark all pending-scope variable as out-of-scope. Otherwise
1807 all pending-scope variables become conditionally scoped.
1810 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1812 ###### ast functions
1814 static struct variable *var_decl(struct parse_context *c, struct text s)
1816 struct binding *b = find_binding(c, s);
1817 struct variable *v = b->var;
1819 switch (v ? v->scope : OutScope) {
1821 /* Caller will report the error */
1825 v && v->scope == CondScope;
1827 v->scope = OutScope;
1831 v = calloc(1, sizeof(*v));
1832 v->previous = b->var;
1836 v->min_depth = v->depth = c->scope_depth;
1838 v->in_scope = c->in_scope;
1839 v->scope_start = c->scope_count;
1845 static struct variable *var_ref(struct parse_context *c, struct text s)
1847 struct binding *b = find_binding(c, s);
1848 struct variable *v = b->var;
1849 struct variable *v2;
1851 switch (v ? v->scope : OutScope) {
1854 /* Caller will report the error */
1857 /* All CondScope variables of this name need to be merged
1858 * and become InScope
1860 v->depth = v->min_depth;
1862 for (v2 = v->previous;
1863 v2 && v2->scope == CondScope;
1865 variable_merge(v, v2);
1873 static int var_refile(struct parse_context *c, struct variable *v)
1875 /* Variable just went out of scope. Add it to the out_scope
1876 * list, sorted by ->scope_start
1878 struct variable **vp = &c->out_scope;
1879 while ((*vp) && (*vp)->scope_start < v->scope_start)
1880 vp = &(*vp)->in_scope;
1886 static void var_block_close(struct parse_context *c, enum closetype ct,
1889 /* Close off all variables that are in_scope.
1890 * Some variables in c->scope may already be not-in-scope,
1891 * such as when a PendingScope variable is hidden by a new
1892 * variable with the same name.
1893 * So we check for v->name->var != v and drop them.
1894 * If we choose to make a variable OutScope, we drop it
1897 struct variable *v, **vp, *v2;
1900 for (vp = &c->in_scope;
1901 (v = *vp) && v->min_depth > c->scope_depth;
1902 (v->scope == OutScope || v->name->var != v)
1903 ? (*vp = v->in_scope, var_refile(c, v))
1904 : ( vp = &v->in_scope, 0)) {
1905 v->min_depth = c->scope_depth;
1906 if (v->name->var != v)
1907 /* This is still in scope, but we haven't just
1911 v->min_depth = c->scope_depth;
1912 if (v->scope == InScope)
1913 v->scope_end = c->scope_count;
1914 if (v->scope == InScope && e && !v->global) {
1915 /* This variable gets cleaned up when 'e' finishes */
1916 variable_unlink_exec(v);
1917 v->cleanup_exec = e;
1918 v->next_free = e->to_free;
1923 case CloseParallel: /* handle PendingScope */
1927 if (c->scope_stack->child_count == 1)
1928 /* first among parallel branches */
1929 v->scope = PendingScope;
1930 else if (v->previous &&
1931 v->previous->scope == PendingScope)
1932 /* all previous branches used name */
1933 v->scope = PendingScope;
1935 v->scope = OutScope;
1936 if (ct == CloseElse) {
1937 /* All Pending variables with this name
1938 * are now Conditional */
1940 v2 && v2->scope == PendingScope;
1942 v2->scope = CondScope;
1946 /* Not possible as it would require
1947 * parallel scope to be nested immediately
1948 * in a parallel scope, and that never
1952 /* Not possible as we already tested for
1959 if (v->scope == CondScope)
1960 /* Condition cannot continue past end of function */
1963 case CloseSequential:
1966 v->scope = OutScope;
1969 /* There was no 'else', so we can only become
1970 * conditional if we know the cases were exhaustive,
1971 * and that doesn't mean anything yet.
1972 * So only labels become conditional..
1975 v2 && v2->scope == PendingScope;
1977 v2->scope = OutScope;
1980 case OutScope: break;
1989 The value of a variable is store separately from the variable, on an
1990 analogue of a stack frame. There are (currently) two frames that can be
1991 active. A global frame which currently only stores constants, and a
1992 stacked frame which stores local variables. Each variable knows if it
1993 is global or not, and what its index into the frame is.
1995 Values in the global frame are known immediately they are relevant, so
1996 the frame needs to be reallocated as it grows so it can store those
1997 values. The local frame doesn't get values until the interpreted phase
1998 is started, so there is no need to allocate until the size is known.
2000 We initialize the `frame_pos` to an impossible value, so that we can
2001 tell if it was set or not later.
2003 ###### variable fields
2007 ###### variable init
2010 ###### parse context
2012 short global_size, global_alloc;
2014 void *global, *local;
2016 ###### forward decls
2017 static struct value *global_alloc(struct parse_context *c, struct type *t,
2018 struct variable *v, struct value *init);
2020 ###### ast functions
2022 static struct value *var_value(struct parse_context *c, struct variable *v)
2025 if (!c->local || !v->type)
2026 return NULL; // UNTESTED
2027 if (v->frame_pos + v->type->size > c->local_size) {
2028 printf("INVALID frame_pos\n"); // NOTEST
2031 return c->local + v->frame_pos;
2033 if (c->global_size > c->global_alloc) {
2034 int old = c->global_alloc;
2035 c->global_alloc = (c->global_size | 1023) + 1024;
2036 c->global = realloc(c->global, c->global_alloc);
2037 memset(c->global + old, 0, c->global_alloc - old);
2039 return c->global + v->frame_pos;
2042 static struct value *global_alloc(struct parse_context *c, struct type *t,
2043 struct variable *v, struct value *init)
2046 struct variable scratch;
2048 if (t->prepare_type)
2049 t->prepare_type(c, t, 1); // NOTEST
2051 if (c->global_size & (t->align - 1))
2052 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
2057 v->frame_pos = c->global_size;
2059 c->global_size += v->type->size;
2060 ret = var_value(c, v);
2062 memcpy(ret, init, t->size);
2064 val_init(t, ret); // NOTEST
2068 As global values are found -- struct field initializers, labels etc --
2069 `global_alloc()` is called to record the value in the global frame.
2071 When the program is fully parsed, each function is analysed, we need to
2072 walk the list of variables local to that function and assign them an
2073 offset in the stack frame. For this we have `scope_finalize()`.
2075 We keep the stack from dense by re-using space for between variables
2076 that are not in scope at the same time. The `out_scope` list is sorted
2077 by `scope_start` and as we process a varible, we move it to an FIFO
2078 stack. For each variable we consider, we first discard any from the
2079 stack anything that went out of scope before the new variable came in.
2080 Then we place the new variable just after the one at the top of the
2083 ###### ast functions
2085 static void scope_finalize(struct parse_context *c, struct type *ft)
2087 int size = ft->function.local_size;
2088 struct variable *next = ft->function.scope;
2089 struct variable *done = NULL;
2092 struct variable *v = next;
2093 struct type *t = v->type;
2100 if (v->frame_pos >= 0)
2102 while (done && done->scope_end < v->scope_start)
2103 done = done->in_scope;
2105 pos = done->frame_pos + done->type->size;
2107 pos = ft->function.local_size;
2108 if (pos & (t->align - 1))
2109 pos = (pos + t->align) & ~(t->align-1);
2111 if (size < pos + v->type->size)
2112 size = pos + v->type->size;
2116 c->out_scope = NULL;
2117 ft->function.local_size = size;
2120 ###### free context storage
2121 free(context.global);
2123 #### Variables as executables
2125 Just as we used a `val` to wrap a value into an `exec`, we similarly
2126 need a `var` to wrap a `variable` into an exec. While each `val`
2127 contained a copy of the value, each `var` holds a link to the variable
2128 because it really is the same variable no matter where it appears.
2129 When a variable is used, we need to remember to follow the `->merged`
2130 link to find the primary instance.
2132 When a variable is declared, it may or may not be given an explicit
2133 type. We need to record which so that we can report the parsed code
2142 struct variable *var;
2145 ###### variable fields
2153 VariableDecl -> IDENTIFIER : ${ {
2154 struct variable *v = var_decl(c, $1.txt);
2155 $0 = new_pos(var, $1);
2160 v = var_ref(c, $1.txt);
2162 type_err(c, "error: variable '%v' redeclared",
2164 type_err(c, "info: this is where '%v' was first declared",
2165 v->where_decl, NULL, 0, NULL);
2168 | IDENTIFIER :: ${ {
2169 struct variable *v = var_decl(c, $1.txt);
2170 $0 = new_pos(var, $1);
2176 v = var_ref(c, $1.txt);
2178 type_err(c, "error: variable '%v' redeclared",
2180 type_err(c, "info: this is where '%v' was first declared",
2181 v->where_decl, NULL, 0, NULL);
2184 | IDENTIFIER : Type ${ {
2185 struct variable *v = var_decl(c, $1.txt);
2186 $0 = new_pos(var, $1);
2192 v->explicit_type = 1;
2194 v = var_ref(c, $1.txt);
2196 type_err(c, "error: variable '%v' redeclared",
2198 type_err(c, "info: this is where '%v' was first declared",
2199 v->where_decl, NULL, 0, NULL);
2202 | IDENTIFIER :: Type ${ {
2203 struct variable *v = var_decl(c, $1.txt);
2204 $0 = new_pos(var, $1);
2211 v->explicit_type = 1;
2213 v = var_ref(c, $1.txt);
2215 type_err(c, "error: variable '%v' redeclared",
2217 type_err(c, "info: this is where '%v' was first declared",
2218 v->where_decl, NULL, 0, NULL);
2223 Variable -> IDENTIFIER ${ {
2224 struct variable *v = var_ref(c, $1.txt);
2225 $0 = new_pos(var, $1);
2227 /* This might be a global const or a label
2228 * Allocate a var with impossible type Tnone,
2229 * which will be adjusted when we find out what it is,
2230 * or will trigger an error.
2232 v = var_decl(c, $1.txt);
2239 cast(var, $0)->var = v;
2242 ###### print exec cases
2245 struct var *v = cast(var, e);
2247 struct binding *b = v->var->name;
2248 printf("%.*s", b->name.len, b->name.txt);
2255 if (loc && loc->type == Xvar) {
2256 struct var *v = cast(var, loc);
2258 struct binding *b = v->var->name;
2259 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2261 fputs("???", stderr); // NOTEST
2263 fputs("NOTVAR", stderr); // NOTEST
2266 ###### propagate exec cases
2270 struct var *var = cast(var, prog);
2271 struct variable *v = var->var;
2273 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2274 return Tnone; // NOTEST
2277 if (v->constant && (rules & Rnoconstant)) {
2278 type_err(c, "error: Cannot assign to a constant: %v",
2279 prog, NULL, 0, NULL);
2280 type_err(c, "info: name was defined as a constant here",
2281 v->where_decl, NULL, 0, NULL);
2284 if (v->type == Tnone && v->where_decl == prog)
2285 type_err(c, "error: variable used but not declared: %v",
2286 prog, NULL, 0, NULL);
2287 if (v->type == NULL) {
2288 if (type && !(*perr & Efail)) {
2290 v->where_set = prog;
2293 } else if (!type_compat(type, v->type, rules)) {
2294 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2295 type, rules, v->type);
2296 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2297 v->type, rules, NULL);
2299 if (!v->global || v->frame_pos < 0)
2306 ###### interp exec cases
2309 struct var *var = cast(var, e);
2310 struct variable *v = var->var;
2313 lrv = var_value(c, v);
2318 ###### ast functions
2320 static void free_var(struct var *v)
2325 ###### free exec cases
2326 case Xvar: free_var(cast(var, e)); break;
2331 Now that we have the shape of the interpreter in place we can add some
2332 complex types and connected them in to the data structures and the
2333 different phases of parse, analyse, print, interpret.
2335 Being "complex" the language will naturally have syntax to access
2336 specifics of objects of these types. These will fit into the grammar as
2337 "Terms" which are the things that are combined with various operators to
2338 form "Expression". Where a Term is formed by some operation on another
2339 Term, the subordinate Term will always come first, so for example a
2340 member of an array will be expressed as the Term for the array followed
2341 by an index in square brackets. The strict rule of using postfix
2342 operations makes precedence irrelevant within terms. To provide a place
2343 to put the grammar for each terms of each type, we will start out by
2344 introducing the "Term" grammar production, with contains at least a
2345 simple "Value" (to be explained later).
2349 Term -> Value ${ $0 = $<1; }$
2350 | Variable ${ $0 = $<1; }$
2353 Thus far the complex types we have are arrays and structs.
2357 Arrays can be declared by giving a size and a type, as `[size]type' so
2358 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2359 size can be either a literal number, or a named constant. Some day an
2360 arbitrary expression will be supported.
2362 As a formal parameter to a function, the array can be declared with a
2363 new variable as the size: `name:[size::number]string`. The `size`
2364 variable is set to the size of the array and must be a constant. As
2365 `number` is the only supported type, it can be left out:
2366 `name:[size::]string`.
2368 Arrays cannot be assigned. When pointers are introduced we will also
2369 introduce array slices which can refer to part or all of an array -
2370 the assignment syntax will create a slice. For now, an array can only
2371 ever be referenced by the name it is declared with. It is likely that
2372 a "`copy`" primitive will eventually be define which can be used to
2373 make a copy of an array with controllable recursive depth.
2375 For now we have two sorts of array, those with fixed size either because
2376 it is given as a literal number or because it is a struct member (which
2377 cannot have a runtime-changing size), and those with a size that is
2378 determined at runtime - local variables with a const size. The former
2379 have their size calculated at parse time, the latter at run time.
2381 For the latter type, the `size` field of the type is the size of a
2382 pointer, and the array is reallocated every time it comes into scope.
2384 We differentiate struct fields with a const size from local variables
2385 with a const size by whether they are prepared at parse time or not.
2387 ###### type union fields
2390 int unspec; // size is unspecified - vsize must be set.
2393 struct variable *vsize;
2394 struct type *member;
2397 ###### value union fields
2398 void *array; // used if not static_size
2400 ###### value functions
2402 static int array_prepare_type(struct parse_context *c, struct type *type,
2405 struct value *vsize;
2407 if (type->array.static_size)
2408 return 1; // UNTESTED
2409 if (type->array.unspec && parse_time)
2410 return 1; // UNTESTED
2411 if (parse_time && type->array.vsize && !type->array.vsize->global)
2412 return 1; // UNTESTED
2414 if (type->array.vsize) {
2415 vsize = var_value(c, type->array.vsize);
2417 return 1; // UNTESTED
2419 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2420 type->array.size = mpz_get_si(q);
2425 if (type->array.member->size <= 0)
2426 return 0; // UNTESTED
2428 type->array.static_size = 1;
2429 type->size = type->array.size * type->array.member->size;
2430 type->align = type->array.member->align;
2435 static void array_init(struct type *type, struct value *val)
2438 void *ptr = val->ptr;
2442 if (!type->array.static_size) {
2443 val->array = calloc(type->array.size,
2444 type->array.member->size);
2447 for (i = 0; i < type->array.size; i++) {
2449 v = (void*)ptr + i * type->array.member->size;
2450 val_init(type->array.member, v);
2454 static void array_free(struct type *type, struct value *val)
2457 void *ptr = val->ptr;
2459 if (!type->array.static_size)
2461 for (i = 0; i < type->array.size; i++) {
2463 v = (void*)ptr + i * type->array.member->size;
2464 free_value(type->array.member, v);
2466 if (!type->array.static_size)
2470 static int array_compat(struct type *require, struct type *have)
2472 if (have->compat != require->compat)
2474 /* Both are arrays, so we can look at details */
2475 if (!type_compat(require->array.member, have->array.member, 0))
2477 if (have->array.unspec && require->array.unspec) {
2478 if (have->array.vsize && require->array.vsize &&
2479 have->array.vsize != require->array.vsize) // UNTESTED
2480 /* sizes might not be the same */
2481 return 0; // UNTESTED
2484 if (have->array.unspec || require->array.unspec)
2485 return 1; // UNTESTED
2486 if (require->array.vsize == NULL && have->array.vsize == NULL)
2487 return require->array.size == have->array.size;
2489 return require->array.vsize == have->array.vsize; // UNTESTED
2492 static void array_print_type(struct type *type, FILE *f)
2495 if (type->array.vsize) {
2496 struct binding *b = type->array.vsize->name;
2497 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2498 type->array.unspec ? "::" : "");
2499 } else if (type->array.size)
2500 fprintf(f, "%d]", type->array.size);
2503 type_print(type->array.member, f);
2506 static struct type array_prototype = {
2508 .prepare_type = array_prepare_type,
2509 .print_type = array_print_type,
2510 .compat = array_compat,
2512 .size = sizeof(void*),
2513 .align = sizeof(void*),
2516 ###### declare terminals
2521 | [ NUMBER ] Type ${ {
2527 if (number_parse(num, tail, $2.txt) == 0)
2528 tok_err(c, "error: unrecognised number", &$2);
2530 tok_err(c, "error: unsupported number suffix", &$2);
2533 elements = mpz_get_ui(mpq_numref(num));
2534 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2535 tok_err(c, "error: array size must be an integer",
2537 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2538 tok_err(c, "error: array size is too large",
2543 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2544 t->array.size = elements;
2545 t->array.member = $<4;
2546 t->array.vsize = NULL;
2549 | [ IDENTIFIER ] Type ${ {
2550 struct variable *v = var_ref(c, $2.txt);
2553 tok_err(c, "error: name undeclared", &$2);
2554 else if (!v->constant)
2555 tok_err(c, "error: array size must be a constant", &$2);
2557 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2558 $0->array.member = $<4;
2560 $0->array.vsize = v;
2565 OptType -> Type ${ $0 = $<1; }$
2568 ###### formal type grammar
2570 | [ IDENTIFIER :: OptType ] Type ${ {
2571 struct variable *v = var_decl(c, $ID.txt);
2577 $0 = add_anon_type(c, &array_prototype, "array[var]");
2578 $0->array.member = $<6;
2580 $0->array.unspec = 1;
2581 $0->array.vsize = v;
2589 | Term [ Expression ] ${ {
2590 struct binode *b = new(binode);
2597 ###### print binode cases
2599 print_exec(b->left, -1, bracket);
2601 print_exec(b->right, -1, bracket);
2605 ###### propagate binode cases
2607 /* left must be an array, right must be a number,
2608 * result is the member type of the array
2610 propagate_types(b->right, c, perr, Tnum, 0);
2611 t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
2612 if (!t || t->compat != array_compat) {
2613 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2616 if (!type_compat(type, t->array.member, rules)) {
2617 type_err(c, "error: have %1 but need %2", prog,
2618 t->array.member, rules, type);
2620 return t->array.member;
2624 ###### interp binode cases
2630 lleft = linterp_exec(c, b->left, <ype);
2631 right = interp_exec(c, b->right, &rtype);
2633 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2637 if (ltype->array.static_size)
2640 ptr = *(void**)lleft;
2641 rvtype = ltype->array.member;
2642 if (i >= 0 && i < ltype->array.size)
2643 lrv = ptr + i * rvtype->size;
2645 val_init(ltype->array.member, &rv); // UNSAFE
2652 A `struct` is a data-type that contains one or more other data-types.
2653 It differs from an array in that each member can be of a different
2654 type, and they are accessed by name rather than by number. Thus you
2655 cannot choose an element by calculation, you need to know what you
2658 The language makes no promises about how a given structure will be
2659 stored in memory - it is free to rearrange fields to suit whatever
2660 criteria seems important.
2662 Structs are declared separately from program code - they cannot be
2663 declared in-line in a variable declaration like arrays can. A struct
2664 is given a name and this name is used to identify the type - the name
2665 is not prefixed by the word `struct` as it would be in C.
2667 Structs are only treated as the same if they have the same name.
2668 Simply having the same fields in the same order is not enough. This
2669 might change once we can create structure initializers from a list of
2672 Each component datum is identified much like a variable is declared,
2673 with a name, one or two colons, and a type. The type cannot be omitted
2674 as there is no opportunity to deduce the type from usage. An initial
2675 value can be given following an equals sign, so
2677 ##### Example: a struct type
2683 would declare a type called "complex" which has two number fields,
2684 each initialised to zero.
2686 Struct will need to be declared separately from the code that uses
2687 them, so we will need to be able to print out the declaration of a
2688 struct when reprinting the whole program. So a `print_type_decl` type
2689 function will be needed.
2691 ###### type union fields
2700 } *fields; // This is created when field_list is analysed.
2702 struct fieldlist *prev;
2705 } *field_list; // This is created during parsing
2708 ###### type functions
2709 void (*print_type_decl)(struct type *type, FILE *f);
2710 struct type *(*fieldref)(struct type *t, struct parse_context *c,
2711 struct fieldref *f, struct value **vp);
2713 ###### value functions
2715 static void structure_init(struct type *type, struct value *val)
2719 for (i = 0; i < type->structure.nfields; i++) {
2721 v = (void*) val->ptr + type->structure.fields[i].offset;
2722 if (type->structure.fields[i].init)
2723 dup_value(type->structure.fields[i].type,
2724 type->structure.fields[i].init,
2727 val_init(type->structure.fields[i].type, v);
2731 static void structure_free(struct type *type, struct value *val)
2735 for (i = 0; i < type->structure.nfields; i++) {
2737 v = (void*)val->ptr + type->structure.fields[i].offset;
2738 free_value(type->structure.fields[i].type, v);
2742 static void free_fieldlist(struct fieldlist *f)
2746 free_fieldlist(f->prev);
2751 static void structure_free_type(struct type *t)
2754 for (i = 0; i < t->structure.nfields; i++)
2755 if (t->structure.fields[i].init) {
2756 free_value(t->structure.fields[i].type,
2757 t->structure.fields[i].init);
2759 free(t->structure.fields);
2760 free_fieldlist(t->structure.field_list);
2763 static int structure_prepare_type(struct parse_context *c,
2764 struct type *t, int parse_time)
2767 struct fieldlist *f;
2769 if (!parse_time || t->structure.fields)
2772 for (f = t->structure.field_list; f; f=f->prev) {
2776 if (f->f.type->size <= 0)
2778 if (f->f.type->prepare_type)
2779 f->f.type->prepare_type(c, f->f.type, parse_time);
2781 if (f->init == NULL)
2785 propagate_types(f->init, c, &perr, f->f.type, 0);
2786 } while (perr & Eretry);
2788 c->parse_error += 1; // NOTEST
2791 t->structure.nfields = cnt;
2792 t->structure.fields = calloc(cnt, sizeof(struct field));
2793 f = t->structure.field_list;
2795 int a = f->f.type->align;
2797 t->structure.fields[cnt] = f->f;
2798 if (t->size & (a-1))
2799 t->size = (t->size | (a-1)) + 1;
2800 t->structure.fields[cnt].offset = t->size;
2801 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2805 if (f->init && !c->parse_error) {
2806 struct value vl = interp_exec(c, f->init, NULL);
2807 t->structure.fields[cnt].init =
2808 global_alloc(c, f->f.type, NULL, &vl);
2816 static int find_struct_index(struct type *type, struct text field)
2819 for (i = 0; i < type->structure.nfields; i++)
2820 if (text_cmp(type->structure.fields[i].name, field) == 0)
2822 return IndexInvalid;
2825 static struct type *structure_fieldref(struct type *t, struct parse_context *c,
2826 struct fieldref *f, struct value **vp)
2828 if (f->index == IndexUnknown) {
2829 f->index = find_struct_index(t, f->name);
2831 type_err(c, "error: cannot find requested field in %1",
2832 f->left, t, 0, NULL);
2837 struct value *v = *vp;
2838 v = (void*)v->ptr + t->structure.fields[f->index].offset;
2841 return t->structure.fields[f->index].type;
2844 static struct type structure_prototype = {
2845 .init = structure_init,
2846 .free = structure_free,
2847 .free_type = structure_free_type,
2848 .print_type_decl = structure_print_type,
2849 .prepare_type = structure_prepare_type,
2850 .fieldref = structure_fieldref,
2863 enum { IndexUnknown = -1, IndexInvalid = -2 };
2865 ###### free exec cases
2867 free_exec(cast(fieldref, e)->left);
2871 ###### declare terminals
2876 | Term . IDENTIFIER ${ {
2877 struct fieldref *fr = new_pos(fieldref, $2);
2880 fr->index = IndexUnknown;
2884 ###### print exec cases
2888 struct fieldref *f = cast(fieldref, e);
2889 print_exec(f->left, -1, bracket);
2890 printf(".%.*s", f->name.len, f->name.txt);
2894 ###### propagate exec cases
2898 struct fieldref *f = cast(fieldref, prog);
2899 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2901 if (!st || !st->fieldref)
2902 type_err(c, "error: field reference on %1 is not supported",
2903 f->left, st, 0, NULL);
2905 t = st->fieldref(st, c, f, NULL);
2906 if (t && !type_compat(type, t, rules))
2907 type_err(c, "error: have %1 but need %2", prog,
2914 ###### interp exec cases
2917 struct fieldref *f = cast(fieldref, e);
2919 struct value *lleft = linterp_exec(c, f->left, <ype);
2921 rvtype = ltype->fieldref(ltype, c, f, &lrv);
2925 ###### top level grammar
2926 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2928 t = find_type(c, $ID.txt);
2930 t = add_type(c, $ID.txt, &structure_prototype);
2931 else if (t->size >= 0) {
2932 tok_err(c, "error: type already declared", &$ID);
2933 tok_err(c, "info: this is location of declartion", &t->first_use);
2934 /* Create a new one - duplicate */
2935 t = add_type(c, $ID.txt, &structure_prototype);
2937 struct type tmp = *t;
2938 *t = structure_prototype;
2942 t->structure.field_list = $<FB;
2947 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2948 | { SimpleFieldList } ${ $0 = $<SFL; }$
2949 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2950 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2952 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2953 | FieldLines SimpleFieldList Newlines ${
2958 SimpleFieldList -> Field ${ $0 = $<F; }$
2959 | SimpleFieldList ; Field ${
2963 | SimpleFieldList ; ${
2966 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2968 Field -> IDENTIFIER : Type = Expression ${ {
2969 $0 = calloc(1, sizeof(struct fieldlist));
2970 $0->f.name = $ID.txt;
2971 $0->f.type = $<Type;
2975 | IDENTIFIER : Type ${
2976 $0 = calloc(1, sizeof(struct fieldlist));
2977 $0->f.name = $ID.txt;
2978 $0->f.type = $<Type;
2981 ###### forward decls
2982 static void structure_print_type(struct type *t, FILE *f);
2984 ###### value functions
2985 static void structure_print_type(struct type *t, FILE *f)
2989 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2991 for (i = 0; i < t->structure.nfields; i++) {
2992 struct field *fl = t->structure.fields + i;
2993 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2994 type_print(fl->type, f);
2995 if (fl->type->print && fl->init) {
2997 if (fl->type == Tstr)
2998 fprintf(f, "\""); // UNTESTED
2999 print_value(fl->type, fl->init, f);
3000 if (fl->type == Tstr)
3001 fprintf(f, "\""); // UNTESTED
3007 ###### print type decls
3012 while (target != 0) {
3014 for (t = context.typelist; t ; t=t->next)
3015 if (!t->anon && t->print_type_decl &&
3025 t->print_type_decl(t, stdout);
3033 References, or pointers, are values that refer to another value. They
3034 can only refer to a `struct`, though as a struct can embed anything they
3035 can effectively refer to anything.
3037 References are potentially dangerous as they might refer to some
3038 variable which no longer exists - either because a stack frame
3039 containing it has been discarded or because the value was allocated on
3040 the heap and has now been free. Ocean does not yet provide any
3041 protection against these problems. It will in due course.
3043 With references comes the opportunity and the need to explicitly
3044 allocate values on the "heap" and to free them. We currently provide
3045 fairly basic support for this.
3047 Reference make use of the `@` symbol in various ways. A type that starts
3048 with `@` is a reference to whatever follows. A reference value
3049 followed by an `@` acts as the referred value, though the `@` is often
3050 not needed. Finally, an expression that starts with `@` is a special
3051 reference related expression. Some examples might help.
3053 ##### Example: Reference examples
3060 bar.number = 23; bar.string = "hello"
3071 Obviously this is very contrived. `ref` is a reference to a `foo` which
3072 is initially set to refer to the value stored in `bar` - no extra syntax
3073 is needed to "Take the address of" `bar` - the fact that `ref` is a
3074 reference means that only the address make sense.
3076 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
3077 The same syntax is used for accessing fields both in structs and in
3078 references to structs. It would be correct to use `ref@.a`, but not
3081 `@new()` creates an object of whatever type is needed for the program
3082 to by type-correct. In future iterations of Ocean, arguments a
3083 constructor will access arguments, so the the syntax now looks like a
3084 function call. `@free` can be assigned any reference that was returned
3085 by `@new()`, and it will be freed. `@nil` is a value of whatever
3086 reference type is appropriate, and is stable and never the address of
3087 anything in the heap or on the stack. A reference can be assigned
3088 `@nil` or compared against that value.
3090 ###### declare terminals
3093 ###### type union fields
3096 struct type *referent;
3099 ###### value union fields
3102 ###### value functions
3104 static void reference_print_type(struct type *t, FILE *f)
3107 type_print(t->reference.referent, f);
3110 static int reference_cmp(struct type *tl, struct type *tr,
3111 struct value *left, struct value *right)
3113 return left->ref == right->ref ? 0 : 1;
3116 static void reference_dup(struct type *t,
3117 struct value *vold, struct value *vnew)
3119 vnew->ref = vold->ref;
3122 static void reference_free(struct type *t, struct value *v)
3124 /* Nothing to do here */
3127 static int reference_compat(struct type *require, struct type *have)
3129 if (have->compat != require->compat)
3131 if (have->reference.referent != require->reference.referent)
3136 static int reference_test(struct type *type, struct value *val)
3138 return val->ref != NULL;
3141 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3142 struct fieldref *f, struct value **vp)
3144 struct type *rt = t->reference.referent;
3149 return rt->fieldref(rt, c, f, vp);
3151 type_err(c, "error: field reference on %1 is not supported",
3152 f->left, rt, 0, NULL);
3157 static struct type reference_prototype = {
3158 .print_type = reference_print_type,
3159 .cmp_eq = reference_cmp,
3160 .dup = reference_dup,
3161 .test = reference_test,
3162 .free = reference_free,
3163 .compat = reference_compat,
3164 .fieldref = reference_fieldref,
3165 .size = sizeof(void*),
3166 .align = sizeof(void*),
3172 struct type *t = find_type(c, $ID.txt);
3174 t = add_type(c, $ID.txt, NULL);
3177 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3178 $ID.txt.len, $ID.txt.txt);
3179 $0->reference.referent = t;
3182 ###### core functions
3183 static int text_is(struct text t, char *s)
3185 return (strlen(s) == t.len &&
3186 strncmp(s, t.txt, t.len) == 0);
3195 enum ref_func { RefNew, RefFree, RefNil } action;
3196 struct type *reftype;
3200 ###### SimpleStatement Grammar
3202 | @ IDENTIFIER = Expression ${ {
3203 struct ref *r = new_pos(ref, $ID);
3205 if (!text_is($ID.txt, "free"))
3206 tok_err(c, "error: only \"@free\" makes sense here",
3210 r->action = RefFree;
3214 ###### expression grammar
3215 | @ IDENTIFIER ( ) ${
3216 // Only 'new' valid here
3217 if (!text_is($ID.txt, "new")) {
3218 tok_err(c, "error: Only reference function is \"@new()\"",
3221 struct ref *r = new_pos(ref,$ID);
3227 // Only 'nil' valid here
3228 if (!text_is($ID.txt, "nil")) {
3229 tok_err(c, "error: Only reference value is \"@nil\"",
3232 struct ref *r = new_pos(ref,$ID);
3238 ###### print exec cases
3240 struct ref *r = cast(ref, e);
3241 switch (r->action) {
3243 printf("@new()"); break;
3245 printf("@nil"); break;
3247 do_indent(indent, "@free = ");
3248 print_exec(r->right, indent, bracket);
3254 ###### propagate exec cases
3256 struct ref *r = cast(ref, prog);
3257 switch (r->action) {
3259 if (type && type->free != reference_free) {
3260 type_err(c, "error: @new() can only be used with references, not %1",
3261 prog, type, 0, NULL);
3264 if (type && !r->reftype) {
3270 if (type && type->free != reference_free)
3271 type_err(c, "error: @nil can only be used with reference, not %1",
3272 prog, type, 0, NULL);
3273 if (type && !r->reftype) {
3279 t = propagate_types(r->right, c, perr, NULL, 0);
3280 if (t && t->free != reference_free)
3281 type_err(c, "error: @free can only be assigned a reference, not %1",
3290 ###### interp exec cases
3292 struct ref *r = cast(ref, e);
3293 switch (r->action) {
3296 rv.ref = calloc(1, r->reftype->reference.referent->size);
3297 rvtype = r->reftype;
3301 rvtype = r->reftype;
3304 rv = interp_exec(c, r->right, &rvtype);
3305 free_value(rvtype->reference.referent, rv.ref);
3313 ###### free exec cases
3315 struct ref *r = cast(ref, e);
3316 free_exec(r->right);
3321 ###### Expressions: dereference
3329 struct binode *b = new(binode);
3335 ###### print binode cases
3337 print_exec(b->left, -1, bracket);
3341 ###### propagate binode cases
3343 /* left must be a reference, and we return what it refers to */
3344 /* FIXME how can I pass the expected type down? */
3345 t = propagate_types(b->left, c, perr, NULL, 0);
3346 if (!t || t->free != reference_free)
3347 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3349 return t->reference.referent;
3352 ###### interp binode cases
3354 left = interp_exec(c, b->left, <ype);
3356 rvtype = ltype->reference.referent;
3363 A function is a chunk of code which can be passed parameters and can
3364 return results. Each function has a type which includes the set of
3365 parameters and the return value. As yet these types cannot be declared
3366 separately from the function itself.
3368 The parameters can be specified either in parentheses as a ';' separated
3371 ##### Example: function 1
3373 func main(av:[ac::number]string; env:[envc::number]string)
3376 or as an indented list of one parameter per line (though each line can
3377 be a ';' separated list)
3379 ##### Example: function 2
3382 argv:[argc::number]string
3383 env:[envc::number]string
3387 In the first case a return type can follow the parentheses after a colon,
3388 in the second it is given on a line starting with the word `return`.
3390 ##### Example: functions that return
3392 func add(a:number; b:number): number
3402 Rather than returning a type, the function can specify a set of local
3403 variables to return as a struct. The values of these variables when the
3404 function exits will be provided to the caller. For this the return type
3405 is replaced with a block of result declarations, either in parentheses
3406 or bracketed by `return` and `do`.
3408 ##### Example: functions returning multiple variables
3410 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3423 For constructing the lists we use a `List` binode, which will be
3424 further detailed when Expression Lists are introduced.
3426 ###### type union fields
3429 struct binode *params;
3430 struct type *return_type;
3431 struct variable *scope;
3432 int inline_result; // return value is at start of 'local'
3436 ###### value union fields
3437 struct exec *function;
3439 ###### type functions
3440 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3441 struct type *require, struct exec *args);
3443 ###### value functions
3445 static void function_free(struct type *type, struct value *val)
3447 free_exec(val->function);
3448 val->function = NULL;
3451 static int function_compat(struct type *require, struct type *have)
3453 // FIXME can I do anything here yet?
3457 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3458 struct type *require, struct exec *args)
3460 /* This should be 'compat', but we don't have a 'tuple' type to
3461 * hold the type of 'args'
3463 struct binode *arg = cast(binode, args);
3464 struct binode *param = require->function.params;
3467 struct var *pv = cast(var, param->left);
3469 type_err(c, "error: insufficient arguments to function.",
3470 args, NULL, 0, NULL);
3474 propagate_types(arg->left, c, perr, pv->var->type, 0);
3475 param = cast(binode, param->right);
3476 arg = cast(binode, arg->right);
3479 type_err(c, "error: too many arguments to function.",
3480 args, NULL, 0, NULL);
3483 static void function_print(struct type *type, struct value *val, FILE *f)
3485 print_exec(val->function, 1, 0);
3488 static void function_print_type_decl(struct type *type, FILE *f)
3492 for (b = type->function.params; b; b = cast(binode, b->right)) {
3493 struct variable *v = cast(var, b->left)->var;
3494 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3495 v->constant ? "::" : ":");
3496 type_print(v->type, f);
3501 if (type->function.return_type != Tnone) {
3503 if (type->function.inline_result) {
3505 struct type *t = type->function.return_type;
3507 for (i = 0; i < t->structure.nfields; i++) {
3508 struct field *fl = t->structure.fields + i;
3511 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3512 type_print(fl->type, f);
3516 type_print(type->function.return_type, f);
3521 static void function_free_type(struct type *t)
3523 free_exec(t->function.params);
3526 static struct type function_prototype = {
3527 .size = sizeof(void*),
3528 .align = sizeof(void*),
3529 .free = function_free,
3530 .compat = function_compat,
3531 .check_args = function_check_args,
3532 .print = function_print,
3533 .print_type_decl = function_print_type_decl,
3534 .free_type = function_free_type,
3537 ###### declare terminals
3547 FuncName -> IDENTIFIER ${ {
3548 struct variable *v = var_decl(c, $1.txt);
3549 struct var *e = new_pos(var, $1);
3556 v = var_ref(c, $1.txt);
3558 type_err(c, "error: function '%v' redeclared",
3560 type_err(c, "info: this is where '%v' was first declared",
3561 v->where_decl, NULL, 0, NULL);
3567 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3568 | Args ArgsLine NEWLINE ${ {
3569 struct binode *b = $<AL;
3570 struct binode **bp = &b;
3572 bp = (struct binode **)&(*bp)->left;
3577 ArgsLine -> ${ $0 = NULL; }$
3578 | Varlist ${ $0 = $<1; }$
3579 | Varlist ; ${ $0 = $<1; }$
3581 Varlist -> Varlist ; ArgDecl ${
3582 $0 = new_pos(binode, $2);
3595 ArgDecl -> IDENTIFIER : FormalType ${ {
3596 struct variable *v = var_decl(c, $ID.txt);
3597 $0 = new_pos(var, $ID);
3604 ##### Function calls
3606 A function call can appear either as an expression or as a statement.
3607 We use a new 'Funcall' binode type to link the function with a list of
3608 arguments, form with the 'List' nodes.
3610 We have already seen the "Term" which is how a function call can appear
3611 in an expression. To parse a function call into a statement we include
3612 it in the "SimpleStatement Grammar" which will be described later.
3618 | Term ( ExpressionList ) ${ {
3619 struct binode *b = new(binode);
3622 b->right = reorder_bilist($<EL);
3626 struct binode *b = new(binode);
3633 ###### SimpleStatement Grammar
3635 | Term ( ExpressionList ) ${ {
3636 struct binode *b = new(binode);
3639 b->right = reorder_bilist($<EL);
3643 ###### print binode cases
3646 do_indent(indent, "");
3647 print_exec(b->left, -1, bracket);
3649 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3652 print_exec(b->left, -1, bracket);
3662 ###### propagate binode cases
3665 /* Every arg must match formal parameter, and result
3666 * is return type of function
3668 struct binode *args = cast(binode, b->right);
3669 struct var *v = cast(var, b->left);
3671 if (!v->var->type || v->var->type->check_args == NULL) {
3672 type_err(c, "error: attempt to call a non-function.",
3673 prog, NULL, 0, NULL);
3677 v->var->type->check_args(c, perr, v->var->type, args);
3678 if (v->var->type->function.inline_result)
3680 return v->var->type->function.return_type;
3683 ###### interp binode cases
3686 struct var *v = cast(var, b->left);
3687 struct type *t = v->var->type;
3688 void *oldlocal = c->local;
3689 int old_size = c->local_size;
3690 void *local = calloc(1, t->function.local_size);
3691 struct value *fbody = var_value(c, v->var);
3692 struct binode *arg = cast(binode, b->right);
3693 struct binode *param = t->function.params;
3696 struct var *pv = cast(var, param->left);
3697 struct type *vtype = NULL;
3698 struct value val = interp_exec(c, arg->left, &vtype);
3700 c->local = local; c->local_size = t->function.local_size;
3701 lval = var_value(c, pv->var);
3702 c->local = oldlocal; c->local_size = old_size;
3703 memcpy(lval, &val, vtype->size);
3704 param = cast(binode, param->right);
3705 arg = cast(binode, arg->right);
3707 c->local = local; c->local_size = t->function.local_size;
3708 if (t->function.inline_result && dtype) {
3709 _interp_exec(c, fbody->function, NULL, NULL);
3710 memcpy(dest, local, dtype->size);
3711 rvtype = ret.type = NULL;
3713 rv = interp_exec(c, fbody->function, &rvtype);
3714 c->local = oldlocal; c->local_size = old_size;
3719 ## Complex executables: statements and expressions
3721 Now that we have types and values and variables and most of the basic
3722 Terms which provide access to these, we can explore the more complex
3723 code that combine all of these to get useful work done. Specifically
3724 statements and expressions.
3726 Expressions are various combinations of Terms. We will use operator
3727 precedence to ensure correct parsing. The simplest Expression is just a
3728 Term - others will follow.
3733 Expression -> Term ${ $0 = $<Term; }$
3734 ## expression grammar
3736 ### Expressions: Conditional
3738 Our first user of the `binode` will be conditional expressions, which
3739 is a bit odd as they actually have three components. That will be
3740 handled by having 2 binodes for each expression. The conditional
3741 expression is the lowest precedence operator which is why we define it
3742 first - to start the precedence list.
3744 Conditional expressions are of the form "value `if` condition `else`
3745 other_value". They associate to the right, so everything to the right
3746 of `else` is part of an else value, while only a higher-precedence to
3747 the left of `if` is the if values. Between `if` and `else` there is no
3748 room for ambiguity, so a full conditional expression is allowed in
3754 ###### declare terminals
3758 ###### expression grammar
3760 | Expression if Expression else Expression $$ifelse ${ {
3761 struct binode *b1 = new(binode);
3762 struct binode *b2 = new(binode);
3772 ###### print binode cases
3775 b2 = cast(binode, b->right);
3776 if (bracket) printf("(");
3777 print_exec(b2->left, -1, bracket);
3779 print_exec(b->left, -1, bracket);
3781 print_exec(b2->right, -1, bracket);
3782 if (bracket) printf(")");
3785 ###### propagate binode cases
3788 /* cond must be Tbool, others must match */
3789 struct binode *b2 = cast(binode, b->right);
3792 propagate_types(b->left, c, perr, Tbool, 0);
3793 t = propagate_types(b2->left, c, perr, type, Rnolabel);
3794 t2 = propagate_types(b2->right, c, perr, type ?: t, Rnolabel);
3798 ###### interp binode cases
3801 struct binode *b2 = cast(binode, b->right);
3802 left = interp_exec(c, b->left, <ype);
3804 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3806 rv = interp_exec(c, b2->right, &rvtype);
3812 We take a brief detour, now that we have expressions, to describe lists
3813 of expressions. These will be needed for function parameters and
3814 possibly other situations. They seem generic enough to introduce here
3815 to be used elsewhere.
3817 And ExpressionList will use the `List` type of `binode`, building up at
3818 the end. And place where they are used will probably call
3819 `reorder_bilist()` to get a more normal first/next arrangement.
3821 ###### declare terminals
3824 `List` execs have no implicit semantics, so they are never propagated or
3825 interpreted. The can be printed as a comma separate list, which is how
3826 they are parsed. Note they are also used for function formal parameter
3827 lists. In that case a separate function is used to print them.
3829 ###### print binode cases
3833 print_exec(b->left, -1, bracket);
3836 b = cast(binode, b->right);
3840 ###### propagate binode cases
3841 case List: abort(); // NOTEST
3842 ###### interp binode cases
3843 case List: abort(); // NOTEST
3848 ExpressionList -> ExpressionList , Expression ${
3861 ### Expressions: Boolean
3863 The next class of expressions to use the `binode` will be Boolean
3864 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3865 have same corresponding precendence. The difference is that they don't
3866 evaluate the second expression if not necessary.
3875 ###### declare terminals
3880 ###### expression grammar
3881 | Expression or Expression ${ {
3882 struct binode *b = new(binode);
3888 | Expression or else Expression ${ {
3889 struct binode *b = new(binode);
3896 | Expression and Expression ${ {
3897 struct binode *b = new(binode);
3903 | Expression and then Expression ${ {
3904 struct binode *b = new(binode);
3911 | not Expression ${ {
3912 struct binode *b = new(binode);
3918 ###### print binode cases
3920 if (bracket) printf("(");
3921 print_exec(b->left, -1, bracket);
3923 print_exec(b->right, -1, bracket);
3924 if (bracket) printf(")");
3927 if (bracket) printf("(");
3928 print_exec(b->left, -1, bracket);
3929 printf(" and then ");
3930 print_exec(b->right, -1, bracket);
3931 if (bracket) printf(")");
3934 if (bracket) printf("(");
3935 print_exec(b->left, -1, bracket);
3937 print_exec(b->right, -1, bracket);
3938 if (bracket) printf(")");
3941 if (bracket) printf("(");
3942 print_exec(b->left, -1, bracket);
3943 printf(" or else ");
3944 print_exec(b->right, -1, bracket);
3945 if (bracket) printf(")");
3948 if (bracket) printf("(");
3950 print_exec(b->right, -1, bracket);
3951 if (bracket) printf(")");
3954 ###### propagate binode cases
3960 /* both must be Tbool, result is Tbool */
3961 propagate_types(b->left, c, perr, Tbool, 0);
3962 propagate_types(b->right, c, perr, Tbool, 0);
3963 if (type && type != Tbool)
3964 type_err(c, "error: %1 operation found where %2 expected", prog,
3968 ###### interp binode cases
3970 rv = interp_exec(c, b->left, &rvtype);
3971 right = interp_exec(c, b->right, &rtype);
3972 rv.bool = rv.bool && right.bool;
3975 rv = interp_exec(c, b->left, &rvtype);
3977 rv = interp_exec(c, b->right, NULL);
3980 rv = interp_exec(c, b->left, &rvtype);
3981 right = interp_exec(c, b->right, &rtype);
3982 rv.bool = rv.bool || right.bool;
3985 rv = interp_exec(c, b->left, &rvtype);
3987 rv = interp_exec(c, b->right, NULL);
3990 rv = interp_exec(c, b->right, &rvtype);
3994 ### Expressions: Comparison
3996 Of slightly higher precedence that Boolean expressions are Comparisons.
3997 A comparison takes arguments of any comparable type, but the two types
4000 To simplify the parsing we introduce an `eop` which can record an
4001 expression operator, and the `CMPop` non-terminal will match one of them.
4008 ###### ast functions
4009 static void free_eop(struct eop *e)
4023 ###### declare terminals
4024 $LEFT < > <= >= == != CMPop
4026 ###### expression grammar
4027 | Expression CMPop Expression ${ {
4028 struct binode *b = new(binode);
4038 CMPop -> < ${ $0.op = Less; }$
4039 | > ${ $0.op = Gtr; }$
4040 | <= ${ $0.op = LessEq; }$
4041 | >= ${ $0.op = GtrEq; }$
4042 | == ${ $0.op = Eql; }$
4043 | != ${ $0.op = NEql; }$
4045 ###### print binode cases
4053 if (bracket) printf("(");
4054 print_exec(b->left, -1, bracket);
4056 case Less: printf(" < "); break;
4057 case LessEq: printf(" <= "); break;
4058 case Gtr: printf(" > "); break;
4059 case GtrEq: printf(" >= "); break;
4060 case Eql: printf(" == "); break;
4061 case NEql: printf(" != "); break;
4062 default: abort(); // NOTEST
4064 print_exec(b->right, -1, bracket);
4065 if (bracket) printf(")");
4068 ###### propagate binode cases
4075 /* Both must match but not be labels, result is Tbool */
4076 t = propagate_types(b->left, c, perr, NULL, Rnolabel);
4078 propagate_types(b->right, c, perr, t, 0);
4080 t = propagate_types(b->right, c, perr, NULL, Rnolabel); // UNTESTED
4082 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
4084 if (!type_compat(type, Tbool, 0))
4085 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4086 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,
4267 /* both must be Tstr, result is Tstr */
4268 propagate_types(b->left, c, perr, Tstr, 0);
4269 propagate_types(b->right, c, perr, Tstr, 0);
4270 if (!type_compat(type, Tstr, 0))
4271 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4276 /* op must be string, result is number */
4277 propagate_types(b->left, c, perr, Tstr, 0);
4278 if (!type_compat(type, Tnum, 0))
4279 type_err(c, // UNTESTED
4280 "error: Can only convert string to number, not %1",
4281 prog, type, 0, NULL);
4285 /* LHS must support ->test, result is Tbool */
4286 t = propagate_types(b->right, c, perr, NULL, 0);
4288 type_err(c, "error: '?' requires a testable value, not %1",
4293 /* LHS and RHS must match and are returned. Must support
4296 t = propagate_types(b->left, c, perr, type, rules);
4297 t = propagate_types(b->right, c, perr, t, rules);
4298 if (t && t->test == NULL)
4299 type_err(c, "error: \"??\" requires a testable value, not %1",
4304 return propagate_types(b->right, c, perr, type, 0);
4306 ###### interp binode cases
4309 rv = interp_exec(c, b->left, &rvtype);
4310 right = interp_exec(c, b->right, &rtype);
4311 mpq_add(rv.num, rv.num, right.num);
4314 rv = interp_exec(c, b->left, &rvtype);
4315 right = interp_exec(c, b->right, &rtype);
4316 mpq_sub(rv.num, rv.num, right.num);
4319 rv = interp_exec(c, b->left, &rvtype);
4320 right = interp_exec(c, b->right, &rtype);
4321 mpq_mul(rv.num, rv.num, right.num);
4324 rv = interp_exec(c, b->left, &rvtype);
4325 right = interp_exec(c, b->right, &rtype);
4326 mpq_div(rv.num, rv.num, right.num);
4331 left = interp_exec(c, b->left, <ype);
4332 right = interp_exec(c, b->right, &rtype);
4333 mpz_init(l); mpz_init(r); mpz_init(rem);
4334 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4335 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4336 mpz_tdiv_r(rem, l, r);
4337 val_init(Tnum, &rv);
4338 mpq_set_z(rv.num, rem);
4339 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4344 rv = interp_exec(c, b->right, &rvtype);
4345 mpq_neg(rv.num, rv.num);
4348 rv = interp_exec(c, b->right, &rvtype);
4349 mpq_abs(rv.num, rv.num);
4352 rv = interp_exec(c, b->right, &rvtype);
4355 left = interp_exec(c, b->left, <ype);
4356 right = interp_exec(c, b->right, &rtype);
4358 rv.str = text_join(left.str, right.str);
4361 right = interp_exec(c, b->right, &rvtype);
4365 struct text tx = right.str;
4368 if (tx.txt[0] == '-') {
4369 neg = 1; // UNTESTED
4370 tx.txt++; // UNTESTED
4371 tx.len--; // UNTESTED
4373 if (number_parse(rv.num, tail, tx) == 0)
4374 mpq_init(rv.num); // UNTESTED
4376 mpq_neg(rv.num, rv.num); // UNTESTED
4378 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4382 right = interp_exec(c, b->right, &rtype);
4384 rv.bool = !!rtype->test(rtype, &right);
4387 left = interp_exec(c, b->left, <ype);
4388 if (ltype->test(ltype, &left)) {
4393 rv = interp_exec(c, b->right, &rvtype);
4396 ###### value functions
4398 static struct text text_join(struct text a, struct text b)
4401 rv.len = a.len + b.len;
4402 rv.txt = malloc(rv.len);
4403 memcpy(rv.txt, a.txt, a.len);
4404 memcpy(rv.txt+a.len, b.txt, b.len);
4408 ### Blocks, Statements, and Statement lists.
4410 Now that we have expressions out of the way we need to turn to
4411 statements. There are simple statements and more complex statements.
4412 Simple statements do not contain (syntactic) newlines, complex statements do.
4414 Statements often come in sequences and we have corresponding simple
4415 statement lists and complex statement lists.
4416 The former comprise only simple statements separated by semicolons.
4417 The later comprise complex statements and simple statement lists. They are
4418 separated by newlines. Thus the semicolon is only used to separate
4419 simple statements on the one line. This may be overly restrictive,
4420 but I'm not sure I ever want a complex statement to share a line with
4423 Note that a simple statement list can still use multiple lines if
4424 subsequent lines are indented, so
4426 ###### Example: wrapped simple statement list
4431 is a single simple statement list. This might allow room for
4432 confusion, so I'm not set on it yet.
4434 A simple statement list needs no extra syntax. A complex statement
4435 list has two syntactic forms. It can be enclosed in braces (much like
4436 C blocks), or it can be introduced by an indent and continue until an
4437 unindented newline (much like Python blocks). With this extra syntax
4438 it is referred to as a block.
4440 Note that a block does not have to include any newlines if it only
4441 contains simple statements. So both of:
4443 if condition: a=b; d=f
4445 if condition { a=b; print f }
4449 In either case the list is constructed from a `binode` list with
4450 `Block` as the operator. When parsing the list it is most convenient
4451 to append to the end, so a list is a list and a statement. When using
4452 the list it is more convenient to consider a list to be a statement
4453 and a list. So we need a function to re-order a list.
4454 `reorder_bilist` serves this purpose.
4456 The only stand-alone statement we introduce at this stage is `pass`
4457 which does nothing and is represented as a `NULL` pointer in a `Block`
4458 list. Other stand-alone statements will follow once the infrastructure
4461 As many statements will use binodes, we declare a binode pointer 'b' in
4462 the common header for all reductions to use.
4464 ###### Parser: reduce
4475 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4476 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4477 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4478 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4479 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4481 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4482 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4483 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4484 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4485 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4487 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4488 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4489 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4491 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4492 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4493 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4494 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4495 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4497 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4499 ComplexStatements -> ComplexStatements ComplexStatement ${
4509 | ComplexStatement ${
4521 ComplexStatement -> SimpleStatements Newlines ${
4522 $0 = reorder_bilist($<SS);
4524 | SimpleStatements ; Newlines ${
4525 $0 = reorder_bilist($<SS);
4527 ## ComplexStatement Grammar
4530 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4536 | SimpleStatement ${
4545 SimpleStatement -> pass ${ $0 = NULL; }$
4546 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4547 ## SimpleStatement Grammar
4549 ###### print binode cases
4553 if (b->left == NULL) // UNTESTED
4554 printf("pass"); // UNTESTED
4556 print_exec(b->left, indent, bracket); // UNTESTED
4557 if (b->right) { // UNTESTED
4558 printf("; "); // UNTESTED
4559 print_exec(b->right, indent, bracket); // UNTESTED
4562 // block, one per line
4563 if (b->left == NULL)
4564 do_indent(indent, "pass\n");
4566 print_exec(b->left, indent, bracket);
4568 print_exec(b->right, indent, bracket);
4572 ###### propagate binode cases
4575 /* If any statement returns something other than Tnone
4576 * or Tbool then all such must return same type.
4577 * As each statement may be Tnone or something else,
4578 * we must always pass NULL (unknown) down, otherwise an incorrect
4579 * error might occur. We never return Tnone unless it is
4584 for (e = b; e; e = cast(binode, e->right)) {
4585 t = propagate_types(e->left, c, perr, NULL, rules);
4586 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4588 if (t == Tnone && e->right)
4589 /* Only the final statement *must* return a value
4597 type_err(c, "error: expected %1%r, found %2",
4598 e->left, type, rules, t);
4604 ###### interp binode cases
4606 while (rvtype == Tnone &&
4609 rv = interp_exec(c, b->left, &rvtype);
4610 b = cast(binode, b->right);
4614 ### The Print statement
4616 `print` is a simple statement that takes a comma-separated list of
4617 expressions and prints the values separated by spaces and terminated
4618 by a newline. No control of formatting is possible.
4620 `print` uses `ExpressionList` to collect the expressions and stores them
4621 on the left side of a `Print` binode unlessthere is a trailing comma
4622 when the list is stored on the `right` side and no trailing newline is
4628 ##### declare terminals
4631 ###### SimpleStatement Grammar
4633 | print ExpressionList ${
4634 $0 = b = new_pos(binode, $1);
4637 b->left = reorder_bilist($<EL);
4639 | print ExpressionList , ${ {
4640 $0 = b = new_pos(binode, $1);
4642 b->right = reorder_bilist($<EL);
4646 $0 = b = new_pos(binode, $1);
4652 ###### print binode cases
4655 do_indent(indent, "print");
4657 print_exec(b->right, -1, bracket);
4660 print_exec(b->left, -1, bracket);
4665 ###### propagate binode cases
4668 /* don't care but all must be consistent */
4670 b = cast(binode, b->left);
4672 b = cast(binode, b->right);
4674 propagate_types(b->left, c, perr, NULL, Rnolabel);
4675 b = cast(binode, b->right);
4679 ###### interp binode cases
4683 struct binode *b2 = cast(binode, b->left);
4685 b2 = cast(binode, b->right);
4686 for (; b2; b2 = cast(binode, b2->right)) {
4687 left = interp_exec(c, b2->left, <ype);
4688 print_value(ltype, &left, stdout);
4689 free_value(ltype, &left);
4693 if (b->right == NULL)
4699 ###### Assignment statement
4701 An assignment will assign a value to a variable, providing it hasn't
4702 been declared as a constant. The analysis phase ensures that the type
4703 will be correct so the interpreter just needs to perform the
4704 calculation. There is a form of assignment which declares a new
4705 variable as well as assigning a value. If a name is used before
4706 it is declared, it is assumed to be a global constant which are allowed to
4707 be declared at any time.
4713 ###### declare terminals
4716 ###### SimpleStatement Grammar
4717 | Term = Expression ${
4718 $0 = b= new(binode);
4723 | VariableDecl = Expression ${
4724 $0 = b= new(binode);
4731 if ($1->var->where_set == NULL) {
4733 "Variable declared with no type or value: %v",
4737 $0 = b = new(binode);
4744 ###### print binode cases
4747 do_indent(indent, "");
4748 print_exec(b->left, -1, bracket);
4750 print_exec(b->right, -1, bracket);
4757 struct variable *v = cast(var, b->left)->var;
4758 do_indent(indent, "");
4759 print_exec(b->left, -1, bracket);
4760 if (cast(var, b->left)->var->constant) {
4762 if (v->explicit_type) {
4763 type_print(v->type, stdout);
4768 if (v->explicit_type) {
4769 type_print(v->type, stdout);
4775 print_exec(b->right, -1, bracket);
4782 ###### propagate binode cases
4786 /* Both must match and not be labels,
4787 * Type must support 'dup',
4788 * For Assign, left must not be constant.
4791 t = propagate_types(b->left, c, perr, NULL,
4792 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
4797 if (propagate_types(b->right, c, perr, t, 0) != t)
4798 if (b->left->type == Xvar)
4799 type_err(c, "info: variable '%v' was set as %1 here.",
4800 cast(var, b->left)->var->where_set, t, rules, NULL);
4802 t = propagate_types(b->right, c, perr, NULL, Rnolabel);
4804 propagate_types(b->left, c, perr, t,
4805 (b->op == Assign ? Rnoconstant : 0));
4807 if (t && t->dup == NULL && !(*perr & Emaycopy))
4808 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4813 ###### interp binode cases
4816 lleft = linterp_exec(c, b->left, <ype);
4818 dinterp_exec(c, b->right, lleft, ltype, 1);
4824 struct variable *v = cast(var, b->left)->var;
4827 val = var_value(c, v);
4828 if (v->type->prepare_type)
4829 v->type->prepare_type(c, v->type, 0);
4831 dinterp_exec(c, b->right, val, v->type, 0);
4833 val_init(v->type, val);
4837 ### The `use` statement
4839 The `use` statement is the last "simple" statement. It is needed when a
4840 statement block can return a value. This includes the body of a
4841 function which has a return type, and the "condition" code blocks in
4842 `if`, `while`, and `switch` statements.
4847 ###### declare terminals
4850 ###### SimpleStatement Grammar
4852 $0 = b = new_pos(binode, $1);
4857 ###### print binode cases
4860 do_indent(indent, "use ");
4861 print_exec(b->right, -1, bracket);
4866 ###### propagate binode cases
4869 /* result matches value */
4870 return propagate_types(b->right, c, perr, type, 0);
4872 ###### interp binode cases
4875 rv = interp_exec(c, b->right, &rvtype);
4878 ### The Conditional Statement
4880 This is the biggy and currently the only complex statement. This
4881 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4882 It is comprised of a number of parts, all of which are optional though
4883 set combinations apply. Each part is (usually) a key word (`then` is
4884 sometimes optional) followed by either an expression or a code block,
4885 except the `casepart` which is a "key word and an expression" followed
4886 by a code block. The code-block option is valid for all parts and,
4887 where an expression is also allowed, the code block can use the `use`
4888 statement to report a value. If the code block does not report a value
4889 the effect is similar to reporting `True`.
4891 The `else` and `case` parts, as well as `then` when combined with
4892 `if`, can contain a `use` statement which will apply to some
4893 containing conditional statement. `for` parts, `do` parts and `then`
4894 parts used with `for` can never contain a `use`, except in some
4895 subordinate conditional statement.
4897 If there is a `forpart`, it is executed first, only once.
4898 If there is a `dopart`, then it is executed repeatedly providing
4899 always that the `condpart` or `cond`, if present, does not return a non-True
4900 value. `condpart` can fail to return any value if it simply executes
4901 to completion. This is treated the same as returning `True`.
4903 If there is a `thenpart` it will be executed whenever the `condpart`
4904 or `cond` returns True (or does not return any value), but this will happen
4905 *after* `dopart` (when present).
4907 If `elsepart` is present it will be executed at most once when the
4908 condition returns `False` or some value that isn't `True` and isn't
4909 matched by any `casepart`. If there are any `casepart`s, they will be
4910 executed when the condition returns a matching value.
4912 The particular sorts of values allowed in case parts has not yet been
4913 determined in the language design, so nothing is prohibited.
4915 The various blocks in this complex statement potentially provide scope
4916 for variables as described earlier. Each such block must include the
4917 "OpenScope" nonterminal before parsing the block, and must call
4918 `var_block_close()` when closing the block.
4920 The code following "`if`", "`switch`" and "`for`" does not get its own
4921 scope, but is in a scope covering the whole statement, so names
4922 declared there cannot be redeclared elsewhere. Similarly the
4923 condition following "`while`" is in a scope the covers the body
4924 ("`do`" part) of the loop, and which does not allow conditional scope
4925 extension. Code following "`then`" (both looping and non-looping),
4926 "`else`" and "`case`" each get their own local scope.
4928 The type requirements on the code block in a `whilepart` are quite
4929 unusal. It is allowed to return a value of some identifiable type, in
4930 which case the loop aborts and an appropriate `casepart` is run, or it
4931 can return a Boolean, in which case the loop either continues to the
4932 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4933 This is different both from the `ifpart` code block which is expected to
4934 return a Boolean, or the `switchpart` code block which is expected to
4935 return the same type as the casepart values. The correct analysis of
4936 the type of the `whilepart` code block is the reason for the
4937 `Rboolok` flag which is passed to `propagate_types()`.
4939 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4940 defined. As there are two scopes which cover multiple parts - one for
4941 the whole statement and one for "while" and "do" - and as we will use
4942 the 'struct exec' to track scopes, we actually need two new types of
4943 exec. One is a `binode` for the looping part, the rest is the
4944 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4945 casepart` to track a list of case parts.
4956 struct exec *action;
4957 struct casepart *next;
4959 struct cond_statement {
4961 struct exec *forpart, *condpart, *thenpart, *elsepart;
4962 struct binode *looppart;
4963 struct casepart *casepart;
4966 ###### ast functions
4968 static void free_casepart(struct casepart *cp)
4972 free_exec(cp->value);
4973 free_exec(cp->action);
4980 static void free_cond_statement(struct cond_statement *s)
4984 free_exec(s->forpart);
4985 free_exec(s->condpart);
4986 free_exec(s->looppart);
4987 free_exec(s->thenpart);
4988 free_exec(s->elsepart);
4989 free_casepart(s->casepart);
4993 ###### free exec cases
4994 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4996 ###### ComplexStatement Grammar
4997 | CondStatement ${ $0 = $<1; }$
4999 ###### declare terminals
5000 $TERM for then while do
5007 // A CondStatement must end with EOL, as does CondSuffix and
5009 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
5010 // may or may not end with EOL
5011 // WhilePart and IfPart include an appropriate Suffix
5013 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
5014 // them. WhilePart opens and closes its own scope.
5015 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
5018 $0->thenpart = $<TP;
5019 $0->looppart = $<WP;
5020 var_block_close(c, CloseSequential, $0);
5022 | ForPart OptNL WhilePart CondSuffix ${
5025 $0->looppart = $<WP;
5026 var_block_close(c, CloseSequential, $0);
5028 | WhilePart CondSuffix ${
5030 $0->looppart = $<WP;
5032 | SwitchPart OptNL CasePart CondSuffix ${
5034 $0->condpart = $<SP;
5035 $CP->next = $0->casepart;
5036 $0->casepart = $<CP;
5037 var_block_close(c, CloseSequential, $0);
5039 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
5041 $0->condpart = $<SP;
5042 $CP->next = $0->casepart;
5043 $0->casepart = $<CP;
5044 var_block_close(c, CloseSequential, $0);
5046 | IfPart IfSuffix ${
5048 $0->condpart = $IP.condpart; $IP.condpart = NULL;
5049 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
5050 // This is where we close an "if" statement
5051 var_block_close(c, CloseSequential, $0);
5054 CondSuffix -> IfSuffix ${
5057 | Newlines CasePart CondSuffix ${
5059 $CP->next = $0->casepart;
5060 $0->casepart = $<CP;
5062 | CasePart CondSuffix ${
5064 $CP->next = $0->casepart;
5065 $0->casepart = $<CP;
5068 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
5069 | Newlines ElsePart ${ $0 = $<EP; }$
5070 | ElsePart ${$0 = $<EP; }$
5072 ElsePart -> else OpenBlock Newlines ${
5073 $0 = new(cond_statement);
5074 $0->elsepart = $<OB;
5075 var_block_close(c, CloseElse, $0->elsepart);
5077 | else OpenScope CondStatement ${
5078 $0 = new(cond_statement);
5079 $0->elsepart = $<CS;
5080 var_block_close(c, CloseElse, $0->elsepart);
5084 CasePart -> case Expression OpenScope ColonBlock ${
5085 $0 = calloc(1,sizeof(struct casepart));
5088 var_block_close(c, CloseParallel, $0->action);
5092 // These scopes are closed in CondStatement
5093 ForPart -> for OpenBlock ${
5097 ThenPart -> then OpenBlock ${
5099 var_block_close(c, CloseSequential, $0);
5103 // This scope is closed in CondStatement
5104 WhilePart -> while UseBlock OptNL do OpenBlock ${
5109 var_block_close(c, CloseSequential, $0->right);
5110 var_block_close(c, CloseSequential, $0);
5112 | while OpenScope Expression OpenScope ColonBlock ${
5117 var_block_close(c, CloseSequential, $0->right);
5118 var_block_close(c, CloseSequential, $0);
5122 IfPart -> if UseBlock OptNL then OpenBlock ${
5125 var_block_close(c, CloseParallel, $0.thenpart);
5127 | if OpenScope Expression OpenScope ColonBlock ${
5130 var_block_close(c, CloseParallel, $0.thenpart);
5132 | if OpenScope Expression OpenScope OptNL then Block ${
5135 var_block_close(c, CloseParallel, $0.thenpart);
5139 // This scope is closed in CondStatement
5140 SwitchPart -> switch OpenScope Expression ${
5143 | switch UseBlock ${
5147 ###### print binode cases
5149 if (b->left && b->left->type == Xbinode &&
5150 cast(binode, b->left)->op == Block) {
5152 do_indent(indent, "while {\n");
5154 do_indent(indent, "while\n");
5155 print_exec(b->left, indent+1, bracket);
5157 do_indent(indent, "} do {\n");
5159 do_indent(indent, "do\n");
5160 print_exec(b->right, indent+1, bracket);
5162 do_indent(indent, "}\n");
5164 do_indent(indent, "while ");
5165 print_exec(b->left, 0, bracket);
5170 print_exec(b->right, indent+1, bracket);
5172 do_indent(indent, "}\n");
5176 ###### print exec cases
5178 case Xcond_statement:
5180 struct cond_statement *cs = cast(cond_statement, e);
5181 struct casepart *cp;
5183 do_indent(indent, "for");
5184 if (bracket) printf(" {\n"); else printf("\n");
5185 print_exec(cs->forpart, indent+1, bracket);
5188 do_indent(indent, "} then {\n");
5190 do_indent(indent, "then\n");
5191 print_exec(cs->thenpart, indent+1, bracket);
5193 if (bracket) do_indent(indent, "}\n");
5196 print_exec(cs->looppart, indent, bracket);
5200 do_indent(indent, "switch");
5202 do_indent(indent, "if");
5203 if (cs->condpart && cs->condpart->type == Xbinode &&
5204 cast(binode, cs->condpart)->op == Block) {
5209 print_exec(cs->condpart, indent+1, bracket);
5211 do_indent(indent, "}\n");
5213 do_indent(indent, "then\n");
5214 print_exec(cs->thenpart, indent+1, bracket);
5218 print_exec(cs->condpart, 0, bracket);
5224 print_exec(cs->thenpart, indent+1, bracket);
5226 do_indent(indent, "}\n");
5231 for (cp = cs->casepart; cp; cp = cp->next) {
5232 do_indent(indent, "case ");
5233 print_exec(cp->value, -1, 0);
5238 print_exec(cp->action, indent+1, bracket);
5240 do_indent(indent, "}\n");
5243 do_indent(indent, "else");
5248 print_exec(cs->elsepart, indent+1, bracket);
5250 do_indent(indent, "}\n");
5255 ###### propagate binode cases
5257 t = propagate_types(b->right, c, perr, Tnone, 0);
5258 if (!type_compat(Tnone, t, 0))
5259 *perr |= Efail; // UNTESTED
5260 return propagate_types(b->left, c, perr, type, rules);
5262 ###### propagate exec cases
5263 case Xcond_statement:
5265 // forpart and looppart->right must return Tnone
5266 // thenpart must return Tnone if there is a loopart,
5267 // otherwise it is like elsepart.
5269 // be bool if there is no casepart
5270 // match casepart->values if there is a switchpart
5271 // either be bool or match casepart->value if there
5273 // elsepart and casepart->action must match the return type
5274 // expected of this statement.
5275 struct cond_statement *cs = cast(cond_statement, prog);
5276 struct casepart *cp;
5278 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5279 if (!type_compat(Tnone, t, 0))
5280 *perr |= Efail; // UNTESTED
5283 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5284 if (!type_compat(Tnone, t, 0))
5285 *perr |= Efail; // UNTESTED
5287 if (cs->casepart == NULL) {
5288 propagate_types(cs->condpart, c, perr, Tbool, 0);
5289 propagate_types(cs->looppart, c, perr, Tbool, 0);
5291 /* Condpart must match case values, with bool permitted */
5293 for (cp = cs->casepart;
5294 cp && !t; cp = cp->next)
5295 t = propagate_types(cp->value, c, perr, NULL, 0);
5296 if (!t && cs->condpart)
5297 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5298 if (!t && cs->looppart)
5299 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5300 // Now we have a type (I hope) push it down
5302 for (cp = cs->casepart; cp; cp = cp->next)
5303 propagate_types(cp->value, c, perr, t, 0);
5304 propagate_types(cs->condpart, c, perr, t, Rboolok);
5305 propagate_types(cs->looppart, c, perr, t, Rboolok);
5308 // (if)then, else, and case parts must return expected type.
5309 if (!cs->looppart && !type)
5310 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5312 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5313 for (cp = cs->casepart;
5315 cp = cp->next) // UNTESTED
5316 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5319 propagate_types(cs->thenpart, c, perr, type, rules);
5320 propagate_types(cs->elsepart, c, perr, type, rules);
5321 for (cp = cs->casepart; cp ; cp = cp->next)
5322 propagate_types(cp->action, c, perr, type, rules);
5328 ###### interp binode cases
5330 // This just performs one iterration of the loop
5331 rv = interp_exec(c, b->left, &rvtype);
5332 if (rvtype == Tnone ||
5333 (rvtype == Tbool && rv.bool != 0))
5334 // rvtype is Tnone or Tbool, doesn't need to be freed
5335 interp_exec(c, b->right, NULL);
5338 ###### interp exec cases
5339 case Xcond_statement:
5341 struct value v, cnd;
5342 struct type *vtype, *cndtype;
5343 struct casepart *cp;
5344 struct cond_statement *cs = cast(cond_statement, e);
5347 interp_exec(c, cs->forpart, NULL);
5349 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5350 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5351 interp_exec(c, cs->thenpart, NULL);
5353 cnd = interp_exec(c, cs->condpart, &cndtype);
5354 if ((cndtype == Tnone ||
5355 (cndtype == Tbool && cnd.bool != 0))) {
5356 // cnd is Tnone or Tbool, doesn't need to be freed
5357 rv = interp_exec(c, cs->thenpart, &rvtype);
5358 // skip else (and cases)
5362 for (cp = cs->casepart; cp; cp = cp->next) {
5363 v = interp_exec(c, cp->value, &vtype);
5364 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5365 free_value(vtype, &v);
5366 free_value(cndtype, &cnd);
5367 rv = interp_exec(c, cp->action, &rvtype);
5370 free_value(vtype, &v);
5372 free_value(cndtype, &cnd);
5374 rv = interp_exec(c, cs->elsepart, &rvtype);
5381 ### Top level structure
5383 All the language elements so far can be used in various places. Now
5384 it is time to clarify what those places are.
5386 At the top level of a file there will be a number of declarations.
5387 Many of the things that can be declared haven't been described yet,
5388 such as functions, procedures, imports, and probably more.
5389 For now there are two sorts of things that can appear at the top
5390 level. They are predefined constants, `struct` types, and the `main`
5391 function. While the syntax will allow the `main` function to appear
5392 multiple times, that will trigger an error if it is actually attempted.
5394 The various declarations do not return anything. They store the
5395 various declarations in the parse context.
5397 ###### Parser: grammar
5400 Ocean -> OptNL DeclarationList
5402 ## declare terminals
5410 DeclarationList -> Declaration
5411 | DeclarationList Declaration
5413 Declaration -> ERROR Newlines ${
5414 tok_err(c, // UNTESTED
5415 "error: unhandled parse error", &$1);
5421 ## top level grammar
5425 ### The `const` section
5427 As well as being defined in with the code that uses them, constants can
5428 be declared at the top level. These have full-file scope, so they are
5429 always `InScope`, even before(!) they have been declared. The value of
5430 a top level constant can be given as an expression, and this is
5431 evaluated after parsing and before execution.
5433 A function call can be used to evaluate a constant, but it will not have
5434 access to any program state, once such statement becomes meaningful.
5435 e.g. arguments and filesystem will not be visible.
5437 Constants are defined in a section that starts with the reserved word
5438 `const` and then has a block with a list of assignment statements.
5439 For syntactic consistency, these must use the double-colon syntax to
5440 make it clear that they are constants. Type can also be given: if
5441 not, the type will be determined during analysis, as with other
5444 ###### parse context
5445 struct binode *constlist;
5447 ###### top level grammar
5451 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5452 | const { SimpleConstList } Newlines
5453 | const IN OptNL ConstList OUT Newlines
5454 | const SimpleConstList Newlines
5456 ConstList -> ConstList SimpleConstLine
5459 SimpleConstList -> SimpleConstList ; Const
5463 SimpleConstLine -> SimpleConstList Newlines
5464 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5467 CType -> Type ${ $0 = $<1; }$
5471 Const -> IDENTIFIER :: CType = Expression ${ {
5473 struct binode *bl, *bv;
5474 struct var *var = new_pos(var, $ID);
5476 v = var_decl(c, $ID.txt);
5478 v->where_decl = var;
5484 v = var_ref(c, $1.txt);
5485 if (v->type == Tnone) {
5486 v->where_decl = var;
5492 tok_err(c, "error: name already declared", &$1);
5493 type_err(c, "info: this is where '%v' was first declared",
5494 v->where_decl, NULL, 0, NULL);
5506 bl->left = c->constlist;
5511 ###### core functions
5512 static void resolve_consts(struct parse_context *c)
5516 enum { none, some, cannot } progress = none;
5518 c->constlist = reorder_bilist(c->constlist);
5521 for (b = cast(binode, c->constlist); b;
5522 b = cast(binode, b->right)) {
5524 struct binode *vb = cast(binode, b->left);
5525 struct var *v = cast(var, vb->left);
5526 if (v->var->frame_pos >= 0)
5530 propagate_types(vb->right, c, &perr,
5532 } while (perr & Eretry);
5534 c->parse_error += 1;
5535 else if (!(perr & Enoconst)) {
5537 struct value res = interp_exec(
5538 c, vb->right, &v->var->type);
5539 global_alloc(c, v->var->type, v->var, &res);
5541 if (progress == cannot)
5542 type_err(c, "error: const %v cannot be resolved.",
5552 progress = cannot; break;
5554 progress = none; break;
5559 ###### print const decls
5564 for (b = cast(binode, context.constlist); b;
5565 b = cast(binode, b->right)) {
5566 struct binode *vb = cast(binode, b->left);
5567 struct var *vr = cast(var, vb->left);
5568 struct variable *v = vr->var;
5574 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5575 type_print(v->type, stdout);
5577 print_exec(vb->right, -1, 0);
5582 ###### free const decls
5583 free_binode(context.constlist);
5585 ### Function declarations
5587 The code in an Ocean program is all stored in function declarations.
5588 One of the functions must be named `main` and it must accept an array of
5589 strings as a parameter - the command line arguments.
5591 As this is the top level, several things are handled a bit differently.
5592 The function is not interpreted by `interp_exec` as that isn't passed
5593 the argument list which the program requires. Similarly type analysis
5594 is a bit more interesting at this level.
5596 ###### ast functions
5598 static struct type *handle_results(struct parse_context *c,
5599 struct binode *results)
5601 /* Create a 'struct' type from the results list, which
5602 * is a list for 'struct var'
5604 struct type *t = add_anon_type(c, &structure_prototype,
5609 for (b = results; b; b = cast(binode, b->right))
5611 t->structure.nfields = cnt;
5612 t->structure.fields = calloc(cnt, sizeof(struct field));
5614 for (b = results; b; b = cast(binode, b->right)) {
5615 struct var *v = cast(var, b->left);
5616 struct field *f = &t->structure.fields[cnt++];
5617 int a = v->var->type->align;
5618 f->name = v->var->name->name;
5619 f->type = v->var->type;
5621 f->offset = t->size;
5622 v->var->frame_pos = f->offset;
5623 t->size += ((f->type->size - 1) | (a-1)) + 1;
5626 variable_unlink_exec(v->var);
5628 free_binode(results);
5632 static struct variable *declare_function(struct parse_context *c,
5633 struct variable *name,
5634 struct binode *args,
5636 struct binode *results,
5640 struct value fn = {.function = code};
5642 var_block_close(c, CloseFunction, code);
5643 t = add_anon_type(c, &function_prototype,
5644 "func %.*s", name->name->name.len,
5645 name->name->name.txt);
5647 t->function.params = reorder_bilist(args);
5649 ret = handle_results(c, reorder_bilist(results));
5650 t->function.inline_result = 1;
5651 t->function.local_size = ret->size;
5653 t->function.return_type = ret;
5654 global_alloc(c, t, name, &fn);
5655 name->type->function.scope = c->out_scope;
5660 var_block_close(c, CloseFunction, NULL);
5662 c->out_scope = NULL;
5666 ###### declare terminals
5669 ###### top level grammar
5672 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5673 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5675 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5676 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5678 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5679 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5681 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5682 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5684 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5685 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5687 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5688 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5690 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5691 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5693 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5694 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5696 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5697 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5700 ###### print func decls
5705 while (target != 0) {
5707 for (v = context.in_scope; v; v=v->in_scope)
5708 if (v->depth == 0 && v->type && v->type->check_args) {
5717 struct value *val = var_value(&context, v);
5718 printf("func %.*s", v->name->name.len, v->name->name.txt);
5719 v->type->print_type_decl(v->type, stdout);
5721 print_exec(val->function, 0, brackets);
5723 print_value(v->type, val, stdout);
5724 printf("/* frame size %d */\n", v->type->function.local_size);
5730 ###### core functions
5732 static int analyse_funcs(struct parse_context *c)
5736 for (v = c->in_scope; v; v = v->in_scope) {
5740 if (v->depth != 0 || !v->type || !v->type->check_args)
5742 ret = v->type->function.inline_result ?
5743 Tnone : v->type->function.return_type;
5744 val = var_value(c, v);
5747 propagate_types(val->function, c, &perr, ret, 0);
5748 } while (!(perr & Efail) && (perr & Eretry));
5749 if (!(perr & Efail))
5750 /* Make sure everything is still consistent */
5751 propagate_types(val->function, c, &perr, ret, 0);
5754 if (!v->type->function.inline_result &&
5755 !v->type->function.return_type->dup) {
5756 type_err(c, "error: function cannot return value of type %1",
5757 v->where_decl, v->type->function.return_type, 0, NULL);
5760 scope_finalize(c, v->type);
5765 static int analyse_main(struct type *type, struct parse_context *c)
5767 struct binode *bp = type->function.params;
5771 struct type *argv_type;
5773 argv_type = add_anon_type(c, &array_prototype, "argv");
5774 argv_type->array.member = Tstr;
5775 argv_type->array.unspec = 1;
5777 for (b = bp; b; b = cast(binode, b->right)) {
5781 propagate_types(b->left, c, &perr, argv_type, 0);
5783 default: /* invalid */ // NOTEST
5784 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5787 c->parse_error += 1;
5790 return !c->parse_error;
5793 static void interp_main(struct parse_context *c, int argc, char **argv)
5795 struct value *progp = NULL;
5796 struct text main_name = { "main", 4 };
5797 struct variable *mainv;
5803 mainv = var_ref(c, main_name);
5805 progp = var_value(c, mainv);
5806 if (!progp || !progp->function) {
5807 fprintf(stderr, "oceani: no main function found.\n");
5808 c->parse_error += 1;
5811 if (!analyse_main(mainv->type, c)) {
5812 fprintf(stderr, "oceani: main has wrong type.\n");
5813 c->parse_error += 1;
5816 al = mainv->type->function.params;
5818 c->local_size = mainv->type->function.local_size;
5819 c->local = calloc(1, c->local_size);
5821 struct var *v = cast(var, al->left);
5822 struct value *vl = var_value(c, v->var);
5832 mpq_set_ui(argcq, argc, 1);
5833 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5834 t->prepare_type(c, t, 0);
5835 array_init(v->var->type, vl);
5836 for (i = 0; i < argc; i++) {
5837 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5839 arg.str.txt = argv[i];
5840 arg.str.len = strlen(argv[i]);
5841 free_value(Tstr, vl2);
5842 dup_value(Tstr, &arg, vl2);
5846 al = cast(binode, al->right);
5848 v = interp_exec(c, progp->function, &vtype);
5849 free_value(vtype, &v);
5854 ###### ast functions
5855 void free_variable(struct variable *v)
5859 ## And now to test it out.
5861 Having a language requires having a "hello world" program. I'll
5862 provide a little more than that: a program that prints "Hello world"
5863 finds the GCD of two numbers, prints the first few elements of
5864 Fibonacci, performs a binary search for a number, and a few other
5865 things which will likely grow as the languages grows.
5867 ###### File: oceani.mk
5870 @echo "===== DEMO ====="
5871 ./oceani --section "demo: hello" oceani.mdc 55 33
5877 four ::= 2 + 2 ; five ::= 10/2
5878 const pie ::= "I like Pie";
5879 cake ::= "The cake is"
5887 func main(argv:[argc::]string)
5888 print "Hello World, what lovely oceans you have!"
5889 print "Are there", five, "?"
5890 print pi, pie, "but", cake
5892 A := $argv[1]; B := $argv[2]
5894 /* When a variable is defined in both branches of an 'if',
5895 * and used afterwards, the variables are merged.
5901 print "Is", A, "bigger than", B,"? ", bigger
5902 /* If a variable is not used after the 'if', no
5903 * merge happens, so types can be different
5906 double:string = "yes"
5907 print A, "is more than twice", B, "?", double
5910 print "double", B, "is", double
5915 if a > 0 and then b > 0:
5921 print "GCD of", A, "and", B,"is", a
5923 print a, "is not positive, cannot calculate GCD"
5925 print b, "is not positive, cannot calculate GCD"
5930 print "Fibonacci:", f1,f2,
5931 then togo = togo - 1
5939 /* Binary search... */
5944 mid := (lo + hi) / 2
5957 print "Yay, I found", target
5959 print "Closest I found was", lo
5964 // "middle square" PRNG. Not particularly good, but one my
5965 // Dad taught me - the first one I ever heard of.
5966 for i:=1; then i = i + 1; while i < size:
5967 n := list[i-1] * list[i-1]
5968 list[i] = (n / 100) % 10 000
5970 print "Before sort:",
5971 for i:=0; then i = i + 1; while i < size:
5975 for i := 1; then i=i+1; while i < size:
5976 for j:=i-1; then j=j-1; while j >= 0:
5977 if list[j] > list[j+1]:
5981 print " After sort:",
5982 for i:=0; then i = i + 1; while i < size:
5986 if 1 == 2 then print "yes"; else print "no"
5990 bob.alive = (bob.name == "Hello")
5991 print "bob", "is" if bob.alive else "isn't", "alive"