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);
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;
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-%p*", 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);
1270 $0 = new_val(Tbool, $1);
1274 $0 = new_val(Tbool, $1);
1279 $0 = new_val(Tnum, $1);
1280 if (number_parse($0->val.num, tail, $1.txt) == 0)
1281 mpq_init($0->val.num); // UNTESTED
1283 tok_err(c, "error: unsupported number suffix",
1288 $0 = new_val(Tstr, $1);
1289 string_parse(&$1, '\\', &$0->val.str, tail);
1291 tok_err(c, "error: unsupported string suffix",
1296 $0 = new_val(Tstr, $1);
1297 string_parse(&$1, '\\', &$0->val.str, tail);
1299 tok_err(c, "error: unsupported string suffix",
1303 ###### print exec cases
1306 struct val *v = cast(val, e);
1307 if (v->vtype == Tstr)
1309 // FIXME how to ensure numbers have same precision.
1310 print_value(v->vtype, &v->val, stdout);
1311 if (v->vtype == Tstr)
1316 ###### propagate exec cases
1319 struct val *val = cast(val, prog);
1320 if (!type_compat(type, val->vtype, rules))
1321 type_err(c, "error: expected %1%r found %2",
1322 prog, type, rules, val->vtype);
1326 ###### interp exec cases
1328 rvtype = cast(val, e)->vtype;
1329 dup_value(rvtype, &cast(val, e)->val, &rv);
1332 ###### ast functions
1333 static void free_val(struct val *v)
1336 free_value(v->vtype, &v->val);
1340 ###### free exec cases
1341 case Xval: free_val(cast(val, e)); break;
1343 ###### ast functions
1344 // Move all nodes from 'b' to 'rv', reversing their order.
1345 // In 'b' 'left' is a list, and 'right' is the last node.
1346 // In 'rv', left' is the first node and 'right' is a list.
1347 static struct binode *reorder_bilist(struct binode *b)
1349 struct binode *rv = NULL;
1352 struct exec *t = b->right;
1356 b = cast(binode, b->left);
1366 Variables are scoped named values. We store the names in a linked list
1367 of "bindings" sorted in lexical order, and use sequential search and
1374 struct binding *next; // in lexical order
1378 This linked list is stored in the parse context so that "reduce"
1379 functions can find or add variables, and so the analysis phase can
1380 ensure that every variable gets a type.
1382 ###### parse context
1384 struct binding *varlist; // In lexical order
1386 ###### ast functions
1388 static struct binding *find_binding(struct parse_context *c, struct text s)
1390 struct binding **l = &c->varlist;
1395 (cmp = text_cmp((*l)->name, s)) < 0)
1399 n = calloc(1, sizeof(*n));
1406 Each name can be linked to multiple variables defined in different
1407 scopes. Each scope starts where the name is declared and continues
1408 until the end of the containing code block. Scopes of a given name
1409 cannot nest, so a declaration while a name is in-scope is an error.
1411 ###### binding fields
1412 struct variable *var;
1416 struct variable *previous;
1418 struct binding *name;
1419 struct exec *where_decl;// where name was declared
1420 struct exec *where_set; // where type was set
1424 When a scope closes, the values of the variables might need to be freed.
1425 This happens in the context of some `struct exec` and each `exec` will
1426 need to know which variables need to be freed when it completes.
1429 struct variable *to_free;
1431 ####### variable fields
1432 struct exec *cleanup_exec;
1433 struct variable *next_free;
1435 ####### interp exec cleanup
1438 for (v = e->to_free; v; v = v->next_free) {
1439 struct value *val = var_value(c, v);
1440 free_value(v->type, val);
1444 ###### ast functions
1445 static void variable_unlink_exec(struct variable *v)
1447 struct variable **vp;
1448 if (!v->cleanup_exec)
1450 for (vp = &v->cleanup_exec->to_free;
1451 *vp; vp = &(*vp)->next_free) {
1455 v->cleanup_exec = NULL;
1460 While the naming seems strange, we include local constants in the
1461 definition of variables. A name declared `var := value` can
1462 subsequently be changed, but a name declared `var ::= value` cannot -
1465 ###### variable fields
1468 Scopes in parallel branches can be partially merged. More
1469 specifically, if a given name is declared in both branches of an
1470 if/else then its scope is a candidate for merging. Similarly if
1471 every branch of an exhaustive switch (e.g. has an "else" clause)
1472 declares a given name, then the scopes from the branches are
1473 candidates for merging.
1475 Note that names declared inside a loop (which is only parallel to
1476 itself) are never visible after the loop. Similarly names defined in
1477 scopes which are not parallel, such as those started by `for` and
1478 `switch`, are never visible after the scope. Only variables defined in
1479 both `then` and `else` (including the implicit then after an `if`, and
1480 excluding `then` used with `for`) and in all `case`s and `else` of a
1481 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1483 Labels, which are a bit like variables, follow different rules.
1484 Labels are not explicitly declared, but if an undeclared name appears
1485 in a context where a label is legal, that effectively declares the
1486 name as a label. The declaration remains in force (or in scope) at
1487 least to the end of the immediately containing block and conditionally
1488 in any larger containing block which does not declare the name in some
1489 other way. Importantly, the conditional scope extension happens even
1490 if the label is only used in one parallel branch of a conditional --
1491 when used in one branch it is treated as having been declared in all
1494 Merge candidates are tentatively visible beyond the end of the
1495 branching statement which creates them. If the name is used, the
1496 merge is affirmed and they become a single variable visible at the
1497 outer layer. If not - if it is redeclared first - the merge lapses.
1499 To track scopes we have an extra stack, implemented as a linked list,
1500 which roughly parallels the parse stack and which is used exclusively
1501 for scoping. When a new scope is opened, a new frame is pushed and
1502 the child-count of the parent frame is incremented. This child-count
1503 is used to distinguish between the first of a set of parallel scopes,
1504 in which declared variables must not be in scope, and subsequent
1505 branches, whether they may already be conditionally scoped.
1507 We need a total ordering of scopes so we can easily compare to variables
1508 to see if they are concurrently in scope. To achieve this we record a
1509 `scope_count` which is actually a count of both beginnings and endings
1510 of scopes. Then each variable has a record of the scope count where it
1511 enters scope, and where it leaves.
1513 To push a new frame *before* any code in the frame is parsed, we need a
1514 grammar reduction. This is most easily achieved with a grammar
1515 element which derives the empty string, and creates the new scope when
1516 it is recognised. This can be placed, for example, between a keyword
1517 like "if" and the code following it.
1521 struct scope *parent;
1525 ###### parse context
1528 struct scope *scope_stack;
1530 ###### variable fields
1531 int scope_start, scope_end;
1533 ###### ast functions
1534 static void scope_pop(struct parse_context *c)
1536 struct scope *s = c->scope_stack;
1538 c->scope_stack = s->parent;
1540 c->scope_depth -= 1;
1541 c->scope_count += 1;
1544 static void scope_push(struct parse_context *c)
1546 struct scope *s = calloc(1, sizeof(*s));
1548 c->scope_stack->child_count += 1;
1549 s->parent = c->scope_stack;
1551 c->scope_depth += 1;
1552 c->scope_count += 1;
1558 OpenScope -> ${ scope_push(c); }$
1560 Each variable records a scope depth and is in one of four states:
1562 - "in scope". This is the case between the declaration of the
1563 variable and the end of the containing block, and also between
1564 the usage with affirms a merge and the end of that block.
1566 The scope depth is not greater than the current parse context scope
1567 nest depth. When the block of that depth closes, the state will
1568 change. To achieve this, all "in scope" variables are linked
1569 together as a stack in nesting order.
1571 - "pending". The "in scope" block has closed, but other parallel
1572 scopes are still being processed. So far, every parallel block at
1573 the same level that has closed has declared the name.
1575 The scope depth is the depth of the last parallel block that
1576 enclosed the declaration, and that has closed.
1578 - "conditionally in scope". The "in scope" block and all parallel
1579 scopes have closed, and no further mention of the name has been seen.
1580 This state includes a secondary nest depth (`min_depth`) which records
1581 the outermost scope seen since the variable became conditionally in
1582 scope. If a use of the name is found, the variable becomes "in scope"
1583 and that secondary depth becomes the recorded scope depth. If the
1584 name is declared as a new variable, the old variable becomes "out of
1585 scope" and the recorded scope depth stays unchanged.
1587 - "out of scope". The variable is neither in scope nor conditionally
1588 in scope. It is permanently out of scope now and can be removed from
1589 the "in scope" stack. When a variable becomes out-of-scope it is
1590 moved to a separate list (`out_scope`) of variables which have fully
1591 known scope. This will be used at the end of each function to assign
1592 each variable a place in the stack frame.
1594 ###### variable fields
1595 int depth, min_depth;
1596 enum { OutScope, PendingScope, CondScope, InScope } scope;
1597 struct variable *in_scope;
1599 ###### parse context
1601 struct variable *in_scope;
1602 struct variable *out_scope;
1604 All variables with the same name are linked together using the
1605 'previous' link. Those variable that have been affirmatively merged all
1606 have a 'merged' pointer that points to one primary variable - the most
1607 recently declared instance. When merging variables, we need to also
1608 adjust the 'merged' pointer on any other variables that had previously
1609 been merged with the one that will no longer be primary.
1611 A variable that is no longer the most recent instance of a name may
1612 still have "pending" scope, if it might still be merged with most
1613 recent instance. These variables don't really belong in the
1614 "in_scope" list, but are not immediately removed when a new instance
1615 is found. Instead, they are detected and ignored when considering the
1616 list of in_scope names.
1618 The storage of the value of a variable will be described later. For now
1619 we just need to know that when a variable goes out of scope, it might
1620 need to be freed. For this we need to be able to find it, so assume that
1621 `var_value()` will provide that.
1623 ###### variable fields
1624 struct variable *merged;
1626 ###### ast functions
1628 static void variable_merge(struct variable *primary, struct variable *secondary)
1632 primary = primary->merged;
1634 for (v = primary->previous; v; v=v->previous)
1635 if (v == secondary || v == secondary->merged ||
1636 v->merged == secondary ||
1637 v->merged == secondary->merged) {
1638 v->scope = OutScope;
1639 v->merged = primary;
1640 if (v->scope_start < primary->scope_start)
1641 primary->scope_start = v->scope_start;
1642 if (v->scope_end > primary->scope_end)
1643 primary->scope_end = v->scope_end; // NOTEST
1644 variable_unlink_exec(v);
1648 ###### forward decls
1649 static struct value *var_value(struct parse_context *c, struct variable *v);
1651 ###### free global vars
1653 while (context.varlist) {
1654 struct binding *b = context.varlist;
1655 struct variable *v = b->var;
1656 context.varlist = b->next;
1659 struct variable *next = v->previous;
1661 if (v->global && v->frame_pos >= 0) {
1662 free_value(v->type, var_value(&context, v));
1663 if (v->depth == 0 && v->type->free == function_free)
1664 // This is a function constant
1665 free_exec(v->where_decl);
1672 #### Manipulating Bindings
1674 When a name is conditionally visible, a new declaration discards the old
1675 binding - the condition lapses. Similarly when we reach the end of a
1676 function (outermost non-global scope) any conditional scope must lapse.
1677 Conversely a usage of the name affirms the visibility and extends it to
1678 the end of the containing block - i.e. the block that contains both the
1679 original declaration and the latest usage. This is determined from
1680 `min_depth`. When a conditionally visible variable gets affirmed like
1681 this, it is also merged with other conditionally visible variables with
1684 When we parse a variable declaration we either report an error if the
1685 name is currently bound, or create a new variable at the current nest
1686 depth if the name is unbound or bound to a conditionally scoped or
1687 pending-scope variable. If the previous variable was conditionally
1688 scoped, it and its homonyms becomes out-of-scope.
1690 When we parse a variable reference (including non-declarative assignment
1691 "foo = bar") we report an error if the name is not bound or is bound to
1692 a pending-scope variable; update the scope if the name is bound to a
1693 conditionally scoped variable; or just proceed normally if the named
1694 variable is in scope.
1696 When we exit a scope, any variables bound at this level are either
1697 marked out of scope or pending-scoped, depending on whether the scope
1698 was sequential or parallel. Here a "parallel" scope means the "then"
1699 or "else" part of a conditional, or any "case" or "else" branch of a
1700 switch. Other scopes are "sequential".
1702 When exiting a parallel scope we check if there are any variables that
1703 were previously pending and are still visible. If there are, then
1704 they weren't redeclared in the most recent scope, so they cannot be
1705 merged and must become out-of-scope. If it is not the first of
1706 parallel scopes (based on `child_count`), we check that there was a
1707 previous binding that is still pending-scope. If there isn't, the new
1708 variable must now be out-of-scope.
1710 When exiting a sequential scope that immediately enclosed parallel
1711 scopes, we need to resolve any pending-scope variables. If there was
1712 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1713 we need to mark all pending-scope variable as out-of-scope. Otherwise
1714 all pending-scope variables become conditionally scoped.
1717 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1719 ###### ast functions
1721 static struct variable *var_decl(struct parse_context *c, struct text s)
1723 struct binding *b = find_binding(c, s);
1724 struct variable *v = b->var;
1726 switch (v ? v->scope : OutScope) {
1728 /* Caller will report the error */
1732 v && v->scope == CondScope;
1734 v->scope = OutScope;
1738 v = calloc(1, sizeof(*v));
1739 v->previous = b->var;
1743 v->min_depth = v->depth = c->scope_depth;
1745 v->in_scope = c->in_scope;
1746 v->scope_start = c->scope_count;
1752 static struct variable *var_ref(struct parse_context *c, struct text s)
1754 struct binding *b = find_binding(c, s);
1755 struct variable *v = b->var;
1756 struct variable *v2;
1758 switch (v ? v->scope : OutScope) {
1761 /* Caller will report the error */
1764 /* All CondScope variables of this name need to be merged
1765 * and become InScope
1767 v->depth = v->min_depth;
1769 for (v2 = v->previous;
1770 v2 && v2->scope == CondScope;
1772 variable_merge(v, v2);
1780 static int var_refile(struct parse_context *c, struct variable *v)
1782 /* Variable just went out of scope. Add it to the out_scope
1783 * list, sorted by ->scope_start
1785 struct variable **vp = &c->out_scope;
1786 while ((*vp) && (*vp)->scope_start < v->scope_start)
1787 vp = &(*vp)->in_scope;
1793 static void var_block_close(struct parse_context *c, enum closetype ct,
1796 /* Close off all variables that are in_scope.
1797 * Some variables in c->scope may already be not-in-scope,
1798 * such as when a PendingScope variable is hidden by a new
1799 * variable with the same name.
1800 * So we check for v->name->var != v and drop them.
1801 * If we choose to make a variable OutScope, we drop it
1804 struct variable *v, **vp, *v2;
1807 for (vp = &c->in_scope;
1808 (v = *vp) && v->min_depth > c->scope_depth;
1809 (v->scope == OutScope || v->name->var != v)
1810 ? (*vp = v->in_scope, var_refile(c, v))
1811 : ( vp = &v->in_scope, 0)) {
1812 v->min_depth = c->scope_depth;
1813 if (v->name->var != v)
1814 /* This is still in scope, but we haven't just
1818 v->min_depth = c->scope_depth;
1819 if (v->scope == InScope)
1820 v->scope_end = c->scope_count;
1821 if (v->scope == InScope && e && !v->global) {
1822 /* This variable gets cleaned up when 'e' finishes */
1823 variable_unlink_exec(v);
1824 v->cleanup_exec = e;
1825 v->next_free = e->to_free;
1830 case CloseParallel: /* handle PendingScope */
1834 if (c->scope_stack->child_count == 1)
1835 /* first among parallel branches */
1836 v->scope = PendingScope;
1837 else if (v->previous &&
1838 v->previous->scope == PendingScope)
1839 /* all previous branches used name */
1840 v->scope = PendingScope;
1841 else if (v->type == Tlabel)
1842 /* Labels remain pending even when not used */
1843 v->scope = PendingScope; // UNTESTED
1845 v->scope = OutScope;
1846 if (ct == CloseElse) {
1847 /* All Pending variables with this name
1848 * are now Conditional */
1850 v2 && v2->scope == PendingScope;
1852 v2->scope = CondScope;
1856 /* Not possible as it would require
1857 * parallel scope to be nested immediately
1858 * in a parallel scope, and that never
1862 /* Not possible as we already tested for
1869 if (v->scope == CondScope)
1870 /* Condition cannot continue past end of function */
1873 case CloseSequential:
1874 if (v->type == Tlabel)
1875 v->scope = PendingScope;
1878 v->scope = OutScope;
1881 /* There was no 'else', so we can only become
1882 * conditional if we know the cases were exhaustive,
1883 * and that doesn't mean anything yet.
1884 * So only labels become conditional..
1887 v2 && v2->scope == PendingScope;
1889 if (v2->type == Tlabel)
1890 v2->scope = CondScope;
1892 v2->scope = OutScope;
1895 case OutScope: break;
1904 The value of a variable is store separately from the variable, on an
1905 analogue of a stack frame. There are (currently) two frames that can be
1906 active. A global frame which currently only stores constants, and a
1907 stacked frame which stores local variables. Each variable knows if it
1908 is global or not, and what its index into the frame is.
1910 Values in the global frame are known immediately they are relevant, so
1911 the frame needs to be reallocated as it grows so it can store those
1912 values. The local frame doesn't get values until the interpreted phase
1913 is started, so there is no need to allocate until the size is known.
1915 We initialize the `frame_pos` to an impossible value, so that we can
1916 tell if it was set or not later.
1918 ###### variable fields
1922 ###### variable init
1925 ###### parse context
1927 short global_size, global_alloc;
1929 void *global, *local;
1931 ###### forward decls
1932 static struct value *global_alloc(struct parse_context *c, struct type *t,
1933 struct variable *v, struct value *init);
1935 ###### ast functions
1937 static struct value *var_value(struct parse_context *c, struct variable *v)
1940 if (!c->local || !v->type)
1941 return NULL; // UNTESTED
1942 if (v->frame_pos + v->type->size > c->local_size) {
1943 printf("INVALID frame_pos\n"); // NOTEST
1946 return c->local + v->frame_pos;
1948 if (c->global_size > c->global_alloc) {
1949 int old = c->global_alloc;
1950 c->global_alloc = (c->global_size | 1023) + 1024;
1951 c->global = realloc(c->global, c->global_alloc);
1952 memset(c->global + old, 0, c->global_alloc - old);
1954 return c->global + v->frame_pos;
1957 static struct value *global_alloc(struct parse_context *c, struct type *t,
1958 struct variable *v, struct value *init)
1961 struct variable scratch;
1963 if (t->prepare_type)
1964 t->prepare_type(c, t, 1); // NOTEST
1966 if (c->global_size & (t->align - 1))
1967 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
1972 v->frame_pos = c->global_size;
1974 c->global_size += v->type->size;
1975 ret = var_value(c, v);
1977 memcpy(ret, init, t->size);
1983 As global values are found -- struct field initializers, labels etc --
1984 `global_alloc()` is called to record the value in the global frame.
1986 When the program is fully parsed, each function is analysed, we need to
1987 walk the list of variables local to that function and assign them an
1988 offset in the stack frame. For this we have `scope_finalize()`.
1990 We keep the stack from dense by re-using space for between variables
1991 that are not in scope at the same time. The `out_scope` list is sorted
1992 by `scope_start` and as we process a varible, we move it to an FIFO
1993 stack. For each variable we consider, we first discard any from the
1994 stack anything that went out of scope before the new variable came in.
1995 Then we place the new variable just after the one at the top of the
1998 ###### ast functions
2000 static void scope_finalize(struct parse_context *c, struct type *ft)
2002 int size = ft->function.local_size;
2003 struct variable *next = ft->function.scope;
2004 struct variable *done = NULL;
2007 struct variable *v = next;
2008 struct type *t = v->type;
2015 if (v->frame_pos >= 0)
2017 while (done && done->scope_end < v->scope_start)
2018 done = done->in_scope;
2020 pos = done->frame_pos + done->type->size;
2022 pos = ft->function.local_size;
2023 if (pos & (t->align - 1))
2024 pos = (pos + t->align) & ~(t->align-1);
2026 if (size < pos + v->type->size)
2027 size = pos + v->type->size;
2031 c->out_scope = NULL;
2032 ft->function.local_size = size;
2035 ###### free context storage
2036 free(context.global);
2038 #### Variables as executables
2040 Just as we used a `val` to wrap a value into an `exec`, we similarly
2041 need a `var` to wrap a `variable` into an exec. While each `val`
2042 contained a copy of the value, each `var` holds a link to the variable
2043 because it really is the same variable no matter where it appears.
2044 When a variable is used, we need to remember to follow the `->merged`
2045 link to find the primary instance.
2047 When a variable is declared, it may or may not be given an explicit
2048 type. We need to record which so that we can report the parsed code
2057 struct variable *var;
2060 ###### variable fields
2068 VariableDecl -> IDENTIFIER : ${ {
2069 struct variable *v = var_decl(c, $1.txt);
2070 $0 = new_pos(var, $1);
2075 v = var_ref(c, $1.txt);
2077 type_err(c, "error: variable '%v' redeclared",
2079 type_err(c, "info: this is where '%v' was first declared",
2080 v->where_decl, NULL, 0, NULL);
2083 | IDENTIFIER :: ${ {
2084 struct variable *v = var_decl(c, $1.txt);
2085 $0 = new_pos(var, $1);
2091 v = var_ref(c, $1.txt);
2093 type_err(c, "error: variable '%v' redeclared",
2095 type_err(c, "info: this is where '%v' was first declared",
2096 v->where_decl, NULL, 0, NULL);
2099 | IDENTIFIER : Type ${ {
2100 struct variable *v = var_decl(c, $1.txt);
2101 $0 = new_pos(var, $1);
2107 v->explicit_type = 1;
2109 v = var_ref(c, $1.txt);
2111 type_err(c, "error: variable '%v' redeclared",
2113 type_err(c, "info: this is where '%v' was first declared",
2114 v->where_decl, NULL, 0, NULL);
2117 | IDENTIFIER :: Type ${ {
2118 struct variable *v = var_decl(c, $1.txt);
2119 $0 = new_pos(var, $1);
2126 v->explicit_type = 1;
2128 v = var_ref(c, $1.txt);
2130 type_err(c, "error: variable '%v' redeclared",
2132 type_err(c, "info: this is where '%v' was first declared",
2133 v->where_decl, NULL, 0, NULL);
2138 Variable -> IDENTIFIER ${ {
2139 struct variable *v = var_ref(c, $1.txt);
2140 $0 = new_pos(var, $1);
2142 /* This might be a global const or a label
2143 * Allocate a var with impossible type Tnone,
2144 * which will be adjusted when we find out what it is,
2145 * or will trigger an error.
2147 v = var_decl(c, $1.txt);
2154 cast(var, $0)->var = v;
2157 ###### print exec cases
2160 struct var *v = cast(var, e);
2162 struct binding *b = v->var->name;
2163 printf("%.*s", b->name.len, b->name.txt);
2170 if (loc && loc->type == Xvar) {
2171 struct var *v = cast(var, loc);
2173 struct binding *b = v->var->name;
2174 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2176 fputs("???", stderr); // NOTEST
2178 fputs("NOTVAR", stderr); // NOTEST
2181 ###### propagate exec cases
2185 struct var *var = cast(var, prog);
2186 struct variable *v = var->var;
2188 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2189 return Tnone; // NOTEST
2192 if (v->constant && (rules & Rnoconstant)) {
2193 type_err(c, "error: Cannot assign to a constant: %v",
2194 prog, NULL, 0, NULL);
2195 type_err(c, "info: name was defined as a constant here",
2196 v->where_decl, NULL, 0, NULL);
2199 if (v->type == Tnone && v->where_decl == prog)
2200 type_err(c, "error: variable used but not declared: %v",
2201 prog, NULL, 0, NULL);
2202 if (v->type == NULL) {
2203 if (type && !(*perr & Efail)) {
2205 v->where_set = prog;
2208 } else if (!type_compat(type, v->type, rules)) {
2209 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2210 type, rules, v->type);
2211 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2212 v->type, rules, NULL);
2214 if (!v->global || v->frame_pos < 0)
2221 ###### interp exec cases
2224 struct var *var = cast(var, e);
2225 struct variable *v = var->var;
2228 lrv = var_value(c, v);
2233 ###### ast functions
2235 static void free_var(struct var *v)
2240 ###### free exec cases
2241 case Xvar: free_var(cast(var, e)); break;
2246 Now that we have the shape of the interpreter in place we can add some
2247 complex types and connected them in to the data structures and the
2248 different phases of parse, analyse, print, interpret.
2250 Being "complex" the language will naturally have syntax to access
2251 specifics of objects of these types. These will fit into the grammar as
2252 "Terms" which are the things that are combined with various operators to
2253 form "Expression". Where a Term is formed by some operation on another
2254 Term, the subordinate Term will always come first, so for example a
2255 member of an array will be expressed as the Term for the array followed
2256 by an index in square brackets. The strict rule of using postfix
2257 operations makes precedence irrelevant within terms. To provide a place
2258 to put the grammar for each terms of each type, we will start out by
2259 introducing the "Term" grammar production, with contains at least a
2260 simple "Value" (to be explained later).
2264 Term -> Value ${ $0 = $<1; }$
2265 | Variable ${ $0 = $<1; }$
2268 Thus far the complex types we have are arrays and structs.
2272 Arrays can be declared by giving a size and a type, as `[size]type' so
2273 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2274 size can be either a literal number, or a named constant. Some day an
2275 arbitrary expression will be supported.
2277 As a formal parameter to a function, the array can be declared with a
2278 new variable as the size: `name:[size::number]string`. The `size`
2279 variable is set to the size of the array and must be a constant. As
2280 `number` is the only supported type, it can be left out:
2281 `name:[size::]string`.
2283 Arrays cannot be assigned. When pointers are introduced we will also
2284 introduce array slices which can refer to part or all of an array -
2285 the assignment syntax will create a slice. For now, an array can only
2286 ever be referenced by the name it is declared with. It is likely that
2287 a "`copy`" primitive will eventually be define which can be used to
2288 make a copy of an array with controllable recursive depth.
2290 For now we have two sorts of array, those with fixed size either because
2291 it is given as a literal number or because it is a struct member (which
2292 cannot have a runtime-changing size), and those with a size that is
2293 determined at runtime - local variables with a const size. The former
2294 have their size calculated at parse time, the latter at run time.
2296 For the latter type, the `size` field of the type is the size of a
2297 pointer, and the array is reallocated every time it comes into scope.
2299 We differentiate struct fields with a const size from local variables
2300 with a const size by whether they are prepared at parse time or not.
2302 ###### type union fields
2305 int unspec; // size is unspecified - vsize must be set.
2308 struct variable *vsize;
2309 struct type *member;
2312 ###### value union fields
2313 void *array; // used if not static_size
2315 ###### value functions
2317 static int array_prepare_type(struct parse_context *c, struct type *type,
2320 struct value *vsize;
2322 if (type->array.static_size)
2323 return 1; // UNTESTED
2324 if (type->array.unspec && parse_time)
2325 return 1; // UNTESTED
2326 if (parse_time && type->array.vsize && !type->array.vsize->global)
2327 return 1; // UNTESTED
2329 if (type->array.vsize) {
2330 vsize = var_value(c, type->array.vsize);
2332 return 1; // UNTESTED
2334 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2335 type->array.size = mpz_get_si(q);
2340 if (type->array.member->size <= 0)
2341 return 0; // UNTESTED
2343 type->array.static_size = 1;
2344 type->size = type->array.size * type->array.member->size;
2345 type->align = type->array.member->align;
2350 static void array_init(struct type *type, struct value *val)
2353 void *ptr = val->ptr;
2357 if (!type->array.static_size) {
2358 val->array = calloc(type->array.size,
2359 type->array.member->size);
2362 for (i = 0; i < type->array.size; i++) {
2364 v = (void*)ptr + i * type->array.member->size;
2365 val_init(type->array.member, v);
2369 static void array_free(struct type *type, struct value *val)
2372 void *ptr = val->ptr;
2374 if (!type->array.static_size)
2376 for (i = 0; i < type->array.size; i++) {
2378 v = (void*)ptr + i * type->array.member->size;
2379 free_value(type->array.member, v);
2381 if (!type->array.static_size)
2385 static int array_compat(struct type *require, struct type *have)
2387 if (have->compat != require->compat)
2389 /* Both are arrays, so we can look at details */
2390 if (!type_compat(require->array.member, have->array.member, 0))
2392 if (have->array.unspec && require->array.unspec) {
2393 if (have->array.vsize && require->array.vsize &&
2394 have->array.vsize != require->array.vsize) // UNTESTED
2395 /* sizes might not be the same */
2396 return 0; // UNTESTED
2399 if (have->array.unspec || require->array.unspec)
2400 return 1; // UNTESTED
2401 if (require->array.vsize == NULL && have->array.vsize == NULL)
2402 return require->array.size == have->array.size;
2404 return require->array.vsize == have->array.vsize; // UNTESTED
2407 static void array_print_type(struct type *type, FILE *f)
2410 if (type->array.vsize) {
2411 struct binding *b = type->array.vsize->name;
2412 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2413 type->array.unspec ? "::" : "");
2414 } else if (type->array.size)
2415 fprintf(f, "%d]", type->array.size);
2418 type_print(type->array.member, f);
2421 static struct type array_prototype = {
2423 .prepare_type = array_prepare_type,
2424 .print_type = array_print_type,
2425 .compat = array_compat,
2427 .size = sizeof(void*),
2428 .align = sizeof(void*),
2431 ###### declare terminals
2436 | [ NUMBER ] Type ${ {
2442 if (number_parse(num, tail, $2.txt) == 0)
2443 tok_err(c, "error: unrecognised number", &$2);
2445 tok_err(c, "error: unsupported number suffix", &$2);
2448 elements = mpz_get_ui(mpq_numref(num));
2449 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2450 tok_err(c, "error: array size must be an integer",
2452 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2453 tok_err(c, "error: array size is too large",
2458 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2459 t->array.size = elements;
2460 t->array.member = $<4;
2461 t->array.vsize = NULL;
2464 | [ IDENTIFIER ] Type ${ {
2465 struct variable *v = var_ref(c, $2.txt);
2468 tok_err(c, "error: name undeclared", &$2);
2469 else if (!v->constant)
2470 tok_err(c, "error: array size must be a constant", &$2);
2472 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2473 $0->array.member = $<4;
2475 $0->array.vsize = v;
2480 OptType -> Type ${ $0 = $<1; }$
2483 ###### formal type grammar
2485 | [ IDENTIFIER :: OptType ] Type ${ {
2486 struct variable *v = var_decl(c, $ID.txt);
2492 $0 = add_anon_type(c, &array_prototype, "array[var]");
2493 $0->array.member = $<6;
2495 $0->array.unspec = 1;
2496 $0->array.vsize = v;
2504 | Term [ Expression ] ${ {
2505 struct binode *b = new(binode);
2512 ###### print binode cases
2514 print_exec(b->left, -1, bracket);
2516 print_exec(b->right, -1, bracket);
2520 ###### propagate binode cases
2522 /* left must be an array, right must be a number,
2523 * result is the member type of the array
2525 propagate_types(b->right, c, perr, Tnum, 0);
2526 t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
2527 if (!t || t->compat != array_compat) {
2528 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2531 if (!type_compat(type, t->array.member, rules)) {
2532 type_err(c, "error: have %1 but need %2", prog,
2533 t->array.member, rules, type);
2535 return t->array.member;
2539 ###### interp binode cases
2545 lleft = linterp_exec(c, b->left, <ype);
2546 right = interp_exec(c, b->right, &rtype);
2548 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2552 if (ltype->array.static_size)
2555 ptr = *(void**)lleft;
2556 rvtype = ltype->array.member;
2557 if (i >= 0 && i < ltype->array.size)
2558 lrv = ptr + i * rvtype->size;
2560 val_init(ltype->array.member, &rv); // UNSAFE
2567 A `struct` is a data-type that contains one or more other data-types.
2568 It differs from an array in that each member can be of a different
2569 type, and they are accessed by name rather than by number. Thus you
2570 cannot choose an element by calculation, you need to know what you
2573 The language makes no promises about how a given structure will be
2574 stored in memory - it is free to rearrange fields to suit whatever
2575 criteria seems important.
2577 Structs are declared separately from program code - they cannot be
2578 declared in-line in a variable declaration like arrays can. A struct
2579 is given a name and this name is used to identify the type - the name
2580 is not prefixed by the word `struct` as it would be in C.
2582 Structs are only treated as the same if they have the same name.
2583 Simply having the same fields in the same order is not enough. This
2584 might change once we can create structure initializers from a list of
2587 Each component datum is identified much like a variable is declared,
2588 with a name, one or two colons, and a type. The type cannot be omitted
2589 as there is no opportunity to deduce the type from usage. An initial
2590 value can be given following an equals sign, so
2592 ##### Example: a struct type
2598 would declare a type called "complex" which has two number fields,
2599 each initialised to zero.
2601 Struct will need to be declared separately from the code that uses
2602 them, so we will need to be able to print out the declaration of a
2603 struct when reprinting the whole program. So a `print_type_decl` type
2604 function will be needed.
2606 ###### type union fields
2615 } *fields; // This is created when field_list is analysed.
2617 struct fieldlist *prev;
2620 } *field_list; // This is created during parsing
2623 ###### type functions
2624 void (*print_type_decl)(struct type *type, FILE *f);
2625 struct type *(*fieldref)(struct type *t, struct parse_context *c,
2626 struct fieldref *f, struct value **vp);
2628 ###### value functions
2630 static void structure_init(struct type *type, struct value *val)
2634 for (i = 0; i < type->structure.nfields; i++) {
2636 v = (void*) val->ptr + type->structure.fields[i].offset;
2637 if (type->structure.fields[i].init)
2638 dup_value(type->structure.fields[i].type,
2639 type->structure.fields[i].init,
2642 val_init(type->structure.fields[i].type, v);
2646 static void structure_free(struct type *type, struct value *val)
2650 for (i = 0; i < type->structure.nfields; i++) {
2652 v = (void*)val->ptr + type->structure.fields[i].offset;
2653 free_value(type->structure.fields[i].type, v);
2657 static void free_fieldlist(struct fieldlist *f)
2661 free_fieldlist(f->prev);
2666 static void structure_free_type(struct type *t)
2669 for (i = 0; i < t->structure.nfields; i++)
2670 if (t->structure.fields[i].init) {
2671 free_value(t->structure.fields[i].type,
2672 t->structure.fields[i].init);
2674 free(t->structure.fields);
2675 free_fieldlist(t->structure.field_list);
2678 static int structure_prepare_type(struct parse_context *c,
2679 struct type *t, int parse_time)
2682 struct fieldlist *f;
2684 if (!parse_time || t->structure.fields)
2687 for (f = t->structure.field_list; f; f=f->prev) {
2691 if (f->f.type->size <= 0)
2693 if (f->f.type->prepare_type)
2694 f->f.type->prepare_type(c, f->f.type, parse_time);
2696 if (f->init == NULL)
2700 propagate_types(f->init, c, &perr, f->f.type, 0);
2701 } while (perr & Eretry);
2703 c->parse_error += 1; // NOTEST
2706 t->structure.nfields = cnt;
2707 t->structure.fields = calloc(cnt, sizeof(struct field));
2708 f = t->structure.field_list;
2710 int a = f->f.type->align;
2712 t->structure.fields[cnt] = f->f;
2713 if (t->size & (a-1))
2714 t->size = (t->size | (a-1)) + 1;
2715 t->structure.fields[cnt].offset = t->size;
2716 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2720 if (f->init && !c->parse_error) {
2721 struct value vl = interp_exec(c, f->init, NULL);
2722 t->structure.fields[cnt].init =
2723 global_alloc(c, f->f.type, NULL, &vl);
2731 static int find_struct_index(struct type *type, struct text field)
2734 for (i = 0; i < type->structure.nfields; i++)
2735 if (text_cmp(type->structure.fields[i].name, field) == 0)
2737 return IndexInvalid;
2740 static struct type *structure_fieldref(struct type *t, struct parse_context *c,
2741 struct fieldref *f, struct value **vp)
2743 if (f->index == IndexUnknown) {
2744 f->index = find_struct_index(t, f->name);
2746 type_err(c, "error: cannot find requested field in %1",
2747 f->left, t, 0, NULL);
2752 struct value *v = *vp;
2753 v = (void*)v->ptr + t->structure.fields[f->index].offset;
2756 return t->structure.fields[f->index].type;
2759 static struct type structure_prototype = {
2760 .init = structure_init,
2761 .free = structure_free,
2762 .free_type = structure_free_type,
2763 .print_type_decl = structure_print_type,
2764 .prepare_type = structure_prepare_type,
2765 .fieldref = structure_fieldref,
2778 enum { IndexUnknown = -1, IndexInvalid = -2 };
2780 ###### free exec cases
2782 free_exec(cast(fieldref, e)->left);
2786 ###### declare terminals
2791 | Term . IDENTIFIER ${ {
2792 struct fieldref *fr = new_pos(fieldref, $2);
2795 fr->index = IndexUnknown;
2799 ###### print exec cases
2803 struct fieldref *f = cast(fieldref, e);
2804 print_exec(f->left, -1, bracket);
2805 printf(".%.*s", f->name.len, f->name.txt);
2809 ###### propagate exec cases
2813 struct fieldref *f = cast(fieldref, prog);
2814 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2816 if (!st || !st->fieldref)
2817 type_err(c, "error: field reference on %1 is not supported",
2818 f->left, st, 0, NULL);
2820 t = st->fieldref(st, c, f, NULL);
2821 if (t && !type_compat(type, t, rules))
2822 type_err(c, "error: have %1 but need %2", prog,
2829 ###### interp exec cases
2832 struct fieldref *f = cast(fieldref, e);
2834 struct value *lleft = linterp_exec(c, f->left, <ype);
2836 rvtype = ltype->fieldref(ltype, c, f, &lrv);
2840 ###### top level grammar
2841 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2843 t = find_type(c, $ID.txt);
2845 t = add_type(c, $ID.txt, &structure_prototype);
2846 else if (t->size >= 0) {
2847 tok_err(c, "error: type already declared", &$ID);
2848 tok_err(c, "info: this is location of declartion", &t->first_use);
2849 /* Create a new one - duplicate */
2850 t = add_type(c, $ID.txt, &structure_prototype);
2852 struct type tmp = *t;
2853 *t = structure_prototype;
2857 t->structure.field_list = $<FB;
2862 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2863 | { SimpleFieldList } ${ $0 = $<SFL; }$
2864 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2865 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2867 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2868 | FieldLines SimpleFieldList Newlines ${
2873 SimpleFieldList -> Field ${ $0 = $<F; }$
2874 | SimpleFieldList ; Field ${
2878 | SimpleFieldList ; ${
2881 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2883 Field -> IDENTIFIER : Type = Expression ${ {
2884 $0 = calloc(1, sizeof(struct fieldlist));
2885 $0->f.name = $ID.txt;
2886 $0->f.type = $<Type;
2890 | IDENTIFIER : Type ${
2891 $0 = calloc(1, sizeof(struct fieldlist));
2892 $0->f.name = $ID.txt;
2893 $0->f.type = $<Type;
2896 ###### forward decls
2897 static void structure_print_type(struct type *t, FILE *f);
2899 ###### value functions
2900 static void structure_print_type(struct type *t, FILE *f)
2904 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2906 for (i = 0; i < t->structure.nfields; i++) {
2907 struct field *fl = t->structure.fields + i;
2908 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2909 type_print(fl->type, f);
2910 if (fl->type->print && fl->init) {
2912 if (fl->type == Tstr)
2913 fprintf(f, "\""); // UNTESTED
2914 print_value(fl->type, fl->init, f);
2915 if (fl->type == Tstr)
2916 fprintf(f, "\""); // UNTESTED
2922 ###### print type decls
2927 while (target != 0) {
2929 for (t = context.typelist; t ; t=t->next)
2930 if (!t->anon && t->print_type_decl &&
2940 t->print_type_decl(t, stdout);
2948 References, or pointers, are values that refer to another value. They
2949 can only refer to a `struct`, though as a struct can embed anything they
2950 can effectively refer to anything.
2952 References are potentially dangerous as they might refer to some
2953 variable which no longer exists - either because a stack frame
2954 containing it has been discarded or because the value was allocated on
2955 the heap and has now been free. Ocean does not yet provide any
2956 protection against these problems. It will in due course.
2958 With references comes the opportunity and the need to explicitly
2959 allocate values on the "heap" and to free them. We currently provide
2960 fairly basic support for this.
2962 Reference make use of the `@` symbol in various ways. A type that starts
2963 with `@` is a reference to whatever follows. A reference value
2964 followed by an `@` acts as the referred value, though the `@` is often
2965 not needed. Finally, an expression that starts with `@` is a special
2966 reference related expression. Some examples might help.
2968 ##### Example: Reference examples
2975 bar.number = 23; bar.string = "hello"
2986 Obviously this is very contrived. `ref` is a reference to a `foo` which
2987 is initially set to refer to the value stored in `bar` - no extra syntax
2988 is needed to "Take the address of" `bar` - the fact that `ref` is a
2989 reference means that only the address make sense.
2991 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
2992 The same syntax is used for accessing fields both in structs and in
2993 references to structs. It would be correct to use `ref@.a`, but not
2996 `@new()` creates an object of whatever type is needed for the program
2997 to by type-correct. In future iterations of Ocean, arguments a
2998 constructor will access arguments, so the the syntax now looks like a
2999 function call. `@free` can be assigned any reference that was returned
3000 by `@new()`, and it will be freed. `@nil` is a value of whatever
3001 reference type is appropriate, and is stable and never the address of
3002 anything in the heap or on the stack. A reference can be assigned
3003 `@nil` or compared against that value.
3005 ###### declare terminals
3008 ###### type union fields
3011 struct type *referent;
3014 ###### value union fields
3017 ###### value functions
3019 static void reference_print_type(struct type *t, FILE *f)
3022 type_print(t->reference.referent, f);
3025 static int reference_cmp(struct type *tl, struct type *tr,
3026 struct value *left, struct value *right)
3028 return left->ref == right->ref ? 0 : 1;
3031 static void reference_dup(struct type *t,
3032 struct value *vold, struct value *vnew)
3034 vnew->ref = vold->ref;
3037 static void reference_free(struct type *t, struct value *v)
3039 /* Nothing to do here */
3042 static int reference_compat(struct type *require, struct type *have)
3044 if (have->compat != require->compat)
3046 if (have->reference.referent != require->reference.referent)
3051 static int reference_test(struct type *type, struct value *val)
3053 return val->ref != NULL;
3056 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3057 struct fieldref *f, struct value **vp)
3059 struct type *rt = t->reference.referent;
3064 return rt->fieldref(rt, c, f, vp);
3066 type_err(c, "error: field reference on %1 is not supported",
3067 f->left, rt, 0, NULL);
3072 static struct type reference_prototype = {
3073 .print_type = reference_print_type,
3074 .cmp_eq = reference_cmp,
3075 .dup = reference_dup,
3076 .test = reference_test,
3077 .free = reference_free,
3078 .compat = reference_compat,
3079 .fieldref = reference_fieldref,
3080 .size = sizeof(void*),
3081 .align = sizeof(void*),
3087 struct type *t = find_type(c, $ID.txt);
3089 t = add_type(c, $ID.txt, NULL);
3092 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3093 $ID.txt.len, $ID.txt.txt);
3094 $0->reference.referent = t;
3097 ###### core functions
3098 static int text_is(struct text t, char *s)
3100 return (strlen(s) == t.len &&
3101 strncmp(s, t.txt, t.len) == 0);
3110 enum ref_func { RefNew, RefFree, RefNil } action;
3111 struct type *reftype;
3115 ###### SimpleStatement Grammar
3117 | @ IDENTIFIER = Expression ${ {
3118 struct ref *r = new_pos(ref, $ID);
3120 if (!text_is($ID.txt, "free"))
3121 tok_err(c, "error: only \"@free\" makes sense here",
3125 r->action = RefFree;
3129 ###### expression grammar
3130 | @ IDENTIFIER ( ) ${
3131 // Only 'new' valid here
3132 if (!text_is($ID.txt, "new")) {
3133 tok_err(c, "error: Only reference function is \"@new()\"",
3136 struct ref *r = new_pos(ref,$ID);
3142 // Only 'nil' valid here
3143 if (!text_is($ID.txt, "nil")) {
3144 tok_err(c, "error: Only reference value is \"@nil\"",
3147 struct ref *r = new_pos(ref,$ID);
3153 ###### print exec cases
3155 struct ref *r = cast(ref, e);
3156 switch (r->action) {
3158 printf("@new()"); break;
3160 printf("@nil"); break;
3162 do_indent(indent, "@free = ");
3163 print_exec(r->right, indent, bracket);
3169 ###### propagate exec cases
3171 struct ref *r = cast(ref, prog);
3172 switch (r->action) {
3174 if (type && type->free != reference_free) {
3175 type_err(c, "error: @new() can only be used with references, not %1",
3176 prog, type, 0, NULL);
3179 if (type && !r->reftype) {
3185 if (type && type->free != reference_free)
3186 type_err(c, "error: @nil can only be used with reference, not %1",
3187 prog, type, 0, NULL);
3188 if (type && !r->reftype) {
3194 t = propagate_types(r->right, c, perr, NULL, 0);
3195 if (t && t->free != reference_free)
3196 type_err(c, "error: @free can only be assigned a reference, not %1",
3205 ###### interp exec cases
3207 struct ref *r = cast(ref, e);
3208 switch (r->action) {
3211 rv.ref = calloc(1, r->reftype->reference.referent->size);
3212 rvtype = r->reftype;
3216 rvtype = r->reftype;
3219 rv = interp_exec(c, r->right, &rvtype);
3220 free_value(rvtype->reference.referent, rv.ref);
3228 ###### free exec cases
3230 struct ref *r = cast(ref, e);
3231 free_exec(r->right);
3236 ###### Expressions: dereference
3244 struct binode *b = new(binode);
3250 ###### print binode cases
3252 print_exec(b->left, -1, bracket);
3256 ###### propagate binode cases
3258 /* left must be a reference, and we return what it refers to */
3259 /* FIXME how can I pass the expected type down? */
3260 t = propagate_types(b->left, c, perr, NULL, 0);
3261 if (!t || t->free != reference_free)
3262 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3264 return t->reference.referent;
3267 ###### interp binode cases
3269 left = interp_exec(c, b->left, <ype);
3271 rvtype = ltype->reference.referent;
3278 A function is a chunk of code which can be passed parameters and can
3279 return results. Each function has a type which includes the set of
3280 parameters and the return value. As yet these types cannot be declared
3281 separately from the function itself.
3283 The parameters can be specified either in parentheses as a ';' separated
3286 ##### Example: function 1
3288 func main(av:[ac::number]string; env:[envc::number]string)
3291 or as an indented list of one parameter per line (though each line can
3292 be a ';' separated list)
3294 ##### Example: function 2
3297 argv:[argc::number]string
3298 env:[envc::number]string
3302 In the first case a return type can follow the parentheses after a colon,
3303 in the second it is given on a line starting with the word `return`.
3305 ##### Example: functions that return
3307 func add(a:number; b:number): number
3317 Rather than returning a type, the function can specify a set of local
3318 variables to return as a struct. The values of these variables when the
3319 function exits will be provided to the caller. For this the return type
3320 is replaced with a block of result declarations, either in parentheses
3321 or bracketed by `return` and `do`.
3323 ##### Example: functions returning multiple variables
3325 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3338 For constructing the lists we use a `List` binode, which will be
3339 further detailed when Expression Lists are introduced.
3341 ###### type union fields
3344 struct binode *params;
3345 struct type *return_type;
3346 struct variable *scope;
3347 int inline_result; // return value is at start of 'local'
3351 ###### value union fields
3352 struct exec *function;
3354 ###### type functions
3355 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3356 struct type *require, struct exec *args);
3358 ###### value functions
3360 static void function_free(struct type *type, struct value *val)
3362 free_exec(val->function);
3363 val->function = NULL;
3366 static int function_compat(struct type *require, struct type *have)
3368 // FIXME can I do anything here yet?
3372 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3373 struct type *require, struct exec *args)
3375 /* This should be 'compat', but we don't have a 'tuple' type to
3376 * hold the type of 'args'
3378 struct binode *arg = cast(binode, args);
3379 struct binode *param = require->function.params;
3382 struct var *pv = cast(var, param->left);
3384 type_err(c, "error: insufficient arguments to function.",
3385 args, NULL, 0, NULL);
3389 propagate_types(arg->left, c, perr, pv->var->type, 0);
3390 param = cast(binode, param->right);
3391 arg = cast(binode, arg->right);
3394 type_err(c, "error: too many arguments to function.",
3395 args, NULL, 0, NULL);
3398 static void function_print(struct type *type, struct value *val, FILE *f)
3400 print_exec(val->function, 1, 0);
3403 static void function_print_type_decl(struct type *type, FILE *f)
3407 for (b = type->function.params; b; b = cast(binode, b->right)) {
3408 struct variable *v = cast(var, b->left)->var;
3409 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3410 v->constant ? "::" : ":");
3411 type_print(v->type, f);
3416 if (type->function.return_type != Tnone) {
3418 if (type->function.inline_result) {
3420 struct type *t = type->function.return_type;
3422 for (i = 0; i < t->structure.nfields; i++) {
3423 struct field *fl = t->structure.fields + i;
3426 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3427 type_print(fl->type, f);
3431 type_print(type->function.return_type, f);
3436 static void function_free_type(struct type *t)
3438 free_exec(t->function.params);
3441 static struct type function_prototype = {
3442 .size = sizeof(void*),
3443 .align = sizeof(void*),
3444 .free = function_free,
3445 .compat = function_compat,
3446 .check_args = function_check_args,
3447 .print = function_print,
3448 .print_type_decl = function_print_type_decl,
3449 .free_type = function_free_type,
3452 ###### declare terminals
3462 FuncName -> IDENTIFIER ${ {
3463 struct variable *v = var_decl(c, $1.txt);
3464 struct var *e = new_pos(var, $1);
3471 v = var_ref(c, $1.txt);
3473 type_err(c, "error: function '%v' redeclared",
3475 type_err(c, "info: this is where '%v' was first declared",
3476 v->where_decl, NULL, 0, NULL);
3482 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3483 | Args ArgsLine NEWLINE ${ {
3484 struct binode *b = $<AL;
3485 struct binode **bp = &b;
3487 bp = (struct binode **)&(*bp)->left;
3492 ArgsLine -> ${ $0 = NULL; }$
3493 | Varlist ${ $0 = $<1; }$
3494 | Varlist ; ${ $0 = $<1; }$
3496 Varlist -> Varlist ; ArgDecl ${
3497 $0 = new_pos(binode, $2);
3510 ArgDecl -> IDENTIFIER : FormalType ${ {
3511 struct variable *v = var_decl(c, $ID.txt);
3512 $0 = new_pos(var, $ID);
3519 ##### Function calls
3521 A function call can appear either as an expression or as a statement.
3522 We use a new 'Funcall' binode type to link the function with a list of
3523 arguments, form with the 'List' nodes.
3525 We have already seen the "Term" which is how a function call can appear
3526 in an expression. To parse a function call into a statement we include
3527 it in the "SimpleStatement Grammar" which will be described later.
3533 | Term ( ExpressionList ) ${ {
3534 struct binode *b = new(binode);
3537 b->right = reorder_bilist($<EL);
3541 struct binode *b = new(binode);
3548 ###### SimpleStatement Grammar
3550 | Term ( ExpressionList ) ${ {
3551 struct binode *b = new(binode);
3554 b->right = reorder_bilist($<EL);
3558 ###### print binode cases
3561 do_indent(indent, "");
3562 print_exec(b->left, -1, bracket);
3564 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3567 print_exec(b->left, -1, bracket);
3577 ###### propagate binode cases
3580 /* Every arg must match formal parameter, and result
3581 * is return type of function
3583 struct binode *args = cast(binode, b->right);
3584 struct var *v = cast(var, b->left);
3586 if (!v->var->type || v->var->type->check_args == NULL) {
3587 type_err(c, "error: attempt to call a non-function.",
3588 prog, NULL, 0, NULL);
3592 v->var->type->check_args(c, perr, v->var->type, args);
3593 if (v->var->type->function.inline_result)
3595 return v->var->type->function.return_type;
3598 ###### interp binode cases
3601 struct var *v = cast(var, b->left);
3602 struct type *t = v->var->type;
3603 void *oldlocal = c->local;
3604 int old_size = c->local_size;
3605 void *local = calloc(1, t->function.local_size);
3606 struct value *fbody = var_value(c, v->var);
3607 struct binode *arg = cast(binode, b->right);
3608 struct binode *param = t->function.params;
3611 struct var *pv = cast(var, param->left);
3612 struct type *vtype = NULL;
3613 struct value val = interp_exec(c, arg->left, &vtype);
3615 c->local = local; c->local_size = t->function.local_size;
3616 lval = var_value(c, pv->var);
3617 c->local = oldlocal; c->local_size = old_size;
3618 memcpy(lval, &val, vtype->size);
3619 param = cast(binode, param->right);
3620 arg = cast(binode, arg->right);
3622 c->local = local; c->local_size = t->function.local_size;
3623 if (t->function.inline_result && dtype) {
3624 _interp_exec(c, fbody->function, NULL, NULL);
3625 memcpy(dest, local, dtype->size);
3626 rvtype = ret.type = NULL;
3628 rv = interp_exec(c, fbody->function, &rvtype);
3629 c->local = oldlocal; c->local_size = old_size;
3634 ## Complex executables: statements and expressions
3636 Now that we have types and values and variables and most of the basic
3637 Terms which provide access to these, we can explore the more complex
3638 code that combine all of these to get useful work done. Specifically
3639 statements and expressions.
3641 Expressions are various combinations of Terms. We will use operator
3642 precedence to ensure correct parsing. The simplest Expression is just a
3643 Term - others will follow.
3648 Expression -> Term ${ $0 = $<Term; }$
3649 ## expression grammar
3651 ### Expressions: Conditional
3653 Our first user of the `binode` will be conditional expressions, which
3654 is a bit odd as they actually have three components. That will be
3655 handled by having 2 binodes for each expression. The conditional
3656 expression is the lowest precedence operator which is why we define it
3657 first - to start the precedence list.
3659 Conditional expressions are of the form "value `if` condition `else`
3660 other_value". They associate to the right, so everything to the right
3661 of `else` is part of an else value, while only a higher-precedence to
3662 the left of `if` is the if values. Between `if` and `else` there is no
3663 room for ambiguity, so a full conditional expression is allowed in
3669 ###### declare terminals
3673 ###### expression grammar
3675 | Expression if Expression else Expression $$ifelse ${ {
3676 struct binode *b1 = new(binode);
3677 struct binode *b2 = new(binode);
3687 ###### print binode cases
3690 b2 = cast(binode, b->right);
3691 if (bracket) printf("(");
3692 print_exec(b2->left, -1, bracket);
3694 print_exec(b->left, -1, bracket);
3696 print_exec(b2->right, -1, bracket);
3697 if (bracket) printf(")");
3700 ###### propagate binode cases
3703 /* cond must be Tbool, others must match */
3704 struct binode *b2 = cast(binode, b->right);
3707 propagate_types(b->left, c, perr, Tbool, 0);
3708 t = propagate_types(b2->left, c, perr, type, Rnolabel);
3709 t2 = propagate_types(b2->right, c, perr, type ?: t, Rnolabel);
3713 ###### interp binode cases
3716 struct binode *b2 = cast(binode, b->right);
3717 left = interp_exec(c, b->left, <ype);
3719 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3721 rv = interp_exec(c, b2->right, &rvtype);
3727 We take a brief detour, now that we have expressions, to describe lists
3728 of expressions. These will be needed for function parameters and
3729 possibly other situations. They seem generic enough to introduce here
3730 to be used elsewhere.
3732 And ExpressionList will use the `List` type of `binode`, building up at
3733 the end. And place where they are used will probably call
3734 `reorder_bilist()` to get a more normal first/next arrangement.
3736 ###### declare terminals
3739 `List` execs have no implicit semantics, so they are never propagated or
3740 interpreted. The can be printed as a comma separate list, which is how
3741 they are parsed. Note they are also used for function formal parameter
3742 lists. In that case a separate function is used to print them.
3744 ###### print binode cases
3748 print_exec(b->left, -1, bracket);
3751 b = cast(binode, b->right);
3755 ###### propagate binode cases
3756 case List: abort(); // NOTEST
3757 ###### interp binode cases
3758 case List: abort(); // NOTEST
3763 ExpressionList -> ExpressionList , Expression ${
3776 ### Expressions: Boolean
3778 The next class of expressions to use the `binode` will be Boolean
3779 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3780 have same corresponding precendence. The difference is that they don't
3781 evaluate the second expression if not necessary.
3790 ###### declare terminals
3795 ###### expression grammar
3796 | Expression or Expression ${ {
3797 struct binode *b = new(binode);
3803 | Expression or else Expression ${ {
3804 struct binode *b = new(binode);
3811 | Expression and Expression ${ {
3812 struct binode *b = new(binode);
3818 | Expression and then Expression ${ {
3819 struct binode *b = new(binode);
3826 | not Expression ${ {
3827 struct binode *b = new(binode);
3833 ###### print binode cases
3835 if (bracket) printf("(");
3836 print_exec(b->left, -1, bracket);
3838 print_exec(b->right, -1, bracket);
3839 if (bracket) printf(")");
3842 if (bracket) printf("(");
3843 print_exec(b->left, -1, bracket);
3844 printf(" and then ");
3845 print_exec(b->right, -1, bracket);
3846 if (bracket) printf(")");
3849 if (bracket) printf("(");
3850 print_exec(b->left, -1, bracket);
3852 print_exec(b->right, -1, bracket);
3853 if (bracket) printf(")");
3856 if (bracket) printf("(");
3857 print_exec(b->left, -1, bracket);
3858 printf(" or else ");
3859 print_exec(b->right, -1, bracket);
3860 if (bracket) printf(")");
3863 if (bracket) printf("(");
3865 print_exec(b->right, -1, bracket);
3866 if (bracket) printf(")");
3869 ###### propagate binode cases
3875 /* both must be Tbool, result is Tbool */
3876 propagate_types(b->left, c, perr, Tbool, 0);
3877 propagate_types(b->right, c, perr, Tbool, 0);
3878 if (type && type != Tbool)
3879 type_err(c, "error: %1 operation found where %2 expected", prog,
3883 ###### interp binode cases
3885 rv = interp_exec(c, b->left, &rvtype);
3886 right = interp_exec(c, b->right, &rtype);
3887 rv.bool = rv.bool && right.bool;
3890 rv = interp_exec(c, b->left, &rvtype);
3892 rv = interp_exec(c, b->right, NULL);
3895 rv = interp_exec(c, b->left, &rvtype);
3896 right = interp_exec(c, b->right, &rtype);
3897 rv.bool = rv.bool || right.bool;
3900 rv = interp_exec(c, b->left, &rvtype);
3902 rv = interp_exec(c, b->right, NULL);
3905 rv = interp_exec(c, b->right, &rvtype);
3909 ### Expressions: Comparison
3911 Of slightly higher precedence that Boolean expressions are Comparisons.
3912 A comparison takes arguments of any comparable type, but the two types
3915 To simplify the parsing we introduce an `eop` which can record an
3916 expression operator, and the `CMPop` non-terminal will match one of them.
3923 ###### ast functions
3924 static void free_eop(struct eop *e)
3938 ###### declare terminals
3939 $LEFT < > <= >= == != CMPop
3941 ###### expression grammar
3942 | Expression CMPop Expression ${ {
3943 struct binode *b = new(binode);
3953 CMPop -> < ${ $0.op = Less; }$
3954 | > ${ $0.op = Gtr; }$
3955 | <= ${ $0.op = LessEq; }$
3956 | >= ${ $0.op = GtrEq; }$
3957 | == ${ $0.op = Eql; }$
3958 | != ${ $0.op = NEql; }$
3960 ###### print binode cases
3968 if (bracket) printf("(");
3969 print_exec(b->left, -1, bracket);
3971 case Less: printf(" < "); break;
3972 case LessEq: printf(" <= "); break;
3973 case Gtr: printf(" > "); break;
3974 case GtrEq: printf(" >= "); break;
3975 case Eql: printf(" == "); break;
3976 case NEql: printf(" != "); break;
3977 default: abort(); // NOTEST
3979 print_exec(b->right, -1, bracket);
3980 if (bracket) printf(")");
3983 ###### propagate binode cases
3990 /* Both must match but not be labels, result is Tbool */
3991 t = propagate_types(b->left, c, perr, NULL, Rnolabel);
3993 propagate_types(b->right, c, perr, t, 0);
3995 t = propagate_types(b->right, c, perr, NULL, Rnolabel); // UNTESTED
3997 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
3999 if (!type_compat(type, Tbool, 0))
4000 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4001 Tbool, rules, type);
4004 ###### interp binode cases
4013 left = interp_exec(c, b->left, <ype);
4014 right = interp_exec(c, b->right, &rtype);
4015 cmp = value_cmp(ltype, rtype, &left, &right);
4018 case Less: rv.bool = cmp < 0; break;
4019 case LessEq: rv.bool = cmp <= 0; break;
4020 case Gtr: rv.bool = cmp > 0; break;
4021 case GtrEq: rv.bool = cmp >= 0; break;
4022 case Eql: rv.bool = cmp == 0; break;
4023 case NEql: rv.bool = cmp != 0; break;
4024 default: rv.bool = 0; break; // NOTEST
4029 ### Expressions: Arithmetic etc.
4031 The remaining expressions with the highest precedence are arithmetic,
4032 string concatenation, string conversion, and testing. String concatenation
4033 (`++`) has the same precedence as multiplication and division, but lower
4036 Testing comes in two forms. A single question mark (`?`) is a uniary
4037 operator which converts come types into Boolean. The general meaning is
4038 "is this a value value" and there will be more uses as the language
4039 develops. A double questionmark (`??`) is a binary operator (Choose),
4040 with same precedence as multiplication, which returns the LHS if it
4041 tests successfully, else returns the RHS.
4043 String conversion is a temporary feature until I get a better type
4044 system. `$` is a prefix operator which expects a string and returns
4047 `+` and `-` are both infix and prefix operations (where they are
4048 absolute value and negation). These have different operator names.
4050 We also have a 'Bracket' operator which records where parentheses were
4051 found. This makes it easy to reproduce these when printing. Possibly I
4052 should only insert brackets were needed for precedence. Putting
4053 parentheses around an expression converts it into a Term,
4059 Absolute, Negate, Test,
4063 ###### declare terminals
4065 $LEFT * / % ++ ?? Top
4069 ###### expression grammar
4070 | Expression Eop Expression ${ {
4071 struct binode *b = new(binode);
4078 | Expression Top Expression ${ {
4079 struct binode *b = new(binode);
4086 | Uop Expression ${ {
4087 struct binode *b = new(binode);
4095 | ( Expression ) ${ {
4096 struct binode *b = new_pos(binode, $1);
4105 Eop -> + ${ $0.op = Plus; }$
4106 | - ${ $0.op = Minus; }$
4108 Uop -> + ${ $0.op = Absolute; }$
4109 | - ${ $0.op = Negate; }$
4110 | $ ${ $0.op = StringConv; }$
4111 | ? ${ $0.op = Test; }$
4113 Top -> * ${ $0.op = Times; }$
4114 | / ${ $0.op = Divide; }$
4115 | % ${ $0.op = Rem; }$
4116 | ++ ${ $0.op = Concat; }$
4117 | ?? ${ $0.op = Choose; }$
4119 ###### print binode cases
4127 if (bracket) printf("(");
4128 print_exec(b->left, indent, bracket);
4130 case Plus: fputs(" + ", stdout); break;
4131 case Minus: fputs(" - ", stdout); break;
4132 case Times: fputs(" * ", stdout); break;
4133 case Divide: fputs(" / ", stdout); break;
4134 case Rem: fputs(" % ", stdout); break;
4135 case Concat: fputs(" ++ ", stdout); break;
4136 case Choose: fputs(" ?? ", stdout); break;
4137 default: abort(); // NOTEST
4139 print_exec(b->right, indent, bracket);
4140 if (bracket) printf(")");
4146 if (bracket) printf("(");
4148 case Absolute: fputs("+", stdout); break;
4149 case Negate: fputs("-", stdout); break;
4150 case StringConv: fputs("$", stdout); break;
4151 case Test: fputs("?", stdout); break;
4152 default: abort(); // NOTEST
4154 print_exec(b->right, indent, bracket);
4155 if (bracket) printf(")");
4159 print_exec(b->right, indent, bracket);
4163 ###### propagate binode cases
4169 /* both must be numbers, result is Tnum */
4172 /* as propagate_types ignores a NULL,
4173 * unary ops fit here too */
4174 propagate_types(b->left, c, perr, Tnum, 0);
4175 propagate_types(b->right, c, perr, Tnum, 0);
4176 if (!type_compat(type, Tnum, 0))
4177 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4182 /* both must be Tstr, result is Tstr */
4183 propagate_types(b->left, c, perr, Tstr, 0);
4184 propagate_types(b->right, c, perr, Tstr, 0);
4185 if (!type_compat(type, Tstr, 0))
4186 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4191 /* op must be string, result is number */
4192 propagate_types(b->left, c, perr, Tstr, 0);
4193 if (!type_compat(type, Tnum, 0))
4194 type_err(c, // UNTESTED
4195 "error: Can only convert string to number, not %1",
4196 prog, type, 0, NULL);
4200 /* LHS must support ->test, result is Tbool */
4201 t = propagate_types(b->right, c, perr, NULL, 0);
4203 type_err(c, "error: '?' requires a testable value, not %1",
4208 /* LHS and RHS must match and are returned. Must support
4211 t = propagate_types(b->left, c, perr, type, rules);
4212 t = propagate_types(b->right, c, perr, t, rules);
4213 if (t && t->test == NULL)
4214 type_err(c, "error: \"??\" requires a testable value, not %1",
4219 return propagate_types(b->right, c, perr, type, 0);
4221 ###### interp binode cases
4224 rv = interp_exec(c, b->left, &rvtype);
4225 right = interp_exec(c, b->right, &rtype);
4226 mpq_add(rv.num, rv.num, right.num);
4229 rv = interp_exec(c, b->left, &rvtype);
4230 right = interp_exec(c, b->right, &rtype);
4231 mpq_sub(rv.num, rv.num, right.num);
4234 rv = interp_exec(c, b->left, &rvtype);
4235 right = interp_exec(c, b->right, &rtype);
4236 mpq_mul(rv.num, rv.num, right.num);
4239 rv = interp_exec(c, b->left, &rvtype);
4240 right = interp_exec(c, b->right, &rtype);
4241 mpq_div(rv.num, rv.num, right.num);
4246 left = interp_exec(c, b->left, <ype);
4247 right = interp_exec(c, b->right, &rtype);
4248 mpz_init(l); mpz_init(r); mpz_init(rem);
4249 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4250 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4251 mpz_tdiv_r(rem, l, r);
4252 val_init(Tnum, &rv);
4253 mpq_set_z(rv.num, rem);
4254 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4259 rv = interp_exec(c, b->right, &rvtype);
4260 mpq_neg(rv.num, rv.num);
4263 rv = interp_exec(c, b->right, &rvtype);
4264 mpq_abs(rv.num, rv.num);
4267 rv = interp_exec(c, b->right, &rvtype);
4270 left = interp_exec(c, b->left, <ype);
4271 right = interp_exec(c, b->right, &rtype);
4273 rv.str = text_join(left.str, right.str);
4276 right = interp_exec(c, b->right, &rvtype);
4280 struct text tx = right.str;
4283 if (tx.txt[0] == '-') {
4284 neg = 1; // UNTESTED
4285 tx.txt++; // UNTESTED
4286 tx.len--; // UNTESTED
4288 if (number_parse(rv.num, tail, tx) == 0)
4289 mpq_init(rv.num); // UNTESTED
4291 mpq_neg(rv.num, rv.num); // UNTESTED
4293 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4297 right = interp_exec(c, b->right, &rtype);
4299 rv.bool = !!rtype->test(rtype, &right);
4302 left = interp_exec(c, b->left, <ype);
4303 if (ltype->test(ltype, &left)) {
4308 rv = interp_exec(c, b->right, &rvtype);
4311 ###### value functions
4313 static struct text text_join(struct text a, struct text b)
4316 rv.len = a.len + b.len;
4317 rv.txt = malloc(rv.len);
4318 memcpy(rv.txt, a.txt, a.len);
4319 memcpy(rv.txt+a.len, b.txt, b.len);
4323 ### Blocks, Statements, and Statement lists.
4325 Now that we have expressions out of the way we need to turn to
4326 statements. There are simple statements and more complex statements.
4327 Simple statements do not contain (syntactic) newlines, complex statements do.
4329 Statements often come in sequences and we have corresponding simple
4330 statement lists and complex statement lists.
4331 The former comprise only simple statements separated by semicolons.
4332 The later comprise complex statements and simple statement lists. They are
4333 separated by newlines. Thus the semicolon is only used to separate
4334 simple statements on the one line. This may be overly restrictive,
4335 but I'm not sure I ever want a complex statement to share a line with
4338 Note that a simple statement list can still use multiple lines if
4339 subsequent lines are indented, so
4341 ###### Example: wrapped simple statement list
4346 is a single simple statement list. This might allow room for
4347 confusion, so I'm not set on it yet.
4349 A simple statement list needs no extra syntax. A complex statement
4350 list has two syntactic forms. It can be enclosed in braces (much like
4351 C blocks), or it can be introduced by an indent and continue until an
4352 unindented newline (much like Python blocks). With this extra syntax
4353 it is referred to as a block.
4355 Note that a block does not have to include any newlines if it only
4356 contains simple statements. So both of:
4358 if condition: a=b; d=f
4360 if condition { a=b; print f }
4364 In either case the list is constructed from a `binode` list with
4365 `Block` as the operator. When parsing the list it is most convenient
4366 to append to the end, so a list is a list and a statement. When using
4367 the list it is more convenient to consider a list to be a statement
4368 and a list. So we need a function to re-order a list.
4369 `reorder_bilist` serves this purpose.
4371 The only stand-alone statement we introduce at this stage is `pass`
4372 which does nothing and is represented as a `NULL` pointer in a `Block`
4373 list. Other stand-alone statements will follow once the infrastructure
4376 As many statements will use binodes, we declare a binode pointer 'b' in
4377 the common header for all reductions to use.
4379 ###### Parser: reduce
4390 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4391 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4392 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4393 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4394 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4396 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4397 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4398 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4399 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4400 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4402 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4403 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4404 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4406 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4407 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4408 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4409 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4410 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4412 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4414 ComplexStatements -> ComplexStatements ComplexStatement ${
4424 | ComplexStatement ${
4436 ComplexStatement -> SimpleStatements Newlines ${
4437 $0 = reorder_bilist($<SS);
4439 | SimpleStatements ; Newlines ${
4440 $0 = reorder_bilist($<SS);
4442 ## ComplexStatement Grammar
4445 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4451 | SimpleStatement ${
4460 SimpleStatement -> pass ${ $0 = NULL; }$
4461 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4462 ## SimpleStatement Grammar
4464 ###### print binode cases
4468 if (b->left == NULL) // UNTESTED
4469 printf("pass"); // UNTESTED
4471 print_exec(b->left, indent, bracket); // UNTESTED
4472 if (b->right) { // UNTESTED
4473 printf("; "); // UNTESTED
4474 print_exec(b->right, indent, bracket); // UNTESTED
4477 // block, one per line
4478 if (b->left == NULL)
4479 do_indent(indent, "pass\n");
4481 print_exec(b->left, indent, bracket);
4483 print_exec(b->right, indent, bracket);
4487 ###### propagate binode cases
4490 /* If any statement returns something other than Tnone
4491 * or Tbool then all such must return same type.
4492 * As each statement may be Tnone or something else,
4493 * we must always pass NULL (unknown) down, otherwise an incorrect
4494 * error might occur. We never return Tnone unless it is
4499 for (e = b; e; e = cast(binode, e->right)) {
4500 t = propagate_types(e->left, c, perr, NULL, rules);
4501 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4503 if (t == Tnone && e->right)
4504 /* Only the final statement *must* return a value
4512 type_err(c, "error: expected %1%r, found %2",
4513 e->left, type, rules, t);
4519 ###### interp binode cases
4521 while (rvtype == Tnone &&
4524 rv = interp_exec(c, b->left, &rvtype);
4525 b = cast(binode, b->right);
4529 ### The Print statement
4531 `print` is a simple statement that takes a comma-separated list of
4532 expressions and prints the values separated by spaces and terminated
4533 by a newline. No control of formatting is possible.
4535 `print` uses `ExpressionList` to collect the expressions and stores them
4536 on the left side of a `Print` binode unlessthere is a trailing comma
4537 when the list is stored on the `right` side and no trailing newline is
4543 ##### declare terminals
4546 ###### SimpleStatement Grammar
4548 | print ExpressionList ${
4549 $0 = b = new_pos(binode, $1);
4552 b->left = reorder_bilist($<EL);
4554 | print ExpressionList , ${ {
4555 $0 = b = new_pos(binode, $1);
4557 b->right = reorder_bilist($<EL);
4561 $0 = b = new_pos(binode, $1);
4567 ###### print binode cases
4570 do_indent(indent, "print");
4572 print_exec(b->right, -1, bracket);
4575 print_exec(b->left, -1, bracket);
4580 ###### propagate binode cases
4583 /* don't care but all must be consistent */
4585 b = cast(binode, b->left);
4587 b = cast(binode, b->right);
4589 propagate_types(b->left, c, perr, NULL, Rnolabel);
4590 b = cast(binode, b->right);
4594 ###### interp binode cases
4598 struct binode *b2 = cast(binode, b->left);
4600 b2 = cast(binode, b->right);
4601 for (; b2; b2 = cast(binode, b2->right)) {
4602 left = interp_exec(c, b2->left, <ype);
4603 print_value(ltype, &left, stdout);
4604 free_value(ltype, &left);
4608 if (b->right == NULL)
4614 ###### Assignment statement
4616 An assignment will assign a value to a variable, providing it hasn't
4617 been declared as a constant. The analysis phase ensures that the type
4618 will be correct so the interpreter just needs to perform the
4619 calculation. There is a form of assignment which declares a new
4620 variable as well as assigning a value. If a name is assigned before
4621 it is declared, and error will be raised as the name is created as
4622 `Tlabel` and it is illegal to assign to such names.
4628 ###### declare terminals
4631 ###### SimpleStatement Grammar
4632 | Term = Expression ${
4633 $0 = b= new(binode);
4638 | VariableDecl = Expression ${
4639 $0 = b= new(binode);
4646 if ($1->var->where_set == NULL) {
4648 "Variable declared with no type or value: %v",
4652 $0 = b = new(binode);
4659 ###### print binode cases
4662 do_indent(indent, "");
4663 print_exec(b->left, -1, bracket);
4665 print_exec(b->right, -1, bracket);
4672 struct variable *v = cast(var, b->left)->var;
4673 do_indent(indent, "");
4674 print_exec(b->left, -1, bracket);
4675 if (cast(var, b->left)->var->constant) {
4677 if (v->explicit_type) {
4678 type_print(v->type, stdout);
4683 if (v->explicit_type) {
4684 type_print(v->type, stdout);
4690 print_exec(b->right, -1, bracket);
4697 ###### propagate binode cases
4701 /* Both must match and not be labels,
4702 * Type must support 'dup',
4703 * For Assign, left must not be constant.
4706 t = propagate_types(b->left, c, perr, NULL,
4707 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
4712 if (propagate_types(b->right, c, perr, t, 0) != t)
4713 if (b->left->type == Xvar)
4714 type_err(c, "info: variable '%v' was set as %1 here.",
4715 cast(var, b->left)->var->where_set, t, rules, NULL);
4717 t = propagate_types(b->right, c, perr, NULL, Rnolabel);
4719 propagate_types(b->left, c, perr, t,
4720 (b->op == Assign ? Rnoconstant : 0));
4722 if (t && t->dup == NULL && !(*perr & Emaycopy))
4723 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4728 ###### interp binode cases
4731 lleft = linterp_exec(c, b->left, <ype);
4733 dinterp_exec(c, b->right, lleft, ltype, 1);
4739 struct variable *v = cast(var, b->left)->var;
4742 val = var_value(c, v);
4743 if (v->type->prepare_type)
4744 v->type->prepare_type(c, v->type, 0);
4746 dinterp_exec(c, b->right, val, v->type, 0);
4748 val_init(v->type, val);
4752 ### The `use` statement
4754 The `use` statement is the last "simple" statement. It is needed when a
4755 statement block can return a value. This includes the body of a
4756 function which has a return type, and the "condition" code blocks in
4757 `if`, `while`, and `switch` statements.
4762 ###### declare terminals
4765 ###### SimpleStatement Grammar
4767 $0 = b = new_pos(binode, $1);
4770 if (b->right->type == Xvar) {
4771 struct var *v = cast(var, b->right);
4772 if (v->var->type == Tnone) {
4773 /* Convert this to a label */
4776 v->var->type = Tlabel;
4777 val = global_alloc(c, Tlabel, v->var, NULL);
4783 ###### print binode cases
4786 do_indent(indent, "use ");
4787 print_exec(b->right, -1, bracket);
4792 ###### propagate binode cases
4795 /* result matches value */
4796 return propagate_types(b->right, c, perr, type, 0);
4798 ###### interp binode cases
4801 rv = interp_exec(c, b->right, &rvtype);
4804 ### The Conditional Statement
4806 This is the biggy and currently the only complex statement. This
4807 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4808 It is comprised of a number of parts, all of which are optional though
4809 set combinations apply. Each part is (usually) a key word (`then` is
4810 sometimes optional) followed by either an expression or a code block,
4811 except the `casepart` which is a "key word and an expression" followed
4812 by a code block. The code-block option is valid for all parts and,
4813 where an expression is also allowed, the code block can use the `use`
4814 statement to report a value. If the code block does not report a value
4815 the effect is similar to reporting `True`.
4817 The `else` and `case` parts, as well as `then` when combined with
4818 `if`, can contain a `use` statement which will apply to some
4819 containing conditional statement. `for` parts, `do` parts and `then`
4820 parts used with `for` can never contain a `use`, except in some
4821 subordinate conditional statement.
4823 If there is a `forpart`, it is executed first, only once.
4824 If there is a `dopart`, then it is executed repeatedly providing
4825 always that the `condpart` or `cond`, if present, does not return a non-True
4826 value. `condpart` can fail to return any value if it simply executes
4827 to completion. This is treated the same as returning `True`.
4829 If there is a `thenpart` it will be executed whenever the `condpart`
4830 or `cond` returns True (or does not return any value), but this will happen
4831 *after* `dopart` (when present).
4833 If `elsepart` is present it will be executed at most once when the
4834 condition returns `False` or some value that isn't `True` and isn't
4835 matched by any `casepart`. If there are any `casepart`s, they will be
4836 executed when the condition returns a matching value.
4838 The particular sorts of values allowed in case parts has not yet been
4839 determined in the language design, so nothing is prohibited.
4841 The various blocks in this complex statement potentially provide scope
4842 for variables as described earlier. Each such block must include the
4843 "OpenScope" nonterminal before parsing the block, and must call
4844 `var_block_close()` when closing the block.
4846 The code following "`if`", "`switch`" and "`for`" does not get its own
4847 scope, but is in a scope covering the whole statement, so names
4848 declared there cannot be redeclared elsewhere. Similarly the
4849 condition following "`while`" is in a scope the covers the body
4850 ("`do`" part) of the loop, and which does not allow conditional scope
4851 extension. Code following "`then`" (both looping and non-looping),
4852 "`else`" and "`case`" each get their own local scope.
4854 The type requirements on the code block in a `whilepart` are quite
4855 unusal. It is allowed to return a value of some identifiable type, in
4856 which case the loop aborts and an appropriate `casepart` is run, or it
4857 can return a Boolean, in which case the loop either continues to the
4858 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4859 This is different both from the `ifpart` code block which is expected to
4860 return a Boolean, or the `switchpart` code block which is expected to
4861 return the same type as the casepart values. The correct analysis of
4862 the type of the `whilepart` code block is the reason for the
4863 `Rboolok` flag which is passed to `propagate_types()`.
4865 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4866 defined. As there are two scopes which cover multiple parts - one for
4867 the whole statement and one for "while" and "do" - and as we will use
4868 the 'struct exec' to track scopes, we actually need two new types of
4869 exec. One is a `binode` for the looping part, the rest is the
4870 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4871 casepart` to track a list of case parts.
4882 struct exec *action;
4883 struct casepart *next;
4885 struct cond_statement {
4887 struct exec *forpart, *condpart, *thenpart, *elsepart;
4888 struct binode *looppart;
4889 struct casepart *casepart;
4892 ###### ast functions
4894 static void free_casepart(struct casepart *cp)
4898 free_exec(cp->value);
4899 free_exec(cp->action);
4906 static void free_cond_statement(struct cond_statement *s)
4910 free_exec(s->forpart);
4911 free_exec(s->condpart);
4912 free_exec(s->looppart);
4913 free_exec(s->thenpart);
4914 free_exec(s->elsepart);
4915 free_casepart(s->casepart);
4919 ###### free exec cases
4920 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4922 ###### ComplexStatement Grammar
4923 | CondStatement ${ $0 = $<1; }$
4925 ###### declare terminals
4926 $TERM for then while do
4933 // A CondStatement must end with EOL, as does CondSuffix and
4935 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4936 // may or may not end with EOL
4937 // WhilePart and IfPart include an appropriate Suffix
4939 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4940 // them. WhilePart opens and closes its own scope.
4941 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4944 $0->thenpart = $<TP;
4945 $0->looppart = $<WP;
4946 var_block_close(c, CloseSequential, $0);
4948 | ForPart OptNL WhilePart CondSuffix ${
4951 $0->looppart = $<WP;
4952 var_block_close(c, CloseSequential, $0);
4954 | WhilePart CondSuffix ${
4956 $0->looppart = $<WP;
4958 | SwitchPart OptNL CasePart CondSuffix ${
4960 $0->condpart = $<SP;
4961 $CP->next = $0->casepart;
4962 $0->casepart = $<CP;
4963 var_block_close(c, CloseSequential, $0);
4965 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4967 $0->condpart = $<SP;
4968 $CP->next = $0->casepart;
4969 $0->casepart = $<CP;
4970 var_block_close(c, CloseSequential, $0);
4972 | IfPart IfSuffix ${
4974 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4975 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4976 // This is where we close an "if" statement
4977 var_block_close(c, CloseSequential, $0);
4980 CondSuffix -> IfSuffix ${
4983 | Newlines CasePart CondSuffix ${
4985 $CP->next = $0->casepart;
4986 $0->casepart = $<CP;
4988 | CasePart CondSuffix ${
4990 $CP->next = $0->casepart;
4991 $0->casepart = $<CP;
4994 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4995 | Newlines ElsePart ${ $0 = $<EP; }$
4996 | ElsePart ${$0 = $<EP; }$
4998 ElsePart -> else OpenBlock Newlines ${
4999 $0 = new(cond_statement);
5000 $0->elsepart = $<OB;
5001 var_block_close(c, CloseElse, $0->elsepart);
5003 | else OpenScope CondStatement ${
5004 $0 = new(cond_statement);
5005 $0->elsepart = $<CS;
5006 var_block_close(c, CloseElse, $0->elsepart);
5010 CasePart -> case Expression OpenScope ColonBlock ${
5011 $0 = calloc(1,sizeof(struct casepart));
5014 var_block_close(c, CloseParallel, $0->action);
5018 // These scopes are closed in CondStatement
5019 ForPart -> for OpenBlock ${
5023 ThenPart -> then OpenBlock ${
5025 var_block_close(c, CloseSequential, $0);
5029 // This scope is closed in CondStatement
5030 WhilePart -> while UseBlock OptNL do OpenBlock ${
5035 var_block_close(c, CloseSequential, $0->right);
5036 var_block_close(c, CloseSequential, $0);
5038 | while OpenScope Expression OpenScope ColonBlock ${
5043 var_block_close(c, CloseSequential, $0->right);
5044 var_block_close(c, CloseSequential, $0);
5048 IfPart -> if UseBlock OptNL then OpenBlock ${
5051 var_block_close(c, CloseParallel, $0.thenpart);
5053 | if OpenScope Expression OpenScope ColonBlock ${
5056 var_block_close(c, CloseParallel, $0.thenpart);
5058 | if OpenScope Expression OpenScope OptNL then Block ${
5061 var_block_close(c, CloseParallel, $0.thenpart);
5065 // This scope is closed in CondStatement
5066 SwitchPart -> switch OpenScope Expression ${
5069 | switch UseBlock ${
5073 ###### print binode cases
5075 if (b->left && b->left->type == Xbinode &&
5076 cast(binode, b->left)->op == Block) {
5078 do_indent(indent, "while {\n");
5080 do_indent(indent, "while\n");
5081 print_exec(b->left, indent+1, bracket);
5083 do_indent(indent, "} do {\n");
5085 do_indent(indent, "do\n");
5086 print_exec(b->right, indent+1, bracket);
5088 do_indent(indent, "}\n");
5090 do_indent(indent, "while ");
5091 print_exec(b->left, 0, bracket);
5096 print_exec(b->right, indent+1, bracket);
5098 do_indent(indent, "}\n");
5102 ###### print exec cases
5104 case Xcond_statement:
5106 struct cond_statement *cs = cast(cond_statement, e);
5107 struct casepart *cp;
5109 do_indent(indent, "for");
5110 if (bracket) printf(" {\n"); else printf("\n");
5111 print_exec(cs->forpart, indent+1, bracket);
5114 do_indent(indent, "} then {\n");
5116 do_indent(indent, "then\n");
5117 print_exec(cs->thenpart, indent+1, bracket);
5119 if (bracket) do_indent(indent, "}\n");
5122 print_exec(cs->looppart, indent, bracket);
5126 do_indent(indent, "switch");
5128 do_indent(indent, "if");
5129 if (cs->condpart && cs->condpart->type == Xbinode &&
5130 cast(binode, cs->condpart)->op == Block) {
5135 print_exec(cs->condpart, indent+1, bracket);
5137 do_indent(indent, "}\n");
5139 do_indent(indent, "then\n");
5140 print_exec(cs->thenpart, indent+1, bracket);
5144 print_exec(cs->condpart, 0, bracket);
5150 print_exec(cs->thenpart, indent+1, bracket);
5152 do_indent(indent, "}\n");
5157 for (cp = cs->casepart; cp; cp = cp->next) {
5158 do_indent(indent, "case ");
5159 print_exec(cp->value, -1, 0);
5164 print_exec(cp->action, indent+1, bracket);
5166 do_indent(indent, "}\n");
5169 do_indent(indent, "else");
5174 print_exec(cs->elsepart, indent+1, bracket);
5176 do_indent(indent, "}\n");
5181 ###### propagate binode cases
5183 t = propagate_types(b->right, c, perr, Tnone, 0);
5184 if (!type_compat(Tnone, t, 0))
5185 *perr |= Efail; // UNTESTED
5186 return propagate_types(b->left, c, perr, type, rules);
5188 ###### propagate exec cases
5189 case Xcond_statement:
5191 // forpart and looppart->right must return Tnone
5192 // thenpart must return Tnone if there is a loopart,
5193 // otherwise it is like elsepart.
5195 // be bool if there is no casepart
5196 // match casepart->values if there is a switchpart
5197 // either be bool or match casepart->value if there
5199 // elsepart and casepart->action must match the return type
5200 // expected of this statement.
5201 struct cond_statement *cs = cast(cond_statement, prog);
5202 struct casepart *cp;
5204 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5205 if (!type_compat(Tnone, t, 0))
5206 *perr |= Efail; // UNTESTED
5209 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5210 if (!type_compat(Tnone, t, 0))
5211 *perr |= Efail; // UNTESTED
5213 if (cs->casepart == NULL) {
5214 propagate_types(cs->condpart, c, perr, Tbool, 0);
5215 propagate_types(cs->looppart, c, perr, Tbool, 0);
5217 /* Condpart must match case values, with bool permitted */
5219 for (cp = cs->casepart;
5220 cp && !t; cp = cp->next)
5221 t = propagate_types(cp->value, c, perr, NULL, 0);
5222 if (!t && cs->condpart)
5223 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5224 if (!t && cs->looppart)
5225 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5226 // Now we have a type (I hope) push it down
5228 for (cp = cs->casepart; cp; cp = cp->next)
5229 propagate_types(cp->value, c, perr, t, 0);
5230 propagate_types(cs->condpart, c, perr, t, Rboolok);
5231 propagate_types(cs->looppart, c, perr, t, Rboolok);
5234 // (if)then, else, and case parts must return expected type.
5235 if (!cs->looppart && !type)
5236 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5238 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5239 for (cp = cs->casepart;
5241 cp = cp->next) // UNTESTED
5242 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5245 propagate_types(cs->thenpart, c, perr, type, rules);
5246 propagate_types(cs->elsepart, c, perr, type, rules);
5247 for (cp = cs->casepart; cp ; cp = cp->next)
5248 propagate_types(cp->action, c, perr, type, rules);
5254 ###### interp binode cases
5256 // This just performs one iterration of the loop
5257 rv = interp_exec(c, b->left, &rvtype);
5258 if (rvtype == Tnone ||
5259 (rvtype == Tbool && rv.bool != 0))
5260 // rvtype is Tnone or Tbool, doesn't need to be freed
5261 interp_exec(c, b->right, NULL);
5264 ###### interp exec cases
5265 case Xcond_statement:
5267 struct value v, cnd;
5268 struct type *vtype, *cndtype;
5269 struct casepart *cp;
5270 struct cond_statement *cs = cast(cond_statement, e);
5273 interp_exec(c, cs->forpart, NULL);
5275 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5276 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5277 interp_exec(c, cs->thenpart, NULL);
5279 cnd = interp_exec(c, cs->condpart, &cndtype);
5280 if ((cndtype == Tnone ||
5281 (cndtype == Tbool && cnd.bool != 0))) {
5282 // cnd is Tnone or Tbool, doesn't need to be freed
5283 rv = interp_exec(c, cs->thenpart, &rvtype);
5284 // skip else (and cases)
5288 for (cp = cs->casepart; cp; cp = cp->next) {
5289 v = interp_exec(c, cp->value, &vtype);
5290 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5291 free_value(vtype, &v);
5292 free_value(cndtype, &cnd);
5293 rv = interp_exec(c, cp->action, &rvtype);
5296 free_value(vtype, &v);
5298 free_value(cndtype, &cnd);
5300 rv = interp_exec(c, cs->elsepart, &rvtype);
5307 ### Top level structure
5309 All the language elements so far can be used in various places. Now
5310 it is time to clarify what those places are.
5312 At the top level of a file there will be a number of declarations.
5313 Many of the things that can be declared haven't been described yet,
5314 such as functions, procedures, imports, and probably more.
5315 For now there are two sorts of things that can appear at the top
5316 level. They are predefined constants, `struct` types, and the `main`
5317 function. While the syntax will allow the `main` function to appear
5318 multiple times, that will trigger an error if it is actually attempted.
5320 The various declarations do not return anything. They store the
5321 various declarations in the parse context.
5323 ###### Parser: grammar
5326 Ocean -> OptNL DeclarationList
5328 ## declare terminals
5336 DeclarationList -> Declaration
5337 | DeclarationList Declaration
5339 Declaration -> ERROR Newlines ${
5340 tok_err(c, // UNTESTED
5341 "error: unhandled parse error", &$1);
5347 ## top level grammar
5351 ### The `const` section
5353 As well as being defined in with the code that uses them, constants can
5354 be declared at the top level. These have full-file scope, so they are
5355 always `InScope`, even before(!) they have been declared. The value of
5356 a top level constant can be given as an expression, and this is
5357 evaluated after parsing and before execution.
5359 A function call can be used to evaluate a constant, but it will not have
5360 access to any program state, once such statement becomes meaningful.
5361 e.g. arguments and filesystem will not be visible.
5363 Constants are defined in a section that starts with the reserved word
5364 `const` and then has a block with a list of assignment statements.
5365 For syntactic consistency, these must use the double-colon syntax to
5366 make it clear that they are constants. Type can also be given: if
5367 not, the type will be determined during analysis, as with other
5370 ###### parse context
5371 struct binode *constlist;
5373 ###### top level grammar
5377 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5378 | const { SimpleConstList } Newlines
5379 | const IN OptNL ConstList OUT Newlines
5380 | const SimpleConstList Newlines
5382 ConstList -> ConstList SimpleConstLine
5385 SimpleConstList -> SimpleConstList ; Const
5389 SimpleConstLine -> SimpleConstList Newlines
5390 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5393 CType -> Type ${ $0 = $<1; }$
5397 Const -> IDENTIFIER :: CType = Expression ${ {
5399 struct binode *bl, *bv;
5400 struct var *var = new_pos(var, $ID);
5402 v = var_decl(c, $ID.txt);
5404 v->where_decl = var;
5410 v = var_ref(c, $1.txt);
5411 if (v->type == Tnone) {
5412 v->where_decl = var;
5418 tok_err(c, "error: name already declared", &$1);
5419 type_err(c, "info: this is where '%v' was first declared",
5420 v->where_decl, NULL, 0, NULL);
5432 bl->left = c->constlist;
5437 ###### core functions
5438 static void resolve_consts(struct parse_context *c)
5442 enum { none, some, cannot } progress = none;
5444 c->constlist = reorder_bilist(c->constlist);
5447 for (b = cast(binode, c->constlist); b;
5448 b = cast(binode, b->right)) {
5450 struct binode *vb = cast(binode, b->left);
5451 struct var *v = cast(var, vb->left);
5452 if (v->var->frame_pos >= 0)
5456 propagate_types(vb->right, c, &perr,
5458 } while (perr & Eretry);
5460 c->parse_error += 1;
5461 else if (!(perr & Enoconst)) {
5463 struct value res = interp_exec(
5464 c, vb->right, &v->var->type);
5465 global_alloc(c, v->var->type, v->var, &res);
5467 if (progress == cannot)
5468 type_err(c, "error: const %v cannot be resolved.",
5478 progress = cannot; break;
5480 progress = none; break;
5485 ###### print const decls
5490 for (b = cast(binode, context.constlist); b;
5491 b = cast(binode, b->right)) {
5492 struct binode *vb = cast(binode, b->left);
5493 struct var *vr = cast(var, vb->left);
5494 struct variable *v = vr->var;
5500 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5501 type_print(v->type, stdout);
5503 print_exec(vb->right, -1, 0);
5508 ###### free const decls
5509 free_binode(context.constlist);
5511 ### Function declarations
5513 The code in an Ocean program is all stored in function declarations.
5514 One of the functions must be named `main` and it must accept an array of
5515 strings as a parameter - the command line arguments.
5517 As this is the top level, several things are handled a bit differently.
5518 The function is not interpreted by `interp_exec` as that isn't passed
5519 the argument list which the program requires. Similarly type analysis
5520 is a bit more interesting at this level.
5522 ###### ast functions
5524 static struct type *handle_results(struct parse_context *c,
5525 struct binode *results)
5527 /* Create a 'struct' type from the results list, which
5528 * is a list for 'struct var'
5530 struct type *t = add_anon_type(c, &structure_prototype,
5535 for (b = results; b; b = cast(binode, b->right))
5537 t->structure.nfields = cnt;
5538 t->structure.fields = calloc(cnt, sizeof(struct field));
5540 for (b = results; b; b = cast(binode, b->right)) {
5541 struct var *v = cast(var, b->left);
5542 struct field *f = &t->structure.fields[cnt++];
5543 int a = v->var->type->align;
5544 f->name = v->var->name->name;
5545 f->type = v->var->type;
5547 f->offset = t->size;
5548 v->var->frame_pos = f->offset;
5549 t->size += ((f->type->size - 1) | (a-1)) + 1;
5552 variable_unlink_exec(v->var);
5554 free_binode(results);
5558 static struct variable *declare_function(struct parse_context *c,
5559 struct variable *name,
5560 struct binode *args,
5562 struct binode *results,
5566 struct value fn = {.function = code};
5568 var_block_close(c, CloseFunction, code);
5569 t = add_anon_type(c, &function_prototype,
5570 "func %.*s", name->name->name.len,
5571 name->name->name.txt);
5573 t->function.params = reorder_bilist(args);
5575 ret = handle_results(c, reorder_bilist(results));
5576 t->function.inline_result = 1;
5577 t->function.local_size = ret->size;
5579 t->function.return_type = ret;
5580 global_alloc(c, t, name, &fn);
5581 name->type->function.scope = c->out_scope;
5586 var_block_close(c, CloseFunction, NULL);
5588 c->out_scope = NULL;
5592 ###### declare terminals
5595 ###### top level grammar
5598 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5599 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5601 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5602 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5604 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5605 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5607 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5608 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5610 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5611 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5613 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5614 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5616 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5617 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5619 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5620 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5622 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5623 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5626 ###### print func decls
5631 while (target != 0) {
5633 for (v = context.in_scope; v; v=v->in_scope)
5634 if (v->depth == 0 && v->type && v->type->check_args) {
5643 struct value *val = var_value(&context, v);
5644 printf("func %.*s", v->name->name.len, v->name->name.txt);
5645 v->type->print_type_decl(v->type, stdout);
5647 print_exec(val->function, 0, brackets);
5649 print_value(v->type, val, stdout);
5650 printf("/* frame size %d */\n", v->type->function.local_size);
5656 ###### core functions
5658 static int analyse_funcs(struct parse_context *c)
5662 for (v = c->in_scope; v; v = v->in_scope) {
5666 if (v->depth != 0 || !v->type || !v->type->check_args)
5668 ret = v->type->function.inline_result ?
5669 Tnone : v->type->function.return_type;
5670 val = var_value(c, v);
5673 propagate_types(val->function, c, &perr, ret, 0);
5674 } while (!(perr & Efail) && (perr & Eretry));
5675 if (!(perr & Efail))
5676 /* Make sure everything is still consistent */
5677 propagate_types(val->function, c, &perr, ret, 0);
5680 if (!v->type->function.inline_result &&
5681 !v->type->function.return_type->dup) {
5682 type_err(c, "error: function cannot return value of type %1",
5683 v->where_decl, v->type->function.return_type, 0, NULL);
5686 scope_finalize(c, v->type);
5691 static int analyse_main(struct type *type, struct parse_context *c)
5693 struct binode *bp = type->function.params;
5697 struct type *argv_type;
5699 argv_type = add_anon_type(c, &array_prototype, "argv");
5700 argv_type->array.member = Tstr;
5701 argv_type->array.unspec = 1;
5703 for (b = bp; b; b = cast(binode, b->right)) {
5707 propagate_types(b->left, c, &perr, argv_type, 0);
5709 default: /* invalid */ // NOTEST
5710 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5713 c->parse_error += 1;
5716 return !c->parse_error;
5719 static void interp_main(struct parse_context *c, int argc, char **argv)
5721 struct value *progp = NULL;
5722 struct text main_name = { "main", 4 };
5723 struct variable *mainv;
5729 mainv = var_ref(c, main_name);
5731 progp = var_value(c, mainv);
5732 if (!progp || !progp->function) {
5733 fprintf(stderr, "oceani: no main function found.\n");
5734 c->parse_error += 1;
5737 if (!analyse_main(mainv->type, c)) {
5738 fprintf(stderr, "oceani: main has wrong type.\n");
5739 c->parse_error += 1;
5742 al = mainv->type->function.params;
5744 c->local_size = mainv->type->function.local_size;
5745 c->local = calloc(1, c->local_size);
5747 struct var *v = cast(var, al->left);
5748 struct value *vl = var_value(c, v->var);
5758 mpq_set_ui(argcq, argc, 1);
5759 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5760 t->prepare_type(c, t, 0);
5761 array_init(v->var->type, vl);
5762 for (i = 0; i < argc; i++) {
5763 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5765 arg.str.txt = argv[i];
5766 arg.str.len = strlen(argv[i]);
5767 free_value(Tstr, vl2);
5768 dup_value(Tstr, &arg, vl2);
5772 al = cast(binode, al->right);
5774 v = interp_exec(c, progp->function, &vtype);
5775 free_value(vtype, &v);
5780 ###### ast functions
5781 void free_variable(struct variable *v)
5785 ## And now to test it out.
5787 Having a language requires having a "hello world" program. I'll
5788 provide a little more than that: a program that prints "Hello world"
5789 finds the GCD of two numbers, prints the first few elements of
5790 Fibonacci, performs a binary search for a number, and a few other
5791 things which will likely grow as the languages grows.
5793 ###### File: oceani.mk
5796 @echo "===== DEMO ====="
5797 ./oceani --section "demo: hello" oceani.mdc 55 33
5803 four ::= 2 + 2 ; five ::= 10/2
5804 const pie ::= "I like Pie";
5805 cake ::= "The cake is"
5813 func main(argv:[argc::]string)
5814 print "Hello World, what lovely oceans you have!"
5815 print "Are there", five, "?"
5816 print pi, pie, "but", cake
5818 A := $argv[1]; B := $argv[2]
5820 /* When a variable is defined in both branches of an 'if',
5821 * and used afterwards, the variables are merged.
5827 print "Is", A, "bigger than", B,"? ", bigger
5828 /* If a variable is not used after the 'if', no
5829 * merge happens, so types can be different
5832 double:string = "yes"
5833 print A, "is more than twice", B, "?", double
5836 print "double", B, "is", double
5841 if a > 0 and then b > 0:
5847 print "GCD of", A, "and", B,"is", a
5849 print a, "is not positive, cannot calculate GCD"
5851 print b, "is not positive, cannot calculate GCD"
5856 print "Fibonacci:", f1,f2,
5857 then togo = togo - 1
5865 /* Binary search... */
5870 mid := (lo + hi) / 2
5883 print "Yay, I found", target
5885 print "Closest I found was", lo
5890 // "middle square" PRNG. Not particularly good, but one my
5891 // Dad taught me - the first one I ever heard of.
5892 for i:=1; then i = i + 1; while i < size:
5893 n := list[i-1] * list[i-1]
5894 list[i] = (n / 100) % 10 000
5896 print "Before sort:",
5897 for i:=0; then i = i + 1; while i < size:
5901 for i := 1; then i=i+1; while i < size:
5902 for j:=i-1; then j=j-1; while j >= 0:
5903 if list[j] > list[j+1]:
5907 print " After sort:",
5908 for i:=0; then i = i + 1; while i < size:
5912 if 1 == 2 then print "yes"; else print "no"
5916 bob.alive = (bob.name == "Hello")
5917 print "bob", "is" if bob.alive else "isn't", "alive"