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;
113 struct parse_context {
114 struct token_config config;
122 #define container_of(ptr, type, member) ({ \
123 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
124 (type *)( (char *)__mptr - offsetof(type,member) );})
126 #define config2context(_conf) container_of(_conf, struct parse_context, \
129 ###### Parser: reduce
130 struct parse_context *c = config2context(config);
138 #include <sys/mman.h>
157 static char Usage[] =
158 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
159 static const struct option long_options[] = {
160 {"trace", 0, NULL, 't'},
161 {"print", 0, NULL, 'p'},
162 {"noexec", 0, NULL, 'n'},
163 {"brackets", 0, NULL, 'b'},
164 {"section", 1, NULL, 's'},
167 const char *options = "tpnbs";
169 static void pr_err(char *msg) // NOTEST
171 fprintf(stderr, "%s\n", msg); // NOTEST
174 int main(int argc, char *argv[])
179 struct section *s = NULL, *ss;
180 char *section = NULL;
181 struct parse_context context = {
183 .ignored = (1 << TK_mark),
184 .number_chars = ".,_+- ",
189 int doprint=0, dotrace=0, doexec=1, brackets=0;
191 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
194 case 't': dotrace=1; break;
195 case 'p': doprint=1; break;
196 case 'n': doexec=0; break;
197 case 'b': brackets=1; break;
198 case 's': section = optarg; break;
199 default: fprintf(stderr, Usage);
203 if (optind >= argc) {
204 fprintf(stderr, "oceani: no input file given\n");
207 fd = open(argv[optind], O_RDONLY);
209 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
212 context.file_name = argv[optind];
213 len = lseek(fd, 0, 2);
214 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
215 s = code_extract(file, file+len, pr_err);
217 fprintf(stderr, "oceani: could not find any code in %s\n",
222 ## context initialization
225 for (ss = s; ss; ss = ss->next) {
226 struct text sec = ss->section;
227 if (sec.len == strlen(section) &&
228 strncmp(sec.txt, section, sec.len) == 0)
232 fprintf(stderr, "oceani: cannot find section %s\n",
239 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
240 goto cleanup; // NOTEST
243 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
245 resolve_consts(&context);
246 prepare_types(&context);
247 if (!context.parse_error && !analyse_funcs(&context)) {
248 fprintf(stderr, "oceani: type error in program - not running.\n");
249 context.parse_error += 1;
257 if (doexec && !context.parse_error)
258 interp_main(&context, argc - optind, argv + optind);
261 struct section *t = s->next;
266 // FIXME parser should pop scope even on error
267 while (context.scope_depth > 0)
271 ## free context types
272 ## free context storage
273 exit(context.parse_error ? 1 : 0);
278 The four requirements of parse, analyse, print, interpret apply to
279 each language element individually so that is how most of the code
282 Three of the four are fairly self explanatory. The one that requires
283 a little explanation is the analysis step.
285 The current language design does not require the types of variables to
286 be declared, but they must still have a single type. Different
287 operations impose different requirements on the variables, for example
288 addition requires both arguments to be numeric, and assignment
289 requires the variable on the left to have the same type as the
290 expression on the right.
292 Analysis involves propagating these type requirements around and
293 consequently setting the type of each variable. If any requirements
294 are violated (e.g. a string is compared with a number) or if a
295 variable needs to have two different types, then an error is raised
296 and the program will not run.
298 If the same variable is declared in both branchs of an 'if/else', or
299 in all cases of a 'switch' then the multiple instances may be merged
300 into just one variable if the variable is referenced after the
301 conditional statement. When this happens, the types must naturally be
302 consistent across all the branches. When the variable is not used
303 outside the if, the variables in the different branches are distinct
304 and can be of different types.
306 Undeclared names may only appear in "use" statements and "case" expressions.
307 These names are given a type of "label" and a unique value.
308 This allows them to fill the role of a name in an enumerated type, which
309 is useful for testing the `switch` statement.
311 As we will see, the condition part of a `while` statement can return
312 either a Boolean or some other type. This requires that the expected
313 type that gets passed around comprises a type and a flag to indicate
314 that `Tbool` is also permitted.
316 As there are, as yet, no distinct types that are compatible, there
317 isn't much subtlety in the analysis. When we have distinct number
318 types, this will become more interesting.
322 When analysis discovers an inconsistency it needs to report an error;
323 just refusing to run the code ensures that the error doesn't cascade,
324 but by itself it isn't very useful. A clear understanding of the sort
325 of error message that are useful will help guide the process of
328 At a simplistic level, the only sort of error that type analysis can
329 report is that the type of some construct doesn't match a contextual
330 requirement. For example, in `4 + "hello"` the addition provides a
331 contextual requirement for numbers, but `"hello"` is not a number. In
332 this particular example no further information is needed as the types
333 are obvious from local information. When a variable is involved that
334 isn't the case. It may be helpful to explain why the variable has a
335 particular type, by indicating the location where the type was set,
336 whether by declaration or usage.
338 Using a recursive-descent analysis we can easily detect a problem at
339 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
340 will detect that one argument is not a number and the usage of `hello`
341 will detect that a number was wanted, but not provided. In this
342 (early) version of the language, we will generate error reports at
343 multiple locations, so the use of `hello` will report an error and
344 explain were the value was set, and the addition will report an error
345 and say why numbers are needed. To be able to report locations for
346 errors, each language element will need to record a file location
347 (line and column) and each variable will need to record the language
348 element where its type was set. For now we will assume that each line
349 of an error message indicates one location in the file, and up to 2
350 types. So we provide a `printf`-like function which takes a format, a
351 location (a `struct exec` which has not yet been introduced), and 2
352 types. "`%1`" reports the first type, "`%2`" reports the second. We
353 will need a function to print the location, once we know how that is
354 stored. e As will be explained later, there are sometimes extra rules for
355 type matching and they might affect error messages, we need to pass those
358 As well as type errors, we sometimes need to report problems with
359 tokens, which might be unexpected or might name a type that has not
360 been defined. For these we have `tok_err()` which reports an error
361 with a given token. Each of the error functions sets the flag in the
362 context so indicate that parsing failed.
366 static void fput_loc(struct exec *loc, FILE *f);
367 static void type_err(struct parse_context *c,
368 char *fmt, struct exec *loc,
369 struct type *t1, int rules, struct type *t2);
370 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
372 ###### core functions
374 static void type_err(struct parse_context *c,
375 char *fmt, struct exec *loc,
376 struct type *t1, int rules, struct type *t2)
378 fprintf(stderr, "%s:", c->file_name);
379 fput_loc(loc, stderr);
380 for (; *fmt ; fmt++) {
387 case '%': fputc(*fmt, stderr); break; // NOTEST
388 default: fputc('?', stderr); break; // NOTEST
390 type_print(t1, stderr);
393 type_print(t2, stderr);
402 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
404 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
405 t->txt.len, t->txt.txt);
409 ## Entities: declared and predeclared.
411 There are various "things" that the language and/or the interpreter
412 needs to know about to parse and execute a program. These include
413 types, variables, values, and executable code. These are all lumped
414 together under the term "entities" (calling them "objects" would be
415 confusing) and introduced here. The following section will present the
416 different specific code elements which comprise or manipulate these
421 Executables can be lots of different things. In many cases an
422 executable is just an operation combined with one or two other
423 executables. This allows for expressions and lists etc. Other times an
424 executable is something quite specific like a constant or variable name.
425 So we define a `struct exec` to be a general executable with a type, and
426 a `struct binode` which is a subclass of `exec`, forms a node in a
427 binary tree, and holds an operation. There will be other subclasses,
428 and to access these we need to be able to `cast` the `exec` into the
429 various other types. The first field in any `struct exec` is the type
430 from the `exec_types` enum.
433 #define cast(structname, pointer) ({ \
434 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
435 if (__mptr && *__mptr != X##structname) abort(); \
436 (struct structname *)( (char *)__mptr);})
438 #define new(structname) ({ \
439 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
440 __ptr->type = X##structname; \
441 __ptr->line = -1; __ptr->column = -1; \
444 #define new_pos(structname, token) ({ \
445 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
446 __ptr->type = X##structname; \
447 __ptr->line = token.line; __ptr->column = token.col; \
456 enum exec_types type;
465 struct exec *left, *right;
470 static int __fput_loc(struct exec *loc, FILE *f)
474 if (loc->line >= 0) {
475 fprintf(f, "%d:%d: ", loc->line, loc->column);
478 if (loc->type == Xbinode)
479 return __fput_loc(cast(binode,loc)->left, f) ||
480 __fput_loc(cast(binode,loc)->right, f); // NOTEST
483 static void fput_loc(struct exec *loc, FILE *f)
485 if (!__fput_loc(loc, f))
486 fprintf(f, "??:??: ");
489 Each different type of `exec` node needs a number of functions defined,
490 a bit like methods. We must be able to free it, print it, analyse it
491 and execute it. Once we have specific `exec` types we will need to
492 parse them too. Let's take this a bit more slowly.
496 The parser generator requires a `free_foo` function for each struct
497 that stores attributes and they will often be `exec`s and subtypes
498 there-of. So we need `free_exec` which can handle all the subtypes,
499 and we need `free_binode`.
503 static void free_binode(struct binode *b)
512 ###### core functions
513 static void free_exec(struct exec *e)
524 static void free_exec(struct exec *e);
526 ###### free exec cases
527 case Xbinode: free_binode(cast(binode, e)); break;
531 Printing an `exec` requires that we know the current indent level for
532 printing line-oriented components. As will become clear later, we
533 also want to know what sort of bracketing to use.
537 static void do_indent(int i, char *str)
544 ###### core functions
545 static void print_binode(struct binode *b, int indent, int bracket)
549 ## print binode cases
553 static void print_exec(struct exec *e, int indent, int bracket)
559 print_binode(cast(binode, e), indent, bracket); break;
564 do_indent(indent, "/* FREE");
565 for (v = e->to_free; v; v = v->next_free) {
566 printf(" %.*s", v->name->name.len, v->name->name.txt);
567 printf("[%d,%d]", v->scope_start, v->scope_end);
568 if (v->frame_pos >= 0)
569 printf("(%d+%d)", v->frame_pos,
570 v->type ? v->type->size:0);
578 static void print_exec(struct exec *e, int indent, int bracket);
582 As discussed, analysis involves propagating type requirements around the
583 program and looking for errors.
585 So `propagate_types` is passed an expected type (being a `struct type`
586 pointer together with some `val_rules` flags) that the `exec` is
587 expected to return, and returns the type that it does return, either of
588 which can be `NULL` signifying "unknown". A `prop_err` flag set is
589 passed by reference. It has `Efail` set when an error is found, and
590 `Eretry` when the type for some element is set via propagation. If
591 any expression cannot be evaluated immediately, `Enoconst` is set.
592 If the expression can be copied, `Emaycopy` is set.
594 If it remains unchanged at `0`, then no more propagation is needed.
598 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 1<<2};
599 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Enoconst = 1<<2,
604 if (rules & Rnolabel)
605 fputs(" (labels not permitted)", stderr);
609 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
610 struct type *type, int rules);
611 ###### core functions
613 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
614 struct type *type, int rules)
621 switch (prog->type) {
624 struct binode *b = cast(binode, prog);
626 ## propagate binode cases
630 ## propagate exec cases
635 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
636 struct type *type, int rules)
638 int pre_err = c->parse_error;
639 struct type *ret = __propagate_types(prog, c, perr, type, rules);
641 if (c->parse_error > pre_err)
648 Interpreting an `exec` doesn't require anything but the `exec`. State
649 is stored in variables and each variable will be directly linked from
650 within the `exec` tree. The exception to this is the `main` function
651 which needs to look at command line arguments. This function will be
652 interpreted separately.
654 Each `exec` can return a value combined with a type in `struct lrval`.
655 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
656 the location of a value, which can be updated, in `lval`. Others will
657 set `lval` to NULL indicating that there is a value of appropriate type
661 static struct value interp_exec(struct parse_context *c, struct exec *e,
662 struct type **typeret);
663 ###### core functions
667 struct value rval, *lval;
670 /* If dest is passed, dtype must give the expected type, and
671 * result can go there, in which case type is returned as NULL.
673 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
674 struct value *dest, struct type *dtype);
676 static struct value interp_exec(struct parse_context *c, struct exec *e,
677 struct type **typeret)
679 struct lrval ret = _interp_exec(c, e, NULL, NULL);
681 if (!ret.type) abort();
685 dup_value(ret.type, ret.lval, &ret.rval);
689 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
690 struct type **typeret)
692 struct lrval ret = _interp_exec(c, e, NULL, NULL);
694 if (!ret.type) abort();
698 free_value(ret.type, &ret.rval);
702 /* dinterp_exec is used when the destination type is certain and
703 * the value has a place to go.
705 static void dinterp_exec(struct parse_context *c, struct exec *e,
706 struct value *dest, struct type *dtype,
709 struct lrval ret = _interp_exec(c, e, dest, dtype);
713 free_value(dtype, dest);
715 dup_value(dtype, ret.lval, dest);
717 memcpy(dest, &ret.rval, dtype->size);
720 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
721 struct value *dest, struct type *dtype)
723 /* If the result is copied to dest, ret.type is set to NULL */
725 struct value rv = {}, *lrv = NULL;
728 rvtype = ret.type = Tnone;
738 struct binode *b = cast(binode, e);
739 struct value left, right, *lleft;
740 struct type *ltype, *rtype;
741 ltype = rtype = Tnone;
743 ## interp binode cases
745 free_value(ltype, &left);
746 free_value(rtype, &right);
756 ## interp exec cleanup
762 Values come in a wide range of types, with more likely to be added.
763 Each type needs to be able to print its own values (for convenience at
764 least) as well as to compare two values, at least for equality and
765 possibly for order. For now, values might need to be duplicated and
766 freed, though eventually such manipulations will be better integrated
769 Rather than requiring every numeric type to support all numeric
770 operations (add, multiply, etc), we allow types to be able to present
771 as one of a few standard types: integer, float, and fraction. The
772 existence of these conversion functions eventually enable types to
773 determine if they are compatible with other types, though such types
774 have not yet been implemented.
776 Named type are stored in a simple linked list. Objects of each type are
777 "values" which are often passed around by value.
779 There are both explicitly named types, and anonymous types. Anonymous
780 cannot be accessed by name, but are used internally and have a name
781 which might be reported in error messages.
788 ## value union fields
795 struct token first_use;
798 void (*init)(struct type *type, struct value *val);
799 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
800 void (*print)(struct type *type, struct value *val, FILE *f);
801 void (*print_type)(struct type *type, FILE *f);
802 int (*cmp_order)(struct type *t1, struct type *t2,
803 struct value *v1, struct value *v2);
804 int (*cmp_eq)(struct type *t1, struct type *t2,
805 struct value *v1, struct value *v2);
806 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
807 void (*free)(struct type *type, struct value *val);
808 void (*free_type)(struct type *t);
809 long long (*to_int)(struct value *v);
810 double (*to_float)(struct value *v);
811 int (*to_mpq)(mpq_t *q, struct value *v);
820 struct type *typelist;
827 static struct type *find_type(struct parse_context *c, struct text s)
829 struct type *t = c->typelist;
831 while (t && (t->anon ||
832 text_cmp(t->name, s) != 0))
837 static struct type *_add_type(struct parse_context *c, struct text s,
838 struct type *proto, int anon)
842 n = calloc(1, sizeof(*n));
849 n->next = c->typelist;
854 static struct type *add_type(struct parse_context *c, struct text s,
857 return _add_type(c, s, proto, 0);
860 static struct type *add_anon_type(struct parse_context *c,
861 struct type *proto, char *name, ...)
867 vasprintf(&t.txt, name, ap);
869 t.len = strlen(name);
870 return _add_type(c, t, proto, 1);
873 static void free_type(struct type *t)
875 /* The type is always a reference to something in the
876 * context, so we don't need to free anything.
880 static void free_value(struct type *type, struct value *v)
884 memset(v, 0x5a, type->size);
888 static void type_print(struct type *type, FILE *f)
891 fputs("*unknown*type*", f); // NOTEST
892 else if (type->name.len && !type->anon)
893 fprintf(f, "%.*s", type->name.len, type->name.txt);
894 else if (type->print_type)
895 type->print_type(type, f);
897 fputs("*invalid*type*", f);
900 static void val_init(struct type *type, struct value *val)
902 if (type && type->init)
903 type->init(type, val);
906 static void dup_value(struct type *type,
907 struct value *vold, struct value *vnew)
909 if (type && type->dup)
910 type->dup(type, vold, vnew);
913 static int value_cmp(struct type *tl, struct type *tr,
914 struct value *left, struct value *right)
916 if (tl && tl->cmp_order)
917 return tl->cmp_order(tl, tr, left, right);
918 if (tl && tl->cmp_eq) // NOTEST
919 return tl->cmp_eq(tl, tr, left, right); // NOTEST
923 static void print_value(struct type *type, struct value *v, FILE *f)
925 if (type && type->print)
926 type->print(type, v, f);
928 fprintf(f, "*Unknown*"); // NOTEST
931 static void prepare_types(struct parse_context *c)
935 enum { none, some, cannot } progress = none;
940 for (t = c->typelist; t; t = t->next) {
942 tok_err(c, "error: type used but not declared",
944 if (t->size == 0 && t->prepare_type) {
945 if (t->prepare_type(c, t, 1))
947 else if (progress == cannot)
948 tok_err(c, "error: type has recursive definition",
958 progress = cannot; break;
960 progress = none; break;
967 static void free_value(struct type *type, struct value *v);
968 static int type_compat(struct type *require, struct type *have, int rules);
969 static void type_print(struct type *type, FILE *f);
970 static void val_init(struct type *type, struct value *v);
971 static void dup_value(struct type *type,
972 struct value *vold, struct value *vnew);
973 static int value_cmp(struct type *tl, struct type *tr,
974 struct value *left, struct value *right);
975 static void print_value(struct type *type, struct value *v, FILE *f);
977 ###### free context types
979 while (context.typelist) {
980 struct type *t = context.typelist;
982 context.typelist = t->next;
990 Type can be specified for local variables, for fields in a structure,
991 for formal parameters to functions, and possibly elsewhere. Different
992 rules may apply in different contexts. As a minimum, a named type may
993 always be used. Currently the type of a formal parameter can be
994 different from types in other contexts, so we have a separate grammar
1000 Type -> IDENTIFIER ${
1001 $0 = find_type(c, $ID.txt);
1003 $0 = add_type(c, $ID.txt, NULL);
1004 $0->first_use = $ID;
1009 FormalType -> Type ${ $0 = $<1; }$
1010 ## formal type grammar
1014 Values of the base types can be numbers, which we represent as
1015 multi-precision fractions, strings, Booleans and labels. When
1016 analysing the program we also need to allow for places where no value
1017 is meaningful (type `Tnone`) and where we don't know what type to
1018 expect yet (type is `NULL`).
1020 Values are never shared, they are always copied when used, and freed
1021 when no longer needed.
1023 When propagating type information around the program, we need to
1024 determine if two types are compatible, where type `NULL` is compatible
1025 with anything. There are two special cases with type compatibility,
1026 both related to the Conditional Statement which will be described
1027 later. In some cases a Boolean can be accepted as well as some other
1028 primary type, and in others any type is acceptable except a label (`Vlabel`).
1029 A separate function encoding these cases will simplify some code later.
1031 ###### type functions
1033 int (*compat)(struct type *this, struct type *other);
1035 ###### ast functions
1037 static int type_compat(struct type *require, struct type *have, int rules)
1039 if ((rules & Rboolok) && have == Tbool)
1041 if ((rules & Rnolabel) && have == Tlabel)
1043 if (!require || !have)
1046 if (require->compat)
1047 return require->compat(require, have);
1049 return require == have;
1054 #include "parse_string.h"
1055 #include "parse_number.h"
1058 myLDLIBS := libnumber.o libstring.o -lgmp
1059 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1061 ###### type union fields
1062 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1064 ###### value union fields
1070 ###### ast functions
1071 static void _free_value(struct type *type, struct value *v)
1075 switch (type->vtype) {
1077 case Vstr: free(v->str.txt); break;
1078 case Vnum: mpq_clear(v->num); break;
1084 ###### value functions
1086 static void _val_init(struct type *type, struct value *val)
1088 switch(type->vtype) {
1089 case Vnone: // NOTEST
1092 mpq_init(val->num); break;
1094 val->str.txt = malloc(1);
1106 static void _dup_value(struct type *type,
1107 struct value *vold, struct value *vnew)
1109 switch (type->vtype) {
1110 case Vnone: // NOTEST
1113 vnew->label = vold->label;
1116 vnew->bool = vold->bool;
1119 mpq_init(vnew->num);
1120 mpq_set(vnew->num, vold->num);
1123 vnew->str.len = vold->str.len;
1124 vnew->str.txt = malloc(vnew->str.len);
1125 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1130 static int _value_cmp(struct type *tl, struct type *tr,
1131 struct value *left, struct value *right)
1135 return tl - tr; // NOTEST
1136 switch (tl->vtype) {
1137 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1138 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1139 case Vstr: cmp = text_cmp(left->str, right->str); break;
1140 case Vbool: cmp = left->bool - right->bool; break;
1141 case Vnone: cmp = 0; // NOTEST
1146 static void _print_value(struct type *type, struct value *v, FILE *f)
1148 switch (type->vtype) {
1149 case Vnone: // NOTEST
1150 fprintf(f, "*no-value*"); break; // NOTEST
1151 case Vlabel: // NOTEST
1152 fprintf(f, "*label-%p*", v->label); break; // NOTEST
1154 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1156 fprintf(f, "%s", v->bool ? "True":"False"); break;
1161 mpf_set_q(fl, v->num);
1162 gmp_fprintf(f, "%.10Fg", fl);
1169 static void _free_value(struct type *type, struct value *v);
1171 static struct type base_prototype = {
1173 .print = _print_value,
1174 .cmp_order = _value_cmp,
1175 .cmp_eq = _value_cmp,
1177 .free = _free_value,
1180 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1182 ###### ast functions
1183 static struct type *add_base_type(struct parse_context *c, char *n,
1184 enum vtype vt, int size)
1186 struct text txt = { n, strlen(n) };
1189 t = add_type(c, txt, &base_prototype);
1192 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1193 if (t->size & (t->align - 1))
1194 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1198 ###### context initialization
1200 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1201 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1202 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1203 Tnone = add_base_type(&context, "none", Vnone, 0);
1204 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1208 We have already met values as separate objects. When manifest constants
1209 appear in the program text, that must result in an executable which has
1210 a constant value. So the `val` structure embeds a value in an
1223 ###### ast functions
1224 struct val *new_val(struct type *T, struct token tk)
1226 struct val *v = new_pos(val, tk);
1237 $0 = new_val(Tbool, $1);
1241 $0 = new_val(Tbool, $1);
1246 $0 = new_val(Tnum, $1);
1247 if (number_parse($0->val.num, tail, $1.txt) == 0)
1248 mpq_init($0->val.num); // UNTESTED
1250 tok_err(c, "error: unsupported number suffix",
1255 $0 = new_val(Tstr, $1);
1256 string_parse(&$1, '\\', &$0->val.str, tail);
1258 tok_err(c, "error: unsupported string suffix",
1263 $0 = new_val(Tstr, $1);
1264 string_parse(&$1, '\\', &$0->val.str, tail);
1266 tok_err(c, "error: unsupported string suffix",
1270 ###### print exec cases
1273 struct val *v = cast(val, e);
1274 if (v->vtype == Tstr)
1276 // FIXME how to ensure numbers have same precision.
1277 print_value(v->vtype, &v->val, stdout);
1278 if (v->vtype == Tstr)
1283 ###### propagate exec cases
1286 struct val *val = cast(val, prog);
1287 if (!type_compat(type, val->vtype, rules))
1288 type_err(c, "error: expected %1%r found %2",
1289 prog, type, rules, val->vtype);
1293 ###### interp exec cases
1295 rvtype = cast(val, e)->vtype;
1296 dup_value(rvtype, &cast(val, e)->val, &rv);
1299 ###### ast functions
1300 static void free_val(struct val *v)
1303 free_value(v->vtype, &v->val);
1307 ###### free exec cases
1308 case Xval: free_val(cast(val, e)); break;
1310 ###### ast functions
1311 // Move all nodes from 'b' to 'rv', reversing their order.
1312 // In 'b' 'left' is a list, and 'right' is the last node.
1313 // In 'rv', left' is the first node and 'right' is a list.
1314 static struct binode *reorder_bilist(struct binode *b)
1316 struct binode *rv = NULL;
1319 struct exec *t = b->right;
1323 b = cast(binode, b->left);
1333 Variables are scoped named values. We store the names in a linked list
1334 of "bindings" sorted in lexical order, and use sequential search and
1341 struct binding *next; // in lexical order
1345 This linked list is stored in the parse context so that "reduce"
1346 functions can find or add variables, and so the analysis phase can
1347 ensure that every variable gets a type.
1349 ###### parse context
1351 struct binding *varlist; // In lexical order
1353 ###### ast functions
1355 static struct binding *find_binding(struct parse_context *c, struct text s)
1357 struct binding **l = &c->varlist;
1362 (cmp = text_cmp((*l)->name, s)) < 0)
1366 n = calloc(1, sizeof(*n));
1373 Each name can be linked to multiple variables defined in different
1374 scopes. Each scope starts where the name is declared and continues
1375 until the end of the containing code block. Scopes of a given name
1376 cannot nest, so a declaration while a name is in-scope is an error.
1378 ###### binding fields
1379 struct variable *var;
1383 struct variable *previous;
1385 struct binding *name;
1386 struct exec *where_decl;// where name was declared
1387 struct exec *where_set; // where type was set
1391 When a scope closes, the values of the variables might need to be freed.
1392 This happens in the context of some `struct exec` and each `exec` will
1393 need to know which variables need to be freed when it completes.
1396 struct variable *to_free;
1398 ####### variable fields
1399 struct exec *cleanup_exec;
1400 struct variable *next_free;
1402 ####### interp exec cleanup
1405 for (v = e->to_free; v; v = v->next_free) {
1406 struct value *val = var_value(c, v);
1407 free_value(v->type, val);
1411 ###### ast functions
1412 static void variable_unlink_exec(struct variable *v)
1414 struct variable **vp;
1415 if (!v->cleanup_exec)
1417 for (vp = &v->cleanup_exec->to_free;
1418 *vp; vp = &(*vp)->next_free) {
1422 v->cleanup_exec = NULL;
1427 While the naming seems strange, we include local constants in the
1428 definition of variables. A name declared `var := value` can
1429 subsequently be changed, but a name declared `var ::= value` cannot -
1432 ###### variable fields
1435 Scopes in parallel branches can be partially merged. More
1436 specifically, if a given name is declared in both branches of an
1437 if/else then its scope is a candidate for merging. Similarly if
1438 every branch of an exhaustive switch (e.g. has an "else" clause)
1439 declares a given name, then the scopes from the branches are
1440 candidates for merging.
1442 Note that names declared inside a loop (which is only parallel to
1443 itself) are never visible after the loop. Similarly names defined in
1444 scopes which are not parallel, such as those started by `for` and
1445 `switch`, are never visible after the scope. Only variables defined in
1446 both `then` and `else` (including the implicit then after an `if`, and
1447 excluding `then` used with `for`) and in all `case`s and `else` of a
1448 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1450 Labels, which are a bit like variables, follow different rules.
1451 Labels are not explicitly declared, but if an undeclared name appears
1452 in a context where a label is legal, that effectively declares the
1453 name as a label. The declaration remains in force (or in scope) at
1454 least to the end of the immediately containing block and conditionally
1455 in any larger containing block which does not declare the name in some
1456 other way. Importantly, the conditional scope extension happens even
1457 if the label is only used in one parallel branch of a conditional --
1458 when used in one branch it is treated as having been declared in all
1461 Merge candidates are tentatively visible beyond the end of the
1462 branching statement which creates them. If the name is used, the
1463 merge is affirmed and they become a single variable visible at the
1464 outer layer. If not - if it is redeclared first - the merge lapses.
1466 To track scopes we have an extra stack, implemented as a linked list,
1467 which roughly parallels the parse stack and which is used exclusively
1468 for scoping. When a new scope is opened, a new frame is pushed and
1469 the child-count of the parent frame is incremented. This child-count
1470 is used to distinguish between the first of a set of parallel scopes,
1471 in which declared variables must not be in scope, and subsequent
1472 branches, whether they may already be conditionally scoped.
1474 We need a total ordering of scopes so we can easily compare to variables
1475 to see if they are concurrently in scope. To achieve this we record a
1476 `scope_count` which is actually a count of both beginnings and endings
1477 of scopes. Then each variable has a record of the scope count where it
1478 enters scope, and where it leaves.
1480 To push a new frame *before* any code in the frame is parsed, we need a
1481 grammar reduction. This is most easily achieved with a grammar
1482 element which derives the empty string, and creates the new scope when
1483 it is recognised. This can be placed, for example, between a keyword
1484 like "if" and the code following it.
1488 struct scope *parent;
1492 ###### parse context
1495 struct scope *scope_stack;
1497 ###### variable fields
1498 int scope_start, scope_end;
1500 ###### ast functions
1501 static void scope_pop(struct parse_context *c)
1503 struct scope *s = c->scope_stack;
1505 c->scope_stack = s->parent;
1507 c->scope_depth -= 1;
1508 c->scope_count += 1;
1511 static void scope_push(struct parse_context *c)
1513 struct scope *s = calloc(1, sizeof(*s));
1515 c->scope_stack->child_count += 1;
1516 s->parent = c->scope_stack;
1518 c->scope_depth += 1;
1519 c->scope_count += 1;
1525 OpenScope -> ${ scope_push(c); }$
1527 Each variable records a scope depth and is in one of four states:
1529 - "in scope". This is the case between the declaration of the
1530 variable and the end of the containing block, and also between
1531 the usage with affirms a merge and the end of that block.
1533 The scope depth is not greater than the current parse context scope
1534 nest depth. When the block of that depth closes, the state will
1535 change. To achieve this, all "in scope" variables are linked
1536 together as a stack in nesting order.
1538 - "pending". The "in scope" block has closed, but other parallel
1539 scopes are still being processed. So far, every parallel block at
1540 the same level that has closed has declared the name.
1542 The scope depth is the depth of the last parallel block that
1543 enclosed the declaration, and that has closed.
1545 - "conditionally in scope". The "in scope" block and all parallel
1546 scopes have closed, and no further mention of the name has been seen.
1547 This state includes a secondary nest depth (`min_depth`) which records
1548 the outermost scope seen since the variable became conditionally in
1549 scope. If a use of the name is found, the variable becomes "in scope"
1550 and that secondary depth becomes the recorded scope depth. If the
1551 name is declared as a new variable, the old variable becomes "out of
1552 scope" and the recorded scope depth stays unchanged.
1554 - "out of scope". The variable is neither in scope nor conditionally
1555 in scope. It is permanently out of scope now and can be removed from
1556 the "in scope" stack. When a variable becomes out-of-scope it is
1557 moved to a separate list (`out_scope`) of variables which have fully
1558 known scope. This will be used at the end of each function to assign
1559 each variable a place in the stack frame.
1561 ###### variable fields
1562 int depth, min_depth;
1563 enum { OutScope, PendingScope, CondScope, InScope } scope;
1564 struct variable *in_scope;
1566 ###### parse context
1568 struct variable *in_scope;
1569 struct variable *out_scope;
1571 All variables with the same name are linked together using the
1572 'previous' link. Those variable that have been affirmatively merged all
1573 have a 'merged' pointer that points to one primary variable - the most
1574 recently declared instance. When merging variables, we need to also
1575 adjust the 'merged' pointer on any other variables that had previously
1576 been merged with the one that will no longer be primary.
1578 A variable that is no longer the most recent instance of a name may
1579 still have "pending" scope, if it might still be merged with most
1580 recent instance. These variables don't really belong in the
1581 "in_scope" list, but are not immediately removed when a new instance
1582 is found. Instead, they are detected and ignored when considering the
1583 list of in_scope names.
1585 The storage of the value of a variable will be described later. For now
1586 we just need to know that when a variable goes out of scope, it might
1587 need to be freed. For this we need to be able to find it, so assume that
1588 `var_value()` will provide that.
1590 ###### variable fields
1591 struct variable *merged;
1593 ###### ast functions
1595 static void variable_merge(struct variable *primary, struct variable *secondary)
1599 primary = primary->merged;
1601 for (v = primary->previous; v; v=v->previous)
1602 if (v == secondary || v == secondary->merged ||
1603 v->merged == secondary ||
1604 v->merged == secondary->merged) {
1605 v->scope = OutScope;
1606 v->merged = primary;
1607 if (v->scope_start < primary->scope_start)
1608 primary->scope_start = v->scope_start;
1609 if (v->scope_end > primary->scope_end)
1610 primary->scope_end = v->scope_end; // NOTEST
1611 variable_unlink_exec(v);
1615 ###### forward decls
1616 static struct value *var_value(struct parse_context *c, struct variable *v);
1618 ###### free global vars
1620 while (context.varlist) {
1621 struct binding *b = context.varlist;
1622 struct variable *v = b->var;
1623 context.varlist = b->next;
1626 struct variable *next = v->previous;
1628 if (v->global && v->frame_pos >= 0) {
1629 free_value(v->type, var_value(&context, v));
1630 if (v->depth == 0 && v->type->free == function_free)
1631 // This is a function constant
1632 free_exec(v->where_decl);
1639 #### Manipulating Bindings
1641 When a name is conditionally visible, a new declaration discards the old
1642 binding - the condition lapses. Similarly when we reach the end of a
1643 function (outermost non-global scope) any conditional scope must lapse.
1644 Conversely a usage of the name affirms the visibility and extends it to
1645 the end of the containing block - i.e. the block that contains both the
1646 original declaration and the latest usage. This is determined from
1647 `min_depth`. When a conditionally visible variable gets affirmed like
1648 this, it is also merged with other conditionally visible variables with
1651 When we parse a variable declaration we either report an error if the
1652 name is currently bound, or create a new variable at the current nest
1653 depth if the name is unbound or bound to a conditionally scoped or
1654 pending-scope variable. If the previous variable was conditionally
1655 scoped, it and its homonyms becomes out-of-scope.
1657 When we parse a variable reference (including non-declarative assignment
1658 "foo = bar") we report an error if the name is not bound or is bound to
1659 a pending-scope variable; update the scope if the name is bound to a
1660 conditionally scoped variable; or just proceed normally if the named
1661 variable is in scope.
1663 When we exit a scope, any variables bound at this level are either
1664 marked out of scope or pending-scoped, depending on whether the scope
1665 was sequential or parallel. Here a "parallel" scope means the "then"
1666 or "else" part of a conditional, or any "case" or "else" branch of a
1667 switch. Other scopes are "sequential".
1669 When exiting a parallel scope we check if there are any variables that
1670 were previously pending and are still visible. If there are, then
1671 they weren't redeclared in the most recent scope, so they cannot be
1672 merged and must become out-of-scope. If it is not the first of
1673 parallel scopes (based on `child_count`), we check that there was a
1674 previous binding that is still pending-scope. If there isn't, the new
1675 variable must now be out-of-scope.
1677 When exiting a sequential scope that immediately enclosed parallel
1678 scopes, we need to resolve any pending-scope variables. If there was
1679 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1680 we need to mark all pending-scope variable as out-of-scope. Otherwise
1681 all pending-scope variables become conditionally scoped.
1684 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1686 ###### ast functions
1688 static struct variable *var_decl(struct parse_context *c, struct text s)
1690 struct binding *b = find_binding(c, s);
1691 struct variable *v = b->var;
1693 switch (v ? v->scope : OutScope) {
1695 /* Caller will report the error */
1699 v && v->scope == CondScope;
1701 v->scope = OutScope;
1705 v = calloc(1, sizeof(*v));
1706 v->previous = b->var;
1710 v->min_depth = v->depth = c->scope_depth;
1712 v->in_scope = c->in_scope;
1713 v->scope_start = c->scope_count;
1719 static struct variable *var_ref(struct parse_context *c, struct text s)
1721 struct binding *b = find_binding(c, s);
1722 struct variable *v = b->var;
1723 struct variable *v2;
1725 switch (v ? v->scope : OutScope) {
1728 /* Caller will report the error */
1731 /* All CondScope variables of this name need to be merged
1732 * and become InScope
1734 v->depth = v->min_depth;
1736 for (v2 = v->previous;
1737 v2 && v2->scope == CondScope;
1739 variable_merge(v, v2);
1747 static int var_refile(struct parse_context *c, struct variable *v)
1749 /* Variable just went out of scope. Add it to the out_scope
1750 * list, sorted by ->scope_start
1752 struct variable **vp = &c->out_scope;
1753 while ((*vp) && (*vp)->scope_start < v->scope_start)
1754 vp = &(*vp)->in_scope;
1760 static void var_block_close(struct parse_context *c, enum closetype ct,
1763 /* Close off all variables that are in_scope.
1764 * Some variables in c->scope may already be not-in-scope,
1765 * such as when a PendingScope variable is hidden by a new
1766 * variable with the same name.
1767 * So we check for v->name->var != v and drop them.
1768 * If we choose to make a variable OutScope, we drop it
1771 struct variable *v, **vp, *v2;
1774 for (vp = &c->in_scope;
1775 (v = *vp) && v->min_depth > c->scope_depth;
1776 (v->scope == OutScope || v->name->var != v)
1777 ? (*vp = v->in_scope, var_refile(c, v))
1778 : ( vp = &v->in_scope, 0)) {
1779 v->min_depth = c->scope_depth;
1780 if (v->name->var != v)
1781 /* This is still in scope, but we haven't just
1785 v->min_depth = c->scope_depth;
1786 if (v->scope == InScope)
1787 v->scope_end = c->scope_count;
1788 if (v->scope == InScope && e && !v->global) {
1789 /* This variable gets cleaned up when 'e' finishes */
1790 variable_unlink_exec(v);
1791 v->cleanup_exec = e;
1792 v->next_free = e->to_free;
1797 case CloseParallel: /* handle PendingScope */
1801 if (c->scope_stack->child_count == 1)
1802 /* first among parallel branches */
1803 v->scope = PendingScope;
1804 else if (v->previous &&
1805 v->previous->scope == PendingScope)
1806 /* all previous branches used name */
1807 v->scope = PendingScope;
1808 else if (v->type == Tlabel)
1809 /* Labels remain pending even when not used */
1810 v->scope = PendingScope; // UNTESTED
1812 v->scope = OutScope;
1813 if (ct == CloseElse) {
1814 /* All Pending variables with this name
1815 * are now Conditional */
1817 v2 && v2->scope == PendingScope;
1819 v2->scope = CondScope;
1823 /* Not possible as it would require
1824 * parallel scope to be nested immediately
1825 * in a parallel scope, and that never
1829 /* Not possible as we already tested for
1836 if (v->scope == CondScope)
1837 /* Condition cannot continue past end of function */
1840 case CloseSequential:
1841 if (v->type == Tlabel)
1842 v->scope = PendingScope;
1845 v->scope = OutScope;
1848 /* There was no 'else', so we can only become
1849 * conditional if we know the cases were exhaustive,
1850 * and that doesn't mean anything yet.
1851 * So only labels become conditional..
1854 v2 && v2->scope == PendingScope;
1856 if (v2->type == Tlabel)
1857 v2->scope = CondScope;
1859 v2->scope = OutScope;
1862 case OutScope: break;
1871 The value of a variable is store separately from the variable, on an
1872 analogue of a stack frame. There are (currently) two frames that can be
1873 active. A global frame which currently only stores constants, and a
1874 stacked frame which stores local variables. Each variable knows if it
1875 is global or not, and what its index into the frame is.
1877 Values in the global frame are known immediately they are relevant, so
1878 the frame needs to be reallocated as it grows so it can store those
1879 values. The local frame doesn't get values until the interpreted phase
1880 is started, so there is no need to allocate until the size is known.
1882 We initialize the `frame_pos` to an impossible value, so that we can
1883 tell if it was set or not later.
1885 ###### variable fields
1889 ###### variable init
1892 ###### parse context
1894 short global_size, global_alloc;
1896 void *global, *local;
1898 ###### forward decls
1899 static struct value *global_alloc(struct parse_context *c, struct type *t,
1900 struct variable *v, struct value *init);
1902 ###### ast functions
1904 static struct value *var_value(struct parse_context *c, struct variable *v)
1907 if (!c->local || !v->type)
1908 return NULL; // UNTESTED
1909 if (v->frame_pos + v->type->size > c->local_size) {
1910 printf("INVALID frame_pos\n"); // NOTEST
1913 return c->local + v->frame_pos;
1915 if (c->global_size > c->global_alloc) {
1916 int old = c->global_alloc;
1917 c->global_alloc = (c->global_size | 1023) + 1024;
1918 c->global = realloc(c->global, c->global_alloc);
1919 memset(c->global + old, 0, c->global_alloc - old);
1921 return c->global + v->frame_pos;
1924 static struct value *global_alloc(struct parse_context *c, struct type *t,
1925 struct variable *v, struct value *init)
1928 struct variable scratch;
1930 if (t->prepare_type)
1931 t->prepare_type(c, t, 1); // NOTEST
1933 if (c->global_size & (t->align - 1))
1934 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
1939 v->frame_pos = c->global_size;
1941 c->global_size += v->type->size;
1942 ret = var_value(c, v);
1944 memcpy(ret, init, t->size);
1950 As global values are found -- struct field initializers, labels etc --
1951 `global_alloc()` is called to record the value in the global frame.
1953 When the program is fully parsed, each function is analysed, we need to
1954 walk the list of variables local to that function and assign them an
1955 offset in the stack frame. For this we have `scope_finalize()`.
1957 We keep the stack from dense by re-using space for between variables
1958 that are not in scope at the same time. The `out_scope` list is sorted
1959 by `scope_start` and as we process a varible, we move it to an FIFO
1960 stack. For each variable we consider, we first discard any from the
1961 stack anything that went out of scope before the new variable came in.
1962 Then we place the new variable just after the one at the top of the
1965 ###### ast functions
1967 static void scope_finalize(struct parse_context *c, struct type *ft)
1969 int size = ft->function.local_size;
1970 struct variable *next = ft->function.scope;
1971 struct variable *done = NULL;
1974 struct variable *v = next;
1975 struct type *t = v->type;
1982 if (v->frame_pos >= 0)
1984 while (done && done->scope_end < v->scope_start)
1985 done = done->in_scope;
1987 pos = done->frame_pos + done->type->size;
1989 pos = ft->function.local_size;
1990 if (pos & (t->align - 1))
1991 pos = (pos + t->align) & ~(t->align-1);
1993 if (size < pos + v->type->size)
1994 size = pos + v->type->size;
1998 c->out_scope = NULL;
1999 ft->function.local_size = size;
2002 ###### free context storage
2003 free(context.global);
2005 #### Variables as executables
2007 Just as we used a `val` to wrap a value into an `exec`, we similarly
2008 need a `var` to wrap a `variable` into an exec. While each `val`
2009 contained a copy of the value, each `var` holds a link to the variable
2010 because it really is the same variable no matter where it appears.
2011 When a variable is used, we need to remember to follow the `->merged`
2012 link to find the primary instance.
2014 When a variable is declared, it may or may not be given an explicit
2015 type. We need to record which so that we can report the parsed code
2024 struct variable *var;
2027 ###### variable fields
2035 VariableDecl -> IDENTIFIER : ${ {
2036 struct variable *v = var_decl(c, $1.txt);
2037 $0 = new_pos(var, $1);
2042 v = var_ref(c, $1.txt);
2044 type_err(c, "error: variable '%v' redeclared",
2046 type_err(c, "info: this is where '%v' was first declared",
2047 v->where_decl, NULL, 0, NULL);
2050 | IDENTIFIER :: ${ {
2051 struct variable *v = var_decl(c, $1.txt);
2052 $0 = new_pos(var, $1);
2058 v = var_ref(c, $1.txt);
2060 type_err(c, "error: variable '%v' redeclared",
2062 type_err(c, "info: this is where '%v' was first declared",
2063 v->where_decl, NULL, 0, NULL);
2066 | IDENTIFIER : Type ${ {
2067 struct variable *v = var_decl(c, $1.txt);
2068 $0 = new_pos(var, $1);
2074 v->explicit_type = 1;
2076 v = var_ref(c, $1.txt);
2078 type_err(c, "error: variable '%v' redeclared",
2080 type_err(c, "info: this is where '%v' was first declared",
2081 v->where_decl, NULL, 0, NULL);
2084 | IDENTIFIER :: Type ${ {
2085 struct variable *v = var_decl(c, $1.txt);
2086 $0 = new_pos(var, $1);
2093 v->explicit_type = 1;
2095 v = var_ref(c, $1.txt);
2097 type_err(c, "error: variable '%v' redeclared",
2099 type_err(c, "info: this is where '%v' was first declared",
2100 v->where_decl, NULL, 0, NULL);
2105 Variable -> IDENTIFIER ${ {
2106 struct variable *v = var_ref(c, $1.txt);
2107 $0 = new_pos(var, $1);
2109 /* This might be a global const or a label
2110 * Allocate a var with impossible type Tnone,
2111 * which will be adjusted when we find out what it is,
2112 * or will trigger an error.
2114 v = var_decl(c, $1.txt);
2121 cast(var, $0)->var = v;
2124 ###### print exec cases
2127 struct var *v = cast(var, e);
2129 struct binding *b = v->var->name;
2130 printf("%.*s", b->name.len, b->name.txt);
2137 if (loc && loc->type == Xvar) {
2138 struct var *v = cast(var, loc);
2140 struct binding *b = v->var->name;
2141 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2143 fputs("???", stderr); // NOTEST
2145 fputs("NOTVAR", stderr);
2148 ###### propagate exec cases
2152 struct var *var = cast(var, prog);
2153 struct variable *v = var->var;
2155 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2156 return Tnone; // NOTEST
2159 if (v->constant && (rules & Rnoconstant)) {
2160 type_err(c, "error: Cannot assign to a constant: %v",
2161 prog, NULL, 0, NULL);
2162 type_err(c, "info: name was defined as a constant here",
2163 v->where_decl, NULL, 0, NULL);
2166 if (v->type == Tnone && v->where_decl == prog)
2167 type_err(c, "error: variable used but not declared: %v",
2168 prog, NULL, 0, NULL);
2169 if (v->type == NULL) {
2170 if (type && !(*perr & Efail)) {
2172 v->where_set = prog;
2175 } else if (!type_compat(type, v->type, rules)) {
2176 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2177 type, rules, v->type);
2178 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2179 v->type, rules, NULL);
2181 if (!v->global || v->frame_pos < 0)
2188 ###### interp exec cases
2191 struct var *var = cast(var, e);
2192 struct variable *v = var->var;
2195 lrv = var_value(c, v);
2200 ###### ast functions
2202 static void free_var(struct var *v)
2207 ###### free exec cases
2208 case Xvar: free_var(cast(var, e)); break;
2213 Now that we have the shape of the interpreter in place we can add some
2214 complex types and connected them in to the data structures and the
2215 different phases of parse, analyse, print, interpret.
2217 Being "complex" the language will naturally have syntax to access
2218 specifics of objects of these types. These will fit into the grammar as
2219 "Terms" which are the things that are combined with various operators to
2220 form "Expression". Where a Term is formed by some operation on another
2221 Term, the subordinate Term will always come first, so for example a
2222 member of an array will be expressed as the Term for the array followed
2223 by an index in square brackets. The strict rule of using postfix
2224 operations makes precedence irrelevant within terms. To provide a place
2225 to put the grammar for each terms of each type, we will start out by
2226 introducing the "Term" grammar production, with contains at least a
2227 simple "Value" (to be explained later).
2231 Term -> Value ${ $0 = $<1; }$
2232 | Variable ${ $0 = $<1; }$
2235 Thus far the complex types we have are arrays and structs.
2239 Arrays can be declared by giving a size and a type, as `[size]type' so
2240 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2241 size can be either a literal number, or a named constant. Some day an
2242 arbitrary expression will be supported.
2244 As a formal parameter to a function, the array can be declared with a
2245 new variable as the size: `name:[size::number]string`. The `size`
2246 variable is set to the size of the array and must be a constant. As
2247 `number` is the only supported type, it can be left out:
2248 `name:[size::]string`.
2250 Arrays cannot be assigned. When pointers are introduced we will also
2251 introduce array slices which can refer to part or all of an array -
2252 the assignment syntax will create a slice. For now, an array can only
2253 ever be referenced by the name it is declared with. It is likely that
2254 a "`copy`" primitive will eventually be define which can be used to
2255 make a copy of an array with controllable recursive depth.
2257 For now we have two sorts of array, those with fixed size either because
2258 it is given as a literal number or because it is a struct member (which
2259 cannot have a runtime-changing size), and those with a size that is
2260 determined at runtime - local variables with a const size. The former
2261 have their size calculated at parse time, the latter at run time.
2263 For the latter type, the `size` field of the type is the size of a
2264 pointer, and the array is reallocated every time it comes into scope.
2266 We differentiate struct fields with a const size from local variables
2267 with a const size by whether they are prepared at parse time or not.
2269 ###### type union fields
2272 int unspec; // size is unspecified - vsize must be set.
2275 struct variable *vsize;
2276 struct type *member;
2279 ###### value union fields
2280 void *array; // used if not static_size
2282 ###### value functions
2284 static int array_prepare_type(struct parse_context *c, struct type *type,
2287 struct value *vsize;
2289 if (type->array.static_size)
2290 return 1; // UNTESTED
2291 if (type->array.unspec && parse_time)
2292 return 1; // UNTESTED
2293 if (parse_time && type->array.vsize && !type->array.vsize->global)
2294 return 1; // UNTESTED
2296 if (type->array.vsize) {
2297 vsize = var_value(c, type->array.vsize);
2299 return 1; // UNTESTED
2301 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2302 type->array.size = mpz_get_si(q);
2307 if (type->array.member->size <= 0)
2310 type->array.static_size = 1;
2311 type->size = type->array.size * type->array.member->size;
2312 type->align = type->array.member->align;
2317 static void array_init(struct type *type, struct value *val)
2320 void *ptr = val->ptr;
2324 if (!type->array.static_size) {
2325 val->array = calloc(type->array.size,
2326 type->array.member->size);
2329 for (i = 0; i < type->array.size; i++) {
2331 v = (void*)ptr + i * type->array.member->size;
2332 val_init(type->array.member, v);
2336 static void array_free(struct type *type, struct value *val)
2339 void *ptr = val->ptr;
2341 if (!type->array.static_size)
2343 for (i = 0; i < type->array.size; i++) {
2345 v = (void*)ptr + i * type->array.member->size;
2346 free_value(type->array.member, v);
2348 if (!type->array.static_size)
2352 static int array_compat(struct type *require, struct type *have)
2354 if (have->compat != require->compat)
2356 /* Both are arrays, so we can look at details */
2357 if (!type_compat(require->array.member, have->array.member, 0))
2359 if (have->array.unspec && require->array.unspec) {
2360 if (have->array.vsize && require->array.vsize &&
2361 have->array.vsize != require->array.vsize) // UNTESTED
2362 /* sizes might not be the same */
2363 return 0; // UNTESTED
2366 if (have->array.unspec || require->array.unspec)
2367 return 1; // UNTESTED
2368 if (require->array.vsize == NULL && have->array.vsize == NULL)
2369 return require->array.size == have->array.size;
2371 return require->array.vsize == have->array.vsize; // UNTESTED
2374 static void array_print_type(struct type *type, FILE *f)
2377 if (type->array.vsize) {
2378 struct binding *b = type->array.vsize->name;
2379 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2380 type->array.unspec ? "::" : "");
2381 } else if (type->array.size)
2382 fprintf(f, "%d]", type->array.size);
2385 type_print(type->array.member, f);
2388 static struct type array_prototype = {
2390 .prepare_type = array_prepare_type,
2391 .print_type = array_print_type,
2392 .compat = array_compat,
2394 .size = sizeof(void*),
2395 .align = sizeof(void*),
2398 ###### declare terminals
2403 | [ NUMBER ] Type ${ {
2409 if (number_parse(num, tail, $2.txt) == 0)
2410 tok_err(c, "error: unrecognised number", &$2);
2412 tok_err(c, "error: unsupported number suffix", &$2);
2415 elements = mpz_get_ui(mpq_numref(num));
2416 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2417 tok_err(c, "error: array size must be an integer",
2419 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2420 tok_err(c, "error: array size is too large",
2425 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2426 t->array.size = elements;
2427 t->array.member = $<4;
2428 t->array.vsize = NULL;
2431 | [ IDENTIFIER ] Type ${ {
2432 struct variable *v = var_ref(c, $2.txt);
2435 tok_err(c, "error: name undeclared", &$2);
2436 else if (!v->constant)
2437 tok_err(c, "error: array size must be a constant", &$2);
2439 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2440 $0->array.member = $<4;
2442 $0->array.vsize = v;
2447 OptType -> Type ${ $0 = $<1; }$
2450 ###### formal type grammar
2452 | [ IDENTIFIER :: OptType ] Type ${ {
2453 struct variable *v = var_decl(c, $ID.txt);
2459 $0 = add_anon_type(c, &array_prototype, "array[var]");
2460 $0->array.member = $<6;
2462 $0->array.unspec = 1;
2463 $0->array.vsize = v;
2471 | Term [ Expression ] ${ {
2472 struct binode *b = new(binode);
2479 ###### print binode cases
2481 print_exec(b->left, -1, bracket);
2483 print_exec(b->right, -1, bracket);
2487 ###### propagate binode cases
2489 /* left must be an array, right must be a number,
2490 * result is the member type of the array
2492 propagate_types(b->right, c, perr, Tnum, 0);
2493 t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
2494 if (!t || t->compat != array_compat) {
2495 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2498 if (!type_compat(type, t->array.member, rules)) {
2499 type_err(c, "error: have %1 but need %2", prog,
2500 t->array.member, rules, type);
2502 return t->array.member;
2506 ###### interp binode cases
2512 lleft = linterp_exec(c, b->left, <ype);
2513 right = interp_exec(c, b->right, &rtype);
2515 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2519 if (ltype->array.static_size)
2522 ptr = *(void**)lleft;
2523 rvtype = ltype->array.member;
2524 if (i >= 0 && i < ltype->array.size)
2525 lrv = ptr + i * rvtype->size;
2527 val_init(ltype->array.member, &rv); // UNSAFE
2534 A `struct` is a data-type that contains one or more other data-types.
2535 It differs from an array in that each member can be of a different
2536 type, and they are accessed by name rather than by number. Thus you
2537 cannot choose an element by calculation, you need to know what you
2540 The language makes no promises about how a given structure will be
2541 stored in memory - it is free to rearrange fields to suit whatever
2542 criteria seems important.
2544 Structs are declared separately from program code - they cannot be
2545 declared in-line in a variable declaration like arrays can. A struct
2546 is given a name and this name is used to identify the type - the name
2547 is not prefixed by the word `struct` as it would be in C.
2549 Structs are only treated as the same if they have the same name.
2550 Simply having the same fields in the same order is not enough. This
2551 might change once we can create structure initializers from a list of
2554 Each component datum is identified much like a variable is declared,
2555 with a name, one or two colons, and a type. The type cannot be omitted
2556 as there is no opportunity to deduce the type from usage. An initial
2557 value can be given following an equals sign, so
2559 ##### Example: a struct type
2565 would declare a type called "complex" which has two number fields,
2566 each initialised to zero.
2568 Struct will need to be declared separately from the code that uses
2569 them, so we will need to be able to print out the declaration of a
2570 struct when reprinting the whole program. So a `print_type_decl` type
2571 function will be needed.
2573 ###### type union fields
2582 } *fields; // This is created when field_list is analysed.
2584 struct fieldlist *prev;
2587 } *field_list; // This is created during parsing
2590 ###### type functions
2591 void (*print_type_decl)(struct type *type, FILE *f);
2593 ###### value functions
2595 static void structure_init(struct type *type, struct value *val)
2599 for (i = 0; i < type->structure.nfields; i++) {
2601 v = (void*) val->ptr + type->structure.fields[i].offset;
2602 if (type->structure.fields[i].init)
2603 dup_value(type->structure.fields[i].type,
2604 type->structure.fields[i].init,
2607 val_init(type->structure.fields[i].type, v);
2611 static void structure_free(struct type *type, struct value *val)
2615 for (i = 0; i < type->structure.nfields; i++) {
2617 v = (void*)val->ptr + type->structure.fields[i].offset;
2618 free_value(type->structure.fields[i].type, v);
2622 static void free_fieldlist(struct fieldlist *f)
2626 free_fieldlist(f->prev);
2631 static void structure_free_type(struct type *t)
2634 for (i = 0; i < t->structure.nfields; i++)
2635 if (t->structure.fields[i].init) {
2636 free_value(t->structure.fields[i].type,
2637 t->structure.fields[i].init);
2639 free(t->structure.fields);
2640 free_fieldlist(t->structure.field_list);
2643 static int structure_prepare_type(struct parse_context *c,
2644 struct type *t, int parse_time)
2647 struct fieldlist *f;
2649 if (!parse_time || t->structure.fields)
2652 for (f = t->structure.field_list; f; f=f->prev) {
2656 if (f->f.type->size <= 0)
2658 if (f->f.type->prepare_type)
2659 f->f.type->prepare_type(c, f->f.type, parse_time);
2661 if (f->init == NULL)
2665 propagate_types(f->init, c, &perr, f->f.type, 0);
2666 } while (perr & Eretry);
2668 c->parse_error += 1; // NOTEST
2671 t->structure.nfields = cnt;
2672 t->structure.fields = calloc(cnt, sizeof(struct field));
2673 f = t->structure.field_list;
2675 int a = f->f.type->align;
2677 t->structure.fields[cnt] = f->f;
2678 if (t->size & (a-1))
2679 t->size = (t->size | (a-1)) + 1;
2680 t->structure.fields[cnt].offset = t->size;
2681 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2685 if (f->init && !c->parse_error) {
2686 struct value vl = interp_exec(c, f->init, NULL);
2687 t->structure.fields[cnt].init =
2688 global_alloc(c, f->f.type, NULL, &vl);
2696 static struct type structure_prototype = {
2697 .init = structure_init,
2698 .free = structure_free,
2699 .free_type = structure_free_type,
2700 .print_type_decl = structure_print_type,
2701 .prepare_type = structure_prepare_type,
2715 ###### free exec cases
2717 free_exec(cast(fieldref, e)->left);
2721 ###### declare terminals
2726 | Term . IDENTIFIER ${ {
2727 struct fieldref *fr = new_pos(fieldref, $2);
2734 ###### print exec cases
2738 struct fieldref *f = cast(fieldref, e);
2739 print_exec(f->left, -1, bracket);
2740 printf(".%.*s", f->name.len, f->name.txt);
2744 ###### ast functions
2745 static int find_struct_index(struct type *type, struct text field)
2748 for (i = 0; i < type->structure.nfields; i++)
2749 if (text_cmp(type->structure.fields[i].name, field) == 0)
2754 ###### propagate exec cases
2758 struct fieldref *f = cast(fieldref, prog);
2759 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2762 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2764 else if (st->init != structure_init)
2765 type_err(c, "error: field reference attempted on %1, not a struct",
2766 f->left, st, 0, NULL);
2767 else if (f->index == -2) {
2768 f->index = find_struct_index(st, f->name);
2770 type_err(c, "error: cannot find requested field in %1",
2771 f->left, st, 0, NULL);
2773 if (f->index >= 0) {
2774 struct type *ft = st->structure.fields[f->index].type;
2775 if (!type_compat(type, ft, rules))
2776 type_err(c, "error: have %1 but need %2", prog,
2783 ###### interp exec cases
2786 struct fieldref *f = cast(fieldref, e);
2788 struct value *lleft = linterp_exec(c, f->left, <ype);
2789 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2790 rvtype = ltype->structure.fields[f->index].type;
2794 ###### top level grammar
2795 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2797 t = find_type(c, $ID.txt);
2799 t = add_type(c, $ID.txt, &structure_prototype);
2800 else if (t->size >= 0) {
2801 tok_err(c, "error: type already declared", &$ID);
2802 tok_err(c, "info: this is location of declartion", &t->first_use);
2803 /* Create a new one - duplicate */
2804 t = add_type(c, $ID.txt, &structure_prototype);
2806 struct type tmp = *t;
2807 *t = structure_prototype;
2811 t->structure.field_list = $<FB;
2816 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2817 | { SimpleFieldList } ${ $0 = $<SFL; }$
2818 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2819 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2821 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2822 | FieldLines SimpleFieldList Newlines ${
2827 SimpleFieldList -> Field ${ $0 = $<F; }$
2828 | SimpleFieldList ; Field ${
2832 | SimpleFieldList ; ${
2835 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2837 Field -> IDENTIFIER : Type = Expression ${ {
2838 $0 = calloc(1, sizeof(struct fieldlist));
2839 $0->f.name = $ID.txt;
2840 $0->f.type = $<Type;
2844 | IDENTIFIER : Type ${
2845 $0 = calloc(1, sizeof(struct fieldlist));
2846 $0->f.name = $ID.txt;
2847 $0->f.type = $<Type;
2850 ###### forward decls
2851 static void structure_print_type(struct type *t, FILE *f);
2853 ###### value functions
2854 static void structure_print_type(struct type *t, FILE *f)
2858 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2860 for (i = 0; i < t->structure.nfields; i++) {
2861 struct field *fl = t->structure.fields + i;
2862 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2863 type_print(fl->type, f);
2864 if (fl->type->print && fl->init) {
2866 if (fl->type == Tstr)
2867 fprintf(f, "\""); // UNTESTED
2868 print_value(fl->type, fl->init, f);
2869 if (fl->type == Tstr)
2870 fprintf(f, "\""); // UNTESTED
2876 ###### print type decls
2881 while (target != 0) {
2883 for (t = context.typelist; t ; t=t->next)
2884 if (!t->anon && t->print_type_decl &&
2894 t->print_type_decl(t, stdout);
2902 A function is a chunk of code which can be passed parameters and can
2903 return results. Each function has a type which includes the set of
2904 parameters and the return value. As yet these types cannot be declared
2905 separately from the function itself.
2907 The parameters can be specified either in parentheses as a ';' separated
2910 ##### Example: function 1
2912 func main(av:[ac::number]string; env:[envc::number]string)
2915 or as an indented list of one parameter per line (though each line can
2916 be a ';' separated list)
2918 ##### Example: function 2
2921 argv:[argc::number]string
2922 env:[envc::number]string
2926 In the first case a return type can follow the parentheses after a colon,
2927 in the second it is given on a line starting with the word `return`.
2929 ##### Example: functions that return
2931 func add(a:number; b:number): number
2941 Rather than returning a type, the function can specify a set of local
2942 variables to return as a struct. The values of these variables when the
2943 function exits will be provided to the caller. For this the return type
2944 is replaced with a block of result declarations, either in parentheses
2945 or bracketed by `return` and `do`.
2947 ##### Example: functions returning multiple variables
2949 func to_cartesian(rho:number; theta:number):(x:number; y:number)
2962 For constructing the lists we use a `List` binode, which will be
2963 further detailed when Expression Lists are introduced.
2965 ###### type union fields
2968 struct binode *params;
2969 struct type *return_type;
2970 struct variable *scope;
2971 int inline_result; // return value is at start of 'local'
2975 ###### value union fields
2976 struct exec *function;
2978 ###### type functions
2979 void (*check_args)(struct parse_context *c, enum prop_err *perr,
2980 struct type *require, struct exec *args);
2982 ###### value functions
2984 static void function_free(struct type *type, struct value *val)
2986 free_exec(val->function);
2987 val->function = NULL;
2990 static int function_compat(struct type *require, struct type *have)
2992 // FIXME can I do anything here yet?
2996 static void function_check_args(struct parse_context *c, enum prop_err *perr,
2997 struct type *require, struct exec *args)
2999 /* This should be 'compat', but we don't have a 'tuple' type to
3000 * hold the type of 'args'
3002 struct binode *arg = cast(binode, args);
3003 struct binode *param = require->function.params;
3006 struct var *pv = cast(var, param->left);
3008 type_err(c, "error: insufficient arguments to function.",
3009 args, NULL, 0, NULL);
3013 propagate_types(arg->left, c, perr, pv->var->type, 0);
3014 param = cast(binode, param->right);
3015 arg = cast(binode, arg->right);
3018 type_err(c, "error: too many arguments to function.",
3019 args, NULL, 0, NULL);
3022 static void function_print(struct type *type, struct value *val, FILE *f)
3024 print_exec(val->function, 1, 0);
3027 static void function_print_type_decl(struct type *type, FILE *f)
3031 for (b = type->function.params; b; b = cast(binode, b->right)) {
3032 struct variable *v = cast(var, b->left)->var;
3033 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3034 v->constant ? "::" : ":");
3035 type_print(v->type, f);
3040 if (type->function.return_type != Tnone) {
3042 if (type->function.inline_result) {
3044 struct type *t = type->function.return_type;
3046 for (i = 0; i < t->structure.nfields; i++) {
3047 struct field *fl = t->structure.fields + i;
3050 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3051 type_print(fl->type, f);
3055 type_print(type->function.return_type, f);
3060 static void function_free_type(struct type *t)
3062 free_exec(t->function.params);
3065 static struct type function_prototype = {
3066 .size = sizeof(void*),
3067 .align = sizeof(void*),
3068 .free = function_free,
3069 .compat = function_compat,
3070 .check_args = function_check_args,
3071 .print = function_print,
3072 .print_type_decl = function_print_type_decl,
3073 .free_type = function_free_type,
3076 ###### declare terminals
3086 FuncName -> IDENTIFIER ${ {
3087 struct variable *v = var_decl(c, $1.txt);
3088 struct var *e = new_pos(var, $1);
3094 v = var_ref(c, $1.txt);
3096 type_err(c, "error: function '%v' redeclared",
3098 type_err(c, "info: this is where '%v' was first declared",
3099 v->where_decl, NULL, 0, NULL);
3105 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3106 | Args ArgsLine NEWLINE ${ {
3107 struct binode *b = $<AL;
3108 struct binode **bp = &b;
3110 bp = (struct binode **)&(*bp)->left;
3115 ArgsLine -> ${ $0 = NULL; }$
3116 | Varlist ${ $0 = $<1; }$
3117 | Varlist ; ${ $0 = $<1; }$
3119 Varlist -> Varlist ; ArgDecl ${
3133 ArgDecl -> IDENTIFIER : FormalType ${ {
3134 struct variable *v = var_decl(c, $1.txt);
3140 ##### Function calls
3142 A function call can appear either as an expression or as a statement.
3143 We use a new 'Funcall' binode type to link the function with a list of
3144 arguments, form with the 'List' nodes.
3146 We have already seen the "Term" which is how a function call can appear
3147 in an expression. To parse a function call into a statement we include
3148 it in the "SimpleStatement Grammar" which will be described later.
3154 | Term ( ExpressionList ) ${ {
3155 struct binode *b = new(binode);
3158 b->right = reorder_bilist($<EL);
3162 struct binode *b = new(binode);
3169 ###### SimpleStatement Grammar
3171 | Term ( ExpressionList ) ${ {
3172 struct binode *b = new(binode);
3175 b->right = reorder_bilist($<EL);
3179 ###### print binode cases
3182 do_indent(indent, "");
3183 print_exec(b->left, -1, bracket);
3185 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3188 print_exec(b->left, -1, bracket);
3198 ###### propagate binode cases
3201 /* Every arg must match formal parameter, and result
3202 * is return type of function
3204 struct binode *args = cast(binode, b->right);
3205 struct var *v = cast(var, b->left);
3207 if (!v->var->type || v->var->type->check_args == NULL) {
3208 type_err(c, "error: attempt to call a non-function.",
3209 prog, NULL, 0, NULL);
3213 v->var->type->check_args(c, perr, v->var->type, args);
3214 if (v->var->type->function.inline_result)
3216 return v->var->type->function.return_type;
3219 ###### interp binode cases
3222 struct var *v = cast(var, b->left);
3223 struct type *t = v->var->type;
3224 void *oldlocal = c->local;
3225 int old_size = c->local_size;
3226 void *local = calloc(1, t->function.local_size);
3227 struct value *fbody = var_value(c, v->var);
3228 struct binode *arg = cast(binode, b->right);
3229 struct binode *param = t->function.params;
3232 struct var *pv = cast(var, param->left);
3233 struct type *vtype = NULL;
3234 struct value val = interp_exec(c, arg->left, &vtype);
3236 c->local = local; c->local_size = t->function.local_size;
3237 lval = var_value(c, pv->var);
3238 c->local = oldlocal; c->local_size = old_size;
3239 memcpy(lval, &val, vtype->size);
3240 param = cast(binode, param->right);
3241 arg = cast(binode, arg->right);
3243 c->local = local; c->local_size = t->function.local_size;
3244 if (t->function.inline_result && dtype) {
3245 _interp_exec(c, fbody->function, NULL, NULL);
3246 memcpy(dest, local, dtype->size);
3247 rvtype = ret.type = NULL;
3249 rv = interp_exec(c, fbody->function, &rvtype);
3250 c->local = oldlocal; c->local_size = old_size;
3255 ## Complex executables: statements and expressions
3257 Now that we have types and values and variables and most of the basic
3258 Terms which provide access to these, we can explore the more complex
3259 code that combine all of these to get useful work done. Specifically
3260 statements and expressions.
3262 Expressions are various combinations of Terms. We will use operator
3263 precedence to ensure correct parsing. The simplest Expression is just a
3264 Term - others will follow.
3269 Expression -> Term ${ $0 = $<Term; }$
3270 ## expression grammar
3272 ### Expressions: Conditional
3274 Our first user of the `binode` will be conditional expressions, which
3275 is a bit odd as they actually have three components. That will be
3276 handled by having 2 binodes for each expression. The conditional
3277 expression is the lowest precedence operator which is why we define it
3278 first - to start the precedence list.
3280 Conditional expressions are of the form "value `if` condition `else`
3281 other_value". They associate to the right, so everything to the right
3282 of `else` is part of an else value, while only a higher-precedence to
3283 the left of `if` is the if values. Between `if` and `else` there is no
3284 room for ambiguity, so a full conditional expression is allowed in
3290 ###### declare terminals
3294 ###### expression grammar
3296 | Expression if Expression else Expression $$ifelse ${ {
3297 struct binode *b1 = new(binode);
3298 struct binode *b2 = new(binode);
3308 ###### print binode cases
3311 b2 = cast(binode, b->right);
3312 if (bracket) printf("(");
3313 print_exec(b2->left, -1, bracket);
3315 print_exec(b->left, -1, bracket);
3317 print_exec(b2->right, -1, bracket);
3318 if (bracket) printf(")");
3321 ###### propagate binode cases
3324 /* cond must be Tbool, others must match */
3325 struct binode *b2 = cast(binode, b->right);
3328 propagate_types(b->left, c, perr, Tbool, 0);
3329 t = propagate_types(b2->left, c, perr, type, Rnolabel);
3330 t2 = propagate_types(b2->right, c, perr, type ?: t, Rnolabel);
3334 ###### interp binode cases
3337 struct binode *b2 = cast(binode, b->right);
3338 left = interp_exec(c, b->left, <ype);
3340 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3342 rv = interp_exec(c, b2->right, &rvtype);
3348 We take a brief detour, now that we have expressions, to describe lists
3349 of expressions. These will be needed for function parameters and
3350 possibly other situations. They seem generic enough to introduce here
3351 to be used elsewhere.
3353 And ExpressionList will use the `List` type of `binode`, building up at
3354 the end. And place where they are used will probably call
3355 `reorder_bilist()` to get a more normal first/next arrangement.
3357 ###### declare terminals
3360 `List` execs have no implicit semantics, so they are never propagated or
3361 interpreted. The can be printed as a comma separate list, which is how
3362 they are parsed. Note they are also used for function formal parameter
3363 lists. In that case a separate function is used to print them.
3365 ###### print binode cases
3369 print_exec(b->left, -1, bracket);
3372 b = cast(binode, b->right);
3376 ###### propagate binode cases
3377 case List: abort(); // NOTEST
3378 ###### interp binode cases
3379 case List: abort(); // NOTEST
3384 ExpressionList -> ExpressionList , Expression ${
3397 ### Expressions: Boolean
3399 The next class of expressions to use the `binode` will be Boolean
3400 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3401 have same corresponding precendence. The difference is that they don't
3402 evaluate the second expression if not necessary.
3411 ###### declare terminals
3416 ###### expression grammar
3417 | Expression or Expression ${ {
3418 struct binode *b = new(binode);
3424 | Expression or else Expression ${ {
3425 struct binode *b = new(binode);
3432 | Expression and Expression ${ {
3433 struct binode *b = new(binode);
3439 | Expression and then Expression ${ {
3440 struct binode *b = new(binode);
3447 | not Expression ${ {
3448 struct binode *b = new(binode);
3454 ###### print binode cases
3456 if (bracket) printf("(");
3457 print_exec(b->left, -1, bracket);
3459 print_exec(b->right, -1, bracket);
3460 if (bracket) printf(")");
3463 if (bracket) printf("(");
3464 print_exec(b->left, -1, bracket);
3465 printf(" and then ");
3466 print_exec(b->right, -1, bracket);
3467 if (bracket) printf(")");
3470 if (bracket) printf("(");
3471 print_exec(b->left, -1, bracket);
3473 print_exec(b->right, -1, bracket);
3474 if (bracket) printf(")");
3477 if (bracket) printf("(");
3478 print_exec(b->left, -1, bracket);
3479 printf(" or else ");
3480 print_exec(b->right, -1, bracket);
3481 if (bracket) printf(")");
3484 if (bracket) printf("(");
3486 print_exec(b->right, -1, bracket);
3487 if (bracket) printf(")");
3490 ###### propagate binode cases
3496 /* both must be Tbool, result is Tbool */
3497 propagate_types(b->left, c, perr, Tbool, 0);
3498 propagate_types(b->right, c, perr, Tbool, 0);
3499 if (type && type != Tbool)
3500 type_err(c, "error: %1 operation found where %2 expected", prog,
3504 ###### interp binode cases
3506 rv = interp_exec(c, b->left, &rvtype);
3507 right = interp_exec(c, b->right, &rtype);
3508 rv.bool = rv.bool && right.bool;
3511 rv = interp_exec(c, b->left, &rvtype);
3513 rv = interp_exec(c, b->right, NULL);
3516 rv = interp_exec(c, b->left, &rvtype);
3517 right = interp_exec(c, b->right, &rtype);
3518 rv.bool = rv.bool || right.bool;
3521 rv = interp_exec(c, b->left, &rvtype);
3523 rv = interp_exec(c, b->right, NULL);
3526 rv = interp_exec(c, b->right, &rvtype);
3530 ### Expressions: Comparison
3532 Of slightly higher precedence that Boolean expressions are Comparisons.
3533 A comparison takes arguments of any comparable type, but the two types
3536 To simplify the parsing we introduce an `eop` which can record an
3537 expression operator, and the `CMPop` non-terminal will match one of them.
3544 ###### ast functions
3545 static void free_eop(struct eop *e)
3559 ###### declare terminals
3560 $LEFT < > <= >= == != CMPop
3562 ###### expression grammar
3563 | Expression CMPop Expression ${ {
3564 struct binode *b = new(binode);
3574 CMPop -> < ${ $0.op = Less; }$
3575 | > ${ $0.op = Gtr; }$
3576 | <= ${ $0.op = LessEq; }$
3577 | >= ${ $0.op = GtrEq; }$
3578 | == ${ $0.op = Eql; }$
3579 | != ${ $0.op = NEql; }$
3581 ###### print binode cases
3589 if (bracket) printf("(");
3590 print_exec(b->left, -1, bracket);
3592 case Less: printf(" < "); break;
3593 case LessEq: printf(" <= "); break;
3594 case Gtr: printf(" > "); break;
3595 case GtrEq: printf(" >= "); break;
3596 case Eql: printf(" == "); break;
3597 case NEql: printf(" != "); break;
3598 default: abort(); // NOTEST
3600 print_exec(b->right, -1, bracket);
3601 if (bracket) printf(")");
3604 ###### propagate binode cases
3611 /* Both must match but not be labels, result is Tbool */
3612 t = propagate_types(b->left, c, perr, NULL, Rnolabel);
3614 propagate_types(b->right, c, perr, t, 0);
3616 t = propagate_types(b->right, c, perr, NULL, Rnolabel); // UNTESTED
3618 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
3620 if (!type_compat(type, Tbool, 0))
3621 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3622 Tbool, rules, type);
3625 ###### interp binode cases
3634 left = interp_exec(c, b->left, <ype);
3635 right = interp_exec(c, b->right, &rtype);
3636 cmp = value_cmp(ltype, rtype, &left, &right);
3639 case Less: rv.bool = cmp < 0; break;
3640 case LessEq: rv.bool = cmp <= 0; break;
3641 case Gtr: rv.bool = cmp > 0; break;
3642 case GtrEq: rv.bool = cmp >= 0; break;
3643 case Eql: rv.bool = cmp == 0; break;
3644 case NEql: rv.bool = cmp != 0; break;
3645 default: rv.bool = 0; break; // NOTEST
3650 ### Expressions: Arithmetic etc.
3652 The remaining expressions with the highest precedence are arithmetic,
3653 string concatenation, and string conversion. String concatenation
3654 (`++`) has the same precedence as multiplication and division, but lower
3657 String conversion is a temporary feature until I get a better type
3658 system. `$` is a prefix operator which expects a string and returns
3661 `+` and `-` are both infix and prefix operations (where they are
3662 absolute value and negation). These have different operator names.
3664 We also have a 'Bracket' operator which records where parentheses were
3665 found. This makes it easy to reproduce these when printing. Possibly I
3666 should only insert brackets were needed for precedence. Putting
3667 parentheses around an expression converts it into a Term,
3677 ###### declare terminals
3683 ###### expression grammar
3684 | Expression Eop Expression ${ {
3685 struct binode *b = new(binode);
3692 | Expression Top Expression ${ {
3693 struct binode *b = new(binode);
3700 | Uop Expression ${ {
3701 struct binode *b = new(binode);
3709 | ( Expression ) ${ {
3710 struct binode *b = new_pos(binode, $1);
3719 Eop -> + ${ $0.op = Plus; }$
3720 | - ${ $0.op = Minus; }$
3722 Uop -> + ${ $0.op = Absolute; }$
3723 | - ${ $0.op = Negate; }$
3724 | $ ${ $0.op = StringConv; }$
3726 Top -> * ${ $0.op = Times; }$
3727 | / ${ $0.op = Divide; }$
3728 | % ${ $0.op = Rem; }$
3729 | ++ ${ $0.op = Concat; }$
3731 ###### print binode cases
3738 if (bracket) printf("(");
3739 print_exec(b->left, indent, bracket);
3741 case Plus: fputs(" + ", stdout); break;
3742 case Minus: fputs(" - ", stdout); break;
3743 case Times: fputs(" * ", stdout); break;
3744 case Divide: fputs(" / ", stdout); break;
3745 case Rem: fputs(" % ", stdout); break;
3746 case Concat: fputs(" ++ ", stdout); break;
3747 default: abort(); // NOTEST
3749 print_exec(b->right, indent, bracket);
3750 if (bracket) printf(")");
3755 if (bracket) printf("(");
3757 case Absolute: fputs("+", stdout); break;
3758 case Negate: fputs("-", stdout); break;
3759 case StringConv: fputs("$", stdout); break;
3760 default: abort(); // NOTEST
3762 print_exec(b->right, indent, bracket);
3763 if (bracket) printf(")");
3767 print_exec(b->right, indent, bracket);
3771 ###### propagate binode cases
3777 /* both must be numbers, result is Tnum */
3780 /* as propagate_types ignores a NULL,
3781 * unary ops fit here too */
3782 propagate_types(b->left, c, perr, Tnum, 0);
3783 propagate_types(b->right, c, perr, Tnum, 0);
3784 if (!type_compat(type, Tnum, 0))
3785 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3790 /* both must be Tstr, result is Tstr */
3791 propagate_types(b->left, c, perr, Tstr, 0);
3792 propagate_types(b->right, c, perr, Tstr, 0);
3793 if (!type_compat(type, Tstr, 0))
3794 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3799 /* op must be string, result is number */
3800 propagate_types(b->left, c, perr, Tstr, 0);
3801 if (!type_compat(type, Tnum, 0))
3802 type_err(c, // UNTESTED
3803 "error: Can only convert string to number, not %1",
3804 prog, type, 0, NULL);
3808 return propagate_types(b->right, c, perr, type, 0);
3810 ###### interp binode cases
3813 rv = interp_exec(c, b->left, &rvtype);
3814 right = interp_exec(c, b->right, &rtype);
3815 mpq_add(rv.num, rv.num, right.num);
3818 rv = interp_exec(c, b->left, &rvtype);
3819 right = interp_exec(c, b->right, &rtype);
3820 mpq_sub(rv.num, rv.num, right.num);
3823 rv = interp_exec(c, b->left, &rvtype);
3824 right = interp_exec(c, b->right, &rtype);
3825 mpq_mul(rv.num, rv.num, right.num);
3828 rv = interp_exec(c, b->left, &rvtype);
3829 right = interp_exec(c, b->right, &rtype);
3830 mpq_div(rv.num, rv.num, right.num);
3835 left = interp_exec(c, b->left, <ype);
3836 right = interp_exec(c, b->right, &rtype);
3837 mpz_init(l); mpz_init(r); mpz_init(rem);
3838 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3839 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3840 mpz_tdiv_r(rem, l, r);
3841 val_init(Tnum, &rv);
3842 mpq_set_z(rv.num, rem);
3843 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3848 rv = interp_exec(c, b->right, &rvtype);
3849 mpq_neg(rv.num, rv.num);
3852 rv = interp_exec(c, b->right, &rvtype);
3853 mpq_abs(rv.num, rv.num);
3856 rv = interp_exec(c, b->right, &rvtype);
3859 left = interp_exec(c, b->left, <ype);
3860 right = interp_exec(c, b->right, &rtype);
3862 rv.str = text_join(left.str, right.str);
3865 right = interp_exec(c, b->right, &rvtype);
3869 struct text tx = right.str;
3872 if (tx.txt[0] == '-') {
3873 neg = 1; // UNTESTED
3874 tx.txt++; // UNTESTED
3875 tx.len--; // UNTESTED
3877 if (number_parse(rv.num, tail, tx) == 0)
3878 mpq_init(rv.num); // UNTESTED
3880 mpq_neg(rv.num, rv.num); // UNTESTED
3882 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3886 ###### value functions
3888 static struct text text_join(struct text a, struct text b)
3891 rv.len = a.len + b.len;
3892 rv.txt = malloc(rv.len);
3893 memcpy(rv.txt, a.txt, a.len);
3894 memcpy(rv.txt+a.len, b.txt, b.len);
3898 ### Blocks, Statements, and Statement lists.
3900 Now that we have expressions out of the way we need to turn to
3901 statements. There are simple statements and more complex statements.
3902 Simple statements do not contain (syntactic) newlines, complex statements do.
3904 Statements often come in sequences and we have corresponding simple
3905 statement lists and complex statement lists.
3906 The former comprise only simple statements separated by semicolons.
3907 The later comprise complex statements and simple statement lists. They are
3908 separated by newlines. Thus the semicolon is only used to separate
3909 simple statements on the one line. This may be overly restrictive,
3910 but I'm not sure I ever want a complex statement to share a line with
3913 Note that a simple statement list can still use multiple lines if
3914 subsequent lines are indented, so
3916 ###### Example: wrapped simple statement list
3921 is a single simple statement list. This might allow room for
3922 confusion, so I'm not set on it yet.
3924 A simple statement list needs no extra syntax. A complex statement
3925 list has two syntactic forms. It can be enclosed in braces (much like
3926 C blocks), or it can be introduced by an indent and continue until an
3927 unindented newline (much like Python blocks). With this extra syntax
3928 it is referred to as a block.
3930 Note that a block does not have to include any newlines if it only
3931 contains simple statements. So both of:
3933 if condition: a=b; d=f
3935 if condition { a=b; print f }
3939 In either case the list is constructed from a `binode` list with
3940 `Block` as the operator. When parsing the list it is most convenient
3941 to append to the end, so a list is a list and a statement. When using
3942 the list it is more convenient to consider a list to be a statement
3943 and a list. So we need a function to re-order a list.
3944 `reorder_bilist` serves this purpose.
3946 The only stand-alone statement we introduce at this stage is `pass`
3947 which does nothing and is represented as a `NULL` pointer in a `Block`
3948 list. Other stand-alone statements will follow once the infrastructure
3951 As many statements will use binodes, we declare a binode pointer 'b' in
3952 the common header for all reductions to use.
3954 ###### Parser: reduce
3965 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3966 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3967 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3968 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3969 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3971 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3972 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3973 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3974 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3975 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3977 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3978 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3979 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3981 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3982 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3983 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3984 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3985 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3987 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3989 ComplexStatements -> ComplexStatements ComplexStatement ${
3999 | ComplexStatement ${
4011 ComplexStatement -> SimpleStatements Newlines ${
4012 $0 = reorder_bilist($<SS);
4014 | SimpleStatements ; Newlines ${
4015 $0 = reorder_bilist($<SS);
4017 ## ComplexStatement Grammar
4020 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4026 | SimpleStatement ${
4035 SimpleStatement -> pass ${ $0 = NULL; }$
4036 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4037 ## SimpleStatement Grammar
4039 ###### print binode cases
4043 if (b->left == NULL) // UNTESTED
4044 printf("pass"); // UNTESTED
4046 print_exec(b->left, indent, bracket); // UNTESTED
4047 if (b->right) { // UNTESTED
4048 printf("; "); // UNTESTED
4049 print_exec(b->right, indent, bracket); // UNTESTED
4052 // block, one per line
4053 if (b->left == NULL)
4054 do_indent(indent, "pass\n");
4056 print_exec(b->left, indent, bracket);
4058 print_exec(b->right, indent, bracket);
4062 ###### propagate binode cases
4065 /* If any statement returns something other than Tnone
4066 * or Tbool then all such must return same type.
4067 * As each statement may be Tnone or something else,
4068 * we must always pass NULL (unknown) down, otherwise an incorrect
4069 * error might occur. We never return Tnone unless it is
4074 for (e = b; e; e = cast(binode, e->right)) {
4075 t = propagate_types(e->left, c, perr, NULL, rules);
4076 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4078 if (t == Tnone && e->right)
4079 /* Only the final statement *must* return a value
4087 type_err(c, "error: expected %1%r, found %2",
4088 e->left, type, rules, t);
4094 ###### interp binode cases
4096 while (rvtype == Tnone &&
4099 rv = interp_exec(c, b->left, &rvtype);
4100 b = cast(binode, b->right);
4104 ### The Print statement
4106 `print` is a simple statement that takes a comma-separated list of
4107 expressions and prints the values separated by spaces and terminated
4108 by a newline. No control of formatting is possible.
4110 `print` uses `ExpressionList` to collect the expressions and stores them
4111 on the left side of a `Print` binode unlessthere is a trailing comma
4112 when the list is stored on the `right` side and no trailing newline is
4118 ##### declare terminals
4121 ###### SimpleStatement Grammar
4123 | print ExpressionList ${
4124 $0 = b = new(binode);
4127 b->left = reorder_bilist($<EL);
4129 | print ExpressionList , ${ {
4130 $0 = b = new(binode);
4132 b->right = reorder_bilist($<EL);
4136 $0 = b = new(binode);
4142 ###### print binode cases
4145 do_indent(indent, "print");
4147 print_exec(b->right, -1, bracket);
4150 print_exec(b->left, -1, bracket);
4155 ###### propagate binode cases
4158 /* don't care but all must be consistent */
4160 b = cast(binode, b->left);
4162 b = cast(binode, b->right);
4164 propagate_types(b->left, c, perr, NULL, Rnolabel);
4165 b = cast(binode, b->right);
4169 ###### interp binode cases
4173 struct binode *b2 = cast(binode, b->left);
4175 b2 = cast(binode, b->right);
4176 for (; b2; b2 = cast(binode, b2->right)) {
4177 left = interp_exec(c, b2->left, <ype);
4178 print_value(ltype, &left, stdout);
4179 free_value(ltype, &left);
4183 if (b->right == NULL)
4189 ###### Assignment statement
4191 An assignment will assign a value to a variable, providing it hasn't
4192 been declared as a constant. The analysis phase ensures that the type
4193 will be correct so the interpreter just needs to perform the
4194 calculation. There is a form of assignment which declares a new
4195 variable as well as assigning a value. If a name is assigned before
4196 it is declared, and error will be raised as the name is created as
4197 `Tlabel` and it is illegal to assign to such names.
4203 ###### declare terminals
4206 ###### SimpleStatement Grammar
4207 | Term = Expression ${
4208 $0 = b= new(binode);
4213 | VariableDecl = Expression ${
4214 $0 = b= new(binode);
4221 if ($1->var->where_set == NULL) {
4223 "Variable declared with no type or value: %v",
4227 $0 = b = new(binode);
4234 ###### print binode cases
4237 do_indent(indent, "");
4238 print_exec(b->left, indent, bracket);
4240 print_exec(b->right, indent, bracket);
4247 struct variable *v = cast(var, b->left)->var;
4248 do_indent(indent, "");
4249 print_exec(b->left, indent, bracket);
4250 if (cast(var, b->left)->var->constant) {
4252 if (v->explicit_type) {
4253 type_print(v->type, stdout);
4258 if (v->explicit_type) {
4259 type_print(v->type, stdout);
4265 print_exec(b->right, indent, bracket);
4272 ###### propagate binode cases
4276 /* Both must match and not be labels,
4277 * Type must support 'dup',
4278 * For Assign, left must not be constant.
4281 t = propagate_types(b->left, c, perr, NULL,
4282 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
4287 if (propagate_types(b->right, c, perr, t, 0) != t)
4288 if (b->left->type == Xvar)
4289 type_err(c, "info: variable '%v' was set as %1 here.",
4290 cast(var, b->left)->var->where_set, t, rules, NULL);
4292 t = propagate_types(b->right, c, perr, NULL, Rnolabel);
4294 propagate_types(b->left, c, perr, t,
4295 (b->op == Assign ? Rnoconstant : 0));
4297 if (t && t->dup == NULL && !(*perr & Emaycopy))
4298 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4303 ###### interp binode cases
4306 lleft = linterp_exec(c, b->left, <ype);
4308 dinterp_exec(c, b->right, lleft, ltype, 1);
4314 struct variable *v = cast(var, b->left)->var;
4317 val = var_value(c, v);
4318 if (v->type->prepare_type)
4319 v->type->prepare_type(c, v->type, 0);
4321 dinterp_exec(c, b->right, val, v->type, 0);
4323 val_init(v->type, val);
4327 ### The `use` statement
4329 The `use` statement is the last "simple" statement. It is needed when a
4330 statement block can return a value. This includes the body of a
4331 function which has a return type, and the "condition" code blocks in
4332 `if`, `while`, and `switch` statements.
4337 ###### declare terminals
4340 ###### SimpleStatement Grammar
4342 $0 = b = new_pos(binode, $1);
4345 if (b->right->type == Xvar) {
4346 struct var *v = cast(var, b->right);
4347 if (v->var->type == Tnone) {
4348 /* Convert this to a label */
4351 v->var->type = Tlabel;
4352 val = global_alloc(c, Tlabel, v->var, NULL);
4358 ###### print binode cases
4361 do_indent(indent, "use ");
4362 print_exec(b->right, -1, bracket);
4367 ###### propagate binode cases
4370 /* result matches value */
4371 return propagate_types(b->right, c, perr, type, 0);
4373 ###### interp binode cases
4376 rv = interp_exec(c, b->right, &rvtype);
4379 ### The Conditional Statement
4381 This is the biggy and currently the only complex statement. This
4382 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4383 It is comprised of a number of parts, all of which are optional though
4384 set combinations apply. Each part is (usually) a key word (`then` is
4385 sometimes optional) followed by either an expression or a code block,
4386 except the `casepart` which is a "key word and an expression" followed
4387 by a code block. The code-block option is valid for all parts and,
4388 where an expression is also allowed, the code block can use the `use`
4389 statement to report a value. If the code block does not report a value
4390 the effect is similar to reporting `True`.
4392 The `else` and `case` parts, as well as `then` when combined with
4393 `if`, can contain a `use` statement which will apply to some
4394 containing conditional statement. `for` parts, `do` parts and `then`
4395 parts used with `for` can never contain a `use`, except in some
4396 subordinate conditional statement.
4398 If there is a `forpart`, it is executed first, only once.
4399 If there is a `dopart`, then it is executed repeatedly providing
4400 always that the `condpart` or `cond`, if present, does not return a non-True
4401 value. `condpart` can fail to return any value if it simply executes
4402 to completion. This is treated the same as returning `True`.
4404 If there is a `thenpart` it will be executed whenever the `condpart`
4405 or `cond` returns True (or does not return any value), but this will happen
4406 *after* `dopart` (when present).
4408 If `elsepart` is present it will be executed at most once when the
4409 condition returns `False` or some value that isn't `True` and isn't
4410 matched by any `casepart`. If there are any `casepart`s, they will be
4411 executed when the condition returns a matching value.
4413 The particular sorts of values allowed in case parts has not yet been
4414 determined in the language design, so nothing is prohibited.
4416 The various blocks in this complex statement potentially provide scope
4417 for variables as described earlier. Each such block must include the
4418 "OpenScope" nonterminal before parsing the block, and must call
4419 `var_block_close()` when closing the block.
4421 The code following "`if`", "`switch`" and "`for`" does not get its own
4422 scope, but is in a scope covering the whole statement, so names
4423 declared there cannot be redeclared elsewhere. Similarly the
4424 condition following "`while`" is in a scope the covers the body
4425 ("`do`" part) of the loop, and which does not allow conditional scope
4426 extension. Code following "`then`" (both looping and non-looping),
4427 "`else`" and "`case`" each get their own local scope.
4429 The type requirements on the code block in a `whilepart` are quite
4430 unusal. It is allowed to return a value of some identifiable type, in
4431 which case the loop aborts and an appropriate `casepart` is run, or it
4432 can return a Boolean, in which case the loop either continues to the
4433 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4434 This is different both from the `ifpart` code block which is expected to
4435 return a Boolean, or the `switchpart` code block which is expected to
4436 return the same type as the casepart values. The correct analysis of
4437 the type of the `whilepart` code block is the reason for the
4438 `Rboolok` flag which is passed to `propagate_types()`.
4440 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4441 defined. As there are two scopes which cover multiple parts - one for
4442 the whole statement and one for "while" and "do" - and as we will use
4443 the 'struct exec' to track scopes, we actually need two new types of
4444 exec. One is a `binode` for the looping part, the rest is the
4445 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4446 casepart` to track a list of case parts.
4457 struct exec *action;
4458 struct casepart *next;
4460 struct cond_statement {
4462 struct exec *forpart, *condpart, *thenpart, *elsepart;
4463 struct binode *looppart;
4464 struct casepart *casepart;
4467 ###### ast functions
4469 static void free_casepart(struct casepart *cp)
4473 free_exec(cp->value);
4474 free_exec(cp->action);
4481 static void free_cond_statement(struct cond_statement *s)
4485 free_exec(s->forpart);
4486 free_exec(s->condpart);
4487 free_exec(s->looppart);
4488 free_exec(s->thenpart);
4489 free_exec(s->elsepart);
4490 free_casepart(s->casepart);
4494 ###### free exec cases
4495 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4497 ###### ComplexStatement Grammar
4498 | CondStatement ${ $0 = $<1; }$
4500 ###### declare terminals
4501 $TERM for then while do
4508 // A CondStatement must end with EOL, as does CondSuffix and
4510 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4511 // may or may not end with EOL
4512 // WhilePart and IfPart include an appropriate Suffix
4514 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4515 // them. WhilePart opens and closes its own scope.
4516 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4519 $0->thenpart = $<TP;
4520 $0->looppart = $<WP;
4521 var_block_close(c, CloseSequential, $0);
4523 | ForPart OptNL WhilePart CondSuffix ${
4526 $0->looppart = $<WP;
4527 var_block_close(c, CloseSequential, $0);
4529 | WhilePart CondSuffix ${
4531 $0->looppart = $<WP;
4533 | SwitchPart OptNL CasePart CondSuffix ${
4535 $0->condpart = $<SP;
4536 $CP->next = $0->casepart;
4537 $0->casepart = $<CP;
4538 var_block_close(c, CloseSequential, $0);
4540 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4542 $0->condpart = $<SP;
4543 $CP->next = $0->casepart;
4544 $0->casepart = $<CP;
4545 var_block_close(c, CloseSequential, $0);
4547 | IfPart IfSuffix ${
4549 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4550 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4551 // This is where we close an "if" statement
4552 var_block_close(c, CloseSequential, $0);
4555 CondSuffix -> IfSuffix ${
4558 | Newlines CasePart CondSuffix ${
4560 $CP->next = $0->casepart;
4561 $0->casepart = $<CP;
4563 | CasePart CondSuffix ${
4565 $CP->next = $0->casepart;
4566 $0->casepart = $<CP;
4569 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4570 | Newlines ElsePart ${ $0 = $<EP; }$
4571 | ElsePart ${$0 = $<EP; }$
4573 ElsePart -> else OpenBlock Newlines ${
4574 $0 = new(cond_statement);
4575 $0->elsepart = $<OB;
4576 var_block_close(c, CloseElse, $0->elsepart);
4578 | else OpenScope CondStatement ${
4579 $0 = new(cond_statement);
4580 $0->elsepart = $<CS;
4581 var_block_close(c, CloseElse, $0->elsepart);
4585 CasePart -> case Expression OpenScope ColonBlock ${
4586 $0 = calloc(1,sizeof(struct casepart));
4589 var_block_close(c, CloseParallel, $0->action);
4593 // These scopes are closed in CondStatement
4594 ForPart -> for OpenBlock ${
4598 ThenPart -> then OpenBlock ${
4600 var_block_close(c, CloseSequential, $0);
4604 // This scope is closed in CondStatement
4605 WhilePart -> while UseBlock OptNL do OpenBlock ${
4610 var_block_close(c, CloseSequential, $0->right);
4611 var_block_close(c, CloseSequential, $0);
4613 | while OpenScope Expression OpenScope ColonBlock ${
4618 var_block_close(c, CloseSequential, $0->right);
4619 var_block_close(c, CloseSequential, $0);
4623 IfPart -> if UseBlock OptNL then OpenBlock ${
4626 var_block_close(c, CloseParallel, $0.thenpart);
4628 | if OpenScope Expression OpenScope ColonBlock ${
4631 var_block_close(c, CloseParallel, $0.thenpart);
4633 | if OpenScope Expression OpenScope OptNL then Block ${
4636 var_block_close(c, CloseParallel, $0.thenpart);
4640 // This scope is closed in CondStatement
4641 SwitchPart -> switch OpenScope Expression ${
4644 | switch UseBlock ${
4648 ###### print binode cases
4650 if (b->left && b->left->type == Xbinode &&
4651 cast(binode, b->left)->op == Block) {
4653 do_indent(indent, "while {\n");
4655 do_indent(indent, "while\n");
4656 print_exec(b->left, indent+1, bracket);
4658 do_indent(indent, "} do {\n");
4660 do_indent(indent, "do\n");
4661 print_exec(b->right, indent+1, bracket);
4663 do_indent(indent, "}\n");
4665 do_indent(indent, "while ");
4666 print_exec(b->left, 0, bracket);
4671 print_exec(b->right, indent+1, bracket);
4673 do_indent(indent, "}\n");
4677 ###### print exec cases
4679 case Xcond_statement:
4681 struct cond_statement *cs = cast(cond_statement, e);
4682 struct casepart *cp;
4684 do_indent(indent, "for");
4685 if (bracket) printf(" {\n"); else printf("\n");
4686 print_exec(cs->forpart, indent+1, bracket);
4689 do_indent(indent, "} then {\n");
4691 do_indent(indent, "then\n");
4692 print_exec(cs->thenpart, indent+1, bracket);
4694 if (bracket) do_indent(indent, "}\n");
4697 print_exec(cs->looppart, indent, bracket);
4701 do_indent(indent, "switch");
4703 do_indent(indent, "if");
4704 if (cs->condpart && cs->condpart->type == Xbinode &&
4705 cast(binode, cs->condpart)->op == Block) {
4710 print_exec(cs->condpart, indent+1, bracket);
4712 do_indent(indent, "}\n");
4714 do_indent(indent, "then\n");
4715 print_exec(cs->thenpart, indent+1, bracket);
4719 print_exec(cs->condpart, 0, bracket);
4725 print_exec(cs->thenpart, indent+1, bracket);
4727 do_indent(indent, "}\n");
4732 for (cp = cs->casepart; cp; cp = cp->next) {
4733 do_indent(indent, "case ");
4734 print_exec(cp->value, -1, 0);
4739 print_exec(cp->action, indent+1, bracket);
4741 do_indent(indent, "}\n");
4744 do_indent(indent, "else");
4749 print_exec(cs->elsepart, indent+1, bracket);
4751 do_indent(indent, "}\n");
4756 ###### propagate binode cases
4758 t = propagate_types(b->right, c, perr, Tnone, 0);
4759 if (!type_compat(Tnone, t, 0))
4760 *perr |= Efail; // UNTESTED
4761 return propagate_types(b->left, c, perr, type, rules);
4763 ###### propagate exec cases
4764 case Xcond_statement:
4766 // forpart and looppart->right must return Tnone
4767 // thenpart must return Tnone if there is a loopart,
4768 // otherwise it is like elsepart.
4770 // be bool if there is no casepart
4771 // match casepart->values if there is a switchpart
4772 // either be bool or match casepart->value if there
4774 // elsepart and casepart->action must match the return type
4775 // expected of this statement.
4776 struct cond_statement *cs = cast(cond_statement, prog);
4777 struct casepart *cp;
4779 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
4780 if (!type_compat(Tnone, t, 0))
4781 *perr |= Efail; // UNTESTED
4784 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
4785 if (!type_compat(Tnone, t, 0))
4786 *perr |= Efail; // UNTESTED
4788 if (cs->casepart == NULL) {
4789 propagate_types(cs->condpart, c, perr, Tbool, 0);
4790 propagate_types(cs->looppart, c, perr, Tbool, 0);
4792 /* Condpart must match case values, with bool permitted */
4794 for (cp = cs->casepart;
4795 cp && !t; cp = cp->next)
4796 t = propagate_types(cp->value, c, perr, NULL, 0);
4797 if (!t && cs->condpart)
4798 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
4799 if (!t && cs->looppart)
4800 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
4801 // Now we have a type (I hope) push it down
4803 for (cp = cs->casepart; cp; cp = cp->next)
4804 propagate_types(cp->value, c, perr, t, 0);
4805 propagate_types(cs->condpart, c, perr, t, Rboolok);
4806 propagate_types(cs->looppart, c, perr, t, Rboolok);
4809 // (if)then, else, and case parts must return expected type.
4810 if (!cs->looppart && !type)
4811 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
4813 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
4814 for (cp = cs->casepart;
4816 cp = cp->next) // UNTESTED
4817 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
4820 propagate_types(cs->thenpart, c, perr, type, rules);
4821 propagate_types(cs->elsepart, c, perr, type, rules);
4822 for (cp = cs->casepart; cp ; cp = cp->next)
4823 propagate_types(cp->action, c, perr, type, rules);
4829 ###### interp binode cases
4831 // This just performs one iterration of the loop
4832 rv = interp_exec(c, b->left, &rvtype);
4833 if (rvtype == Tnone ||
4834 (rvtype == Tbool && rv.bool != 0))
4835 // rvtype is Tnone or Tbool, doesn't need to be freed
4836 interp_exec(c, b->right, NULL);
4839 ###### interp exec cases
4840 case Xcond_statement:
4842 struct value v, cnd;
4843 struct type *vtype, *cndtype;
4844 struct casepart *cp;
4845 struct cond_statement *cs = cast(cond_statement, e);
4848 interp_exec(c, cs->forpart, NULL);
4850 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4851 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4852 interp_exec(c, cs->thenpart, NULL);
4854 cnd = interp_exec(c, cs->condpart, &cndtype);
4855 if ((cndtype == Tnone ||
4856 (cndtype == Tbool && cnd.bool != 0))) {
4857 // cnd is Tnone or Tbool, doesn't need to be freed
4858 rv = interp_exec(c, cs->thenpart, &rvtype);
4859 // skip else (and cases)
4863 for (cp = cs->casepart; cp; cp = cp->next) {
4864 v = interp_exec(c, cp->value, &vtype);
4865 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4866 free_value(vtype, &v);
4867 free_value(cndtype, &cnd);
4868 rv = interp_exec(c, cp->action, &rvtype);
4871 free_value(vtype, &v);
4873 free_value(cndtype, &cnd);
4875 rv = interp_exec(c, cs->elsepart, &rvtype);
4882 ### Top level structure
4884 All the language elements so far can be used in various places. Now
4885 it is time to clarify what those places are.
4887 At the top level of a file there will be a number of declarations.
4888 Many of the things that can be declared haven't been described yet,
4889 such as functions, procedures, imports, and probably more.
4890 For now there are two sorts of things that can appear at the top
4891 level. They are predefined constants, `struct` types, and the `main`
4892 function. While the syntax will allow the `main` function to appear
4893 multiple times, that will trigger an error if it is actually attempted.
4895 The various declarations do not return anything. They store the
4896 various declarations in the parse context.
4898 ###### Parser: grammar
4901 Ocean -> OptNL DeclarationList
4903 ## declare terminals
4911 DeclarationList -> Declaration
4912 | DeclarationList Declaration
4914 Declaration -> ERROR Newlines ${
4915 tok_err(c, // UNTESTED
4916 "error: unhandled parse error", &$1);
4922 ## top level grammar
4926 ### The `const` section
4928 As well as being defined in with the code that uses them, constants can
4929 be declared at the top level. These have full-file scope, so they are
4930 always `InScope`, even before(!) they have been declared. The value of
4931 a top level constant can be given as an expression, and this is
4932 evaluated after parsing and before execution.
4934 A function call can be used to evaluate a constant, but it will not have
4935 access to any program state, once such statement becomes meaningful.
4936 e.g. arguments and filesystem will not be visible.
4938 Constants are defined in a section that starts with the reserved word
4939 `const` and then has a block with a list of assignment statements.
4940 For syntactic consistency, these must use the double-colon syntax to
4941 make it clear that they are constants. Type can also be given: if
4942 not, the type will be determined during analysis, as with other
4945 ###### parse context
4946 struct binode *constlist;
4948 ###### top level grammar
4952 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4953 | const { SimpleConstList } Newlines
4954 | const IN OptNL ConstList OUT Newlines
4955 | const SimpleConstList Newlines
4957 ConstList -> ConstList SimpleConstLine
4960 SimpleConstList -> SimpleConstList ; Const
4964 SimpleConstLine -> SimpleConstList Newlines
4965 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4968 CType -> Type ${ $0 = $<1; }$
4972 Const -> IDENTIFIER :: CType = Expression ${ {
4974 struct binode *bl, *bv;
4975 struct var *var = new_pos(var, $ID);
4977 v = var_decl(c, $ID.txt);
4979 v->where_decl = var;
4985 v = var_ref(c, $1.txt);
4986 if (v->type == Tnone) {
4987 v->where_decl = var;
4993 tok_err(c, "error: name already declared", &$1);
4994 type_err(c, "info: this is where '%v' was first declared",
4995 v->where_decl, NULL, 0, NULL);
5007 bl->left = c->constlist;
5012 ###### core functions
5013 static void resolve_consts(struct parse_context *c)
5017 enum { none, some, cannot } progress = none;
5019 c->constlist = reorder_bilist(c->constlist);
5022 for (b = cast(binode, c->constlist); b;
5023 b = cast(binode, b->right)) {
5025 struct binode *vb = cast(binode, b->left);
5026 struct var *v = cast(var, vb->left);
5027 if (v->var->frame_pos >= 0)
5031 propagate_types(vb->right, c, &perr,
5033 } while (perr & Eretry);
5035 c->parse_error += 1;
5036 else if (!(perr & Enoconst)) {
5038 struct value res = interp_exec(
5039 c, vb->right, &v->var->type);
5040 global_alloc(c, v->var->type, v->var, &res);
5042 if (progress == cannot)
5043 type_err(c, "error: const %v cannot be resolved.",
5053 progress = cannot; break;
5055 progress = none; break;
5060 ###### print const decls
5065 for (b = cast(binode, context.constlist); b;
5066 b = cast(binode, b->right)) {
5067 struct binode *vb = cast(binode, b->left);
5068 struct var *vr = cast(var, vb->left);
5069 struct variable *v = vr->var;
5075 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5076 type_print(v->type, stdout);
5078 print_exec(vb->right, -1, 0);
5083 ###### free const decls
5084 free_binode(context.constlist);
5086 ### Function declarations
5088 The code in an Ocean program is all stored in function declarations.
5089 One of the functions must be named `main` and it must accept an array of
5090 strings as a parameter - the command line arguments.
5092 As this is the top level, several things are handled a bit differently.
5093 The function is not interpreted by `interp_exec` as that isn't passed
5094 the argument list which the program requires. Similarly type analysis
5095 is a bit more interesting at this level.
5097 ###### ast functions
5099 static struct type *handle_results(struct parse_context *c,
5100 struct binode *results)
5102 /* Create a 'struct' type from the results list, which
5103 * is a list for 'struct var'
5105 struct type *t = add_anon_type(c, &structure_prototype,
5110 for (b = results; b; b = cast(binode, b->right))
5112 t->structure.nfields = cnt;
5113 t->structure.fields = calloc(cnt, sizeof(struct field));
5115 for (b = results; b; b = cast(binode, b->right)) {
5116 struct var *v = cast(var, b->left);
5117 struct field *f = &t->structure.fields[cnt++];
5118 int a = v->var->type->align;
5119 f->name = v->var->name->name;
5120 f->type = v->var->type;
5122 f->offset = t->size;
5123 v->var->frame_pos = f->offset;
5124 t->size += ((f->type->size - 1) | (a-1)) + 1;
5127 variable_unlink_exec(v->var);
5129 free_binode(results);
5133 static struct variable *declare_function(struct parse_context *c,
5134 struct variable *name,
5135 struct binode *args,
5137 struct binode *results,
5141 struct value fn = {.function = code};
5143 var_block_close(c, CloseFunction, code);
5144 t = add_anon_type(c, &function_prototype,
5145 "func %.*s", name->name->name.len,
5146 name->name->name.txt);
5148 t->function.params = reorder_bilist(args);
5150 ret = handle_results(c, reorder_bilist(results));
5151 t->function.inline_result = 1;
5152 t->function.local_size = ret->size;
5154 t->function.return_type = ret;
5155 global_alloc(c, t, name, &fn);
5156 name->type->function.scope = c->out_scope;
5161 var_block_close(c, CloseFunction, NULL);
5163 c->out_scope = NULL;
5167 ###### declare terminals
5170 ###### top level grammar
5173 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5174 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5176 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5177 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5179 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5180 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5182 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5183 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5185 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5186 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5188 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5189 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5191 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5192 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5194 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5195 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5197 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5198 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5201 ###### print func decls
5206 while (target != 0) {
5208 for (v = context.in_scope; v; v=v->in_scope)
5209 if (v->depth == 0 && v->type && v->type->check_args) {
5218 struct value *val = var_value(&context, v);
5219 printf("func %.*s", v->name->name.len, v->name->name.txt);
5220 v->type->print_type_decl(v->type, stdout);
5222 print_exec(val->function, 0, brackets);
5224 print_value(v->type, val, stdout);
5225 printf("/* frame size %d */\n", v->type->function.local_size);
5231 ###### core functions
5233 static int analyse_funcs(struct parse_context *c)
5237 for (v = c->in_scope; v; v = v->in_scope) {
5241 if (v->depth != 0 || !v->type || !v->type->check_args)
5243 ret = v->type->function.inline_result ?
5244 Tnone : v->type->function.return_type;
5245 val = var_value(c, v);
5248 propagate_types(val->function, c, &perr, ret, 0);
5249 } while (!(perr & Efail) && (perr & Eretry));
5250 if (!(perr & Efail))
5251 /* Make sure everything is still consistent */
5252 propagate_types(val->function, c, &perr, ret, 0);
5255 if (!v->type->function.inline_result &&
5256 !v->type->function.return_type->dup) {
5257 type_err(c, "error: function cannot return value of type %1",
5258 v->where_decl, v->type->function.return_type, 0, NULL);
5261 scope_finalize(c, v->type);
5266 static int analyse_main(struct type *type, struct parse_context *c)
5268 struct binode *bp = type->function.params;
5272 struct type *argv_type;
5274 argv_type = add_anon_type(c, &array_prototype, "argv");
5275 argv_type->array.member = Tstr;
5276 argv_type->array.unspec = 1;
5278 for (b = bp; b; b = cast(binode, b->right)) {
5282 propagate_types(b->left, c, &perr, argv_type, 0);
5284 default: /* invalid */ // NOTEST
5285 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5288 c->parse_error += 1;
5291 return !c->parse_error;
5294 static void interp_main(struct parse_context *c, int argc, char **argv)
5296 struct value *progp = NULL;
5297 struct text main_name = { "main", 4 };
5298 struct variable *mainv;
5304 mainv = var_ref(c, main_name);
5306 progp = var_value(c, mainv);
5307 if (!progp || !progp->function) {
5308 fprintf(stderr, "oceani: no main function found.\n");
5309 c->parse_error += 1;
5312 if (!analyse_main(mainv->type, c)) {
5313 fprintf(stderr, "oceani: main has wrong type.\n");
5314 c->parse_error += 1;
5317 al = mainv->type->function.params;
5319 c->local_size = mainv->type->function.local_size;
5320 c->local = calloc(1, c->local_size);
5322 struct var *v = cast(var, al->left);
5323 struct value *vl = var_value(c, v->var);
5333 mpq_set_ui(argcq, argc, 1);
5334 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5335 t->prepare_type(c, t, 0);
5336 array_init(v->var->type, vl);
5337 for (i = 0; i < argc; i++) {
5338 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5340 arg.str.txt = argv[i];
5341 arg.str.len = strlen(argv[i]);
5342 free_value(Tstr, vl2);
5343 dup_value(Tstr, &arg, vl2);
5347 al = cast(binode, al->right);
5349 v = interp_exec(c, progp->function, &vtype);
5350 free_value(vtype, &v);
5355 ###### ast functions
5356 void free_variable(struct variable *v)
5360 ## And now to test it out.
5362 Having a language requires having a "hello world" program. I'll
5363 provide a little more than that: a program that prints "Hello world"
5364 finds the GCD of two numbers, prints the first few elements of
5365 Fibonacci, performs a binary search for a number, and a few other
5366 things which will likely grow as the languages grows.
5368 ###### File: oceani.mk
5371 @echo "===== DEMO ====="
5372 ./oceani --section "demo: hello" oceani.mdc 55 33
5378 four ::= 2 + 2 ; five ::= 10/2
5379 const pie ::= "I like Pie";
5380 cake ::= "The cake is"
5388 func main(argv:[argc::]string)
5389 print "Hello World, what lovely oceans you have!"
5390 print "Are there", five, "?"
5391 print pi, pie, "but", cake
5393 A := $argv[1]; B := $argv[2]
5395 /* When a variable is defined in both branches of an 'if',
5396 * and used afterwards, the variables are merged.
5402 print "Is", A, "bigger than", B,"? ", bigger
5403 /* If a variable is not used after the 'if', no
5404 * merge happens, so types can be different
5407 double:string = "yes"
5408 print A, "is more than twice", B, "?", double
5411 print "double", B, "is", double
5416 if a > 0 and then b > 0:
5422 print "GCD of", A, "and", B,"is", a
5424 print a, "is not positive, cannot calculate GCD"
5426 print b, "is not positive, cannot calculate GCD"
5431 print "Fibonacci:", f1,f2,
5432 then togo = togo - 1
5440 /* Binary search... */
5445 mid := (lo + hi) / 2
5458 print "Yay, I found", target
5460 print "Closest I found was", lo
5465 // "middle square" PRNG. Not particularly good, but one my
5466 // Dad taught me - the first one I ever heard of.
5467 for i:=1; then i = i + 1; while i < size:
5468 n := list[i-1] * list[i-1]
5469 list[i] = (n / 100) % 10 000
5471 print "Before sort:",
5472 for i:=0; then i = i + 1; while i < size:
5476 for i := 1; then i=i+1; while i < size:
5477 for j:=i-1; then j=j-1; while j >= 0:
5478 if list[j] > list[j+1]:
5482 print " After sort:",
5483 for i:=0; then i = i + 1; while i < size:
5487 if 1 == 2 then print "yes"; else print "no"
5491 bob.alive = (bob.name == "Hello")
5492 print "bob", "is" if bob.alive else "isn't", "alive"