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 (which have since been remove), and the
41 "if ... else" trinary operator which can select between two expressions
42 based on a third (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
114 struct parse_context {
115 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s = NULL, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
240 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
241 goto cleanup; // NOTEST
244 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
246 resolve_consts(&context);
247 prepare_types(&context);
248 if (!context.parse_error && !analyse_funcs(&context)) {
249 fprintf(stderr, "oceani: type error in program - not running.\n");
250 context.parse_error += 1;
258 if (doexec && !context.parse_error)
259 interp_main(&context, argc - optind, argv + optind);
262 struct section *t = s->next;
267 // FIXME parser should pop scope even on error
268 while (context.scope_depth > 0)
272 ## free context types
273 ## free context storage
274 exit(context.parse_error ? 1 : 0);
279 The four requirements of parse, analyse, print, interpret apply to
280 each language element individually so that is how most of the code
283 Three of the four are fairly self explanatory. The one that requires
284 a little explanation is the analysis step.
286 The current language design does not require the types of variables to
287 be declared, but they must still have a single type. Different
288 operations impose different requirements on the variables, for example
289 addition requires both arguments to be numeric, and assignment
290 requires the variable on the left to have the same type as the
291 expression on the right.
293 Analysis involves propagating these type requirements around and
294 consequently setting the type of each variable. If any requirements
295 are violated (e.g. a string is compared with a number) or if a
296 variable needs to have two different types, then an error is raised
297 and the program will not run.
299 If the same variable is declared in both branchs of an 'if/else', or
300 in all cases of a 'switch' then the multiple instances may be merged
301 into just one variable if the variable is referenced after the
302 conditional statement. When this happens, the types must naturally be
303 consistent across all the branches. When the variable is not used
304 outside the if, the variables in the different branches are distinct
305 and can be of different types.
307 Undeclared names may only appear in "use" statements and "case" expressions.
308 These names are given a type of "label" and a unique value.
309 This allows them to fill the role of a name in an enumerated type, which
310 is useful for testing the `switch` statement.
312 As we will see, the condition part of a `while` statement can return
313 either a Boolean or some other type. This requires that the expected
314 type that gets passed around comprises a type and a flag to indicate
315 that `Tbool` is also permitted.
317 As there are, as yet, no distinct types that are compatible, there
318 isn't much subtlety in the analysis. When we have distinct number
319 types, this will become more interesting.
323 When analysis discovers an inconsistency it needs to report an error;
324 just refusing to run the code ensures that the error doesn't cascade,
325 but by itself it isn't very useful. A clear understanding of the sort
326 of error message that are useful will help guide the process of
329 At a simplistic level, the only sort of error that type analysis can
330 report is that the type of some construct doesn't match a contextual
331 requirement. For example, in `4 + "hello"` the addition provides a
332 contextual requirement for numbers, but `"hello"` is not a number. In
333 this particular example no further information is needed as the types
334 are obvious from local information. When a variable is involved that
335 isn't the case. It may be helpful to explain why the variable has a
336 particular type, by indicating the location where the type was set,
337 whether by declaration or usage.
339 Using a recursive-descent analysis we can easily detect a problem at
340 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
341 will detect that one argument is not a number and the usage of `hello`
342 will detect that a number was wanted, but not provided. In this
343 (early) version of the language, we will generate error reports at
344 multiple locations, so the use of `hello` will report an error and
345 explain were the value was set, and the addition will report an error
346 and say why numbers are needed. To be able to report locations for
347 errors, each language element will need to record a file location
348 (line and column) and each variable will need to record the language
349 element where its type was set. For now we will assume that each line
350 of an error message indicates one location in the file, and up to 2
351 types. So we provide a `printf`-like function which takes a format, a
352 location (a `struct exec` which has not yet been introduced), and 2
353 types. "`%1`" reports the first type, "`%2`" reports the second. We
354 will need a function to print the location, once we know how that is
355 stored. e As will be explained later, there are sometimes extra rules for
356 type matching and they might affect error messages, we need to pass those
359 As well as type errors, we sometimes need to report problems with
360 tokens, which might be unexpected or might name a type that has not
361 been defined. For these we have `tok_err()` which reports an error
362 with a given token. Each of the error functions sets the flag in the
363 context so indicate that parsing failed.
367 static void fput_loc(struct exec *loc, FILE *f);
368 static void type_err(struct parse_context *c,
369 char *fmt, struct exec *loc,
370 struct type *t1, enum val_rules rules, struct type *t2);
371 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
373 ###### core functions
375 static void type_err(struct parse_context *c,
376 char *fmt, struct exec *loc,
377 struct type *t1, enum val_rules rules, struct type *t2)
379 fprintf(stderr, "%s:", c->file_name);
380 fput_loc(loc, stderr);
381 for (; *fmt ; fmt++) {
388 case '%': fputc(*fmt, stderr); break; // NOTEST
389 default: fputc('?', stderr); break; // NOTEST
391 type_print(t1, stderr);
394 type_print(t2, stderr);
403 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
405 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
406 t->txt.len, t->txt.txt);
410 ## Entities: declared and predeclared.
412 There are various "things" that the language and/or the interpreter
413 needs to know about to parse and execute a program. These include
414 types, variables, values, and executable code. These are all lumped
415 together under the term "entities" (calling them "objects" would be
416 confusing) and introduced here. The following section will present the
417 different specific code elements which comprise or manipulate these
422 Executables can be lots of different things. In many cases an
423 executable is just an operation combined with one or two other
424 executables. This allows for expressions and lists etc. Other times an
425 executable is something quite specific like a constant or variable name.
426 So we define a `struct exec` to be a general executable with a type, and
427 a `struct binode` which is a subclass of `exec`, forms a node in a
428 binary tree, and holds an operation. The simplest operation is "List"
429 which can be used to combine several execs together.
431 There will be other subclasses, and to access these we need to be able
432 to `cast` the `exec` into the various other types. The first field in
433 any `struct exec` is the type from the `exec_types` enum.
436 #define cast(structname, pointer) ({ \
437 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
438 if (__mptr && *__mptr != X##structname) abort(); \
439 (struct structname *)( (char *)__mptr);})
441 #define new(structname) ({ \
442 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
443 __ptr->type = X##structname; \
444 __ptr->line = -1; __ptr->column = -1; \
447 #define new_pos(structname, token) ({ \
448 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
449 __ptr->type = X##structname; \
450 __ptr->line = token.line; __ptr->column = token.col; \
459 enum exec_types type;
469 struct exec *left, *right;
474 static int __fput_loc(struct exec *loc, FILE *f)
478 if (loc->line >= 0) {
479 fprintf(f, "%d:%d: ", loc->line, loc->column);
482 if (loc->type == Xbinode)
483 return __fput_loc(cast(binode,loc)->left, f) ||
484 __fput_loc(cast(binode,loc)->right, f); // NOTEST
487 static void fput_loc(struct exec *loc, FILE *f)
489 if (!__fput_loc(loc, f))
490 fprintf(f, "??:??: "); // NOTEST
493 Each different type of `exec` node needs a number of functions defined,
494 a bit like methods. We must be able to free it, print it, analyse it
495 and execute it. Once we have specific `exec` types we will need to
496 parse them too. Let's take this a bit more slowly.
500 The parser generator requires a `free_foo` function for each struct
501 that stores attributes and they will often be `exec`s and subtypes
502 there-of. So we need `free_exec` which can handle all the subtypes,
503 and we need `free_binode`.
507 static void free_binode(struct binode *b)
516 ###### core functions
517 static void free_exec(struct exec *e)
528 static void free_exec(struct exec *e);
530 ###### free exec cases
531 case Xbinode: free_binode(cast(binode, e)); break;
535 Printing an `exec` requires that we know the current indent level for
536 printing line-oriented components. As will become clear later, we
537 also want to know what sort of bracketing to use. It will also be used
538 to sometime print comments after an exec to explain some of the results
543 static void do_indent(int i, char *str)
550 ###### core functions
551 static void print_binode(struct binode *b, int indent, int bracket)
555 case List: abort(); // must be handled by parent NOTEST
556 ## print binode cases
560 static void print_exec(struct exec *e, int indent, int bracket)
566 print_binode(cast(binode, e), indent, bracket); break;
574 static void print_exec(struct exec *e, int indent, int bracket);
578 As discussed, analysis involves propagating type requirements around the
579 program and looking for errors.
581 So `propagate_types` is passed an expected type (being a `struct type`
582 pointer together with some `val_rules` flags) that the `exec` is
583 expected to return, and returns the type that it does return, either of
584 which can be `NULL` signifying "unknown". A `prop_err` flag set is
585 passed by reference. It has `Efail` set when an error is found, and
586 `Eretry` when the type for some element is set via propagation. If
587 any expression cannot be evaluated a compile time, `Eruntime` is set.
588 If the expression can be copied, `Emaycopy` is set.
590 If `Erval` is set, then the value cannot be assigned to because it is
591 a temporary result. If `Erval` is clear but `Econst` is set, then
592 the value can only be assigned once, when the variable is declared.
596 enum val_rules {Rboolok = 1<<0, Rrefok = 1<<1,};
597 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Eruntime = 1<<2,
598 Emaycopy = 1<<3, Erval = 1<<4, Econst = 1<<5};
601 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
602 struct type *type, enum val_rules rules);
603 ###### core functions
605 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
606 enum prop_err *perr_local,
607 struct type *type, enum val_rules rules)
614 switch (prog->type) {
617 struct binode *b = cast(binode, prog);
619 case List: abort(); // NOTEST
620 ## propagate binode cases
624 ## propagate exec cases
629 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
630 struct type *type, enum val_rules rules)
632 int pre_err = c->parse_error;
633 enum prop_err perr_local = 0;
634 struct type *ret = __propagate_types(prog, c, perr, &perr_local, type, rules);
636 *perr |= perr_local & (Efail | Eretry);
637 if (c->parse_error > pre_err)
644 Interpreting an `exec` doesn't require anything but the `exec`. State
645 is stored in variables and each variable will be directly linked from
646 within the `exec` tree. The exception to this is the `main` function
647 which needs to look at command line arguments. This function will be
648 interpreted separately.
650 Each `exec` can return a value combined with a type in `struct lrval`.
651 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
652 the location of a value, which can be updated, in `lval`. Others will
653 set `lval` to NULL indicating that there is a value of appropriate type
657 static struct value interp_exec(struct parse_context *c, struct exec *e,
658 struct type **typeret);
659 ###### core functions
663 struct value rval, *lval;
666 /* If dest is passed, dtype must give the expected type, and
667 * result can go there, in which case type is returned as NULL.
669 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
670 struct value *dest, struct type *dtype);
672 static struct value interp_exec(struct parse_context *c, struct exec *e,
673 struct type **typeret)
675 struct lrval ret = _interp_exec(c, e, NULL, NULL);
677 if (!ret.type) abort();
681 dup_value(ret.type, ret.lval, &ret.rval);
685 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
686 struct type **typeret)
688 struct lrval ret = _interp_exec(c, e, NULL, NULL);
690 if (!ret.type) abort();
694 free_value(ret.type, &ret.rval);
698 /* dinterp_exec is used when the destination type is certain and
699 * the value has a place to go.
701 static void dinterp_exec(struct parse_context *c, struct exec *e,
702 struct value *dest, struct type *dtype,
705 struct lrval ret = _interp_exec(c, e, dest, dtype);
709 free_value(dtype, dest);
711 dup_value(dtype, ret.lval, dest);
713 memcpy(dest, &ret.rval, dtype->size);
716 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
717 struct value *dest, struct type *dtype)
719 /* If the result is copied to dest, ret.type is set to NULL */
721 struct value rv = {}, *lrv = NULL;
724 rvtype = ret.type = Tnone;
734 struct binode *b = cast(binode, e);
735 struct value left, right, *lleft;
736 struct type *ltype, *rtype;
737 ltype = rtype = Tnone;
739 case List: abort(); // NOTEST
740 ## interp binode cases
742 free_value(ltype, &left);
743 free_value(rtype, &right);
753 ## interp exec cleanup
759 Values come in a wide range of types, with more likely to be added.
760 Each type needs to be able to print its own values (for convenience at
761 least) as well as to compare two values, at least for equality and
762 possibly for order. For now, values might need to be duplicated and
763 freed, though eventually such manipulations will be better integrated
766 Named type are stored in a simple linked list. Objects of each type are
767 "values" which are often passed around by value.
769 There are both explicitly named types, and anonymous types. Anonymous
770 cannot be accessed by name, but are used internally and have a name
771 which might be reported in error messages.
778 ## value union fields
786 struct token first_use;
789 void (*init)(struct type *type, struct value *val);
790 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
791 void (*print)(struct type *type, struct value *val, FILE *f);
792 void (*print_type)(struct type *type, FILE *f);
793 int (*cmp_order)(struct type *t1, struct type *t2,
794 struct value *v1, struct value *v2);
795 int (*cmp_eq)(struct type *t1, struct type *t2,
796 struct value *v1, struct value *v2);
797 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
798 int (*test)(struct type *type, struct value *val);
799 void (*free)(struct type *type, struct value *val);
800 void (*free_type)(struct type *t);
809 struct type *typelist;
816 static struct type *find_type(struct parse_context *c, struct text s)
818 struct type *t = c->typelist;
820 while (t && (t->anon ||
821 text_cmp(t->name, s) != 0))
826 static struct type *_add_type(struct parse_context *c, struct text s,
827 struct type *proto, int anon)
831 n = calloc(1, sizeof(*n));
838 n->next = c->typelist;
843 static struct type *add_type(struct parse_context *c, struct text s,
846 return _add_type(c, s, proto, 0);
849 static struct type *add_anon_type(struct parse_context *c,
850 struct type *proto, char *name, ...)
856 vasprintf(&t.txt, name, ap);
858 t.len = strlen(t.txt);
859 return _add_type(c, t, proto, 1);
862 static struct type *find_anon_type(struct parse_context *c,
863 struct type *proto, char *name, ...)
865 struct type *t = c->typelist;
870 vasprintf(&nm.txt, name, ap);
872 nm.len = strlen(name);
874 while (t && (!t->anon ||
875 text_cmp(t->name, nm) != 0))
881 return _add_type(c, nm, proto, 1);
884 static void free_type(struct type *t)
886 /* The type is always a reference to something in the
887 * context, so we don't need to free anything.
891 static void free_value(struct type *type, struct value *v)
895 memset(v, 0x5a, type->size);
899 static void type_print(struct type *type, FILE *f)
902 fputs("*unknown*type*", f); // NOTEST
903 else if (type->name.len && !type->anon)
904 fprintf(f, "%.*s", type->name.len, type->name.txt);
905 else if (type->print_type)
906 type->print_type(type, f);
907 else if (type->name.len && type->anon)
908 fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
910 fputs("*invalid*type*", f); // NOTEST
913 static void val_init(struct type *type, struct value *val)
915 if (type && type->init)
916 type->init(type, val);
919 static void dup_value(struct type *type,
920 struct value *vold, struct value *vnew)
922 if (type && type->dup)
923 type->dup(type, vold, vnew);
926 static int value_cmp(struct type *tl, struct type *tr,
927 struct value *left, struct value *right)
929 if (tl && tl->cmp_order)
930 return tl->cmp_order(tl, tr, left, right);
931 if (tl && tl->cmp_eq)
932 return tl->cmp_eq(tl, tr, left, right);
936 static void print_value(struct type *type, struct value *v, FILE *f)
938 if (type && type->print)
939 type->print(type, v, f);
941 fprintf(f, "*Unknown*"); // NOTEST
944 static void prepare_types(struct parse_context *c)
948 enum { none, some, cannot } progress = none;
953 for (t = c->typelist; t; t = t->next) {
955 tok_err(c, "error: type used but not declared",
957 if (t->size == 0 && t->prepare_type) {
958 if (t->prepare_type(c, t, 1))
960 else if (progress == cannot)
961 tok_err(c, "error: type has recursive definition",
971 progress = cannot; break;
973 progress = none; break;
980 static void free_value(struct type *type, struct value *v);
981 static int type_compat(struct type *require, struct type *have, enum val_rules rules);
982 static void type_print(struct type *type, FILE *f);
983 static void val_init(struct type *type, struct value *v);
984 static void dup_value(struct type *type,
985 struct value *vold, struct value *vnew);
986 static int value_cmp(struct type *tl, struct type *tr,
987 struct value *left, struct value *right);
988 static void print_value(struct type *type, struct value *v, FILE *f);
990 ###### free context types
992 while (context.typelist) {
993 struct type *t = context.typelist;
995 context.typelist = t->next;
1003 Type can be specified for local variables, for fields in a structure,
1004 for formal parameters to functions, and possibly elsewhere. Different
1005 rules may apply in different contexts. As a minimum, a named type may
1006 always be used. Currently the type of a formal parameter can be
1007 different from types in other contexts, so we have a separate grammar
1013 Type -> IDENTIFIER ${
1014 $0 = find_type(c, $ID.txt);
1016 $0 = add_type(c, $ID.txt, NULL);
1017 $0->first_use = $ID;
1022 FormalType -> Type ${ $0 = $<1; }$
1023 ## formal type grammar
1027 Values of the base types can be numbers, which we represent as
1028 multi-precision fractions, strings, Booleans and labels. When
1029 analysing the program we also need to allow for places where no value
1030 is meaningful (type `Tnone`) and where we don't know what type to
1031 expect yet (type is `NULL`).
1033 Values are never shared, they are always copied when used, and freed
1034 when no longer needed.
1036 When propagating type information around the program, we need to
1037 determine if two types are compatible, where type `NULL` is compatible
1038 with anything. There are two special cases with type compatibility,
1039 both related to the Conditional Statement which will be described
1040 later. In some cases a Boolean can be accepted as well as some other
1041 primary type, and in others any type is acceptable except a label (`Vlabel`).
1042 A separate function encoding these cases will simplify some code later.
1044 ###### type functions
1046 int (*compat)(struct type *this, struct type *other, enum val_rules rules);
1048 ###### ast functions
1050 static int type_compat(struct type *require, struct type *have,
1051 enum val_rules rules)
1053 if ((rules & Rboolok) && have == Tbool)
1055 if (!require || !have)
1058 if (require->compat)
1059 return require->compat(require, have, rules);
1061 return require == have;
1066 #include "parse_string.h"
1067 #include "parse_number.h"
1070 myLDLIBS := libnumber.o libstring.o -lgmp
1071 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1073 ###### type union fields
1074 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1076 ###### value union fields
1082 ###### ast functions
1083 static void _free_value(struct type *type, struct value *v)
1087 switch (type->vtype) {
1089 case Vstr: free(v->str.txt); break;
1090 case Vnum: mpq_clear(v->num); break;
1096 ###### value functions
1098 static void _val_init(struct type *type, struct value *val)
1100 switch(type->vtype) {
1101 case Vnone: // NOTEST
1104 mpq_init(val->num); break;
1106 val->str.txt = malloc(1);
1113 val->label = 0; // NOTEST
1118 static void _dup_value(struct type *type,
1119 struct value *vold, struct value *vnew)
1121 switch (type->vtype) {
1122 case Vnone: // NOTEST
1125 vnew->label = vold->label; // NOTEST
1128 vnew->bool = vold->bool;
1131 mpq_init(vnew->num);
1132 mpq_set(vnew->num, vold->num);
1135 vnew->str.len = vold->str.len;
1136 vnew->str.txt = malloc(vnew->str.len);
1137 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1142 static int _value_cmp(struct type *tl, struct type *tr,
1143 struct value *left, struct value *right)
1148 switch (tl->vtype) {
1149 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1150 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1151 case Vstr: cmp = text_cmp(left->str, right->str); break;
1152 case Vbool: cmp = left->bool - right->bool; break;
1153 case Vnone: cmp = 0; // NOTEST
1158 static void _print_value(struct type *type, struct value *v, FILE *f)
1160 switch (type->vtype) {
1161 case Vnone: // NOTEST
1162 fprintf(f, "*no-value*"); break; // NOTEST
1163 case Vlabel: // NOTEST
1164 fprintf(f, "*label-%d*", v->label); break; // NOTEST
1166 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1168 fprintf(f, "%s", v->bool ? "True":"False"); break;
1173 mpf_set_q(fl, v->num);
1174 gmp_fprintf(f, "%.10Fg", fl);
1181 static void _free_value(struct type *type, struct value *v);
1183 static int bool_test(struct type *type, struct value *v)
1188 static struct type base_prototype = {
1190 .print = _print_value,
1191 .cmp_order = _value_cmp,
1192 .cmp_eq = _value_cmp,
1194 .free = _free_value,
1197 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1199 ###### ast functions
1200 static struct type *add_base_type(struct parse_context *c, char *n,
1201 enum vtype vt, int size)
1203 struct text txt = { n, strlen(n) };
1206 t = add_type(c, txt, &base_prototype);
1209 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1210 if (t->size & (t->align - 1))
1211 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1215 ###### context initialization
1217 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1218 Tbool->test = bool_test;
1219 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1220 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1221 Tnone = add_base_type(&context, "none", Vnone, 0);
1222 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1226 We have already met values as separate objects. When manifest constants
1227 appear in the program text, that must result in an executable which has
1228 a constant value. So the `val` structure embeds a value in an
1241 ###### ast functions
1242 struct val *new_val(struct type *T, struct token tk)
1244 struct val *v = new_pos(val, tk);
1249 ###### declare terminals
1256 $0 = new_val(Tbool, $1);
1260 $0 = new_val(Tbool, $1);
1265 $0 = new_val(Tnum, $1);
1266 if (number_parse($0->val.num, tail, $1.txt) == 0) {
1267 mpq_init($0->val.num);
1268 tok_err(c, "error: unsupported number format", &$NUM);
1270 tok_err(c, "error: unsupported number suffix", &$1);
1274 $0 = new_val(Tstr, $1);
1275 string_parse(&$1, '\\', &$0->val.str, tail);
1277 tok_err(c, "error: unsupported string suffix",
1282 $0 = new_val(Tstr, $1);
1283 string_parse(&$1, '\\', &$0->val.str, tail);
1285 tok_err(c, "error: unsupported string suffix",
1289 ###### print exec cases
1292 struct val *v = cast(val, e);
1293 if (v->vtype == Tstr)
1295 // FIXME how to ensure numbers have same precision.
1296 print_value(v->vtype, &v->val, stdout);
1297 if (v->vtype == Tstr)
1302 ###### propagate exec cases
1305 struct val *val = cast(val, prog);
1306 if (!type_compat(type, val->vtype, rules))
1307 type_err(c, "error: expected %1 found %2",
1308 prog, type, rules, val->vtype);
1313 ###### interp exec cases
1315 rvtype = cast(val, e)->vtype;
1316 dup_value(rvtype, &cast(val, e)->val, &rv);
1319 ###### ast functions
1320 static void free_val(struct val *v)
1323 free_value(v->vtype, &v->val);
1327 ###### free exec cases
1328 case Xval: free_val(cast(val, e)); break;
1330 ###### ast functions
1331 // Move all nodes from 'b' to 'rv', reversing their order.
1332 // In 'b' 'left' is a list, and 'right' is the last node.
1333 // In 'rv', left' is the first node and 'right' is a list.
1334 static struct binode *reorder_bilist(struct binode *b)
1336 struct binode *rv = NULL;
1339 struct exec *t = b->right;
1343 b = cast(binode, b->left);
1353 Labels are a temporary concept until I implement enums. There are an
1354 anonymous enum which is declared by usage. Thet are only allowed in
1355 `use` statements and corresponding `case` entries. They appear as a
1356 period followed by an identifier. All identifiers that are "used" must
1359 For now, we have a global list of labels, and don't check that all "use"
1371 ###### free exec cases
1375 ###### print exec cases
1377 struct label *l = cast(label, e);
1378 printf(".%.*s", l->name.len, l->name.txt);
1384 struct labels *next;
1388 ###### parse context
1389 struct labels *labels;
1391 ###### ast functions
1392 static int label_lookup(struct parse_context *c, struct text name)
1394 struct labels *l, **lp = &c->labels;
1395 while (*lp && text_cmp((*lp)->name, name) < 0)
1397 if (*lp && text_cmp((*lp)->name, name) == 0)
1398 return (*lp)->value;
1399 l = calloc(1, sizeof(*l));
1402 if (c->next_label == 0)
1404 l->value = c->next_label;
1410 ###### free context storage
1411 while (context.labels) {
1412 struct labels *l = context.labels;
1413 context.labels = l->next;
1417 ###### declare terminals
1421 struct label *l = new_pos(label, $ID);
1425 ###### propagate exec cases
1427 struct label *l = cast(label, prog);
1428 l->value = label_lookup(c, l->name);
1429 if (!type_compat(type, Tlabel, rules))
1430 type_err(c, "error: expected %1 found %2",
1431 prog, type, rules, Tlabel);
1435 ###### interp exec cases
1437 struct label *l = cast(label, e);
1438 rv.label = l->value;
1446 Variables are scoped named values. We store the names in a linked list
1447 of "bindings" sorted in lexical order, and use sequential search and
1454 struct binding *next; // in lexical order
1458 This linked list is stored in the parse context so that "reduce"
1459 functions can find or add variables, and so the analysis phase can
1460 ensure that every variable gets a type.
1462 ###### parse context
1464 struct binding *varlist; // In lexical order
1466 ###### ast functions
1468 static struct binding *find_binding(struct parse_context *c, struct text s)
1470 struct binding **l = &c->varlist;
1475 (cmp = text_cmp((*l)->name, s)) < 0)
1479 n = calloc(1, sizeof(*n));
1486 Each name can be linked to multiple variables defined in different
1487 scopes. Each scope starts where the name is declared and continues
1488 until the end of the containing code block. Scopes of a given name
1489 cannot nest, so a declaration while a name is in-scope is an error.
1491 ###### binding fields
1492 struct variable *var;
1496 struct variable *previous;
1498 struct binding *name;
1499 struct exec *where_decl;// where name was declared
1500 struct exec *where_set; // where type was set
1504 When a scope closes, the values of the variables might need to be freed.
1505 This happens in the context of some `struct exec` and each `exec` will
1506 need to know which variables need to be freed when it completes. To
1507 improve visibility, we add a comment when printing any `exec` that
1508 embodies a scope to list the variables that must be freed when it ends.
1511 struct variable *to_free;
1513 ####### variable fields
1514 struct exec *cleanup_exec;
1515 struct variable *next_free;
1517 ####### interp exec cleanup
1520 for (v = e->to_free; v; v = v->next_free) {
1521 struct value *val = var_value(c, v);
1522 free_value(v->type, val);
1526 ###### print exec extras
1529 do_indent(indent, "/* FREE");
1530 for (v = e->to_free; v; v = v->next_free) {
1531 printf(" %.*s", v->name->name.len, v->name->name.txt);
1532 printf("[%d,%d]", v->scope_start, v->scope_end);
1533 if (v->frame_pos >= 0)
1534 printf("(%d+%d)", v->frame_pos,
1535 v->type ? v->type->size:0);
1540 ###### ast functions
1541 static void variable_unlink_exec(struct variable *v)
1543 struct variable **vp;
1544 if (!v->cleanup_exec)
1546 for (vp = &v->cleanup_exec->to_free;
1547 *vp; vp = &(*vp)->next_free) {
1551 v->cleanup_exec = NULL;
1556 While the naming seems strange, we include local constants in the
1557 definition of variables. A name declared `var := value` can
1558 subsequently be changed, but a name declared `var ::= value` cannot -
1561 ###### variable fields
1564 Scopes in parallel branches can be partially merged. More
1565 specifically, if a given name is declared in both branches of an
1566 if/else then its scope is a candidate for merging. Similarly if
1567 every branch of an exhaustive switch (e.g. has an "else" clause)
1568 declares a given name, then the scopes from the branches are
1569 candidates for merging.
1571 Note that names declared inside a loop (which is only parallel to
1572 itself) are never visible after the loop. Similarly names defined in
1573 scopes which are not parallel, such as those started by `for` and
1574 `switch`, are never visible after the scope. Only variables defined in
1575 both `then` and `else` (including the implicit then after an `if`, and
1576 excluding `then` used with `for`) and in all `case`s and `else` of a
1577 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1579 Labels, which are a bit like variables, follow different rules.
1580 Labels are not explicitly declared, but if an undeclared name appears
1581 in a context where a label is legal, that effectively declares the
1582 name as a label. The declaration remains in force (or in scope) at
1583 least to the end of the immediately containing block and conditionally
1584 in any larger containing block which does not declare the name in some
1585 other way. Importantly, the conditional scope extension happens even
1586 if the label is only used in one parallel branch of a conditional --
1587 when used in one branch it is treated as having been declared in all
1590 Merge candidates are tentatively visible beyond the end of the
1591 branching statement which creates them. If the name is used, the
1592 merge is affirmed and they become a single variable visible at the
1593 outer layer. If not - if it is redeclared first - the merge lapses.
1595 To track scopes we have an extra stack, implemented as a linked list,
1596 which roughly parallels the parse stack and which is used exclusively
1597 for scoping. When a new scope is opened, a new frame is pushed and
1598 the child-count of the parent frame is incremented. This child-count
1599 is used to distinguish between the first of a set of parallel scopes,
1600 in which declared variables must not be in scope, and subsequent
1601 branches, whether they may already be conditionally scoped.
1603 We need a total ordering of scopes so we can easily compare to variables
1604 to see if they are concurrently in scope. To achieve this we record a
1605 `scope_count` which is actually a count of both beginnings and endings
1606 of scopes. Then each variable has a record of the scope count where it
1607 enters scope, and where it leaves.
1609 To push a new frame *before* any code in the frame is parsed, we need a
1610 grammar reduction. This is most easily achieved with a grammar
1611 element which derives the empty string, and creates the new scope when
1612 it is recognised. This can be placed, for example, between a keyword
1613 like "if" and the code following it.
1617 struct scope *parent;
1621 ###### parse context
1624 struct scope *scope_stack;
1626 ###### variable fields
1627 int scope_start, scope_end;
1629 ###### ast functions
1630 static void scope_pop(struct parse_context *c)
1632 struct scope *s = c->scope_stack;
1634 c->scope_stack = s->parent;
1636 c->scope_depth -= 1;
1637 c->scope_count += 1;
1640 static void scope_push(struct parse_context *c)
1642 struct scope *s = calloc(1, sizeof(*s));
1644 c->scope_stack->child_count += 1;
1645 s->parent = c->scope_stack;
1647 c->scope_depth += 1;
1648 c->scope_count += 1;
1654 OpenScope -> ${ scope_push(c); }$
1656 Each variable records a scope depth and is in one of four states:
1658 - "in scope". This is the case between the declaration of the
1659 variable and the end of the containing block, and also between
1660 the usage with affirms a merge and the end of that block.
1662 The scope depth is not greater than the current parse context scope
1663 nest depth. When the block of that depth closes, the state will
1664 change. To achieve this, all "in scope" variables are linked
1665 together as a stack in nesting order.
1667 - "pending". The "in scope" block has closed, but other parallel
1668 scopes are still being processed. So far, every parallel block at
1669 the same level that has closed has declared the name.
1671 The scope depth is the depth of the last parallel block that
1672 enclosed the declaration, and that has closed.
1674 - "conditionally in scope". The "in scope" block and all parallel
1675 scopes have closed, and no further mention of the name has been seen.
1676 This state includes a secondary nest depth (`min_depth`) which records
1677 the outermost scope seen since the variable became conditionally in
1678 scope. If a use of the name is found, the variable becomes "in scope"
1679 and that secondary depth becomes the recorded scope depth. If the
1680 name is declared as a new variable, the old variable becomes "out of
1681 scope" and the recorded scope depth stays unchanged.
1683 - "out of scope". The variable is neither in scope nor conditionally
1684 in scope. It is permanently out of scope now and can be removed from
1685 the "in scope" stack. When a variable becomes out-of-scope it is
1686 moved to a separate list (`out_scope`) of variables which have fully
1687 known scope. This will be used at the end of each function to assign
1688 each variable a place in the stack frame.
1690 ###### variable fields
1691 int depth, min_depth;
1692 enum { OutScope, PendingScope, CondScope, InScope } scope;
1693 struct variable *in_scope;
1695 ###### parse context
1697 struct variable *in_scope;
1698 struct variable *out_scope;
1700 All variables with the same name are linked together using the
1701 'previous' link. Those variable that have been affirmatively merged all
1702 have a 'merged' pointer that points to one primary variable - the most
1703 recently declared instance. When merging variables, we need to also
1704 adjust the 'merged' pointer on any other variables that had previously
1705 been merged with the one that will no longer be primary.
1707 A variable that is no longer the most recent instance of a name may
1708 still have "pending" scope, if it might still be merged with most
1709 recent instance. These variables don't really belong in the
1710 "in_scope" list, but are not immediately removed when a new instance
1711 is found. Instead, they are detected and ignored when considering the
1712 list of in_scope names.
1714 The storage of the value of a variable will be described later. For now
1715 we just need to know that when a variable goes out of scope, it might
1716 need to be freed. For this we need to be able to find it, so assume that
1717 `var_value()` will provide that.
1719 ###### variable fields
1720 struct variable *merged;
1722 ###### ast functions
1724 static void variable_merge(struct variable *primary, struct variable *secondary)
1728 primary = primary->merged;
1730 for (v = primary->previous; v; v=v->previous)
1731 if (v == secondary || v == secondary->merged ||
1732 v->merged == secondary ||
1733 v->merged == secondary->merged) {
1734 v->scope = OutScope;
1735 v->merged = primary;
1736 if (v->scope_start < primary->scope_start)
1737 primary->scope_start = v->scope_start;
1738 if (v->scope_end > primary->scope_end)
1739 primary->scope_end = v->scope_end; // NOTEST
1740 variable_unlink_exec(v);
1744 ###### forward decls
1745 static struct value *var_value(struct parse_context *c, struct variable *v);
1747 ###### free global vars
1749 while (context.varlist) {
1750 struct binding *b = context.varlist;
1751 struct variable *v = b->var;
1752 context.varlist = b->next;
1755 struct variable *next = v->previous;
1757 if (v->global && v->frame_pos >= 0) {
1758 free_value(v->type, var_value(&context, v));
1759 if (v->depth == 0 && v->type->free == function_free)
1760 // This is a function constant
1761 free_exec(v->where_decl);
1768 #### Manipulating Bindings
1770 When a name is conditionally visible, a new declaration discards the old
1771 binding - the condition lapses. Similarly when we reach the end of a
1772 function (outermost non-global scope) any conditional scope must lapse.
1773 Conversely a usage of the name affirms the visibility and extends it to
1774 the end of the containing block - i.e. the block that contains both the
1775 original declaration and the latest usage. This is determined from
1776 `min_depth`. When a conditionally visible variable gets affirmed like
1777 this, it is also merged with other conditionally visible variables with
1780 When we parse a variable declaration we either report an error if the
1781 name is currently bound, or create a new variable at the current nest
1782 depth if the name is unbound or bound to a conditionally scoped or
1783 pending-scope variable. If the previous variable was conditionally
1784 scoped, it and its homonyms becomes out-of-scope.
1786 When we parse a variable reference (including non-declarative assignment
1787 "foo = bar") we report an error if the name is not bound or is bound to
1788 a pending-scope variable; update the scope if the name is bound to a
1789 conditionally scoped variable; or just proceed normally if the named
1790 variable is in scope.
1792 When we exit a scope, any variables bound at this level are either
1793 marked out of scope or pending-scoped, depending on whether the scope
1794 was sequential or parallel. Here a "parallel" scope means the "then"
1795 or "else" part of a conditional, or any "case" or "else" branch of a
1796 switch. Other scopes are "sequential".
1798 When exiting a parallel scope we check if there are any variables that
1799 were previously pending and are still visible. If there are, then
1800 they weren't redeclared in the most recent scope, so they cannot be
1801 merged and must become out-of-scope. If it is not the first of
1802 parallel scopes (based on `child_count`), we check that there was a
1803 previous binding that is still pending-scope. If there isn't, the new
1804 variable must now be out-of-scope.
1806 When exiting a sequential scope that immediately enclosed parallel
1807 scopes, we need to resolve any pending-scope variables. If there was
1808 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1809 we need to mark all pending-scope variable as out-of-scope. Otherwise
1810 all pending-scope variables become conditionally scoped.
1813 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1815 ###### ast functions
1817 static struct variable *var_decl(struct parse_context *c, struct text s)
1819 struct binding *b = find_binding(c, s);
1820 struct variable *v = b->var;
1822 switch (v ? v->scope : OutScope) {
1824 /* Caller will report the error */
1828 v && v->scope == CondScope;
1830 v->scope = OutScope;
1834 v = calloc(1, sizeof(*v));
1835 v->previous = b->var;
1839 v->min_depth = v->depth = c->scope_depth;
1841 v->in_scope = c->in_scope;
1842 v->scope_start = c->scope_count;
1848 static struct variable *var_ref(struct parse_context *c, struct text s)
1850 struct binding *b = find_binding(c, s);
1851 struct variable *v = b->var;
1852 struct variable *v2;
1854 switch (v ? v->scope : OutScope) {
1857 /* Caller will report the error */
1860 /* All CondScope variables of this name need to be merged
1861 * and become InScope
1863 v->depth = v->min_depth;
1865 for (v2 = v->previous;
1866 v2 && v2->scope == CondScope;
1868 variable_merge(v, v2);
1876 static int var_refile(struct parse_context *c, struct variable *v)
1878 /* Variable just went out of scope. Add it to the out_scope
1879 * list, sorted by ->scope_start
1881 struct variable **vp = &c->out_scope;
1882 while ((*vp) && (*vp)->scope_start < v->scope_start)
1883 vp = &(*vp)->in_scope;
1889 static void var_block_close(struct parse_context *c, enum closetype ct,
1892 /* Close off all variables that are in_scope.
1893 * Some variables in c->scope may already be not-in-scope,
1894 * such as when a PendingScope variable is hidden by a new
1895 * variable with the same name.
1896 * So we check for v->name->var != v and drop them.
1897 * If we choose to make a variable OutScope, we drop it
1900 struct variable *v, **vp, *v2;
1903 for (vp = &c->in_scope;
1904 (v = *vp) && v->min_depth > c->scope_depth;
1905 (v->scope == OutScope || v->name->var != v)
1906 ? (*vp = v->in_scope, var_refile(c, v))
1907 : ( vp = &v->in_scope, 0)) {
1908 v->min_depth = c->scope_depth;
1909 if (v->name->var != v)
1910 /* This is still in scope, but we haven't just
1914 v->min_depth = c->scope_depth;
1915 if (v->scope == InScope)
1916 v->scope_end = c->scope_count;
1917 if (v->scope == InScope && e && !v->global) {
1918 /* This variable gets cleaned up when 'e' finishes */
1919 variable_unlink_exec(v);
1920 v->cleanup_exec = e;
1921 v->next_free = e->to_free;
1926 case CloseParallel: /* handle PendingScope */
1930 if (c->scope_stack->child_count == 1)
1931 /* first among parallel branches */
1932 v->scope = PendingScope;
1933 else if (v->previous &&
1934 v->previous->scope == PendingScope)
1935 /* all previous branches used name */
1936 v->scope = PendingScope;
1938 v->scope = OutScope;
1939 if (ct == CloseElse) {
1940 /* All Pending variables with this name
1941 * are now Conditional */
1943 v2 && v2->scope == PendingScope;
1945 v2->scope = CondScope;
1949 /* Not possible as it would require
1950 * parallel scope to be nested immediately
1951 * in a parallel scope, and that never
1955 /* Not possible as we already tested for
1962 if (v->scope == CondScope)
1963 /* Condition cannot continue past end of function */
1966 case CloseSequential:
1969 v->scope = OutScope;
1972 /* There was no 'else', so we can only become
1973 * conditional if we know the cases were exhaustive,
1974 * and that doesn't mean anything yet.
1975 * So only labels become conditional..
1978 v2 && v2->scope == PendingScope;
1980 v2->scope = OutScope;
1983 case OutScope: break;
1992 The value of a variable is store separately from the variable, on an
1993 analogue of a stack frame. There are (currently) two frames that can be
1994 active. A global frame which currently only stores constants, and a
1995 stacked frame which stores local variables. Each variable knows if it
1996 is global or not, and what its index into the frame is.
1998 Values in the global frame are known immediately they are relevant, so
1999 the frame needs to be reallocated as it grows so it can store those
2000 values. The local frame doesn't get values until the interpreted phase
2001 is started, so there is no need to allocate until the size is known.
2003 We initialize the `frame_pos` to an impossible value, so that we can
2004 tell if it was set or not later.
2006 ###### variable fields
2010 ###### variable init
2013 ###### parse context
2015 short global_size, global_alloc;
2017 void *global, *local;
2019 ###### forward decls
2020 static struct value *global_alloc(struct parse_context *c, struct type *t,
2021 struct variable *v, struct value *init);
2023 ###### ast functions
2025 static struct value *var_value(struct parse_context *c, struct variable *v)
2028 if (!c->local || !v->type)
2029 return NULL; // NOTEST
2030 if (v->frame_pos + v->type->size > c->local_size) {
2031 printf("INVALID frame_pos\n"); // NOTEST
2034 return c->local + v->frame_pos;
2036 if (c->global_size > c->global_alloc) {
2037 int old = c->global_alloc;
2038 c->global_alloc = (c->global_size | 1023) + 1024;
2039 c->global = realloc(c->global, c->global_alloc);
2040 memset(c->global + old, 0, c->global_alloc - old);
2042 return c->global + v->frame_pos;
2045 static struct value *global_alloc(struct parse_context *c, struct type *t,
2046 struct variable *v, struct value *init)
2049 struct variable scratch;
2051 if (t->prepare_type)
2052 t->prepare_type(c, t, 1); // NOTEST
2054 if (c->global_size & (t->align - 1))
2055 c->global_size = (c->global_size + t->align) & ~(t->align-1);
2060 v->frame_pos = c->global_size;
2062 c->global_size += v->type->size;
2063 ret = var_value(c, v);
2065 memcpy(ret, init, t->size);
2067 val_init(t, ret); // NOTEST
2071 As global values are found -- struct field initializers, labels etc --
2072 `global_alloc()` is called to record the value in the global frame.
2074 When the program is fully parsed, each function is analysed, we need to
2075 walk the list of variables local to that function and assign them an
2076 offset in the stack frame. For this we have `scope_finalize()`.
2078 We keep the stack from dense by re-using space for between variables
2079 that are not in scope at the same time. The `out_scope` list is sorted
2080 by `scope_start` and as we process a varible, we move it to an FIFO
2081 stack. For each variable we consider, we first discard any from the
2082 stack anything that went out of scope before the new variable came in.
2083 Then we place the new variable just after the one at the top of the
2086 ###### ast functions
2088 static void scope_finalize(struct parse_context *c, struct type *ft)
2090 int size = ft->function.local_size;
2091 struct variable *next = ft->function.scope;
2092 struct variable *done = NULL;
2095 struct variable *v = next;
2096 struct type *t = v->type;
2103 if (v->frame_pos >= 0)
2105 while (done && done->scope_end < v->scope_start)
2106 done = done->in_scope;
2108 pos = done->frame_pos + done->type->size;
2110 pos = ft->function.local_size;
2111 if (pos & (t->align - 1))
2112 pos = (pos + t->align) & ~(t->align-1);
2114 if (size < pos + v->type->size)
2115 size = pos + v->type->size;
2119 c->out_scope = NULL;
2120 ft->function.local_size = size;
2123 ###### free context storage
2124 free(context.global);
2126 #### Variables as executables
2128 Just as we used a `val` to wrap a value into an `exec`, we similarly
2129 need a `var` to wrap a `variable` into an exec. While each `val`
2130 contained a copy of the value, each `var` holds a link to the variable
2131 because it really is the same variable no matter where it appears.
2132 When a variable is used, we need to remember to follow the `->merged`
2133 link to find the primary instance.
2135 When a variable is declared, it may or may not be given an explicit
2136 type. We need to record which so that we can report the parsed code
2145 struct variable *var;
2148 ###### variable fields
2156 VariableDecl -> IDENTIFIER : ${ {
2157 struct variable *v = var_decl(c, $1.txt);
2158 $0 = new_pos(var, $1);
2163 v = var_ref(c, $1.txt);
2165 type_err(c, "error: variable '%v' redeclared",
2167 type_err(c, "info: this is where '%v' was first declared",
2168 v->where_decl, NULL, 0, NULL);
2171 | IDENTIFIER :: ${ {
2172 struct variable *v = var_decl(c, $1.txt);
2173 $0 = new_pos(var, $1);
2179 v = var_ref(c, $1.txt);
2181 type_err(c, "error: variable '%v' redeclared",
2183 type_err(c, "info: this is where '%v' was first declared",
2184 v->where_decl, NULL, 0, NULL);
2187 | IDENTIFIER : Type ${ {
2188 struct variable *v = var_decl(c, $1.txt);
2189 $0 = new_pos(var, $1);
2195 v->explicit_type = 1;
2197 v = var_ref(c, $1.txt);
2199 type_err(c, "error: variable '%v' redeclared",
2201 type_err(c, "info: this is where '%v' was first declared",
2202 v->where_decl, NULL, 0, NULL);
2205 | IDENTIFIER :: Type ${ {
2206 struct variable *v = var_decl(c, $1.txt);
2207 $0 = new_pos(var, $1);
2214 v->explicit_type = 1;
2216 v = var_ref(c, $1.txt);
2218 type_err(c, "error: variable '%v' redeclared",
2220 type_err(c, "info: this is where '%v' was first declared",
2221 v->where_decl, NULL, 0, NULL);
2226 Variable -> IDENTIFIER ${ {
2227 struct variable *v = var_ref(c, $1.txt);
2228 $0 = new_pos(var, $1);
2230 /* This might be a global const or a label
2231 * Allocate a var with impossible type Tnone,
2232 * which will be adjusted when we find out what it is,
2233 * or will trigger an error.
2235 v = var_decl(c, $1.txt);
2242 cast(var, $0)->var = v;
2245 ###### print exec cases
2248 struct var *v = cast(var, e);
2250 struct binding *b = v->var->name;
2251 printf("%.*s", b->name.len, b->name.txt);
2258 if (loc && loc->type == Xvar) {
2259 struct var *v = cast(var, loc);
2261 struct binding *b = v->var->name;
2262 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2264 fputs("???", stderr); // NOTEST
2266 fputs("NOTVAR", stderr); // NOTEST
2269 ###### propagate exec cases
2273 struct var *var = cast(var, prog);
2274 struct variable *v = var->var;
2276 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2277 return Tnone; // NOTEST
2280 if (v->type == Tnone && v->where_decl == prog)
2281 type_err(c, "error: variable used but not declared: %v",
2282 prog, NULL, 0, NULL);
2283 if (v->type == NULL) {
2284 if (type && !(*perr & Efail)) {
2286 v->where_set = prog;
2289 } else if (!type_compat(type, v->type, rules)) {
2290 type_err(c, "error: expected %1 but variable '%v' is %2", prog,
2291 type, rules, v->type);
2292 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2293 v->type, rules, NULL);
2295 if (!v->global || v->frame_pos < 0)
2302 ###### interp exec cases
2305 struct var *var = cast(var, e);
2306 struct variable *v = var->var;
2309 lrv = var_value(c, v);
2314 ###### ast functions
2316 static void free_var(struct var *v)
2321 ###### free exec cases
2322 case Xvar: free_var(cast(var, e)); break;
2327 Now that we have the shape of the interpreter in place we can add some
2328 complex types and connected them in to the data structures and the
2329 different phases of parse, analyse, print, interpret.
2331 Being "complex" the language will naturally have syntax to access
2332 specifics of objects of these types. These will fit into the grammar as
2333 "Terms" which are the things that are combined with various operators to
2334 form an "Expression". Where a Term is formed by some operation on another
2335 Term, the subordinate Term will always come first, so for example a
2336 member of an array will be expressed as the Term for the array followed
2337 by an index in square brackets. The strict rule of using postfix
2338 operations makes precedence irrelevant within terms. To provide a place
2339 to put the grammar for terms of each type, we will start out by
2340 introducing the "Term" grammar production, with contains at least a
2341 simple "Value" (to be explained later).
2343 We also take this opportunity to introduce the "ExpressionsList" which
2344 is a simple comma-separated list of expressions - it may be used in
2347 ###### declare terminals
2352 Term -> Value ${ $0 = $<1; }$
2353 | Variable ${ $0 = $<1; }$
2357 ExpressionList -> ExpressionList , Expression ${
2370 Thus far the complex types we have are arrays and structs.
2374 Arrays can be declared by giving a size and a type, as `[size]type' so
2375 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2376 size can be either a literal number, or a named constant. Some day an
2377 arbitrary expression will be supported.
2379 As a formal parameter to a function, the array can be declared with a
2380 new variable as the size: `name:[size::number]string`. The `size`
2381 variable is set to the size of the array and must be a constant. As
2382 `number` is the only supported type, it can be left out:
2383 `name:[size::]string`.
2385 Arrays cannot be assigned. When pointers are introduced we will also
2386 introduce array slices which can refer to part or all of an array -
2387 the assignment syntax will create a slice. For now, an array can only
2388 ever be referenced by the name it is declared with. It is likely that
2389 a "`copy`" primitive will eventually be define which can be used to
2390 make a copy of an array with controllable recursive depth.
2392 For now we have two sorts of array, those with fixed size either because
2393 it is given as a literal number or because it is a struct member (which
2394 cannot have a runtime-changing size), and those with a size that is
2395 determined at runtime - local variables with a const size. The former
2396 have their size calculated at parse time, the latter at run time.
2398 For the latter type, the `size` field of the type is the size of a
2399 pointer, and the array is reallocated every time it comes into scope.
2401 We differentiate struct fields with a const size from local variables
2402 with a const size by whether they are prepared at parse time or not.
2404 ###### type union fields
2407 int unspec; // size is unspecified - vsize must be set.
2410 struct variable *vsize;
2411 struct type *member;
2414 ###### value union fields
2415 void *array; // used if not static_size
2417 ###### value functions
2419 static int array_prepare_type(struct parse_context *c, struct type *type,
2422 struct value *vsize;
2424 if (type->array.static_size)
2425 return 1; // NOTEST - guard against reentry
2426 if (type->array.unspec && parse_time)
2427 return 1; // NOTEST - unspec is still incomplete
2428 if (parse_time && type->array.vsize && !type->array.vsize->global)
2429 return 1; // NOTEST - should be impossible
2431 if (type->array.vsize) {
2432 vsize = var_value(c, type->array.vsize);
2434 return 1; // NOTEST - should be impossible
2436 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2437 type->array.size = mpz_get_si(q);
2442 if (type->array.member->size <= 0)
2443 return 0; // NOTEST - error caught before here
2445 type->array.static_size = 1;
2446 type->size = type->array.size * type->array.member->size;
2447 type->align = type->array.member->align;
2452 static void array_init(struct type *type, struct value *val)
2455 void *ptr = val->ptr;
2459 if (!type->array.static_size) {
2460 val->array = calloc(type->array.size,
2461 type->array.member->size);
2464 for (i = 0; i < type->array.size; i++) {
2466 v = (void*)ptr + i * type->array.member->size;
2467 val_init(type->array.member, v);
2471 static void array_free(struct type *type, struct value *val)
2474 void *ptr = val->ptr;
2476 if (!type->array.static_size)
2478 for (i = 0; i < type->array.size; i++) {
2480 v = (void*)ptr + i * type->array.member->size;
2481 free_value(type->array.member, v);
2483 if (!type->array.static_size)
2487 static int array_compat(struct type *require, struct type *have,
2488 enum val_rules rules)
2490 if (have->compat != require->compat)
2492 /* Both are arrays, so we can look at details */
2493 if (!type_compat(require->array.member, have->array.member, 0))
2495 if (have->array.unspec && require->array.unspec &&
2496 have->array.size != require->array.size)
2498 if (have->array.unspec || require->array.unspec)
2500 if (require->array.vsize == NULL && have->array.vsize == NULL)
2501 return require->array.size == have->array.size;
2503 return require->array.vsize == have->array.vsize;
2506 static void array_print_type(struct type *type, FILE *f)
2509 if (type->array.vsize) {
2510 struct binding *b = type->array.vsize->name;
2511 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2512 type->array.unspec ? "::" : "");
2513 } else if (type->array.size)
2514 fprintf(f, "%d]", type->array.size);
2517 type_print(type->array.member, f);
2520 static struct type array_prototype = {
2522 .prepare_type = array_prepare_type,
2523 .print_type = array_print_type,
2524 .compat = array_compat,
2526 .size = sizeof(void*),
2527 .align = sizeof(void*),
2530 ###### declare terminals
2535 | [ NUMBER ] Type ${ {
2541 if (number_parse(num, tail, $2.txt) == 0)
2542 tok_err(c, "error: unrecognised number", &$2);
2544 tok_err(c, "error: unsupported number suffix", &$2);
2547 elements = mpz_get_ui(mpq_numref(num));
2548 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2549 tok_err(c, "error: array size must be an integer",
2551 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2552 tok_err(c, "error: array size is too large",
2557 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2558 t->array.size = elements;
2559 t->array.member = $<4;
2560 t->array.vsize = NULL;
2563 | [ IDENTIFIER ] Type ${ {
2564 struct variable *v = var_ref(c, $2.txt);
2567 tok_err(c, "error: name undeclared", &$2);
2568 else if (!v->constant)
2569 tok_err(c, "error: array size must be a constant", &$2);
2571 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2572 $0->array.member = $<4;
2574 $0->array.vsize = v;
2577 ###### formal type grammar
2580 $0 = add_anon_type(c, &array_prototype, "array[]");
2581 $0->array.member = $<Type;
2583 $0->array.unspec = 1;
2584 $0->array.vsize = NULL;
2592 | Term [ Expression ] ${ {
2593 struct binode *b = new(binode);
2601 struct binode *b = new(binode);
2607 ###### print binode cases
2609 print_exec(b->left, -1, bracket);
2611 print_exec(b->right, -1, bracket);
2616 print_exec(b->left, -1, bracket);
2620 ###### propagate binode cases
2622 /* left must be an array, right must be a number,
2623 * result is the member type of the array
2625 propagate_types(b->right, c, perr_local, Tnum, 0);
2626 t = propagate_types(b->left, c, perr, NULL, 0);
2627 if (!t || t->compat != array_compat) {
2628 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2631 if (!type_compat(type, t->array.member, rules)) {
2632 type_err(c, "error: have %1 but need %2", prog,
2633 t->array.member, rules, type);
2635 return t->array.member;
2640 /* left must be an array, result is a number
2642 t = propagate_types(b->left, c, perr, NULL, 0);
2643 if (!t || t->compat != array_compat) {
2644 type_err(c, "error: %1 cannot provide length", prog, t, 0, NULL);
2647 if (!type_compat(type, Tnum, rules))
2648 type_err(c, "error: have %1 but need %2", prog,
2653 ###### interp binode cases
2659 lleft = linterp_exec(c, b->left, <ype);
2660 right = interp_exec(c, b->right, &rtype);
2662 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2666 if (ltype->array.static_size)
2669 ptr = *(void**)lleft;
2670 rvtype = ltype->array.member;
2671 if (i >= 0 && i < ltype->array.size)
2672 lrv = ptr + i * rvtype->size;
2674 val_init(ltype->array.member, &rv); // UNSAFE
2679 lleft = linterp_exec(c, b->left, <ype);
2680 mpq_set_ui(rv.num, ltype->array.size, 1);
2688 A `struct` is a data-type that contains one or more other data-types.
2689 It differs from an array in that each member can be of a different
2690 type, and they are accessed by name rather than by number. Thus you
2691 cannot choose an element by calculation, you need to know what you
2694 The language makes no promises about how a given structure will be
2695 stored in memory - it is free to rearrange fields to suit whatever
2696 criteria seems important.
2698 Structs are declared separately from program code - they cannot be
2699 declared in-line in a variable declaration like arrays can. A struct
2700 is given a name and this name is used to identify the type - the name
2701 is not prefixed by the word `struct` as it would be in C.
2703 Structs are only treated as the same if they have the same name.
2704 Simply having the same fields in the same order is not enough. This
2705 might change once we can create structure initializers from a list of
2708 Each component datum is identified much like a variable is declared,
2709 with a name, one or two colons, and a type. The type cannot be omitted
2710 as there is no opportunity to deduce the type from usage. An initial
2711 value can be given following an equals sign, so
2713 ##### Example: a struct type
2719 would declare a type called "complex" which has two number fields,
2720 each initialised to zero.
2722 Struct will need to be declared separately from the code that uses
2723 them, so we will need to be able to print out the declaration of a
2724 struct when reprinting the whole program. So a `print_type_decl` type
2725 function will be needed.
2727 ###### type union fields
2736 } *fields; // This is created when field_list is analysed.
2738 struct fieldlist *prev;
2741 } *field_list; // This is created during parsing
2744 ###### type functions
2745 void (*print_type_decl)(struct type *type, FILE *f);
2746 struct type *(*fieldref)(struct type *t, struct parse_context *c,
2747 struct fieldref *f, struct value **vp);
2749 ###### value functions
2751 static void structure_init(struct type *type, struct value *val)
2755 for (i = 0; i < type->structure.nfields; i++) {
2757 v = (void*) val->ptr + type->structure.fields[i].offset;
2758 if (type->structure.fields[i].init)
2759 dup_value(type->structure.fields[i].type,
2760 type->structure.fields[i].init,
2763 val_init(type->structure.fields[i].type, v);
2767 static void structure_free(struct type *type, struct value *val)
2771 for (i = 0; i < type->structure.nfields; i++) {
2773 v = (void*)val->ptr + type->structure.fields[i].offset;
2774 free_value(type->structure.fields[i].type, v);
2778 static void free_fieldlist(struct fieldlist *f)
2782 free_fieldlist(f->prev);
2787 static void structure_free_type(struct type *t)
2790 for (i = 0; i < t->structure.nfields; i++)
2791 if (t->structure.fields[i].init) {
2792 free_value(t->structure.fields[i].type,
2793 t->structure.fields[i].init);
2795 free(t->structure.fields);
2796 free_fieldlist(t->structure.field_list);
2799 static int structure_prepare_type(struct parse_context *c,
2800 struct type *t, int parse_time)
2803 struct fieldlist *f;
2805 if (!parse_time || t->structure.fields)
2808 for (f = t->structure.field_list; f; f=f->prev) {
2812 if (f->f.type->size <= 0)
2814 if (f->f.type->prepare_type)
2815 f->f.type->prepare_type(c, f->f.type, parse_time);
2817 if (f->init == NULL)
2821 propagate_types(f->init, c, &perr, f->f.type, 0);
2822 } while (perr & Eretry);
2824 c->parse_error += 1; // NOTEST
2827 t->structure.nfields = cnt;
2828 t->structure.fields = calloc(cnt, sizeof(struct field));
2829 f = t->structure.field_list;
2831 int a = f->f.type->align;
2833 t->structure.fields[cnt] = f->f;
2834 if (t->size & (a-1))
2835 t->size = (t->size | (a-1)) + 1;
2836 t->structure.fields[cnt].offset = t->size;
2837 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2841 if (f->init && !c->parse_error) {
2842 struct value vl = interp_exec(c, f->init, NULL);
2843 t->structure.fields[cnt].init =
2844 global_alloc(c, f->f.type, NULL, &vl);
2852 static int find_struct_index(struct type *type, struct text field)
2855 for (i = 0; i < type->structure.nfields; i++)
2856 if (text_cmp(type->structure.fields[i].name, field) == 0)
2858 return IndexInvalid;
2861 static struct type *structure_fieldref(struct type *t, struct parse_context *c,
2862 struct fieldref *f, struct value **vp)
2864 if (f->index == IndexUnknown) {
2865 f->index = find_struct_index(t, f->name);
2867 type_err(c, "error: cannot find requested field in %1",
2868 f->left, t, 0, NULL);
2873 struct value *v = *vp;
2874 v = (void*)v->ptr + t->structure.fields[f->index].offset;
2877 return t->structure.fields[f->index].type;
2880 static struct type structure_prototype = {
2881 .init = structure_init,
2882 .free = structure_free,
2883 .free_type = structure_free_type,
2884 .print_type_decl = structure_print_type,
2885 .prepare_type = structure_prepare_type,
2886 .fieldref = structure_fieldref,
2899 enum { IndexUnknown = -1, IndexInvalid = -2 };
2901 ###### free exec cases
2903 free_exec(cast(fieldref, e)->left);
2907 ###### declare terminals
2912 | Term . IDENTIFIER ${ {
2913 struct fieldref *fr = new_pos(fieldref, $2);
2916 fr->index = IndexUnknown;
2920 ###### print exec cases
2924 struct fieldref *f = cast(fieldref, e);
2925 print_exec(f->left, -1, bracket);
2926 printf(".%.*s", f->name.len, f->name.txt);
2930 ###### propagate exec cases
2934 struct fieldref *f = cast(fieldref, prog);
2935 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2937 if (!st || !st->fieldref)
2938 type_err(c, "error: field reference on %1 is not supported",
2939 f->left, st, 0, NULL);
2941 t = st->fieldref(st, c, f, NULL);
2942 if (t && !type_compat(type, t, rules))
2943 type_err(c, "error: have %1 but need %2", prog,
2950 ###### interp exec cases
2953 struct fieldref *f = cast(fieldref, e);
2955 struct value *lleft = linterp_exec(c, f->left, <ype);
2957 rvtype = ltype->fieldref(ltype, c, f, &lrv);
2961 ###### top level grammar
2963 StructName -> IDENTIFIER ${ {
2964 struct type *t = find_type(c, $ID.txt);
2966 if (t && t->size >= 0) {
2967 tok_err(c, "error: type already declared", &$ID);
2968 tok_err(c, "info: this is location of declartion", &t->first_use);
2972 t = add_type(c, $ID.txt, NULL);
2977 DeclareStruct -> struct StructName FieldBlock Newlines ${ {
2978 struct type *t = $<SN;
2979 struct type tmp = *t;
2981 *t = structure_prototype;
2984 t->first_use = tmp.first_use;
2986 t->structure.field_list = $<FB;
2990 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2991 | { SimpleFieldList } ${ $0 = $<SFL; }$
2992 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2993 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2995 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2996 | FieldLines SimpleFieldList Newlines ${ {
2997 struct fieldlist *f = $<SFL;
3008 SimpleFieldList -> Field ${ $0 = $<F; }$
3009 | SimpleFieldList ; Field ${
3013 | SimpleFieldList ; ${
3016 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
3018 Field -> IDENTIFIER : Type = Expression ${ {
3019 $0 = calloc(1, sizeof(struct fieldlist));
3020 $0->f.name = $ID.txt;
3021 $0->f.type = $<Type;
3025 | IDENTIFIER : Type ${
3026 $0 = calloc(1, sizeof(struct fieldlist));
3027 $0->f.name = $ID.txt;
3028 $0->f.type = $<Type;
3031 ###### forward decls
3032 static void structure_print_type(struct type *t, FILE *f);
3034 ###### value functions
3035 static void structure_print_type(struct type *t, FILE *f)
3039 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
3041 for (i = 0; i < t->structure.nfields; i++) {
3042 struct field *fl = t->structure.fields + i;
3043 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
3044 type_print(fl->type, f);
3045 if (fl->type->print && fl->init) {
3047 if (fl->type == Tstr)
3049 print_value(fl->type, fl->init, f);
3050 if (fl->type == Tstr)
3057 ###### print type decls
3062 while (target != 0) {
3064 for (t = context.typelist; t ; t=t->next)
3065 if (!t->anon && t->print_type_decl &&
3075 t->print_type_decl(t, stdout);
3083 References, or pointers, are values that refer to another value. They
3084 can only refer to a `struct`, though as a struct can embed anything they
3085 can effectively refer to anything.
3087 References are potentially dangerous as they might refer to some
3088 variable which no longer exists - either because a stack frame
3089 containing it has been discarded or because the value was allocated on
3090 the heap and has now been free. Ocean does not yet provide any
3091 protection against these problems. It will in due course.
3093 With references comes the opportunity and the need to explicitly
3094 allocate values on the "heap" and to free them. We currently provide
3095 fairly basic support for this.
3097 Reference make use of the `@` symbol in various ways. A type that starts
3098 with `@` is a reference to whatever follows. A reference value
3099 followed by an `@` acts as the referred value, though the `@` is often
3100 not needed. Finally, an expression that starts with `@` is a special
3101 reference related expression. Some examples might help.
3103 ##### Example: Reference examples
3110 bar.number = 23; bar.string = "hello"
3121 Obviously this is very contrived. `ref` is a reference to a `foo` which
3122 is initially set to refer to the value stored in `bar` - no extra syntax
3123 is needed to "Take the address of" `bar` - the fact that `ref` is a
3124 reference means that only the address make sense.
3126 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
3127 The same syntax is used for accessing fields both in structs and in
3128 references to structs. It would be correct to use `ref@.a`, but not
3131 `@new()` creates an object of whatever type is needed for the program
3132 to by type-correct. In future iterations of Ocean, arguments a
3133 constructor will access arguments, so the the syntax now looks like a
3134 function call. `@free` can be assigned any reference that was returned
3135 by `@new()`, and it will be freed. `@nil` is a value of whatever
3136 reference type is appropriate, and is stable and never the address of
3137 anything in the heap or on the stack. A reference can be assigned
3138 `@nil` or compared against that value.
3140 ###### declare terminals
3143 ###### type union fields
3146 struct type *referent;
3149 ###### value union fields
3152 ###### value functions
3154 static void reference_print_type(struct type *t, FILE *f)
3157 type_print(t->reference.referent, f);
3160 static int reference_cmp(struct type *tl, struct type *tr,
3161 struct value *left, struct value *right)
3163 return left->ref == right->ref ? 0 : 1;
3166 static void reference_dup(struct type *t,
3167 struct value *vold, struct value *vnew)
3169 vnew->ref = vold->ref;
3172 static void reference_free(struct type *t, struct value *v)
3174 /* Nothing to do here */
3177 static int reference_compat(struct type *require, struct type *have,
3178 enum val_rules rules)
3181 if (require->reference.referent == have)
3183 if (have->compat != require->compat)
3185 if (have->reference.referent != require->reference.referent)
3190 static int reference_test(struct type *type, struct value *val)
3192 return val->ref != NULL;
3195 static struct type *reference_fieldref(struct type *t, struct parse_context *c,
3196 struct fieldref *f, struct value **vp)
3198 struct type *rt = t->reference.referent;
3203 return rt->fieldref(rt, c, f, vp);
3205 type_err(c, "error: field reference on %1 is not supported",
3206 f->left, rt, 0, NULL);
3210 static struct type reference_prototype = {
3211 .print_type = reference_print_type,
3212 .cmp_eq = reference_cmp,
3213 .dup = reference_dup,
3214 .test = reference_test,
3215 .free = reference_free,
3216 .compat = reference_compat,
3217 .fieldref = reference_fieldref,
3218 .size = sizeof(void*),
3219 .align = sizeof(void*),
3225 struct type *t = find_type(c, $ID.txt);
3227 t = add_type(c, $ID.txt, NULL);
3230 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3231 $ID.txt.len, $ID.txt.txt);
3232 $0->reference.referent = t;
3235 ###### core functions
3236 static int text_is(struct text t, char *s)
3238 return (strlen(s) == t.len &&
3239 strncmp(s, t.txt, t.len) == 0);
3248 enum ref_func { RefNew, RefFree, RefNil } action;
3249 struct type *reftype;
3253 ###### SimpleStatement Grammar
3255 | @ IDENTIFIER = Expression ${ {
3256 struct ref *r = new_pos(ref, $ID);
3258 if (!text_is($ID.txt, "free"))
3259 tok_err(c, "error: only \"@free\" makes sense here",
3263 r->action = RefFree;
3267 ###### expression grammar
3268 | @ IDENTIFIER ( ) ${
3269 // Only 'new' valid here
3270 if (!text_is($ID.txt, "new")) {
3271 tok_err(c, "error: Only reference function is \"@new()\"",
3274 struct ref *r = new_pos(ref,$ID);
3280 // Only 'nil' valid here
3281 if (!text_is($ID.txt, "nil")) {
3282 tok_err(c, "error: Only reference value is \"@nil\"",
3285 struct ref *r = new_pos(ref,$ID);
3291 ###### print exec cases
3293 struct ref *r = cast(ref, e);
3294 switch (r->action) {
3296 printf("@new()"); break;
3298 printf("@nil"); break;
3300 do_indent(indent, "@free = ");
3301 print_exec(r->right, indent, bracket);
3307 ###### propagate exec cases
3309 struct ref *r = cast(ref, prog);
3310 switch (r->action) {
3312 if (type && type->free != reference_free) {
3313 type_err(c, "error: @new() can only be used with references, not %1",
3314 prog, type, 0, NULL);
3317 if (type && !r->reftype) {
3324 if (type && type->free != reference_free)
3325 type_err(c, "error: @nil can only be used with reference, not %1",
3326 prog, type, 0, NULL);
3327 if (type && !r->reftype) {
3334 t = propagate_types(r->right, c, perr_local, NULL, 0);
3335 if (t && t->free != reference_free)
3336 type_err(c, "error: @free can only be assigned a reference, not %1",
3345 ###### interp exec cases
3347 struct ref *r = cast(ref, e);
3348 switch (r->action) {
3351 rv.ref = calloc(1, r->reftype->reference.referent->size);
3352 rvtype = r->reftype;
3356 rvtype = r->reftype;
3359 rv = interp_exec(c, r->right, &rvtype);
3360 free_value(rvtype->reference.referent, rv.ref);
3368 ###### free exec cases
3370 struct ref *r = cast(ref, e);
3371 free_exec(r->right);
3376 ###### Expressions: dereference
3384 struct binode *b = new(binode);
3390 ###### print binode cases
3392 print_exec(b->left, -1, bracket);
3396 print_exec(b->left, -1, bracket);
3399 ###### propagate binode cases
3401 /* left must be a reference, and we return what it refers to */
3402 /* FIXME how can I pass the expected type down? */
3403 t = propagate_types(b->left, c, perr, NULL, 0);
3405 if (!t || t->free != reference_free)
3406 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3408 return t->reference.referent;
3412 /* left must be lval, we create reference to it */
3413 if (!type || type->free != reference_free)
3414 t = propagate_types(b->left, c, perr, type, 0); // NOTEST impossible
3416 t = propagate_types(b->left, c, perr,
3417 type->reference.referent, 0);
3419 t = find_anon_type(c, &reference_prototype, "@%.*s",
3420 t->name.len, t->name.txt);
3423 ###### interp binode cases
3425 left = interp_exec(c, b->left, <ype);
3427 rvtype = ltype->reference.referent;
3431 rv.ref = linterp_exec(c, b->left, &rvtype);
3432 rvtype = find_anon_type(c, &reference_prototype, "@%.*s",
3433 rvtype->name.len, rvtype->name.txt);
3439 A function is a chunk of code which can be passed parameters and can
3440 return results. Each function has a type which includes the set of
3441 parameters and the return value. As yet these types cannot be declared
3442 separately from the function itself.
3444 The parameters can be specified either in parentheses as a ';' separated
3447 ##### Example: function 1
3449 func main(av:[ac::number]string; env:[envc::number]string)
3452 or as an indented list of one parameter per line (though each line can
3453 be a ';' separated list)
3455 ##### Example: function 2
3458 argv:[argc::number]string
3459 env:[envc::number]string
3463 In the first case a return type can follow the parentheses after a colon,
3464 in the second it is given on a line starting with the word `return`.
3466 ##### Example: functions that return
3468 func add(a:number; b:number): number
3478 Rather than returning a type, the function can specify a set of local
3479 variables to return as a struct. The values of these variables when the
3480 function exits will be provided to the caller. For this the return type
3481 is replaced with a block of result declarations, either in parentheses
3482 or bracketed by `return` and `do`.
3484 ##### Example: functions returning multiple variables
3486 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3499 For constructing the lists we use a `List` binode, which will be
3500 further detailed when Expression Lists are introduced.
3502 ###### type union fields
3505 struct binode *params;
3506 struct type *return_type;
3507 struct variable *scope;
3508 int inline_result; // return value is at start of 'local'
3512 ###### value union fields
3513 struct exec *function;
3515 ###### type functions
3516 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3517 struct type *require, struct exec *args);
3519 ###### value functions
3521 static void function_free(struct type *type, struct value *val)
3523 free_exec(val->function);
3524 val->function = NULL;
3527 static int function_compat(struct type *require, struct type *have,
3528 enum val_rules rules)
3530 // FIXME can I do anything here yet?
3534 static struct exec *take_addr(struct exec *e)
3536 struct binode *rv = new(binode);
3542 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3543 struct type *require, struct exec *args)
3545 /* This should be 'compat', but we don't have a 'tuple' type to
3546 * hold the type of 'args'
3548 struct binode *arg = cast(binode, args);
3549 struct binode *param = require->function.params;
3552 struct var *pv = cast(var, param->left);
3553 struct type *t = pv->var->type, *t2;
3555 type_err(c, "error: insufficient arguments to function.",
3556 args, NULL, 0, NULL);
3560 t2 = propagate_types(arg->left, c, perr, t, Rrefok);
3561 if (t->free == reference_free &&
3562 t->reference.referent == t2 &&
3564 arg->left = take_addr(arg->left);
3565 } else if (!(*perr & Efail) && !type_compat(t2, t, 0)) {
3566 type_err(c, "error: cannot pass rval when reference expected",
3567 arg->left, NULL, 0, NULL);
3569 param = cast(binode, param->right);
3570 arg = cast(binode, arg->right);
3573 type_err(c, "error: too many arguments to function.",
3574 args, NULL, 0, NULL);
3577 static void function_print(struct type *type, struct value *val, FILE *f)
3580 print_exec(val->function, 1, 0);
3583 static void function_print_type_decl(struct type *type, FILE *f)
3587 for (b = type->function.params; b; b = cast(binode, b->right)) {
3588 struct variable *v = cast(var, b->left)->var;
3589 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3590 v->constant ? "::" : ":");
3591 type_print(v->type, f);
3596 if (type->function.return_type != Tnone) {
3598 if (type->function.inline_result) {
3600 struct type *t = type->function.return_type;
3602 for (i = 0; i < t->structure.nfields; i++) {
3603 struct field *fl = t->structure.fields + i;
3606 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3607 type_print(fl->type, f);
3611 type_print(type->function.return_type, f);
3615 static void function_free_type(struct type *t)
3617 free_exec(t->function.params);
3620 static struct type function_prototype = {
3621 .size = sizeof(void*),
3622 .align = sizeof(void*),
3623 .free = function_free,
3624 .compat = function_compat,
3625 .check_args = function_check_args,
3626 .print = function_print,
3627 .print_type_decl = function_print_type_decl,
3628 .free_type = function_free_type,
3631 ###### declare terminals
3638 FuncName -> IDENTIFIER ${ {
3639 struct variable *v = var_decl(c, $1.txt);
3640 struct var *e = new_pos(var, $1);
3647 v = var_ref(c, $1.txt);
3649 type_err(c, "error: function '%v' redeclared",
3651 type_err(c, "info: this is where '%v' was first declared",
3652 v->where_decl, NULL, 0, NULL);
3658 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3659 | Args ArgsLine NEWLINE ${ {
3660 struct binode *b = $<AL;
3661 struct binode **bp = &b;
3663 bp = (struct binode **)&(*bp)->left;
3668 ArgsLine -> ${ $0 = NULL; }$
3669 | Varlist ${ $0 = $<1; }$
3670 | Varlist ; ${ $0 = $<1; }$
3672 Varlist -> Varlist ; ArgDecl ${
3673 $0 = new_pos(binode, $2);
3686 ArgDecl -> IDENTIFIER : FormalType ${ {
3687 struct variable *v = var_decl(c, $ID.txt);
3688 $0 = new_pos(var, $ID);
3695 ##### Function calls
3697 A function call can appear either as an expression or as a statement.
3698 We use a new 'Funcall' binode type to link the function with a list of
3699 arguments, form with the 'List' nodes.
3701 We have already seen the "Term" which is how a function call can appear
3702 in an expression. To parse a function call into a statement we include
3703 it in the "SimpleStatement Grammar" which will be described later.
3709 | Term ( ExpressionList ) ${ {
3710 struct binode *b = new(binode);
3713 b->right = reorder_bilist($<EL);
3717 struct binode *b = new(binode);
3724 ###### SimpleStatement Grammar
3726 | Term ( ExpressionList ) ${ {
3727 struct binode *b = new(binode);
3730 b->right = reorder_bilist($<EL);
3734 ###### print binode cases
3737 do_indent(indent, "");
3738 print_exec(b->left, -1, bracket);
3740 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3743 print_exec(b->left, -1, bracket);
3753 ###### propagate binode cases
3756 /* Every arg must match formal parameter, and result
3757 * is return type of function
3759 struct binode *args = cast(binode, b->right);
3760 struct var *v = cast(var, b->left);
3762 if (!v->var->type || v->var->type->check_args == NULL) {
3763 type_err(c, "error: attempt to call a non-function.",
3764 prog, NULL, 0, NULL);
3768 v->var->type->check_args(c, perr_local, v->var->type, args);
3769 if (v->var->type->function.inline_result)
3772 return v->var->type->function.return_type;
3775 ###### interp binode cases
3778 struct var *v = cast(var, b->left);
3779 struct type *t = v->var->type;
3780 void *oldlocal = c->local;
3781 int old_size = c->local_size;
3782 void *local = calloc(1, t->function.local_size);
3783 struct value *fbody = var_value(c, v->var);
3784 struct binode *arg = cast(binode, b->right);
3785 struct binode *param = t->function.params;
3788 struct var *pv = cast(var, param->left);
3789 struct type *vtype = NULL;
3790 struct value val = interp_exec(c, arg->left, &vtype);
3792 c->local = local; c->local_size = t->function.local_size;
3793 lval = var_value(c, pv->var);
3794 c->local = oldlocal; c->local_size = old_size;
3795 memcpy(lval, &val, vtype->size);
3796 param = cast(binode, param->right);
3797 arg = cast(binode, arg->right);
3799 c->local = local; c->local_size = t->function.local_size;
3800 if (t->function.inline_result && dtype) {
3801 _interp_exec(c, fbody->function, NULL, NULL);
3802 memcpy(dest, local, dtype->size);
3803 rvtype = ret.type = NULL;
3805 rv = interp_exec(c, fbody->function, &rvtype);
3806 c->local = oldlocal; c->local_size = old_size;
3811 ## Complex executables: statements and expressions
3813 Now that we have types and values and variables and most of the basic
3814 Terms which provide access to these, we can explore the more complex
3815 code that combine all of these to get useful work done. Specifically
3816 statements and expressions.
3818 Expressions are various combinations of Terms. We will use operator
3819 precedence to ensure correct parsing. The simplest Expression is just a
3820 Term - others will follow.
3825 Expression -> Term ${ $0 = $<Term; }$
3826 ## expression grammar
3828 ### Expressions: Conditional
3830 Our first user of the `binode` will be conditional expressions, which
3831 is a bit odd as they actually have three components. That will be
3832 handled by having 2 binodes for each expression. The conditional
3833 expression is the lowest precedence operator which is why we define it
3834 first - to start the precedence list.
3836 Conditional expressions are of the form "value `if` condition `else`
3837 other_value". They associate to the right, so everything to the right
3838 of `else` is part of an else value, while only a higher-precedence to
3839 the left of `if` is the if values. Between `if` and `else` there is no
3840 room for ambiguity, so a full conditional expression is allowed in
3846 ###### declare terminals
3850 ###### expression grammar
3852 | Expression if Expression else Expression $$ifelse ${ {
3853 struct binode *b1 = new(binode);
3854 struct binode *b2 = new(binode);
3864 ###### print binode cases
3867 b2 = cast(binode, b->right);
3868 if (bracket) printf("(");
3869 print_exec(b2->left, -1, bracket);
3871 print_exec(b->left, -1, bracket);
3873 print_exec(b2->right, -1, bracket);
3874 if (bracket) printf(")");
3877 ###### propagate binode cases
3880 /* cond must be Tbool, others must match */
3881 struct binode *b2 = cast(binode, b->right);
3884 propagate_types(b->left, c, perr_local, Tbool, 0);
3885 t = propagate_types(b2->left, c, perr, type, 0);
3886 t2 = propagate_types(b2->right, c, perr, type ?: t, 0);
3890 ###### interp binode cases
3893 struct binode *b2 = cast(binode, b->right);
3894 left = interp_exec(c, b->left, <ype);
3896 rv = interp_exec(c, b2->left, &rvtype);
3898 rv = interp_exec(c, b2->right, &rvtype);
3902 ### Expressions: Boolean
3904 The next class of expressions to use the `binode` will be Boolean
3905 expressions. `and` and `or` are short-circuit operators that don't
3906 evaluate the second expression if not necessary.
3913 ###### declare terminals
3918 ###### expression grammar
3919 | Expression or Expression ${ {
3920 struct binode *b = new(binode);
3926 | Expression and Expression ${ {
3927 struct binode *b = new(binode);
3933 | not Expression ${ {
3934 struct binode *b = new(binode);
3940 ###### print binode cases
3942 if (bracket) printf("(");
3943 print_exec(b->left, -1, bracket);
3945 print_exec(b->right, -1, bracket);
3946 if (bracket) printf(")");
3949 if (bracket) printf("(");
3950 print_exec(b->left, -1, bracket);
3952 print_exec(b->right, -1, bracket);
3953 if (bracket) printf(")");
3956 if (bracket) printf("(");
3958 print_exec(b->right, -1, bracket);
3959 if (bracket) printf(")");
3962 ###### propagate binode cases
3966 /* both must be Tbool, result is Tbool */
3967 propagate_types(b->left, c, perr, Tbool, 0);
3968 propagate_types(b->right, c, perr, Tbool, 0);
3969 if (type && type != Tbool)
3970 type_err(c, "error: %1 operation found where %2 expected", prog,
3975 ###### interp binode cases
3977 rv = interp_exec(c, b->left, &rvtype);
3979 rv = interp_exec(c, b->right, NULL);
3982 rv = interp_exec(c, b->left, &rvtype);
3984 rv = interp_exec(c, b->right, NULL);
3987 rv = interp_exec(c, b->right, &rvtype);
3991 ### Expressions: Comparison
3993 Of slightly higher precedence that Boolean expressions are Comparisons.
3994 A comparison takes arguments of any comparable type, but the two types
3997 To simplify the parsing we introduce an `eop` which can record an
3998 expression operator, and the `CMPop` non-terminal will match one of them.
4005 ###### ast functions
4006 static void free_eop(struct eop *e)
4020 ###### declare terminals
4021 $LEFT < > <= >= == != CMPop
4023 ###### expression grammar
4024 | Expression CMPop Expression ${ {
4025 struct binode *b = new(binode);
4035 CMPop -> < ${ $0.op = Less; }$
4036 | > ${ $0.op = Gtr; }$
4037 | <= ${ $0.op = LessEq; }$
4038 | >= ${ $0.op = GtrEq; }$
4039 | == ${ $0.op = Eql; }$
4040 | != ${ $0.op = NEql; }$
4042 ###### print binode cases
4050 if (bracket) printf("(");
4051 print_exec(b->left, -1, bracket);
4053 case Less: printf(" < "); break;
4054 case LessEq: printf(" <= "); break;
4055 case Gtr: printf(" > "); break;
4056 case GtrEq: printf(" >= "); break;
4057 case Eql: printf(" == "); break;
4058 case NEql: printf(" != "); break;
4059 default: abort(); // NOTEST
4061 print_exec(b->right, -1, bracket);
4062 if (bracket) printf(")");
4065 ###### propagate binode cases
4072 /* Both must match but not be labels, result is Tbool */
4073 t = propagate_types(b->left, c, perr, NULL, 0);
4075 propagate_types(b->right, c, perr, t, 0);
4077 t = propagate_types(b->right, c, perr, NULL, 0); // NOTEST
4079 t = propagate_types(b->left, c, perr, t, 0); // NOTEST
4081 if (!type_compat(type, Tbool, 0))
4082 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
4083 Tbool, rules, type);
4087 ###### interp binode cases
4096 left = interp_exec(c, b->left, <ype);
4097 right = interp_exec(c, b->right, &rtype);
4098 cmp = value_cmp(ltype, rtype, &left, &right);
4101 case Less: rv.bool = cmp < 0; break;
4102 case LessEq: rv.bool = cmp <= 0; break;
4103 case Gtr: rv.bool = cmp > 0; break;
4104 case GtrEq: rv.bool = cmp >= 0; break;
4105 case Eql: rv.bool = cmp == 0; break;
4106 case NEql: rv.bool = cmp != 0; break;
4107 default: rv.bool = 0; break; // NOTEST
4112 ### Expressions: Arithmetic etc.
4114 The remaining expressions with the highest precedence are arithmetic,
4115 string concatenation, string conversion, and testing. String concatenation
4116 (`++`) has the same precedence as multiplication and division, but lower
4119 Testing comes in two forms. A single question mark (`?`) is a uniary
4120 operator which converts come types into Boolean. The general meaning is
4121 "is this a value value" and there will be more uses as the language
4122 develops. A double questionmark (`??`) is a binary operator (Choose),
4123 with same precedence as multiplication, which returns the LHS if it
4124 tests successfully, else returns the RHS.
4126 String conversion is a temporary feature until I get a better type
4127 system. `$` is a prefix operator which expects a string and returns
4130 `+` and `-` are both infix and prefix operations (where they are
4131 absolute value and negation). These have different operator names.
4133 We also have a 'Bracket' operator which records where parentheses were
4134 found. This makes it easy to reproduce these when printing. Possibly I
4135 should only insert brackets were needed for precedence. Putting
4136 parentheses around an expression converts it into a Term,
4142 Absolute, Negate, Test,
4146 ###### declare terminals
4148 $LEFT * / % ++ ?? Top
4152 ###### expression grammar
4153 | Expression Eop Expression ${ {
4154 struct binode *b = new(binode);
4161 | Expression Top Expression ${ {
4162 struct binode *b = new(binode);
4169 | Uop Expression ${ {
4170 struct binode *b = new(binode);
4178 | ( Expression ) ${ {
4179 struct binode *b = new_pos(binode, $1);
4188 Eop -> + ${ $0.op = Plus; }$
4189 | - ${ $0.op = Minus; }$
4191 Uop -> + ${ $0.op = Absolute; }$
4192 | - ${ $0.op = Negate; }$
4193 | $ ${ $0.op = StringConv; }$
4194 | ? ${ $0.op = Test; }$
4196 Top -> * ${ $0.op = Times; }$
4197 | / ${ $0.op = Divide; }$
4198 | % ${ $0.op = Rem; }$
4199 | ++ ${ $0.op = Concat; }$
4200 | ?? ${ $0.op = Choose; }$
4202 ###### print binode cases
4210 if (bracket) printf("(");
4211 print_exec(b->left, indent, bracket);
4213 case Plus: fputs(" + ", stdout); break;
4214 case Minus: fputs(" - ", stdout); break;
4215 case Times: fputs(" * ", stdout); break;
4216 case Divide: fputs(" / ", stdout); break;
4217 case Rem: fputs(" % ", stdout); break;
4218 case Concat: fputs(" ++ ", stdout); break;
4219 case Choose: fputs(" ?? ", stdout); break;
4220 default: abort(); // NOTEST
4222 print_exec(b->right, indent, bracket);
4223 if (bracket) printf(")");
4229 if (bracket) printf("(");
4231 case Absolute: fputs("+", stdout); break;
4232 case Negate: fputs("-", stdout); break;
4233 case StringConv: fputs("$", stdout); break;
4234 case Test: fputs("?", stdout); break;
4235 default: abort(); // NOTEST
4237 print_exec(b->right, indent, bracket);
4238 if (bracket) printf(")");
4241 /* Avoid double brackets... */
4242 if (!bracket) printf("(");
4243 print_exec(b->right, indent, bracket);
4244 if (!bracket) printf(")");
4247 ###### propagate binode cases
4253 /* both must be numbers, result is Tnum */
4256 /* as propagate_types ignores a NULL,
4257 * unary ops fit here too */
4258 propagate_types(b->left, c, perr, Tnum, 0);
4259 propagate_types(b->right, c, perr, Tnum, 0);
4260 if (!type_compat(type, Tnum, 0))
4261 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4267 /* both must be Tstr, result is Tstr */
4268 propagate_types(b->left, c, perr, Tstr, 0);
4269 propagate_types(b->right, c, perr, Tstr, 0);
4270 if (!type_compat(type, Tstr, 0))
4271 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4277 /* op must be string, result is number */
4278 propagate_types(b->left, c, perr, Tstr, 0);
4279 if (!type_compat(type, Tnum, 0))
4281 "error: Can only convert string to number, not %1",
4282 prog, type, 0, NULL);
4287 /* LHS must support ->test, result is Tbool */
4288 t = propagate_types(b->right, c, perr, NULL, 0);
4290 type_err(c, "error: '?' requires a testable value, not %1",
4296 /* LHS and RHS must match and are returned. Must support
4299 t = propagate_types(b->left, c, perr, type, rules);
4300 t = propagate_types(b->right, c, perr, t, rules);
4301 if (t && t->test == NULL)
4302 type_err(c, "error: \"??\" requires a testable value, not %1",
4308 return propagate_types(b->right, c, perr, type, rules);
4310 ###### interp binode cases
4313 rv = interp_exec(c, b->left, &rvtype);
4314 right = interp_exec(c, b->right, &rtype);
4315 mpq_add(rv.num, rv.num, right.num);
4318 rv = interp_exec(c, b->left, &rvtype);
4319 right = interp_exec(c, b->right, &rtype);
4320 mpq_sub(rv.num, rv.num, right.num);
4323 rv = interp_exec(c, b->left, &rvtype);
4324 right = interp_exec(c, b->right, &rtype);
4325 mpq_mul(rv.num, rv.num, right.num);
4328 rv = interp_exec(c, b->left, &rvtype);
4329 right = interp_exec(c, b->right, &rtype);
4330 mpq_div(rv.num, rv.num, right.num);
4335 left = interp_exec(c, b->left, <ype);
4336 right = interp_exec(c, b->right, &rtype);
4337 mpz_init(l); mpz_init(r); mpz_init(rem);
4338 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4339 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4340 mpz_tdiv_r(rem, l, r);
4341 val_init(Tnum, &rv);
4342 mpq_set_z(rv.num, rem);
4343 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4348 rv = interp_exec(c, b->right, &rvtype);
4349 mpq_neg(rv.num, rv.num);
4352 rv = interp_exec(c, b->right, &rvtype);
4353 mpq_abs(rv.num, rv.num);
4356 rv = interp_exec(c, b->right, &rvtype);
4359 left = interp_exec(c, b->left, <ype);
4360 right = interp_exec(c, b->right, &rtype);
4362 rv.str = text_join(left.str, right.str);
4365 right = interp_exec(c, b->right, &rvtype);
4369 struct text tx = right.str;
4372 if (tx.txt[0] == '-') {
4377 if (number_parse(rv.num, tail, tx) == 0)
4380 mpq_neg(rv.num, rv.num);
4382 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt);
4386 right = interp_exec(c, b->right, &rtype);
4388 rv.bool = !!rtype->test(rtype, &right);
4391 left = interp_exec(c, b->left, <ype);
4392 if (ltype->test(ltype, &left)) {
4397 rv = interp_exec(c, b->right, &rvtype);
4400 ###### value functions
4402 static struct text text_join(struct text a, struct text b)
4405 rv.len = a.len + b.len;
4406 rv.txt = malloc(rv.len);
4407 memcpy(rv.txt, a.txt, a.len);
4408 memcpy(rv.txt+a.len, b.txt, b.len);
4412 ### Blocks, Statements, and Statement lists.
4414 Now that we have expressions out of the way we need to turn to
4415 statements. There are simple statements and more complex statements.
4416 Simple statements do not contain (syntactic) newlines, complex statements do.
4418 Statements often come in sequences and we have corresponding simple
4419 statement lists and complex statement lists.
4420 The former comprise only simple statements separated by semicolons.
4421 The later comprise complex statements and simple statement lists. They are
4422 separated by newlines. Thus the semicolon is only used to separate
4423 simple statements on the one line. This may be overly restrictive,
4424 but I'm not sure I ever want a complex statement to share a line with
4427 Note that a simple statement list can still use multiple lines if
4428 subsequent lines are indented, so
4430 ###### Example: wrapped simple statement list
4435 is a single simple statement list. This might allow room for
4436 confusion, so I'm not set on it yet.
4438 A simple statement list needs no extra syntax. A complex statement
4439 list has two syntactic forms. It can be enclosed in braces (much like
4440 C blocks), or it can be introduced by an indent and continue until an
4441 unindented newline (much like Python blocks). With this extra syntax
4442 it is referred to as a block.
4444 Note that a block does not have to include any newlines if it only
4445 contains simple statements. So both of:
4447 if condition: a=b; d=f
4449 if condition { a=b; print f }
4453 In either case the list is constructed from a `binode` list with
4454 `Block` as the operator. When parsing the list it is most convenient
4455 to append to the end, so a list is a list and a statement. When using
4456 the list it is more convenient to consider a list to be a statement
4457 and a list. So we need a function to re-order a list.
4458 `reorder_bilist` serves this purpose.
4460 The only stand-alone statement we introduce at this stage is `pass`
4461 which does nothing and is represented as a `NULL` pointer in a `Block`
4462 list. Other stand-alone statements will follow once the infrastructure
4465 As many statements will use binodes, we declare a binode pointer 'b' in
4466 the common header for all reductions to use.
4468 ###### Parser: reduce
4479 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4480 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4481 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4482 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS);
4484 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4486 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4487 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4488 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4489 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4490 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4492 UseBlock -> { IN OpenScope OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4493 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4494 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4496 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4497 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4498 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4499 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4500 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4502 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4504 ComplexStatements -> ComplexStatements ComplexStatement ${
4506 $0 = $<1; // NOTEST - impossible
4514 | ComplexStatement ${
4516 $0 = NULL; // NOTEST - impossible
4526 ComplexStatement -> SimpleStatements Newlines ${
4527 $0 = reorder_bilist($<SS);
4529 | SimpleStatements ; Newlines ${
4530 $0 = reorder_bilist($<SS);
4532 ## ComplexStatement Grammar
4535 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4541 | SimpleStatement ${
4550 SimpleStatement -> pass ${ $0 = NULL; }$
4551 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4552 ## SimpleStatement Grammar
4554 ###### print binode cases
4556 // block, one per line
4557 if (b->left == NULL)
4558 do_indent(indent, "pass\n");
4560 print_exec(b->left, indent, bracket);
4562 print_exec(b->right, indent, bracket);
4565 ###### propagate binode cases
4568 /* If any statement returns something other than Tnone
4569 * or Tbool then all such must return same type.
4570 * As each statement may be Tnone or something else,
4571 * we must always pass NULL (unknown) down, otherwise an incorrect
4572 * error might occur. We never return Tnone unless it is
4577 for (e = b; e; e = cast(binode, e->right)) {
4578 *perr |= *perr_local;
4580 t = propagate_types(e->left, c, perr_local, NULL, rules);
4581 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4583 if (t == Tnone && e->right)
4584 /* Only the final statement *must* return a value
4592 type_err(c, "error: expected %1, found %2",
4593 e->left, type, rules, t);
4599 ###### interp binode cases
4601 while (rvtype == Tnone &&
4604 rv = interp_exec(c, b->left, &rvtype);
4605 b = cast(binode, b->right);
4609 ### The Print statement
4611 `print` is a simple statement that takes a comma-separated list of
4612 expressions and prints the values separated by spaces and terminated
4613 by a newline. No control of formatting is possible.
4615 `print` uses `ExpressionList` to collect the expressions and stores them
4616 on the left side of a `Print` binode unlessthere is a trailing comma
4617 when the list is stored on the `right` side and no trailing newline is
4623 ##### declare terminals
4626 ###### SimpleStatement Grammar
4628 | print ExpressionList ${
4629 $0 = b = new_pos(binode, $1);
4632 b->left = reorder_bilist($<EL);
4634 | print ExpressionList , ${ {
4635 $0 = b = new_pos(binode, $1);
4637 b->right = reorder_bilist($<EL);
4641 $0 = b = new_pos(binode, $1);
4647 ###### print binode cases
4650 do_indent(indent, "print");
4651 b2 = cast(binode, b->left ?: b->right);
4654 print_exec(b2->left, -1, bracket);
4657 b2 = cast(binode, b2->right);
4665 ###### propagate binode cases
4668 /* don't care but all must be consistent */
4670 b = cast(binode, b->left);
4672 b = cast(binode, b->right);
4674 propagate_types(b->left, c, perr_local, NULL, 0);
4675 b = cast(binode, b->right);
4679 ###### interp binode cases
4683 struct binode *b2 = cast(binode, b->left);
4685 b2 = cast(binode, b->right);
4686 for (; b2; b2 = cast(binode, b2->right)) {
4687 left = interp_exec(c, b2->left, <ype);
4688 print_value(ltype, &left, stdout);
4689 free_value(ltype, &left);
4693 if (b->right == NULL)
4699 ###### Assignment statement
4701 An assignment will assign a value to a variable, providing it hasn't
4702 been declared as a constant. The analysis phase ensures that the type
4703 will be correct so the interpreter just needs to perform the
4704 calculation. There is a form of assignment which declares a new
4705 variable as well as assigning a value. If a name is used before
4706 it is declared, it is assumed to be a global constant which are allowed to
4707 be declared at any time.
4713 ###### declare terminals
4716 ###### SimpleStatement Grammar
4717 | Term = Expression ${
4718 $0 = b= new(binode);
4723 | VariableDecl = Expression ${
4724 $0 = b= new(binode);
4731 if ($1->var->where_set == NULL) {
4733 "Variable declared with no type or value: %v",
4737 $0 = b = new(binode);
4744 ###### print binode cases
4747 do_indent(indent, "");
4748 print_exec(b->left, -1, bracket);
4750 print_exec(b->right, -1, bracket);
4757 struct variable *v = cast(var, b->left)->var;
4758 do_indent(indent, "");
4759 print_exec(b->left, -1, bracket);
4760 if (cast(var, b->left)->var->constant) {
4762 if (v->explicit_type) {
4763 type_print(v->type, stdout);
4768 if (v->explicit_type) {
4769 type_print(v->type, stdout);
4775 print_exec(b->right, -1, bracket);
4782 ###### propagate binode cases
4786 /* Both must match, or left may be ref and right an lval
4787 * Type must support 'dup',
4788 * For Assign, left must not be constant.
4791 *perr &= ~(Erval | Econst);
4792 t = propagate_types(b->left, c, perr, NULL, 0);
4797 struct type *t2 = propagate_types(b->right, c, perr_local,
4799 if (!t2 || t2 == t || (*perr_local & Efail))
4800 ; // No more effort needed
4801 else if (t->free == reference_free &&
4802 t->reference.referent == t2 &&
4803 !(*perr_local & Erval))
4804 b->right = take_addr(b->right);
4805 else if (t->free == reference_free &&
4806 t->reference.referent == t2 &&
4807 (*perr_local & Erval))
4808 type_err(c, "error: Cannot assign an rval to a reference.",
4811 t = propagate_types(b->right, c, perr_local, NULL, 0);
4813 propagate_types(b->left, c, perr, t, 0);
4816 type_err(c, "error: cannot assign to an rval", b,
4818 else if (b->op == Assign && (*perr & Econst)) {
4819 type_err(c, "error: Cannot assign to a constant: %v",
4820 b->left, NULL, 0, NULL);
4821 if (b->left->type == Xvar) {
4822 struct var *var = cast(var, b->left);
4823 struct variable *v = var->var;
4824 type_err(c, "info: name was defined as a constant here",
4825 v->where_decl, NULL, 0, NULL);
4828 if (t && t->dup == NULL && !(*perr_local & Emaycopy))
4829 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4830 if (b->left->type == Xvar && (*perr_local & Efail))
4831 type_err(c, "info: variable '%v' was set as %1 here.",
4832 cast(var, b->left)->var->where_set, t, rules, NULL);
4837 ###### interp binode cases
4840 lleft = linterp_exec(c, b->left, <ype);
4842 dinterp_exec(c, b->right, lleft, ltype, 1);
4848 struct variable *v = cast(var, b->left)->var;
4851 val = var_value(c, v);
4852 if (v->type->prepare_type)
4853 v->type->prepare_type(c, v->type, 0);
4855 val_init(v->type, val);
4857 dinterp_exec(c, b->right, val, v->type, 0);
4861 ### The `use` statement
4863 The `use` statement is the last "simple" statement. It is needed when a
4864 statement block can return a value. This includes the body of a
4865 function which has a return type, and the "condition" code blocks in
4866 `if`, `while`, and `switch` statements.
4871 ###### declare terminals
4874 ###### SimpleStatement Grammar
4876 $0 = b = new_pos(binode, $1);
4881 ###### print binode cases
4884 do_indent(indent, "use ");
4885 print_exec(b->right, -1, bracket);
4890 ###### propagate binode cases
4893 /* result matches value */
4894 return propagate_types(b->right, c, perr, type, 0);
4896 ###### interp binode cases
4899 rv = interp_exec(c, b->right, &rvtype);
4902 ### The Conditional Statement
4904 This is the biggy and currently the only complex statement. This
4905 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4906 It is comprised of a number of parts, all of which are optional though
4907 set combinations apply. Each part is (usually) a key word (`then` is
4908 sometimes optional) followed by either an expression or a code block,
4909 except the `casepart` which is a "key word and an expression" followed
4910 by a code block. The code-block option is valid for all parts and,
4911 where an expression is also allowed, the code block can use the `use`
4912 statement to report a value. If the code block does not report a value
4913 the effect is similar to reporting `True`.
4915 The `else` and `case` parts, as well as `then` when combined with
4916 `if`, can contain a `use` statement which will apply to some
4917 containing conditional statement. `for` parts, `do` parts and `then`
4918 parts used with `for` can never contain a `use`, except in some
4919 subordinate conditional statement.
4921 If there is a `forpart`, it is executed first, only once.
4922 If there is a `dopart`, then it is executed repeatedly providing
4923 always that the `condpart` or `cond`, if present, does not return a non-True
4924 value. `condpart` can fail to return any value if it simply executes
4925 to completion. This is treated the same as returning `True`.
4927 If there is a `thenpart` it will be executed whenever the `condpart`
4928 or `cond` returns True (or does not return any value), but this will happen
4929 *after* `dopart` (when present).
4931 If `elsepart` is present it will be executed at most once when the
4932 condition returns `False` or some value that isn't `True` and isn't
4933 matched by any `casepart`. If there are any `casepart`s, they will be
4934 executed when the condition returns a matching value.
4936 The particular sorts of values allowed in case parts has not yet been
4937 determined in the language design, so nothing is prohibited.
4939 The various blocks in this complex statement potentially provide scope
4940 for variables as described earlier. Each such block must include the
4941 "OpenScope" nonterminal before parsing the block, and must call
4942 `var_block_close()` when closing the block.
4944 The code following "`if`", "`switch`" and "`for`" does not get its own
4945 scope, but is in a scope covering the whole statement, so names
4946 declared there cannot be redeclared elsewhere. Similarly the
4947 condition following "`while`" is in a scope the covers the body
4948 ("`do`" part) of the loop, and which does not allow conditional scope
4949 extension. Code following "`then`" (both looping and non-looping),
4950 "`else`" and "`case`" each get their own local scope.
4952 The type requirements on the code block in a `whilepart` are quite
4953 unusal. It is allowed to return a value of some identifiable type, in
4954 which case the loop aborts and an appropriate `casepart` is run, or it
4955 can return a Boolean, in which case the loop either continues to the
4956 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4957 This is different both from the `ifpart` code block which is expected to
4958 return a Boolean, or the `switchpart` code block which is expected to
4959 return the same type as the casepart values. The correct analysis of
4960 the type of the `whilepart` code block is the reason for the
4961 `Rboolok` flag which is passed to `propagate_types()`.
4963 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4964 defined. As there are two scopes which cover multiple parts - one for
4965 the whole statement and one for "while" and "do" - and as we will use
4966 the 'struct exec' to track scopes, we actually need two new types of
4967 exec. One is a `binode` for the looping part, the rest is the
4968 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4969 casepart` to track a list of case parts.
4980 struct exec *action;
4981 struct casepart *next;
4983 struct cond_statement {
4985 struct exec *forpart, *condpart, *thenpart, *elsepart;
4986 struct binode *looppart;
4987 struct casepart *casepart;
4990 ###### ast functions
4992 static void free_casepart(struct casepart *cp)
4996 free_exec(cp->value);
4997 free_exec(cp->action);
5004 static void free_cond_statement(struct cond_statement *s)
5008 free_exec(s->forpart);
5009 free_exec(s->condpart);
5010 free_exec(s->looppart);
5011 free_exec(s->thenpart);
5012 free_exec(s->elsepart);
5013 free_casepart(s->casepart);
5017 ###### free exec cases
5018 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
5020 ###### ComplexStatement Grammar
5021 | CondStatement ${ $0 = $<1; }$
5023 ###### declare terminals
5024 $TERM for then while do
5031 // A CondStatement must end with EOL, as does CondSuffix and
5033 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
5034 // may or may not end with EOL
5035 // WhilePart and IfPart include an appropriate Suffix
5037 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
5038 // them. WhilePart opens and closes its own scope.
5039 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
5042 $0->thenpart = $<TP;
5043 $0->looppart = $<WP;
5044 var_block_close(c, CloseSequential, $0);
5046 | ForPart OptNL WhilePart CondSuffix ${
5049 $0->looppart = $<WP;
5050 var_block_close(c, CloseSequential, $0);
5052 | WhilePart CondSuffix ${
5054 $0->looppart = $<WP;
5056 | SwitchPart OptNL CasePart CondSuffix ${
5058 $0->condpart = $<SP;
5059 $CP->next = $0->casepart;
5060 $0->casepart = $<CP;
5061 var_block_close(c, CloseSequential, $0);
5063 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
5065 $0->condpart = $<SP;
5066 $CP->next = $0->casepart;
5067 $0->casepart = $<CP;
5068 var_block_close(c, CloseSequential, $0);
5070 | IfPart IfSuffix ${
5072 $0->condpart = $IP.condpart; $IP.condpart = NULL;
5073 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
5074 // This is where we close an "if" statement
5075 var_block_close(c, CloseSequential, $0);
5078 CondSuffix -> IfSuffix ${
5081 | Newlines CasePart CondSuffix ${
5083 $CP->next = $0->casepart;
5084 $0->casepart = $<CP;
5086 | CasePart CondSuffix ${
5088 $CP->next = $0->casepart;
5089 $0->casepart = $<CP;
5092 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
5093 | Newlines ElsePart ${ $0 = $<EP; }$
5094 | ElsePart ${$0 = $<EP; }$
5096 ElsePart -> else OpenBlock Newlines ${
5097 $0 = new(cond_statement);
5098 $0->elsepart = $<OB;
5099 var_block_close(c, CloseElse, $0->elsepart);
5101 | else OpenScope CondStatement ${
5102 $0 = new(cond_statement);
5103 $0->elsepart = $<CS;
5104 var_block_close(c, CloseElse, $0->elsepart);
5108 CasePart -> case Expression OpenScope ColonBlock ${
5109 $0 = calloc(1,sizeof(struct casepart));
5112 var_block_close(c, CloseParallel, $0->action);
5116 // These scopes are closed in CondStatement
5117 ForPart -> for OpenBlock ${
5121 ThenPart -> then OpenBlock ${
5123 var_block_close(c, CloseSequential, $0);
5127 // This scope is closed in CondStatement
5128 WhilePart -> while UseBlock OptNL do OpenBlock ${
5133 var_block_close(c, CloseSequential, $0->right);
5134 var_block_close(c, CloseSequential, $0);
5136 | while OpenScope Expression OpenScope ColonBlock ${
5141 var_block_close(c, CloseSequential, $0->right);
5142 var_block_close(c, CloseSequential, $0);
5146 IfPart -> if UseBlock OptNL then OpenBlock ${
5149 var_block_close(c, CloseParallel, $0.thenpart);
5151 | if OpenScope Expression OpenScope ColonBlock ${
5154 var_block_close(c, CloseParallel, $0.thenpart);
5156 | if OpenScope Expression OpenScope OptNL then Block ${
5159 var_block_close(c, CloseParallel, $0.thenpart);
5163 // This scope is closed in CondStatement
5164 SwitchPart -> switch OpenScope Expression ${
5167 | switch UseBlock ${
5171 ###### print binode cases
5173 if (b->left && b->left->type == Xbinode &&
5174 cast(binode, b->left)->op == Block) {
5176 do_indent(indent, "while {\n");
5178 do_indent(indent, "while\n");
5179 print_exec(b->left, indent+1, bracket);
5181 do_indent(indent, "} do {\n");
5183 do_indent(indent, "do\n");
5184 print_exec(b->right, indent+1, bracket);
5186 do_indent(indent, "}\n");
5188 do_indent(indent, "while ");
5189 print_exec(b->left, 0, bracket);
5194 print_exec(b->right, indent+1, bracket);
5196 do_indent(indent, "}\n");
5200 ###### print exec cases
5202 case Xcond_statement:
5204 struct cond_statement *cs = cast(cond_statement, e);
5205 struct casepart *cp;
5207 do_indent(indent, "for");
5208 if (bracket) printf(" {\n"); else printf("\n");
5209 print_exec(cs->forpart, indent+1, bracket);
5212 do_indent(indent, "} then {\n");
5214 do_indent(indent, "then\n");
5215 print_exec(cs->thenpart, indent+1, bracket);
5217 if (bracket) do_indent(indent, "}\n");
5220 print_exec(cs->looppart, indent, bracket);
5224 do_indent(indent, "switch");
5226 do_indent(indent, "if");
5227 if (cs->condpart && cs->condpart->type == Xbinode &&
5228 cast(binode, cs->condpart)->op == Block) {
5233 print_exec(cs->condpart, indent+1, bracket);
5235 do_indent(indent, "}\n");
5237 do_indent(indent, "then\n");
5238 print_exec(cs->thenpart, indent+1, bracket);
5242 print_exec(cs->condpart, 0, bracket);
5248 print_exec(cs->thenpart, indent+1, bracket);
5250 do_indent(indent, "}\n");
5255 for (cp = cs->casepart; cp; cp = cp->next) {
5256 do_indent(indent, "case ");
5257 print_exec(cp->value, -1, 0);
5262 print_exec(cp->action, indent+1, bracket);
5264 do_indent(indent, "}\n");
5267 do_indent(indent, "else");
5272 print_exec(cs->elsepart, indent+1, bracket);
5274 do_indent(indent, "}\n");
5279 ###### propagate binode cases
5281 propagate_types(b->right, c, perr_local, Tnone, 0);
5282 return propagate_types(b->left, c, perr, type, rules);
5284 ###### propagate exec cases
5285 case Xcond_statement:
5287 // forpart and looppart->right must return Tnone
5288 // thenpart must return Tnone if there is a loopart,
5289 // otherwise it is like elsepart.
5291 // be bool if there is no casepart
5292 // match casepart->values if there is a switchpart
5293 // either be bool or match casepart->value if there
5295 // elsepart and casepart->action must match the return type
5296 // expected of this statement.
5297 struct cond_statement *cs = cast(cond_statement, prog);
5298 struct casepart *cp;
5300 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5303 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5305 if (cs->casepart == NULL) {
5306 propagate_types(cs->condpart, c, perr, Tbool, 0);
5307 propagate_types(cs->looppart, c, perr, Tbool, 0);
5309 /* Condpart must match case values, with bool permitted */
5311 for (cp = cs->casepart;
5312 cp && !t; cp = cp->next)
5313 t = propagate_types(cp->value, c, perr, NULL, 0);
5314 if (!t && cs->condpart)
5315 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // NOTEST
5316 if (!t && cs->looppart)
5317 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // NOTEST
5318 // Now we have a type (I hope) push it down
5320 for (cp = cs->casepart; cp; cp = cp->next)
5321 propagate_types(cp->value, c, perr, t, 0);
5322 propagate_types(cs->condpart, c, perr, t, Rboolok);
5323 propagate_types(cs->looppart, c, perr, t, Rboolok);
5326 // (if)then, else, and case parts must return expected type.
5327 if (!cs->looppart && !type)
5328 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5330 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5331 for (cp = cs->casepart;
5333 cp = cp->next) // NOTEST
5334 type = propagate_types(cp->action, c, perr, NULL, rules); // NOTEST
5337 propagate_types(cs->thenpart, c, perr, type, rules);
5338 propagate_types(cs->elsepart, c, perr, type, rules);
5339 for (cp = cs->casepart; cp ; cp = cp->next)
5340 propagate_types(cp->action, c, perr, type, rules);
5346 ###### interp binode cases
5348 // This just performs one iterration of the loop
5349 rv = interp_exec(c, b->left, &rvtype);
5350 if (rvtype == Tnone ||
5351 (rvtype == Tbool && rv.bool != 0))
5352 // rvtype is Tnone or Tbool, doesn't need to be freed
5353 interp_exec(c, b->right, NULL);
5356 ###### interp exec cases
5357 case Xcond_statement:
5359 struct value v, cnd;
5360 struct type *vtype, *cndtype;
5361 struct casepart *cp;
5362 struct cond_statement *cs = cast(cond_statement, e);
5365 interp_exec(c, cs->forpart, NULL);
5367 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5368 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5369 interp_exec(c, cs->thenpart, NULL);
5371 cnd = interp_exec(c, cs->condpart, &cndtype);
5372 if ((cndtype == Tnone ||
5373 (cndtype == Tbool && cnd.bool != 0))) {
5374 // cnd is Tnone or Tbool, doesn't need to be freed
5375 rv = interp_exec(c, cs->thenpart, &rvtype);
5376 // skip else (and cases)
5380 for (cp = cs->casepart; cp; cp = cp->next) {
5381 v = interp_exec(c, cp->value, &vtype);
5382 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5383 free_value(vtype, &v);
5384 free_value(cndtype, &cnd);
5385 rv = interp_exec(c, cp->action, &rvtype);
5388 free_value(vtype, &v);
5390 free_value(cndtype, &cnd);
5392 rv = interp_exec(c, cs->elsepart, &rvtype);
5399 ### Top level structure
5401 All the language elements so far can be used in various places. Now
5402 it is time to clarify what those places are.
5404 At the top level of a file there will be a number of declarations.
5405 Many of the things that can be declared haven't been described yet,
5406 such as functions, procedures, imports, and probably more.
5407 For now there are two sorts of things that can appear at the top
5408 level. They are predefined constants, `struct` types, and the `main`
5409 function. While the syntax will allow the `main` function to appear
5410 multiple times, that will trigger an error if it is actually attempted.
5412 The various declarations do not return anything. They store the
5413 various declarations in the parse context.
5415 ###### Parser: grammar
5418 Ocean -> OptNL DeclarationList
5420 ## declare terminals
5428 DeclarationList -> Declaration
5429 | DeclarationList Declaration
5431 Declaration -> ERROR Newlines ${
5432 tok_err(c, // NOTEST
5433 "error: unhandled parse error", &$1);
5439 ## top level grammar
5443 ### The `const` section
5445 As well as being defined in with the code that uses them, constants can
5446 be declared at the top level. These have full-file scope, so they are
5447 always `InScope`, even before(!) they have been declared. The value of
5448 a top level constant can be given as an expression, and this is
5449 evaluated after parsing and before execution.
5451 A function call can be used to evaluate a constant, but it will not have
5452 access to any program state, once such statement becomes meaningful.
5453 e.g. arguments and filesystem will not be visible.
5455 Constants are defined in a section that starts with the reserved word
5456 `const` and then has a block with a list of assignment statements.
5457 For syntactic consistency, these must use the double-colon syntax to
5458 make it clear that they are constants. Type can also be given: if
5459 not, the type will be determined during analysis, as with other
5462 ###### parse context
5463 struct binode *constlist;
5465 ###### top level grammar
5469 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5470 | const { SimpleConstList } Newlines
5471 | const IN OptNL ConstList OUT Newlines
5472 | const SimpleConstList Newlines
5474 ConstList -> ConstList SimpleConstLine
5477 SimpleConstList -> SimpleConstList ; Const
5481 SimpleConstLine -> SimpleConstList Newlines
5482 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5485 CType -> Type ${ $0 = $<1; }$
5489 Const -> IDENTIFIER :: CType = Expression ${ {
5491 struct binode *bl, *bv;
5492 struct var *var = new_pos(var, $ID);
5494 v = var_decl(c, $ID.txt);
5496 v->where_decl = var;
5502 v = var_ref(c, $1.txt);
5503 if (v->type == Tnone) {
5504 v->where_decl = var;
5510 tok_err(c, "error: name already declared", &$1);
5511 type_err(c, "info: this is where '%v' was first declared",
5512 v->where_decl, NULL, 0, NULL);
5524 bl->left = c->constlist;
5529 ###### core functions
5530 static void resolve_consts(struct parse_context *c)
5534 enum { none, some, cannot } progress = none;
5536 c->constlist = reorder_bilist(c->constlist);
5539 for (b = cast(binode, c->constlist); b;
5540 b = cast(binode, b->right)) {
5542 struct binode *vb = cast(binode, b->left);
5543 struct var *v = cast(var, vb->left);
5544 if (v->var->frame_pos >= 0)
5548 propagate_types(vb->right, c, &perr,
5550 } while (perr & Eretry);
5552 c->parse_error += 1;
5553 else if (!(perr & Eruntime)) {
5555 struct value res = interp_exec(
5556 c, vb->right, &v->var->type);
5557 global_alloc(c, v->var->type, v->var, &res);
5559 if (progress == cannot)
5560 type_err(c, "error: const %v cannot be resolved.",
5570 progress = cannot; break;
5572 progress = none; break;
5577 ###### print const decls
5582 for (b = cast(binode, context.constlist); b;
5583 b = cast(binode, b->right)) {
5584 struct binode *vb = cast(binode, b->left);
5585 struct var *vr = cast(var, vb->left);
5586 struct variable *v = vr->var;
5592 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5593 type_print(v->type, stdout);
5595 print_exec(vb->right, -1, 0);
5600 ###### free const decls
5601 free_binode(context.constlist);
5603 ### Function declarations
5605 The code in an Ocean program is all stored in function declarations.
5606 One of the functions must be named `main` and it must accept an array of
5607 strings as a parameter - the command line arguments.
5609 As this is the top level, several things are handled a bit differently.
5610 The function is not interpreted by `interp_exec` as that isn't passed
5611 the argument list which the program requires. Similarly type analysis
5612 is a bit more interesting at this level.
5614 ###### ast functions
5616 static struct type *handle_results(struct parse_context *c,
5617 struct binode *results)
5619 /* Create a 'struct' type from the results list, which
5620 * is a list for 'struct var'
5622 struct type *t = add_anon_type(c, &structure_prototype,
5627 for (b = results; b; b = cast(binode, b->right))
5629 t->structure.nfields = cnt;
5630 t->structure.fields = calloc(cnt, sizeof(struct field));
5632 for (b = results; b; b = cast(binode, b->right)) {
5633 struct var *v = cast(var, b->left);
5634 struct field *f = &t->structure.fields[cnt++];
5635 int a = v->var->type->align;
5636 f->name = v->var->name->name;
5637 f->type = v->var->type;
5639 f->offset = t->size;
5640 v->var->frame_pos = f->offset;
5641 t->size += ((f->type->size - 1) | (a-1)) + 1;
5644 variable_unlink_exec(v->var);
5646 free_binode(results);
5650 static struct variable *declare_function(struct parse_context *c,
5651 struct variable *name,
5652 struct binode *args,
5654 struct binode *results,
5658 struct value fn = {.function = code};
5660 var_block_close(c, CloseFunction, code);
5661 t = add_anon_type(c, &function_prototype,
5662 "func %.*s", name->name->name.len,
5663 name->name->name.txt);
5665 t->function.params = reorder_bilist(args);
5667 ret = handle_results(c, reorder_bilist(results));
5668 t->function.inline_result = 1;
5669 t->function.local_size = ret->size;
5671 t->function.return_type = ret;
5672 global_alloc(c, t, name, &fn);
5673 name->type->function.scope = c->out_scope;
5678 var_block_close(c, CloseFunction, NULL);
5680 c->out_scope = NULL;
5684 ###### declare terminals
5687 ###### top level grammar
5690 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5691 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5693 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5694 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5696 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5697 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5699 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5700 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5702 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5703 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5705 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5706 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5708 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5709 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5711 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5712 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5714 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5715 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5718 ###### print func decls
5723 while (target != 0) {
5725 for (v = context.in_scope; v; v=v->in_scope)
5726 if (v->depth == 0 && v->type && v->type->check_args) {
5735 struct value *val = var_value(&context, v);
5736 printf("func %.*s", v->name->name.len, v->name->name.txt);
5737 v->type->print_type_decl(v->type, stdout);
5740 print_exec(val->function, 1, brackets);
5743 print_value(v->type, val, stdout);
5745 printf("/* frame size %d */\n", v->type->function.local_size);
5751 ###### core functions
5753 static int analyse_funcs(struct parse_context *c)
5757 for (v = c->in_scope; v; v = v->in_scope) {
5761 if (v->depth != 0 || !v->type || !v->type->check_args)
5763 ret = v->type->function.inline_result ?
5764 Tnone : v->type->function.return_type;
5765 val = var_value(c, v);
5768 propagate_types(val->function, c, &perr, ret, 0);
5769 } while (!(perr & Efail) && (perr & Eretry));
5770 if (!(perr & Efail))
5771 /* Make sure everything is still consistent */
5772 propagate_types(val->function, c, &perr, ret, 0);
5775 if (!v->type->function.inline_result &&
5776 !v->type->function.return_type->dup) {
5777 type_err(c, "error: function cannot return value of type %1",
5778 v->where_decl, v->type->function.return_type, 0, NULL);
5781 scope_finalize(c, v->type);
5786 static int analyse_main(struct type *type, struct parse_context *c)
5788 struct binode *bp = type->function.params;
5792 struct type *argv_type;
5794 argv_type = add_anon_type(c, &array_prototype, "argv");
5795 argv_type->array.member = Tstr;
5796 argv_type->array.unspec = 1;
5798 for (b = bp; b; b = cast(binode, b->right)) {
5802 propagate_types(b->left, c, &perr, argv_type, 0);
5804 default: /* invalid */ // NOTEST
5805 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5808 c->parse_error += 1;
5811 return !c->parse_error;
5814 static void interp_main(struct parse_context *c, int argc, char **argv)
5816 struct value *progp = NULL;
5817 struct text main_name = { "main", 4 };
5818 struct variable *mainv;
5824 mainv = var_ref(c, main_name);
5826 progp = var_value(c, mainv);
5827 if (!progp || !progp->function) {
5828 fprintf(stderr, "oceani: no main function found.\n");
5829 c->parse_error += 1;
5832 if (!analyse_main(mainv->type, c)) {
5833 fprintf(stderr, "oceani: main has wrong type.\n");
5834 c->parse_error += 1;
5837 al = mainv->type->function.params;
5839 c->local_size = mainv->type->function.local_size;
5840 c->local = calloc(1, c->local_size);
5842 struct var *v = cast(var, al->left);
5843 struct value *vl = var_value(c, v->var);
5851 t->array.size = argc;
5852 t->prepare_type(c, t, 0);
5853 array_init(v->var->type, vl);
5854 for (i = 0; i < argc; i++) {
5855 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5857 arg.str.txt = argv[i];
5858 arg.str.len = strlen(argv[i]);
5859 free_value(Tstr, vl2);
5860 dup_value(Tstr, &arg, vl2);
5864 al = cast(binode, al->right);
5866 v = interp_exec(c, progp->function, &vtype);
5867 free_value(vtype, &v);
5872 ###### ast functions
5873 void free_variable(struct variable *v)
5877 ## And now to test it out.
5879 Having a language requires having a "hello world" program. I'll
5880 provide a little more than that: a program that prints "Hello world"
5881 finds the GCD of two numbers, prints the first few elements of
5882 Fibonacci, performs a binary search for a number, and a few other
5883 things which will likely grow as the languages grows.
5885 ###### File: oceani.mk
5888 @echo "===== DEMO ====="
5889 ./oceani --section "demo: hello" oceani.mdc 55 33
5895 four ::= 2 + 2 ; five ::= 10/2
5896 const pie ::= "I like Pie";
5897 cake ::= "The cake is"
5905 func main(argv:[]string)
5906 print "Hello World, what lovely oceans you have!"
5907 print "Are there", five, "?"
5908 print pi, pie, "but", cake
5910 A := $argv[1]; B := $argv[2]
5912 /* When a variable is defined in both branches of an 'if',
5913 * and used afterwards, the variables are merged.
5919 print "Is", A, "bigger than", B,"? ", bigger
5920 /* If a variable is not used after the 'if', no
5921 * merge happens, so types can be different
5924 double:string = "yes"
5925 print A, "is more than twice", B, "?", double
5928 print "double", B, "is", double
5939 print "GCD of", A, "and", B,"is", a
5941 print a, "is not positive, cannot calculate GCD"
5943 print b, "is not positive, cannot calculate GCD"
5948 print "Fibonacci:", f1,f2,
5949 then togo = togo - 1
5957 /* Binary search... */
5962 mid := (lo + hi) / 2
5975 print "Yay, I found", target
5977 print "Closest I found was", lo
5982 // "middle square" PRNG. Not particularly good, but one my
5983 // Dad taught me - the first one I ever heard of.
5984 for i:=1; then i = i + 1; while i < size:
5985 n := list[i-1] * list[i-1]
5986 list[i] = (n / 100) % 10 000
5988 print "Before sort:",
5989 for i:=0; then i = i + 1; while i < size:
5993 for i := 1; then i=i+1; while i < size:
5994 for j:=i-1; then j=j-1; while j >= 0:
5995 if list[j] > list[j+1]:
5999 print " After sort:",
6000 for i:=0; then i = i + 1; while i < size:
6004 if 1 == 2 then print "yes"; else print "no"
6008 bob.alive = (bob.name == "Hello")
6009 print "bob", "is" if bob.alive else "isn't", "alive"