1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
113 struct parse_context {
114 struct token_config config;
122 #define container_of(ptr, type, member) ({ \
123 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
124 (type *)( (char *)__mptr - offsetof(type,member) );})
126 #define config2context(_conf) container_of(_conf, struct parse_context, \
129 ###### Parser: reduce
130 struct parse_context *c = config2context(config);
138 #include <sys/mman.h>
157 static char Usage[] =
158 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
159 static const struct option long_options[] = {
160 {"trace", 0, NULL, 't'},
161 {"print", 0, NULL, 'p'},
162 {"noexec", 0, NULL, 'n'},
163 {"brackets", 0, NULL, 'b'},
164 {"section", 1, NULL, 's'},
167 const char *options = "tpnbs";
169 static void pr_err(char *msg) // NOTEST
171 fprintf(stderr, "%s\n", msg); // NOTEST
174 int main(int argc, char *argv[])
179 struct section *s = NULL, *ss;
180 char *section = NULL;
181 struct parse_context context = {
183 .ignored = (1 << TK_mark),
184 .number_chars = ".,_+- ",
189 int doprint=0, dotrace=0, doexec=1, brackets=0;
191 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
194 case 't': dotrace=1; break;
195 case 'p': doprint=1; break;
196 case 'n': doexec=0; break;
197 case 'b': brackets=1; break;
198 case 's': section = optarg; break;
199 default: fprintf(stderr, Usage);
203 if (optind >= argc) {
204 fprintf(stderr, "oceani: no input file given\n");
207 fd = open(argv[optind], O_RDONLY);
209 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
212 context.file_name = argv[optind];
213 len = lseek(fd, 0, 2);
214 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
215 s = code_extract(file, file+len, pr_err);
217 fprintf(stderr, "oceani: could not find any code in %s\n",
222 ## context initialization
225 for (ss = s; ss; ss = ss->next) {
226 struct text sec = ss->section;
227 if (sec.len == strlen(section) &&
228 strncmp(sec.txt, section, sec.len) == 0)
232 fprintf(stderr, "oceani: cannot find section %s\n",
239 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
240 goto cleanup; // NOTEST
243 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
245 resolve_consts(&context);
246 prepare_types(&context);
247 if (!context.parse_error && !analyse_funcs(&context)) {
248 fprintf(stderr, "oceani: type error in program - not running.\n");
249 context.parse_error += 1;
257 if (doexec && !context.parse_error)
258 interp_main(&context, argc - optind, argv + optind);
261 struct section *t = s->next;
266 // FIXME parser should pop scope even on error
267 while (context.scope_depth > 0)
271 ## free context types
272 ## free context storage
273 exit(context.parse_error ? 1 : 0);
278 The four requirements of parse, analyse, print, interpret apply to
279 each language element individually so that is how most of the code
282 Three of the four are fairly self explanatory. The one that requires
283 a little explanation is the analysis step.
285 The current language design does not require the types of variables to
286 be declared, but they must still have a single type. Different
287 operations impose different requirements on the variables, for example
288 addition requires both arguments to be numeric, and assignment
289 requires the variable on the left to have the same type as the
290 expression on the right.
292 Analysis involves propagating these type requirements around and
293 consequently setting the type of each variable. If any requirements
294 are violated (e.g. a string is compared with a number) or if a
295 variable needs to have two different types, then an error is raised
296 and the program will not run.
298 If the same variable is declared in both branchs of an 'if/else', or
299 in all cases of a 'switch' then the multiple instances may be merged
300 into just one variable if the variable is referenced after the
301 conditional statement. When this happens, the types must naturally be
302 consistent across all the branches. When the variable is not used
303 outside the if, the variables in the different branches are distinct
304 and can be of different types.
306 Undeclared names may only appear in "use" statements and "case" expressions.
307 These names are given a type of "label" and a unique value.
308 This allows them to fill the role of a name in an enumerated type, which
309 is useful for testing the `switch` statement.
311 As we will see, the condition part of a `while` statement can return
312 either a Boolean or some other type. This requires that the expected
313 type that gets passed around comprises a type and a flag to indicate
314 that `Tbool` is also permitted.
316 As there are, as yet, no distinct types that are compatible, there
317 isn't much subtlety in the analysis. When we have distinct number
318 types, this will become more interesting.
322 When analysis discovers an inconsistency it needs to report an error;
323 just refusing to run the code ensures that the error doesn't cascade,
324 but by itself it isn't very useful. A clear understanding of the sort
325 of error message that are useful will help guide the process of
328 At a simplistic level, the only sort of error that type analysis can
329 report is that the type of some construct doesn't match a contextual
330 requirement. For example, in `4 + "hello"` the addition provides a
331 contextual requirement for numbers, but `"hello"` is not a number. In
332 this particular example no further information is needed as the types
333 are obvious from local information. When a variable is involved that
334 isn't the case. It may be helpful to explain why the variable has a
335 particular type, by indicating the location where the type was set,
336 whether by declaration or usage.
338 Using a recursive-descent analysis we can easily detect a problem at
339 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
340 will detect that one argument is not a number and the usage of `hello`
341 will detect that a number was wanted, but not provided. In this
342 (early) version of the language, we will generate error reports at
343 multiple locations, so the use of `hello` will report an error and
344 explain were the value was set, and the addition will report an error
345 and say why numbers are needed. To be able to report locations for
346 errors, each language element will need to record a file location
347 (line and column) and each variable will need to record the language
348 element where its type was set. For now we will assume that each line
349 of an error message indicates one location in the file, and up to 2
350 types. So we provide a `printf`-like function which takes a format, a
351 location (a `struct exec` which has not yet been introduced), and 2
352 types. "`%1`" reports the first type, "`%2`" reports the second. We
353 will need a function to print the location, once we know how that is
354 stored. e As will be explained later, there are sometimes extra rules for
355 type matching and they might affect error messages, we need to pass those
358 As well as type errors, we sometimes need to report problems with
359 tokens, which might be unexpected or might name a type that has not
360 been defined. For these we have `tok_err()` which reports an error
361 with a given token. Each of the error functions sets the flag in the
362 context so indicate that parsing failed.
366 static void fput_loc(struct exec *loc, FILE *f);
367 static void type_err(struct parse_context *c,
368 char *fmt, struct exec *loc,
369 struct type *t1, int rules, struct type *t2);
370 static void tok_err(struct parse_context *c, char *fmt, struct token *t);
372 ###### core functions
374 static void type_err(struct parse_context *c,
375 char *fmt, struct exec *loc,
376 struct type *t1, int rules, struct type *t2)
378 fprintf(stderr, "%s:", c->file_name);
379 fput_loc(loc, stderr);
380 for (; *fmt ; fmt++) {
387 case '%': fputc(*fmt, stderr); break; // NOTEST
388 default: fputc('?', stderr); break; // NOTEST
390 type_print(t1, stderr);
393 type_print(t2, stderr);
402 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
404 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
405 t->txt.len, t->txt.txt);
409 ## Entities: declared and predeclared.
411 There are various "things" that the language and/or the interpreter
412 needs to know about to parse and execute a program. These include
413 types, variables, values, and executable code. These are all lumped
414 together under the term "entities" (calling them "objects" would be
415 confusing) and introduced here. The following section will present the
416 different specific code elements which comprise or manipulate these
421 Executables can be lots of different things. In many cases an
422 executable is just an operation combined with one or two other
423 executables. This allows for expressions and lists etc. Other times an
424 executable is something quite specific like a constant or variable name.
425 So we define a `struct exec` to be a general executable with a type, and
426 a `struct binode` which is a subclass of `exec`, forms a node in a
427 binary tree, and holds an operation. There will be other subclasses,
428 and to access these we need to be able to `cast` the `exec` into the
429 various other types. The first field in any `struct exec` is the type
430 from the `exec_types` enum.
433 #define cast(structname, pointer) ({ \
434 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
435 if (__mptr && *__mptr != X##structname) abort(); \
436 (struct structname *)( (char *)__mptr);})
438 #define new(structname) ({ \
439 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
440 __ptr->type = X##structname; \
441 __ptr->line = -1; __ptr->column = -1; \
444 #define new_pos(structname, token) ({ \
445 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
446 __ptr->type = X##structname; \
447 __ptr->line = token.line; __ptr->column = token.col; \
456 enum exec_types type;
465 struct exec *left, *right;
470 static int __fput_loc(struct exec *loc, FILE *f)
474 if (loc->line >= 0) {
475 fprintf(f, "%d:%d: ", loc->line, loc->column);
478 if (loc->type == Xbinode)
479 return __fput_loc(cast(binode,loc)->left, f) ||
480 __fput_loc(cast(binode,loc)->right, f); // NOTEST
483 static void fput_loc(struct exec *loc, FILE *f)
485 if (!__fput_loc(loc, f))
486 fprintf(f, "??:??: "); // NOTEST
489 Each different type of `exec` node needs a number of functions defined,
490 a bit like methods. We must be able to free it, print it, analyse it
491 and execute it. Once we have specific `exec` types we will need to
492 parse them too. Let's take this a bit more slowly.
496 The parser generator requires a `free_foo` function for each struct
497 that stores attributes and they will often be `exec`s and subtypes
498 there-of. So we need `free_exec` which can handle all the subtypes,
499 and we need `free_binode`.
503 static void free_binode(struct binode *b)
512 ###### core functions
513 static void free_exec(struct exec *e)
524 static void free_exec(struct exec *e);
526 ###### free exec cases
527 case Xbinode: free_binode(cast(binode, e)); break;
531 Printing an `exec` requires that we know the current indent level for
532 printing line-oriented components. As will become clear later, we
533 also want to know what sort of bracketing to use.
537 static void do_indent(int i, char *str)
544 ###### core functions
545 static void print_binode(struct binode *b, int indent, int bracket)
549 ## print binode cases
553 static void print_exec(struct exec *e, int indent, int bracket)
559 print_binode(cast(binode, e), indent, bracket); break;
564 do_indent(indent, "/* FREE");
565 for (v = e->to_free; v; v = v->next_free) {
566 printf(" %.*s", v->name->name.len, v->name->name.txt);
567 printf("[%d,%d]", v->scope_start, v->scope_end);
568 if (v->frame_pos >= 0)
569 printf("(%d+%d)", v->frame_pos,
570 v->type ? v->type->size:0);
578 static void print_exec(struct exec *e, int indent, int bracket);
582 As discussed, analysis involves propagating type requirements around the
583 program and looking for errors.
585 So `propagate_types` is passed an expected type (being a `struct type`
586 pointer together with some `val_rules` flags) that the `exec` is
587 expected to return, and returns the type that it does return, either of
588 which can be `NULL` signifying "unknown". A `prop_err` flag set is
589 passed by reference. It has `Efail` set when an error is found, and
590 `Eretry` when the type for some element is set via propagation. If
591 any expression cannot be evaluated immediately, `Enoconst` is set.
592 If the expression can be copied, `Emaycopy` is set.
594 If it remains unchanged at `0`, then no more propagation is needed.
598 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 1<<2};
599 enum prop_err {Efail = 1<<0, Eretry = 1<<1, Enoconst = 1<<2,
604 if (rules & Rnolabel)
605 fputs(" (labels not permitted)", stderr);
609 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
610 struct type *type, int rules);
611 ###### core functions
613 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
614 struct type *type, int rules)
621 switch (prog->type) {
624 struct binode *b = cast(binode, prog);
626 ## propagate binode cases
630 ## propagate exec cases
635 static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
636 struct type *type, int rules)
638 int pre_err = c->parse_error;
639 struct type *ret = __propagate_types(prog, c, perr, type, rules);
641 if (c->parse_error > pre_err)
648 Interpreting an `exec` doesn't require anything but the `exec`. State
649 is stored in variables and each variable will be directly linked from
650 within the `exec` tree. The exception to this is the `main` function
651 which needs to look at command line arguments. This function will be
652 interpreted separately.
654 Each `exec` can return a value combined with a type in `struct lrval`.
655 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
656 the location of a value, which can be updated, in `lval`. Others will
657 set `lval` to NULL indicating that there is a value of appropriate type
661 static struct value interp_exec(struct parse_context *c, struct exec *e,
662 struct type **typeret);
663 ###### core functions
667 struct value rval, *lval;
670 /* If dest is passed, dtype must give the expected type, and
671 * result can go there, in which case type is returned as NULL.
673 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
674 struct value *dest, struct type *dtype);
676 static struct value interp_exec(struct parse_context *c, struct exec *e,
677 struct type **typeret)
679 struct lrval ret = _interp_exec(c, e, NULL, NULL);
681 if (!ret.type) abort();
685 dup_value(ret.type, ret.lval, &ret.rval);
689 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
690 struct type **typeret)
692 struct lrval ret = _interp_exec(c, e, NULL, NULL);
694 if (!ret.type) abort();
698 free_value(ret.type, &ret.rval);
702 /* dinterp_exec is used when the destination type is certain and
703 * the value has a place to go.
705 static void dinterp_exec(struct parse_context *c, struct exec *e,
706 struct value *dest, struct type *dtype,
709 struct lrval ret = _interp_exec(c, e, dest, dtype);
713 free_value(dtype, dest);
715 dup_value(dtype, ret.lval, dest);
717 memcpy(dest, &ret.rval, dtype->size);
720 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
721 struct value *dest, struct type *dtype)
723 /* If the result is copied to dest, ret.type is set to NULL */
725 struct value rv = {}, *lrv = NULL;
728 rvtype = ret.type = Tnone;
738 struct binode *b = cast(binode, e);
739 struct value left, right, *lleft;
740 struct type *ltype, *rtype;
741 ltype = rtype = Tnone;
743 ## interp binode cases
745 free_value(ltype, &left);
746 free_value(rtype, &right);
756 ## interp exec cleanup
762 Values come in a wide range of types, with more likely to be added.
763 Each type needs to be able to print its own values (for convenience at
764 least) as well as to compare two values, at least for equality and
765 possibly for order. For now, values might need to be duplicated and
766 freed, though eventually such manipulations will be better integrated
769 Rather than requiring every numeric type to support all numeric
770 operations (add, multiply, etc), we allow types to be able to present
771 as one of a few standard types: integer, float, and fraction. The
772 existence of these conversion functions eventually enable types to
773 determine if they are compatible with other types, though such types
774 have not yet been implemented.
776 Named type are stored in a simple linked list. Objects of each type are
777 "values" which are often passed around by value.
779 There are both explicitly named types, and anonymous types. Anonymous
780 cannot be accessed by name, but are used internally and have a name
781 which might be reported in error messages.
788 ## value union fields
795 struct token first_use;
798 void (*init)(struct type *type, struct value *val);
799 int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
800 void (*print)(struct type *type, struct value *val, FILE *f);
801 void (*print_type)(struct type *type, FILE *f);
802 int (*cmp_order)(struct type *t1, struct type *t2,
803 struct value *v1, struct value *v2);
804 int (*cmp_eq)(struct type *t1, struct type *t2,
805 struct value *v1, struct value *v2);
806 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
807 int (*test)(struct type *type, struct value *val);
808 void (*free)(struct type *type, struct value *val);
809 void (*free_type)(struct type *t);
810 long long (*to_int)(struct value *v);
811 double (*to_float)(struct value *v);
812 int (*to_mpq)(mpq_t *q, struct value *v);
821 struct type *typelist;
828 static struct type *find_type(struct parse_context *c, struct text s)
830 struct type *t = c->typelist;
832 while (t && (t->anon ||
833 text_cmp(t->name, s) != 0))
838 static struct type *_add_type(struct parse_context *c, struct text s,
839 struct type *proto, int anon)
843 n = calloc(1, sizeof(*n));
850 n->next = c->typelist;
855 static struct type *add_type(struct parse_context *c, struct text s,
858 return _add_type(c, s, proto, 0);
861 static struct type *add_anon_type(struct parse_context *c,
862 struct type *proto, char *name, ...)
868 vasprintf(&t.txt, name, ap);
870 t.len = strlen(t.txt);
871 return _add_type(c, t, proto, 1);
874 static struct type *find_anon_type(struct parse_context *c,
875 struct type *proto, char *name, ...)
877 struct type *t = c->typelist;
882 vasprintf(&nm.txt, name, ap);
884 nm.len = strlen(name);
886 while (t && (!t->anon ||
887 text_cmp(t->name, nm) != 0))
893 return _add_type(c, nm, proto, 1);
896 static void free_type(struct type *t)
898 /* The type is always a reference to something in the
899 * context, so we don't need to free anything.
903 static void free_value(struct type *type, struct value *v)
907 memset(v, 0x5a, type->size);
911 static void type_print(struct type *type, FILE *f)
914 fputs("*unknown*type*", f); // NOTEST
915 else if (type->name.len && !type->anon)
916 fprintf(f, "%.*s", type->name.len, type->name.txt);
917 else if (type->print_type)
918 type->print_type(type, f);
919 else if (type->name.len && type->anon)
920 fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
922 fputs("*invalid*type*", f); // NOTEST
925 static void val_init(struct type *type, struct value *val)
927 if (type && type->init)
928 type->init(type, val);
931 static void dup_value(struct type *type,
932 struct value *vold, struct value *vnew)
934 if (type && type->dup)
935 type->dup(type, vold, vnew);
938 static int value_cmp(struct type *tl, struct type *tr,
939 struct value *left, struct value *right)
941 if (tl && tl->cmp_order)
942 return tl->cmp_order(tl, tr, left, right);
943 if (tl && tl->cmp_eq)
944 return tl->cmp_eq(tl, tr, left, right);
948 static void print_value(struct type *type, struct value *v, FILE *f)
950 if (type && type->print)
951 type->print(type, v, f);
953 fprintf(f, "*Unknown*"); // NOTEST
956 static void prepare_types(struct parse_context *c)
960 enum { none, some, cannot } progress = none;
965 for (t = c->typelist; t; t = t->next) {
967 tok_err(c, "error: type used but not declared",
969 if (t->size == 0 && t->prepare_type) {
970 if (t->prepare_type(c, t, 1))
972 else if (progress == cannot)
973 tok_err(c, "error: type has recursive definition",
983 progress = cannot; break;
985 progress = none; break;
992 static void free_value(struct type *type, struct value *v);
993 static int type_compat(struct type *require, struct type *have, int rules);
994 static void type_print(struct type *type, FILE *f);
995 static void val_init(struct type *type, struct value *v);
996 static void dup_value(struct type *type,
997 struct value *vold, struct value *vnew);
998 static int value_cmp(struct type *tl, struct type *tr,
999 struct value *left, struct value *right);
1000 static void print_value(struct type *type, struct value *v, FILE *f);
1002 ###### free context types
1004 while (context.typelist) {
1005 struct type *t = context.typelist;
1007 context.typelist = t->next;
1015 Type can be specified for local variables, for fields in a structure,
1016 for formal parameters to functions, and possibly elsewhere. Different
1017 rules may apply in different contexts. As a minimum, a named type may
1018 always be used. Currently the type of a formal parameter can be
1019 different from types in other contexts, so we have a separate grammar
1025 Type -> IDENTIFIER ${
1026 $0 = find_type(c, $ID.txt);
1028 $0 = add_type(c, $ID.txt, NULL);
1029 $0->first_use = $ID;
1034 FormalType -> Type ${ $0 = $<1; }$
1035 ## formal type grammar
1039 Values of the base types can be numbers, which we represent as
1040 multi-precision fractions, strings, Booleans and labels. When
1041 analysing the program we also need to allow for places where no value
1042 is meaningful (type `Tnone`) and where we don't know what type to
1043 expect yet (type is `NULL`).
1045 Values are never shared, they are always copied when used, and freed
1046 when no longer needed.
1048 When propagating type information around the program, we need to
1049 determine if two types are compatible, where type `NULL` is compatible
1050 with anything. There are two special cases with type compatibility,
1051 both related to the Conditional Statement which will be described
1052 later. In some cases a Boolean can be accepted as well as some other
1053 primary type, and in others any type is acceptable except a label (`Vlabel`).
1054 A separate function encoding these cases will simplify some code later.
1056 ###### type functions
1058 int (*compat)(struct type *this, struct type *other);
1060 ###### ast functions
1062 static int type_compat(struct type *require, struct type *have, int rules)
1064 if ((rules & Rboolok) && have == Tbool)
1066 if ((rules & Rnolabel) && have == Tlabel)
1068 if (!require || !have)
1071 if (require->compat)
1072 return require->compat(require, have);
1074 return require == have;
1079 #include "parse_string.h"
1080 #include "parse_number.h"
1083 myLDLIBS := libnumber.o libstring.o -lgmp
1084 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
1086 ###### type union fields
1087 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
1089 ###### value union fields
1095 ###### ast functions
1096 static void _free_value(struct type *type, struct value *v)
1100 switch (type->vtype) {
1102 case Vstr: free(v->str.txt); break;
1103 case Vnum: mpq_clear(v->num); break;
1109 ###### value functions
1111 static void _val_init(struct type *type, struct value *val)
1113 switch(type->vtype) {
1114 case Vnone: // NOTEST
1117 mpq_init(val->num); break;
1119 val->str.txt = malloc(1);
1131 static void _dup_value(struct type *type,
1132 struct value *vold, struct value *vnew)
1134 switch (type->vtype) {
1135 case Vnone: // NOTEST
1138 vnew->label = vold->label;
1141 vnew->bool = vold->bool;
1144 mpq_init(vnew->num);
1145 mpq_set(vnew->num, vold->num);
1148 vnew->str.len = vold->str.len;
1149 vnew->str.txt = malloc(vnew->str.len);
1150 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
1155 static int _value_cmp(struct type *tl, struct type *tr,
1156 struct value *left, struct value *right)
1160 return tl - tr; // NOTEST
1161 switch (tl->vtype) {
1162 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
1163 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
1164 case Vstr: cmp = text_cmp(left->str, right->str); break;
1165 case Vbool: cmp = left->bool - right->bool; break;
1166 case Vnone: cmp = 0; // NOTEST
1171 static void _print_value(struct type *type, struct value *v, FILE *f)
1173 switch (type->vtype) {
1174 case Vnone: // NOTEST
1175 fprintf(f, "*no-value*"); break; // NOTEST
1176 case Vlabel: // NOTEST
1177 fprintf(f, "*label-%p*", v->label); break; // NOTEST
1179 fprintf(f, "%.*s", v->str.len, v->str.txt); break;
1181 fprintf(f, "%s", v->bool ? "True":"False"); break;
1186 mpf_set_q(fl, v->num);
1187 gmp_fprintf(f, "%.10Fg", fl);
1194 static void _free_value(struct type *type, struct value *v);
1196 static int bool_test(struct type *type, struct value *v)
1201 static struct type base_prototype = {
1203 .print = _print_value,
1204 .cmp_order = _value_cmp,
1205 .cmp_eq = _value_cmp,
1207 .free = _free_value,
1210 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
1212 ###### ast functions
1213 static struct type *add_base_type(struct parse_context *c, char *n,
1214 enum vtype vt, int size)
1216 struct text txt = { n, strlen(n) };
1219 t = add_type(c, txt, &base_prototype);
1222 t->align = size > sizeof(void*) ? sizeof(void*) : size;
1223 if (t->size & (t->align - 1))
1224 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
1228 ###### context initialization
1230 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
1231 Tbool->test = bool_test;
1232 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
1233 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
1234 Tnone = add_base_type(&context, "none", Vnone, 0);
1235 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
1239 We have already met values as separate objects. When manifest constants
1240 appear in the program text, that must result in an executable which has
1241 a constant value. So the `val` structure embeds a value in an
1254 ###### ast functions
1255 struct val *new_val(struct type *T, struct token tk)
1257 struct val *v = new_pos(val, tk);
1268 $0 = new_val(Tbool, $1);
1272 $0 = new_val(Tbool, $1);
1277 $0 = new_val(Tnum, $1);
1278 if (number_parse($0->val.num, tail, $1.txt) == 0)
1279 mpq_init($0->val.num); // UNTESTED
1281 tok_err(c, "error: unsupported number suffix",
1286 $0 = new_val(Tstr, $1);
1287 string_parse(&$1, '\\', &$0->val.str, tail);
1289 tok_err(c, "error: unsupported string suffix",
1294 $0 = new_val(Tstr, $1);
1295 string_parse(&$1, '\\', &$0->val.str, tail);
1297 tok_err(c, "error: unsupported string suffix",
1301 ###### print exec cases
1304 struct val *v = cast(val, e);
1305 if (v->vtype == Tstr)
1307 // FIXME how to ensure numbers have same precision.
1308 print_value(v->vtype, &v->val, stdout);
1309 if (v->vtype == Tstr)
1314 ###### propagate exec cases
1317 struct val *val = cast(val, prog);
1318 if (!type_compat(type, val->vtype, rules))
1319 type_err(c, "error: expected %1%r found %2",
1320 prog, type, rules, val->vtype);
1324 ###### interp exec cases
1326 rvtype = cast(val, e)->vtype;
1327 dup_value(rvtype, &cast(val, e)->val, &rv);
1330 ###### ast functions
1331 static void free_val(struct val *v)
1334 free_value(v->vtype, &v->val);
1338 ###### free exec cases
1339 case Xval: free_val(cast(val, e)); break;
1341 ###### ast functions
1342 // Move all nodes from 'b' to 'rv', reversing their order.
1343 // In 'b' 'left' is a list, and 'right' is the last node.
1344 // In 'rv', left' is the first node and 'right' is a list.
1345 static struct binode *reorder_bilist(struct binode *b)
1347 struct binode *rv = NULL;
1350 struct exec *t = b->right;
1354 b = cast(binode, b->left);
1364 Variables are scoped named values. We store the names in a linked list
1365 of "bindings" sorted in lexical order, and use sequential search and
1372 struct binding *next; // in lexical order
1376 This linked list is stored in the parse context so that "reduce"
1377 functions can find or add variables, and so the analysis phase can
1378 ensure that every variable gets a type.
1380 ###### parse context
1382 struct binding *varlist; // In lexical order
1384 ###### ast functions
1386 static struct binding *find_binding(struct parse_context *c, struct text s)
1388 struct binding **l = &c->varlist;
1393 (cmp = text_cmp((*l)->name, s)) < 0)
1397 n = calloc(1, sizeof(*n));
1404 Each name can be linked to multiple variables defined in different
1405 scopes. Each scope starts where the name is declared and continues
1406 until the end of the containing code block. Scopes of a given name
1407 cannot nest, so a declaration while a name is in-scope is an error.
1409 ###### binding fields
1410 struct variable *var;
1414 struct variable *previous;
1416 struct binding *name;
1417 struct exec *where_decl;// where name was declared
1418 struct exec *where_set; // where type was set
1422 When a scope closes, the values of the variables might need to be freed.
1423 This happens in the context of some `struct exec` and each `exec` will
1424 need to know which variables need to be freed when it completes.
1427 struct variable *to_free;
1429 ####### variable fields
1430 struct exec *cleanup_exec;
1431 struct variable *next_free;
1433 ####### interp exec cleanup
1436 for (v = e->to_free; v; v = v->next_free) {
1437 struct value *val = var_value(c, v);
1438 free_value(v->type, val);
1442 ###### ast functions
1443 static void variable_unlink_exec(struct variable *v)
1445 struct variable **vp;
1446 if (!v->cleanup_exec)
1448 for (vp = &v->cleanup_exec->to_free;
1449 *vp; vp = &(*vp)->next_free) {
1453 v->cleanup_exec = NULL;
1458 While the naming seems strange, we include local constants in the
1459 definition of variables. A name declared `var := value` can
1460 subsequently be changed, but a name declared `var ::= value` cannot -
1463 ###### variable fields
1466 Scopes in parallel branches can be partially merged. More
1467 specifically, if a given name is declared in both branches of an
1468 if/else then its scope is a candidate for merging. Similarly if
1469 every branch of an exhaustive switch (e.g. has an "else" clause)
1470 declares a given name, then the scopes from the branches are
1471 candidates for merging.
1473 Note that names declared inside a loop (which is only parallel to
1474 itself) are never visible after the loop. Similarly names defined in
1475 scopes which are not parallel, such as those started by `for` and
1476 `switch`, are never visible after the scope. Only variables defined in
1477 both `then` and `else` (including the implicit then after an `if`, and
1478 excluding `then` used with `for`) and in all `case`s and `else` of a
1479 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
1481 Labels, which are a bit like variables, follow different rules.
1482 Labels are not explicitly declared, but if an undeclared name appears
1483 in a context where a label is legal, that effectively declares the
1484 name as a label. The declaration remains in force (or in scope) at
1485 least to the end of the immediately containing block and conditionally
1486 in any larger containing block which does not declare the name in some
1487 other way. Importantly, the conditional scope extension happens even
1488 if the label is only used in one parallel branch of a conditional --
1489 when used in one branch it is treated as having been declared in all
1492 Merge candidates are tentatively visible beyond the end of the
1493 branching statement which creates them. If the name is used, the
1494 merge is affirmed and they become a single variable visible at the
1495 outer layer. If not - if it is redeclared first - the merge lapses.
1497 To track scopes we have an extra stack, implemented as a linked list,
1498 which roughly parallels the parse stack and which is used exclusively
1499 for scoping. When a new scope is opened, a new frame is pushed and
1500 the child-count of the parent frame is incremented. This child-count
1501 is used to distinguish between the first of a set of parallel scopes,
1502 in which declared variables must not be in scope, and subsequent
1503 branches, whether they may already be conditionally scoped.
1505 We need a total ordering of scopes so we can easily compare to variables
1506 to see if they are concurrently in scope. To achieve this we record a
1507 `scope_count` which is actually a count of both beginnings and endings
1508 of scopes. Then each variable has a record of the scope count where it
1509 enters scope, and where it leaves.
1511 To push a new frame *before* any code in the frame is parsed, we need a
1512 grammar reduction. This is most easily achieved with a grammar
1513 element which derives the empty string, and creates the new scope when
1514 it is recognised. This can be placed, for example, between a keyword
1515 like "if" and the code following it.
1519 struct scope *parent;
1523 ###### parse context
1526 struct scope *scope_stack;
1528 ###### variable fields
1529 int scope_start, scope_end;
1531 ###### ast functions
1532 static void scope_pop(struct parse_context *c)
1534 struct scope *s = c->scope_stack;
1536 c->scope_stack = s->parent;
1538 c->scope_depth -= 1;
1539 c->scope_count += 1;
1542 static void scope_push(struct parse_context *c)
1544 struct scope *s = calloc(1, sizeof(*s));
1546 c->scope_stack->child_count += 1;
1547 s->parent = c->scope_stack;
1549 c->scope_depth += 1;
1550 c->scope_count += 1;
1556 OpenScope -> ${ scope_push(c); }$
1558 Each variable records a scope depth and is in one of four states:
1560 - "in scope". This is the case between the declaration of the
1561 variable and the end of the containing block, and also between
1562 the usage with affirms a merge and the end of that block.
1564 The scope depth is not greater than the current parse context scope
1565 nest depth. When the block of that depth closes, the state will
1566 change. To achieve this, all "in scope" variables are linked
1567 together as a stack in nesting order.
1569 - "pending". The "in scope" block has closed, but other parallel
1570 scopes are still being processed. So far, every parallel block at
1571 the same level that has closed has declared the name.
1573 The scope depth is the depth of the last parallel block that
1574 enclosed the declaration, and that has closed.
1576 - "conditionally in scope". The "in scope" block and all parallel
1577 scopes have closed, and no further mention of the name has been seen.
1578 This state includes a secondary nest depth (`min_depth`) which records
1579 the outermost scope seen since the variable became conditionally in
1580 scope. If a use of the name is found, the variable becomes "in scope"
1581 and that secondary depth becomes the recorded scope depth. If the
1582 name is declared as a new variable, the old variable becomes "out of
1583 scope" and the recorded scope depth stays unchanged.
1585 - "out of scope". The variable is neither in scope nor conditionally
1586 in scope. It is permanently out of scope now and can be removed from
1587 the "in scope" stack. When a variable becomes out-of-scope it is
1588 moved to a separate list (`out_scope`) of variables which have fully
1589 known scope. This will be used at the end of each function to assign
1590 each variable a place in the stack frame.
1592 ###### variable fields
1593 int depth, min_depth;
1594 enum { OutScope, PendingScope, CondScope, InScope } scope;
1595 struct variable *in_scope;
1597 ###### parse context
1599 struct variable *in_scope;
1600 struct variable *out_scope;
1602 All variables with the same name are linked together using the
1603 'previous' link. Those variable that have been affirmatively merged all
1604 have a 'merged' pointer that points to one primary variable - the most
1605 recently declared instance. When merging variables, we need to also
1606 adjust the 'merged' pointer on any other variables that had previously
1607 been merged with the one that will no longer be primary.
1609 A variable that is no longer the most recent instance of a name may
1610 still have "pending" scope, if it might still be merged with most
1611 recent instance. These variables don't really belong in the
1612 "in_scope" list, but are not immediately removed when a new instance
1613 is found. Instead, they are detected and ignored when considering the
1614 list of in_scope names.
1616 The storage of the value of a variable will be described later. For now
1617 we just need to know that when a variable goes out of scope, it might
1618 need to be freed. For this we need to be able to find it, so assume that
1619 `var_value()` will provide that.
1621 ###### variable fields
1622 struct variable *merged;
1624 ###### ast functions
1626 static void variable_merge(struct variable *primary, struct variable *secondary)
1630 primary = primary->merged;
1632 for (v = primary->previous; v; v=v->previous)
1633 if (v == secondary || v == secondary->merged ||
1634 v->merged == secondary ||
1635 v->merged == secondary->merged) {
1636 v->scope = OutScope;
1637 v->merged = primary;
1638 if (v->scope_start < primary->scope_start)
1639 primary->scope_start = v->scope_start;
1640 if (v->scope_end > primary->scope_end)
1641 primary->scope_end = v->scope_end; // NOTEST
1642 variable_unlink_exec(v);
1646 ###### forward decls
1647 static struct value *var_value(struct parse_context *c, struct variable *v);
1649 ###### free global vars
1651 while (context.varlist) {
1652 struct binding *b = context.varlist;
1653 struct variable *v = b->var;
1654 context.varlist = b->next;
1657 struct variable *next = v->previous;
1659 if (v->global && v->frame_pos >= 0) {
1660 free_value(v->type, var_value(&context, v));
1661 if (v->depth == 0 && v->type->free == function_free)
1662 // This is a function constant
1663 free_exec(v->where_decl);
1670 #### Manipulating Bindings
1672 When a name is conditionally visible, a new declaration discards the old
1673 binding - the condition lapses. Similarly when we reach the end of a
1674 function (outermost non-global scope) any conditional scope must lapse.
1675 Conversely a usage of the name affirms the visibility and extends it to
1676 the end of the containing block - i.e. the block that contains both the
1677 original declaration and the latest usage. This is determined from
1678 `min_depth`. When a conditionally visible variable gets affirmed like
1679 this, it is also merged with other conditionally visible variables with
1682 When we parse a variable declaration we either report an error if the
1683 name is currently bound, or create a new variable at the current nest
1684 depth if the name is unbound or bound to a conditionally scoped or
1685 pending-scope variable. If the previous variable was conditionally
1686 scoped, it and its homonyms becomes out-of-scope.
1688 When we parse a variable reference (including non-declarative assignment
1689 "foo = bar") we report an error if the name is not bound or is bound to
1690 a pending-scope variable; update the scope if the name is bound to a
1691 conditionally scoped variable; or just proceed normally if the named
1692 variable is in scope.
1694 When we exit a scope, any variables bound at this level are either
1695 marked out of scope or pending-scoped, depending on whether the scope
1696 was sequential or parallel. Here a "parallel" scope means the "then"
1697 or "else" part of a conditional, or any "case" or "else" branch of a
1698 switch. Other scopes are "sequential".
1700 When exiting a parallel scope we check if there are any variables that
1701 were previously pending and are still visible. If there are, then
1702 they weren't redeclared in the most recent scope, so they cannot be
1703 merged and must become out-of-scope. If it is not the first of
1704 parallel scopes (based on `child_count`), we check that there was a
1705 previous binding that is still pending-scope. If there isn't, the new
1706 variable must now be out-of-scope.
1708 When exiting a sequential scope that immediately enclosed parallel
1709 scopes, we need to resolve any pending-scope variables. If there was
1710 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1711 we need to mark all pending-scope variable as out-of-scope. Otherwise
1712 all pending-scope variables become conditionally scoped.
1715 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1717 ###### ast functions
1719 static struct variable *var_decl(struct parse_context *c, struct text s)
1721 struct binding *b = find_binding(c, s);
1722 struct variable *v = b->var;
1724 switch (v ? v->scope : OutScope) {
1726 /* Caller will report the error */
1730 v && v->scope == CondScope;
1732 v->scope = OutScope;
1736 v = calloc(1, sizeof(*v));
1737 v->previous = b->var;
1741 v->min_depth = v->depth = c->scope_depth;
1743 v->in_scope = c->in_scope;
1744 v->scope_start = c->scope_count;
1750 static struct variable *var_ref(struct parse_context *c, struct text s)
1752 struct binding *b = find_binding(c, s);
1753 struct variable *v = b->var;
1754 struct variable *v2;
1756 switch (v ? v->scope : OutScope) {
1759 /* Caller will report the error */
1762 /* All CondScope variables of this name need to be merged
1763 * and become InScope
1765 v->depth = v->min_depth;
1767 for (v2 = v->previous;
1768 v2 && v2->scope == CondScope;
1770 variable_merge(v, v2);
1778 static int var_refile(struct parse_context *c, struct variable *v)
1780 /* Variable just went out of scope. Add it to the out_scope
1781 * list, sorted by ->scope_start
1783 struct variable **vp = &c->out_scope;
1784 while ((*vp) && (*vp)->scope_start < v->scope_start)
1785 vp = &(*vp)->in_scope;
1791 static void var_block_close(struct parse_context *c, enum closetype ct,
1794 /* Close off all variables that are in_scope.
1795 * Some variables in c->scope may already be not-in-scope,
1796 * such as when a PendingScope variable is hidden by a new
1797 * variable with the same name.
1798 * So we check for v->name->var != v and drop them.
1799 * If we choose to make a variable OutScope, we drop it
1802 struct variable *v, **vp, *v2;
1805 for (vp = &c->in_scope;
1806 (v = *vp) && v->min_depth > c->scope_depth;
1807 (v->scope == OutScope || v->name->var != v)
1808 ? (*vp = v->in_scope, var_refile(c, v))
1809 : ( vp = &v->in_scope, 0)) {
1810 v->min_depth = c->scope_depth;
1811 if (v->name->var != v)
1812 /* This is still in scope, but we haven't just
1816 v->min_depth = c->scope_depth;
1817 if (v->scope == InScope)
1818 v->scope_end = c->scope_count;
1819 if (v->scope == InScope && e && !v->global) {
1820 /* This variable gets cleaned up when 'e' finishes */
1821 variable_unlink_exec(v);
1822 v->cleanup_exec = e;
1823 v->next_free = e->to_free;
1828 case CloseParallel: /* handle PendingScope */
1832 if (c->scope_stack->child_count == 1)
1833 /* first among parallel branches */
1834 v->scope = PendingScope;
1835 else if (v->previous &&
1836 v->previous->scope == PendingScope)
1837 /* all previous branches used name */
1838 v->scope = PendingScope;
1839 else if (v->type == Tlabel)
1840 /* Labels remain pending even when not used */
1841 v->scope = PendingScope; // UNTESTED
1843 v->scope = OutScope;
1844 if (ct == CloseElse) {
1845 /* All Pending variables with this name
1846 * are now Conditional */
1848 v2 && v2->scope == PendingScope;
1850 v2->scope = CondScope;
1854 /* Not possible as it would require
1855 * parallel scope to be nested immediately
1856 * in a parallel scope, and that never
1860 /* Not possible as we already tested for
1867 if (v->scope == CondScope)
1868 /* Condition cannot continue past end of function */
1871 case CloseSequential:
1872 if (v->type == Tlabel)
1873 v->scope = PendingScope;
1876 v->scope = OutScope;
1879 /* There was no 'else', so we can only become
1880 * conditional if we know the cases were exhaustive,
1881 * and that doesn't mean anything yet.
1882 * So only labels become conditional..
1885 v2 && v2->scope == PendingScope;
1887 if (v2->type == Tlabel)
1888 v2->scope = CondScope;
1890 v2->scope = OutScope;
1893 case OutScope: break;
1902 The value of a variable is store separately from the variable, on an
1903 analogue of a stack frame. There are (currently) two frames that can be
1904 active. A global frame which currently only stores constants, and a
1905 stacked frame which stores local variables. Each variable knows if it
1906 is global or not, and what its index into the frame is.
1908 Values in the global frame are known immediately they are relevant, so
1909 the frame needs to be reallocated as it grows so it can store those
1910 values. The local frame doesn't get values until the interpreted phase
1911 is started, so there is no need to allocate until the size is known.
1913 We initialize the `frame_pos` to an impossible value, so that we can
1914 tell if it was set or not later.
1916 ###### variable fields
1920 ###### variable init
1923 ###### parse context
1925 short global_size, global_alloc;
1927 void *global, *local;
1929 ###### forward decls
1930 static struct value *global_alloc(struct parse_context *c, struct type *t,
1931 struct variable *v, struct value *init);
1933 ###### ast functions
1935 static struct value *var_value(struct parse_context *c, struct variable *v)
1938 if (!c->local || !v->type)
1939 return NULL; // UNTESTED
1940 if (v->frame_pos + v->type->size > c->local_size) {
1941 printf("INVALID frame_pos\n"); // NOTEST
1944 return c->local + v->frame_pos;
1946 if (c->global_size > c->global_alloc) {
1947 int old = c->global_alloc;
1948 c->global_alloc = (c->global_size | 1023) + 1024;
1949 c->global = realloc(c->global, c->global_alloc);
1950 memset(c->global + old, 0, c->global_alloc - old);
1952 return c->global + v->frame_pos;
1955 static struct value *global_alloc(struct parse_context *c, struct type *t,
1956 struct variable *v, struct value *init)
1959 struct variable scratch;
1961 if (t->prepare_type)
1962 t->prepare_type(c, t, 1); // NOTEST
1964 if (c->global_size & (t->align - 1))
1965 c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
1970 v->frame_pos = c->global_size;
1972 c->global_size += v->type->size;
1973 ret = var_value(c, v);
1975 memcpy(ret, init, t->size);
1981 As global values are found -- struct field initializers, labels etc --
1982 `global_alloc()` is called to record the value in the global frame.
1984 When the program is fully parsed, each function is analysed, we need to
1985 walk the list of variables local to that function and assign them an
1986 offset in the stack frame. For this we have `scope_finalize()`.
1988 We keep the stack from dense by re-using space for between variables
1989 that are not in scope at the same time. The `out_scope` list is sorted
1990 by `scope_start` and as we process a varible, we move it to an FIFO
1991 stack. For each variable we consider, we first discard any from the
1992 stack anything that went out of scope before the new variable came in.
1993 Then we place the new variable just after the one at the top of the
1996 ###### ast functions
1998 static void scope_finalize(struct parse_context *c, struct type *ft)
2000 int size = ft->function.local_size;
2001 struct variable *next = ft->function.scope;
2002 struct variable *done = NULL;
2005 struct variable *v = next;
2006 struct type *t = v->type;
2013 if (v->frame_pos >= 0)
2015 while (done && done->scope_end < v->scope_start)
2016 done = done->in_scope;
2018 pos = done->frame_pos + done->type->size;
2020 pos = ft->function.local_size;
2021 if (pos & (t->align - 1))
2022 pos = (pos + t->align) & ~(t->align-1);
2024 if (size < pos + v->type->size)
2025 size = pos + v->type->size;
2029 c->out_scope = NULL;
2030 ft->function.local_size = size;
2033 ###### free context storage
2034 free(context.global);
2036 #### Variables as executables
2038 Just as we used a `val` to wrap a value into an `exec`, we similarly
2039 need a `var` to wrap a `variable` into an exec. While each `val`
2040 contained a copy of the value, each `var` holds a link to the variable
2041 because it really is the same variable no matter where it appears.
2042 When a variable is used, we need to remember to follow the `->merged`
2043 link to find the primary instance.
2045 When a variable is declared, it may or may not be given an explicit
2046 type. We need to record which so that we can report the parsed code
2055 struct variable *var;
2058 ###### variable fields
2066 VariableDecl -> IDENTIFIER : ${ {
2067 struct variable *v = var_decl(c, $1.txt);
2068 $0 = new_pos(var, $1);
2073 v = var_ref(c, $1.txt);
2075 type_err(c, "error: variable '%v' redeclared",
2077 type_err(c, "info: this is where '%v' was first declared",
2078 v->where_decl, NULL, 0, NULL);
2081 | IDENTIFIER :: ${ {
2082 struct variable *v = var_decl(c, $1.txt);
2083 $0 = new_pos(var, $1);
2089 v = var_ref(c, $1.txt);
2091 type_err(c, "error: variable '%v' redeclared",
2093 type_err(c, "info: this is where '%v' was first declared",
2094 v->where_decl, NULL, 0, NULL);
2097 | IDENTIFIER : Type ${ {
2098 struct variable *v = var_decl(c, $1.txt);
2099 $0 = new_pos(var, $1);
2105 v->explicit_type = 1;
2107 v = var_ref(c, $1.txt);
2109 type_err(c, "error: variable '%v' redeclared",
2111 type_err(c, "info: this is where '%v' was first declared",
2112 v->where_decl, NULL, 0, NULL);
2115 | IDENTIFIER :: Type ${ {
2116 struct variable *v = var_decl(c, $1.txt);
2117 $0 = new_pos(var, $1);
2124 v->explicit_type = 1;
2126 v = var_ref(c, $1.txt);
2128 type_err(c, "error: variable '%v' redeclared",
2130 type_err(c, "info: this is where '%v' was first declared",
2131 v->where_decl, NULL, 0, NULL);
2136 Variable -> IDENTIFIER ${ {
2137 struct variable *v = var_ref(c, $1.txt);
2138 $0 = new_pos(var, $1);
2140 /* This might be a global const or a label
2141 * Allocate a var with impossible type Tnone,
2142 * which will be adjusted when we find out what it is,
2143 * or will trigger an error.
2145 v = var_decl(c, $1.txt);
2152 cast(var, $0)->var = v;
2155 ###### print exec cases
2158 struct var *v = cast(var, e);
2160 struct binding *b = v->var->name;
2161 printf("%.*s", b->name.len, b->name.txt);
2168 if (loc && loc->type == Xvar) {
2169 struct var *v = cast(var, loc);
2171 struct binding *b = v->var->name;
2172 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2174 fputs("???", stderr); // NOTEST
2176 fputs("NOTVAR", stderr); // NOTEST
2179 ###### propagate exec cases
2183 struct var *var = cast(var, prog);
2184 struct variable *v = var->var;
2186 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2187 return Tnone; // NOTEST
2190 if (v->constant && (rules & Rnoconstant)) {
2191 type_err(c, "error: Cannot assign to a constant: %v",
2192 prog, NULL, 0, NULL);
2193 type_err(c, "info: name was defined as a constant here",
2194 v->where_decl, NULL, 0, NULL);
2197 if (v->type == Tnone && v->where_decl == prog)
2198 type_err(c, "error: variable used but not declared: %v",
2199 prog, NULL, 0, NULL);
2200 if (v->type == NULL) {
2201 if (type && !(*perr & Efail)) {
2203 v->where_set = prog;
2206 } else if (!type_compat(type, v->type, rules)) {
2207 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2208 type, rules, v->type);
2209 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2210 v->type, rules, NULL);
2212 if (!v->global || v->frame_pos < 0)
2219 ###### interp exec cases
2222 struct var *var = cast(var, e);
2223 struct variable *v = var->var;
2226 lrv = var_value(c, v);
2231 ###### ast functions
2233 static void free_var(struct var *v)
2238 ###### free exec cases
2239 case Xvar: free_var(cast(var, e)); break;
2244 Now that we have the shape of the interpreter in place we can add some
2245 complex types and connected them in to the data structures and the
2246 different phases of parse, analyse, print, interpret.
2248 Being "complex" the language will naturally have syntax to access
2249 specifics of objects of these types. These will fit into the grammar as
2250 "Terms" which are the things that are combined with various operators to
2251 form "Expression". Where a Term is formed by some operation on another
2252 Term, the subordinate Term will always come first, so for example a
2253 member of an array will be expressed as the Term for the array followed
2254 by an index in square brackets. The strict rule of using postfix
2255 operations makes precedence irrelevant within terms. To provide a place
2256 to put the grammar for each terms of each type, we will start out by
2257 introducing the "Term" grammar production, with contains at least a
2258 simple "Value" (to be explained later).
2262 Term -> Value ${ $0 = $<1; }$
2263 | Variable ${ $0 = $<1; }$
2266 Thus far the complex types we have are arrays and structs.
2270 Arrays can be declared by giving a size and a type, as `[size]type' so
2271 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
2272 size can be either a literal number, or a named constant. Some day an
2273 arbitrary expression will be supported.
2275 As a formal parameter to a function, the array can be declared with a
2276 new variable as the size: `name:[size::number]string`. The `size`
2277 variable is set to the size of the array and must be a constant. As
2278 `number` is the only supported type, it can be left out:
2279 `name:[size::]string`.
2281 Arrays cannot be assigned. When pointers are introduced we will also
2282 introduce array slices which can refer to part or all of an array -
2283 the assignment syntax will create a slice. For now, an array can only
2284 ever be referenced by the name it is declared with. It is likely that
2285 a "`copy`" primitive will eventually be define which can be used to
2286 make a copy of an array with controllable recursive depth.
2288 For now we have two sorts of array, those with fixed size either because
2289 it is given as a literal number or because it is a struct member (which
2290 cannot have a runtime-changing size), and those with a size that is
2291 determined at runtime - local variables with a const size. The former
2292 have their size calculated at parse time, the latter at run time.
2294 For the latter type, the `size` field of the type is the size of a
2295 pointer, and the array is reallocated every time it comes into scope.
2297 We differentiate struct fields with a const size from local variables
2298 with a const size by whether they are prepared at parse time or not.
2300 ###### type union fields
2303 int unspec; // size is unspecified - vsize must be set.
2306 struct variable *vsize;
2307 struct type *member;
2310 ###### value union fields
2311 void *array; // used if not static_size
2313 ###### value functions
2315 static int array_prepare_type(struct parse_context *c, struct type *type,
2318 struct value *vsize;
2320 if (type->array.static_size)
2321 return 1; // UNTESTED
2322 if (type->array.unspec && parse_time)
2323 return 1; // UNTESTED
2324 if (parse_time && type->array.vsize && !type->array.vsize->global)
2325 return 1; // UNTESTED
2327 if (type->array.vsize) {
2328 vsize = var_value(c, type->array.vsize);
2330 return 1; // UNTESTED
2332 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
2333 type->array.size = mpz_get_si(q);
2338 if (type->array.member->size <= 0)
2339 return 0; // UNTESTED
2341 type->array.static_size = 1;
2342 type->size = type->array.size * type->array.member->size;
2343 type->align = type->array.member->align;
2348 static void array_init(struct type *type, struct value *val)
2351 void *ptr = val->ptr;
2355 if (!type->array.static_size) {
2356 val->array = calloc(type->array.size,
2357 type->array.member->size);
2360 for (i = 0; i < type->array.size; i++) {
2362 v = (void*)ptr + i * type->array.member->size;
2363 val_init(type->array.member, v);
2367 static void array_free(struct type *type, struct value *val)
2370 void *ptr = val->ptr;
2372 if (!type->array.static_size)
2374 for (i = 0; i < type->array.size; i++) {
2376 v = (void*)ptr + i * type->array.member->size;
2377 free_value(type->array.member, v);
2379 if (!type->array.static_size)
2383 static int array_compat(struct type *require, struct type *have)
2385 if (have->compat != require->compat)
2387 /* Both are arrays, so we can look at details */
2388 if (!type_compat(require->array.member, have->array.member, 0))
2390 if (have->array.unspec && require->array.unspec) {
2391 if (have->array.vsize && require->array.vsize &&
2392 have->array.vsize != require->array.vsize) // UNTESTED
2393 /* sizes might not be the same */
2394 return 0; // UNTESTED
2397 if (have->array.unspec || require->array.unspec)
2398 return 1; // UNTESTED
2399 if (require->array.vsize == NULL && have->array.vsize == NULL)
2400 return require->array.size == have->array.size;
2402 return require->array.vsize == have->array.vsize; // UNTESTED
2405 static void array_print_type(struct type *type, FILE *f)
2408 if (type->array.vsize) {
2409 struct binding *b = type->array.vsize->name;
2410 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
2411 type->array.unspec ? "::" : "");
2412 } else if (type->array.size)
2413 fprintf(f, "%d]", type->array.size);
2416 type_print(type->array.member, f);
2419 static struct type array_prototype = {
2421 .prepare_type = array_prepare_type,
2422 .print_type = array_print_type,
2423 .compat = array_compat,
2425 .size = sizeof(void*),
2426 .align = sizeof(void*),
2429 ###### declare terminals
2434 | [ NUMBER ] Type ${ {
2440 if (number_parse(num, tail, $2.txt) == 0)
2441 tok_err(c, "error: unrecognised number", &$2);
2443 tok_err(c, "error: unsupported number suffix", &$2);
2446 elements = mpz_get_ui(mpq_numref(num));
2447 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
2448 tok_err(c, "error: array size must be an integer",
2450 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
2451 tok_err(c, "error: array size is too large",
2456 $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
2457 t->array.size = elements;
2458 t->array.member = $<4;
2459 t->array.vsize = NULL;
2462 | [ IDENTIFIER ] Type ${ {
2463 struct variable *v = var_ref(c, $2.txt);
2466 tok_err(c, "error: name undeclared", &$2);
2467 else if (!v->constant)
2468 tok_err(c, "error: array size must be a constant", &$2);
2470 $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
2471 $0->array.member = $<4;
2473 $0->array.vsize = v;
2478 OptType -> Type ${ $0 = $<1; }$
2481 ###### formal type grammar
2483 | [ IDENTIFIER :: OptType ] Type ${ {
2484 struct variable *v = var_decl(c, $ID.txt);
2490 $0 = add_anon_type(c, &array_prototype, "array[var]");
2491 $0->array.member = $<6;
2493 $0->array.unspec = 1;
2494 $0->array.vsize = v;
2502 | Term [ Expression ] ${ {
2503 struct binode *b = new(binode);
2510 ###### print binode cases
2512 print_exec(b->left, -1, bracket);
2514 print_exec(b->right, -1, bracket);
2518 ###### propagate binode cases
2520 /* left must be an array, right must be a number,
2521 * result is the member type of the array
2523 propagate_types(b->right, c, perr, Tnum, 0);
2524 t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
2525 if (!t || t->compat != array_compat) {
2526 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2529 if (!type_compat(type, t->array.member, rules)) {
2530 type_err(c, "error: have %1 but need %2", prog,
2531 t->array.member, rules, type);
2533 return t->array.member;
2537 ###### interp binode cases
2543 lleft = linterp_exec(c, b->left, <ype);
2544 right = interp_exec(c, b->right, &rtype);
2546 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2550 if (ltype->array.static_size)
2553 ptr = *(void**)lleft;
2554 rvtype = ltype->array.member;
2555 if (i >= 0 && i < ltype->array.size)
2556 lrv = ptr + i * rvtype->size;
2558 val_init(ltype->array.member, &rv); // UNSAFE
2565 A `struct` is a data-type that contains one or more other data-types.
2566 It differs from an array in that each member can be of a different
2567 type, and they are accessed by name rather than by number. Thus you
2568 cannot choose an element by calculation, you need to know what you
2571 The language makes no promises about how a given structure will be
2572 stored in memory - it is free to rearrange fields to suit whatever
2573 criteria seems important.
2575 Structs are declared separately from program code - they cannot be
2576 declared in-line in a variable declaration like arrays can. A struct
2577 is given a name and this name is used to identify the type - the name
2578 is not prefixed by the word `struct` as it would be in C.
2580 Structs are only treated as the same if they have the same name.
2581 Simply having the same fields in the same order is not enough. This
2582 might change once we can create structure initializers from a list of
2585 Each component datum is identified much like a variable is declared,
2586 with a name, one or two colons, and a type. The type cannot be omitted
2587 as there is no opportunity to deduce the type from usage. An initial
2588 value can be given following an equals sign, so
2590 ##### Example: a struct type
2596 would declare a type called "complex" which has two number fields,
2597 each initialised to zero.
2599 Struct will need to be declared separately from the code that uses
2600 them, so we will need to be able to print out the declaration of a
2601 struct when reprinting the whole program. So a `print_type_decl` type
2602 function will be needed.
2604 ###### type union fields
2613 } *fields; // This is created when field_list is analysed.
2615 struct fieldlist *prev;
2618 } *field_list; // This is created during parsing
2621 ###### type functions
2622 void (*print_type_decl)(struct type *type, FILE *f);
2624 ###### value functions
2626 static void structure_init(struct type *type, struct value *val)
2630 for (i = 0; i < type->structure.nfields; i++) {
2632 v = (void*) val->ptr + type->structure.fields[i].offset;
2633 if (type->structure.fields[i].init)
2634 dup_value(type->structure.fields[i].type,
2635 type->structure.fields[i].init,
2638 val_init(type->structure.fields[i].type, v);
2642 static void structure_free(struct type *type, struct value *val)
2646 for (i = 0; i < type->structure.nfields; i++) {
2648 v = (void*)val->ptr + type->structure.fields[i].offset;
2649 free_value(type->structure.fields[i].type, v);
2653 static void free_fieldlist(struct fieldlist *f)
2657 free_fieldlist(f->prev);
2662 static void structure_free_type(struct type *t)
2665 for (i = 0; i < t->structure.nfields; i++)
2666 if (t->structure.fields[i].init) {
2667 free_value(t->structure.fields[i].type,
2668 t->structure.fields[i].init);
2670 free(t->structure.fields);
2671 free_fieldlist(t->structure.field_list);
2674 static int structure_prepare_type(struct parse_context *c,
2675 struct type *t, int parse_time)
2678 struct fieldlist *f;
2680 if (!parse_time || t->structure.fields)
2683 for (f = t->structure.field_list; f; f=f->prev) {
2687 if (f->f.type->size <= 0)
2689 if (f->f.type->prepare_type)
2690 f->f.type->prepare_type(c, f->f.type, parse_time);
2692 if (f->init == NULL)
2696 propagate_types(f->init, c, &perr, f->f.type, 0);
2697 } while (perr & Eretry);
2699 c->parse_error += 1; // NOTEST
2702 t->structure.nfields = cnt;
2703 t->structure.fields = calloc(cnt, sizeof(struct field));
2704 f = t->structure.field_list;
2706 int a = f->f.type->align;
2708 t->structure.fields[cnt] = f->f;
2709 if (t->size & (a-1))
2710 t->size = (t->size | (a-1)) + 1;
2711 t->structure.fields[cnt].offset = t->size;
2712 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2716 if (f->init && !c->parse_error) {
2717 struct value vl = interp_exec(c, f->init, NULL);
2718 t->structure.fields[cnt].init =
2719 global_alloc(c, f->f.type, NULL, &vl);
2727 static struct type structure_prototype = {
2728 .init = structure_init,
2729 .free = structure_free,
2730 .free_type = structure_free_type,
2731 .print_type_decl = structure_print_type,
2732 .prepare_type = structure_prepare_type,
2746 ###### free exec cases
2748 free_exec(cast(fieldref, e)->left);
2752 ###### declare terminals
2757 | Term . IDENTIFIER ${ {
2758 struct fieldref *fr = new_pos(fieldref, $2);
2765 ###### print exec cases
2769 struct fieldref *f = cast(fieldref, e);
2770 print_exec(f->left, -1, bracket);
2771 printf(".%.*s", f->name.len, f->name.txt);
2775 ###### ast functions
2776 static int find_struct_index(struct type *type, struct text field)
2779 for (i = 0; i < type->structure.nfields; i++)
2780 if (text_cmp(type->structure.fields[i].name, field) == 0)
2785 ###### propagate exec cases
2789 struct fieldref *f = cast(fieldref, prog);
2790 struct type *st = propagate_types(f->left, c, perr, NULL, 0);
2793 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2795 else if (st->init != structure_init)
2796 type_err(c, "error: field reference attempted on %1, not a struct",
2797 f->left, st, 0, NULL);
2798 else if (f->index == -2) {
2799 f->index = find_struct_index(st, f->name);
2801 type_err(c, "error: cannot find requested field in %1",
2802 f->left, st, 0, NULL);
2804 if (f->index >= 0) {
2805 struct type *ft = st->structure.fields[f->index].type;
2806 if (!type_compat(type, ft, rules))
2807 type_err(c, "error: have %1 but need %2", prog,
2814 ###### interp exec cases
2817 struct fieldref *f = cast(fieldref, e);
2819 struct value *lleft = linterp_exec(c, f->left, <ype);
2820 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2821 rvtype = ltype->structure.fields[f->index].type;
2825 ###### top level grammar
2826 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2828 t = find_type(c, $ID.txt);
2830 t = add_type(c, $ID.txt, &structure_prototype);
2831 else if (t->size >= 0) {
2832 tok_err(c, "error: type already declared", &$ID);
2833 tok_err(c, "info: this is location of declartion", &t->first_use);
2834 /* Create a new one - duplicate */
2835 t = add_type(c, $ID.txt, &structure_prototype);
2837 struct type tmp = *t;
2838 *t = structure_prototype;
2842 t->structure.field_list = $<FB;
2847 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2848 | { SimpleFieldList } ${ $0 = $<SFL; }$
2849 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2850 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2852 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2853 | FieldLines SimpleFieldList Newlines ${
2858 SimpleFieldList -> Field ${ $0 = $<F; }$
2859 | SimpleFieldList ; Field ${
2863 | SimpleFieldList ; ${
2866 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2868 Field -> IDENTIFIER : Type = Expression ${ {
2869 $0 = calloc(1, sizeof(struct fieldlist));
2870 $0->f.name = $ID.txt;
2871 $0->f.type = $<Type;
2875 | IDENTIFIER : Type ${
2876 $0 = calloc(1, sizeof(struct fieldlist));
2877 $0->f.name = $ID.txt;
2878 $0->f.type = $<Type;
2881 ###### forward decls
2882 static void structure_print_type(struct type *t, FILE *f);
2884 ###### value functions
2885 static void structure_print_type(struct type *t, FILE *f)
2889 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2891 for (i = 0; i < t->structure.nfields; i++) {
2892 struct field *fl = t->structure.fields + i;
2893 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2894 type_print(fl->type, f);
2895 if (fl->type->print && fl->init) {
2897 if (fl->type == Tstr)
2898 fprintf(f, "\""); // UNTESTED
2899 print_value(fl->type, fl->init, f);
2900 if (fl->type == Tstr)
2901 fprintf(f, "\""); // UNTESTED
2907 ###### print type decls
2912 while (target != 0) {
2914 for (t = context.typelist; t ; t=t->next)
2915 if (!t->anon && t->print_type_decl &&
2925 t->print_type_decl(t, stdout);
2933 References, or pointers, are values that refer to another value. They
2934 can only refer to a `struct`, though as a struct can embed anything they
2935 can effectively refer to anything.
2937 References are potentially dangerous as they might refer to some
2938 variable which no longer exists - either because a stack frame
2939 containing it has been discarded or because the value was allocated on
2940 the heap and has now been free. Ocean does not yet provide any
2941 protection against these problems. It will in due course.
2943 With references comes the opportunity and the need to explicitly
2944 allocate values on the "heap" and to free them. We currently provide
2945 fairly basic support for this.
2947 Reference make use of the `@` symbol in various ways. A type that starts
2948 with `@` is a reference to whatever follows. A reference value
2949 followed by an `@` acts as the referred value, though the `@` is often
2950 not needed. Finally, an expression that starts with `@` is a special
2951 reference related expression. Some examples might help.
2953 ##### Example: Reference examples
2960 bar.number = 23; bar.string = "hello"
2971 Obviously this is very contrived. `ref` is a reference to a `foo` which
2972 is initially set to refer to the value stored in `bar` - no extra syntax
2973 is needed to "Take the address of" `bar` - the fact that `ref` is a
2974 reference means that only the address make sense.
2976 When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
2977 The same syntax is used for accessing fields both in structs and in
2978 references to structs. It would be correct to use `ref@.a`, but not
2981 `@new()` creates an object of whatever type is needed for the program
2982 to by type-correct. In future iterations of Ocean, arguments a
2983 constructor will access arguments, so the the syntax now looks like a
2984 function call. `@free` can be assigned any reference that was returned
2985 by `@new()`, and it will be freed. `@nil` is a value of whatever
2986 reference type is appropriate, and is stable and never the address of
2987 anything in the heap or on the stack. A reference can be assigned
2988 `@nil` or compared against that value.
2990 ###### declare terminals
2993 ###### type union fields
2996 struct type *referent;
2999 ###### value union fields
3002 ###### value functions
3004 static void reference_print_type(struct type *t, FILE *f)
3007 type_print(t->reference.referent, f);
3010 static int reference_cmp(struct type *tl, struct type *tr,
3011 struct value *left, struct value *right)
3013 return left->ref == right->ref ? 0 : 1;
3016 static void reference_dup(struct type *t,
3017 struct value *vold, struct value *vnew)
3019 vnew->ref = vold->ref;
3022 static void reference_free(struct type *t, struct value *v)
3024 /* Nothing to do here */
3027 static int reference_compat(struct type *require, struct type *have)
3029 if (have->compat != require->compat)
3031 if (have->reference.referent != require->reference.referent)
3036 static int reference_test(struct type *type, struct value *val)
3038 return val->ref != NULL;
3041 static struct type reference_prototype = {
3042 .print_type = reference_print_type,
3043 .cmp_eq = reference_cmp,
3044 .dup = reference_dup,
3045 .test = reference_test,
3046 .free = reference_free,
3047 .compat = reference_compat,
3048 .size = sizeof(void*),
3049 .align = sizeof(void*),
3055 struct type *t = find_type(c, $ID.txt);
3057 t = add_type(c, $ID.txt, NULL);
3060 $0 = find_anon_type(c, &reference_prototype, "@%.*s",
3061 $ID.txt.len, $ID.txt.txt);
3062 $0->reference.referent = t;
3065 ###### core functions
3066 static int text_is(struct text t, char *s)
3068 return (strlen(s) == t.len &&
3069 strncmp(s, t.txt, t.len) == 0);
3078 enum ref_func { RefNew, RefFree, RefNil } action;
3079 struct type *reftype;
3083 ###### SimpleStatement Grammar
3085 | @ IDENTIFIER = Expression ${ {
3086 struct ref *r = new_pos(ref, $ID);
3088 if (!text_is($ID.txt, "free"))
3089 tok_err(c, "error: only \"@free\" makes sense here",
3093 r->action = RefFree;
3097 ###### expression grammar
3098 | @ IDENTIFIER ( ) ${
3099 // Only 'new' valid here
3100 if (!text_is($ID.txt, "new")) {
3101 tok_err(c, "error: Only reference function is \"@new()\"",
3104 struct ref *r = new_pos(ref,$ID);
3110 // Only 'nil' valid here
3111 if (!text_is($ID.txt, "nil")) {
3112 tok_err(c, "error: Only reference value is \"@nil\"",
3115 struct ref *r = new_pos(ref,$ID);
3121 ###### print exec cases
3123 struct ref *r = cast(ref, e);
3124 switch (r->action) {
3126 printf("@new()"); break;
3128 printf("@nil"); break;
3130 do_indent(indent, "@free = ");
3131 print_exec(r->right, indent, bracket);
3137 ###### propagate exec cases
3139 struct ref *r = cast(ref, prog);
3140 switch (r->action) {
3142 if (type && type->free != reference_free) {
3143 type_err(c, "error: @new() can only be used with references, not %1",
3144 prog, type, 0, NULL);
3147 if (type && !r->reftype) {
3153 if (type && type->free != reference_free)
3154 type_err(c, "error: @nil can only be used with reference, not %1",
3155 prog, type, 0, NULL);
3156 if (type && !r->reftype) {
3162 t = propagate_types(r->right, c, perr, NULL, 0);
3163 if (t && t->free != reference_free)
3164 type_err(c, "error: @free can only be assigned a reference, not %1",
3173 ###### interp exec cases
3175 struct ref *r = cast(ref, e);
3176 switch (r->action) {
3179 rv.ref = calloc(1, r->reftype->reference.referent->size);
3180 rvtype = r->reftype;
3184 rvtype = r->reftype;
3187 rv = interp_exec(c, r->right, &rvtype);
3188 free_value(rvtype->reference.referent, rv.ref);
3196 ###### free exec cases
3198 struct ref *r = cast(ref, e);
3199 free_exec(r->right);
3204 ###### Expressions: dereference
3212 struct binode *b = new(binode);
3218 ###### print binode cases
3220 print_exec(b->left, -1, bracket);
3224 ###### propagate binode cases
3226 /* left must be a reference, and we return what it refers to */
3227 /* FIXME how can I pass the expected type down? */
3228 t = propagate_types(b->left, c, perr, NULL, 0);
3229 if (!t || t->free != reference_free)
3230 type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
3232 return t->reference.referent;
3235 ###### interp binode cases
3237 left = interp_exec(c, b->left, <ype);
3239 rvtype = ltype->reference.referent;
3246 A function is a chunk of code which can be passed parameters and can
3247 return results. Each function has a type which includes the set of
3248 parameters and the return value. As yet these types cannot be declared
3249 separately from the function itself.
3251 The parameters can be specified either in parentheses as a ';' separated
3254 ##### Example: function 1
3256 func main(av:[ac::number]string; env:[envc::number]string)
3259 or as an indented list of one parameter per line (though each line can
3260 be a ';' separated list)
3262 ##### Example: function 2
3265 argv:[argc::number]string
3266 env:[envc::number]string
3270 In the first case a return type can follow the parentheses after a colon,
3271 in the second it is given on a line starting with the word `return`.
3273 ##### Example: functions that return
3275 func add(a:number; b:number): number
3285 Rather than returning a type, the function can specify a set of local
3286 variables to return as a struct. The values of these variables when the
3287 function exits will be provided to the caller. For this the return type
3288 is replaced with a block of result declarations, either in parentheses
3289 or bracketed by `return` and `do`.
3291 ##### Example: functions returning multiple variables
3293 func to_cartesian(rho:number; theta:number):(x:number; y:number)
3306 For constructing the lists we use a `List` binode, which will be
3307 further detailed when Expression Lists are introduced.
3309 ###### type union fields
3312 struct binode *params;
3313 struct type *return_type;
3314 struct variable *scope;
3315 int inline_result; // return value is at start of 'local'
3319 ###### value union fields
3320 struct exec *function;
3322 ###### type functions
3323 void (*check_args)(struct parse_context *c, enum prop_err *perr,
3324 struct type *require, struct exec *args);
3326 ###### value functions
3328 static void function_free(struct type *type, struct value *val)
3330 free_exec(val->function);
3331 val->function = NULL;
3334 static int function_compat(struct type *require, struct type *have)
3336 // FIXME can I do anything here yet?
3340 static void function_check_args(struct parse_context *c, enum prop_err *perr,
3341 struct type *require, struct exec *args)
3343 /* This should be 'compat', but we don't have a 'tuple' type to
3344 * hold the type of 'args'
3346 struct binode *arg = cast(binode, args);
3347 struct binode *param = require->function.params;
3350 struct var *pv = cast(var, param->left);
3352 type_err(c, "error: insufficient arguments to function.",
3353 args, NULL, 0, NULL);
3357 propagate_types(arg->left, c, perr, pv->var->type, 0);
3358 param = cast(binode, param->right);
3359 arg = cast(binode, arg->right);
3362 type_err(c, "error: too many arguments to function.",
3363 args, NULL, 0, NULL);
3366 static void function_print(struct type *type, struct value *val, FILE *f)
3368 print_exec(val->function, 1, 0);
3371 static void function_print_type_decl(struct type *type, FILE *f)
3375 for (b = type->function.params; b; b = cast(binode, b->right)) {
3376 struct variable *v = cast(var, b->left)->var;
3377 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
3378 v->constant ? "::" : ":");
3379 type_print(v->type, f);
3384 if (type->function.return_type != Tnone) {
3386 if (type->function.inline_result) {
3388 struct type *t = type->function.return_type;
3390 for (i = 0; i < t->structure.nfields; i++) {
3391 struct field *fl = t->structure.fields + i;
3394 fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
3395 type_print(fl->type, f);
3399 type_print(type->function.return_type, f);
3404 static void function_free_type(struct type *t)
3406 free_exec(t->function.params);
3409 static struct type function_prototype = {
3410 .size = sizeof(void*),
3411 .align = sizeof(void*),
3412 .free = function_free,
3413 .compat = function_compat,
3414 .check_args = function_check_args,
3415 .print = function_print,
3416 .print_type_decl = function_print_type_decl,
3417 .free_type = function_free_type,
3420 ###### declare terminals
3430 FuncName -> IDENTIFIER ${ {
3431 struct variable *v = var_decl(c, $1.txt);
3432 struct var *e = new_pos(var, $1);
3439 v = var_ref(c, $1.txt);
3441 type_err(c, "error: function '%v' redeclared",
3443 type_err(c, "info: this is where '%v' was first declared",
3444 v->where_decl, NULL, 0, NULL);
3450 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
3451 | Args ArgsLine NEWLINE ${ {
3452 struct binode *b = $<AL;
3453 struct binode **bp = &b;
3455 bp = (struct binode **)&(*bp)->left;
3460 ArgsLine -> ${ $0 = NULL; }$
3461 | Varlist ${ $0 = $<1; }$
3462 | Varlist ; ${ $0 = $<1; }$
3464 Varlist -> Varlist ; ArgDecl ${
3465 $0 = new_pos(binode, $2);
3478 ArgDecl -> IDENTIFIER : FormalType ${ {
3479 struct variable *v = var_decl(c, $ID.txt);
3480 $0 = new_pos(var, $ID);
3487 ##### Function calls
3489 A function call can appear either as an expression or as a statement.
3490 We use a new 'Funcall' binode type to link the function with a list of
3491 arguments, form with the 'List' nodes.
3493 We have already seen the "Term" which is how a function call can appear
3494 in an expression. To parse a function call into a statement we include
3495 it in the "SimpleStatement Grammar" which will be described later.
3501 | Term ( ExpressionList ) ${ {
3502 struct binode *b = new(binode);
3505 b->right = reorder_bilist($<EL);
3509 struct binode *b = new(binode);
3516 ###### SimpleStatement Grammar
3518 | Term ( ExpressionList ) ${ {
3519 struct binode *b = new(binode);
3522 b->right = reorder_bilist($<EL);
3526 ###### print binode cases
3529 do_indent(indent, "");
3530 print_exec(b->left, -1, bracket);
3532 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3535 print_exec(b->left, -1, bracket);
3545 ###### propagate binode cases
3548 /* Every arg must match formal parameter, and result
3549 * is return type of function
3551 struct binode *args = cast(binode, b->right);
3552 struct var *v = cast(var, b->left);
3554 if (!v->var->type || v->var->type->check_args == NULL) {
3555 type_err(c, "error: attempt to call a non-function.",
3556 prog, NULL, 0, NULL);
3560 v->var->type->check_args(c, perr, v->var->type, args);
3561 if (v->var->type->function.inline_result)
3563 return v->var->type->function.return_type;
3566 ###### interp binode cases
3569 struct var *v = cast(var, b->left);
3570 struct type *t = v->var->type;
3571 void *oldlocal = c->local;
3572 int old_size = c->local_size;
3573 void *local = calloc(1, t->function.local_size);
3574 struct value *fbody = var_value(c, v->var);
3575 struct binode *arg = cast(binode, b->right);
3576 struct binode *param = t->function.params;
3579 struct var *pv = cast(var, param->left);
3580 struct type *vtype = NULL;
3581 struct value val = interp_exec(c, arg->left, &vtype);
3583 c->local = local; c->local_size = t->function.local_size;
3584 lval = var_value(c, pv->var);
3585 c->local = oldlocal; c->local_size = old_size;
3586 memcpy(lval, &val, vtype->size);
3587 param = cast(binode, param->right);
3588 arg = cast(binode, arg->right);
3590 c->local = local; c->local_size = t->function.local_size;
3591 if (t->function.inline_result && dtype) {
3592 _interp_exec(c, fbody->function, NULL, NULL);
3593 memcpy(dest, local, dtype->size);
3594 rvtype = ret.type = NULL;
3596 rv = interp_exec(c, fbody->function, &rvtype);
3597 c->local = oldlocal; c->local_size = old_size;
3602 ## Complex executables: statements and expressions
3604 Now that we have types and values and variables and most of the basic
3605 Terms which provide access to these, we can explore the more complex
3606 code that combine all of these to get useful work done. Specifically
3607 statements and expressions.
3609 Expressions are various combinations of Terms. We will use operator
3610 precedence to ensure correct parsing. The simplest Expression is just a
3611 Term - others will follow.
3616 Expression -> Term ${ $0 = $<Term; }$
3617 ## expression grammar
3619 ### Expressions: Conditional
3621 Our first user of the `binode` will be conditional expressions, which
3622 is a bit odd as they actually have three components. That will be
3623 handled by having 2 binodes for each expression. The conditional
3624 expression is the lowest precedence operator which is why we define it
3625 first - to start the precedence list.
3627 Conditional expressions are of the form "value `if` condition `else`
3628 other_value". They associate to the right, so everything to the right
3629 of `else` is part of an else value, while only a higher-precedence to
3630 the left of `if` is the if values. Between `if` and `else` there is no
3631 room for ambiguity, so a full conditional expression is allowed in
3637 ###### declare terminals
3641 ###### expression grammar
3643 | Expression if Expression else Expression $$ifelse ${ {
3644 struct binode *b1 = new(binode);
3645 struct binode *b2 = new(binode);
3655 ###### print binode cases
3658 b2 = cast(binode, b->right);
3659 if (bracket) printf("(");
3660 print_exec(b2->left, -1, bracket);
3662 print_exec(b->left, -1, bracket);
3664 print_exec(b2->right, -1, bracket);
3665 if (bracket) printf(")");
3668 ###### propagate binode cases
3671 /* cond must be Tbool, others must match */
3672 struct binode *b2 = cast(binode, b->right);
3675 propagate_types(b->left, c, perr, Tbool, 0);
3676 t = propagate_types(b2->left, c, perr, type, Rnolabel);
3677 t2 = propagate_types(b2->right, c, perr, type ?: t, Rnolabel);
3681 ###### interp binode cases
3684 struct binode *b2 = cast(binode, b->right);
3685 left = interp_exec(c, b->left, <ype);
3687 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3689 rv = interp_exec(c, b2->right, &rvtype);
3695 We take a brief detour, now that we have expressions, to describe lists
3696 of expressions. These will be needed for function parameters and
3697 possibly other situations. They seem generic enough to introduce here
3698 to be used elsewhere.
3700 And ExpressionList will use the `List` type of `binode`, building up at
3701 the end. And place where they are used will probably call
3702 `reorder_bilist()` to get a more normal first/next arrangement.
3704 ###### declare terminals
3707 `List` execs have no implicit semantics, so they are never propagated or
3708 interpreted. The can be printed as a comma separate list, which is how
3709 they are parsed. Note they are also used for function formal parameter
3710 lists. In that case a separate function is used to print them.
3712 ###### print binode cases
3716 print_exec(b->left, -1, bracket);
3719 b = cast(binode, b->right);
3723 ###### propagate binode cases
3724 case List: abort(); // NOTEST
3725 ###### interp binode cases
3726 case List: abort(); // NOTEST
3731 ExpressionList -> ExpressionList , Expression ${
3744 ### Expressions: Boolean
3746 The next class of expressions to use the `binode` will be Boolean
3747 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3748 have same corresponding precendence. The difference is that they don't
3749 evaluate the second expression if not necessary.
3758 ###### declare terminals
3763 ###### expression grammar
3764 | Expression or Expression ${ {
3765 struct binode *b = new(binode);
3771 | Expression or else Expression ${ {
3772 struct binode *b = new(binode);
3779 | Expression and Expression ${ {
3780 struct binode *b = new(binode);
3786 | Expression and then Expression ${ {
3787 struct binode *b = new(binode);
3794 | not Expression ${ {
3795 struct binode *b = new(binode);
3801 ###### print binode cases
3803 if (bracket) printf("(");
3804 print_exec(b->left, -1, bracket);
3806 print_exec(b->right, -1, bracket);
3807 if (bracket) printf(")");
3810 if (bracket) printf("(");
3811 print_exec(b->left, -1, bracket);
3812 printf(" and then ");
3813 print_exec(b->right, -1, bracket);
3814 if (bracket) printf(")");
3817 if (bracket) printf("(");
3818 print_exec(b->left, -1, bracket);
3820 print_exec(b->right, -1, bracket);
3821 if (bracket) printf(")");
3824 if (bracket) printf("(");
3825 print_exec(b->left, -1, bracket);
3826 printf(" or else ");
3827 print_exec(b->right, -1, bracket);
3828 if (bracket) printf(")");
3831 if (bracket) printf("(");
3833 print_exec(b->right, -1, bracket);
3834 if (bracket) printf(")");
3837 ###### propagate binode cases
3843 /* both must be Tbool, result is Tbool */
3844 propagate_types(b->left, c, perr, Tbool, 0);
3845 propagate_types(b->right, c, perr, Tbool, 0);
3846 if (type && type != Tbool)
3847 type_err(c, "error: %1 operation found where %2 expected", prog,
3851 ###### interp binode cases
3853 rv = interp_exec(c, b->left, &rvtype);
3854 right = interp_exec(c, b->right, &rtype);
3855 rv.bool = rv.bool && right.bool;
3858 rv = interp_exec(c, b->left, &rvtype);
3860 rv = interp_exec(c, b->right, NULL);
3863 rv = interp_exec(c, b->left, &rvtype);
3864 right = interp_exec(c, b->right, &rtype);
3865 rv.bool = rv.bool || right.bool;
3868 rv = interp_exec(c, b->left, &rvtype);
3870 rv = interp_exec(c, b->right, NULL);
3873 rv = interp_exec(c, b->right, &rvtype);
3877 ### Expressions: Comparison
3879 Of slightly higher precedence that Boolean expressions are Comparisons.
3880 A comparison takes arguments of any comparable type, but the two types
3883 To simplify the parsing we introduce an `eop` which can record an
3884 expression operator, and the `CMPop` non-terminal will match one of them.
3891 ###### ast functions
3892 static void free_eop(struct eop *e)
3906 ###### declare terminals
3907 $LEFT < > <= >= == != CMPop
3909 ###### expression grammar
3910 | Expression CMPop Expression ${ {
3911 struct binode *b = new(binode);
3921 CMPop -> < ${ $0.op = Less; }$
3922 | > ${ $0.op = Gtr; }$
3923 | <= ${ $0.op = LessEq; }$
3924 | >= ${ $0.op = GtrEq; }$
3925 | == ${ $0.op = Eql; }$
3926 | != ${ $0.op = NEql; }$
3928 ###### print binode cases
3936 if (bracket) printf("(");
3937 print_exec(b->left, -1, bracket);
3939 case Less: printf(" < "); break;
3940 case LessEq: printf(" <= "); break;
3941 case Gtr: printf(" > "); break;
3942 case GtrEq: printf(" >= "); break;
3943 case Eql: printf(" == "); break;
3944 case NEql: printf(" != "); break;
3945 default: abort(); // NOTEST
3947 print_exec(b->right, -1, bracket);
3948 if (bracket) printf(")");
3951 ###### propagate binode cases
3958 /* Both must match but not be labels, result is Tbool */
3959 t = propagate_types(b->left, c, perr, NULL, Rnolabel);
3961 propagate_types(b->right, c, perr, t, 0);
3963 t = propagate_types(b->right, c, perr, NULL, Rnolabel); // UNTESTED
3965 t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
3967 if (!type_compat(type, Tbool, 0))
3968 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3969 Tbool, rules, type);
3972 ###### interp binode cases
3981 left = interp_exec(c, b->left, <ype);
3982 right = interp_exec(c, b->right, &rtype);
3983 cmp = value_cmp(ltype, rtype, &left, &right);
3986 case Less: rv.bool = cmp < 0; break;
3987 case LessEq: rv.bool = cmp <= 0; break;
3988 case Gtr: rv.bool = cmp > 0; break;
3989 case GtrEq: rv.bool = cmp >= 0; break;
3990 case Eql: rv.bool = cmp == 0; break;
3991 case NEql: rv.bool = cmp != 0; break;
3992 default: rv.bool = 0; break; // NOTEST
3997 ### Expressions: Arithmetic etc.
3999 The remaining expressions with the highest precedence are arithmetic,
4000 string concatenation, string conversion, and testing. String concatenation
4001 (`++`) has the same precedence as multiplication and division, but lower
4004 Testing comes in two forms. A single question mark (`?`) is a uniary
4005 operator which converts come types into Boolean. The general meaning is
4006 "is this a value value" and there will be more uses as the language
4007 develops. A double questionmark (`??`) is a binary operator (Choose),
4008 with same precedence as multiplication, which returns the LHS if it
4009 tests successfully, else returns the RHS.
4011 String conversion is a temporary feature until I get a better type
4012 system. `$` is a prefix operator which expects a string and returns
4015 `+` and `-` are both infix and prefix operations (where they are
4016 absolute value and negation). These have different operator names.
4018 We also have a 'Bracket' operator which records where parentheses were
4019 found. This makes it easy to reproduce these when printing. Possibly I
4020 should only insert brackets were needed for precedence. Putting
4021 parentheses around an expression converts it into a Term,
4027 Absolute, Negate, Test,
4031 ###### declare terminals
4033 $LEFT * / % ++ ?? Top
4037 ###### expression grammar
4038 | Expression Eop Expression ${ {
4039 struct binode *b = new(binode);
4046 | Expression Top Expression ${ {
4047 struct binode *b = new(binode);
4054 | Uop Expression ${ {
4055 struct binode *b = new(binode);
4063 | ( Expression ) ${ {
4064 struct binode *b = new_pos(binode, $1);
4073 Eop -> + ${ $0.op = Plus; }$
4074 | - ${ $0.op = Minus; }$
4076 Uop -> + ${ $0.op = Absolute; }$
4077 | - ${ $0.op = Negate; }$
4078 | $ ${ $0.op = StringConv; }$
4079 | ? ${ $0.op = Test; }$
4081 Top -> * ${ $0.op = Times; }$
4082 | / ${ $0.op = Divide; }$
4083 | % ${ $0.op = Rem; }$
4084 | ++ ${ $0.op = Concat; }$
4085 | ?? ${ $0.op = Choose; }$
4087 ###### print binode cases
4095 if (bracket) printf("(");
4096 print_exec(b->left, indent, bracket);
4098 case Plus: fputs(" + ", stdout); break;
4099 case Minus: fputs(" - ", stdout); break;
4100 case Times: fputs(" * ", stdout); break;
4101 case Divide: fputs(" / ", stdout); break;
4102 case Rem: fputs(" % ", stdout); break;
4103 case Concat: fputs(" ++ ", stdout); break;
4104 case Choose: fputs(" ?? ", stdout); break;
4105 default: abort(); // NOTEST
4107 print_exec(b->right, indent, bracket);
4108 if (bracket) printf(")");
4114 if (bracket) printf("(");
4116 case Absolute: fputs("+", stdout); break;
4117 case Negate: fputs("-", stdout); break;
4118 case StringConv: fputs("$", stdout); break;
4119 case Test: fputs("?", stdout); break;
4120 default: abort(); // NOTEST
4122 print_exec(b->right, indent, bracket);
4123 if (bracket) printf(")");
4127 print_exec(b->right, indent, bracket);
4131 ###### propagate binode cases
4137 /* both must be numbers, result is Tnum */
4140 /* as propagate_types ignores a NULL,
4141 * unary ops fit here too */
4142 propagate_types(b->left, c, perr, Tnum, 0);
4143 propagate_types(b->right, c, perr, Tnum, 0);
4144 if (!type_compat(type, Tnum, 0))
4145 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
4150 /* both must be Tstr, result is Tstr */
4151 propagate_types(b->left, c, perr, Tstr, 0);
4152 propagate_types(b->right, c, perr, Tstr, 0);
4153 if (!type_compat(type, Tstr, 0))
4154 type_err(c, "error: Concat returns %1 but %2 expected", prog,
4159 /* op must be string, result is number */
4160 propagate_types(b->left, c, perr, Tstr, 0);
4161 if (!type_compat(type, Tnum, 0))
4162 type_err(c, // UNTESTED
4163 "error: Can only convert string to number, not %1",
4164 prog, type, 0, NULL);
4168 /* LHS must support ->test, result is Tbool */
4169 t = propagate_types(b->right, c, perr, NULL, 0);
4171 type_err(c, "error: '?' requires a testable value, not %1",
4176 /* LHS and RHS must match and are returned. Must support
4179 t = propagate_types(b->left, c, perr, type, rules);
4180 t = propagate_types(b->right, c, perr, t, rules);
4181 if (t && t->test == NULL)
4182 type_err(c, "error: \"??\" requires a testable value, not %1",
4187 return propagate_types(b->right, c, perr, type, 0);
4189 ###### interp binode cases
4192 rv = interp_exec(c, b->left, &rvtype);
4193 right = interp_exec(c, b->right, &rtype);
4194 mpq_add(rv.num, rv.num, right.num);
4197 rv = interp_exec(c, b->left, &rvtype);
4198 right = interp_exec(c, b->right, &rtype);
4199 mpq_sub(rv.num, rv.num, right.num);
4202 rv = interp_exec(c, b->left, &rvtype);
4203 right = interp_exec(c, b->right, &rtype);
4204 mpq_mul(rv.num, rv.num, right.num);
4207 rv = interp_exec(c, b->left, &rvtype);
4208 right = interp_exec(c, b->right, &rtype);
4209 mpq_div(rv.num, rv.num, right.num);
4214 left = interp_exec(c, b->left, <ype);
4215 right = interp_exec(c, b->right, &rtype);
4216 mpz_init(l); mpz_init(r); mpz_init(rem);
4217 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
4218 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
4219 mpz_tdiv_r(rem, l, r);
4220 val_init(Tnum, &rv);
4221 mpq_set_z(rv.num, rem);
4222 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
4227 rv = interp_exec(c, b->right, &rvtype);
4228 mpq_neg(rv.num, rv.num);
4231 rv = interp_exec(c, b->right, &rvtype);
4232 mpq_abs(rv.num, rv.num);
4235 rv = interp_exec(c, b->right, &rvtype);
4238 left = interp_exec(c, b->left, <ype);
4239 right = interp_exec(c, b->right, &rtype);
4241 rv.str = text_join(left.str, right.str);
4244 right = interp_exec(c, b->right, &rvtype);
4248 struct text tx = right.str;
4251 if (tx.txt[0] == '-') {
4252 neg = 1; // UNTESTED
4253 tx.txt++; // UNTESTED
4254 tx.len--; // UNTESTED
4256 if (number_parse(rv.num, tail, tx) == 0)
4257 mpq_init(rv.num); // UNTESTED
4259 mpq_neg(rv.num, rv.num); // UNTESTED
4261 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
4265 right = interp_exec(c, b->right, &rtype);
4267 rv.bool = !!rtype->test(rtype, &right);
4270 left = interp_exec(c, b->left, <ype);
4271 if (ltype->test(ltype, &left)) {
4276 rv = interp_exec(c, b->right, &rvtype);
4279 ###### value functions
4281 static struct text text_join(struct text a, struct text b)
4284 rv.len = a.len + b.len;
4285 rv.txt = malloc(rv.len);
4286 memcpy(rv.txt, a.txt, a.len);
4287 memcpy(rv.txt+a.len, b.txt, b.len);
4291 ### Blocks, Statements, and Statement lists.
4293 Now that we have expressions out of the way we need to turn to
4294 statements. There are simple statements and more complex statements.
4295 Simple statements do not contain (syntactic) newlines, complex statements do.
4297 Statements often come in sequences and we have corresponding simple
4298 statement lists and complex statement lists.
4299 The former comprise only simple statements separated by semicolons.
4300 The later comprise complex statements and simple statement lists. They are
4301 separated by newlines. Thus the semicolon is only used to separate
4302 simple statements on the one line. This may be overly restrictive,
4303 but I'm not sure I ever want a complex statement to share a line with
4306 Note that a simple statement list can still use multiple lines if
4307 subsequent lines are indented, so
4309 ###### Example: wrapped simple statement list
4314 is a single simple statement list. This might allow room for
4315 confusion, so I'm not set on it yet.
4317 A simple statement list needs no extra syntax. A complex statement
4318 list has two syntactic forms. It can be enclosed in braces (much like
4319 C blocks), or it can be introduced by an indent and continue until an
4320 unindented newline (much like Python blocks). With this extra syntax
4321 it is referred to as a block.
4323 Note that a block does not have to include any newlines if it only
4324 contains simple statements. So both of:
4326 if condition: a=b; d=f
4328 if condition { a=b; print f }
4332 In either case the list is constructed from a `binode` list with
4333 `Block` as the operator. When parsing the list it is most convenient
4334 to append to the end, so a list is a list and a statement. When using
4335 the list it is more convenient to consider a list to be a statement
4336 and a list. So we need a function to re-order a list.
4337 `reorder_bilist` serves this purpose.
4339 The only stand-alone statement we introduce at this stage is `pass`
4340 which does nothing and is represented as a `NULL` pointer in a `Block`
4341 list. Other stand-alone statements will follow once the infrastructure
4344 As many statements will use binodes, we declare a binode pointer 'b' in
4345 the common header for all reductions to use.
4347 ###### Parser: reduce
4358 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4359 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4360 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4361 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4362 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4364 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4365 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4366 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4367 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4368 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4370 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4371 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4372 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
4374 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
4375 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
4376 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
4377 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
4378 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
4380 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
4382 ComplexStatements -> ComplexStatements ComplexStatement ${
4392 | ComplexStatement ${
4404 ComplexStatement -> SimpleStatements Newlines ${
4405 $0 = reorder_bilist($<SS);
4407 | SimpleStatements ; Newlines ${
4408 $0 = reorder_bilist($<SS);
4410 ## ComplexStatement Grammar
4413 SimpleStatements -> SimpleStatements ; SimpleStatement ${
4419 | SimpleStatement ${
4428 SimpleStatement -> pass ${ $0 = NULL; }$
4429 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
4430 ## SimpleStatement Grammar
4432 ###### print binode cases
4436 if (b->left == NULL) // UNTESTED
4437 printf("pass"); // UNTESTED
4439 print_exec(b->left, indent, bracket); // UNTESTED
4440 if (b->right) { // UNTESTED
4441 printf("; "); // UNTESTED
4442 print_exec(b->right, indent, bracket); // UNTESTED
4445 // block, one per line
4446 if (b->left == NULL)
4447 do_indent(indent, "pass\n");
4449 print_exec(b->left, indent, bracket);
4451 print_exec(b->right, indent, bracket);
4455 ###### propagate binode cases
4458 /* If any statement returns something other than Tnone
4459 * or Tbool then all such must return same type.
4460 * As each statement may be Tnone or something else,
4461 * we must always pass NULL (unknown) down, otherwise an incorrect
4462 * error might occur. We never return Tnone unless it is
4467 for (e = b; e; e = cast(binode, e->right)) {
4468 t = propagate_types(e->left, c, perr, NULL, rules);
4469 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
4471 if (t == Tnone && e->right)
4472 /* Only the final statement *must* return a value
4480 type_err(c, "error: expected %1%r, found %2",
4481 e->left, type, rules, t);
4487 ###### interp binode cases
4489 while (rvtype == Tnone &&
4492 rv = interp_exec(c, b->left, &rvtype);
4493 b = cast(binode, b->right);
4497 ### The Print statement
4499 `print` is a simple statement that takes a comma-separated list of
4500 expressions and prints the values separated by spaces and terminated
4501 by a newline. No control of formatting is possible.
4503 `print` uses `ExpressionList` to collect the expressions and stores them
4504 on the left side of a `Print` binode unlessthere is a trailing comma
4505 when the list is stored on the `right` side and no trailing newline is
4511 ##### declare terminals
4514 ###### SimpleStatement Grammar
4516 | print ExpressionList ${
4517 $0 = b = new_pos(binode, $1);
4520 b->left = reorder_bilist($<EL);
4522 | print ExpressionList , ${ {
4523 $0 = b = new_pos(binode, $1);
4525 b->right = reorder_bilist($<EL);
4529 $0 = b = new_pos(binode, $1);
4535 ###### print binode cases
4538 do_indent(indent, "print");
4540 print_exec(b->right, -1, bracket);
4543 print_exec(b->left, -1, bracket);
4548 ###### propagate binode cases
4551 /* don't care but all must be consistent */
4553 b = cast(binode, b->left);
4555 b = cast(binode, b->right);
4557 propagate_types(b->left, c, perr, NULL, Rnolabel);
4558 b = cast(binode, b->right);
4562 ###### interp binode cases
4566 struct binode *b2 = cast(binode, b->left);
4568 b2 = cast(binode, b->right);
4569 for (; b2; b2 = cast(binode, b2->right)) {
4570 left = interp_exec(c, b2->left, <ype);
4571 print_value(ltype, &left, stdout);
4572 free_value(ltype, &left);
4576 if (b->right == NULL)
4582 ###### Assignment statement
4584 An assignment will assign a value to a variable, providing it hasn't
4585 been declared as a constant. The analysis phase ensures that the type
4586 will be correct so the interpreter just needs to perform the
4587 calculation. There is a form of assignment which declares a new
4588 variable as well as assigning a value. If a name is assigned before
4589 it is declared, and error will be raised as the name is created as
4590 `Tlabel` and it is illegal to assign to such names.
4596 ###### declare terminals
4599 ###### SimpleStatement Grammar
4600 | Term = Expression ${
4601 $0 = b= new(binode);
4606 | VariableDecl = Expression ${
4607 $0 = b= new(binode);
4614 if ($1->var->where_set == NULL) {
4616 "Variable declared with no type or value: %v",
4620 $0 = b = new(binode);
4627 ###### print binode cases
4630 do_indent(indent, "");
4631 print_exec(b->left, -1, bracket);
4633 print_exec(b->right, -1, bracket);
4640 struct variable *v = cast(var, b->left)->var;
4641 do_indent(indent, "");
4642 print_exec(b->left, -1, bracket);
4643 if (cast(var, b->left)->var->constant) {
4645 if (v->explicit_type) {
4646 type_print(v->type, stdout);
4651 if (v->explicit_type) {
4652 type_print(v->type, stdout);
4658 print_exec(b->right, -1, bracket);
4665 ###### propagate binode cases
4669 /* Both must match and not be labels,
4670 * Type must support 'dup',
4671 * For Assign, left must not be constant.
4674 t = propagate_types(b->left, c, perr, NULL,
4675 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
4680 if (propagate_types(b->right, c, perr, t, 0) != t)
4681 if (b->left->type == Xvar)
4682 type_err(c, "info: variable '%v' was set as %1 here.",
4683 cast(var, b->left)->var->where_set, t, rules, NULL);
4685 t = propagate_types(b->right, c, perr, NULL, Rnolabel);
4687 propagate_types(b->left, c, perr, t,
4688 (b->op == Assign ? Rnoconstant : 0));
4690 if (t && t->dup == NULL && !(*perr & Emaycopy))
4691 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4696 ###### interp binode cases
4699 lleft = linterp_exec(c, b->left, <ype);
4701 dinterp_exec(c, b->right, lleft, ltype, 1);
4707 struct variable *v = cast(var, b->left)->var;
4710 val = var_value(c, v);
4711 if (v->type->prepare_type)
4712 v->type->prepare_type(c, v->type, 0);
4714 dinterp_exec(c, b->right, val, v->type, 0);
4716 val_init(v->type, val);
4720 ### The `use` statement
4722 The `use` statement is the last "simple" statement. It is needed when a
4723 statement block can return a value. This includes the body of a
4724 function which has a return type, and the "condition" code blocks in
4725 `if`, `while`, and `switch` statements.
4730 ###### declare terminals
4733 ###### SimpleStatement Grammar
4735 $0 = b = new_pos(binode, $1);
4738 if (b->right->type == Xvar) {
4739 struct var *v = cast(var, b->right);
4740 if (v->var->type == Tnone) {
4741 /* Convert this to a label */
4744 v->var->type = Tlabel;
4745 val = global_alloc(c, Tlabel, v->var, NULL);
4751 ###### print binode cases
4754 do_indent(indent, "use ");
4755 print_exec(b->right, -1, bracket);
4760 ###### propagate binode cases
4763 /* result matches value */
4764 return propagate_types(b->right, c, perr, type, 0);
4766 ###### interp binode cases
4769 rv = interp_exec(c, b->right, &rvtype);
4772 ### The Conditional Statement
4774 This is the biggy and currently the only complex statement. This
4775 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4776 It is comprised of a number of parts, all of which are optional though
4777 set combinations apply. Each part is (usually) a key word (`then` is
4778 sometimes optional) followed by either an expression or a code block,
4779 except the `casepart` which is a "key word and an expression" followed
4780 by a code block. The code-block option is valid for all parts and,
4781 where an expression is also allowed, the code block can use the `use`
4782 statement to report a value. If the code block does not report a value
4783 the effect is similar to reporting `True`.
4785 The `else` and `case` parts, as well as `then` when combined with
4786 `if`, can contain a `use` statement which will apply to some
4787 containing conditional statement. `for` parts, `do` parts and `then`
4788 parts used with `for` can never contain a `use`, except in some
4789 subordinate conditional statement.
4791 If there is a `forpart`, it is executed first, only once.
4792 If there is a `dopart`, then it is executed repeatedly providing
4793 always that the `condpart` or `cond`, if present, does not return a non-True
4794 value. `condpart` can fail to return any value if it simply executes
4795 to completion. This is treated the same as returning `True`.
4797 If there is a `thenpart` it will be executed whenever the `condpart`
4798 or `cond` returns True (or does not return any value), but this will happen
4799 *after* `dopart` (when present).
4801 If `elsepart` is present it will be executed at most once when the
4802 condition returns `False` or some value that isn't `True` and isn't
4803 matched by any `casepart`. If there are any `casepart`s, they will be
4804 executed when the condition returns a matching value.
4806 The particular sorts of values allowed in case parts has not yet been
4807 determined in the language design, so nothing is prohibited.
4809 The various blocks in this complex statement potentially provide scope
4810 for variables as described earlier. Each such block must include the
4811 "OpenScope" nonterminal before parsing the block, and must call
4812 `var_block_close()` when closing the block.
4814 The code following "`if`", "`switch`" and "`for`" does not get its own
4815 scope, but is in a scope covering the whole statement, so names
4816 declared there cannot be redeclared elsewhere. Similarly the
4817 condition following "`while`" is in a scope the covers the body
4818 ("`do`" part) of the loop, and which does not allow conditional scope
4819 extension. Code following "`then`" (both looping and non-looping),
4820 "`else`" and "`case`" each get their own local scope.
4822 The type requirements on the code block in a `whilepart` are quite
4823 unusal. It is allowed to return a value of some identifiable type, in
4824 which case the loop aborts and an appropriate `casepart` is run, or it
4825 can return a Boolean, in which case the loop either continues to the
4826 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4827 This is different both from the `ifpart` code block which is expected to
4828 return a Boolean, or the `switchpart` code block which is expected to
4829 return the same type as the casepart values. The correct analysis of
4830 the type of the `whilepart` code block is the reason for the
4831 `Rboolok` flag which is passed to `propagate_types()`.
4833 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4834 defined. As there are two scopes which cover multiple parts - one for
4835 the whole statement and one for "while" and "do" - and as we will use
4836 the 'struct exec' to track scopes, we actually need two new types of
4837 exec. One is a `binode` for the looping part, the rest is the
4838 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4839 casepart` to track a list of case parts.
4850 struct exec *action;
4851 struct casepart *next;
4853 struct cond_statement {
4855 struct exec *forpart, *condpart, *thenpart, *elsepart;
4856 struct binode *looppart;
4857 struct casepart *casepart;
4860 ###### ast functions
4862 static void free_casepart(struct casepart *cp)
4866 free_exec(cp->value);
4867 free_exec(cp->action);
4874 static void free_cond_statement(struct cond_statement *s)
4878 free_exec(s->forpart);
4879 free_exec(s->condpart);
4880 free_exec(s->looppart);
4881 free_exec(s->thenpart);
4882 free_exec(s->elsepart);
4883 free_casepart(s->casepart);
4887 ###### free exec cases
4888 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4890 ###### ComplexStatement Grammar
4891 | CondStatement ${ $0 = $<1; }$
4893 ###### declare terminals
4894 $TERM for then while do
4901 // A CondStatement must end with EOL, as does CondSuffix and
4903 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4904 // may or may not end with EOL
4905 // WhilePart and IfPart include an appropriate Suffix
4907 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4908 // them. WhilePart opens and closes its own scope.
4909 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4912 $0->thenpart = $<TP;
4913 $0->looppart = $<WP;
4914 var_block_close(c, CloseSequential, $0);
4916 | ForPart OptNL WhilePart CondSuffix ${
4919 $0->looppart = $<WP;
4920 var_block_close(c, CloseSequential, $0);
4922 | WhilePart CondSuffix ${
4924 $0->looppart = $<WP;
4926 | SwitchPart OptNL CasePart CondSuffix ${
4928 $0->condpart = $<SP;
4929 $CP->next = $0->casepart;
4930 $0->casepart = $<CP;
4931 var_block_close(c, CloseSequential, $0);
4933 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4935 $0->condpart = $<SP;
4936 $CP->next = $0->casepart;
4937 $0->casepart = $<CP;
4938 var_block_close(c, CloseSequential, $0);
4940 | IfPart IfSuffix ${
4942 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4943 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4944 // This is where we close an "if" statement
4945 var_block_close(c, CloseSequential, $0);
4948 CondSuffix -> IfSuffix ${
4951 | Newlines CasePart CondSuffix ${
4953 $CP->next = $0->casepart;
4954 $0->casepart = $<CP;
4956 | CasePart CondSuffix ${
4958 $CP->next = $0->casepart;
4959 $0->casepart = $<CP;
4962 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4963 | Newlines ElsePart ${ $0 = $<EP; }$
4964 | ElsePart ${$0 = $<EP; }$
4966 ElsePart -> else OpenBlock Newlines ${
4967 $0 = new(cond_statement);
4968 $0->elsepart = $<OB;
4969 var_block_close(c, CloseElse, $0->elsepart);
4971 | else OpenScope CondStatement ${
4972 $0 = new(cond_statement);
4973 $0->elsepart = $<CS;
4974 var_block_close(c, CloseElse, $0->elsepart);
4978 CasePart -> case Expression OpenScope ColonBlock ${
4979 $0 = calloc(1,sizeof(struct casepart));
4982 var_block_close(c, CloseParallel, $0->action);
4986 // These scopes are closed in CondStatement
4987 ForPart -> for OpenBlock ${
4991 ThenPart -> then OpenBlock ${
4993 var_block_close(c, CloseSequential, $0);
4997 // This scope is closed in CondStatement
4998 WhilePart -> while UseBlock OptNL do OpenBlock ${
5003 var_block_close(c, CloseSequential, $0->right);
5004 var_block_close(c, CloseSequential, $0);
5006 | while OpenScope Expression OpenScope ColonBlock ${
5011 var_block_close(c, CloseSequential, $0->right);
5012 var_block_close(c, CloseSequential, $0);
5016 IfPart -> if UseBlock OptNL then OpenBlock ${
5019 var_block_close(c, CloseParallel, $0.thenpart);
5021 | if OpenScope Expression OpenScope ColonBlock ${
5024 var_block_close(c, CloseParallel, $0.thenpart);
5026 | if OpenScope Expression OpenScope OptNL then Block ${
5029 var_block_close(c, CloseParallel, $0.thenpart);
5033 // This scope is closed in CondStatement
5034 SwitchPart -> switch OpenScope Expression ${
5037 | switch UseBlock ${
5041 ###### print binode cases
5043 if (b->left && b->left->type == Xbinode &&
5044 cast(binode, b->left)->op == Block) {
5046 do_indent(indent, "while {\n");
5048 do_indent(indent, "while\n");
5049 print_exec(b->left, indent+1, bracket);
5051 do_indent(indent, "} do {\n");
5053 do_indent(indent, "do\n");
5054 print_exec(b->right, indent+1, bracket);
5056 do_indent(indent, "}\n");
5058 do_indent(indent, "while ");
5059 print_exec(b->left, 0, bracket);
5064 print_exec(b->right, indent+1, bracket);
5066 do_indent(indent, "}\n");
5070 ###### print exec cases
5072 case Xcond_statement:
5074 struct cond_statement *cs = cast(cond_statement, e);
5075 struct casepart *cp;
5077 do_indent(indent, "for");
5078 if (bracket) printf(" {\n"); else printf("\n");
5079 print_exec(cs->forpart, indent+1, bracket);
5082 do_indent(indent, "} then {\n");
5084 do_indent(indent, "then\n");
5085 print_exec(cs->thenpart, indent+1, bracket);
5087 if (bracket) do_indent(indent, "}\n");
5090 print_exec(cs->looppart, indent, bracket);
5094 do_indent(indent, "switch");
5096 do_indent(indent, "if");
5097 if (cs->condpart && cs->condpart->type == Xbinode &&
5098 cast(binode, cs->condpart)->op == Block) {
5103 print_exec(cs->condpart, indent+1, bracket);
5105 do_indent(indent, "}\n");
5107 do_indent(indent, "then\n");
5108 print_exec(cs->thenpart, indent+1, bracket);
5112 print_exec(cs->condpart, 0, bracket);
5118 print_exec(cs->thenpart, indent+1, bracket);
5120 do_indent(indent, "}\n");
5125 for (cp = cs->casepart; cp; cp = cp->next) {
5126 do_indent(indent, "case ");
5127 print_exec(cp->value, -1, 0);
5132 print_exec(cp->action, indent+1, bracket);
5134 do_indent(indent, "}\n");
5137 do_indent(indent, "else");
5142 print_exec(cs->elsepart, indent+1, bracket);
5144 do_indent(indent, "}\n");
5149 ###### propagate binode cases
5151 t = propagate_types(b->right, c, perr, Tnone, 0);
5152 if (!type_compat(Tnone, t, 0))
5153 *perr |= Efail; // UNTESTED
5154 return propagate_types(b->left, c, perr, type, rules);
5156 ###### propagate exec cases
5157 case Xcond_statement:
5159 // forpart and looppart->right must return Tnone
5160 // thenpart must return Tnone if there is a loopart,
5161 // otherwise it is like elsepart.
5163 // be bool if there is no casepart
5164 // match casepart->values if there is a switchpart
5165 // either be bool or match casepart->value if there
5167 // elsepart and casepart->action must match the return type
5168 // expected of this statement.
5169 struct cond_statement *cs = cast(cond_statement, prog);
5170 struct casepart *cp;
5172 t = propagate_types(cs->forpart, c, perr, Tnone, 0);
5173 if (!type_compat(Tnone, t, 0))
5174 *perr |= Efail; // UNTESTED
5177 t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
5178 if (!type_compat(Tnone, t, 0))
5179 *perr |= Efail; // UNTESTED
5181 if (cs->casepart == NULL) {
5182 propagate_types(cs->condpart, c, perr, Tbool, 0);
5183 propagate_types(cs->looppart, c, perr, Tbool, 0);
5185 /* Condpart must match case values, with bool permitted */
5187 for (cp = cs->casepart;
5188 cp && !t; cp = cp->next)
5189 t = propagate_types(cp->value, c, perr, NULL, 0);
5190 if (!t && cs->condpart)
5191 t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
5192 if (!t && cs->looppart)
5193 t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
5194 // Now we have a type (I hope) push it down
5196 for (cp = cs->casepart; cp; cp = cp->next)
5197 propagate_types(cp->value, c, perr, t, 0);
5198 propagate_types(cs->condpart, c, perr, t, Rboolok);
5199 propagate_types(cs->looppart, c, perr, t, Rboolok);
5202 // (if)then, else, and case parts must return expected type.
5203 if (!cs->looppart && !type)
5204 type = propagate_types(cs->thenpart, c, perr, NULL, rules);
5206 type = propagate_types(cs->elsepart, c, perr, NULL, rules);
5207 for (cp = cs->casepart;
5209 cp = cp->next) // UNTESTED
5210 type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
5213 propagate_types(cs->thenpart, c, perr, type, rules);
5214 propagate_types(cs->elsepart, c, perr, type, rules);
5215 for (cp = cs->casepart; cp ; cp = cp->next)
5216 propagate_types(cp->action, c, perr, type, rules);
5222 ###### interp binode cases
5224 // This just performs one iterration of the loop
5225 rv = interp_exec(c, b->left, &rvtype);
5226 if (rvtype == Tnone ||
5227 (rvtype == Tbool && rv.bool != 0))
5228 // rvtype is Tnone or Tbool, doesn't need to be freed
5229 interp_exec(c, b->right, NULL);
5232 ###### interp exec cases
5233 case Xcond_statement:
5235 struct value v, cnd;
5236 struct type *vtype, *cndtype;
5237 struct casepart *cp;
5238 struct cond_statement *cs = cast(cond_statement, e);
5241 interp_exec(c, cs->forpart, NULL);
5243 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
5244 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
5245 interp_exec(c, cs->thenpart, NULL);
5247 cnd = interp_exec(c, cs->condpart, &cndtype);
5248 if ((cndtype == Tnone ||
5249 (cndtype == Tbool && cnd.bool != 0))) {
5250 // cnd is Tnone or Tbool, doesn't need to be freed
5251 rv = interp_exec(c, cs->thenpart, &rvtype);
5252 // skip else (and cases)
5256 for (cp = cs->casepart; cp; cp = cp->next) {
5257 v = interp_exec(c, cp->value, &vtype);
5258 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
5259 free_value(vtype, &v);
5260 free_value(cndtype, &cnd);
5261 rv = interp_exec(c, cp->action, &rvtype);
5264 free_value(vtype, &v);
5266 free_value(cndtype, &cnd);
5268 rv = interp_exec(c, cs->elsepart, &rvtype);
5275 ### Top level structure
5277 All the language elements so far can be used in various places. Now
5278 it is time to clarify what those places are.
5280 At the top level of a file there will be a number of declarations.
5281 Many of the things that can be declared haven't been described yet,
5282 such as functions, procedures, imports, and probably more.
5283 For now there are two sorts of things that can appear at the top
5284 level. They are predefined constants, `struct` types, and the `main`
5285 function. While the syntax will allow the `main` function to appear
5286 multiple times, that will trigger an error if it is actually attempted.
5288 The various declarations do not return anything. They store the
5289 various declarations in the parse context.
5291 ###### Parser: grammar
5294 Ocean -> OptNL DeclarationList
5296 ## declare terminals
5304 DeclarationList -> Declaration
5305 | DeclarationList Declaration
5307 Declaration -> ERROR Newlines ${
5308 tok_err(c, // UNTESTED
5309 "error: unhandled parse error", &$1);
5315 ## top level grammar
5319 ### The `const` section
5321 As well as being defined in with the code that uses them, constants can
5322 be declared at the top level. These have full-file scope, so they are
5323 always `InScope`, even before(!) they have been declared. The value of
5324 a top level constant can be given as an expression, and this is
5325 evaluated after parsing and before execution.
5327 A function call can be used to evaluate a constant, but it will not have
5328 access to any program state, once such statement becomes meaningful.
5329 e.g. arguments and filesystem will not be visible.
5331 Constants are defined in a section that starts with the reserved word
5332 `const` and then has a block with a list of assignment statements.
5333 For syntactic consistency, these must use the double-colon syntax to
5334 make it clear that they are constants. Type can also be given: if
5335 not, the type will be determined during analysis, as with other
5338 ###### parse context
5339 struct binode *constlist;
5341 ###### top level grammar
5345 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
5346 | const { SimpleConstList } Newlines
5347 | const IN OptNL ConstList OUT Newlines
5348 | const SimpleConstList Newlines
5350 ConstList -> ConstList SimpleConstLine
5353 SimpleConstList -> SimpleConstList ; Const
5357 SimpleConstLine -> SimpleConstList Newlines
5358 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
5361 CType -> Type ${ $0 = $<1; }$
5365 Const -> IDENTIFIER :: CType = Expression ${ {
5367 struct binode *bl, *bv;
5368 struct var *var = new_pos(var, $ID);
5370 v = var_decl(c, $ID.txt);
5372 v->where_decl = var;
5378 v = var_ref(c, $1.txt);
5379 if (v->type == Tnone) {
5380 v->where_decl = var;
5386 tok_err(c, "error: name already declared", &$1);
5387 type_err(c, "info: this is where '%v' was first declared",
5388 v->where_decl, NULL, 0, NULL);
5400 bl->left = c->constlist;
5405 ###### core functions
5406 static void resolve_consts(struct parse_context *c)
5410 enum { none, some, cannot } progress = none;
5412 c->constlist = reorder_bilist(c->constlist);
5415 for (b = cast(binode, c->constlist); b;
5416 b = cast(binode, b->right)) {
5418 struct binode *vb = cast(binode, b->left);
5419 struct var *v = cast(var, vb->left);
5420 if (v->var->frame_pos >= 0)
5424 propagate_types(vb->right, c, &perr,
5426 } while (perr & Eretry);
5428 c->parse_error += 1;
5429 else if (!(perr & Enoconst)) {
5431 struct value res = interp_exec(
5432 c, vb->right, &v->var->type);
5433 global_alloc(c, v->var->type, v->var, &res);
5435 if (progress == cannot)
5436 type_err(c, "error: const %v cannot be resolved.",
5446 progress = cannot; break;
5448 progress = none; break;
5453 ###### print const decls
5458 for (b = cast(binode, context.constlist); b;
5459 b = cast(binode, b->right)) {
5460 struct binode *vb = cast(binode, b->left);
5461 struct var *vr = cast(var, vb->left);
5462 struct variable *v = vr->var;
5468 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
5469 type_print(v->type, stdout);
5471 print_exec(vb->right, -1, 0);
5476 ###### free const decls
5477 free_binode(context.constlist);
5479 ### Function declarations
5481 The code in an Ocean program is all stored in function declarations.
5482 One of the functions must be named `main` and it must accept an array of
5483 strings as a parameter - the command line arguments.
5485 As this is the top level, several things are handled a bit differently.
5486 The function is not interpreted by `interp_exec` as that isn't passed
5487 the argument list which the program requires. Similarly type analysis
5488 is a bit more interesting at this level.
5490 ###### ast functions
5492 static struct type *handle_results(struct parse_context *c,
5493 struct binode *results)
5495 /* Create a 'struct' type from the results list, which
5496 * is a list for 'struct var'
5498 struct type *t = add_anon_type(c, &structure_prototype,
5503 for (b = results; b; b = cast(binode, b->right))
5505 t->structure.nfields = cnt;
5506 t->structure.fields = calloc(cnt, sizeof(struct field));
5508 for (b = results; b; b = cast(binode, b->right)) {
5509 struct var *v = cast(var, b->left);
5510 struct field *f = &t->structure.fields[cnt++];
5511 int a = v->var->type->align;
5512 f->name = v->var->name->name;
5513 f->type = v->var->type;
5515 f->offset = t->size;
5516 v->var->frame_pos = f->offset;
5517 t->size += ((f->type->size - 1) | (a-1)) + 1;
5520 variable_unlink_exec(v->var);
5522 free_binode(results);
5526 static struct variable *declare_function(struct parse_context *c,
5527 struct variable *name,
5528 struct binode *args,
5530 struct binode *results,
5534 struct value fn = {.function = code};
5536 var_block_close(c, CloseFunction, code);
5537 t = add_anon_type(c, &function_prototype,
5538 "func %.*s", name->name->name.len,
5539 name->name->name.txt);
5541 t->function.params = reorder_bilist(args);
5543 ret = handle_results(c, reorder_bilist(results));
5544 t->function.inline_result = 1;
5545 t->function.local_size = ret->size;
5547 t->function.return_type = ret;
5548 global_alloc(c, t, name, &fn);
5549 name->type->function.scope = c->out_scope;
5554 var_block_close(c, CloseFunction, NULL);
5556 c->out_scope = NULL;
5560 ###### declare terminals
5563 ###### top level grammar
5566 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
5567 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5569 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
5570 $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
5572 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
5573 $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
5575 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
5576 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5578 | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
5579 $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
5581 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
5582 $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
5584 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
5585 $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
5587 | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
5588 $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
5590 | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
5591 $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
5594 ###### print func decls
5599 while (target != 0) {
5601 for (v = context.in_scope; v; v=v->in_scope)
5602 if (v->depth == 0 && v->type && v->type->check_args) {
5611 struct value *val = var_value(&context, v);
5612 printf("func %.*s", v->name->name.len, v->name->name.txt);
5613 v->type->print_type_decl(v->type, stdout);
5615 print_exec(val->function, 0, brackets);
5617 print_value(v->type, val, stdout);
5618 printf("/* frame size %d */\n", v->type->function.local_size);
5624 ###### core functions
5626 static int analyse_funcs(struct parse_context *c)
5630 for (v = c->in_scope; v; v = v->in_scope) {
5634 if (v->depth != 0 || !v->type || !v->type->check_args)
5636 ret = v->type->function.inline_result ?
5637 Tnone : v->type->function.return_type;
5638 val = var_value(c, v);
5641 propagate_types(val->function, c, &perr, ret, 0);
5642 } while (!(perr & Efail) && (perr & Eretry));
5643 if (!(perr & Efail))
5644 /* Make sure everything is still consistent */
5645 propagate_types(val->function, c, &perr, ret, 0);
5648 if (!v->type->function.inline_result &&
5649 !v->type->function.return_type->dup) {
5650 type_err(c, "error: function cannot return value of type %1",
5651 v->where_decl, v->type->function.return_type, 0, NULL);
5654 scope_finalize(c, v->type);
5659 static int analyse_main(struct type *type, struct parse_context *c)
5661 struct binode *bp = type->function.params;
5665 struct type *argv_type;
5667 argv_type = add_anon_type(c, &array_prototype, "argv");
5668 argv_type->array.member = Tstr;
5669 argv_type->array.unspec = 1;
5671 for (b = bp; b; b = cast(binode, b->right)) {
5675 propagate_types(b->left, c, &perr, argv_type, 0);
5677 default: /* invalid */ // NOTEST
5678 propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
5681 c->parse_error += 1;
5684 return !c->parse_error;
5687 static void interp_main(struct parse_context *c, int argc, char **argv)
5689 struct value *progp = NULL;
5690 struct text main_name = { "main", 4 };
5691 struct variable *mainv;
5697 mainv = var_ref(c, main_name);
5699 progp = var_value(c, mainv);
5700 if (!progp || !progp->function) {
5701 fprintf(stderr, "oceani: no main function found.\n");
5702 c->parse_error += 1;
5705 if (!analyse_main(mainv->type, c)) {
5706 fprintf(stderr, "oceani: main has wrong type.\n");
5707 c->parse_error += 1;
5710 al = mainv->type->function.params;
5712 c->local_size = mainv->type->function.local_size;
5713 c->local = calloc(1, c->local_size);
5715 struct var *v = cast(var, al->left);
5716 struct value *vl = var_value(c, v->var);
5726 mpq_set_ui(argcq, argc, 1);
5727 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5728 t->prepare_type(c, t, 0);
5729 array_init(v->var->type, vl);
5730 for (i = 0; i < argc; i++) {
5731 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5733 arg.str.txt = argv[i];
5734 arg.str.len = strlen(argv[i]);
5735 free_value(Tstr, vl2);
5736 dup_value(Tstr, &arg, vl2);
5740 al = cast(binode, al->right);
5742 v = interp_exec(c, progp->function, &vtype);
5743 free_value(vtype, &v);
5748 ###### ast functions
5749 void free_variable(struct variable *v)
5753 ## And now to test it out.
5755 Having a language requires having a "hello world" program. I'll
5756 provide a little more than that: a program that prints "Hello world"
5757 finds the GCD of two numbers, prints the first few elements of
5758 Fibonacci, performs a binary search for a number, and a few other
5759 things which will likely grow as the languages grows.
5761 ###### File: oceani.mk
5764 @echo "===== DEMO ====="
5765 ./oceani --section "demo: hello" oceani.mdc 55 33
5771 four ::= 2 + 2 ; five ::= 10/2
5772 const pie ::= "I like Pie";
5773 cake ::= "The cake is"
5781 func main(argv:[argc::]string)
5782 print "Hello World, what lovely oceans you have!"
5783 print "Are there", five, "?"
5784 print pi, pie, "but", cake
5786 A := $argv[1]; B := $argv[2]
5788 /* When a variable is defined in both branches of an 'if',
5789 * and used afterwards, the variables are merged.
5795 print "Is", A, "bigger than", B,"? ", bigger
5796 /* If a variable is not used after the 'if', no
5797 * merge happens, so types can be different
5800 double:string = "yes"
5801 print A, "is more than twice", B, "?", double
5804 print "double", B, "is", double
5809 if a > 0 and then b > 0:
5815 print "GCD of", A, "and", B,"is", a
5817 print a, "is not positive, cannot calculate GCD"
5819 print b, "is not positive, cannot calculate GCD"
5824 print "Fibonacci:", f1,f2,
5825 then togo = togo - 1
5833 /* Binary search... */
5838 mid := (lo + hi) / 2
5851 print "Yay, I found", target
5853 print "Closest I found was", lo
5858 // "middle square" PRNG. Not particularly good, but one my
5859 // Dad taught me - the first one I ever heard of.
5860 for i:=1; then i = i + 1; while i < size:
5861 n := list[i-1] * list[i-1]
5862 list[i] = (n / 100) % 10 000
5864 print "Before sort:",
5865 for i:=0; then i = i + 1; while i < size:
5869 for i := 1; then i=i+1; while i < size:
5870 for j:=i-1; then j=j-1; while j >= 0:
5871 if list[j] > list[j+1]:
5875 print " After sort:",
5876 for i:=0; then i = i + 1; while i < size:
5880 if 1 == 2 then print "yes"; else print "no"
5884 bob.alive = (bob.name == "Hello")
5885 print "bob", "is" if bob.alive else "isn't", "alive"