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 Elements that are present purely to make a usable language, and
45 without any expectation that they will remain, are the "program'
46 clause, which provides a list of variables to received command-line
47 arguments, and the "print" statement which performs simple output.
49 The current scalar types are "number", "Boolean", and "string".
50 Boolean will likely stay in its current form, the other two might, but
51 could just as easily be changed.
55 Versions of the interpreter which obviously do not support a complete
56 language will be named after creeks and streams. This one is Jamison
59 Once we have something reasonably resembling a complete language, the
60 names of rivers will be used.
61 Early versions of the compiler will be named after seas. Major
62 releases of the compiler will be named after oceans. Hopefully I will
63 be finished once I get to the Pacific Ocean release.
67 As well as parsing and executing a program, the interpreter can print
68 out the program from the parsed internal structure. This is useful
69 for validating the parsing.
70 So the main requirements of the interpreter are:
72 - Parse the program, possibly with tracing,
73 - Analyse the parsed program to ensure consistency,
75 - Execute the program, if no parsing or consistency errors were found.
77 This is all performed by a single C program extracted with
80 There will be two formats for printing the program: a default and one
81 that uses bracketing. So a `--bracket` command line option is needed
82 for that. Normally the first code section found is used, however an
83 alternate section can be requested so that a file (such as this one)
84 can contain multiple programs This is effected with the `--section`
87 This code must be compiled with `-fplan9-extensions` so that anonymous
88 structures can be used.
90 ###### File: oceani.mk
92 myCFLAGS := -Wall -g -fplan9-extensions
93 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
94 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
95 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
97 all :: $(LDLIBS) oceani
98 oceani.c oceani.h : oceani.mdc parsergen
99 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
100 oceani.mk: oceani.mdc md2c
103 oceani: oceani.o $(LDLIBS)
104 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
106 ###### Parser: header
109 struct parse_context {
110 struct token_config config;
119 #define container_of(ptr, type, member) ({ \
120 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
121 (type *)( (char *)__mptr - offsetof(type,member) );})
123 #define config2context(_conf) container_of(_conf, struct parse_context, \
126 ###### Parser: reduce
127 struct parse_context *c = config2context(config);
135 #include <sys/mman.h>
154 static char Usage[] = "Usage: oceani --trace --print --noexec --brackets"
155 "--section=SectionName prog.ocn\n";
156 static const struct option long_options[] = {
157 {"trace", 0, NULL, 't'},
158 {"print", 0, NULL, 'p'},
159 {"noexec", 0, NULL, 'n'},
160 {"brackets", 0, NULL, 'b'},
161 {"section", 1, NULL, 's'},
164 const char *options = "tpnbs";
165 int main(int argc, char *argv[])
170 struct section *s, *ss;
171 char *section = NULL;
172 struct parse_context context = {
174 .ignored = (1 << TK_line_comment)
175 | (1 << TK_block_comment)
177 .number_chars = ".,_+- ",
182 int doprint=0, dotrace=0, doexec=1, brackets=0;
184 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
187 case 't': dotrace=1; break;
188 case 'p': doprint=1; break;
189 case 'n': doexec=0; break;
190 case 'b': brackets=1; break;
191 case 's': section = optarg; break;
192 default: fprintf(stderr, Usage);
196 if (optind >= argc) {
197 fprintf(stderr, "oceani: no input file given\n");
200 fd = open(argv[optind], O_RDONLY);
202 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
205 context.file_name = argv[optind];
206 len = lseek(fd, 0, 2);
207 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
208 s = code_extract(file, file+len, NULL);
210 fprintf(stderr, "oceani: could not find any code in %s\n",
215 ## context initialization
218 for (ss = s; ss; ss = ss->next) {
219 struct text sec = ss->section;
220 if (sec.len == strlen(section) &&
221 strncmp(sec.txt, section, sec.len) == 0)
225 fprintf(stderr, "oceani: cannot find section %s\n",
231 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
234 fprintf(stderr, "oceani: no program found.\n");
235 context.parse_error = 1;
237 if (context.prog && doprint) {
240 print_exec(context.prog, 0, brackets);
242 if (context.prog && doexec && !context.parse_error) {
243 if (!analyse_prog(context.prog, &context)) {
244 fprintf(stderr, "oceani: type error in program - not running.\n");
247 interp_prog(context.prog, argv+optind+1);
249 free_exec(context.prog);
252 struct section *t = s->next;
258 ## free context types
259 exit(context.parse_error ? 1 : 0);
264 The four requirements of parse, analyse, print, interpret apply to
265 each language element individually so that is how most of the code
268 Three of the four are fairly self explanatory. The one that requires
269 a little explanation is the analysis step.
271 The current language design does not require the types of variables to
272 be declared, but they must still have a single type. Different
273 operations impose different requirements on the variables, for example
274 addition requires both arguments to be numeric, and assignment
275 requires the variable on the left to have the same type as the
276 expression on the right.
278 Analysis involves propagating these type requirements around and
279 consequently setting the type of each variable. If any requirements
280 are violated (e.g. a string is compared with a number) or if a
281 variable needs to have two different types, then an error is raised
282 and the program will not run.
284 If the same variable is declared in both branchs of an 'if/else', or
285 in all cases of a 'switch' then the multiple instances may be merged
286 into just one variable if the variable is references after the
287 conditional statement. When this happens, the types must naturally be
288 consistent across all the branches. When the variable is not used
289 outside the if, the variables in the different branches are distinct
290 and can be of different types.
292 Determining the types of all variables early is important for
293 processing command line arguments. These can be assigned to any of
294 several types of variable, but we must first know the correct type so
295 any required conversion can happen. If a variable is associated with
296 a command line argument but no type can be interpreted (e.g. the
297 variable is only ever used in a `print` statement), then the type is
300 Undeclared names may only appear in "use" statements and "case" expressions.
301 These names are given a type of "label" and a unique value.
302 This allows them to fill the role of a name in an enumerated type, which
303 is useful for testing the `switch` statement.
305 As we will see, the condition part of a `while` statement can return
306 either a Boolean or some other type. This requires that the expected
307 type that gets passed around comprises a type and a flag to indicate
308 that `Tbool` is also permitted.
310 As there are, as yet, no distinct types that are compatible, there
311 isn't much subtlety in the analysis. When we have distinct number
312 types, this will become more interesting.
316 When analysis discovers an inconsistency it needs to report an error;
317 just refusing to run the code ensures that the error doesn't cascade,
318 but by itself it isn't very useful. A clear understanding of the sort
319 of error message that are useful will help guide the process of
322 At a simplistic level, the only sort of error that type analysis can
323 report is that the type of some construct doesn't match a contextual
324 requirement. For example, in `4 + "hello"` the addition provides a
325 contextual requirement for numbers, but `"hello"` is not a number. In
326 this particular example no further information is needed as the types
327 are obvious from local information. When a variable is involved that
328 isn't the case. It may be helpful to explain why the variable has a
329 particular type, by indicating the location where the type was set,
330 whether by declaration or usage.
332 Using a recursive-descent analysis we can easily detect a problem at
333 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
334 will detect that one argument is not a number and the usage of `hello`
335 will detect that a number was wanted, but not provided. In this
336 (early) version of the language, we will generate error reports at
337 multiple locations, so the use of `hello` will report an error and
338 explain were the value was set, and the addition will report an error
339 and say why numbers are needed. To be able to report locations for
340 errors, each language element will need to record a file location
341 (line and column) and each variable will need to record the language
342 element where its type was set. For now we will assume that each line
343 of an error message indicates one location in the file, and up to 2
344 types. So we provide a `printf`-like function which takes a format, a
345 language (a `struct exec` which has not yet been introduced), and 2
346 types. "`%1`" reports the first type, "`%2`" reports the second. We
347 will need a function to print the location, once we know how that is
348 stored. As will be explained later, there are sometimes extra rules for
349 type matching and they might affect error messages, we need to pass those
352 As well as type errors, we sometimes need to report problems with
353 tokens, which might be unexpected or might name a type that has not
354 been defined. For these we have `tok_err()` which reports an error
355 with a given token. Each of the error functions sets the flag in the
356 context so indicate that parsing failed.
360 static void fput_loc(struct exec *loc, FILE *f);
362 ###### core functions
364 static void type_err(struct parse_context *c,
365 char *fmt, struct exec *loc,
366 struct type *t1, int rules, struct type *t2)
368 fprintf(stderr, "%s:", c->file_name);
369 fput_loc(loc, stderr);
370 for (; *fmt ; fmt++) {
377 case '%': fputc(*fmt, stderr); break; // NOTEST
378 default: fputc('?', stderr); break; // NOTEST
380 type_print(t1, stderr);
383 type_print(t2, stderr);
392 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
394 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
395 t->txt.len, t->txt.txt);
399 ## Entities: declared and predeclared.
401 There are various "things" that the language and/or the interpreter
402 needs to know about to parse and execute a program. These include
403 types, variables, values, and executable code. These are all lumped
404 together under the term "entities" (calling them "objects" would be
405 confusing) and introduced here. These will introduced and described
406 here. The following section will present the different specific code
407 elements which comprise or manipulate these various entities.
411 Values come in a wide range of types, with more likely to be added.
412 Each type needs to be able to parse and print its own values (for
413 convenience at least) as well as to compare two values, at least for
414 equality and possibly for order. For now, values might need to be
415 duplicated and freed, though eventually such manipulations will be
416 better integrated into the language.
418 Rather than requiring every numeric type to support all numeric
419 operations (add, multiple, etc), we allow types to be able to present
420 as one of a few standard types: integer, float, and fraction. The
421 existence of these conversion functions eventaully enable types to
422 determine if they are compatible with other types, though such types
423 have not yet been implemented.
425 Named type are stored in a simple linked list. Objects of each type are "values"
426 which are often passed around by value.
433 ## value union fields
440 struct value (*init)(struct type *type);
441 struct value (*prepare)(struct type *type);
442 struct value (*parse)(struct type *type, char *str);
443 void (*print)(struct value val);
444 void (*print_type)(struct type *type, FILE *f);
445 int (*cmp_order)(struct value v1, struct value v2);
446 int (*cmp_eq)(struct value v1, struct value v2);
447 struct value (*dup)(struct value val);
448 void (*free)(struct value val);
449 void (*free_type)(struct type *t);
450 int (*compat)(struct type *this, struct type *other);
451 long long (*to_int)(struct value *v);
452 double (*to_float)(struct value *v);
453 int (*to_mpq)(mpq_t *q, struct value *v);
462 struct type *typelist;
466 static struct type *find_type(struct parse_context *c, struct text s)
468 struct type *l = c->typelist;
471 text_cmp(l->name, s) != 0)
476 static struct type *add_type(struct parse_context *c, struct text s,
481 n = calloc(1, sizeof(*n));
484 n->next = c->typelist;
489 static void free_type(struct type *t)
491 /* The type is always a reference to something in the
492 * context, so we don't need to free anything.
496 static void free_value(struct value v)
502 static int type_compat(struct type *require, struct type *have, int rules)
504 if ((rules & Rboolok) && have == Tbool)
506 if ((rules & Rnolabel) && have == Tlabel)
508 if (!require || !have)
512 return require->compat(require, have);
514 return require == have;
517 static void type_print(struct type *type, FILE *f)
520 fputs("*unknown*type*", f);
521 else if (type->name.len)
522 fprintf(f, "%.*s", type->name.len, type->name.txt);
523 else if (type->print_type)
524 type->print_type(type, f);
526 fputs("*invalid*type*", f); // NOTEST
529 static struct value val_prepare(struct type *type)
534 return type->prepare(type);
539 static struct value val_init(struct type *type)
544 return type->init(type);
549 static struct value dup_value(struct value v)
552 return v.type->dup(v);
556 static int value_cmp(struct value left, struct value right)
558 if (left.type && left.type->cmp_order)
559 return left.type->cmp_order(left, right);
560 if (left.type && left.type->cmp_eq)
561 return left.type->cmp_eq(left, right);
565 static void print_value(struct value v)
567 if (v.type && v.type->print)
570 printf("*Unknown*"); // NOTEST
573 static struct value parse_value(struct type *type, char *arg)
577 if (type && type->parse)
578 return type->parse(type, arg);
579 rv.type = NULL; // NOTEST
585 static void free_value(struct value v);
586 static int type_compat(struct type *require, struct type *have, int rules);
587 static void type_print(struct type *type, FILE *f);
588 static struct value val_init(struct type *type);
589 static struct value dup_value(struct value v);
590 static int value_cmp(struct value left, struct value right);
591 static void print_value(struct value v);
592 static struct value parse_value(struct type *type, char *arg);
594 ###### free context types
596 while (context.typelist) {
597 struct type *t = context.typelist;
599 context.typelist = t->next;
607 Values of the base types can be numbers, which we represent as
608 multi-precision fractions, strings, Booleans and labels. When
609 analysing the program we also need to allow for places where no value
610 is meaningful (type `Tnone`) and where we don't know what type to
611 expect yet (type is `NULL`).
613 Values are never shared, they are always copied when used, and freed
614 when no longer needed.
616 When propagating type information around the program, we need to
617 determine if two types are compatible, where type `NULL` is compatible
618 with anything. There are two special cases with type compatibility,
619 both related to the Conditional Statement which will be described
620 later. In some cases a Boolean can be accepted as well as some other
621 primary type, and in others any type is acceptable except a label (`Vlabel`).
622 A separate function encoding these cases will simplify some code later.
624 When assigning command line arguments to variables, we need to be able
625 to parse each type from a string.
627 The distinction beteen "prepare" and "init" needs to be explained.
628 "init" sets up an initial value, such as "zero" or the empty string.
629 "prepare" simply prepares the data structure so that if "free" gets
630 called on it, it won't do something silly. Normally a value will be
631 stored after "prepare" but before "free", but this might not happen if
640 myLDLIBS := libnumber.o libstring.o -lgmp
641 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
643 ###### type union fields
644 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
646 ###### value union fields
653 static void _free_value(struct value v)
655 switch (v.type->vtype) {
657 case Vstr: free(v.str.txt); break;
658 case Vnum: mpq_clear(v.num); break;
664 ###### value functions
666 static struct value _val_prepare(struct type *type)
671 switch(type->vtype) {
675 memset(&rv.num, 0, sizeof(rv.num));
691 static struct value _val_init(struct type *type)
696 switch(type->vtype) {
697 case Vnone: // NOTEST
700 mpq_init(rv.num); break;
702 rv.str.txt = malloc(1);
708 case Vlabel: // NOTEST
709 rv.label = NULL; // NOTEST
715 static struct value _dup_value(struct value v)
719 switch (rv.type->vtype) {
720 case Vnone: // NOTEST
730 mpq_set(rv.num, v.num);
733 rv.str.len = v.str.len;
734 rv.str.txt = malloc(rv.str.len);
735 memcpy(rv.str.txt, v.str.txt, v.str.len);
741 static int _value_cmp(struct value left, struct value right)
744 if (left.type != right.type)
745 return left.type - right.type; // NOTEST
746 switch (left.type->vtype) {
747 case Vlabel: cmp = left.label == right.label ? 0 : 1; break;
748 case Vnum: cmp = mpq_cmp(left.num, right.num); break;
749 case Vstr: cmp = text_cmp(left.str, right.str); break;
750 case Vbool: cmp = left.bool - right.bool; break;
751 case Vnone: cmp = 0; // NOTEST
756 static void _print_value(struct value v)
758 switch (v.type->vtype) {
759 case Vnone: // NOTEST
760 printf("*no-value*"); break; // NOTEST
761 case Vlabel: // NOTEST
762 printf("*label-%p*", v.label); break; // NOTEST
764 printf("%.*s", v.str.len, v.str.txt); break;
766 printf("%s", v.bool ? "True":"False"); break;
771 mpf_set_q(fl, v.num);
772 gmp_printf("%Fg", fl);
779 static struct value _parse_value(struct type *type, char *arg)
787 switch(type->vtype) {
788 case Vlabel: // NOTEST
789 case Vnone: // NOTEST
790 val.type = NULL; // NOTEST
793 val.str.len = strlen(arg);
794 val.str.txt = malloc(val.str.len);
795 memcpy(val.str.txt, arg, val.str.len);
802 tx.txt = arg; tx.len = strlen(tx.txt);
803 if (number_parse(val.num, tail, tx) == 0)
806 mpq_neg(val.num, val.num);
808 printf("Unsupported suffix: %s\n", arg);
813 if (strcasecmp(arg, "true") == 0 ||
814 strcmp(arg, "1") == 0)
816 else if (strcasecmp(arg, "false") == 0 ||
817 strcmp(arg, "0") == 0)
820 printf("Bad bool: %s\n", arg);
828 static void _free_value(struct value v);
830 static struct type base_prototype = {
832 .prepare = _val_prepare,
833 .parse = _parse_value,
834 .print = _print_value,
835 .cmp_order = _value_cmp,
836 .cmp_eq = _value_cmp,
841 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
844 static struct type *add_base_type(struct parse_context *c, char *n, enum vtype vt)
846 struct text txt = { n, strlen(n) };
849 t = add_type(c, txt, &base_prototype);
854 ###### context initialization
856 Tbool = add_base_type(&context, "Boolean", Vbool);
857 Tstr = add_base_type(&context, "string", Vstr);
858 Tnum = add_base_type(&context, "number", Vnum);
859 Tnone = add_base_type(&context, "none", Vnone);
860 Tlabel = add_base_type(&context, "label", Vlabel);
864 Variables are scoped named values. We store the names in a linked
865 list of "bindings" sorted lexically, and use sequential search and
872 struct binding *next; // in lexical order
876 This linked list is stored in the parse context so that "reduce"
877 functions can find or add variables, and so the analysis phase can
878 ensure that every variable gets a type.
882 struct binding *varlist; // In lexical order
886 static struct binding *find_binding(struct parse_context *c, struct text s)
888 struct binding **l = &c->varlist;
893 (cmp = text_cmp((*l)->name, s)) < 0)
897 n = calloc(1, sizeof(*n));
904 Each name can be linked to multiple variables defined in different
905 scopes. Each scope starts where the name is declared and continues
906 until the end of the containing code block. Scopes of a given name
907 cannot nest, so a declaration while a name is in-scope is an error.
909 ###### binding fields
910 struct variable *var;
914 struct variable *previous;
916 struct binding *name;
917 struct exec *where_decl;// where name was declared
918 struct exec *where_set; // where type was set
922 While the naming seems strange, we include local constants in the
923 definition of variables. A name declared `var := value` can
924 subsequently be changed, but a name declared `var ::= value` cannot -
927 ###### variable fields
930 Scopes in parallel branches can be partially merged. More
931 specifically, if a given name is declared in both branches of an
932 if/else then its scope is a candidate for merging. Similarly if
933 every branch of an exhaustive switch (e.g. has an "else" clause)
934 declares a given name, then the scopes from the branches are
935 candidates for merging.
937 Note that names declared inside a loop (which is only parallel to
938 itself) are never visible after the loop. Similarly names defined in
939 scopes which are not parallel, such as those started by `for` and
940 `switch`, are never visible after the scope. Only variables defined in
941 both `then` and `else` (including the implicit then after an `if`, and
942 excluding `then` used with `for`) and in all `case`s and `else` of a
943 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
945 Labels, which are a bit like variables, follow different rules.
946 Labels are not explicitly declared, but if an undeclared name appears
947 in a context where a label is legal, that effectively declares the
948 name as a label. The declaration remains in force (or in scope) at
949 least to the end of the immediately containing block and conditionally
950 in any larger containing block which does not declare the name in some
951 other way. Importantly, the conditional scope extension happens even
952 if the label is only used in one parallel branch of a conditional --
953 when used in one branch it is treated as having been declared in all
956 Merge candidates are tentatively visible beyond the end of the
957 branching statement which creates them. If the name is used, the
958 merge is affirmed and they become a single variable visible at the
959 outer layer. If not - if it is redeclared first - the merge lapses.
961 To track scopes we have an extra stack, implemented as a linked list,
962 which roughly parallels the parse stack and which is used exclusively
963 for scoping. When a new scope is opened, a new frame is pushed and
964 the child-count of the parent frame is incremented. This child-count
965 is used to distinguish between the first of a set of parallel scopes,
966 in which declared variables must not be in scope, and subsequent
967 branches, whether they must already be conditionally scoped.
969 To push a new frame *before* any code in the frame is parsed, we need a
970 grammar reduction. This is most easily achieved with a grammar
971 element which derives the empty string, and creates the new scope when
972 it is recognized. This can be placed, for example, between a keyword
973 like "if" and the code following it.
977 struct scope *parent;
983 struct scope *scope_stack;
986 static void scope_pop(struct parse_context *c)
988 struct scope *s = c->scope_stack;
990 c->scope_stack = s->parent;
995 static void scope_push(struct parse_context *c)
997 struct scope *s = calloc(1, sizeof(*s));
999 c->scope_stack->child_count += 1;
1000 s->parent = c->scope_stack;
1002 c->scope_depth += 1;
1008 OpenScope -> ${ scope_push(c); }$
1010 Each variable records a scope depth and is in one of four states:
1012 - "in scope". This is the case between the declaration of the
1013 variable and the end of the containing block, and also between
1014 the usage with affirms a merge and the end of that block.
1016 The scope depth is not greater than the current parse context scope
1017 nest depth. When the block of that depth closes, the state will
1018 change. To achieve this, all "in scope" variables are linked
1019 together as a stack in nesting order.
1021 - "pending". The "in scope" block has closed, but other parallel
1022 scopes are still being processed. So far, every parallel block at
1023 the same level that has closed has declared the name.
1025 The scope depth is the depth of the last parallel block that
1026 enclosed the declaration, and that has closed.
1028 - "conditionally in scope". The "in scope" block and all parallel
1029 scopes have closed, and no further mention of the name has been
1030 seen. This state includes a secondary nest depth which records the
1031 outermost scope seen since the variable became conditionally in
1032 scope. If a use of the name is found, the variable becomes "in
1033 scope" and that secondary depth becomes the recorded scope depth.
1034 If the name is declared as a new variable, the old variable becomes
1035 "out of scope" and the recorded scope depth stays unchanged.
1037 - "out of scope". The variable is neither in scope nor conditionally
1038 in scope. It is permanently out of scope now and can be removed from
1039 the "in scope" stack.
1041 ###### variable fields
1042 int depth, min_depth;
1043 enum { OutScope, PendingScope, CondScope, InScope } scope;
1044 struct variable *in_scope;
1046 ###### parse context
1048 struct variable *in_scope;
1050 All variables with the same name are linked together using the
1051 'previous' link. Those variable that have
1052 been affirmatively merged all have a 'merged' pointer that points to
1053 one primary variable - the most recently declared instance. When
1054 merging variables, we need to also adjust the 'merged' pointer on any
1055 other variables that had previously been merged with the one that will
1056 no longer be primary.
1058 A variable that is no longer the most recent instance of a name may
1059 still have "pending" scope, if it might still be merged with most
1060 recent instance. These variables don't really belong in the
1061 "in_scope" list, but are not immediately removed when a new instance
1062 is found. Instead, they are detected and ignored when considering the
1063 list of in_scope names.
1065 ###### variable fields
1066 struct variable *merged;
1068 ###### ast functions
1070 static void variable_merge(struct variable *primary, struct variable *secondary)
1074 if (primary->merged)
1076 primary = primary->merged;
1078 for (v = primary->previous; v; v=v->previous)
1079 if (v == secondary || v == secondary->merged ||
1080 v->merged == secondary ||
1081 (v->merged && v->merged == secondary->merged)) {
1082 v->scope = OutScope;
1083 v->merged = primary;
1087 ###### free context vars
1089 while (context.varlist) {
1090 struct binding *b = context.varlist;
1091 struct variable *v = b->var;
1092 context.varlist = b->next;
1095 struct variable *t = v;
1100 // This is a global constant
1101 free_exec(t->where_decl);
1106 #### Manipulating Bindings
1108 When a name is conditionally visible, a new declaration discards the
1109 old binding - the condition lapses. Conversely a usage of the name
1110 affirms the visibility and extends it to the end of the containing
1111 block - i.e. the block that contains both the original declaration and
1112 the latest usage. This is determined from `min_depth`. When a
1113 conditionally visible variable gets affirmed like this, it is also
1114 merged with other conditionally visible variables with the same name.
1116 When we parse a variable declaration we either report an error if the
1117 name is currently bound, or create a new variable at the current nest
1118 depth if the name is unbound or bound to a conditionally scoped or
1119 pending-scope variable. If the previous variable was conditionally
1120 scoped, it and its homonyms becomes out-of-scope.
1122 When we parse a variable reference (including non-declarative
1123 assignment) we report an error if the name is not bound or is bound to
1124 a pending-scope variable; update the scope if the name is bound to a
1125 conditionally scoped variable; or just proceed normally if the named
1126 variable is in scope.
1128 When we exit a scope, any variables bound at this level are either
1129 marked out of scope or pending-scoped, depending on whether the scope
1130 was sequential or parallel. Here a "parallel" scope means the "then"
1131 or "else" part of a conditional, or any "case" or "else" branch of a
1132 switch. Other scopes are "sequential".
1134 When exiting a parallel scope we check if there are any variables that
1135 were previously pending and are still visible. If there are, then
1136 there weren't redeclared in the most recent scope, so they cannot be
1137 merged and must become out-of-scope. If it is not the first of
1138 parallel scopes (based on `child_count`), we check that there was a
1139 previous binding that is still pending-scope. If there isn't, the new
1140 variable must now be out-of-scope.
1142 When exiting a sequential scope that immediately enclosed parallel
1143 scopes, we need to resolve any pending-scope variables. If there was
1144 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1145 we need to mark all pending-scope variable as out-of-scope. Otherwise
1146 all pending-scope variables become conditionally scoped.
1149 enum closetype { CloseSequential, CloseParallel, CloseElse };
1151 ###### ast functions
1153 static struct variable *var_decl(struct parse_context *c, struct text s)
1155 struct binding *b = find_binding(c, s);
1156 struct variable *v = b->var;
1158 switch (v ? v->scope : OutScope) {
1160 /* Caller will report the error */
1164 v && v->scope == CondScope;
1166 v->scope = OutScope;
1170 v = calloc(1, sizeof(*v));
1171 v->previous = b->var;
1174 v->min_depth = v->depth = c->scope_depth;
1176 v->in_scope = c->in_scope;
1178 v->val = val_prepare(NULL);
1182 static struct variable *var_ref(struct parse_context *c, struct text s)
1184 struct binding *b = find_binding(c, s);
1185 struct variable *v = b->var;
1186 struct variable *v2;
1188 switch (v ? v->scope : OutScope) {
1191 /* Caller will report the error */
1194 /* All CondScope variables of this name need to be merged
1195 * and become InScope
1197 v->depth = v->min_depth;
1199 for (v2 = v->previous;
1200 v2 && v2->scope == CondScope;
1202 variable_merge(v, v2);
1210 static void var_block_close(struct parse_context *c, enum closetype ct)
1212 /* Close off all variables that are in_scope */
1213 struct variable *v, **vp, *v2;
1216 for (vp = &c->in_scope;
1217 v = *vp, v && v->depth > c->scope_depth && v->min_depth > c->scope_depth;
1219 if (v->name->var == v) switch (ct) {
1221 case CloseParallel: /* handle PendingScope */
1225 if (c->scope_stack->child_count == 1)
1226 v->scope = PendingScope;
1227 else if (v->previous &&
1228 v->previous->scope == PendingScope)
1229 v->scope = PendingScope;
1230 else if (v->val.type == Tlabel)
1231 v->scope = PendingScope;
1232 else if (v->name->var == v)
1233 v->scope = OutScope;
1234 if (ct == CloseElse) {
1235 /* All Pending variables with this name
1236 * are now Conditional */
1238 v2 && v2->scope == PendingScope;
1240 v2->scope = CondScope;
1245 v2 && v2->scope == PendingScope;
1247 if (v2->val.type != Tlabel)
1248 v2->scope = OutScope;
1250 case OutScope: break;
1253 case CloseSequential:
1254 if (v->val.type == Tlabel)
1255 v->scope = PendingScope;
1258 v->scope = OutScope;
1261 /* There was no 'else', so we can only become
1262 * conditional if we know the cases were exhaustive,
1263 * and that doesn't mean anything yet.
1264 * So only labels become conditional..
1267 v2 && v2->scope == PendingScope;
1269 if (v2->val.type == Tlabel) {
1270 v2->scope = CondScope;
1271 v2->min_depth = c->scope_depth;
1273 v2->scope = OutScope;
1276 case OutScope: break;
1280 if (v->scope == OutScope || v->name->var != v)
1289 Executables can be lots of different things. In many cases an
1290 executable is just an operation combined with one or two other
1291 executables. This allows for expressions and lists etc. Other times
1292 an executable is something quite specific like a constant or variable
1293 name. So we define a `struct exec` to be a general executable with a
1294 type, and a `struct binode` which is a subclass of `exec`, forms a
1295 node in a binary tree, and holds an operation. There will be other
1296 subclasses, and to access these we need to be able to `cast` the
1297 `exec` into the various other types.
1300 #define cast(structname, pointer) ({ \
1301 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1302 if (__mptr && *__mptr != X##structname) abort(); \
1303 (struct structname *)( (char *)__mptr);})
1305 #define new(structname) ({ \
1306 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1307 __ptr->type = X##structname; \
1308 __ptr->line = -1; __ptr->column = -1; \
1311 #define new_pos(structname, token) ({ \
1312 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1313 __ptr->type = X##structname; \
1314 __ptr->line = token.line; __ptr->column = token.col; \
1323 enum exec_types type;
1331 struct exec *left, *right;
1334 ###### ast functions
1336 static int __fput_loc(struct exec *loc, FILE *f)
1340 if (loc->line >= 0) {
1341 fprintf(f, "%d:%d: ", loc->line, loc->column);
1344 if (loc->type == Xbinode)
1345 return __fput_loc(cast(binode,loc)->left, f) ||
1346 __fput_loc(cast(binode,loc)->right, f);
1349 static void fput_loc(struct exec *loc, FILE *f)
1351 if (!__fput_loc(loc, f))
1352 fprintf(f, "??:??: "); // NOTEST
1355 Each different type of `exec` node needs a number of functions
1356 defined, a bit like methods. We must be able to be able to free it,
1357 print it, analyse it and execute it. Once we have specific `exec`
1358 types we will need to parse them too. Let's take this a bit more
1363 The parser generator requires a `free_foo` function for each struct
1364 that stores attributes and they will often be `exec`s and subtypes
1365 there-of. So we need `free_exec` which can handle all the subtypes,
1366 and we need `free_binode`.
1368 ###### ast functions
1370 static void free_binode(struct binode *b)
1375 free_exec(b->right);
1379 ###### core functions
1380 static void free_exec(struct exec *e)
1389 ###### forward decls
1391 static void free_exec(struct exec *e);
1393 ###### free exec cases
1394 case Xbinode: free_binode(cast(binode, e)); break;
1398 Printing an `exec` requires that we know the current indent level for
1399 printing line-oriented components. As will become clear later, we
1400 also want to know what sort of bracketing to use.
1402 ###### ast functions
1404 static void do_indent(int i, char *str)
1411 ###### core functions
1412 static void print_binode(struct binode *b, int indent, int bracket)
1416 ## print binode cases
1420 static void print_exec(struct exec *e, int indent, int bracket)
1426 print_binode(cast(binode, e), indent, bracket); break;
1431 ###### forward decls
1433 static void print_exec(struct exec *e, int indent, int bracket);
1437 As discussed, analysis involves propagating type requirements around
1438 the program and looking for errors.
1440 So `propagate_types` is passed an expected type (being a `struct type`
1441 pointer together with some `val_rules` flags) that the `exec` is
1442 expected to return, and returns the type that it does return, either
1443 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1444 by reference. It is set to `0` when an error is found, and `2` when
1445 any change is made. If it remains unchanged at `1`, then no more
1446 propagation is needed.
1450 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1454 if (rules & Rnolabel)
1455 fputs(" (labels not permitted)", stderr);
1458 ###### core functions
1460 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1461 struct type *type, int rules);
1462 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1463 struct type *type, int rules)
1470 switch (prog->type) {
1473 struct binode *b = cast(binode, prog);
1475 ## propagate binode cases
1479 ## propagate exec cases
1484 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1485 struct type *type, int rules)
1487 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1496 Interpreting an `exec` doesn't require anything but the `exec`. State
1497 is stored in variables and each variable will be directly linked from
1498 within the `exec` tree. The exception to this is the whole `program`
1499 which needs to look at command line arguments. The `program` will be
1500 interpreted separately.
1502 Each `exec` can return a value, which may be `Tnone` but must be
1503 non-NULL; Some `exec`s will return the location of a value, which can
1504 be updates. To support this, each exec case must store either a value
1505 in `val` or the pointer to a value in `lval`. If `lval` is set, but a
1506 simple value is required, `inter_exec()` will dereference `lval` to
1509 ###### core functions
1512 struct value val, *lval;
1515 static struct lrval _interp_exec(struct exec *e);
1517 static struct value interp_exec(struct exec *e)
1519 struct lrval ret = _interp_exec(e);
1522 return dup_value(*ret.lval);
1527 static struct value *linterp_exec(struct exec *e)
1529 struct lrval ret = _interp_exec(e);
1534 static struct lrval _interp_exec(struct exec *e)
1537 struct value rv, *lrv = NULL;
1548 struct binode *b = cast(binode, e);
1549 struct value left, right, *lleft;
1550 left.type = right.type = Tnone;
1552 ## interp binode cases
1554 free_value(left); free_value(right);
1557 ## interp exec cases
1566 Now that we have the shape of the interpreter in place we can add some
1567 complex types and connected them in to the data structures and the
1568 different phases of parse, analyse, print, interpret.
1570 Thus far we have arrays and structs.
1574 Arrays can be declared by giving a size and a type, as `[size]type' so
1575 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1576 size can be an arbitrary expression which is evaluated when the name
1579 Arrays cannot be assigned. When pointers are introduced we will also
1580 introduce array slices which can refer to part or all of an array -
1581 the assignment syntax will create a slice. For now, an array can only
1582 ever be referenced by the name it is declared with. It is likely that
1583 a "`copy`" primitive will eventually be define which can be used to
1584 make a copy of an array with controllable depth.
1586 ###### type union fields
1590 struct variable *vsize;
1591 struct type *member;
1594 ###### value union fields
1596 struct value *elmnts;
1599 ###### value functions
1601 static struct value array_prepare(struct type *type)
1606 ret.array.elmnts = NULL;
1610 static struct value array_init(struct type *type)
1616 if (type->array.vsize) {
1619 mpz_tdiv_q(q, mpq_numref(type->array.vsize->val.num),
1620 mpq_denref(type->array.vsize->val.num));
1621 type->array.size = mpz_get_si(q);
1624 ret.array.elmnts = calloc(type->array.size,
1625 sizeof(ret.array.elmnts[0]));
1626 for (i = 0; ret.array.elmnts && i < type->array.size; i++)
1627 ret.array.elmnts[i] = val_init(type->array.member);
1631 static void array_free(struct value val)
1635 if (val.array.elmnts)
1636 for (i = 0; i < val.type->array.size; i++)
1637 free_value(val.array.elmnts[i]);
1638 free(val.array.elmnts);
1641 static int array_compat(struct type *require, struct type *have)
1643 if (have->compat != require->compat)
1645 /* Both are arrays, so we can look at details */
1646 if (!type_compat(require->array.member, have->array.member, 0))
1648 if (require->array.vsize == NULL && have->array.vsize == NULL)
1649 return require->array.size == have->array.size;
1651 return require->array.vsize == have->array.vsize;
1654 static void array_print_type(struct type *type, FILE *f)
1657 if (type->array.vsize) {
1658 struct binding *b = type->array.vsize->name;
1659 fprintf(f, "%.*s]", b->name.len, b->name.txt);
1661 fprintf(f, "%d]", type->array.size);
1662 type_print(type->array.member, f);
1665 static struct type array_prototype = {
1666 .prepare = array_prepare,
1668 .print_type = array_print_type,
1669 .compat = array_compat,
1675 | [ NUMBER ] Type ${
1676 $0 = calloc(1, sizeof(struct type));
1677 *($0) = array_prototype;
1678 $0->array.member = $<4;
1679 $0->array.vsize = NULL;
1683 if (number_parse(num, tail, $2.txt) == 0)
1684 tok_err(c, "error: unrecognised number", &$2);
1686 tok_err(c, "error: unsupported number suffix", &$2);
1688 $0->array.size = mpz_get_ui(mpq_numref(num));
1689 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1690 tok_err(c, "error: array size must be an integer",
1692 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1693 tok_err(c, "error: array size is too large",
1697 $0->next= c->anon_typelist;
1698 c->anon_typelist = $0;
1702 | [ IDENTIFIER ] Type ${ {
1703 struct variable *v = var_ref(c, $2.txt);
1706 tok_err(c, "error: name undeclared", &$2);
1707 else if (!v->constant)
1708 tok_err(c, "error: array size must be a constant", &$2);
1710 $0 = calloc(1, sizeof(struct type));
1711 *($0) = array_prototype;
1712 $0->array.member = $<4;
1714 $0->array.vsize = v;
1715 $0->next= c->anon_typelist;
1716 c->anon_typelist = $0;
1719 ###### parse context
1721 struct type *anon_typelist;
1723 ###### free context types
1725 while (context.anon_typelist) {
1726 struct type *t = context.anon_typelist;
1728 context.anon_typelist = t->next;
1735 ###### variable grammar
1737 | Variable [ Expression ] ${ {
1738 struct binode *b = new(binode);
1745 ###### print binode cases
1747 print_exec(b->left, -1, bracket);
1749 print_exec(b->right, -1, bracket);
1753 ###### propagate binode cases
1755 /* left must be an array, right must be a number,
1756 * result is the member type of the array
1758 propagate_types(b->right, c, ok, Tnum, 0);
1759 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1760 if (!t || t->compat != array_compat) {
1761 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1764 if (!type_compat(type, t->array.member, rules)) {
1765 type_err(c, "error: have %1 but need %2", prog,
1766 t->array.member, rules, type);
1768 return t->array.member;
1772 ###### interp binode cases
1777 lleft = linterp_exec(b->left);
1778 right = interp_exec(b->right);
1780 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1784 if (i >= 0 && i < lleft->type->array.size)
1785 lrv = &lleft->array.elmnts[i];
1787 rv = val_init(lleft->type->array.member);
1793 A `struct` is a data-type that contains one or more other data-types.
1794 It differs from an array in that each member can be of a different
1795 type, and they are accessed by name rather than by number. Thus you
1796 cannot choose an element by calculation, you need to know what you
1799 The language makes no promises about how a given structure will be
1800 stored in memory - it is free to rearrange fields to suit whatever
1801 criteria seems important.
1803 Structs are declared separately from program code - they cannot be
1804 declared in-line in a variable declaration like arrays can. A struct
1805 is given a name and this name is used to identify the type - the name
1806 is not prefixed by the word `struct` as it would be in C.
1808 Structs are only treated as the same if they have the same name.
1809 Simply having the same fields in the same order is not enough. This
1810 might change once we can create structure initializes from a list of
1813 Each component datum is identified much like a variable is declared,
1814 with a name, one or two colons, and a type. The type cannot be omitted
1815 as there is no opportunity to deduce the type from usage. An initial
1816 value can be given following an equals sign, so
1818 ##### Example: a struct type
1824 would declare a type called "complex" which has two number fields,
1825 each initialised to zero.
1827 Struct will need to be declared separately from the code that uses
1828 them, so we will need to be able to print out the declaration of a
1829 struct when reprinting the whole program. So a `print_type_decl` type
1830 function will be needed.
1832 ###### type union fields
1843 ###### value union fields
1845 struct value *fields;
1848 ###### type functions
1849 void (*print_type_decl)(struct type *type, FILE *f);
1851 ###### value functions
1853 static struct value structure_prepare(struct type *type)
1858 ret.structure.fields = NULL;
1862 static struct value structure_init(struct type *type)
1868 ret.structure.fields = calloc(type->structure.nfields,
1869 sizeof(ret.structure.fields[0]));
1870 for (i = 0; ret.structure.fields && i < type->structure.nfields; i++)
1871 ret.structure.fields[i] = val_init(type->structure.fields[i].type);
1875 static void structure_free(struct value val)
1879 if (val.structure.fields)
1880 for (i = 0; i < val.type->structure.nfields; i++)
1881 free_value(val.structure.fields[i]);
1882 free(val.structure.fields);
1885 static void structure_free_type(struct type *t)
1888 for (i = 0; i < t->structure.nfields; i++)
1889 free_value(t->structure.fields[i].init);
1890 free(t->structure.fields);
1893 static struct type structure_prototype = {
1894 .prepare = structure_prepare,
1895 .init = structure_init,
1896 .free = structure_free,
1897 .free_type = structure_free_type,
1898 .print_type_decl = structure_print_type,
1912 ###### free exec cases
1914 free_exec(cast(fieldref, e)->left);
1918 ###### variable grammar
1920 | Variable . IDENTIFIER ${ {
1921 struct fieldref *fr = new_pos(fieldref, $2);
1928 ###### print exec cases
1932 struct fieldref *f = cast(fieldref, e);
1933 print_exec(f->left, -1, bracket);
1934 printf(".%.*s", f->name.len, f->name.txt);
1938 ###### ast functions
1939 static int find_struct_index(struct type *type, struct text field)
1942 for (i = 0; i < type->structure.nfields; i++)
1943 if (text_cmp(type->structure.fields[i].name, field) == 0)
1948 ###### propagate exec cases
1952 struct fieldref *f = cast(fieldref, prog);
1953 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
1956 type_err(c, "error: unknown type for field access", f->left,
1958 else if (st->prepare != structure_prepare)
1959 type_err(c, "error: field reference attempted on %1, not a struct",
1960 f->left, st, 0, NULL);
1961 else if (f->index == -2) {
1962 f->index = find_struct_index(st, f->name);
1964 type_err(c, "error: cannot find requested field in %1",
1965 f->left, st, 0, NULL);
1967 if (f->index >= 0) {
1968 struct type *ft = st->structure.fields[f->index].type;
1969 if (!type_compat(type, ft, rules))
1970 type_err(c, "error: have %1 but need %2", prog,
1977 ###### interp exec cases
1980 struct fieldref *f = cast(fieldref, e);
1981 struct value *lleft = linterp_exec(f->left);
1982 lrv = &lleft->structure.fields[f->index];
1988 struct fieldlist *prev;
1992 ###### ast functions
1993 static void free_fieldlist(struct fieldlist *f)
1997 free_fieldlist(f->prev);
1998 free_value(f->f.init);
2002 ###### top level grammar
2003 DeclareStruct -> struct IDENTIFIER FieldBlock ${ {
2005 add_type(c, $2.txt, &structure_prototype);
2007 struct fieldlist *f;
2009 for (f = $3; f; f=f->prev)
2012 t->structure.nfields = cnt;
2013 t->structure.fields = calloc(cnt, sizeof(struct field));
2017 t->structure.fields[cnt] = f->f;
2018 f->f.init = val_prepare(Tnone);
2029 FieldBlock -> Open FieldList Close ${ $0 = $<2; }$
2030 | Open SimpleFieldList } ${ $0 = $<2; }$
2031 | : FieldList ${ $0 = $<2; }$
2033 FieldList -> SimpleFieldList NEWLINE ${ $0 = $<1; }$
2034 | FieldList SimpleFieldList NEWLINE ${
2039 SimpleFieldList -> Field ${ $0 = $<1; }$
2040 | SimpleFieldList ; Field ${
2044 | SimpleFieldList ; ${
2048 Field -> IDENTIFIER : Type = Expression ${ {
2051 $0 = calloc(1, sizeof(struct fieldlist));
2052 $0->f.name = $1.txt;
2054 $0->f.init = val_prepare($0->f.type);
2057 propagate_types($<5, c, &ok, $3, 0);
2062 $0->f.init = interp_exec($5);
2064 | IDENTIFIER : Type ${
2065 $0 = calloc(1, sizeof(struct fieldlist));
2066 $0->f.name = $1.txt;
2068 $0->f.init = val_init($3);
2070 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2072 ###### forward decls
2073 static void structure_print_type(struct type *t, FILE *f);
2075 ###### value functions
2076 static void structure_print_type(struct type *t, FILE *f)
2080 fprintf(f, "struct %.*s:\n", t->name.len, t->name.txt);
2082 for (i = 0; i < t->structure.nfields; i++) {
2083 struct field *fl = t->structure.fields + i;
2084 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2085 type_print(fl->type, f);
2086 if (fl->init.type->print) {
2088 if (fl->init.type == Tstr)
2090 print_value(fl->init);
2091 if (fl->init.type == Tstr)
2098 ###### print type decls
2103 while (target != 0) {
2105 for (t = context.typelist; t ; t=t->next)
2106 if (t->print_type_decl) {
2115 t->print_type_decl(t, stdout);
2121 ## Executables: the elements of code
2123 Each code element needs to be parsed, printed, analysed,
2124 interpreted, and freed. There are several, so let's just start with
2125 the easy ones and work our way up.
2129 We have already met values as separate objects. When manifest
2130 constants appear in the program text, that must result in an executable
2131 which has a constant value. So the `val` structure embeds a value in
2147 $0 = new_pos(val, $1);
2148 $0->val.type = Tbool;
2152 $0 = new_pos(val, $1);
2153 $0->val.type = Tbool;
2157 $0 = new_pos(val, $1);
2158 $0->val.type = Tnum;
2161 if (number_parse($0->val.num, tail, $1.txt) == 0)
2162 mpq_init($0->val.num);
2164 tok_err(c, "error: unsupported number suffix",
2169 $0 = new_pos(val, $1);
2170 $0->val.type = Tstr;
2173 string_parse(&$1, '\\', &$0->val.str, tail);
2175 tok_err(c, "error: unsupported string suffix",
2180 $0 = new_pos(val, $1);
2181 $0->val.type = Tstr;
2184 string_parse(&$1, '\\', &$0->val.str, tail);
2186 tok_err(c, "error: unsupported string suffix",
2191 ###### print exec cases
2194 struct val *v = cast(val, e);
2195 if (v->val.type == Tstr)
2197 print_value(v->val);
2198 if (v->val.type == Tstr)
2203 ###### propagate exec cases
2206 struct val *val = cast(val, prog);
2207 if (!type_compat(type, val->val.type, rules))
2208 type_err(c, "error: expected %1%r found %2",
2209 prog, type, rules, val->val.type);
2210 return val->val.type;
2213 ###### interp exec cases
2215 rv = dup_value(cast(val, e)->val);
2218 ###### ast functions
2219 static void free_val(struct val *v)
2227 ###### free exec cases
2228 case Xval: free_val(cast(val, e)); break;
2230 ###### ast functions
2231 // Move all nodes from 'b' to 'rv', reversing the order.
2232 // In 'b' 'left' is a list, and 'right' is the last node.
2233 // In 'rv', left' is the first node and 'right' is a list.
2234 static struct binode *reorder_bilist(struct binode *b)
2236 struct binode *rv = NULL;
2239 struct exec *t = b->right;
2243 b = cast(binode, b->left);
2253 Just as we used a `val` to wrap a value into an `exec`, we similarly
2254 need a `var` to wrap a `variable` into an exec. While each `val`
2255 contained a copy of the value, each `var` hold a link to the variable
2256 because it really is the same variable no matter where it appears.
2257 When a variable is used, we need to remember to follow the `->merged`
2258 link to find the primary instance.
2266 struct variable *var;
2272 VariableDecl -> IDENTIFIER : ${ {
2273 struct variable *v = var_decl(c, $1.txt);
2274 $0 = new_pos(var, $1);
2279 v = var_ref(c, $1.txt);
2281 type_err(c, "error: variable '%v' redeclared",
2283 type_err(c, "info: this is where '%v' was first declared",
2284 v->where_decl, NULL, 0, NULL);
2287 | IDENTIFIER :: ${ {
2288 struct variable *v = var_decl(c, $1.txt);
2289 $0 = new_pos(var, $1);
2295 v = var_ref(c, $1.txt);
2297 type_err(c, "error: variable '%v' redeclared",
2299 type_err(c, "info: this is where '%v' was first declared",
2300 v->where_decl, NULL, 0, NULL);
2303 | IDENTIFIER : Type ${ {
2304 struct variable *v = var_decl(c, $1.txt);
2305 $0 = new_pos(var, $1);
2310 v->val = val_prepare($<3);
2312 v = var_ref(c, $1.txt);
2314 type_err(c, "error: variable '%v' redeclared",
2316 type_err(c, "info: this is where '%v' was first declared",
2317 v->where_decl, NULL, 0, NULL);
2320 | IDENTIFIER :: Type ${ {
2321 struct variable *v = var_decl(c, $1.txt);
2322 $0 = new_pos(var, $1);
2327 v->val = val_prepare($<3);
2330 v = var_ref(c, $1.txt);
2332 type_err(c, "error: variable '%v' redeclared",
2334 type_err(c, "info: this is where '%v' was first declared",
2335 v->where_decl, NULL, 0, NULL);
2340 Variable -> IDENTIFIER ${ {
2341 struct variable *v = var_ref(c, $1.txt);
2342 $0 = new_pos(var, $1);
2344 /* This might be a label - allocate a var just in case */
2345 v = var_decl(c, $1.txt);
2347 v->val = val_prepare(Tnone);
2352 cast(var, $0)->var = v;
2357 Type -> IDENTIFIER ${
2358 $0 = find_type(c, $1.txt);
2361 "error: undefined type", &$1);
2368 ###### print exec cases
2371 struct var *v = cast(var, e);
2373 struct binding *b = v->var->name;
2374 printf("%.*s", b->name.len, b->name.txt);
2381 if (loc->type == Xvar) {
2382 struct var *v = cast(var, loc);
2384 struct binding *b = v->var->name;
2385 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2387 fputs("???", stderr); // NOTEST
2389 fputs("NOTVAR", stderr); // NOTEST
2392 ###### propagate exec cases
2396 struct var *var = cast(var, prog);
2397 struct variable *v = var->var;
2399 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2400 return Tnone; // NOTEST
2404 if (v->constant && (rules & Rnoconstant)) {
2405 type_err(c, "error: Cannot assign to a constant: %v",
2406 prog, NULL, 0, NULL);
2407 type_err(c, "info: name was defined as a constant here",
2408 v->where_decl, NULL, 0, NULL);
2411 if (v->val.type == Tnone && v->where_decl == prog)
2412 type_err(c, "error: variable used but not declared: %v",
2413 prog, NULL, 0, NULL);
2414 if (v->val.type == NULL) {
2415 if (type && *ok != 0) {
2416 v->val = val_prepare(type);
2417 v->where_set = prog;
2422 if (!type_compat(type, v->val.type, rules)) {
2423 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2424 type, rules, v->val.type);
2425 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2426 v->val.type, rules, NULL);
2433 ###### interp exec cases
2436 struct var *var = cast(var, e);
2437 struct variable *v = var->var;
2445 ###### ast functions
2447 static void free_var(struct var *v)
2452 ###### free exec cases
2453 case Xvar: free_var(cast(var, e)); break;
2455 ### Expressions: Conditional
2457 Our first user of the `binode` will be conditional expressions, which
2458 is a bit odd as they actually have three components. That will be
2459 handled by having 2 binodes for each expression. The conditional
2460 expression is the lowest precedence operatior, so it gets to define
2461 what an "Expression" is. The next level up is "BoolExpr", which
2464 Conditional expressions are of the form "value `if` condition `else`
2465 other_value". They associate to the right, so everything to the right
2466 of `else` is part of an else value, while only the BoolExpr to the
2467 left of `if` is the if values. Between `if` and `else` there is no
2468 room for ambiguity, so a full conditional expression is allowed in there.
2479 Expression -> Expression if Expression else Expression $$ifelse ${ {
2480 struct binode *b1 = new(binode);
2481 struct binode *b2 = new(binode);
2490 ## expression grammar
2492 ###### print binode cases
2495 b2 = cast(binode, b->right);
2496 if (bracket) printf("(");
2497 print_exec(b2->left, -1, bracket);
2499 print_exec(b->left, -1, bracket);
2501 print_exec(b2->right, -1, bracket);
2502 if (bracket) printf(")");
2505 ###### propagate binode cases
2508 /* cond must be Tbool, others must match */
2509 struct binode *b2 = cast(binode, b->right);
2512 propagate_types(b->left, c, ok, Tbool, 0);
2513 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2514 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2518 ###### interp binode cases
2521 struct binode *b2 = cast(binode, b->right);
2522 left = interp_exec(b->left);
2524 rv = interp_exec(b2->left);
2526 rv = interp_exec(b2->right);
2530 ### Expressions: Boolean
2532 The next class of expressions to use the `binode` will be Boolean
2533 expressions. As I haven't implemented precedence in the parser
2534 generator yet, we need different names for each precedence level used
2535 by expressions. The outer most or lowest level precedence after
2536 conditional expressions are Boolean operators which form an `BoolExpr`
2537 out of `BTerm`s and `BFact`s. As well as `or` `and`, and `not` we
2538 have `and then` and `or else` which only evaluate the second operand
2539 if the result would make a difference.
2548 ###### expr precedence
2553 ###### expression grammar
2554 | Expression or Expression ${ {
2555 struct binode *b = new(binode);
2561 | Expression or else Expression ${ {
2562 struct binode *b = new(binode);
2569 | Expression and Expression ${ {
2570 struct binode *b = new(binode);
2576 | Expression and then Expression ${ {
2577 struct binode *b = new(binode);
2584 | not Expression ${ {
2585 struct binode *b = new(binode);
2591 ###### print binode cases
2593 if (bracket) printf("(");
2594 print_exec(b->left, -1, bracket);
2596 print_exec(b->right, -1, bracket);
2597 if (bracket) printf(")");
2600 if (bracket) printf("(");
2601 print_exec(b->left, -1, bracket);
2602 printf(" and then ");
2603 print_exec(b->right, -1, bracket);
2604 if (bracket) printf(")");
2607 if (bracket) printf("(");
2608 print_exec(b->left, -1, bracket);
2610 print_exec(b->right, -1, bracket);
2611 if (bracket) printf(")");
2614 if (bracket) printf("(");
2615 print_exec(b->left, -1, bracket);
2616 printf(" or else ");
2617 print_exec(b->right, -1, bracket);
2618 if (bracket) printf(")");
2621 if (bracket) printf("(");
2623 print_exec(b->right, -1, bracket);
2624 if (bracket) printf(")");
2627 ###### propagate binode cases
2633 /* both must be Tbool, result is Tbool */
2634 propagate_types(b->left, c, ok, Tbool, 0);
2635 propagate_types(b->right, c, ok, Tbool, 0);
2636 if (type && type != Tbool)
2637 type_err(c, "error: %1 operation found where %2 expected", prog,
2641 ###### interp binode cases
2643 rv = interp_exec(b->left);
2644 right = interp_exec(b->right);
2645 rv.bool = rv.bool && right.bool;
2648 rv = interp_exec(b->left);
2650 rv = interp_exec(b->right);
2653 rv = interp_exec(b->left);
2654 right = interp_exec(b->right);
2655 rv.bool = rv.bool || right.bool;
2658 rv = interp_exec(b->left);
2660 rv = interp_exec(b->right);
2663 rv = interp_exec(b->right);
2667 ### Expressions: Comparison
2669 Of slightly higher precedence that Boolean expressions are
2671 A comparison takes arguments of any comparable type, but the two types must be
2674 To simplify the parsing we introduce an `eop` which can record an
2675 expression operator.
2682 ###### ast functions
2683 static void free_eop(struct eop *e)
2697 ###### expr precedence
2700 ###### expression grammar
2701 | Expression CMPop Expression ${ {
2702 struct binode *b = new(binode);
2712 CMPop -> < ${ $0.op = Less; }$
2713 | > ${ $0.op = Gtr; }$
2714 | <= ${ $0.op = LessEq; }$
2715 | >= ${ $0.op = GtrEq; }$
2716 | == ${ $0.op = Eql; }$
2717 | != ${ $0.op = NEql; }$
2719 ###### print binode cases
2727 if (bracket) printf("(");
2728 print_exec(b->left, -1, bracket);
2730 case Less: printf(" < "); break;
2731 case LessEq: printf(" <= "); break;
2732 case Gtr: printf(" > "); break;
2733 case GtrEq: printf(" >= "); break;
2734 case Eql: printf(" == "); break;
2735 case NEql: printf(" != "); break;
2736 default: abort(); // NOTEST
2738 print_exec(b->right, -1, bracket);
2739 if (bracket) printf(")");
2742 ###### propagate binode cases
2749 /* Both must match but not be labels, result is Tbool */
2750 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2752 propagate_types(b->right, c, ok, t, 0);
2754 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2756 t = propagate_types(b->left, c, ok, t, 0);
2758 if (!type_compat(type, Tbool, 0))
2759 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2760 Tbool, rules, type);
2763 ###### interp binode cases
2772 left = interp_exec(b->left);
2773 right = interp_exec(b->right);
2774 cmp = value_cmp(left, right);
2777 case Less: rv.bool = cmp < 0; break;
2778 case LessEq: rv.bool = cmp <= 0; break;
2779 case Gtr: rv.bool = cmp > 0; break;
2780 case GtrEq: rv.bool = cmp >= 0; break;
2781 case Eql: rv.bool = cmp == 0; break;
2782 case NEql: rv.bool = cmp != 0; break;
2783 default: rv.bool = 0; break; // NOTEST
2788 ### Expressions: The rest
2790 The remaining expressions with the highest precedence are arithmetic
2791 and string concatenation. They are `Expr`, `Term`, and `Factor`.
2792 The `Factor` is where the `Value` and `Variable` that we already have
2795 `+` and `-` are both infix and prefix operations (where they are
2796 absolute value and negation). These have different operator names.
2798 We also have a 'Bracket' operator which records where parentheses were
2799 found. This makes it easy to reproduce these when printing. Once
2800 precedence is handled better I might be able to discard this.
2809 ###### expr precedence
2814 ###### expression grammar
2815 | Expression Eop Expression ${ {
2816 struct binode *b = new(binode);
2823 | Expression Top Expression ${ {
2824 struct binode *b = new(binode);
2831 | ( Expression ) ${ {
2832 struct binode *b = new_pos(binode, $1);
2837 | Uop Expression ${ {
2838 struct binode *b = new(binode);
2843 | Value ${ $0 = $<1; }$
2844 | Variable ${ $0 = $<1; }$
2847 Eop -> + ${ $0.op = Plus; }$
2848 | - ${ $0.op = Minus; }$
2850 Uop -> + ${ $0.op = Absolute; }$
2851 | - ${ $0.op = Negate; }$
2853 Top -> * ${ $0.op = Times; }$
2854 | / ${ $0.op = Divide; }$
2855 | % ${ $0.op = Rem; }$
2856 | ++ ${ $0.op = Concat; }$
2858 ###### print binode cases
2865 if (bracket) printf("(");
2866 print_exec(b->left, indent, bracket);
2868 case Plus: fputs(" + ", stdout); break;
2869 case Minus: fputs(" - ", stdout); break;
2870 case Times: fputs(" * ", stdout); break;
2871 case Divide: fputs(" / ", stdout); break;
2872 case Rem: fputs(" % ", stdout); break;
2873 case Concat: fputs(" ++ ", stdout); break;
2874 default: abort(); // NOTEST
2876 print_exec(b->right, indent, bracket);
2877 if (bracket) printf(")");
2880 if (bracket) printf("(");
2882 print_exec(b->right, indent, bracket);
2883 if (bracket) printf(")");
2886 if (bracket) printf("(");
2888 print_exec(b->right, indent, bracket);
2889 if (bracket) printf(")");
2893 print_exec(b->right, indent, bracket);
2897 ###### propagate binode cases
2903 /* both must be numbers, result is Tnum */
2906 /* as propagate_types ignores a NULL,
2907 * unary ops fit here too */
2908 propagate_types(b->left, c, ok, Tnum, 0);
2909 propagate_types(b->right, c, ok, Tnum, 0);
2910 if (!type_compat(type, Tnum, 0))
2911 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
2916 /* both must be Tstr, result is Tstr */
2917 propagate_types(b->left, c, ok, Tstr, 0);
2918 propagate_types(b->right, c, ok, Tstr, 0);
2919 if (!type_compat(type, Tstr, 0))
2920 type_err(c, "error: Concat returns %1 but %2 expected", prog,
2925 return propagate_types(b->right, c, ok, type, 0);
2927 ###### interp binode cases
2930 rv = interp_exec(b->left);
2931 right = interp_exec(b->right);
2932 mpq_add(rv.num, rv.num, right.num);
2935 rv = interp_exec(b->left);
2936 right = interp_exec(b->right);
2937 mpq_sub(rv.num, rv.num, right.num);
2940 rv = interp_exec(b->left);
2941 right = interp_exec(b->right);
2942 mpq_mul(rv.num, rv.num, right.num);
2945 rv = interp_exec(b->left);
2946 right = interp_exec(b->right);
2947 mpq_div(rv.num, rv.num, right.num);
2952 left = interp_exec(b->left);
2953 right = interp_exec(b->right);
2954 mpz_init(l); mpz_init(r); mpz_init(rem);
2955 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
2956 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
2957 mpz_tdiv_r(rem, l, r);
2958 rv = val_init(Tnum);
2959 mpq_set_z(rv.num, rem);
2960 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
2964 rv = interp_exec(b->right);
2965 mpq_neg(rv.num, rv.num);
2968 rv = interp_exec(b->right);
2969 mpq_abs(rv.num, rv.num);
2972 rv = interp_exec(b->right);
2975 left = interp_exec(b->left);
2976 right = interp_exec(b->right);
2978 rv.str = text_join(left.str, right.str);
2981 ###### value functions
2983 static struct text text_join(struct text a, struct text b)
2986 rv.len = a.len + b.len;
2987 rv.txt = malloc(rv.len);
2988 memcpy(rv.txt, a.txt, a.len);
2989 memcpy(rv.txt+a.len, b.txt, b.len);
2993 ### Blocks, Statements, and Statement lists.
2995 Now that we have expressions out of the way we need to turn to
2996 statements. There are simple statements and more complex statements.
2997 Simple statements do not contain (syntactic) newlines, complex statements do.
2999 Statements often come in sequences and we have corresponding simple
3000 statement lists and complex statement lists.
3001 The former comprise only simple statements separated by semicolons.
3002 The later comprise complex statements and simple statement lists. They are
3003 separated by newlines. Thus the semicolon is only used to separate
3004 simple statements on the one line. This may be overly restrictive,
3005 but I'm not sure I ever want a complex statement to share a line with
3008 Note that a simple statement list can still use multiple lines if
3009 subsequent lines are indented, so
3011 ###### Example: wrapped simple statement list
3016 is a single simple statement list. This might allow room for
3017 confusion, so I'm not set on it yet.
3019 A simple statement list needs no extra syntax. A complex statement
3020 list has two syntactic forms. It can be enclosed in braces (much like
3021 C blocks), or it can be introduced by a colon and continue until an
3022 unindented newline (much like Python blocks). With this extra syntax
3023 it is referred to as a block.
3025 Note that a block does not have to include any newlines if it only
3026 contains simple statements. So both of:
3028 if condition: a=b; d=f
3030 if condition { a=b; print f }
3034 In either case the list is constructed from a `binode` list with
3035 `Block` as the operator. When parsing the list it is most convenient
3036 to append to the end, so a list is a list and a statement. When using
3037 the list it is more convenient to consider a list to be a statement
3038 and a list. So we need a function to re-order a list.
3039 `reorder_bilist` serves this purpose.
3041 The only stand-alone statement we introduce at this stage is `pass`
3042 which does nothing and is represented as a `NULL` pointer in a `Block`
3043 list. Other stand-alone statements will follow once the infrastructure
3056 Block -> Open Statementlist Close ${ $0 = $<2; }$
3057 | Open SimpleStatements } ${ $0 = reorder_bilist($<2); }$
3058 | : SimpleStatements ${ $0 = reorder_bilist($<2); }$
3059 | : Statementlist ${ $0 = $<2; }$
3061 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<1); }$
3063 ComplexStatements -> ComplexStatements ComplexStatement ${
3073 | ComplexStatement ${
3085 ComplexStatement -> SimpleStatements NEWLINE ${
3086 $0 = reorder_bilist($<1);
3088 | Newlines ${ $0 = NULL; }$
3089 ## ComplexStatement Grammar
3092 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3098 | SimpleStatement ${
3104 | SimpleStatements ; ${ $0 = $<1; }$
3106 SimpleStatement -> pass ${ $0 = NULL; }$
3107 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3108 ## SimpleStatement Grammar
3110 ###### print binode cases
3114 if (b->left == NULL)
3117 print_exec(b->left, indent, bracket);
3120 print_exec(b->right, indent, bracket);
3123 // block, one per line
3124 if (b->left == NULL)
3125 do_indent(indent, "pass\n");
3127 print_exec(b->left, indent, bracket);
3129 print_exec(b->right, indent, bracket);
3133 ###### propagate binode cases
3136 /* If any statement returns something other than Tnone
3137 * or Tbool then all such must return same type.
3138 * As each statement may be Tnone or something else,
3139 * we must always pass NULL (unknown) down, otherwise an incorrect
3140 * error might occur. We never return Tnone unless it is
3145 for (e = b; e; e = cast(binode, e->right)) {
3146 t = propagate_types(e->left, c, ok, NULL, rules);
3147 if ((rules & Rboolok) && t == Tbool)
3149 if (t && t != Tnone && t != Tbool) {
3153 type_err(c, "error: expected %1%r, found %2",
3154 e->left, type, rules, t);
3160 ###### interp binode cases
3162 while (rv.type == Tnone &&
3165 rv = interp_exec(b->left);
3166 b = cast(binode, b->right);
3170 ### The Print statement
3172 `print` is a simple statement that takes a comma-separated list of
3173 expressions and prints the values separated by spaces and terminated
3174 by a newline. No control of formatting is possible.
3176 `print` faces the same list-ordering issue as blocks, and uses the
3182 ###### SimpleStatement Grammar
3184 | print ExpressionList ${
3185 $0 = reorder_bilist($<2);
3187 | print ExpressionList , ${
3192 $0 = reorder_bilist($0);
3203 ExpressionList -> ExpressionList , Expression ${
3216 ###### print binode cases
3219 do_indent(indent, "print");
3223 print_exec(b->left, -1, bracket);
3227 b = cast(binode, b->right);
3233 ###### propagate binode cases
3236 /* don't care but all must be consistent */
3237 propagate_types(b->left, c, ok, NULL, Rnolabel);
3238 propagate_types(b->right, c, ok, NULL, Rnolabel);
3241 ###### interp binode cases
3247 for ( ; b; b = cast(binode, b->right))
3251 left = interp_exec(b->left);
3264 ###### Assignment statement
3266 An assignment will assign a value to a variable, providing it hasn't
3267 be declared as a constant. The analysis phase ensures that the type
3268 will be correct so the interpreter just needs to perform the
3269 calculation. There is a form of assignment which declares a new
3270 variable as well as assigning a value. If a name is assigned before
3271 it is declared, and error will be raised as the name is created as
3272 `Tlabel` and it is illegal to assign to such names.
3278 ###### SimpleStatement Grammar
3279 | Variable = Expression ${
3285 | VariableDecl = Expression ${
3293 if ($1->var->where_set == NULL) {
3295 "Variable declared with no type or value: %v",
3305 ###### print binode cases
3308 do_indent(indent, "");
3309 print_exec(b->left, indent, bracket);
3311 print_exec(b->right, indent, bracket);
3318 struct variable *v = cast(var, b->left)->var;
3319 do_indent(indent, "");
3320 print_exec(b->left, indent, bracket);
3321 if (cast(var, b->left)->var->constant) {
3322 if (v->where_decl == v->where_set) {
3324 type_print(v->val.type, stdout);
3329 if (v->where_decl == v->where_set) {
3331 type_print(v->val.type, stdout);
3338 print_exec(b->right, indent, bracket);
3345 ###### propagate binode cases
3349 /* Both must match and not be labels,
3350 * Type must support 'dup',
3351 * For Assign, left must not be constant.
3354 t = propagate_types(b->left, c, ok, NULL,
3355 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3360 if (propagate_types(b->right, c, ok, t, 0) != t)
3361 if (b->left->type == Xvar)
3362 type_err(c, "info: variable '%v' was set as %1 here.",
3363 cast(var, b->left)->var->where_set, t, rules, NULL);
3365 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3367 propagate_types(b->left, c, ok, t,
3368 (b->op == Assign ? Rnoconstant : 0));
3370 if (t && t->dup == NULL)
3371 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3376 ###### interp binode cases
3379 lleft = linterp_exec(b->left);
3380 right = interp_exec(b->right);
3385 free_value(right); // NOTEST
3391 struct variable *v = cast(var, b->left)->var;
3395 right = interp_exec(b->right);
3397 right = val_init(v->val.type);
3404 ### The `use` statement
3406 The `use` statement is the last "simple" statement. It is needed when
3407 the condition in a conditional statement is a block. `use` works much
3408 like `return` in C, but only completes the `condition`, not the whole
3414 ###### SimpleStatement Grammar
3416 $0 = new_pos(binode, $1);
3419 if ($0->right->type == Xvar) {
3420 struct var *v = cast(var, $0->right);
3421 if (v->var->val.type == Tnone) {
3422 /* Convert this to a label */
3423 v->var->val = val_prepare(Tlabel);
3424 v->var->val.label = &v->var->val;
3429 ###### print binode cases
3432 do_indent(indent, "use ");
3433 print_exec(b->right, -1, bracket);
3438 ###### propagate binode cases
3441 /* result matches value */
3442 return propagate_types(b->right, c, ok, type, 0);
3444 ###### interp binode cases
3447 rv = interp_exec(b->right);
3450 ### The Conditional Statement
3452 This is the biggy and currently the only complex statement. This
3453 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3454 It is comprised of a number of parts, all of which are optional though
3455 set combinations apply. Each part is (usually) a key word (`then` is
3456 sometimes optional) followed by either an expression or a code block,
3457 except the `casepart` which is a "key word and an expression" followed
3458 by a code block. The code-block option is valid for all parts and,
3459 where an expression is also allowed, the code block can use the `use`
3460 statement to report a value. If the code block does not report a value
3461 the effect is similar to reporting `True`.
3463 The `else` and `case` parts, as well as `then` when combined with
3464 `if`, can contain a `use` statement which will apply to some
3465 containing conditional statement. `for` parts, `do` parts and `then`
3466 parts used with `for` can never contain a `use`, except in some
3467 subordinate conditional statement.
3469 If there is a `forpart`, it is executed first, only once.
3470 If there is a `dopart`, then it is executed repeatedly providing
3471 always that the `condpart` or `cond`, if present, does not return a non-True
3472 value. `condpart` can fail to return any value if it simply executes
3473 to completion. This is treated the same as returning `True`.
3475 If there is a `thenpart` it will be executed whenever the `condpart`
3476 or `cond` returns True (or does not return any value), but this will happen
3477 *after* `dopart` (when present).
3479 If `elsepart` is present it will be executed at most once when the
3480 condition returns `False` or some value that isn't `True` and isn't
3481 matched by any `casepart`. If there are any `casepart`s, they will be
3482 executed when the condition returns a matching value.
3484 The particular sorts of values allowed in case parts has not yet been
3485 determined in the language design, so nothing is prohibited.
3487 The various blocks in this complex statement potentially provide scope
3488 for variables as described earlier. Each such block must include the
3489 "OpenScope" nonterminal before parsing the block, and must call
3490 `var_block_close()` when closing the block.
3492 The code following "`if`", "`switch`" and "`for`" does not get its own
3493 scope, but is in a scope covering the whole statement, so names
3494 declared there cannot be redeclared elsewhere. Similarly the
3495 condition following "`while`" is in a scope the covers the body
3496 ("`do`" part) of the loop, and which does not allow conditional scope
3497 extension. Code following "`then`" (both looping and non-looping),
3498 "`else`" and "`case`" each get their own local scope.
3500 The type requirements on the code block in a `whilepart` are quite
3501 unusal. It is allowed to return a value of some identifiable type, in
3502 which case the loop aborts and an appropriate `casepart` is run, or it
3503 can return a Boolean, in which case the loop either continues to the
3504 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3505 This is different both from the `ifpart` code block which is expected to
3506 return a Boolean, or the `switchpart` code block which is expected to
3507 return the same type as the casepart values. The correct analysis of
3508 the type of the `whilepart` code block is the reason for the
3509 `Rboolok` flag which is passed to `propagate_types()`.
3511 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3520 struct exec *action;
3521 struct casepart *next;
3523 struct cond_statement {
3525 struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
3526 struct casepart *casepart;
3529 ###### ast functions
3531 static void free_casepart(struct casepart *cp)
3535 free_exec(cp->value);
3536 free_exec(cp->action);
3543 static void free_cond_statement(struct cond_statement *s)
3547 free_exec(s->forpart);
3548 free_exec(s->condpart);
3549 free_exec(s->dopart);
3550 free_exec(s->thenpart);
3551 free_exec(s->elsepart);
3552 free_casepart(s->casepart);
3556 ###### free exec cases
3557 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3559 ###### ComplexStatement Grammar
3560 | CondStatement ${ $0 = $<1; }$
3565 // both ForThen and Whilepart open scopes, and CondSuffix only
3566 // closes one - so in the first branch here we have another to close.
3567 CondStatement -> forPart ThenPart WhilePart CondSuffix ${
3571 $0->condpart = $3.condpart; $3.condpart = NULL;
3572 $0->dopart = $3.dopart; $3.dopart = NULL;
3573 var_block_close(c, CloseSequential);
3575 | forPart WhilePart CondSuffix ${
3578 $0->thenpart = NULL;
3579 $0->condpart = $2.condpart; $2.condpart = NULL;
3580 $0->dopart = $2.dopart; $2.dopart = NULL;
3581 var_block_close(c, CloseSequential);
3583 | whilePart CondSuffix ${
3585 $0->condpart = $1.condpart; $1.condpart = NULL;
3586 $0->dopart = $1.dopart; $1.dopart = NULL;
3588 | switchPart CondSuffix ${
3592 | ifPart IfSuffix ${
3594 $0->condpart = $1.condpart; $1.condpart = NULL;
3595 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
3596 // This is where we close an "if" statement
3597 var_block_close(c, CloseSequential);
3600 CondSuffix -> IfSuffix ${
3602 // This is where we close scope of the whole
3603 // "for" or "while" statement
3604 var_block_close(c, CloseSequential);
3606 | CasePart CondSuffix ${
3608 $1->next = $0->casepart;
3616 CasePart -> Case Expression OpenScope Block ${
3617 $0 = calloc(1,sizeof(struct casepart));
3620 var_block_close(c, CloseParallel);
3624 IfSuffix -> ${ $0 = new(cond_statement); }$
3625 | NEWLINE IfSuffix ${ $0 = $<2; }$
3626 | else OpenScope Block ${
3627 $0 = new(cond_statement);
3629 var_block_close(c, CloseElse);
3631 | else OpenScope CondStatement ${
3632 $0 = new(cond_statement);
3634 var_block_close(c, CloseElse);
3645 // These scopes are closed in CondSuffix
3646 forPart -> for OpenScope SimpleStatements ${
3647 $0 = reorder_bilist($<3);
3649 | for OpenScope Block ${
3653 ThenPart -> Then OpenScope SimpleStatements ${
3654 $0 = reorder_bilist($<3);
3655 var_block_close(c, CloseSequential);
3657 | Then OpenScope Block ${
3659 var_block_close(c, CloseSequential);
3662 // This scope is closed in CondSuffix
3663 WhileHead -> While OpenScope Block ${
3666 whileHead -> while OpenScope Block ${
3671 // This scope is closed in CondSuffix
3672 whilePart -> while OpenScope Expression Block ${
3673 $0.type = Xcond_statement;
3677 | whileHead Do Block ${
3678 $0.type = Xcond_statement;
3682 WhilePart -> While OpenScope Expression Block ${
3683 $0.type = Xcond_statement;
3687 | WhileHead Do Block ${
3688 $0.type = Xcond_statement;
3693 ifPart -> if OpenScope Expression OpenScope Block ${
3694 $0.type = Xcond_statement;
3697 var_block_close(c, CloseParallel);
3699 | if OpenScope Block Then OpenScope Block ${
3700 $0.type = Xcond_statement;
3703 var_block_close(c, CloseParallel);
3707 // This scope is closed in CondSuffix
3708 switchPart -> switch OpenScope Expression ${
3711 | switch OpenScope Block ${
3715 ###### print exec cases
3717 case Xcond_statement:
3719 struct cond_statement *cs = cast(cond_statement, e);
3720 struct casepart *cp;
3722 do_indent(indent, "for");
3723 if (bracket) printf(" {\n"); else printf(":\n");
3724 print_exec(cs->forpart, indent+1, bracket);
3727 do_indent(indent, "} then {\n");
3729 do_indent(indent, "then:\n");
3730 print_exec(cs->thenpart, indent+1, bracket);
3732 if (bracket) do_indent(indent, "}\n");
3736 if (cs->condpart && cs->condpart->type == Xbinode &&
3737 cast(binode, cs->condpart)->op == Block) {
3739 do_indent(indent, "while {\n");
3741 do_indent(indent, "while:\n");
3742 print_exec(cs->condpart, indent+1, bracket);
3744 do_indent(indent, "} do {\n");
3746 do_indent(indent, "do:\n");
3747 print_exec(cs->dopart, indent+1, bracket);
3749 do_indent(indent, "}\n");
3751 do_indent(indent, "while ");
3752 print_exec(cs->condpart, 0, bracket);
3757 print_exec(cs->dopart, indent+1, bracket);
3759 do_indent(indent, "}\n");
3764 do_indent(indent, "switch");
3766 do_indent(indent, "if");
3767 if (cs->condpart && cs->condpart->type == Xbinode &&
3768 cast(binode, cs->condpart)->op == Block) {
3773 print_exec(cs->condpart, indent+1, bracket);
3775 do_indent(indent, "}\n");
3777 do_indent(indent, "then:\n");
3778 print_exec(cs->thenpart, indent+1, bracket);
3782 print_exec(cs->condpart, 0, bracket);
3788 print_exec(cs->thenpart, indent+1, bracket);
3790 do_indent(indent, "}\n");
3795 for (cp = cs->casepart; cp; cp = cp->next) {
3796 do_indent(indent, "case ");
3797 print_exec(cp->value, -1, 0);
3802 print_exec(cp->action, indent+1, bracket);
3804 do_indent(indent, "}\n");
3807 do_indent(indent, "else");
3812 print_exec(cs->elsepart, indent+1, bracket);
3814 do_indent(indent, "}\n");
3819 ###### propagate exec cases
3820 case Xcond_statement:
3822 // forpart and dopart must return Tnone
3823 // thenpart must return Tnone if there is a dopart,
3824 // otherwise it is like elsepart.
3826 // be bool if there is no casepart
3827 // match casepart->values if there is a switchpart
3828 // either be bool or match casepart->value if there
3830 // elsepart and casepart->action must match the return type
3831 // expected of this statement.
3832 struct cond_statement *cs = cast(cond_statement, prog);
3833 struct casepart *cp;
3835 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
3836 if (!type_compat(Tnone, t, 0))
3838 t = propagate_types(cs->dopart, c, ok, Tnone, 0);
3839 if (!type_compat(Tnone, t, 0))
3842 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
3843 if (!type_compat(Tnone, t, 0))
3846 if (cs->casepart == NULL)
3847 propagate_types(cs->condpart, c, ok, Tbool, 0);
3849 /* Condpart must match case values, with bool permitted */
3851 for (cp = cs->casepart;
3852 cp && !t; cp = cp->next)
3853 t = propagate_types(cp->value, c, ok, NULL, 0);
3854 if (!t && cs->condpart)
3855 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok);
3856 // Now we have a type (I hope) push it down
3858 for (cp = cs->casepart; cp; cp = cp->next)
3859 propagate_types(cp->value, c, ok, t, 0);
3860 propagate_types(cs->condpart, c, ok, t, Rboolok);
3863 // (if)then, else, and case parts must return expected type.
3864 if (!cs->dopart && !type)
3865 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
3867 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
3868 for (cp = cs->casepart;
3871 type = propagate_types(cp->action, c, ok, NULL, rules);
3874 propagate_types(cs->thenpart, c, ok, type, rules);
3875 propagate_types(cs->elsepart, c, ok, type, rules);
3876 for (cp = cs->casepart; cp ; cp = cp->next)
3877 propagate_types(cp->action, c, ok, type, rules);
3883 ###### interp exec cases
3884 case Xcond_statement:
3886 struct value v, cnd;
3887 struct casepart *cp;
3888 struct cond_statement *c = cast(cond_statement, e);
3891 interp_exec(c->forpart);
3894 cnd = interp_exec(c->condpart);
3897 if (!(cnd.type == Tnone ||
3898 (cnd.type == Tbool && cnd.bool != 0)))
3900 // cnd is Tnone or Tbool, doesn't need to be freed
3902 interp_exec(c->dopart);
3905 rv = interp_exec(c->thenpart);
3906 if (rv.type != Tnone || !c->dopart)
3910 } while (c->dopart);
3912 for (cp = c->casepart; cp; cp = cp->next) {
3913 v = interp_exec(cp->value);
3914 if (value_cmp(v, cnd) == 0) {
3917 rv = interp_exec(cp->action);
3924 rv = interp_exec(c->elsepart);
3931 ### Top level structure
3933 All the language elements so far can be used in various places. Now
3934 it is time to clarify what those places are.
3936 At the top level of a file there will be a number of declarations.
3937 Many of the things that can be declared haven't been described yet,
3938 such as functions, procedures, imports, and probably more.
3939 For now there are two sorts of things that can appear at the top
3940 level. They are predefined constants, `struct` types, and the main
3941 program. While the syntax will allow the main program to appear
3942 multiple times, that will trigger an error if it is actually attempted.
3944 The various declarations do not return anything. They store the
3945 various declarations in the parse context.
3947 ###### Parser: grammar
3950 Ocean -> DeclarationList
3952 DeclarationList -> Declaration
3953 | DeclarationList Declaration
3955 Declaration -> DeclareConstant
3961 "error: unhandled parse error", &$1);
3964 ## top level grammar
3966 ### The `const` section
3968 As well as being defined in with the code that uses them, constants
3969 can be declared at the top level. These have full-file scope, so they
3970 are always `InScope`. The value of a top level constant can be given
3971 as an expression, and this is evaluated immediately rather than in the
3972 later interpretation stage. Once we add functions to the language, we
3973 will need rules concern which, if any, can be used to define a top
3976 Constants are defined in a section that starts with the reserved word
3977 `const` and then has a block with a list of assignment statements.
3978 For syntactic consistency, these must use the double-colon syntax to
3979 make it clear that they are constants. Type can also be given: if
3980 not, the type will be determined during analysis, as with other
3983 As the types constants are inserted at the head of a list, printing
3984 them in the same order that they were read is not straight forward.
3985 We take a quadratic approach here and count the number of constants
3986 (variables of depth 0), then count down from there, each time
3987 searching through for the Nth constant for decreasing N.
3989 ###### top level grammar
3991 DeclareConstant -> const Open ConstList Close
3992 | const Open SimpleConstList }
3994 | const SimpleConstList NEWLINE
3996 ConstList -> ComplexConsts
3998 ComplexConsts -> ComplexConst ComplexConsts
4000 ComplexConst -> SimpleConstList NEWLINE
4001 SimpleConstList -> SimpleConstList ; Const
4006 CType -> Type ${ $0 = $<1; }$
4009 Const -> IDENTIFIER :: CType = Expression ${ {
4013 v = var_decl(c, $1.txt);
4015 struct var *var = new_pos(var, $1);
4016 v->where_decl = var;
4021 v = var_ref(c, $1.txt);
4022 tok_err(c, "error: name already declared", &$1);
4023 type_err(c, "info: this is where '%v' was first declared",
4024 v->where_decl, NULL, 0, NULL);
4028 propagate_types($5, c, &ok, $3, 0);
4033 v->val = interp_exec($5);
4036 | ERROR NEWLINE ${ tok_err(c, "Syntax error in constant", &$1); }$
4038 ###### print const decls
4043 while (target != 0) {
4045 for (v = context.in_scope; v; v=v->in_scope)
4046 if (v->depth == 0) {
4057 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4058 type_print(v->val.type, stdout);
4060 if (v->val.type == Tstr)
4062 print_value(v->val);
4063 if (v->val.type == Tstr)
4071 ### Finally the whole program.
4073 Somewhat reminiscent of Pascal a (current) Ocean program starts with
4074 the keyword "program" and a list of variable names which are assigned
4075 values from command line arguments. Following this is a `block` which
4076 is the code to execute. Unlike Pascal, constants and other
4077 declarations come *before* the program.
4079 As this is the top level, several things are handled a bit
4081 The whole program is not interpreted by `interp_exec` as that isn't
4082 passed the argument list which the program requires. Similarly type
4083 analysis is a bit more interesting at this level.
4088 ###### top level grammar
4090 DeclareProgram -> Program ${ {
4092 type_err(c, "Program defined a second time",
4099 Program -> program OpenScope Varlist Block ${
4102 $0->left = reorder_bilist($<3);
4104 var_block_close(c, CloseSequential);
4105 if (c->scope_stack && !c->parse_error) abort();
4109 "error: unhandled parse error", &$1);
4112 Varlist -> Varlist ArgDecl ${
4121 ArgDecl -> IDENTIFIER ${ {
4122 struct variable *v = var_decl(c, $1.txt);
4129 ###### print binode cases
4131 do_indent(indent, "program");
4132 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4134 print_exec(b2->left, 0, 0);
4140 print_exec(b->right, indent+1, bracket);
4142 do_indent(indent, "}\n");
4145 ###### propagate binode cases
4146 case Program: abort(); // NOTEST
4148 ###### core functions
4150 static int analyse_prog(struct exec *prog, struct parse_context *c)
4152 struct binode *b = cast(binode, prog);
4159 propagate_types(b->right, c, &ok, Tnone, 0);
4164 for (b = cast(binode, b->left); b; b = cast(binode, b->right)) {
4165 struct var *v = cast(var, b->left);
4166 if (!v->var->val.type) {
4167 v->var->where_set = b;
4168 v->var->val = val_prepare(Tstr);
4171 b = cast(binode, prog);
4174 propagate_types(b->right, c, &ok, Tnone, 0);
4179 /* Make sure everything is still consistent */
4180 propagate_types(b->right, c, &ok, Tnone, 0);
4184 static void interp_prog(struct exec *prog, char **argv)
4186 struct binode *p = cast(binode, prog);
4192 al = cast(binode, p->left);
4194 struct var *v = cast(var, al->left);
4195 struct value *vl = &v->var->val;
4197 if (argv[0] == NULL) {
4198 printf("Not enough args\n");
4201 al = cast(binode, al->right);
4203 *vl = parse_value(vl->type, argv[0]);
4204 if (vl->type == NULL)
4208 v = interp_exec(p->right);
4212 ###### interp binode cases
4213 case Program: abort(); // NOTEST
4215 ## And now to test it out.
4217 Having a language requires having a "hello world" program. I'll
4218 provide a little more than that: a program that prints "Hello world"
4219 finds the GCD of two numbers, prints the first few elements of
4220 Fibonacci, performs a binary search for a number, and a few other
4221 things which will likely grow as the languages grows.
4223 ###### File: oceani.mk
4226 @echo "===== DEMO ====="
4227 ./oceani --section "demo: hello" oceani.mdc 55 33
4233 four ::= 2 + 2 ; five ::= 10/2
4234 const pie ::= "I like Pie";
4235 cake ::= "The cake is"
4244 print "Hello World, what lovely oceans you have!"
4245 print "Are there", five, "?"
4246 print pi, pie, "but", cake
4248 /* When a variable is defined in both branches of an 'if',
4249 * and used afterwards, the variables are merged.
4255 print "Is", A, "bigger than", B,"? ", bigger
4256 /* If a variable is not used after the 'if', no
4257 * merge happens, so types can be different
4260 double:string = "yes"
4261 print A, "is more than twice", B, "?", double
4264 print "double", B, "is", double
4269 if a > 0 and then b > 0:
4275 print "GCD of", A, "and", B,"is", a
4277 print a, "is not positive, cannot calculate GCD"
4279 print b, "is not positive, cannot calculate GCD"
4284 print "Fibonacci:", f1,f2,
4285 then togo = togo - 1
4293 /* Binary search... */
4298 mid := (lo + hi) / 2
4310 print "Yay, I found", target
4312 print "Closest I found was", mid
4317 // "middle square" PRNG. Not particularly good, but one my
4318 // Dad taught me - the first one I ever heard of.
4319 for i:=1; then i = i + 1; while i < size:
4320 n := list[i-1] * list[i-1]
4321 list[i] = (n / 100) % 10 000
4323 print "Before sort:",
4324 for i:=0; then i = i + 1; while i < size:
4328 for i := 1; then i=i+1; while i < size:
4329 for j:=i-1; then j=j-1; while j >= 0:
4330 if list[j] > list[j+1]:
4334 print " After sort:",
4335 for i:=0; then i = i + 1; while i < size:
4341 bob.alive = (bob.name == "Hello")
4342 print "bob", "is" if bob.alive else "isn't", "alive"