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, \
132 #include <sys/mman.h>
151 static char Usage[] = "Usage: oceani --trace --print --noexec --brackets"
152 "--section=SectionName prog.ocn\n";
153 static const struct option long_options[] = {
154 {"trace", 0, NULL, 't'},
155 {"print", 0, NULL, 'p'},
156 {"noexec", 0, NULL, 'n'},
157 {"brackets", 0, NULL, 'b'},
158 {"section", 1, NULL, 's'},
161 const char *options = "tpnbs";
162 int main(int argc, char *argv[])
167 struct section *s, *ss;
168 char *section = NULL;
169 struct parse_context context = {
171 .ignored = (1 << TK_line_comment)
172 | (1 << TK_block_comment)
174 .number_chars = ".,_+- ",
179 int doprint=0, dotrace=0, doexec=1, brackets=0;
181 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
184 case 't': dotrace=1; break;
185 case 'p': doprint=1; break;
186 case 'n': doexec=0; break;
187 case 'b': brackets=1; break;
188 case 's': section = optarg; break;
189 default: fprintf(stderr, Usage);
193 if (optind >= argc) {
194 fprintf(stderr, "oceani: no input file given\n");
197 fd = open(argv[optind], O_RDONLY);
199 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
202 context.file_name = argv[optind];
203 len = lseek(fd, 0, 2);
204 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
205 s = code_extract(file, file+len, NULL);
207 fprintf(stderr, "oceani: could not find any code in %s\n",
212 ## context initialization
215 for (ss = s; ss; ss = ss->next) {
216 struct text sec = ss->section;
217 if (sec.len == strlen(section) &&
218 strncmp(sec.txt, section, sec.len) == 0)
222 fprintf(stderr, "oceani: cannot find section %s\n",
228 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
231 fprintf(stderr, "oceani: no program found.\n");
232 context.parse_error = 1;
234 if (context.prog && doprint) {
237 print_exec(context.prog, 0, brackets);
239 if (context.prog && doexec && !context.parse_error) {
240 if (!analyse_prog(context.prog, &context)) {
241 fprintf(stderr, "oceani: type error in program - not running.\n");
244 interp_prog(context.prog, argv+optind+1);
246 free_exec(context.prog);
249 struct section *t = s->next;
255 ## free context types
256 exit(context.parse_error ? 1 : 0);
261 The four requirements of parse, analyse, print, interpret apply to
262 each language element individually so that is how most of the code
265 Three of the four are fairly self explanatory. The one that requires
266 a little explanation is the analysis step.
268 The current language design does not require the types of variables to
269 be declared, but they must still have a single type. Different
270 operations impose different requirements on the variables, for example
271 addition requires both arguments to be numeric, and assignment
272 requires the variable on the left to have the same type as the
273 expression on the right.
275 Analysis involves propagating these type requirements around and
276 consequently setting the type of each variable. If any requirements
277 are violated (e.g. a string is compared with a number) or if a
278 variable needs to have two different types, then an error is raised
279 and the program will not run.
281 If the same variable is declared in both branchs of an 'if/else', or
282 in all cases of a 'switch' then the multiple instances may be merged
283 into just one variable if the variable is references after the
284 conditional statement. When this happens, the types must naturally be
285 consistent across all the branches. When the variable is not used
286 outside the if, the variables in the different branches are distinct
287 and can be of different types.
289 Determining the types of all variables early is important for
290 processing command line arguments. These can be assigned to any of
291 several types of variable, but we must first know the correct type so
292 any required conversion can happen. If a variable is associated with
293 a command line argument but no type can be interpreted (e.g. the
294 variable is only ever used in a `print` statement), then the type is
297 Undeclared names may only appear in "use" statements and "case" expressions.
298 These names are given a type of "label" and a unique value.
299 This allows them to fill the role of a name in an enumerated type, which
300 is useful for testing the `switch` statement.
302 As we will see, the condition part of a `while` statement can return
303 either a Boolean or some other type. This requires that the expected
304 type that gets passed around comprises a type and a flag to indicate
305 that `Tbool` is also permitted.
307 As there are, as yet, no distinct types that are compatible, there
308 isn't much subtlety in the analysis. When we have distinct number
309 types, this will become more interesting.
313 When analysis discovers an inconsistency it needs to report an error;
314 just refusing to run the code ensures that the error doesn't cascade,
315 but by itself it isn't very useful. A clear understanding of the sort
316 of error message that are useful will help guide the process of
319 At a simplistic level, the only sort of error that type analysis can
320 report is that the type of some construct doesn't match a contextual
321 requirement. For example, in `4 + "hello"` the addition provides a
322 contextual requirement for numbers, but `"hello"` is not a number. In
323 this particular example no further information is needed as the types
324 are obvious from local information. When a variable is involved that
325 isn't the case. It may be helpful to explain why the variable has a
326 particular type, by indicating the location where the type was set,
327 whether by declaration or usage.
329 Using a recursive-descent analysis we can easily detect a problem at
330 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
331 will detect that one argument is not a number and the usage of `hello`
332 will detect that a number was wanted, but not provided. In this
333 (early) version of the language, we will generate error reports at
334 multiple locations, so the use of `hello` will report an error and
335 explain were the value was set, and the addition will report an error
336 and say why numbers are needed. To be able to report locations for
337 errors, each language element will need to record a file location
338 (line and column) and each variable will need to record the language
339 element where its type was set. For now we will assume that each line
340 of an error message indicates one location in the file, and up to 2
341 types. So we provide a `printf`-like function which takes a format, a
342 language (a `struct exec` which has not yet been introduced), and 2
343 types. "`%1`" reports the first type, "`%2`" reports the second. We
344 will need a function to print the location, once we know how that is
345 stored. As will be explained later, there are sometimes extra rules for
346 type matching and they might affect error messages, we need to pass those
349 As well as type errors, we sometimes need to report problems with
350 tokens, which might be unexpected or might name a type that has not
351 been defined. For these we have `tok_err()` which reports an error
352 with a given token. Each of the error functions sets the flag in the
353 context so indicate that parsing failed.
357 static void fput_loc(struct exec *loc, FILE *f);
359 ###### core functions
361 static void type_err(struct parse_context *c,
362 char *fmt, struct exec *loc,
363 struct type *t1, int rules, struct type *t2)
365 fprintf(stderr, "%s:", c->file_name);
366 fput_loc(loc, stderr);
367 for (; *fmt ; fmt++) {
374 case '%': fputc(*fmt, stderr); break; // NOTEST
375 default: fputc('?', stderr); break; // NOTEST
377 type_print(t1, stderr);
380 type_print(t2, stderr);
389 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
391 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
392 t->txt.len, t->txt.txt);
396 ## Entities: declared and predeclared.
398 There are various "things" that the language and/or the interpreter
399 needs to know about to parse and execute a program. These include
400 types, variables, values, and executable code. These are all lumped
401 together under the term "entities" (calling them "objects" would be
402 confusing) and introduced here. These will introduced and described
403 here. The following section will present the different specific code
404 elements which comprise or manipulate these various entities.
408 Values come in a wide range of types, with more likely to be added.
409 Each type needs to be able to parse and print its own values (for
410 convenience at least) as well as to compare two values, at least for
411 equality and possibly for order. For now, values might need to be
412 duplicated and freed, though eventually such manipulations will be
413 better integrated into the language.
415 Rather than requiring every numeric type to support all numeric
416 operations (add, multiple, etc), we allow types to be able to present
417 as one of a few standard types: integer, float, and fraction. The
418 existence of these conversion functions eventaully enable types to
419 determine if they are compatible with other types, though such types
420 have not yet been implemented.
422 Named type are stored in a simple linked list. Objects of each type are "values"
423 which are often passed around by value.
430 ## value union fields
437 struct value (*init)(struct type *type);
438 struct value (*prepare)(struct type *type);
439 struct value (*parse)(struct type *type, char *str);
440 void (*print)(struct value val);
441 void (*print_type)(struct type *type, FILE *f);
442 int (*cmp_order)(struct value v1, struct value v2);
443 int (*cmp_eq)(struct value v1, struct value v2);
444 struct value (*dup)(struct value val);
445 void (*free)(struct value val);
446 void (*free_type)(struct type *t);
447 int (*compat)(struct type *this, struct type *other);
448 long long (*to_int)(struct value *v);
449 double (*to_float)(struct value *v);
450 int (*to_mpq)(mpq_t *q, struct value *v);
459 struct type *typelist;
463 static struct type *find_type(struct parse_context *c, struct text s)
465 struct type *l = c->typelist;
468 text_cmp(l->name, s) != 0)
473 static struct type *add_type(struct parse_context *c, struct text s,
478 n = calloc(1, sizeof(*n));
481 n->next = c->typelist;
486 static void free_type(struct type *t)
488 /* The type is always a reference to something in the
489 * context, so we don't need to free anything.
493 static void free_value(struct value v)
499 static int type_compat(struct type *require, struct type *have, int rules)
501 if ((rules & Rboolok) && have == Tbool)
503 if ((rules & Rnolabel) && have == Tlabel)
505 if (!require || !have)
509 return require->compat(require, have);
511 return require == have;
514 static void type_print(struct type *type, FILE *f)
517 fputs("*unknown*type*", f);
518 else if (type->name.len)
519 fprintf(f, "%.*s", type->name.len, type->name.txt);
520 else if (type->print_type)
521 type->print_type(type, f);
523 fputs("*invalid*type*", f); // NOTEST
526 static struct value val_prepare(struct type *type)
531 return type->prepare(type);
536 static struct value val_init(struct type *type)
541 return type->init(type);
546 static struct value dup_value(struct value v)
549 return v.type->dup(v);
553 static int value_cmp(struct value left, struct value right)
555 if (left.type && left.type->cmp_order)
556 return left.type->cmp_order(left, right);
557 if (left.type && left.type->cmp_eq)
558 return left.type->cmp_eq(left, right);
562 static void print_value(struct value v)
564 if (v.type && v.type->print)
567 printf("*Unknown*"); // NOTEST
570 static struct value parse_value(struct type *type, char *arg)
574 if (type && type->parse)
575 return type->parse(type, arg);
576 rv.type = NULL; // NOTEST
582 static void free_value(struct value v);
583 static int type_compat(struct type *require, struct type *have, int rules);
584 static void type_print(struct type *type, FILE *f);
585 static struct value val_init(struct type *type);
586 static struct value dup_value(struct value v);
587 static int value_cmp(struct value left, struct value right);
588 static void print_value(struct value v);
589 static struct value parse_value(struct type *type, char *arg);
591 ###### free context types
593 while (context.typelist) {
594 struct type *t = context.typelist;
596 context.typelist = t->next;
604 Values of the base types can be numbers, which we represent as
605 multi-precision fractions, strings, Booleans and labels. When
606 analysing the program we also need to allow for places where no value
607 is meaningful (type `Tnone`) and where we don't know what type to
608 expect yet (type is `NULL`).
610 Values are never shared, they are always copied when used, and freed
611 when no longer needed.
613 When propagating type information around the program, we need to
614 determine if two types are compatible, where type `NULL` is compatible
615 with anything. There are two special cases with type compatibility,
616 both related to the Conditional Statement which will be described
617 later. In some cases a Boolean can be accepted as well as some other
618 primary type, and in others any type is acceptable except a label (`Vlabel`).
619 A separate function encoding these cases will simplify some code later.
621 When assigning command line arguments to variables, we need to be able
622 to parse each type from a string.
624 The distinction beteen "prepare" and "init" needs to be explained.
625 "init" sets up an initial value, such as "zero" or the empty string.
626 "prepare" simply prepares the data structure so that if "free" gets
627 called on it, it won't do something silly. Normally a value will be
628 stored after "prepare" but before "free", but this might not happen if
637 myLDLIBS := libnumber.o libstring.o -lgmp
638 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
640 ###### type union fields
641 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
643 ###### value union fields
650 static void _free_value(struct value v)
652 switch (v.type->vtype) {
654 case Vstr: free(v.str.txt); break;
655 case Vnum: mpq_clear(v.num); break;
661 ###### value functions
663 static struct value _val_prepare(struct type *type)
668 switch(type->vtype) {
672 memset(&rv.num, 0, sizeof(rv.num));
688 static struct value _val_init(struct type *type)
693 switch(type->vtype) {
694 case Vnone: // NOTEST
697 mpq_init(rv.num); break;
699 rv.str.txt = malloc(1);
705 case Vlabel: // NOTEST
706 rv.label = NULL; // NOTEST
712 static struct value _dup_value(struct value v)
716 switch (rv.type->vtype) {
717 case Vnone: // NOTEST
727 mpq_set(rv.num, v.num);
730 rv.str.len = v.str.len;
731 rv.str.txt = malloc(rv.str.len);
732 memcpy(rv.str.txt, v.str.txt, v.str.len);
738 static int _value_cmp(struct value left, struct value right)
741 if (left.type != right.type)
742 return left.type - right.type; // NOTEST
743 switch (left.type->vtype) {
744 case Vlabel: cmp = left.label == right.label ? 0 : 1; break;
745 case Vnum: cmp = mpq_cmp(left.num, right.num); break;
746 case Vstr: cmp = text_cmp(left.str, right.str); break;
747 case Vbool: cmp = left.bool - right.bool; break;
748 case Vnone: cmp = 0; // NOTEST
753 static void _print_value(struct value v)
755 switch (v.type->vtype) {
756 case Vnone: // NOTEST
757 printf("*no-value*"); break; // NOTEST
758 case Vlabel: // NOTEST
759 printf("*label-%p*", v.label); break; // NOTEST
761 printf("%.*s", v.str.len, v.str.txt); break;
763 printf("%s", v.bool ? "True":"False"); break;
768 mpf_set_q(fl, v.num);
769 gmp_printf("%Fg", fl);
776 static struct value _parse_value(struct type *type, char *arg)
784 switch(type->vtype) {
785 case Vlabel: // NOTEST
786 case Vnone: // NOTEST
787 val.type = NULL; // NOTEST
790 val.str.len = strlen(arg);
791 val.str.txt = malloc(val.str.len);
792 memcpy(val.str.txt, arg, val.str.len);
799 tx.txt = arg; tx.len = strlen(tx.txt);
800 if (number_parse(val.num, tail, tx) == 0)
803 mpq_neg(val.num, val.num);
805 printf("Unsupported suffix: %s\n", arg);
810 if (strcasecmp(arg, "true") == 0 ||
811 strcmp(arg, "1") == 0)
813 else if (strcasecmp(arg, "false") == 0 ||
814 strcmp(arg, "0") == 0)
817 printf("Bad bool: %s\n", arg);
825 static void _free_value(struct value v);
827 static struct type base_prototype = {
829 .prepare = _val_prepare,
830 .parse = _parse_value,
831 .print = _print_value,
832 .cmp_order = _value_cmp,
833 .cmp_eq = _value_cmp,
838 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
841 static struct type *add_base_type(struct parse_context *c, char *n, enum vtype vt)
843 struct text txt = { n, strlen(n) };
846 t = add_type(c, txt, &base_prototype);
851 ###### context initialization
853 Tbool = add_base_type(&context, "Boolean", Vbool);
854 Tstr = add_base_type(&context, "string", Vstr);
855 Tnum = add_base_type(&context, "number", Vnum);
856 Tnone = add_base_type(&context, "none", Vnone);
857 Tlabel = add_base_type(&context, "label", Vlabel);
861 Variables are scoped named values. We store the names in a linked
862 list of "bindings" sorted lexically, and use sequential search and
869 struct binding *next; // in lexical order
873 This linked list is stored in the parse context so that "reduce"
874 functions can find or add variables, and so the analysis phase can
875 ensure that every variable gets a type.
879 struct binding *varlist; // In lexical order
883 static struct binding *find_binding(struct parse_context *c, struct text s)
885 struct binding **l = &c->varlist;
890 (cmp = text_cmp((*l)->name, s)) < 0)
894 n = calloc(1, sizeof(*n));
901 Each name can be linked to multiple variables defined in different
902 scopes. Each scope starts where the name is declared and continues
903 until the end of the containing code block. Scopes of a given name
904 cannot nest, so a declaration while a name is in-scope is an error.
906 ###### binding fields
907 struct variable *var;
911 struct variable *previous;
913 struct binding *name;
914 struct exec *where_decl;// where name was declared
915 struct exec *where_set; // where type was set
919 While the naming seems strange, we include local constants in the
920 definition of variables. A name declared `var := value` can
921 subsequently be changed, but a name declared `var ::= value` cannot -
924 ###### variable fields
927 Scopes in parallel branches can be partially merged. More
928 specifically, if a given name is declared in both branches of an
929 if/else then its scope is a candidate for merging. Similarly if
930 every branch of an exhaustive switch (e.g. has an "else" clause)
931 declares a given name, then the scopes from the branches are
932 candidates for merging.
934 Note that names declared inside a loop (which is only parallel to
935 itself) are never visible after the loop. Similarly names defined in
936 scopes which are not parallel, such as those started by `for` and
937 `switch`, are never visible after the scope. Only variables defined in
938 both `then` and `else` (including the implicit then after an `if`, and
939 excluding `then` used with `for`) and in all `case`s and `else` of a
940 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
942 Labels, which are a bit like variables, follow different rules.
943 Labels are not explicitly declared, but if an undeclared name appears
944 in a context where a label is legal, that effectively declares the
945 name as a label. The declaration remains in force (or in scope) at
946 least to the end of the immediately containing block and conditionally
947 in any larger containing block which does not declare the name in some
948 other way. Importantly, the conditional scope extension happens even
949 if the label is only used in one parallel branch of a conditional --
950 when used in one branch it is treated as having been declared in all
953 Merge candidates are tentatively visible beyond the end of the
954 branching statement which creates them. If the name is used, the
955 merge is affirmed and they become a single variable visible at the
956 outer layer. If not - if it is redeclared first - the merge lapses.
958 To track scopes we have an extra stack, implemented as a linked list,
959 which roughly parallels the parse stack and which is used exclusively
960 for scoping. When a new scope is opened, a new frame is pushed and
961 the child-count of the parent frame is incremented. This child-count
962 is used to distinguish between the first of a set of parallel scopes,
963 in which declared variables must not be in scope, and subsequent
964 branches, whether they must already be conditionally scoped.
966 To push a new frame *before* any code in the frame is parsed, we need a
967 grammar reduction. This is most easily achieved with a grammar
968 element which derives the empty string, and creates the new scope when
969 it is recognized. This can be placed, for example, between a keyword
970 like "if" and the code following it.
974 struct scope *parent;
980 struct scope *scope_stack;
983 static void scope_pop(struct parse_context *c)
985 struct scope *s = c->scope_stack;
987 c->scope_stack = s->parent;
992 static void scope_push(struct parse_context *c)
994 struct scope *s = calloc(1, sizeof(*s));
996 c->scope_stack->child_count += 1;
997 s->parent = c->scope_stack;
1005 OpenScope -> ${ scope_push(config2context(config)); }$
1007 Each variable records a scope depth and is in one of four states:
1009 - "in scope". This is the case between the declaration of the
1010 variable and the end of the containing block, and also between
1011 the usage with affirms a merge and the end of that block.
1013 The scope depth is not greater than the current parse context scope
1014 nest depth. When the block of that depth closes, the state will
1015 change. To achieve this, all "in scope" variables are linked
1016 together as a stack in nesting order.
1018 - "pending". The "in scope" block has closed, but other parallel
1019 scopes are still being processed. So far, every parallel block at
1020 the same level that has closed has declared the name.
1022 The scope depth is the depth of the last parallel block that
1023 enclosed the declaration, and that has closed.
1025 - "conditionally in scope". The "in scope" block and all parallel
1026 scopes have closed, and no further mention of the name has been
1027 seen. This state includes a secondary nest depth which records the
1028 outermost scope seen since the variable became conditionally in
1029 scope. If a use of the name is found, the variable becomes "in
1030 scope" and that secondary depth becomes the recorded scope depth.
1031 If the name is declared as a new variable, the old variable becomes
1032 "out of scope" and the recorded scope depth stays unchanged.
1034 - "out of scope". The variable is neither in scope nor conditionally
1035 in scope. It is permanently out of scope now and can be removed from
1036 the "in scope" stack.
1038 ###### variable fields
1039 int depth, min_depth;
1040 enum { OutScope, PendingScope, CondScope, InScope } scope;
1041 struct variable *in_scope;
1043 ###### parse context
1045 struct variable *in_scope;
1047 All variables with the same name are linked together using the
1048 'previous' link. Those variable that have
1049 been affirmatively merged all have a 'merged' pointer that points to
1050 one primary variable - the most recently declared instance. When
1051 merging variables, we need to also adjust the 'merged' pointer on any
1052 other variables that had previously been merged with the one that will
1053 no longer be primary.
1055 A variable that is no longer the most recent instance of a name may
1056 still have "pending" scope, if it might still be merged with most
1057 recent instance. These variables don't really belong in the
1058 "in_scope" list, but are not immediately removed when a new instance
1059 is found. Instead, they are detected and ignored when considering the
1060 list of in_scope names.
1062 ###### variable fields
1063 struct variable *merged;
1065 ###### ast functions
1067 static void variable_merge(struct variable *primary, struct variable *secondary)
1071 if (primary->merged)
1073 primary = primary->merged;
1075 for (v = primary->previous; v; v=v->previous)
1076 if (v == secondary || v == secondary->merged ||
1077 v->merged == secondary ||
1078 (v->merged && v->merged == secondary->merged)) {
1079 v->scope = OutScope;
1080 v->merged = primary;
1084 ###### free context vars
1086 while (context.varlist) {
1087 struct binding *b = context.varlist;
1088 struct variable *v = b->var;
1089 context.varlist = b->next;
1092 struct variable *t = v;
1096 if (t->min_depth == 0)
1097 // This is a global constant
1098 free_exec(t->where_decl);
1103 #### Manipulating Bindings
1105 When a name is conditionally visible, a new declaration discards the
1106 old binding - the condition lapses. Conversely a usage of the name
1107 affirms the visibility and extends it to the end of the containing
1108 block - i.e. the block that contains both the original declaration and
1109 the latest usage. This is determined from `min_depth`. When a
1110 conditionally visible variable gets affirmed like this, it is also
1111 merged with other conditionally visible variables with the same name.
1113 When we parse a variable declaration we either report an error if the
1114 name is currently bound, or create a new variable at the current nest
1115 depth if the name is unbound or bound to a conditionally scoped or
1116 pending-scope variable. If the previous variable was conditionally
1117 scoped, it and its homonyms becomes out-of-scope.
1119 When we parse a variable reference (including non-declarative
1120 assignment) we report an error if the name is not bound or is bound to
1121 a pending-scope variable; update the scope if the name is bound to a
1122 conditionally scoped variable; or just proceed normally if the named
1123 variable is in scope.
1125 When we exit a scope, any variables bound at this level are either
1126 marked out of scope or pending-scoped, depending on whether the scope
1127 was sequential or parallel. Here a "parallel" scope means the "then"
1128 or "else" part of a conditional, or any "case" or "else" branch of a
1129 switch. Other scopes are "sequential".
1131 When exiting a parallel scope we check if there are any variables that
1132 were previously pending and are still visible. If there are, then
1133 there weren't redeclared in the most recent scope, so they cannot be
1134 merged and must become out-of-scope. If it is not the first of
1135 parallel scopes (based on `child_count`), we check that there was a
1136 previous binding that is still pending-scope. If there isn't, the new
1137 variable must now be out-of-scope.
1139 When exiting a sequential scope that immediately enclosed parallel
1140 scopes, we need to resolve any pending-scope variables. If there was
1141 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1142 we need to mark all pending-scope variable as out-of-scope. Otherwise
1143 all pending-scope variables become conditionally scoped.
1146 enum closetype { CloseSequential, CloseParallel, CloseElse };
1148 ###### ast functions
1150 static struct variable *var_decl(struct parse_context *c, struct text s)
1152 struct binding *b = find_binding(c, s);
1153 struct variable *v = b->var;
1155 switch (v ? v->scope : OutScope) {
1157 /* Caller will report the error */
1161 v && v->scope == CondScope;
1163 v->scope = OutScope;
1167 v = calloc(1, sizeof(*v));
1168 v->previous = b->var;
1171 v->min_depth = v->depth = c->scope_depth;
1173 v->in_scope = c->in_scope;
1175 v->val = val_prepare(NULL);
1179 static struct variable *var_ref(struct parse_context *c, struct text s)
1181 struct binding *b = find_binding(c, s);
1182 struct variable *v = b->var;
1183 struct variable *v2;
1185 switch (v ? v->scope : OutScope) {
1188 /* Caller will report the error */
1191 /* All CondScope variables of this name need to be merged
1192 * and become InScope
1194 v->depth = v->min_depth;
1196 for (v2 = v->previous;
1197 v2 && v2->scope == CondScope;
1199 variable_merge(v, v2);
1207 static void var_block_close(struct parse_context *c, enum closetype ct)
1209 /* Close off all variables that are in_scope */
1210 struct variable *v, **vp, *v2;
1213 for (vp = &c->in_scope;
1214 v = *vp, v && v->depth > c->scope_depth && v->min_depth > c->scope_depth;
1216 if (v->name->var == v) switch (ct) {
1218 case CloseParallel: /* handle PendingScope */
1222 if (c->scope_stack->child_count == 1)
1223 v->scope = PendingScope;
1224 else if (v->previous &&
1225 v->previous->scope == PendingScope)
1226 v->scope = PendingScope;
1227 else if (v->val.type == Tlabel)
1228 v->scope = PendingScope;
1229 else if (v->name->var == v)
1230 v->scope = OutScope;
1231 if (ct == CloseElse) {
1232 /* All Pending variables with this name
1233 * are now Conditional */
1235 v2 && v2->scope == PendingScope;
1237 v2->scope = CondScope;
1242 v2 && v2->scope == PendingScope;
1244 if (v2->val.type != Tlabel)
1245 v2->scope = OutScope;
1247 case OutScope: break;
1250 case CloseSequential:
1251 if (v->val.type == Tlabel)
1252 v->scope = PendingScope;
1255 v->scope = OutScope;
1258 /* There was no 'else', so we can only become
1259 * conditional if we know the cases were exhaustive,
1260 * and that doesn't mean anything yet.
1261 * So only labels become conditional..
1264 v2 && v2->scope == PendingScope;
1266 if (v2->val.type == Tlabel) {
1267 v2->scope = CondScope;
1268 v2->min_depth = c->scope_depth;
1270 v2->scope = OutScope;
1273 case OutScope: break;
1277 if (v->scope == OutScope || v->name->var != v)
1286 Executables can be lots of different things. In many cases an
1287 executable is just an operation combined with one or two other
1288 executables. This allows for expressions and lists etc. Other times
1289 an executable is something quite specific like a constant or variable
1290 name. So we define a `struct exec` to be a general executable with a
1291 type, and a `struct binode` which is a subclass of `exec`, forms a
1292 node in a binary tree, and holds an operation. There will be other
1293 subclasses, and to access these we need to be able to `cast` the
1294 `exec` into the various other types.
1297 #define cast(structname, pointer) ({ \
1298 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1299 if (__mptr && *__mptr != X##structname) abort(); \
1300 (struct structname *)( (char *)__mptr);})
1302 #define new(structname) ({ \
1303 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1304 __ptr->type = X##structname; \
1305 __ptr->line = -1; __ptr->column = -1; \
1308 #define new_pos(structname, token) ({ \
1309 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1310 __ptr->type = X##structname; \
1311 __ptr->line = token.line; __ptr->column = token.col; \
1320 enum exec_types type;
1328 struct exec *left, *right;
1331 ###### ast functions
1333 static int __fput_loc(struct exec *loc, FILE *f)
1337 if (loc->line >= 0) {
1338 fprintf(f, "%d:%d: ", loc->line, loc->column);
1341 if (loc->type == Xbinode)
1342 return __fput_loc(cast(binode,loc)->left, f) ||
1343 __fput_loc(cast(binode,loc)->right, f);
1346 static void fput_loc(struct exec *loc, FILE *f)
1348 if (!__fput_loc(loc, f))
1349 fprintf(f, "??:??: "); // NOTEST
1352 Each different type of `exec` node needs a number of functions
1353 defined, a bit like methods. We must be able to be able to free it,
1354 print it, analyse it and execute it. Once we have specific `exec`
1355 types we will need to parse them too. Let's take this a bit more
1360 The parser generator requires a `free_foo` function for each struct
1361 that stores attributes and they will often be `exec`s and subtypes
1362 there-of. So we need `free_exec` which can handle all the subtypes,
1363 and we need `free_binode`.
1365 ###### ast functions
1367 static void free_binode(struct binode *b)
1372 free_exec(b->right);
1376 ###### core functions
1377 static void free_exec(struct exec *e)
1386 ###### forward decls
1388 static void free_exec(struct exec *e);
1390 ###### free exec cases
1391 case Xbinode: free_binode(cast(binode, e)); break;
1395 Printing an `exec` requires that we know the current indent level for
1396 printing line-oriented components. As will become clear later, we
1397 also want to know what sort of bracketing to use.
1399 ###### ast functions
1401 static void do_indent(int i, char *str)
1408 ###### core functions
1409 static void print_binode(struct binode *b, int indent, int bracket)
1413 ## print binode cases
1417 static void print_exec(struct exec *e, int indent, int bracket)
1423 print_binode(cast(binode, e), indent, bracket); break;
1428 ###### forward decls
1430 static void print_exec(struct exec *e, int indent, int bracket);
1434 As discussed, analysis involves propagating type requirements around
1435 the program and looking for errors.
1437 So `propagate_types` is passed an expected type (being a `struct type`
1438 pointer together with some `val_rules` flags) that the `exec` is
1439 expected to return, and returns the type that it does return, either
1440 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1441 by reference. It is set to `0` when an error is found, and `2` when
1442 any change is made. If it remains unchanged at `1`, then no more
1443 propagation is needed.
1447 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1451 if (rules & Rnolabel)
1452 fputs(" (labels not permitted)", stderr);
1455 ###### core functions
1457 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1458 struct type *type, int rules);
1459 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1460 struct type *type, int rules)
1467 switch (prog->type) {
1470 struct binode *b = cast(binode, prog);
1472 ## propagate binode cases
1476 ## propagate exec cases
1481 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1482 struct type *type, int rules)
1484 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1493 Interpreting an `exec` doesn't require anything but the `exec`. State
1494 is stored in variables and each variable will be directly linked from
1495 within the `exec` tree. The exception to this is the whole `program`
1496 which needs to look at command line arguments. The `program` will be
1497 interpreted separately.
1499 Each `exec` can return a value, which may be `Tnone` but must be
1500 non-NULL; Some `exec`s will return the location of a value, which can
1501 be updates. To support this, each exec case must store either a value
1502 in `val` or the pointer to a value in `lval`. If `lval` is set, but a
1503 simple value is required, `inter_exec()` will dereference `lval` to
1506 ###### core functions
1509 struct value val, *lval;
1512 static struct lrval _interp_exec(struct exec *e);
1514 static struct value interp_exec(struct exec *e)
1516 struct lrval ret = _interp_exec(e);
1519 return dup_value(*ret.lval);
1524 static struct value *linterp_exec(struct exec *e)
1526 struct lrval ret = _interp_exec(e);
1531 static struct lrval _interp_exec(struct exec *e)
1534 struct value rv, *lrv = NULL;
1545 struct binode *b = cast(binode, e);
1546 struct value left, right, *lleft;
1547 left.type = right.type = Tnone;
1549 ## interp binode cases
1551 free_value(left); free_value(right);
1554 ## interp exec cases
1563 Now that we have the shape of the interpreter in place we can add some
1564 complex types and connected them in to the data structures and the
1565 different phases of parse, analyse, print, interpret.
1567 Thus far we have arrays and structs.
1571 Arrays can be declared by giving a size and a type, as `[size]type' so
1572 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1573 size can be an arbitrary expression which is evaluated when the name
1576 Arrays cannot be assigned. When pointers are introduced we will also
1577 introduce array slices which can refer to part or all of an array -
1578 the assignment syntax will create a slice. For now, an array can only
1579 ever be referenced by the name it is declared with. It is likely that
1580 a "`copy`" primitive will eventually be define which can be used to
1581 make a copy of an array with controllable depth.
1583 ###### type union fields
1587 struct variable *vsize;
1588 struct type *member;
1591 ###### value union fields
1593 struct value *elmnts;
1596 ###### value functions
1598 static struct value array_prepare(struct type *type)
1603 ret.array.elmnts = NULL;
1607 static struct value array_init(struct type *type)
1613 if (type->array.vsize) {
1616 mpz_tdiv_q(q, mpq_numref(type->array.vsize->val.num),
1617 mpq_denref(type->array.vsize->val.num));
1618 type->array.size = mpz_get_si(q);
1621 ret.array.elmnts = calloc(type->array.size,
1622 sizeof(ret.array.elmnts[0]));
1623 for (i = 0; ret.array.elmnts && i < type->array.size; i++)
1624 ret.array.elmnts[i] = val_init(type->array.member);
1628 static void array_free(struct value val)
1632 if (val.array.elmnts)
1633 for (i = 0; i < val.type->array.size; i++)
1634 free_value(val.array.elmnts[i]);
1635 free(val.array.elmnts);
1638 static int array_compat(struct type *require, struct type *have)
1640 if (have->compat != require->compat)
1642 /* Both are arrays, so we can look at details */
1643 if (!type_compat(require->array.member, have->array.member, 0))
1645 if (require->array.vsize == NULL && have->array.vsize == NULL)
1646 return require->array.size == have->array.size;
1648 return require->array.vsize == have->array.vsize;
1651 static void array_print_type(struct type *type, FILE *f)
1654 if (type->array.vsize) {
1655 struct binding *b = type->array.vsize->name;
1656 fprintf(f, "%.*s]", b->name.len, b->name.txt);
1658 fprintf(f, "%d]", type->array.size);
1659 type_print(type->array.member, f);
1662 static struct type array_prototype = {
1663 .prepare = array_prepare,
1665 .print_type = array_print_type,
1666 .compat = array_compat,
1672 | [ NUMBER ] Type ${
1673 $0 = calloc(1, sizeof(struct type));
1674 *($0) = array_prototype;
1675 $0->array.member = $<4;
1676 $0->array.vsize = NULL;
1678 struct parse_context *c = config2context(config);
1681 if (number_parse(num, tail, $2.txt) == 0)
1682 tok_err(c, "error: unrecognised number", &$2);
1684 tok_err(c, "error: unsupported number suffix", &$2);
1686 $0->array.size = mpz_get_ui(mpq_numref(num));
1687 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1688 tok_err(c, "error: array size must be an integer",
1690 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1691 tok_err(c, "error: array size is too large",
1695 $0->next= c->anon_typelist;
1696 c->anon_typelist = $0;
1700 | [ IDENTIFIER ] Type ${ {
1701 struct parse_context *c = config2context(config);
1702 struct variable *v = var_ref(c, $2.txt);
1705 tok_err(config2context(config), "error: name undeclared", &$2);
1706 else if (!v->constant)
1707 tok_err(config2context(config), "error: array size must be a constant", &$2);
1709 $0 = calloc(1, sizeof(struct type));
1710 *($0) = array_prototype;
1711 $0->array.member = $<4;
1713 $0->array.vsize = v;
1714 $0->next= c->anon_typelist;
1715 c->anon_typelist = $0;
1718 ###### parse context
1720 struct type *anon_typelist;
1722 ###### free context types
1724 while (context.anon_typelist) {
1725 struct type *t = context.anon_typelist;
1727 context.anon_typelist = t->next;
1734 ###### variable grammar
1736 | Variable [ Expression ] ${ {
1737 struct binode *b = new(binode);
1744 ###### print binode cases
1746 print_exec(b->left, -1, 0);
1748 print_exec(b->right, -1, 0);
1752 ###### propagate binode cases
1754 /* left must be an array, right must be a number,
1755 * result is the member type of the array
1757 propagate_types(b->right, c, ok, Tnum, 0);
1758 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1759 if (!t || t->compat != array_compat) {
1760 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1763 if (!type_compat(type, t->array.member, rules)) {
1764 type_err(c, "error: have %1 but need %2", prog,
1765 t->array.member, rules, type);
1767 return t->array.member;
1771 ###### interp binode cases
1776 lleft = linterp_exec(b->left);
1777 right = interp_exec(b->right);
1779 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1783 if (i >= 0 && i < lleft->type->array.size)
1784 lrv = &lleft->array.elmnts[i];
1786 rv = val_init(lleft->type->array.member);
1792 A `struct` is a data-type that contains one or more other data-types.
1793 It differs from an array in that each member can be of a different
1794 type, and they are accessed by name rather than by number. Thus you
1795 cannot choose an element by calculation, you need to know what you
1798 The language makes no promises about how a given structure will be
1799 stored in memory - it is free to rearrange fields to suit whatever
1800 criteria seems important.
1802 Structs are declared separately from program code - they cannot be
1803 declared in-line in a variable declaration like arrays can. A struct
1804 is given a name and this name is used to identify the type - the name
1805 is not prefixed by the word `struct` as it would be in C.
1807 Structs are only treated as the same if they have the same name.
1808 Simply having the same fields in the same order is not enough. This
1809 might change once we can create structure initializes from a list of
1812 Each component datum is identified much like a variable is declared,
1813 with a name, one or two colons, and a type. The type cannot be omitted
1814 as there is no opportunity to deduce the type from usage. An initial
1815 value can be given following an equals sign, so
1817 ##### Example: a struct type
1823 would declare a type called "complex" which has two number fields,
1824 each initialised to zero.
1826 Struct will need to be declared separately from the code that uses
1827 them, so we will need to be able to print out the declaration of a
1828 struct when reprinting the whole program. So a `print_type_decl` type
1829 function will be needed.
1831 ###### type union fields
1842 ###### value union fields
1844 struct value *fields;
1847 ###### type functions
1848 void (*print_type_decl)(struct type *type, FILE *f);
1850 ###### value functions
1852 static struct value structure_prepare(struct type *type)
1857 ret.structure.fields = NULL;
1861 static struct value structure_init(struct type *type)
1867 ret.structure.fields = calloc(type->structure.nfields,
1868 sizeof(ret.structure.fields[0]));
1869 for (i = 0; ret.structure.fields && i < type->structure.nfields; i++)
1870 ret.structure.fields[i] = val_init(type->structure.fields[i].type);
1874 static void structure_free(struct value val)
1878 if (val.structure.fields)
1879 for (i = 0; i < val.type->structure.nfields; i++)
1880 free_value(val.structure.fields[i]);
1881 free(val.structure.fields);
1884 static void structure_free_type(struct type *t)
1887 for (i = 0; i < t->structure.nfields; i++)
1888 free_value(t->structure.fields[i].init);
1889 free(t->structure.fields);
1892 static struct type structure_prototype = {
1893 .prepare = structure_prepare,
1894 .init = structure_init,
1895 .free = structure_free,
1896 .free_type = structure_free_type,
1897 .print_type_decl = structure_print_type,
1911 ###### free exec cases
1913 free_exec(cast(fieldref, e)->left);
1917 ###### variable grammar
1919 | Variable . IDENTIFIER ${ {
1920 struct fieldref *fr = new_pos(fieldref, $2);
1927 ###### print exec cases
1931 struct fieldref *f = cast(fieldref, e);
1932 print_exec(f->left, -1, 0);
1933 printf(".%.*s", f->name.len, f->name.txt);
1937 ###### ast functions
1938 static int find_struct_index(struct type *type, struct text field)
1941 for (i = 0; i < type->structure.nfields; i++)
1942 if (text_cmp(type->structure.fields[i].name, field) == 0)
1947 ###### propagate exec cases
1951 struct fieldref *f = cast(fieldref, prog);
1952 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
1955 type_err(c, "error: unknown type for field access", f->left,
1957 else if (st->prepare != structure_prepare)
1958 type_err(c, "error: field reference attempted on %1, not a struct",
1959 f->left, st, 0, NULL);
1960 else if (f->index == -2) {
1961 f->index = find_struct_index(st, f->name);
1963 type_err(c, "error: cannot find requested field in %1",
1964 f->left, st, 0, NULL);
1966 if (f->index >= 0) {
1967 struct type *ft = st->structure.fields[f->index].type;
1968 if (!type_compat(type, ft, rules))
1969 type_err(c, "error: have %1 but need %2", prog,
1976 ###### interp exec cases
1979 struct fieldref *f = cast(fieldref, e);
1980 struct value *lleft = linterp_exec(f->left);
1981 lrv = &lleft->structure.fields[f->index];
1987 struct fieldlist *prev;
1991 ###### ast functions
1992 static void free_fieldlist(struct fieldlist *f)
1996 free_fieldlist(f->prev);
1997 free_value(f->f.init);
2001 ###### top level grammar
2002 DeclareStruct -> struct IDENTIFIER FieldBlock ${ {
2004 add_type(config2context(config), $2.txt, &structure_prototype);
2006 struct fieldlist *f;
2008 for (f = $3; f; f=f->prev)
2011 t->structure.nfields = cnt;
2012 t->structure.fields = calloc(cnt, sizeof(struct field));
2016 t->structure.fields[cnt] = f->f;
2017 f->f.init = val_prepare(Tnone);
2028 FieldBlock -> Open FieldList Close ${ $0 = $<2; }$
2029 | Open SimpleFieldList } ${ $0 = $<2; }$
2030 | : FieldList ${ $0 = $<2; }$
2032 FieldList -> SimpleFieldList NEWLINE ${ $0 = $<1; }$
2033 | FieldList SimpleFieldList NEWLINE ${
2038 SimpleFieldList -> Field ${ $0 = $<1; }$
2039 | SimpleFieldList ; Field ${
2043 | SimpleFieldList ; ${
2047 Field -> IDENTIFIER : Type = Expression ${ {
2050 $0 = calloc(1, sizeof(struct fieldlist));
2051 $0->f.name = $1.txt;
2053 $0->f.init = val_prepare($0->f.type);
2056 propagate_types($<5, config2context(config), &ok, $3, 0);
2059 config2context(config)->parse_error = 1;
2061 $0->f.init = interp_exec($5);
2063 | IDENTIFIER : Type ${
2064 $0 = calloc(1, sizeof(struct fieldlist));
2065 $0->f.name = $1.txt;
2067 $0->f.init = val_init($3);
2070 ###### forward decls
2071 static void structure_print_type(struct type *t, FILE *f);
2073 ###### value functions
2074 static void structure_print_type(struct type *t, FILE *f)
2078 fprintf(f, "struct %.*s:\n", t->name.len, t->name.txt);
2080 for (i = 0; i < t->structure.nfields; i++) {
2081 struct field *fl = t->structure.fields + i;
2082 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2083 type_print(fl->type, f);
2084 if (fl->init.type->print) {
2086 if (fl->init.type == Tstr)
2088 print_value(fl->init);
2089 if (fl->init.type == Tstr)
2096 ###### print type decls
2101 while (target != 0) {
2103 for (t = context.typelist; t ; t=t->next)
2104 if (t->print_type_decl) {
2113 t->print_type_decl(t, stdout);
2119 ## Executables: the elements of code
2121 Each code element needs to be parsed, printed, analysed,
2122 interpreted, and freed. There are several, so let's just start with
2123 the easy ones and work our way up.
2127 We have already met values as separate objects. When manifest
2128 constants appear in the program text, that must result in an executable
2129 which has a constant value. So the `val` structure embeds a value in
2145 $0 = new_pos(val, $1);
2146 $0->val.type = Tbool;
2150 $0 = new_pos(val, $1);
2151 $0->val.type = Tbool;
2155 $0 = new_pos(val, $1);
2156 $0->val.type = Tnum;
2159 if (number_parse($0->val.num, tail, $1.txt) == 0)
2160 mpq_init($0->val.num);
2162 tok_err(config2context(config), "error: unsupported number suffix",
2167 $0 = new_pos(val, $1);
2168 $0->val.type = Tstr;
2171 string_parse(&$1, '\\', &$0->val.str, tail);
2173 tok_err(config2context(config), "error: unsupported string suffix",
2178 $0 = new_pos(val, $1);
2179 $0->val.type = Tstr;
2182 string_parse(&$1, '\\', &$0->val.str, tail);
2184 tok_err(config2context(config), "error: unsupported string suffix",
2189 ###### print exec cases
2192 struct val *v = cast(val, e);
2193 if (v->val.type == Tstr)
2195 print_value(v->val);
2196 if (v->val.type == Tstr)
2201 ###### propagate exec cases
2204 struct val *val = cast(val, prog);
2205 if (!type_compat(type, val->val.type, rules))
2206 type_err(c, "error: expected %1%r found %2",
2207 prog, type, rules, val->val.type);
2208 return val->val.type;
2211 ###### interp exec cases
2213 rv = dup_value(cast(val, e)->val);
2216 ###### ast functions
2217 static void free_val(struct val *v)
2225 ###### free exec cases
2226 case Xval: free_val(cast(val, e)); break;
2228 ###### ast functions
2229 // Move all nodes from 'b' to 'rv', reversing the order.
2230 // In 'b' 'left' is a list, and 'right' is the last node.
2231 // In 'rv', left' is the first node and 'right' is a list.
2232 static struct binode *reorder_bilist(struct binode *b)
2234 struct binode *rv = NULL;
2237 struct exec *t = b->right;
2241 b = cast(binode, b->left);
2251 Just as we used a `val` to wrap a value into an `exec`, we similarly
2252 need a `var` to wrap a `variable` into an exec. While each `val`
2253 contained a copy of the value, each `var` hold a link to the variable
2254 because it really is the same variable no matter where it appears.
2255 When a variable is used, we need to remember to follow the `->merged`
2256 link to find the primary instance.
2264 struct variable *var;
2270 VariableDecl -> IDENTIFIER : ${ {
2271 struct variable *v = var_decl(config2context(config), $1.txt);
2272 $0 = new_pos(var, $1);
2277 v = var_ref(config2context(config), $1.txt);
2279 type_err(config2context(config), "error: variable '%v' redeclared",
2281 type_err(config2context(config), "info: this is where '%v' was first declared",
2282 v->where_decl, NULL, 0, NULL);
2285 | IDENTIFIER :: ${ {
2286 struct variable *v = var_decl(config2context(config), $1.txt);
2287 $0 = new_pos(var, $1);
2293 v = var_ref(config2context(config), $1.txt);
2295 type_err(config2context(config), "error: variable '%v' redeclared",
2297 type_err(config2context(config), "info: this is where '%v' was first declared",
2298 v->where_decl, NULL, 0, NULL);
2301 | IDENTIFIER : Type ${ {
2302 struct variable *v = var_decl(config2context(config), $1.txt);
2303 $0 = new_pos(var, $1);
2308 v->val = val_prepare($<3);
2310 v = var_ref(config2context(config), $1.txt);
2312 type_err(config2context(config), "error: variable '%v' redeclared",
2314 type_err(config2context(config), "info: this is where '%v' was first declared",
2315 v->where_decl, NULL, 0, NULL);
2318 | IDENTIFIER :: Type ${ {
2319 struct variable *v = var_decl(config2context(config), $1.txt);
2320 $0 = new_pos(var, $1);
2325 v->val = val_prepare($<3);
2328 v = var_ref(config2context(config), $1.txt);
2330 type_err(config2context(config), "error: variable '%v' redeclared",
2332 type_err(config2context(config), "info: this is where '%v' was first declared",
2333 v->where_decl, NULL, 0, NULL);
2338 Variable -> IDENTIFIER ${ {
2339 struct variable *v = var_ref(config2context(config), $1.txt);
2340 $0 = new_pos(var, $1);
2342 /* This might be a label - allocate a var just in case */
2343 v = var_decl(config2context(config), $1.txt);
2345 v->val = val_prepare(Tlabel);
2346 v->val.label = &v->val;
2350 cast(var, $0)->var = v;
2355 Type -> IDENTIFIER ${
2356 $0 = find_type(config2context(config), $1.txt);
2358 tok_err(config2context(config),
2359 "error: undefined type", &$1);
2366 ###### print exec cases
2369 struct var *v = cast(var, e);
2371 struct binding *b = v->var->name;
2372 printf("%.*s", b->name.len, b->name.txt);
2379 if (loc->type == Xvar) {
2380 struct var *v = cast(var, loc);
2382 struct binding *b = v->var->name;
2383 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2385 fputs("???", stderr); // NOTEST
2387 fputs("NOTVAR", stderr); // NOTEST
2390 ###### propagate exec cases
2394 struct var *var = cast(var, prog);
2395 struct variable *v = var->var;
2397 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2398 return Tnone; // NOTEST
2402 if (v->constant && (rules & Rnoconstant)) {
2403 type_err(c, "error: Cannot assign to a constant: %v",
2404 prog, NULL, 0, NULL);
2405 type_err(c, "info: name was defined as a constant here",
2406 v->where_decl, NULL, 0, NULL);
2409 if (v->val.type == NULL) {
2410 if (type && *ok != 0) {
2411 v->val = val_prepare(type);
2412 v->where_set = prog;
2417 if (!type_compat(type, v->val.type, rules)) {
2418 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2419 type, rules, v->val.type);
2420 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2421 v->val.type, rules, NULL);
2428 ###### interp exec cases
2431 struct var *var = cast(var, e);
2432 struct variable *v = var->var;
2440 ###### ast functions
2442 static void free_var(struct var *v)
2447 ###### free exec cases
2448 case Xvar: free_var(cast(var, e)); break;
2450 ### Expressions: Conditional
2452 Our first user of the `binode` will be conditional expressions, which
2453 is a bit odd as they actually have three components. That will be
2454 handled by having 2 binodes for each expression. The conditional
2455 expression is the lowest precedence operatior, so it gets to define
2456 what an "Expression" is. The next level up is "BoolExpr", which
2459 Conditional expressions are of the form "value `if` condition `else`
2460 other_value". They associate to the right, so everything to the right
2461 of `else` is part of an else value, while only the BoolExpr to the
2462 left of `if` is the if values. Between `if` and `else` there is no
2463 room for ambiguity, so a full conditional expression is allowed in there.
2471 Expression -> BoolExpr if Expression else Expression ${ {
2472 struct binode *b1 = new(binode);
2473 struct binode *b2 = new(binode);
2482 | BoolExpr ${ $0 = $<1; }$
2484 ###### print binode cases
2487 b2 = cast(binode, b->right);
2488 print_exec(b2->left, -1, 0);
2490 print_exec(b->left, -1, 0);
2492 print_exec(b2->right, -1, 0);
2495 ###### propagate binode cases
2498 /* cond must be Tbool, others must match */
2499 struct binode *b2 = cast(binode, b->right);
2502 propagate_types(b->left, c, ok, Tbool, 0);
2503 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2504 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2508 ###### interp binode cases
2511 struct binode *b2 = cast(binode, b->right);
2512 left = interp_exec(b->left);
2514 rv = interp_exec(b2->left);
2516 rv = interp_exec(b2->right);
2520 ### Expressions: Boolean
2522 The next class of expressions to use the `binode` will be Boolean
2523 expressions. As I haven't implemented precedence in the parser
2524 generator yet, we need different names for each precedence level used
2525 by expressions. The outer most or lowest level precedence after
2526 conditional expressions are Boolean operators which form an `BoolExpr`
2527 out of `BTerm`s and `BFact`s. As well as `or` `and`, and `not` we
2528 have `and then` and `or else` which only evaluate the second operand
2529 if the result would make a difference.
2541 BoolExpr -> BoolExpr or BTerm ${ {
2542 struct binode *b = new(binode);
2548 | BoolExpr or else BTerm ${ {
2549 struct binode *b = new(binode);
2555 | BTerm ${ $0 = $<1; }$
2557 BTerm -> BTerm and BFact ${ {
2558 struct binode *b = new(binode);
2564 | BTerm and then BFact ${ {
2565 struct binode *b = new(binode);
2571 | BFact ${ $0 = $<1; }$
2573 BFact -> not BFact ${ {
2574 struct binode *b = new(binode);
2581 ###### print binode cases
2583 print_exec(b->left, -1, 0);
2585 print_exec(b->right, -1, 0);
2588 print_exec(b->left, -1, 0);
2589 printf(" and then ");
2590 print_exec(b->right, -1, 0);
2593 print_exec(b->left, -1, 0);
2595 print_exec(b->right, -1, 0);
2598 print_exec(b->left, -1, 0);
2599 printf(" or else ");
2600 print_exec(b->right, -1, 0);
2604 print_exec(b->right, -1, 0);
2607 ###### propagate binode cases
2613 /* both must be Tbool, result is Tbool */
2614 propagate_types(b->left, c, ok, Tbool, 0);
2615 propagate_types(b->right, c, ok, Tbool, 0);
2616 if (type && type != Tbool)
2617 type_err(c, "error: %1 operation found where %2 expected", prog,
2621 ###### interp binode cases
2623 rv = interp_exec(b->left);
2624 right = interp_exec(b->right);
2625 rv.bool = rv.bool && right.bool;
2628 rv = interp_exec(b->left);
2630 rv = interp_exec(b->right);
2633 rv = interp_exec(b->left);
2634 right = interp_exec(b->right);
2635 rv.bool = rv.bool || right.bool;
2638 rv = interp_exec(b->left);
2640 rv = interp_exec(b->right);
2643 rv = interp_exec(b->right);
2647 ### Expressions: Comparison
2649 Of slightly higher precedence that Boolean expressions are
2651 A comparison takes arguments of any comparable type, but the two types must be
2654 To simplify the parsing we introduce an `eop` which can record an
2655 expression operator.
2662 ###### ast functions
2663 static void free_eop(struct eop *e)
2678 | Expr CMPop Expr ${ {
2679 struct binode *b = new(binode);
2685 | Expr ${ $0 = $<1; }$
2690 CMPop -> < ${ $0.op = Less; }$
2691 | > ${ $0.op = Gtr; }$
2692 | <= ${ $0.op = LessEq; }$
2693 | >= ${ $0.op = GtrEq; }$
2694 | == ${ $0.op = Eql; }$
2695 | != ${ $0.op = NEql; }$
2697 ###### print binode cases
2705 print_exec(b->left, -1, 0);
2707 case Less: printf(" < "); break;
2708 case LessEq: printf(" <= "); break;
2709 case Gtr: printf(" > "); break;
2710 case GtrEq: printf(" >= "); break;
2711 case Eql: printf(" == "); break;
2712 case NEql: printf(" != "); break;
2713 default: abort(); // NOTEST
2715 print_exec(b->right, -1, 0);
2718 ###### propagate binode cases
2725 /* Both must match but not be labels, result is Tbool */
2726 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2728 propagate_types(b->right, c, ok, t, 0);
2730 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2732 t = propagate_types(b->left, c, ok, t, 0);
2734 if (!type_compat(type, Tbool, 0))
2735 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2736 Tbool, rules, type);
2739 ###### interp binode cases
2748 left = interp_exec(b->left);
2749 right = interp_exec(b->right);
2750 cmp = value_cmp(left, right);
2753 case Less: rv.bool = cmp < 0; break;
2754 case LessEq: rv.bool = cmp <= 0; break;
2755 case Gtr: rv.bool = cmp > 0; break;
2756 case GtrEq: rv.bool = cmp >= 0; break;
2757 case Eql: rv.bool = cmp == 0; break;
2758 case NEql: rv.bool = cmp != 0; break;
2759 default: rv.bool = 0; break; // NOTEST
2764 ### Expressions: The rest
2766 The remaining expressions with the highest precedence are arithmetic
2767 and string concatenation. They are `Expr`, `Term`, and `Factor`.
2768 The `Factor` is where the `Value` and `Variable` that we already have
2771 `+` and `-` are both infix and prefix operations (where they are
2772 absolute value and negation). These have different operator names.
2774 We also have a 'Bracket' operator which records where parentheses were
2775 found. This makes it easy to reproduce these when printing. Once
2776 precedence is handled better I might be able to discard this.
2788 Expr -> Expr Eop Term ${ {
2789 struct binode *b = new(binode);
2795 | Term ${ $0 = $<1; }$
2797 Term -> Term Top Factor ${ {
2798 struct binode *b = new(binode);
2804 | Factor ${ $0 = $<1; }$
2806 Factor -> ( Expression ) ${ {
2807 struct binode *b = new_pos(binode, $1);
2813 struct binode *b = new(binode);
2818 | Value ${ $0 = $<1; }$
2819 | Variable ${ $0 = $<1; }$
2822 Eop -> + ${ $0.op = Plus; }$
2823 | - ${ $0.op = Minus; }$
2825 Uop -> + ${ $0.op = Absolute; }$
2826 | - ${ $0.op = Negate; }$
2828 Top -> * ${ $0.op = Times; }$
2829 | / ${ $0.op = Divide; }$
2830 | % ${ $0.op = Rem; }$
2831 | ++ ${ $0.op = Concat; }$
2833 ###### print binode cases
2840 print_exec(b->left, indent, 0);
2842 case Plus: fputs(" + ", stdout); break;
2843 case Minus: fputs(" - ", stdout); break;
2844 case Times: fputs(" * ", stdout); break;
2845 case Divide: fputs(" / ", stdout); break;
2846 case Rem: fputs(" % ", stdout); break;
2847 case Concat: fputs(" ++ ", stdout); break;
2848 default: abort(); // NOTEST
2850 print_exec(b->right, indent, 0);
2854 print_exec(b->right, indent, 0);
2858 print_exec(b->right, indent, 0);
2862 print_exec(b->right, indent, 0);
2866 ###### propagate binode cases
2872 /* both must be numbers, result is Tnum */
2875 /* as propagate_types ignores a NULL,
2876 * unary ops fit here too */
2877 propagate_types(b->left, c, ok, Tnum, 0);
2878 propagate_types(b->right, c, ok, Tnum, 0);
2879 if (!type_compat(type, Tnum, 0))
2880 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
2885 /* both must be Tstr, result is Tstr */
2886 propagate_types(b->left, c, ok, Tstr, 0);
2887 propagate_types(b->right, c, ok, Tstr, 0);
2888 if (!type_compat(type, Tstr, 0))
2889 type_err(c, "error: Concat returns %1 but %2 expected", prog,
2894 return propagate_types(b->right, c, ok, type, 0);
2896 ###### interp binode cases
2899 rv = interp_exec(b->left);
2900 right = interp_exec(b->right);
2901 mpq_add(rv.num, rv.num, right.num);
2904 rv = interp_exec(b->left);
2905 right = interp_exec(b->right);
2906 mpq_sub(rv.num, rv.num, right.num);
2909 rv = interp_exec(b->left);
2910 right = interp_exec(b->right);
2911 mpq_mul(rv.num, rv.num, right.num);
2914 rv = interp_exec(b->left);
2915 right = interp_exec(b->right);
2916 mpq_div(rv.num, rv.num, right.num);
2921 left = interp_exec(b->left);
2922 right = interp_exec(b->right);
2923 mpz_init(l); mpz_init(r); mpz_init(rem);
2924 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
2925 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
2926 mpz_tdiv_r(rem, l, r);
2927 rv = val_init(Tnum);
2928 mpq_set_z(rv.num, rem);
2929 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
2933 rv = interp_exec(b->right);
2934 mpq_neg(rv.num, rv.num);
2937 rv = interp_exec(b->right);
2938 mpq_abs(rv.num, rv.num);
2941 rv = interp_exec(b->right);
2944 left = interp_exec(b->left);
2945 right = interp_exec(b->right);
2947 rv.str = text_join(left.str, right.str);
2950 ###### value functions
2952 static struct text text_join(struct text a, struct text b)
2955 rv.len = a.len + b.len;
2956 rv.txt = malloc(rv.len);
2957 memcpy(rv.txt, a.txt, a.len);
2958 memcpy(rv.txt+a.len, b.txt, b.len);
2962 ### Blocks, Statements, and Statement lists.
2964 Now that we have expressions out of the way we need to turn to
2965 statements. There are simple statements and more complex statements.
2966 Simple statements do not contain (syntactic) newlines, complex statements do.
2968 Statements often come in sequences and we have corresponding simple
2969 statement lists and complex statement lists.
2970 The former comprise only simple statements separated by semicolons.
2971 The later comprise complex statements and simple statement lists. They are
2972 separated by newlines. Thus the semicolon is only used to separate
2973 simple statements on the one line. This may be overly restrictive,
2974 but I'm not sure I ever want a complex statement to share a line with
2977 Note that a simple statement list can still use multiple lines if
2978 subsequent lines are indented, so
2980 ###### Example: wrapped simple statement list
2985 is a single simple statement list. This might allow room for
2986 confusion, so I'm not set on it yet.
2988 A simple statement list needs no extra syntax. A complex statement
2989 list has two syntactic forms. It can be enclosed in braces (much like
2990 C blocks), or it can be introduced by a colon and continue until an
2991 unindented newline (much like Python blocks). With this extra syntax
2992 it is referred to as a block.
2994 Note that a block does not have to include any newlines if it only
2995 contains simple statements. So both of:
2997 if condition: a=b; d=f
2999 if condition { a=b; print f }
3003 In either case the list is constructed from a `binode` list with
3004 `Block` as the operator. When parsing the list it is most convenient
3005 to append to the end, so a list is a list and a statement. When using
3006 the list it is more convenient to consider a list to be a statement
3007 and a list. So we need a function to re-order a list.
3008 `reorder_bilist` serves this purpose.
3010 The only stand-alone statement we introduce at this stage is `pass`
3011 which does nothing and is represented as a `NULL` pointer in a `Block`
3012 list. Other stand-alone statements will follow once the infrastructure
3025 Block -> Open Statementlist Close ${ $0 = $<2; }$
3026 | Open SimpleStatements } ${ $0 = reorder_bilist($<2); }$
3027 | : SimpleStatements ${ $0 = reorder_bilist($<2); }$
3028 | : Statementlist ${ $0 = $<2; }$
3030 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<1); }$
3032 ComplexStatements -> ComplexStatements ComplexStatement ${
3042 | ComplexStatement ${
3054 ComplexStatement -> SimpleStatements NEWLINE ${
3055 $0 = reorder_bilist($<1);
3057 | Newlines ${ $0 = NULL; }$
3058 ## ComplexStatement Grammar
3061 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3067 | SimpleStatement ${
3073 | SimpleStatements ; ${ $0 = $<1; }$
3075 SimpleStatement -> pass ${ $0 = NULL; }$
3076 ## SimpleStatement Grammar
3078 ###### print binode cases
3082 if (b->left == NULL)
3085 print_exec(b->left, indent, 0);
3088 print_exec(b->right, indent, 0);
3091 // block, one per line
3092 if (b->left == NULL)
3093 do_indent(indent, "pass\n");
3095 print_exec(b->left, indent, bracket);
3097 print_exec(b->right, indent, bracket);
3101 ###### propagate binode cases
3104 /* If any statement returns something other than Tnone
3105 * or Tbool then all such must return same type.
3106 * As each statement may be Tnone or something else,
3107 * we must always pass NULL (unknown) down, otherwise an incorrect
3108 * error might occur. We never return Tnone unless it is
3113 for (e = b; e; e = cast(binode, e->right)) {
3114 t = propagate_types(e->left, c, ok, NULL, rules);
3115 if ((rules & Rboolok) && t == Tbool)
3117 if (t && t != Tnone && t != Tbool) {
3121 type_err(c, "error: expected %1%r, found %2",
3122 e->left, type, rules, t);
3128 ###### interp binode cases
3130 while (rv.type == Tnone &&
3133 rv = interp_exec(b->left);
3134 b = cast(binode, b->right);
3138 ### The Print statement
3140 `print` is a simple statement that takes a comma-separated list of
3141 expressions and prints the values separated by spaces and terminated
3142 by a newline. No control of formatting is possible.
3144 `print` faces the same list-ordering issue as blocks, and uses the
3150 ###### SimpleStatement Grammar
3152 | print ExpressionList ${
3153 $0 = reorder_bilist($<2);
3155 | print ExpressionList , ${
3160 $0 = reorder_bilist($0);
3171 ExpressionList -> ExpressionList , Expression ${
3184 ###### print binode cases
3187 do_indent(indent, "print");
3191 print_exec(b->left, -1, 0);
3195 b = cast(binode, b->right);
3201 ###### propagate binode cases
3204 /* don't care but all must be consistent */
3205 propagate_types(b->left, c, ok, NULL, Rnolabel);
3206 propagate_types(b->right, c, ok, NULL, Rnolabel);
3209 ###### interp binode cases
3215 for ( ; b; b = cast(binode, b->right))
3219 left = interp_exec(b->left);
3232 ###### Assignment statement
3234 An assignment will assign a value to a variable, providing it hasn't
3235 be declared as a constant. The analysis phase ensures that the type
3236 will be correct so the interpreter just needs to perform the
3237 calculation. There is a form of assignment which declares a new
3238 variable as well as assigning a value. If a name is assigned before
3239 it is declared, and error will be raised as the name is created as
3240 `Tlabel` and it is illegal to assign to such names.
3246 ###### SimpleStatement Grammar
3247 | Variable = Expression ${
3253 | VariableDecl = Expression ${
3261 if ($1->var->where_set == NULL) {
3262 type_err(config2context(config),
3263 "Variable declared with no type or value: %v",
3273 ###### print binode cases
3276 do_indent(indent, "");
3277 print_exec(b->left, indent, 0);
3279 print_exec(b->right, indent, 0);
3286 struct variable *v = cast(var, b->left)->var;
3287 do_indent(indent, "");
3288 print_exec(b->left, indent, 0);
3289 if (cast(var, b->left)->var->constant) {
3290 if (v->where_decl == v->where_set) {
3292 type_print(v->val.type, stdout);
3297 if (v->where_decl == v->where_set) {
3299 type_print(v->val.type, stdout);
3306 print_exec(b->right, indent, 0);
3313 ###### propagate binode cases
3317 /* Both must match and not be labels,
3318 * Type must support 'dup',
3319 * For Assign, left must not be constant.
3322 t = propagate_types(b->left, c, ok, NULL,
3323 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3328 if (propagate_types(b->right, c, ok, t, 0) != t)
3329 if (b->left->type == Xvar)
3330 type_err(c, "info: variable '%v' was set as %1 here.",
3331 cast(var, b->left)->var->where_set, t, rules, NULL);
3333 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3335 propagate_types(b->left, c, ok, t,
3336 (b->op == Assign ? Rnoconstant : 0));
3338 if (t && t->dup == NULL)
3339 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3344 ###### interp binode cases
3347 lleft = linterp_exec(b->left);
3348 right = interp_exec(b->right);
3353 free_value(right); // NOTEST
3359 struct variable *v = cast(var, b->left)->var;
3363 right = interp_exec(b->right);
3365 right = val_init(v->val.type);
3372 ### The `use` statement
3374 The `use` statement is the last "simple" statement. It is needed when
3375 the condition in a conditional statement is a block. `use` works much
3376 like `return` in C, but only completes the `condition`, not the whole
3382 ###### SimpleStatement Grammar
3384 $0 = new_pos(binode, $1);
3389 ###### print binode cases
3392 do_indent(indent, "use ");
3393 print_exec(b->right, -1, 0);
3398 ###### propagate binode cases
3401 /* result matches value */
3402 return propagate_types(b->right, c, ok, type, 0);
3404 ###### interp binode cases
3407 rv = interp_exec(b->right);
3410 ### The Conditional Statement
3412 This is the biggy and currently the only complex statement. This
3413 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3414 It is comprised of a number of parts, all of which are optional though
3415 set combinations apply. Each part is (usually) a key word (`then` is
3416 sometimes optional) followed by either an expression or a code block,
3417 except the `casepart` which is a "key word and an expression" followed
3418 by a code block. The code-block option is valid for all parts and,
3419 where an expression is also allowed, the code block can use the `use`
3420 statement to report a value. If the code block does not report a value
3421 the effect is similar to reporting `True`.
3423 The `else` and `case` parts, as well as `then` when combined with
3424 `if`, can contain a `use` statement which will apply to some
3425 containing conditional statement. `for` parts, `do` parts and `then`
3426 parts used with `for` can never contain a `use`, except in some
3427 subordinate conditional statement.
3429 If there is a `forpart`, it is executed first, only once.
3430 If there is a `dopart`, then it is executed repeatedly providing
3431 always that the `condpart` or `cond`, if present, does not return a non-True
3432 value. `condpart` can fail to return any value if it simply executes
3433 to completion. This is treated the same as returning `True`.
3435 If there is a `thenpart` it will be executed whenever the `condpart`
3436 or `cond` returns True (or does not return any value), but this will happen
3437 *after* `dopart` (when present).
3439 If `elsepart` is present it will be executed at most once when the
3440 condition returns `False` or some value that isn't `True` and isn't
3441 matched by any `casepart`. If there are any `casepart`s, they will be
3442 executed when the condition returns a matching value.
3444 The particular sorts of values allowed in case parts has not yet been
3445 determined in the language design, so nothing is prohibited.
3447 The various blocks in this complex statement potentially provide scope
3448 for variables as described earlier. Each such block must include the
3449 "OpenScope" nonterminal before parsing the block, and must call
3450 `var_block_close()` when closing the block.
3452 The code following "`if`", "`switch`" and "`for`" does not get its own
3453 scope, but is in a scope covering the whole statement, so names
3454 declared there cannot be redeclared elsewhere. Similarly the
3455 condition following "`while`" is in a scope the covers the body
3456 ("`do`" part) of the loop, and which does not allow conditional scope
3457 extension. Code following "`then`" (both looping and non-looping),
3458 "`else`" and "`case`" each get their own local scope.
3460 The type requirements on the code block in a `whilepart` are quite
3461 unusal. It is allowed to return a value of some identifiable type, in
3462 which case the loop aborts and an appropriate `casepart` is run, or it
3463 can return a Boolean, in which case the loop either continues to the
3464 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3465 This is different both from the `ifpart` code block which is expected to
3466 return a Boolean, or the `switchpart` code block which is expected to
3467 return the same type as the casepart values. The correct analysis of
3468 the type of the `whilepart` code block is the reason for the
3469 `Rboolok` flag which is passed to `propagate_types()`.
3471 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3480 struct exec *action;
3481 struct casepart *next;
3483 struct cond_statement {
3485 struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
3486 struct casepart *casepart;
3489 ###### ast functions
3491 static void free_casepart(struct casepart *cp)
3495 free_exec(cp->value);
3496 free_exec(cp->action);
3503 static void free_cond_statement(struct cond_statement *s)
3507 free_exec(s->forpart);
3508 free_exec(s->condpart);
3509 free_exec(s->dopart);
3510 free_exec(s->thenpart);
3511 free_exec(s->elsepart);
3512 free_casepart(s->casepart);
3516 ###### free exec cases
3517 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3519 ###### ComplexStatement Grammar
3520 | CondStatement ${ $0 = $<1; }$
3525 // both ForThen and Whilepart open scopes, and CondSuffix only
3526 // closes one - so in the first branch here we have another to close.
3527 CondStatement -> forPart ThenPart WhilePart CondSuffix ${
3531 $0->condpart = $3.condpart; $3.condpart = NULL;
3532 $0->dopart = $3.dopart; $3.dopart = NULL;
3533 var_block_close(config2context(config), CloseSequential);
3535 | forPart WhilePart CondSuffix ${
3538 $0->thenpart = NULL;
3539 $0->condpart = $2.condpart; $2.condpart = NULL;
3540 $0->dopart = $2.dopart; $2.dopart = NULL;
3541 var_block_close(config2context(config), CloseSequential);
3543 | whilePart CondSuffix ${
3545 $0->condpart = $1.condpart; $1.condpart = NULL;
3546 $0->dopart = $1.dopart; $1.dopart = NULL;
3548 | switchPart CondSuffix ${
3552 | ifPart IfSuffix ${
3554 $0->condpart = $1.condpart; $1.condpart = NULL;
3555 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
3556 // This is where we close an "if" statement
3557 var_block_close(config2context(config), CloseSequential);
3560 CondSuffix -> IfSuffix ${
3562 // This is where we close scope of the whole
3563 // "for" or "while" statement
3564 var_block_close(config2context(config), CloseSequential);
3566 | CasePart CondSuffix ${
3568 $1->next = $0->casepart;
3576 CasePart -> Case Expression OpenScope Block ${
3577 $0 = calloc(1,sizeof(struct casepart));
3580 var_block_close(config2context(config), CloseParallel);
3584 IfSuffix -> ${ $0 = new(cond_statement); }$
3585 | NEWLINE IfSuffix ${ $0 = $<2; }$
3586 | else OpenScope Block ${
3587 $0 = new(cond_statement);
3589 var_block_close(config2context(config), CloseElse);
3591 | else OpenScope CondStatement ${
3592 $0 = new(cond_statement);
3594 var_block_close(config2context(config), CloseElse);
3605 // These scopes are closed in CondSuffix
3606 forPart -> for OpenScope SimpleStatements ${
3607 $0 = reorder_bilist($<3);
3609 | for OpenScope Block ${
3613 ThenPart -> Then OpenScope SimpleStatements ${
3614 $0 = reorder_bilist($<3);
3615 var_block_close(config2context(config), CloseSequential);
3617 | Then OpenScope Block ${
3619 var_block_close(config2context(config), CloseSequential);
3622 // This scope is closed in CondSuffix
3623 WhileHead -> While OpenScope Block ${
3626 whileHead -> while OpenScope Block ${
3631 // This scope is closed in CondSuffix
3632 whilePart -> while OpenScope Expression Block ${
3633 $0.type = Xcond_statement;
3637 | whileHead Do Block ${
3638 $0.type = Xcond_statement;
3642 WhilePart -> While OpenScope Expression Block ${
3643 $0.type = Xcond_statement;
3647 | WhileHead Do Block ${
3648 $0.type = Xcond_statement;
3653 ifPart -> if OpenScope Expression OpenScope Block ${
3654 $0.type = Xcond_statement;
3657 var_block_close(config2context(config), CloseParallel);
3659 | if OpenScope Block Then OpenScope Block ${
3660 $0.type = Xcond_statement;
3663 var_block_close(config2context(config), CloseParallel);
3667 // This scope is closed in CondSuffix
3668 switchPart -> switch OpenScope Expression ${
3671 | switch OpenScope Block ${
3675 ###### print exec cases
3677 case Xcond_statement:
3679 struct cond_statement *cs = cast(cond_statement, e);
3680 struct casepart *cp;
3682 do_indent(indent, "for");
3683 if (bracket) printf(" {\n"); else printf(":\n");
3684 print_exec(cs->forpart, indent+1, bracket);
3687 do_indent(indent, "} then {\n");
3689 do_indent(indent, "then:\n");
3690 print_exec(cs->thenpart, indent+1, bracket);
3692 if (bracket) do_indent(indent, "}\n");
3696 if (cs->condpart && cs->condpart->type == Xbinode &&
3697 cast(binode, cs->condpart)->op == Block) {
3699 do_indent(indent, "while {\n");
3701 do_indent(indent, "while:\n");
3702 print_exec(cs->condpart, indent+1, bracket);
3704 do_indent(indent, "} do {\n");
3706 do_indent(indent, "do:\n");
3707 print_exec(cs->dopart, indent+1, bracket);
3709 do_indent(indent, "}\n");
3711 do_indent(indent, "while ");
3712 print_exec(cs->condpart, 0, bracket);
3717 print_exec(cs->dopart, indent+1, bracket);
3719 do_indent(indent, "}\n");
3724 do_indent(indent, "switch");
3726 do_indent(indent, "if");
3727 if (cs->condpart && cs->condpart->type == Xbinode &&
3728 cast(binode, cs->condpart)->op == Block) {
3733 print_exec(cs->condpart, indent+1, bracket);
3735 do_indent(indent, "}\n");
3737 do_indent(indent, "then:\n");
3738 print_exec(cs->thenpart, indent+1, bracket);
3742 print_exec(cs->condpart, 0, bracket);
3748 print_exec(cs->thenpart, indent+1, bracket);
3750 do_indent(indent, "}\n");
3755 for (cp = cs->casepart; cp; cp = cp->next) {
3756 do_indent(indent, "case ");
3757 print_exec(cp->value, -1, 0);
3762 print_exec(cp->action, indent+1, bracket);
3764 do_indent(indent, "}\n");
3767 do_indent(indent, "else");
3772 print_exec(cs->elsepart, indent+1, bracket);
3774 do_indent(indent, "}\n");
3779 ###### propagate exec cases
3780 case Xcond_statement:
3782 // forpart and dopart must return Tnone
3783 // thenpart must return Tnone if there is a dopart,
3784 // otherwise it is like elsepart.
3786 // be bool if there is no casepart
3787 // match casepart->values if there is a switchpart
3788 // either be bool or match casepart->value if there
3790 // elsepart and casepart->action must match the return type
3791 // expected of this statement.
3792 struct cond_statement *cs = cast(cond_statement, prog);
3793 struct casepart *cp;
3795 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
3796 if (!type_compat(Tnone, t, 0))
3798 t = propagate_types(cs->dopart, c, ok, Tnone, 0);
3799 if (!type_compat(Tnone, t, 0))
3802 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
3803 if (!type_compat(Tnone, t, 0))
3806 if (cs->casepart == NULL)
3807 propagate_types(cs->condpart, c, ok, Tbool, 0);
3809 /* Condpart must match case values, with bool permitted */
3811 for (cp = cs->casepart;
3812 cp && !t; cp = cp->next)
3813 t = propagate_types(cp->value, c, ok, NULL, 0);
3814 if (!t && cs->condpart)
3815 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok);
3816 // Now we have a type (I hope) push it down
3818 for (cp = cs->casepart; cp; cp = cp->next)
3819 propagate_types(cp->value, c, ok, t, 0);
3820 propagate_types(cs->condpart, c, ok, t, Rboolok);
3823 // (if)then, else, and case parts must return expected type.
3824 if (!cs->dopart && !type)
3825 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
3827 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
3828 for (cp = cs->casepart;
3831 type = propagate_types(cp->action, c, ok, NULL, rules);
3834 propagate_types(cs->thenpart, c, ok, type, rules);
3835 propagate_types(cs->elsepart, c, ok, type, rules);
3836 for (cp = cs->casepart; cp ; cp = cp->next)
3837 propagate_types(cp->action, c, ok, type, rules);
3843 ###### interp exec cases
3844 case Xcond_statement:
3846 struct value v, cnd;
3847 struct casepart *cp;
3848 struct cond_statement *c = cast(cond_statement, e);
3851 interp_exec(c->forpart);
3854 cnd = interp_exec(c->condpart);
3857 if (!(cnd.type == Tnone ||
3858 (cnd.type == Tbool && cnd.bool != 0)))
3860 // cnd is Tnone or Tbool, doesn't need to be freed
3862 interp_exec(c->dopart);
3865 rv = interp_exec(c->thenpart);
3866 if (rv.type != Tnone || !c->dopart)
3870 } while (c->dopart);
3872 for (cp = c->casepart; cp; cp = cp->next) {
3873 v = interp_exec(cp->value);
3874 if (value_cmp(v, cnd) == 0) {
3877 rv = interp_exec(cp->action);
3884 rv = interp_exec(c->elsepart);
3891 ### Top level structure
3893 All the language elements so far can be used in various places. Now
3894 it is time to clarify what those places are.
3896 At the top level of a file there will be a number of declarations.
3897 Many of the things that can be declared haven't been described yet,
3898 such as functions, procedures, imports, and probably more.
3899 For now there are two sorts of things that can appear at the top
3900 level. They are predefined constants, `struct` types, and the main
3901 program. While the syntax will allow the main program to appear
3902 multiple times, that will trigger an error if it is actually attempted.
3904 The various declarations do not return anything. They store the
3905 various declarations in the parse context.
3907 ###### Parser: grammar
3910 Ocean -> DeclarationList
3912 DeclarationList -> Declaration
3913 | DeclarationList Declaration
3915 Declaration -> DeclareConstant
3920 ## top level grammar
3922 ### The `const` section
3924 As well as being defined in with the code that uses them, constants
3925 can be declared at the top level. These have full-file scope, so they
3926 are always `InScope`. The value of a top level constant can be given
3927 as an expression, and this is evaluated immediately rather than in the
3928 later interpretation stage. Once we add functions to the language, we
3929 will need rules concern which, if any, can be used to define a top
3932 Constants are defined in a section that starts with the reserved word
3933 `const` and then has a block with a list of assignment statements.
3934 For syntactic consistency, these must use the double-colon syntax to
3935 make it clear that they are constants. Type can also be given: if
3936 not, the type will be determined during analysis, as with other
3939 As the types constants are inserted at the head of a list, printing
3940 them in the same order that they were read is not straight forward.
3941 We take a quadratic approach here and count the number of constants
3942 (variables of depth 0), then count down from there, each time
3943 searching through for the Nth constant for decreasing N.
3945 ###### top level grammar
3947 DeclareConstant -> const Open ConstList Close
3948 | const Open SimpleConstList }
3950 | const SimpleConstList NEWLINE
3952 ConstList -> ComplexConsts
3954 ComplexConsts -> ComplexConst ComplexConsts
3956 ComplexConst -> SimpleConstList NEWLINE
3957 SimpleConstList -> SimpleConstList ; Const
3962 CType -> Type ${ $0 = $<1; }$
3965 Const -> IDENTIFIER :: CType = Expression ${ {
3969 v = var_decl(config2context(config), $1.txt);
3971 struct var *var = new_pos(var, $1);
3972 v->where_decl = var;
3977 v = var_ref(config2context(config), $1.txt);
3978 tok_err(config2context(config), "error: name already declared", &$1);
3979 type_err(config2context(config), "info: this is where '%v' was first declared",
3980 v->where_decl, NULL, 0, NULL);
3984 propagate_types($5, config2context(config), &ok, $3, 0);
3987 config2context(config)->parse_error = 1;
3989 v->val = interp_exec($5);
3993 ###### print const decls
3998 while (target != 0) {
4000 for (v = context.in_scope; v; v=v->in_scope)
4001 if (v->depth == 0) {
4012 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4013 type_print(v->val.type, stdout);
4015 if (v->val.type == Tstr)
4017 print_value(v->val);
4018 if (v->val.type == Tstr)
4026 ### Finally the whole program.
4028 Somewhat reminiscent of Pascal a (current) Ocean program starts with
4029 the keyword "program" and a list of variable names which are assigned
4030 values from command line arguments. Following this is a `block` which
4031 is the code to execute. Unlike Pascal, constants and other
4032 declarations come *before* the program.
4034 As this is the top level, several things are handled a bit
4036 The whole program is not interpreted by `interp_exec` as that isn't
4037 passed the argument list which the program requires. Similarly type
4038 analysis is a bit more interesting at this level.
4043 ###### top level grammar
4045 DeclareProgram -> Program ${ {
4046 struct parse_context *c = config2context(config);
4048 type_err(c, "Program defined a second time",
4055 Program -> program OpenScope Varlist Block ${
4058 $0->left = reorder_bilist($<3);
4060 var_block_close(config2context(config), CloseSequential);
4061 if (config2context(config)->scope_stack) abort();
4064 tok_err(config2context(config),
4065 "error: unhandled parse error", &$1);
4068 Varlist -> Varlist ArgDecl ${
4077 ArgDecl -> IDENTIFIER ${ {
4078 struct variable *v = var_decl(config2context(config), $1.txt);
4085 ###### print binode cases
4087 do_indent(indent, "program");
4088 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4090 print_exec(b2->left, 0, 0);
4096 print_exec(b->right, indent+1, bracket);
4098 do_indent(indent, "}\n");
4101 ###### propagate binode cases
4102 case Program: abort(); // NOTEST
4104 ###### core functions
4106 static int analyse_prog(struct exec *prog, struct parse_context *c)
4108 struct binode *b = cast(binode, prog);
4115 propagate_types(b->right, c, &ok, Tnone, 0);
4120 for (b = cast(binode, b->left); b; b = cast(binode, b->right)) {
4121 struct var *v = cast(var, b->left);
4122 if (!v->var->val.type) {
4123 v->var->where_set = b;
4124 v->var->val = val_prepare(Tstr);
4127 b = cast(binode, prog);
4130 propagate_types(b->right, c, &ok, Tnone, 0);
4135 /* Make sure everything is still consistent */
4136 propagate_types(b->right, c, &ok, Tnone, 0);
4140 static void interp_prog(struct exec *prog, char **argv)
4142 struct binode *p = cast(binode, prog);
4148 al = cast(binode, p->left);
4150 struct var *v = cast(var, al->left);
4151 struct value *vl = &v->var->val;
4153 if (argv[0] == NULL) {
4154 printf("Not enough args\n");
4157 al = cast(binode, al->right);
4159 *vl = parse_value(vl->type, argv[0]);
4160 if (vl->type == NULL)
4164 v = interp_exec(p->right);
4168 ###### interp binode cases
4169 case Program: abort(); // NOTEST
4171 ## And now to test it out.
4173 Having a language requires having a "hello world" program. I'll
4174 provide a little more than that: a program that prints "Hello world"
4175 finds the GCD of two numbers, prints the first few elements of
4176 Fibonacci, performs a binary search for a number, and a few other
4177 things which will likely grow as the languages grows.
4179 ###### File: oceani.mk
4182 @echo "===== DEMO ====="
4183 ./oceani --section "demo: hello" oceani.mdc 55 33
4189 four ::= 2 + 2 ; five ::= 10/2
4190 const pie ::= "I like Pie";
4191 cake ::= "The cake is"
4200 print "Hello World, what lovely oceans you have!"
4201 print "Are there", five, "?"
4202 print pi, pie, "but", cake
4204 /* When a variable is defined in both branches of an 'if',
4205 * and used afterwards, the variables are merged.
4211 print "Is", A, "bigger than", B,"? ", bigger
4212 /* If a variable is not used after the 'if', no
4213 * merge happens, so types can be different
4216 double:string = "yes"
4217 print A, "is more than twice", B, "?", double
4220 print "double", B, "is", double
4225 if a > 0 and then b > 0:
4231 print "GCD of", A, "and", B,"is", a
4233 print a, "is not positive, cannot calculate GCD"
4235 print b, "is not positive, cannot calculate GCD"
4240 print "Fibonacci:", f1,f2,
4241 then togo = togo - 1
4249 /* Binary search... */
4254 mid := (lo + hi) / 2
4266 print "Yay, I found", target
4268 print "Closest I found was", mid
4273 // "middle square" PRNG. Not particularly good, but one my
4274 // Dad taught me - the first one I ever heard of.
4275 for i:=1; then i = i + 1; while i < size:
4276 n := list[i-1] * list[i-1]
4277 list[i] = (n / 100) % 10 000
4279 print "Before sort:",
4280 for i:=0; then i = i + 1; while i < size:
4284 for i := 1; then i=i+1; while i < size:
4285 for j:=i-1; then j=j-1; while j >= 0:
4286 if list[j] > list[j+1]:
4290 print " After sort:",
4291 for i:=0; then i = i + 1; while i < size:
4297 bob.alive = (bob.name == "Hello")
4298 print "bob", "is" if bob.alive else "isn't", "alive"