1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
113 struct parse_context {
114 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
240 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
244 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
247 fprintf(stderr, "oceani: no main function found.\n");
248 context.parse_error = 1;
250 if (context.prog && !context.parse_error) {
251 if (!analyse_prog(context.prog, &context)) {
252 fprintf(stderr, "oceani: type error in program - not running.\n");
253 context.parse_error = 1;
256 if (context.prog && doprint) {
259 print_exec(context.prog, 0, brackets);
261 if (context.prog && doexec && !context.parse_error)
262 interp_prog(&context, context.prog, argc - optind, argv+optind);
263 free_exec(context.prog);
266 struct section *t = s->next;
272 ## free context types
273 ## free context storage
274 exit(context.parse_error ? 1 : 0);
279 The four requirements of parse, analyse, print, interpret apply to
280 each language element individually so that is how most of the code
283 Three of the four are fairly self explanatory. The one that requires
284 a little explanation is the analysis step.
286 The current language design does not require the types of variables to
287 be declared, but they must still have a single type. Different
288 operations impose different requirements on the variables, for example
289 addition requires both arguments to be numeric, and assignment
290 requires the variable on the left to have the same type as the
291 expression on the right.
293 Analysis involves propagating these type requirements around and
294 consequently setting the type of each variable. If any requirements
295 are violated (e.g. a string is compared with a number) or if a
296 variable needs to have two different types, then an error is raised
297 and the program will not run.
299 If the same variable is declared in both branchs of an 'if/else', or
300 in all cases of a 'switch' then the multiple instances may be merged
301 into just one variable if the variable is referenced after the
302 conditional statement. When this happens, the types must naturally be
303 consistent across all the branches. When the variable is not used
304 outside the if, the variables in the different branches are distinct
305 and can be of different types.
307 Undeclared names may only appear in "use" statements and "case" expressions.
308 These names are given a type of "label" and a unique value.
309 This allows them to fill the role of a name in an enumerated type, which
310 is useful for testing the `switch` statement.
312 As we will see, the condition part of a `while` statement can return
313 either a Boolean or some other type. This requires that the expected
314 type that gets passed around comprises a type and a flag to indicate
315 that `Tbool` is also permitted.
317 As there are, as yet, no distinct types that are compatible, there
318 isn't much subtlety in the analysis. When we have distinct number
319 types, this will become more interesting.
323 When analysis discovers an inconsistency it needs to report an error;
324 just refusing to run the code ensures that the error doesn't cascade,
325 but by itself it isn't very useful. A clear understanding of the sort
326 of error message that are useful will help guide the process of
329 At a simplistic level, the only sort of error that type analysis can
330 report is that the type of some construct doesn't match a contextual
331 requirement. For example, in `4 + "hello"` the addition provides a
332 contextual requirement for numbers, but `"hello"` is not a number. In
333 this particular example no further information is needed as the types
334 are obvious from local information. When a variable is involved that
335 isn't the case. It may be helpful to explain why the variable has a
336 particular type, by indicating the location where the type was set,
337 whether by declaration or usage.
339 Using a recursive-descent analysis we can easily detect a problem at
340 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
341 will detect that one argument is not a number and the usage of `hello`
342 will detect that a number was wanted, but not provided. In this
343 (early) version of the language, we will generate error reports at
344 multiple locations, so the use of `hello` will report an error and
345 explain were the value was set, and the addition will report an error
346 and say why numbers are needed. To be able to report locations for
347 errors, each language element will need to record a file location
348 (line and column) and each variable will need to record the language
349 element where its type was set. For now we will assume that each line
350 of an error message indicates one location in the file, and up to 2
351 types. So we provide a `printf`-like function which takes a format, a
352 location (a `struct exec` which has not yet been introduced), and 2
353 types. "`%1`" reports the first type, "`%2`" reports the second. We
354 will need a function to print the location, once we know how that is
355 stored. e As will be explained later, there are sometimes extra rules for
356 type matching and they might affect error messages, we need to pass those
359 As well as type errors, we sometimes need to report problems with
360 tokens, which might be unexpected or might name a type that has not
361 been defined. For these we have `tok_err()` which reports an error
362 with a given token. Each of the error functions sets the flag in the
363 context so indicate that parsing failed.
367 static void fput_loc(struct exec *loc, FILE *f);
369 ###### core functions
371 static void type_err(struct parse_context *c,
372 char *fmt, struct exec *loc,
373 struct type *t1, int rules, struct type *t2)
375 fprintf(stderr, "%s:", c->file_name);
376 fput_loc(loc, stderr);
377 for (; *fmt ; fmt++) {
384 case '%': fputc(*fmt, stderr); break; // NOTEST
385 default: fputc('?', stderr); break; // NOTEST
387 type_print(t1, stderr);
390 type_print(t2, stderr);
399 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
401 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
402 t->txt.len, t->txt.txt);
406 ## Entities: declared and predeclared.
408 There are various "things" that the language and/or the interpreter
409 needs to know about to parse and execute a program. These include
410 types, variables, values, and executable code. These are all lumped
411 together under the term "entities" (calling them "objects" would be
412 confusing) and introduced here. The following section will present the
413 different specific code elements which comprise or manipulate these
418 Values come in a wide range of types, with more likely to be added.
419 Each type needs to be able to print its own values (for convenience at
420 least) as well as to compare two values, at least for equality and
421 possibly for order. For now, values might need to be duplicated and
422 freed, though eventually such manipulations will be better integrated
425 Rather than requiring every numeric type to support all numeric
426 operations (add, multiple, etc), we allow types to be able to present
427 as one of a few standard types: integer, float, and fraction. The
428 existence of these conversion functions eventually enable types to
429 determine if they are compatible with other types, though such types
430 have not yet been implemented.
432 Named type are stored in a simple linked list. Objects of each type are
433 "values" which are often passed around by value.
440 ## value union fields
448 void (*init)(struct type *type, struct value *val);
449 void (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
450 void (*print)(struct type *type, struct value *val);
451 void (*print_type)(struct type *type, FILE *f);
452 int (*cmp_order)(struct type *t1, struct type *t2,
453 struct value *v1, struct value *v2);
454 int (*cmp_eq)(struct type *t1, struct type *t2,
455 struct value *v1, struct value *v2);
456 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
457 void (*free)(struct type *type, struct value *val);
458 void (*free_type)(struct type *t);
459 long long (*to_int)(struct value *v);
460 double (*to_float)(struct value *v);
461 int (*to_mpq)(mpq_t *q, struct value *v);
470 struct type *typelist;
474 static struct type *find_type(struct parse_context *c, struct text s)
476 struct type *l = c->typelist;
479 text_cmp(l->name, s) != 0)
484 static struct type *add_type(struct parse_context *c, struct text s,
489 n = calloc(1, sizeof(*n));
492 n->next = c->typelist;
497 static void free_type(struct type *t)
499 /* The type is always a reference to something in the
500 * context, so we don't need to free anything.
504 static void free_value(struct type *type, struct value *v)
508 memset(v, 0x5a, type->size);
512 static void type_print(struct type *type, FILE *f)
515 fputs("*unknown*type*", f); // NOTEST
516 else if (type->name.len)
517 fprintf(f, "%.*s", type->name.len, type->name.txt);
518 else if (type->print_type)
519 type->print_type(type, f);
521 fputs("*invalid*type*", f); // NOTEST
524 static void val_init(struct type *type, struct value *val)
526 if (type && type->init)
527 type->init(type, val);
530 static void dup_value(struct type *type,
531 struct value *vold, struct value *vnew)
533 if (type && type->dup)
534 type->dup(type, vold, vnew);
537 static int value_cmp(struct type *tl, struct type *tr,
538 struct value *left, struct value *right)
540 if (tl && tl->cmp_order)
541 return tl->cmp_order(tl, tr, left, right);
542 if (tl && tl->cmp_eq) // NOTEST
543 return tl->cmp_eq(tl, tr, left, right); // NOTEST
547 static void print_value(struct type *type, struct value *v)
549 if (type && type->print)
550 type->print(type, v);
552 printf("*Unknown*"); // NOTEST
557 static void free_value(struct type *type, struct value *v);
558 static int type_compat(struct type *require, struct type *have, int rules);
559 static void type_print(struct type *type, FILE *f);
560 static void val_init(struct type *type, struct value *v);
561 static void dup_value(struct type *type,
562 struct value *vold, struct value *vnew);
563 static int value_cmp(struct type *tl, struct type *tr,
564 struct value *left, struct value *right);
565 static void print_value(struct type *type, struct value *v);
567 ###### free context types
569 while (context.typelist) {
570 struct type *t = context.typelist;
572 context.typelist = t->next;
578 Type can be specified for local variables, for fields in a structure,
579 for formal parameters to functions, and possibly elsewhere. Different
580 rules may apply in different contexts. As a minimum, a named type may
581 always be used. Currently the type of a formal parameter can be
582 different from types in other contexts, so we have a separate grammar
588 Type -> IDENTIFIER ${
589 $0 = find_type(c, $1.txt);
592 "error: undefined type", &$1);
599 FormalType -> Type ${ $0 = $<1; }$
600 ## formal type grammar
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 ###### type functions
623 int (*compat)(struct type *this, struct type *other);
627 static int type_compat(struct type *require, struct type *have, int rules)
629 if ((rules & Rboolok) && have == Tbool)
631 if ((rules & Rnolabel) && have == Tlabel)
633 if (!require || !have)
637 return require->compat(require, have);
639 return require == have;
644 #include "parse_string.h"
645 #include "parse_number.h"
648 myLDLIBS := libnumber.o libstring.o -lgmp
649 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
651 ###### type union fields
652 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
654 ###### value union fields
661 static void _free_value(struct type *type, struct value *v)
665 switch (type->vtype) {
667 case Vstr: free(v->str.txt); break;
668 case Vnum: mpq_clear(v->num); break;
674 ###### value functions
676 static void _val_init(struct type *type, struct value *val)
678 switch(type->vtype) {
679 case Vnone: // NOTEST
682 mpq_init(val->num); break;
684 val->str.txt = malloc(1);
696 static void _dup_value(struct type *type,
697 struct value *vold, struct value *vnew)
699 switch (type->vtype) {
700 case Vnone: // NOTEST
703 vnew->label = vold->label;
706 vnew->bool = vold->bool;
710 mpq_set(vnew->num, vold->num);
713 vnew->str.len = vold->str.len;
714 vnew->str.txt = malloc(vnew->str.len);
715 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
720 static int _value_cmp(struct type *tl, struct type *tr,
721 struct value *left, struct value *right)
725 return tl - tr; // NOTEST
727 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
728 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
729 case Vstr: cmp = text_cmp(left->str, right->str); break;
730 case Vbool: cmp = left->bool - right->bool; break;
731 case Vnone: cmp = 0; // NOTEST
736 static void _print_value(struct type *type, struct value *v)
738 switch (type->vtype) {
739 case Vnone: // NOTEST
740 printf("*no-value*"); break; // NOTEST
741 case Vlabel: // NOTEST
742 printf("*label-%p*", v->label); break; // NOTEST
744 printf("%.*s", v->str.len, v->str.txt); break;
746 printf("%s", v->bool ? "True":"False"); break;
751 mpf_set_q(fl, v->num);
752 gmp_printf("%Fg", fl);
759 static void _free_value(struct type *type, struct value *v);
761 static struct type base_prototype = {
763 .print = _print_value,
764 .cmp_order = _value_cmp,
765 .cmp_eq = _value_cmp,
770 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
773 static struct type *add_base_type(struct parse_context *c, char *n,
774 enum vtype vt, int size)
776 struct text txt = { n, strlen(n) };
779 t = add_type(c, txt, &base_prototype);
782 t->align = size > sizeof(void*) ? sizeof(void*) : size;
783 if (t->size & (t->align - 1))
784 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
788 ###### context initialization
790 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
791 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
792 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
793 Tnone = add_base_type(&context, "none", Vnone, 0);
794 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
798 Variables are scoped named values. We store the names in a linked list
799 of "bindings" sorted in lexical order, and use sequential search and
806 struct binding *next; // in lexical order
810 This linked list is stored in the parse context so that "reduce"
811 functions can find or add variables, and so the analysis phase can
812 ensure that every variable gets a type.
816 struct binding *varlist; // In lexical order
820 static struct binding *find_binding(struct parse_context *c, struct text s)
822 struct binding **l = &c->varlist;
827 (cmp = text_cmp((*l)->name, s)) < 0)
831 n = calloc(1, sizeof(*n));
838 Each name can be linked to multiple variables defined in different
839 scopes. Each scope starts where the name is declared and continues
840 until the end of the containing code block. Scopes of a given name
841 cannot nest, so a declaration while a name is in-scope is an error.
843 ###### binding fields
844 struct variable *var;
848 struct variable *previous;
850 struct binding *name;
851 struct exec *where_decl;// where name was declared
852 struct exec *where_set; // where type was set
856 When a scope closes, the values of the variables might need to be freed.
857 This happens in the context of some `struct exec` and each `exec` will
858 need to know which variables need to be freed when it completes.
861 struct variable *to_free;
863 ####### variable fields
864 struct exec *cleanup_exec;
865 struct variable *next_free;
867 ####### interp exec cleanup
870 for (v = e->to_free; v; v = v->next_free) {
871 struct value *val = var_value(c, v);
872 free_value(v->type, val);
877 static void variable_unlink_exec(struct variable *v)
879 struct variable **vp;
880 if (!v->cleanup_exec)
882 for (vp = &v->cleanup_exec->to_free;
883 *vp; vp = &(*vp)->next_free) {
887 v->cleanup_exec = NULL;
892 While the naming seems strange, we include local constants in the
893 definition of variables. A name declared `var := value` can
894 subsequently be changed, but a name declared `var ::= value` cannot -
897 ###### variable fields
900 Scopes in parallel branches can be partially merged. More
901 specifically, if a given name is declared in both branches of an
902 if/else then its scope is a candidate for merging. Similarly if
903 every branch of an exhaustive switch (e.g. has an "else" clause)
904 declares a given name, then the scopes from the branches are
905 candidates for merging.
907 Note that names declared inside a loop (which is only parallel to
908 itself) are never visible after the loop. Similarly names defined in
909 scopes which are not parallel, such as those started by `for` and
910 `switch`, are never visible after the scope. Only variables defined in
911 both `then` and `else` (including the implicit then after an `if`, and
912 excluding `then` used with `for`) and in all `case`s and `else` of a
913 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
915 Labels, which are a bit like variables, follow different rules.
916 Labels are not explicitly declared, but if an undeclared name appears
917 in a context where a label is legal, that effectively declares the
918 name as a label. The declaration remains in force (or in scope) at
919 least to the end of the immediately containing block and conditionally
920 in any larger containing block which does not declare the name in some
921 other way. Importantly, the conditional scope extension happens even
922 if the label is only used in one parallel branch of a conditional --
923 when used in one branch it is treated as having been declared in all
926 Merge candidates are tentatively visible beyond the end of the
927 branching statement which creates them. If the name is used, the
928 merge is affirmed and they become a single variable visible at the
929 outer layer. If not - if it is redeclared first - the merge lapses.
931 To track scopes we have an extra stack, implemented as a linked list,
932 which roughly parallels the parse stack and which is used exclusively
933 for scoping. When a new scope is opened, a new frame is pushed and
934 the child-count of the parent frame is incremented. This child-count
935 is used to distinguish between the first of a set of parallel scopes,
936 in which declared variables must not be in scope, and subsequent
937 branches, whether they may already be conditionally scoped.
939 To push a new frame *before* any code in the frame is parsed, we need a
940 grammar reduction. This is most easily achieved with a grammar
941 element which derives the empty string, and creates the new scope when
942 it is recognised. This can be placed, for example, between a keyword
943 like "if" and the code following it.
947 struct scope *parent;
953 struct scope *scope_stack;
956 static void scope_pop(struct parse_context *c)
958 struct scope *s = c->scope_stack;
960 c->scope_stack = s->parent;
965 static void scope_push(struct parse_context *c)
967 struct scope *s = calloc(1, sizeof(*s));
969 c->scope_stack->child_count += 1;
970 s->parent = c->scope_stack;
978 OpenScope -> ${ scope_push(c); }$
980 Each variable records a scope depth and is in one of four states:
982 - "in scope". This is the case between the declaration of the
983 variable and the end of the containing block, and also between
984 the usage with affirms a merge and the end of that block.
986 The scope depth is not greater than the current parse context scope
987 nest depth. When the block of that depth closes, the state will
988 change. To achieve this, all "in scope" variables are linked
989 together as a stack in nesting order.
991 - "pending". The "in scope" block has closed, but other parallel
992 scopes are still being processed. So far, every parallel block at
993 the same level that has closed has declared the name.
995 The scope depth is the depth of the last parallel block that
996 enclosed the declaration, and that has closed.
998 - "conditionally in scope". The "in scope" block and all parallel
999 scopes have closed, and no further mention of the name has been seen.
1000 This state includes a secondary nest depth (`min_depth`) which records
1001 the outermost scope seen since the variable became conditionally in
1002 scope. If a use of the name is found, the variable becomes "in scope"
1003 and that secondary depth becomes the recorded scope depth. If the
1004 name is declared as a new variable, the old variable becomes "out of
1005 scope" and the recorded scope depth stays unchanged.
1007 - "out of scope". The variable is neither in scope nor conditionally
1008 in scope. It is permanently out of scope now and can be removed from
1009 the "in scope" stack.
1011 ###### variable fields
1012 int depth, min_depth;
1013 enum { OutScope, PendingScope, CondScope, InScope } scope;
1014 struct variable *in_scope;
1016 ###### parse context
1018 struct variable *in_scope;
1020 All variables with the same name are linked together using the
1021 'previous' link. Those variable that have been affirmatively merged all
1022 have a 'merged' pointer that points to one primary variable - the most
1023 recently declared instance. When merging variables, we need to also
1024 adjust the 'merged' pointer on any other variables that had previously
1025 been merged with the one that will no longer be primary.
1027 A variable that is no longer the most recent instance of a name may
1028 still have "pending" scope, if it might still be merged with most
1029 recent instance. These variables don't really belong in the
1030 "in_scope" list, but are not immediately removed when a new instance
1031 is found. Instead, they are detected and ignored when considering the
1032 list of in_scope names.
1034 The storage of the value of a variable will be described later. For now
1035 we just need to know that when a variable goes out of scope, it might
1036 need to be freed. For this we need to be able to find it, so assume that
1037 `var_value()` will provide that.
1039 ###### variable fields
1040 struct variable *merged;
1042 ###### ast functions
1044 static void variable_merge(struct variable *primary, struct variable *secondary)
1048 primary = primary->merged;
1050 for (v = primary->previous; v; v=v->previous)
1051 if (v == secondary || v == secondary->merged ||
1052 v->merged == secondary ||
1053 v->merged == secondary->merged) {
1054 v->scope = OutScope;
1055 v->merged = primary;
1056 variable_unlink_exec(v);
1060 ###### forward decls
1061 static struct value *var_value(struct parse_context *c, struct variable *v);
1063 ###### free global vars
1065 while (context.varlist) {
1066 struct binding *b = context.varlist;
1067 struct variable *v = b->var;
1068 context.varlist = b->next;
1071 struct variable *t = v;
1075 free_value(t->type, var_value(&context, t));
1077 free_exec(t->where_decl);
1083 #### Manipulating Bindings
1085 When a name is conditionally visible, a new declaration discards the
1086 old binding - the condition lapses. Conversely a usage of the name
1087 affirms the visibility and extends it to the end of the containing
1088 block - i.e. the block that contains both the original declaration and
1089 the latest usage. This is determined from `min_depth`. When a
1090 conditionally visible variable gets affirmed like this, it is also
1091 merged with other conditionally visible variables with the same name.
1093 When we parse a variable declaration we either report an error if the
1094 name is currently bound, or create a new variable at the current nest
1095 depth if the name is unbound or bound to a conditionally scoped or
1096 pending-scope variable. If the previous variable was conditionally
1097 scoped, it and its homonyms becomes out-of-scope.
1099 When we parse a variable reference (including non-declarative assignment
1100 "foo = bar") we report an error if the name is not bound or is bound to
1101 a pending-scope variable; update the scope if the name is bound to a
1102 conditionally scoped variable; or just proceed normally if the named
1103 variable is in scope.
1105 When we exit a scope, any variables bound at this level are either
1106 marked out of scope or pending-scoped, depending on whether the scope
1107 was sequential or parallel. Here a "parallel" scope means the "then"
1108 or "else" part of a conditional, or any "case" or "else" branch of a
1109 switch. Other scopes are "sequential".
1111 When exiting a parallel scope we check if there are any variables that
1112 were previously pending and are still visible. If there are, then
1113 they weren't redeclared in the most recent scope, so they cannot be
1114 merged and must become out-of-scope. If it is not the first of
1115 parallel scopes (based on `child_count`), we check that there was a
1116 previous binding that is still pending-scope. If there isn't, the new
1117 variable must now be out-of-scope.
1119 When exiting a sequential scope that immediately enclosed parallel
1120 scopes, we need to resolve any pending-scope variables. If there was
1121 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1122 we need to mark all pending-scope variable as out-of-scope. Otherwise
1123 all pending-scope variables become conditionally scoped.
1126 enum closetype { CloseSequential, CloseParallel, CloseElse };
1128 ###### ast functions
1130 static struct variable *var_decl(struct parse_context *c, struct text s)
1132 struct binding *b = find_binding(c, s);
1133 struct variable *v = b->var;
1135 switch (v ? v->scope : OutScope) {
1137 /* Caller will report the error */
1141 v && v->scope == CondScope;
1143 v->scope = OutScope;
1147 v = calloc(1, sizeof(*v));
1148 v->previous = b->var;
1152 v->min_depth = v->depth = c->scope_depth;
1154 v->in_scope = c->in_scope;
1159 static struct variable *var_ref(struct parse_context *c, struct text s)
1161 struct binding *b = find_binding(c, s);
1162 struct variable *v = b->var;
1163 struct variable *v2;
1165 switch (v ? v->scope : OutScope) {
1168 /* Caller will report the error */
1171 /* All CondScope variables of this name need to be merged
1172 * and become InScope
1174 v->depth = v->min_depth;
1176 for (v2 = v->previous;
1177 v2 && v2->scope == CondScope;
1179 variable_merge(v, v2);
1187 static void var_block_close(struct parse_context *c, enum closetype ct,
1190 /* Close off all variables that are in_scope.
1191 * Some variables in c->scope may already be not-in-scope,
1192 * such as when a PendingScope variable is hidden by a new
1193 * variable with the same name.
1194 * So we check for v->name->var != v and drop them.
1195 * If we choose to make a variable OutScope, we drop it
1198 struct variable *v, **vp, *v2;
1201 for (vp = &c->in_scope;
1202 (v = *vp) && v->min_depth > c->scope_depth;
1203 (v->scope == OutScope || v->name->var != v)
1204 ? (*vp = v->in_scope, 0)
1205 : ( vp = &v->in_scope, 0)) {
1206 v->min_depth = c->scope_depth;
1207 if (v->name->var != v)
1208 /* This is still in scope, but we haven't just
1212 v->min_depth = c->scope_depth;
1213 if (v->scope == InScope) {
1214 /* This variable gets cleaned up when 'e' finishes */
1215 variable_unlink_exec(v);
1216 v->cleanup_exec = e;
1217 v->next_free = e->to_free;
1222 case CloseParallel: /* handle PendingScope */
1226 if (c->scope_stack->child_count == 1)
1227 /* first among parallel branches */
1228 v->scope = PendingScope;
1229 else if (v->previous &&
1230 v->previous->scope == PendingScope)
1231 /* all previous branches used name */
1232 v->scope = PendingScope;
1233 else if (v->type == Tlabel)
1234 /* Labels remain pending even when not used */
1235 v->scope = PendingScope; // UNTESTED
1237 v->scope = OutScope;
1238 if (ct == CloseElse) {
1239 /* All Pending variables with this name
1240 * are now Conditional */
1242 v2 && v2->scope == PendingScope;
1244 v2->scope = CondScope;
1248 /* Not possible as it would require
1249 * parallel scope to be nested immediately
1250 * in a parallel scope, and that never
1254 /* Not possible as we already tested for
1260 case CloseSequential:
1261 if (v->type == Tlabel)
1262 v->scope = PendingScope;
1265 v->scope = OutScope;
1268 /* There was no 'else', so we can only become
1269 * conditional if we know the cases were exhaustive,
1270 * and that doesn't mean anything yet.
1271 * So only labels become conditional..
1274 v2 && v2->scope == PendingScope;
1276 if (v2->type == Tlabel)
1277 v2->scope = CondScope;
1279 v2->scope = OutScope;
1282 case OutScope: break;
1291 The value of a variable is store separately from the variable, on an
1292 analogue of a stack frame. There are (currently) two frames that can be
1293 active. A global frame which currently only stores constants, and a
1294 stacked frame which stores local variables. Each variable knows if it
1295 is global or not, and what its index into the frame is.
1297 Values in the global frame are known immediately they are relevant, so
1298 the frame needs to be reallocated as it grows so it can store those
1299 values. The local frame doesn't get values until the interpreted phase
1300 is started, so there is no need to allocate until the size is known.
1302 ###### variable fields
1306 ###### parse context
1308 short global_size, global_alloc;
1310 void *global, *local;
1312 ###### ast functions
1314 static struct value *var_value(struct parse_context *c, struct variable *v)
1317 if (!c->local || !v->type)
1318 return NULL; // NOTEST
1319 if (v->frame_pos + v->type->size > c->local_size) {
1320 printf("INVALID frame_pos\n"); // NOTEST
1323 return c->local + v->frame_pos;
1325 if (c->global_size > c->global_alloc) {
1326 int old = c->global_alloc;
1327 c->global_alloc = (c->global_size | 1023) + 1024;
1328 c->global = realloc(c->global, c->global_alloc);
1329 memset(c->global + old, 0, c->global_alloc - old);
1331 return c->global + v->frame_pos;
1334 static struct value *global_alloc(struct parse_context *c, struct type *t,
1335 struct variable *v, struct value *init)
1338 struct variable scratch;
1340 if (t->prepare_type)
1341 t->prepare_type(c, t, 1); // NOTEST
1343 if (c->global_size & (t->align - 1))
1344 c->global_size = (c->global_size + t->align) & ~(t->align-1); // UNTESTED
1349 v->frame_pos = c->global_size;
1351 c->global_size += v->type->size;
1352 ret = var_value(c, v);
1354 memcpy(ret, init, t->size);
1360 As global values are found -- struct field initializers, labels etc --
1361 `global_alloc()` is called to record the value in the global frame.
1363 When the program is fully parsed, we need to walk the list of variables
1364 to find any that weren't merged away and that aren't global, and to
1365 calculate the frame size and assign a frame position for each variable.
1366 For this we have `scope_finalize()`.
1368 ###### ast functions
1370 static void scope_finalize(struct parse_context *c)
1374 for (b = c->varlist; b; b = b->next) {
1376 for (v = b->var; v; v = v->previous) {
1377 struct type *t = v->type;
1382 if (c->local_size & (t->align - 1))
1383 c->local_size = (c->local_size + t->align) & ~(t->align-1);
1384 v->frame_pos = c->local_size;
1385 c->local_size += v->type->size;
1388 c->local = calloc(1, c->local_size);
1391 ###### free context storage
1392 free(context.global);
1393 free(context.local);
1397 Executables can be lots of different things. In many cases an
1398 executable is just an operation combined with one or two other
1399 executables. This allows for expressions and lists etc. Other times an
1400 executable is something quite specific like a constant or variable name.
1401 So we define a `struct exec` to be a general executable with a type, and
1402 a `struct binode` which is a subclass of `exec`, forms a node in a
1403 binary tree, and holds an operation. There will be other subclasses,
1404 and to access these we need to be able to `cast` the `exec` into the
1405 various other types. The first field in any `struct exec` is the type
1406 from the `exec_types` enum.
1409 #define cast(structname, pointer) ({ \
1410 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1411 if (__mptr && *__mptr != X##structname) abort(); \
1412 (struct structname *)( (char *)__mptr);})
1414 #define new(structname) ({ \
1415 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1416 __ptr->type = X##structname; \
1417 __ptr->line = -1; __ptr->column = -1; \
1420 #define new_pos(structname, token) ({ \
1421 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1422 __ptr->type = X##structname; \
1423 __ptr->line = token.line; __ptr->column = token.col; \
1432 enum exec_types type;
1441 struct exec *left, *right;
1444 ###### ast functions
1446 static int __fput_loc(struct exec *loc, FILE *f)
1450 if (loc->line >= 0) {
1451 fprintf(f, "%d:%d: ", loc->line, loc->column);
1454 if (loc->type == Xbinode)
1455 return __fput_loc(cast(binode,loc)->left, f) ||
1456 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1459 static void fput_loc(struct exec *loc, FILE *f)
1461 if (!__fput_loc(loc, f))
1462 fprintf(f, "??:??: "); // NOTEST
1465 Each different type of `exec` node needs a number of functions defined,
1466 a bit like methods. We must be able to free it, print it, analyse it
1467 and execute it. Once we have specific `exec` types we will need to
1468 parse them too. Let's take this a bit more slowly.
1472 The parser generator requires a `free_foo` function for each struct
1473 that stores attributes and they will often be `exec`s and subtypes
1474 there-of. So we need `free_exec` which can handle all the subtypes,
1475 and we need `free_binode`.
1477 ###### ast functions
1479 static void free_binode(struct binode *b)
1484 free_exec(b->right);
1488 ###### core functions
1489 static void free_exec(struct exec *e)
1498 ###### forward decls
1500 static void free_exec(struct exec *e);
1502 ###### free exec cases
1503 case Xbinode: free_binode(cast(binode, e)); break;
1507 Printing an `exec` requires that we know the current indent level for
1508 printing line-oriented components. As will become clear later, we
1509 also want to know what sort of bracketing to use.
1511 ###### ast functions
1513 static void do_indent(int i, char *str)
1520 ###### core functions
1521 static void print_binode(struct binode *b, int indent, int bracket)
1525 ## print binode cases
1529 static void print_exec(struct exec *e, int indent, int bracket)
1535 print_binode(cast(binode, e), indent, bracket); break;
1540 do_indent(indent, "/* FREE");
1541 for (v = e->to_free; v; v = v->next_free)
1542 printf(" %.*s(%c%d+%d)", v->name->name.len, v->name->name.txt,
1543 v->global ? 'G':'L',
1544 v->frame_pos, v->type ? v->type->size:0);
1549 ###### forward decls
1551 static void print_exec(struct exec *e, int indent, int bracket);
1555 As discussed, analysis involves propagating type requirements around the
1556 program and looking for errors.
1558 So `propagate_types` is passed an expected type (being a `struct type`
1559 pointer together with some `val_rules` flags) that the `exec` is
1560 expected to return, and returns the type that it does return, either
1561 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1562 by reference. It is set to `0` when an error is found, and `2` when
1563 any change is made. If it remains unchanged at `1`, then no more
1564 propagation is needed.
1568 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1572 if (rules & Rnolabel)
1573 fputs(" (labels not permitted)", stderr);
1576 ###### core functions
1578 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1579 struct type *type, int rules);
1580 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1581 struct type *type, int rules)
1588 switch (prog->type) {
1591 struct binode *b = cast(binode, prog);
1593 ## propagate binode cases
1597 ## propagate exec cases
1602 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1603 struct type *type, int rules)
1605 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1614 Interpreting an `exec` doesn't require anything but the `exec`. State
1615 is stored in variables and each variable will be directly linked from
1616 within the `exec` tree. The exception to this is the `main` function
1617 which needs to look at command line arguments. This function will be
1618 interpreted separately.
1620 Each `exec` can return a value combined with a type in `struct lrval`.
1621 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1622 the location of a value, which can be updated, in `lval`. Others will
1623 set `lval` to NULL indicating that there is a value of appropriate type
1626 ###### core functions
1630 struct value rval, *lval;
1633 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1635 static struct value interp_exec(struct parse_context *c, struct exec *e,
1636 struct type **typeret)
1638 struct lrval ret = _interp_exec(c, e);
1640 if (!ret.type) abort();
1642 *typeret = ret.type;
1644 dup_value(ret.type, ret.lval, &ret.rval);
1648 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1649 struct type **typeret)
1651 struct lrval ret = _interp_exec(c, e);
1654 *typeret = ret.type;
1656 free_value(ret.type, &ret.rval);
1660 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1663 struct value rv = {}, *lrv = NULL;
1664 struct type *rvtype;
1666 rvtype = ret.type = Tnone;
1676 struct binode *b = cast(binode, e);
1677 struct value left, right, *lleft;
1678 struct type *ltype, *rtype;
1679 ltype = rtype = Tnone;
1681 ## interp binode cases
1683 free_value(ltype, &left);
1684 free_value(rtype, &right);
1687 ## interp exec cases
1692 ## interp exec cleanup
1698 Now that we have the shape of the interpreter in place we can add some
1699 complex types and connected them in to the data structures and the
1700 different phases of parse, analyse, print, interpret.
1702 Thus far we have arrays and structs.
1706 Arrays can be declared by giving a size and a type, as `[size]type' so
1707 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1708 size can be either a literal number, or a named constant. Some day an
1709 arbitrary expression will be supported.
1711 As a formal parameter to a function, the array can be declared with a
1712 new variable as the size: `name:[size::number]string`. The `size`
1713 variable is set to the size of the array and must be a constant. As
1714 `number` is the only supported type, it can be left out:
1715 `name:[size::]string`.
1717 Arrays cannot be assigned. When pointers are introduced we will also
1718 introduce array slices which can refer to part or all of an array -
1719 the assignment syntax will create a slice. For now, an array can only
1720 ever be referenced by the name it is declared with. It is likely that
1721 a "`copy`" primitive will eventually be define which can be used to
1722 make a copy of an array with controllable recursive depth.
1724 For now we have two sorts of array, those with fixed size either because
1725 it is given as a literal number or because it is a struct member (which
1726 cannot have a runtime-changing size), and those with a size that is
1727 determined at runtime - local variables with a const size. The former
1728 have their size calculated at parse time, the latter at run time.
1730 For the latter type, the `size` field of the type is the size of a
1731 pointer, and the array is reallocated every time it comes into scope.
1733 We differentiate struct fields with a const size from local variables
1734 with a const size by whether they are prepared at parse time or not.
1736 ###### type union fields
1739 int unspec; // size is unspecified - vsize must be set.
1742 struct variable *vsize;
1743 struct type *member;
1746 ###### value union fields
1747 void *array; // used if not static_size
1749 ###### value functions
1751 static void array_prepare_type(struct parse_context *c, struct type *type,
1754 struct value *vsize;
1756 if (!type->array.vsize || type->array.static_size)
1759 vsize = var_value(c, type->array.vsize);
1761 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1762 type->array.size = mpz_get_si(q);
1766 type->array.static_size = 1;
1767 type->size = type->array.size * type->array.member->size;
1768 type->align = type->array.member->align;
1772 static void array_init(struct type *type, struct value *val)
1775 void *ptr = val->ptr;
1779 if (!type->array.static_size) {
1780 val->array = calloc(type->array.size,
1781 type->array.member->size);
1784 for (i = 0; i < type->array.size; i++) {
1786 v = (void*)ptr + i * type->array.member->size;
1787 val_init(type->array.member, v);
1791 static void array_free(struct type *type, struct value *val)
1794 void *ptr = val->ptr;
1796 if (!type->array.static_size)
1798 for (i = 0; i < type->array.size; i++) {
1800 v = (void*)ptr + i * type->array.member->size;
1801 free_value(type->array.member, v);
1803 if (!type->array.static_size)
1807 static int array_compat(struct type *require, struct type *have)
1809 if (have->compat != require->compat)
1810 return 0; // UNTESTED
1811 /* Both are arrays, so we can look at details */
1812 if (!type_compat(require->array.member, have->array.member, 0))
1814 if (have->array.unspec && require->array.unspec) {
1815 if (have->array.vsize && require->array.vsize &&
1816 have->array.vsize != require->array.vsize) // UNTESTED
1817 /* sizes might not be the same */
1818 return 0; // UNTESTED
1821 if (have->array.unspec || require->array.unspec)
1822 return 1; // UNTESTED
1823 if (require->array.vsize == NULL && have->array.vsize == NULL)
1824 return require->array.size == have->array.size;
1826 return require->array.vsize == have->array.vsize; // UNTESTED
1829 static void array_print_type(struct type *type, FILE *f)
1832 if (type->array.vsize) {
1833 struct binding *b = type->array.vsize->name;
1834 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1835 type->array.unspec ? "::" : "");
1837 fprintf(f, "%d]", type->array.size);
1838 type_print(type->array.member, f);
1841 static struct type array_prototype = {
1843 .prepare_type = array_prepare_type,
1844 .print_type = array_print_type,
1845 .compat = array_compat,
1847 .size = sizeof(void*),
1848 .align = sizeof(void*),
1851 ###### declare terminals
1856 | [ NUMBER ] Type ${ {
1859 struct text noname = { "", 0 };
1862 $0 = t = add_type(c, noname, &array_prototype);
1863 t->array.member = $<4;
1864 t->array.vsize = NULL;
1865 if (number_parse(num, tail, $2.txt) == 0)
1866 tok_err(c, "error: unrecognised number", &$2);
1868 tok_err(c, "error: unsupported number suffix", &$2);
1870 t->array.size = mpz_get_ui(mpq_numref(num));
1871 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1872 tok_err(c, "error: array size must be an integer",
1874 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1875 tok_err(c, "error: array size is too large",
1879 t->array.static_size = 1;
1880 t->size = t->array.size * t->array.member->size;
1881 t->align = t->array.member->align;
1884 | [ IDENTIFIER ] Type ${ {
1885 struct variable *v = var_ref(c, $2.txt);
1886 struct text noname = { "", 0 };
1889 tok_err(c, "error: name undeclared", &$2);
1890 else if (!v->constant)
1891 tok_err(c, "error: array size must be a constant", &$2);
1893 $0 = add_type(c, noname, &array_prototype);
1894 $0->array.member = $<4;
1896 $0->array.vsize = v;
1901 OptType -> Type ${ $0 = $<1; }$
1904 ###### formal type grammar
1906 | [ IDENTIFIER :: OptType ] Type ${ {
1907 struct variable *v = var_decl(c, $ID.txt);
1908 struct text noname = { "", 0 };
1914 $0 = add_type(c, noname, &array_prototype);
1915 $0->array.member = $<6;
1917 $0->array.unspec = 1;
1918 $0->array.vsize = v;
1924 ###### variable grammar
1926 | Variable [ Expression ] ${ {
1927 struct binode *b = new(binode);
1934 ###### print binode cases
1936 print_exec(b->left, -1, bracket);
1938 print_exec(b->right, -1, bracket);
1942 ###### propagate binode cases
1944 /* left must be an array, right must be a number,
1945 * result is the member type of the array
1947 propagate_types(b->right, c, ok, Tnum, 0);
1948 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1949 if (!t || t->compat != array_compat) {
1950 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1953 if (!type_compat(type, t->array.member, rules)) {
1954 type_err(c, "error: have %1 but need %2", prog,
1955 t->array.member, rules, type);
1957 return t->array.member;
1961 ###### interp binode cases
1967 lleft = linterp_exec(c, b->left, <ype);
1968 right = interp_exec(c, b->right, &rtype);
1970 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1974 if (ltype->array.static_size)
1977 ptr = *(void**)lleft;
1978 rvtype = ltype->array.member;
1979 if (i >= 0 && i < ltype->array.size)
1980 lrv = ptr + i * rvtype->size;
1982 val_init(ltype->array.member, &rv);
1989 A `struct` is a data-type that contains one or more other data-types.
1990 It differs from an array in that each member can be of a different
1991 type, and they are accessed by name rather than by number. Thus you
1992 cannot choose an element by calculation, you need to know what you
1995 The language makes no promises about how a given structure will be
1996 stored in memory - it is free to rearrange fields to suit whatever
1997 criteria seems important.
1999 Structs are declared separately from program code - they cannot be
2000 declared in-line in a variable declaration like arrays can. A struct
2001 is given a name and this name is used to identify the type - the name
2002 is not prefixed by the word `struct` as it would be in C.
2004 Structs are only treated as the same if they have the same name.
2005 Simply having the same fields in the same order is not enough. This
2006 might change once we can create structure initializers from a list of
2009 Each component datum is identified much like a variable is declared,
2010 with a name, one or two colons, and a type. The type cannot be omitted
2011 as there is no opportunity to deduce the type from usage. An initial
2012 value can be given following an equals sign, so
2014 ##### Example: a struct type
2020 would declare a type called "complex" which has two number fields,
2021 each initialised to zero.
2023 Struct will need to be declared separately from the code that uses
2024 them, so we will need to be able to print out the declaration of a
2025 struct when reprinting the whole program. So a `print_type_decl` type
2026 function will be needed.
2028 ###### type union fields
2040 ###### type functions
2041 void (*print_type_decl)(struct type *type, FILE *f);
2043 ###### value functions
2045 static void structure_init(struct type *type, struct value *val)
2049 for (i = 0; i < type->structure.nfields; i++) {
2051 v = (void*) val->ptr + type->structure.fields[i].offset;
2052 if (type->structure.fields[i].init)
2053 dup_value(type->structure.fields[i].type,
2054 type->structure.fields[i].init,
2057 val_init(type->structure.fields[i].type, v);
2061 static void structure_free(struct type *type, struct value *val)
2065 for (i = 0; i < type->structure.nfields; i++) {
2067 v = (void*)val->ptr + type->structure.fields[i].offset;
2068 free_value(type->structure.fields[i].type, v);
2072 static void structure_free_type(struct type *t)
2075 for (i = 0; i < t->structure.nfields; i++)
2076 if (t->structure.fields[i].init) {
2077 free_value(t->structure.fields[i].type,
2078 t->structure.fields[i].init);
2080 free(t->structure.fields);
2083 static struct type structure_prototype = {
2084 .init = structure_init,
2085 .free = structure_free,
2086 .free_type = structure_free_type,
2087 .print_type_decl = structure_print_type,
2101 ###### free exec cases
2103 free_exec(cast(fieldref, e)->left);
2107 ###### declare terminals
2110 ###### variable grammar
2112 | Variable . IDENTIFIER ${ {
2113 struct fieldref *fr = new_pos(fieldref, $2);
2120 ###### print exec cases
2124 struct fieldref *f = cast(fieldref, e);
2125 print_exec(f->left, -1, bracket);
2126 printf(".%.*s", f->name.len, f->name.txt);
2130 ###### ast functions
2131 static int find_struct_index(struct type *type, struct text field)
2134 for (i = 0; i < type->structure.nfields; i++)
2135 if (text_cmp(type->structure.fields[i].name, field) == 0)
2140 ###### propagate exec cases
2144 struct fieldref *f = cast(fieldref, prog);
2145 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2148 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2150 else if (st->init != structure_init)
2151 type_err(c, "error: field reference attempted on %1, not a struct",
2152 f->left, st, 0, NULL);
2153 else if (f->index == -2) {
2154 f->index = find_struct_index(st, f->name);
2156 type_err(c, "error: cannot find requested field in %1",
2157 f->left, st, 0, NULL);
2159 if (f->index >= 0) {
2160 struct type *ft = st->structure.fields[f->index].type;
2161 if (!type_compat(type, ft, rules))
2162 type_err(c, "error: have %1 but need %2", prog,
2169 ###### interp exec cases
2172 struct fieldref *f = cast(fieldref, e);
2174 struct value *lleft = linterp_exec(c, f->left, <ype);
2175 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2176 rvtype = ltype->structure.fields[f->index].type;
2182 struct fieldlist *prev;
2186 ###### ast functions
2187 static void free_fieldlist(struct fieldlist *f)
2191 free_fieldlist(f->prev);
2193 free_value(f->f.type, f->f.init); // UNTESTED
2194 free(f->f.init); // UNTESTED
2199 ###### top level grammar
2200 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2202 add_type(c, $2.txt, &structure_prototype);
2204 struct fieldlist *f;
2206 for (f = $3; f; f=f->prev)
2209 t->structure.nfields = cnt;
2210 t->structure.fields = calloc(cnt, sizeof(struct field));
2213 int a = f->f.type->align;
2215 t->structure.fields[cnt] = f->f;
2216 if (t->size & (a-1))
2217 t->size = (t->size | (a-1)) + 1;
2218 t->structure.fields[cnt].offset = t->size;
2219 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2228 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2229 | { SimpleFieldList } ${ $0 = $<SFL; }$
2230 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2231 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2233 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2234 | FieldLines SimpleFieldList Newlines ${
2239 SimpleFieldList -> Field ${ $0 = $<F; }$
2240 | SimpleFieldList ; Field ${
2244 | SimpleFieldList ; ${
2247 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2249 Field -> IDENTIFIER : Type = Expression ${ {
2252 $0 = calloc(1, sizeof(struct fieldlist));
2253 $0->f.name = $1.txt;
2258 propagate_types($<5, c, &ok, $3, 0);
2261 c->parse_error = 1; // UNTESTED
2263 struct value vl = interp_exec(c, $5, NULL);
2264 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2267 | IDENTIFIER : Type ${
2268 $0 = calloc(1, sizeof(struct fieldlist));
2269 $0->f.name = $1.txt;
2271 if ($0->f.type->prepare_type)
2272 $0->f.type->prepare_type(c, $0->f.type, 1);
2275 ###### forward decls
2276 static void structure_print_type(struct type *t, FILE *f);
2278 ###### value functions
2279 static void structure_print_type(struct type *t, FILE *f) // UNTESTED
2283 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2285 for (i = 0; i < t->structure.nfields; i++) {
2286 struct field *fl = t->structure.fields + i;
2287 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2288 type_print(fl->type, f);
2289 if (fl->type->print && fl->init) {
2291 if (fl->type == Tstr)
2292 fprintf(f, "\""); // UNTESTED
2293 print_value(fl->type, fl->init);
2294 if (fl->type == Tstr)
2295 fprintf(f, "\""); // UNTESTED
2301 ###### print type decls
2303 struct type *t; // UNTESTED
2306 while (target != 0) {
2308 for (t = context.typelist; t ; t=t->next)
2309 if (t->print_type_decl) {
2318 t->print_type_decl(t, stdout);
2326 A function is a named chunk of code which can be passed parameters and
2327 can return results. Each function has an implicit type which includes
2328 the set of parameters and the return value. As yet these types cannot
2329 be declared separate from the function itself.
2331 In fact, only one function is currently possible - `main`. `main` is
2332 passed an array of strings together with the size of the array, and
2333 doesn't return anything. The strings are command line arguments.
2335 The parameters can be specified either in parentheses as a list, such as
2337 ##### Example: function 1
2339 func main(av:[ac::number]string)
2342 or as an indented list of one parameter per line
2344 ##### Example: function 2
2347 argv:[argc::number]string
2351 For constructing these lists we use a `List` binode, which will be
2352 further detailed when Expression Lists are introduced.
2362 MainFunction -> func main ( OpenScope Args ) Block Newlines ${
2365 $0->left = reorder_bilist($<Ar);
2367 var_block_close(c, CloseSequential, $0);
2368 if (c->scope_stack && !c->parse_error) abort();
2370 | func main IN OpenScope OptNL Args OUT OptNL do Block Newlines ${
2373 $0->left = reorder_bilist($<Ar);
2375 var_block_close(c, CloseSequential, $0);
2376 if (c->scope_stack && !c->parse_error) abort();
2378 | func main NEWLINE OpenScope OptNL do Block Newlines ${
2383 var_block_close(c, CloseSequential, $0);
2384 if (c->scope_stack && !c->parse_error) abort();
2387 Args -> ${ $0 = NULL; }$
2388 | Varlist ${ $0 = $<1; }$
2389 | Varlist ; ${ $0 = $<1; }$
2390 | Varlist NEWLINE ${ $0 = $<1; }$
2392 Varlist -> Varlist ; ArgDecl ${ // UNTESTED
2406 ArgDecl -> IDENTIFIER : FormalType ${ {
2407 struct variable *v = var_decl(c, $1.txt);
2413 ## Executables: the elements of code
2415 Each code element needs to be parsed, printed, analysed,
2416 interpreted, and freed. There are several, so let's just start with
2417 the easy ones and work our way up.
2421 We have already met values as separate objects. When manifest
2422 constants appear in the program text, that must result in an executable
2423 which has a constant value. So the `val` structure embeds a value in
2436 ###### ast functions
2437 struct val *new_val(struct type *T, struct token tk)
2439 struct val *v = new_pos(val, tk);
2450 $0 = new_val(Tbool, $1);
2454 $0 = new_val(Tbool, $1);
2458 $0 = new_val(Tnum, $1);
2461 if (number_parse($0->val.num, tail, $1.txt) == 0)
2462 mpq_init($0->val.num); // UNTESTED
2464 tok_err(c, "error: unsupported number suffix",
2469 $0 = new_val(Tstr, $1);
2472 string_parse(&$1, '\\', &$0->val.str, tail);
2474 tok_err(c, "error: unsupported string suffix",
2479 $0 = new_val(Tstr, $1);
2482 string_parse(&$1, '\\', &$0->val.str, tail);
2484 tok_err(c, "error: unsupported string suffix",
2489 ###### print exec cases
2492 struct val *v = cast(val, e);
2493 if (v->vtype == Tstr)
2495 print_value(v->vtype, &v->val);
2496 if (v->vtype == Tstr)
2501 ###### propagate exec cases
2504 struct val *val = cast(val, prog);
2505 if (!type_compat(type, val->vtype, rules))
2506 type_err(c, "error: expected %1%r found %2",
2507 prog, type, rules, val->vtype);
2511 ###### interp exec cases
2513 rvtype = cast(val, e)->vtype;
2514 dup_value(rvtype, &cast(val, e)->val, &rv);
2517 ###### ast functions
2518 static void free_val(struct val *v)
2521 free_value(v->vtype, &v->val);
2525 ###### free exec cases
2526 case Xval: free_val(cast(val, e)); break;
2528 ###### ast functions
2529 // Move all nodes from 'b' to 'rv', reversing their order.
2530 // In 'b' 'left' is a list, and 'right' is the last node.
2531 // In 'rv', left' is the first node and 'right' is a list.
2532 static struct binode *reorder_bilist(struct binode *b)
2534 struct binode *rv = NULL;
2537 struct exec *t = b->right;
2541 b = cast(binode, b->left);
2551 Just as we used a `val` to wrap a value into an `exec`, we similarly
2552 need a `var` to wrap a `variable` into an exec. While each `val`
2553 contained a copy of the value, each `var` holds a link to the variable
2554 because it really is the same variable no matter where it appears.
2555 When a variable is used, we need to remember to follow the `->merged`
2556 link to find the primary instance.
2564 struct variable *var;
2572 VariableDecl -> IDENTIFIER : ${ {
2573 struct variable *v = var_decl(c, $1.txt);
2574 $0 = new_pos(var, $1);
2579 v = var_ref(c, $1.txt);
2581 type_err(c, "error: variable '%v' redeclared",
2583 type_err(c, "info: this is where '%v' was first declared",
2584 v->where_decl, NULL, 0, NULL);
2587 | IDENTIFIER :: ${ {
2588 struct variable *v = var_decl(c, $1.txt);
2589 $0 = new_pos(var, $1);
2595 v = var_ref(c, $1.txt);
2597 type_err(c, "error: variable '%v' redeclared",
2599 type_err(c, "info: this is where '%v' was first declared",
2600 v->where_decl, NULL, 0, NULL);
2603 | IDENTIFIER : Type ${ {
2604 struct variable *v = var_decl(c, $1.txt);
2605 $0 = new_pos(var, $1);
2612 v = var_ref(c, $1.txt);
2614 type_err(c, "error: variable '%v' redeclared",
2616 type_err(c, "info: this is where '%v' was first declared",
2617 v->where_decl, NULL, 0, NULL);
2620 | IDENTIFIER :: Type ${ {
2621 struct variable *v = var_decl(c, $1.txt);
2622 $0 = new_pos(var, $1);
2630 v = var_ref(c, $1.txt);
2632 type_err(c, "error: variable '%v' redeclared",
2634 type_err(c, "info: this is where '%v' was first declared",
2635 v->where_decl, NULL, 0, NULL);
2640 Variable -> IDENTIFIER ${ {
2641 struct variable *v = var_ref(c, $1.txt);
2642 $0 = new_pos(var, $1);
2644 /* This might be a label - allocate a var just in case */
2645 v = var_decl(c, $1.txt);
2652 cast(var, $0)->var = v;
2656 ###### print exec cases
2659 struct var *v = cast(var, e);
2661 struct binding *b = v->var->name;
2662 printf("%.*s", b->name.len, b->name.txt);
2669 if (loc && loc->type == Xvar) {
2670 struct var *v = cast(var, loc);
2672 struct binding *b = v->var->name;
2673 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2675 fputs("???", stderr); // NOTEST
2677 fputs("NOTVAR", stderr); // NOTEST
2680 ###### propagate exec cases
2684 struct var *var = cast(var, prog);
2685 struct variable *v = var->var;
2687 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2688 return Tnone; // NOTEST
2691 if (v->constant && (rules & Rnoconstant)) {
2692 type_err(c, "error: Cannot assign to a constant: %v",
2693 prog, NULL, 0, NULL);
2694 type_err(c, "info: name was defined as a constant here",
2695 v->where_decl, NULL, 0, NULL);
2698 if (v->type == Tnone && v->where_decl == prog)
2699 type_err(c, "error: variable used but not declared: %v",
2700 prog, NULL, 0, NULL);
2701 if (v->type == NULL) {
2702 if (type && *ok != 0) {
2704 v->where_set = prog;
2709 if (!type_compat(type, v->type, rules)) {
2710 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2711 type, rules, v->type);
2712 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2713 v->type, rules, NULL);
2720 ###### interp exec cases
2723 struct var *var = cast(var, e);
2724 struct variable *v = var->var;
2727 lrv = var_value(c, v);
2732 ###### ast functions
2734 static void free_var(struct var *v)
2739 ###### free exec cases
2740 case Xvar: free_var(cast(var, e)); break;
2742 ### Expressions: Conditional
2744 Our first user of the `binode` will be conditional expressions, which
2745 is a bit odd as they actually have three components. That will be
2746 handled by having 2 binodes for each expression. The conditional
2747 expression is the lowest precedence operator which is why we define it
2748 first - to start the precedence list.
2750 Conditional expressions are of the form "value `if` condition `else`
2751 other_value". They associate to the right, so everything to the right
2752 of `else` is part of an else value, while only a higher-precedence to
2753 the left of `if` is the if values. Between `if` and `else` there is no
2754 room for ambiguity, so a full conditional expression is allowed in
2766 Expression -> Expression if Expression else Expression $$ifelse ${ {
2767 struct binode *b1 = new(binode);
2768 struct binode *b2 = new(binode);
2777 ## expression grammar
2779 ###### print binode cases
2782 b2 = cast(binode, b->right);
2783 if (bracket) printf("(");
2784 print_exec(b2->left, -1, bracket);
2786 print_exec(b->left, -1, bracket);
2788 print_exec(b2->right, -1, bracket);
2789 if (bracket) printf(")");
2792 ###### propagate binode cases
2795 /* cond must be Tbool, others must match */
2796 struct binode *b2 = cast(binode, b->right);
2799 propagate_types(b->left, c, ok, Tbool, 0);
2800 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2801 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2805 ###### interp binode cases
2808 struct binode *b2 = cast(binode, b->right);
2809 left = interp_exec(c, b->left, <ype);
2811 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
2813 rv = interp_exec(c, b2->right, &rvtype);
2819 We take a brief detour, now that we have expressions, to describe lists
2820 of expressions. These will be needed for function parameters and
2821 possibly other situations. They seem generic enough to introduce here
2822 to be used elsewhere.
2824 And ExpressionList will use the `List` type of `binode`, building up at
2825 the end. And place where they are used will probably call
2826 `reorder_bilist()` to get a more normal first/next arrangement.
2828 ###### declare terminals
2831 `List` execs have no implicit semantics, so they are never propagated or
2832 interpreted. The can be printed as a comma separate list, which is how
2833 they are parsed. Note they are also used for function formal parameter
2834 lists. In that case a separate function is used to print them.
2836 ###### print binode cases
2840 print_exec(b->left, -1, bracket);
2843 b = cast(binode, b->right);
2847 ###### propagate binode cases
2848 case List: abort(); // NOTEST
2849 ###### interp binode cases
2850 case List: abort(); // NOTEST
2855 ExpressionList -> ExpressionList , Expression ${
2868 ### Expressions: Boolean
2870 The next class of expressions to use the `binode` will be Boolean
2871 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
2872 have same corresponding precendence. The difference is that they don't
2873 evaluate the second expression if not necessary.
2882 ###### expr precedence
2887 ###### expression grammar
2888 | Expression or Expression ${ {
2889 struct binode *b = new(binode);
2895 | Expression or else Expression ${ {
2896 struct binode *b = new(binode);
2903 | Expression and Expression ${ {
2904 struct binode *b = new(binode);
2910 | Expression and then Expression ${ {
2911 struct binode *b = new(binode);
2918 | not Expression ${ {
2919 struct binode *b = new(binode);
2925 ###### print binode cases
2927 if (bracket) printf("(");
2928 print_exec(b->left, -1, bracket);
2930 print_exec(b->right, -1, bracket);
2931 if (bracket) printf(")");
2934 if (bracket) printf("(");
2935 print_exec(b->left, -1, bracket);
2936 printf(" and then ");
2937 print_exec(b->right, -1, bracket);
2938 if (bracket) printf(")");
2941 if (bracket) printf("(");
2942 print_exec(b->left, -1, bracket);
2944 print_exec(b->right, -1, bracket);
2945 if (bracket) printf(")");
2948 if (bracket) printf("(");
2949 print_exec(b->left, -1, bracket);
2950 printf(" or else ");
2951 print_exec(b->right, -1, bracket);
2952 if (bracket) printf(")");
2955 if (bracket) printf("(");
2957 print_exec(b->right, -1, bracket);
2958 if (bracket) printf(")");
2961 ###### propagate binode cases
2967 /* both must be Tbool, result is Tbool */
2968 propagate_types(b->left, c, ok, Tbool, 0);
2969 propagate_types(b->right, c, ok, Tbool, 0);
2970 if (type && type != Tbool)
2971 type_err(c, "error: %1 operation found where %2 expected", prog,
2975 ###### interp binode cases
2977 rv = interp_exec(c, b->left, &rvtype);
2978 right = interp_exec(c, b->right, &rtype);
2979 rv.bool = rv.bool && right.bool;
2982 rv = interp_exec(c, b->left, &rvtype);
2984 rv = interp_exec(c, b->right, NULL);
2987 rv = interp_exec(c, b->left, &rvtype);
2988 right = interp_exec(c, b->right, &rtype);
2989 rv.bool = rv.bool || right.bool;
2992 rv = interp_exec(c, b->left, &rvtype);
2994 rv = interp_exec(c, b->right, NULL);
2997 rv = interp_exec(c, b->right, &rvtype);
3001 ### Expressions: Comparison
3003 Of slightly higher precedence that Boolean expressions are Comparisons.
3004 A comparison takes arguments of any comparable type, but the two types
3007 To simplify the parsing we introduce an `eop` which can record an
3008 expression operator, and the `CMPop` non-terminal will match one of them.
3015 ###### ast functions
3016 static void free_eop(struct eop *e)
3030 ###### expr precedence
3031 $LEFT < > <= >= == != CMPop
3033 ###### expression grammar
3034 | Expression CMPop Expression ${ {
3035 struct binode *b = new(binode);
3045 CMPop -> < ${ $0.op = Less; }$
3046 | > ${ $0.op = Gtr; }$
3047 | <= ${ $0.op = LessEq; }$
3048 | >= ${ $0.op = GtrEq; }$
3049 | == ${ $0.op = Eql; }$
3050 | != ${ $0.op = NEql; }$
3052 ###### print binode cases
3060 if (bracket) printf("(");
3061 print_exec(b->left, -1, bracket);
3063 case Less: printf(" < "); break;
3064 case LessEq: printf(" <= "); break;
3065 case Gtr: printf(" > "); break;
3066 case GtrEq: printf(" >= "); break;
3067 case Eql: printf(" == "); break;
3068 case NEql: printf(" != "); break;
3069 default: abort(); // NOTEST
3071 print_exec(b->right, -1, bracket);
3072 if (bracket) printf(")");
3075 ###### propagate binode cases
3082 /* Both must match but not be labels, result is Tbool */
3083 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
3085 propagate_types(b->right, c, ok, t, 0);
3087 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
3089 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
3091 if (!type_compat(type, Tbool, 0))
3092 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3093 Tbool, rules, type);
3096 ###### interp binode cases
3105 left = interp_exec(c, b->left, <ype);
3106 right = interp_exec(c, b->right, &rtype);
3107 cmp = value_cmp(ltype, rtype, &left, &right);
3110 case Less: rv.bool = cmp < 0; break;
3111 case LessEq: rv.bool = cmp <= 0; break;
3112 case Gtr: rv.bool = cmp > 0; break;
3113 case GtrEq: rv.bool = cmp >= 0; break;
3114 case Eql: rv.bool = cmp == 0; break;
3115 case NEql: rv.bool = cmp != 0; break;
3116 default: rv.bool = 0; break; // NOTEST
3121 ### Expressions: The rest
3123 The remaining expressions with the highest precedence are arithmetic,
3124 string concatenation, and string conversion. String concatenation
3125 (`++`) has the same precedence as multiplication and division, but lower
3128 String conversion is a temporary feature until I get a better type
3129 system. `$` is a prefix operator which expects a string and returns
3132 `+` and `-` are both infix and prefix operations (where they are
3133 absolute value and negation). These have different operator names.
3135 We also have a 'Bracket' operator which records where parentheses were
3136 found. This makes it easy to reproduce these when printing. Possibly I
3137 should only insert brackets were needed for precedence.
3147 ###### expr precedence
3153 ###### expression grammar
3154 | Expression Eop Expression ${ {
3155 struct binode *b = new(binode);
3162 | Expression Top Expression ${ {
3163 struct binode *b = new(binode);
3170 | ( Expression ) ${ {
3171 struct binode *b = new_pos(binode, $1);
3176 | Uop Expression ${ {
3177 struct binode *b = new(binode);
3182 | Value ${ $0 = $<1; }$
3183 | Variable ${ $0 = $<1; }$
3186 Eop -> + ${ $0.op = Plus; }$
3187 | - ${ $0.op = Minus; }$
3189 Uop -> + ${ $0.op = Absolute; }$
3190 | - ${ $0.op = Negate; }$
3191 | $ ${ $0.op = StringConv; }$
3193 Top -> * ${ $0.op = Times; }$
3194 | / ${ $0.op = Divide; }$
3195 | % ${ $0.op = Rem; }$
3196 | ++ ${ $0.op = Concat; }$
3198 ###### print binode cases
3205 if (bracket) printf("(");
3206 print_exec(b->left, indent, bracket);
3208 case Plus: fputs(" + ", stdout); break;
3209 case Minus: fputs(" - ", stdout); break;
3210 case Times: fputs(" * ", stdout); break;
3211 case Divide: fputs(" / ", stdout); break;
3212 case Rem: fputs(" % ", stdout); break;
3213 case Concat: fputs(" ++ ", stdout); break;
3214 default: abort(); // NOTEST
3216 print_exec(b->right, indent, bracket);
3217 if (bracket) printf(")");
3222 if (bracket) printf("(");
3224 case Absolute: fputs("+", stdout); break;
3225 case Negate: fputs("-", stdout); break;
3226 case StringConv: fputs("$", stdout); break;
3227 default: abort(); // NOTEST
3229 print_exec(b->right, indent, bracket);
3230 if (bracket) printf(")");
3234 print_exec(b->right, indent, bracket);
3238 ###### propagate binode cases
3244 /* both must be numbers, result is Tnum */
3247 /* as propagate_types ignores a NULL,
3248 * unary ops fit here too */
3249 propagate_types(b->left, c, ok, Tnum, 0);
3250 propagate_types(b->right, c, ok, Tnum, 0);
3251 if (!type_compat(type, Tnum, 0))
3252 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3257 /* both must be Tstr, result is Tstr */
3258 propagate_types(b->left, c, ok, Tstr, 0);
3259 propagate_types(b->right, c, ok, Tstr, 0);
3260 if (!type_compat(type, Tstr, 0))
3261 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3266 /* op must be string, result is number */
3267 propagate_types(b->left, c, ok, Tstr, 0);
3268 if (!type_compat(type, Tnum, 0))
3269 type_err(c, // UNTESTED
3270 "error: Can only convert string to number, not %1",
3271 prog, type, 0, NULL);
3275 return propagate_types(b->right, c, ok, type, 0);
3277 ###### interp binode cases
3280 rv = interp_exec(c, b->left, &rvtype);
3281 right = interp_exec(c, b->right, &rtype);
3282 mpq_add(rv.num, rv.num, right.num);
3285 rv = interp_exec(c, b->left, &rvtype);
3286 right = interp_exec(c, b->right, &rtype);
3287 mpq_sub(rv.num, rv.num, right.num);
3290 rv = interp_exec(c, b->left, &rvtype);
3291 right = interp_exec(c, b->right, &rtype);
3292 mpq_mul(rv.num, rv.num, right.num);
3295 rv = interp_exec(c, b->left, &rvtype);
3296 right = interp_exec(c, b->right, &rtype);
3297 mpq_div(rv.num, rv.num, right.num);
3302 left = interp_exec(c, b->left, <ype);
3303 right = interp_exec(c, b->right, &rtype);
3304 mpz_init(l); mpz_init(r); mpz_init(rem);
3305 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3306 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3307 mpz_tdiv_r(rem, l, r);
3308 val_init(Tnum, &rv);
3309 mpq_set_z(rv.num, rem);
3310 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3315 rv = interp_exec(c, b->right, &rvtype);
3316 mpq_neg(rv.num, rv.num);
3319 rv = interp_exec(c, b->right, &rvtype);
3320 mpq_abs(rv.num, rv.num);
3323 rv = interp_exec(c, b->right, &rvtype);
3326 left = interp_exec(c, b->left, <ype);
3327 right = interp_exec(c, b->right, &rtype);
3329 rv.str = text_join(left.str, right.str);
3332 right = interp_exec(c, b->right, &rvtype);
3336 struct text tx = right.str;
3339 if (tx.txt[0] == '-') {
3340 neg = 1; // UNTESTED
3341 tx.txt++; // UNTESTED
3342 tx.len--; // UNTESTED
3344 if (number_parse(rv.num, tail, tx) == 0)
3345 mpq_init(rv.num); // UNTESTED
3347 mpq_neg(rv.num, rv.num); // UNTESTED
3349 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3353 ###### value functions
3355 static struct text text_join(struct text a, struct text b)
3358 rv.len = a.len + b.len;
3359 rv.txt = malloc(rv.len);
3360 memcpy(rv.txt, a.txt, a.len);
3361 memcpy(rv.txt+a.len, b.txt, b.len);
3365 ### Blocks, Statements, and Statement lists.
3367 Now that we have expressions out of the way we need to turn to
3368 statements. There are simple statements and more complex statements.
3369 Simple statements do not contain (syntactic) newlines, complex statements do.
3371 Statements often come in sequences and we have corresponding simple
3372 statement lists and complex statement lists.
3373 The former comprise only simple statements separated by semicolons.
3374 The later comprise complex statements and simple statement lists. They are
3375 separated by newlines. Thus the semicolon is only used to separate
3376 simple statements on the one line. This may be overly restrictive,
3377 but I'm not sure I ever want a complex statement to share a line with
3380 Note that a simple statement list can still use multiple lines if
3381 subsequent lines are indented, so
3383 ###### Example: wrapped simple statement list
3388 is a single simple statement list. This might allow room for
3389 confusion, so I'm not set on it yet.
3391 A simple statement list needs no extra syntax. A complex statement
3392 list has two syntactic forms. It can be enclosed in braces (much like
3393 C blocks), or it can be introduced by an indent and continue until an
3394 unindented newline (much like Python blocks). With this extra syntax
3395 it is referred to as a block.
3397 Note that a block does not have to include any newlines if it only
3398 contains simple statements. So both of:
3400 if condition: a=b; d=f
3402 if condition { a=b; print f }
3406 In either case the list is constructed from a `binode` list with
3407 `Block` as the operator. When parsing the list it is most convenient
3408 to append to the end, so a list is a list and a statement. When using
3409 the list it is more convenient to consider a list to be a statement
3410 and a list. So we need a function to re-order a list.
3411 `reorder_bilist` serves this purpose.
3413 The only stand-alone statement we introduce at this stage is `pass`
3414 which does nothing and is represented as a `NULL` pointer in a `Block`
3415 list. Other stand-alone statements will follow once the infrastructure
3426 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3427 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3428 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3429 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3430 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3432 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3433 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3434 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3435 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3436 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3438 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3439 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3440 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3442 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3443 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3444 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3445 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3446 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3448 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3450 ComplexStatements -> ComplexStatements ComplexStatement ${
3460 | ComplexStatement ${
3472 ComplexStatement -> SimpleStatements Newlines ${
3473 $0 = reorder_bilist($<SS);
3475 | SimpleStatements ; Newlines ${
3476 $0 = reorder_bilist($<SS);
3478 ## ComplexStatement Grammar
3481 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3487 | SimpleStatement ${
3495 SimpleStatement -> pass ${ $0 = NULL; }$
3496 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3497 ## SimpleStatement Grammar
3499 ###### print binode cases
3503 if (b->left == NULL) // UNTESTED
3504 printf("pass"); // UNTESTED
3506 print_exec(b->left, indent, bracket); // UNTESTED
3507 if (b->right) { // UNTESTED
3508 printf("; "); // UNTESTED
3509 print_exec(b->right, indent, bracket); // UNTESTED
3512 // block, one per line
3513 if (b->left == NULL)
3514 do_indent(indent, "pass\n");
3516 print_exec(b->left, indent, bracket);
3518 print_exec(b->right, indent, bracket);
3522 ###### propagate binode cases
3525 /* If any statement returns something other than Tnone
3526 * or Tbool then all such must return same type.
3527 * As each statement may be Tnone or something else,
3528 * we must always pass NULL (unknown) down, otherwise an incorrect
3529 * error might occur. We never return Tnone unless it is
3534 for (e = b; e; e = cast(binode, e->right)) {
3535 t = propagate_types(e->left, c, ok, NULL, rules);
3536 if ((rules & Rboolok) && t == Tbool)
3538 if (t && t != Tnone && t != Tbool) {
3542 type_err(c, "error: expected %1%r, found %2",
3543 e->left, type, rules, t);
3549 ###### interp binode cases
3551 while (rvtype == Tnone &&
3554 rv = interp_exec(c, b->left, &rvtype);
3555 b = cast(binode, b->right);
3559 ### The Print statement
3561 `print` is a simple statement that takes a comma-separated list of
3562 expressions and prints the values separated by spaces and terminated
3563 by a newline. No control of formatting is possible.
3565 `print` uses `ExpressionList` to collect the expressions and stores them
3566 on the left side of a `Print` binode unlessthere is a trailing comma
3567 when the list is stored on the `right` side and no trailing newline is
3573 ##### expr precedence
3576 ###### SimpleStatement Grammar
3578 | print ExpressionList ${
3582 $0->left = reorder_bilist($<EL);
3584 | print ExpressionList , ${ {
3587 $0->right = reorder_bilist($<EL);
3597 ###### print binode cases
3600 do_indent(indent, "print");
3602 print_exec(b->right, -1, bracket);
3605 print_exec(b->left, -1, bracket);
3610 ###### propagate binode cases
3613 /* don't care but all must be consistent */
3615 b = cast(binode, b->left);
3617 b = cast(binode, b->right);
3619 propagate_types(b->left, c, ok, NULL, Rnolabel);
3620 b = cast(binode, b->right);
3624 ###### interp binode cases
3628 struct binode *b2 = cast(binode, b->left);
3630 b2 = cast(binode, b->right);
3631 for (; b2; b2 = cast(binode, b2->right)) {
3632 left = interp_exec(c, b2->left, <ype);
3633 print_value(ltype, &left);
3634 free_value(ltype, &left);
3638 if (b->right == NULL)
3644 ###### Assignment statement
3646 An assignment will assign a value to a variable, providing it hasn't
3647 been declared as a constant. The analysis phase ensures that the type
3648 will be correct so the interpreter just needs to perform the
3649 calculation. There is a form of assignment which declares a new
3650 variable as well as assigning a value. If a name is assigned before
3651 it is declared, and error will be raised as the name is created as
3652 `Tlabel` and it is illegal to assign to such names.
3658 ###### declare terminals
3661 ###### SimpleStatement Grammar
3662 | Variable = Expression ${
3668 | VariableDecl = Expression ${
3676 if ($1->var->where_set == NULL) {
3678 "Variable declared with no type or value: %v",
3688 ###### print binode cases
3691 do_indent(indent, "");
3692 print_exec(b->left, indent, bracket);
3694 print_exec(b->right, indent, bracket);
3701 struct variable *v = cast(var, b->left)->var;
3702 do_indent(indent, "");
3703 print_exec(b->left, indent, bracket);
3704 if (cast(var, b->left)->var->constant) {
3706 if (v->where_decl == v->where_set) {
3707 type_print(v->type, stdout);
3712 if (v->where_decl == v->where_set) {
3713 type_print(v->type, stdout);
3719 print_exec(b->right, indent, bracket);
3726 ###### propagate binode cases
3730 /* Both must match and not be labels,
3731 * Type must support 'dup',
3732 * For Assign, left must not be constant.
3735 t = propagate_types(b->left, c, ok, NULL,
3736 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3741 if (propagate_types(b->right, c, ok, t, 0) != t)
3742 if (b->left->type == Xvar)
3743 type_err(c, "info: variable '%v' was set as %1 here.",
3744 cast(var, b->left)->var->where_set, t, rules, NULL);
3746 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3748 propagate_types(b->left, c, ok, t,
3749 (b->op == Assign ? Rnoconstant : 0));
3751 if (t && t->dup == NULL)
3752 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3757 ###### interp binode cases
3760 lleft = linterp_exec(c, b->left, <ype);
3761 right = interp_exec(c, b->right, &rtype);
3763 free_value(ltype, lleft);
3764 dup_value(ltype, &right, lleft);
3771 struct variable *v = cast(var, b->left)->var;
3774 val = var_value(c, v);
3775 if (v->type->prepare_type)
3776 v->type->prepare_type(c, v->type, 0);
3778 right = interp_exec(c, b->right, &rtype);
3779 memcpy(val, &right, rtype->size);
3782 val_init(v->type, val);
3787 ### The `use` statement
3789 The `use` statement is the last "simple" statement. It is needed when
3790 the condition in a conditional statement is a block. `use` works much
3791 like `return` in C, but only completes the `condition`, not the whole
3797 ###### expr precedence
3800 ###### SimpleStatement Grammar
3802 $0 = new_pos(binode, $1);
3805 if ($0->right->type == Xvar) {
3806 struct var *v = cast(var, $0->right);
3807 if (v->var->type == Tnone) {
3808 /* Convert this to a label */
3811 v->var->type = Tlabel;
3812 val = global_alloc(c, Tlabel, v->var, NULL);
3818 ###### print binode cases
3821 do_indent(indent, "use ");
3822 print_exec(b->right, -1, bracket);
3827 ###### propagate binode cases
3830 /* result matches value */
3831 return propagate_types(b->right, c, ok, type, 0);
3833 ###### interp binode cases
3836 rv = interp_exec(c, b->right, &rvtype);
3839 ### The Conditional Statement
3841 This is the biggy and currently the only complex statement. This
3842 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3843 It is comprised of a number of parts, all of which are optional though
3844 set combinations apply. Each part is (usually) a key word (`then` is
3845 sometimes optional) followed by either an expression or a code block,
3846 except the `casepart` which is a "key word and an expression" followed
3847 by a code block. The code-block option is valid for all parts and,
3848 where an expression is also allowed, the code block can use the `use`
3849 statement to report a value. If the code block does not report a value
3850 the effect is similar to reporting `True`.
3852 The `else` and `case` parts, as well as `then` when combined with
3853 `if`, can contain a `use` statement which will apply to some
3854 containing conditional statement. `for` parts, `do` parts and `then`
3855 parts used with `for` can never contain a `use`, except in some
3856 subordinate conditional statement.
3858 If there is a `forpart`, it is executed first, only once.
3859 If there is a `dopart`, then it is executed repeatedly providing
3860 always that the `condpart` or `cond`, if present, does not return a non-True
3861 value. `condpart` can fail to return any value if it simply executes
3862 to completion. This is treated the same as returning `True`.
3864 If there is a `thenpart` it will be executed whenever the `condpart`
3865 or `cond` returns True (or does not return any value), but this will happen
3866 *after* `dopart` (when present).
3868 If `elsepart` is present it will be executed at most once when the
3869 condition returns `False` or some value that isn't `True` and isn't
3870 matched by any `casepart`. If there are any `casepart`s, they will be
3871 executed when the condition returns a matching value.
3873 The particular sorts of values allowed in case parts has not yet been
3874 determined in the language design, so nothing is prohibited.
3876 The various blocks in this complex statement potentially provide scope
3877 for variables as described earlier. Each such block must include the
3878 "OpenScope" nonterminal before parsing the block, and must call
3879 `var_block_close()` when closing the block.
3881 The code following "`if`", "`switch`" and "`for`" does not get its own
3882 scope, but is in a scope covering the whole statement, so names
3883 declared there cannot be redeclared elsewhere. Similarly the
3884 condition following "`while`" is in a scope the covers the body
3885 ("`do`" part) of the loop, and which does not allow conditional scope
3886 extension. Code following "`then`" (both looping and non-looping),
3887 "`else`" and "`case`" each get their own local scope.
3889 The type requirements on the code block in a `whilepart` are quite
3890 unusal. It is allowed to return a value of some identifiable type, in
3891 which case the loop aborts and an appropriate `casepart` is run, or it
3892 can return a Boolean, in which case the loop either continues to the
3893 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3894 This is different both from the `ifpart` code block which is expected to
3895 return a Boolean, or the `switchpart` code block which is expected to
3896 return the same type as the casepart values. The correct analysis of
3897 the type of the `whilepart` code block is the reason for the
3898 `Rboolok` flag which is passed to `propagate_types()`.
3900 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3901 defined. As there are two scopes which cover multiple parts - one for
3902 the whole statement and one for "while" and "do" - and as we will use
3903 the 'struct exec' to track scopes, we actually need two new types of
3904 exec. One is a `binode` for the looping part, the rest is the
3905 `cond_statement`. The `cond_statement` will use an auxilliary `struct
3906 casepart` to track a list of case parts.
3917 struct exec *action;
3918 struct casepart *next;
3920 struct cond_statement {
3922 struct exec *forpart, *condpart, *thenpart, *elsepart;
3923 struct binode *looppart;
3924 struct casepart *casepart;
3927 ###### ast functions
3929 static void free_casepart(struct casepart *cp)
3933 free_exec(cp->value);
3934 free_exec(cp->action);
3941 static void free_cond_statement(struct cond_statement *s)
3945 free_exec(s->forpart);
3946 free_exec(s->condpart);
3947 free_exec(s->looppart);
3948 free_exec(s->thenpart);
3949 free_exec(s->elsepart);
3950 free_casepart(s->casepart);
3954 ###### free exec cases
3955 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3957 ###### ComplexStatement Grammar
3958 | CondStatement ${ $0 = $<1; }$
3960 ###### expr precedence
3961 $TERM for then while do
3968 // A CondStatement must end with EOL, as does CondSuffix and
3970 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
3971 // may or may not end with EOL
3972 // WhilePart and IfPart include an appropriate Suffix
3974 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
3975 // them. WhilePart opens and closes its own scope.
3976 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
3979 $0->thenpart = $<TP;
3980 $0->looppart = $<WP;
3981 var_block_close(c, CloseSequential, $0);
3983 | ForPart OptNL WhilePart CondSuffix ${
3986 $0->looppart = $<WP;
3987 var_block_close(c, CloseSequential, $0);
3989 | WhilePart CondSuffix ${
3991 $0->looppart = $<WP;
3993 | SwitchPart OptNL CasePart CondSuffix ${
3995 $0->condpart = $<SP;
3996 $CP->next = $0->casepart;
3997 $0->casepart = $<CP;
3998 var_block_close(c, CloseSequential, $0);
4000 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4002 $0->condpart = $<SP;
4003 $CP->next = $0->casepart;
4004 $0->casepart = $<CP;
4005 var_block_close(c, CloseSequential, $0);
4007 | IfPart IfSuffix ${
4009 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4010 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4011 // This is where we close an "if" statement
4012 var_block_close(c, CloseSequential, $0);
4015 CondSuffix -> IfSuffix ${
4018 | Newlines CasePart CondSuffix ${
4020 $CP->next = $0->casepart;
4021 $0->casepart = $<CP;
4023 | CasePart CondSuffix ${
4025 $CP->next = $0->casepart;
4026 $0->casepart = $<CP;
4029 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4030 | Newlines ElsePart ${ $0 = $<EP; }$
4031 | ElsePart ${$0 = $<EP; }$
4033 ElsePart -> else OpenBlock Newlines ${
4034 $0 = new(cond_statement);
4035 $0->elsepart = $<OB;
4036 var_block_close(c, CloseElse, $0->elsepart);
4038 | else OpenScope CondStatement ${
4039 $0 = new(cond_statement);
4040 $0->elsepart = $<CS;
4041 var_block_close(c, CloseElse, $0->elsepart);
4045 CasePart -> case Expression OpenScope ColonBlock ${
4046 $0 = calloc(1,sizeof(struct casepart));
4049 var_block_close(c, CloseParallel, $0->action);
4053 // These scopes are closed in CondStatement
4054 ForPart -> for OpenBlock ${
4058 ThenPart -> then OpenBlock ${
4060 var_block_close(c, CloseSequential, $0);
4064 // This scope is closed in CondStatement
4065 WhilePart -> while UseBlock OptNL do OpenBlock ${
4070 var_block_close(c, CloseSequential, $0->right);
4071 var_block_close(c, CloseSequential, $0);
4073 | while OpenScope Expression OpenScope ColonBlock ${
4078 var_block_close(c, CloseSequential, $0->right);
4079 var_block_close(c, CloseSequential, $0);
4083 IfPart -> if UseBlock OptNL then OpenBlock ${
4086 var_block_close(c, CloseParallel, $0.thenpart);
4088 | if OpenScope Expression OpenScope ColonBlock ${
4091 var_block_close(c, CloseParallel, $0.thenpart);
4093 | if OpenScope Expression OpenScope OptNL then Block ${
4096 var_block_close(c, CloseParallel, $0.thenpart);
4100 // This scope is closed in CondStatement
4101 SwitchPart -> switch OpenScope Expression ${
4104 | switch UseBlock ${
4108 ###### print binode cases
4110 if (b->left && b->left->type == Xbinode &&
4111 cast(binode, b->left)->op == Block) {
4113 do_indent(indent, "while {\n");
4115 do_indent(indent, "while\n");
4116 print_exec(b->left, indent+1, bracket);
4118 do_indent(indent, "} do {\n");
4120 do_indent(indent, "do\n");
4121 print_exec(b->right, indent+1, bracket);
4123 do_indent(indent, "}\n");
4125 do_indent(indent, "while ");
4126 print_exec(b->left, 0, bracket);
4131 print_exec(b->right, indent+1, bracket);
4133 do_indent(indent, "}\n");
4137 ###### print exec cases
4139 case Xcond_statement:
4141 struct cond_statement *cs = cast(cond_statement, e);
4142 struct casepart *cp;
4144 do_indent(indent, "for");
4145 if (bracket) printf(" {\n"); else printf("\n");
4146 print_exec(cs->forpart, indent+1, bracket);
4149 do_indent(indent, "} then {\n");
4151 do_indent(indent, "then\n");
4152 print_exec(cs->thenpart, indent+1, bracket);
4154 if (bracket) do_indent(indent, "}\n");
4157 print_exec(cs->looppart, indent, bracket);
4161 do_indent(indent, "switch");
4163 do_indent(indent, "if");
4164 if (cs->condpart && cs->condpart->type == Xbinode &&
4165 cast(binode, cs->condpart)->op == Block) {
4170 print_exec(cs->condpart, indent+1, bracket);
4172 do_indent(indent, "}\n");
4174 do_indent(indent, "then\n");
4175 print_exec(cs->thenpart, indent+1, bracket);
4179 print_exec(cs->condpart, 0, bracket);
4185 print_exec(cs->thenpart, indent+1, bracket);
4187 do_indent(indent, "}\n");
4192 for (cp = cs->casepart; cp; cp = cp->next) {
4193 do_indent(indent, "case ");
4194 print_exec(cp->value, -1, 0);
4199 print_exec(cp->action, indent+1, bracket);
4201 do_indent(indent, "}\n");
4204 do_indent(indent, "else");
4209 print_exec(cs->elsepart, indent+1, bracket);
4211 do_indent(indent, "}\n");
4216 ###### propagate binode cases
4218 t = propagate_types(b->right, c, ok, Tnone, 0);
4219 if (!type_compat(Tnone, t, 0))
4220 *ok = 0; // UNTESTED
4221 return propagate_types(b->left, c, ok, type, rules);
4223 ###### propagate exec cases
4224 case Xcond_statement:
4226 // forpart and looppart->right must return Tnone
4227 // thenpart must return Tnone if there is a loopart,
4228 // otherwise it is like elsepart.
4230 // be bool if there is no casepart
4231 // match casepart->values if there is a switchpart
4232 // either be bool or match casepart->value if there
4234 // elsepart and casepart->action must match the return type
4235 // expected of this statement.
4236 struct cond_statement *cs = cast(cond_statement, prog);
4237 struct casepart *cp;
4239 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4240 if (!type_compat(Tnone, t, 0))
4241 *ok = 0; // UNTESTED
4244 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4245 if (!type_compat(Tnone, t, 0))
4246 *ok = 0; // UNTESTED
4248 if (cs->casepart == NULL) {
4249 propagate_types(cs->condpart, c, ok, Tbool, 0);
4250 propagate_types(cs->looppart, c, ok, Tbool, 0);
4252 /* Condpart must match case values, with bool permitted */
4254 for (cp = cs->casepart;
4255 cp && !t; cp = cp->next)
4256 t = propagate_types(cp->value, c, ok, NULL, 0);
4257 if (!t && cs->condpart)
4258 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4259 if (!t && cs->looppart)
4260 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4261 // Now we have a type (I hope) push it down
4263 for (cp = cs->casepart; cp; cp = cp->next)
4264 propagate_types(cp->value, c, ok, t, 0);
4265 propagate_types(cs->condpart, c, ok, t, Rboolok);
4266 propagate_types(cs->looppart, c, ok, t, Rboolok);
4269 // (if)then, else, and case parts must return expected type.
4270 if (!cs->looppart && !type)
4271 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4273 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4274 for (cp = cs->casepart;
4276 cp = cp->next) // UNTESTED
4277 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4280 propagate_types(cs->thenpart, c, ok, type, rules);
4281 propagate_types(cs->elsepart, c, ok, type, rules);
4282 for (cp = cs->casepart; cp ; cp = cp->next)
4283 propagate_types(cp->action, c, ok, type, rules);
4289 ###### interp binode cases
4291 // This just performs one iterration of the loop
4292 rv = interp_exec(c, b->left, &rvtype);
4293 if (rvtype == Tnone ||
4294 (rvtype == Tbool && rv.bool != 0))
4295 // cnd is Tnone or Tbool, doesn't need to be freed
4296 interp_exec(c, b->right, NULL);
4299 ###### interp exec cases
4300 case Xcond_statement:
4302 struct value v, cnd;
4303 struct type *vtype, *cndtype;
4304 struct casepart *cp;
4305 struct cond_statement *cs = cast(cond_statement, e);
4308 interp_exec(c, cs->forpart, NULL);
4310 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4311 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4312 interp_exec(c, cs->thenpart, NULL);
4314 cnd = interp_exec(c, cs->condpart, &cndtype);
4315 if ((cndtype == Tnone ||
4316 (cndtype == Tbool && cnd.bool != 0))) {
4317 // cnd is Tnone or Tbool, doesn't need to be freed
4318 rv = interp_exec(c, cs->thenpart, &rvtype);
4319 // skip else (and cases)
4323 for (cp = cs->casepart; cp; cp = cp->next) {
4324 v = interp_exec(c, cp->value, &vtype);
4325 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4326 free_value(vtype, &v);
4327 free_value(cndtype, &cnd);
4328 rv = interp_exec(c, cp->action, &rvtype);
4331 free_value(vtype, &v);
4333 free_value(cndtype, &cnd);
4335 rv = interp_exec(c, cs->elsepart, &rvtype);
4342 ### Top level structure
4344 All the language elements so far can be used in various places. Now
4345 it is time to clarify what those places are.
4347 At the top level of a file there will be a number of declarations.
4348 Many of the things that can be declared haven't been described yet,
4349 such as functions, procedures, imports, and probably more.
4350 For now there are two sorts of things that can appear at the top
4351 level. They are predefined constants, `struct` types, and the `main`
4352 function. While the syntax will allow the `main` function to appear
4353 multiple times, that will trigger an error if it is actually attempted.
4355 The various declarations do not return anything. They store the
4356 various declarations in the parse context.
4358 ###### Parser: grammar
4361 Ocean -> OptNL DeclarationList
4363 ## declare terminals
4370 DeclarationList -> Declaration
4371 | DeclarationList Declaration
4373 Declaration -> ERROR Newlines ${
4374 tok_err(c, // UNTESTED
4375 "error: unhandled parse error", &$1);
4381 ## top level grammar
4385 ### The `const` section
4387 As well as being defined in with the code that uses them, constants
4388 can be declared at the top level. These have full-file scope, so they
4389 are always `InScope`. The value of a top level constant can be given
4390 as an expression, and this is evaluated immediately rather than in the
4391 later interpretation stage. Once we add functions to the language, we
4392 will need rules concern which, if any, can be used to define a top
4395 Constants are defined in a section that starts with the reserved word
4396 `const` and then has a block with a list of assignment statements.
4397 For syntactic consistency, these must use the double-colon syntax to
4398 make it clear that they are constants. Type can also be given: if
4399 not, the type will be determined during analysis, as with other
4402 As the types constants are inserted at the head of a list, printing
4403 them in the same order that they were read is not straight forward.
4404 We take a quadratic approach here and count the number of constants
4405 (variables of depth 0), then count down from there, each time
4406 searching through for the Nth constant for decreasing N.
4408 ###### top level grammar
4412 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4413 | const { SimpleConstList } Newlines
4414 | const IN OptNL ConstList OUT Newlines
4415 | const SimpleConstList Newlines
4417 ConstList -> ConstList SimpleConstLine
4419 SimpleConstList -> SimpleConstList ; Const
4422 SimpleConstLine -> SimpleConstList Newlines
4423 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4426 CType -> Type ${ $0 = $<1; }$
4429 Const -> IDENTIFIER :: CType = Expression ${ {
4433 v = var_decl(c, $1.txt);
4435 struct var *var = new_pos(var, $1);
4436 v->where_decl = var;
4441 v = var_ref(c, $1.txt);
4442 tok_err(c, "error: name already declared", &$1);
4443 type_err(c, "info: this is where '%v' was first declared",
4444 v->where_decl, NULL, 0, NULL);
4448 propagate_types($5, c, &ok, $3, 0);
4453 struct value res = interp_exec(c, $5, &v->type);
4454 global_alloc(c, v->type, v, &res);
4458 ###### print const decls
4463 while (target != 0) {
4465 for (v = context.in_scope; v; v=v->in_scope)
4466 if (v->depth == 0) {
4477 struct value *val = var_value(&context, v);
4478 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4479 type_print(v->type, stdout);
4481 if (v->type == Tstr)
4483 print_value(v->type, val);
4484 if (v->type == Tstr)
4492 ### Finally the whole `main` function.
4494 An Ocean program can currently have only one function - `main` - and
4495 that must exist. It expects an array of strings with a provided size.
4496 Following this is a `block` which is the code to execute.
4498 As this is the top level, several things are handled a bit
4500 The function is not interpreted by `interp_exec` as that isn't
4501 passed the argument list which the program requires. Similarly type
4502 analysis is a bit more interesting at this level.
4504 ###### top level grammar
4506 DeclareFunction -> MainFunction ${ {
4508 type_err(c, "\"main\" defined a second time",
4514 ###### print binode cases
4516 do_indent(indent, "func main(");
4517 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4518 struct variable *v = cast(var, b2->left)->var;
4520 print_exec(b2->left, 0, 0);
4522 type_print(v->type, stdout);
4528 print_exec(b->right, indent+1, bracket);
4530 do_indent(indent, "}\n");
4533 ###### propagate binode cases
4534 case Func: abort(); // NOTEST
4536 ###### core functions
4538 static int analyse_prog(struct exec *prog, struct parse_context *c)
4540 struct binode *bp = cast(binode, prog);
4544 struct type *argv_type;
4545 struct text argv_type_name = { " argv", 5 };
4550 argv_type = add_type(c, argv_type_name, &array_prototype);
4551 argv_type->array.member = Tstr;
4552 argv_type->array.unspec = 1;
4554 for (b = cast(binode, bp->left); b; b = cast(binode, b->right)) {
4558 propagate_types(b->left, c, &ok, argv_type, 0);
4560 default: /* invalid */ // NOTEST
4561 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4567 propagate_types(bp->right, c, &ok, Tnone, 0);
4572 /* Make sure everything is still consistent */
4573 propagate_types(bp->right, c, &ok, Tnone, 0);
4575 return 0; // UNTESTED
4580 static void interp_prog(struct parse_context *c, struct exec *prog,
4581 int argc, char **argv)
4583 struct binode *p = cast(binode, prog);
4591 al = cast(binode, p->left);
4593 struct var *v = cast(var, al->left);
4594 struct value *vl = var_value(c, v->var);
4604 mpq_set_ui(argcq, argc, 1);
4605 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4606 t->prepare_type(c, t, 0);
4607 array_init(v->var->type, vl);
4608 for (i = 0; i < argc; i++) {
4609 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4612 arg.str.txt = argv[i];
4613 arg.str.len = strlen(argv[i]);
4614 free_value(Tstr, vl2);
4615 dup_value(Tstr, &arg, vl2);
4619 al = cast(binode, al->right);
4621 v = interp_exec(c, p, &vtype);
4622 free_value(vtype, &v);
4625 ###### interp binode cases
4627 rv = interp_exec(c, b->right, &rvtype);
4630 ## And now to test it out.
4632 Having a language requires having a "hello world" program. I'll
4633 provide a little more than that: a program that prints "Hello world"
4634 finds the GCD of two numbers, prints the first few elements of
4635 Fibonacci, performs a binary search for a number, and a few other
4636 things which will likely grow as the languages grows.
4638 ###### File: oceani.mk
4641 @echo "===== DEMO ====="
4642 ./oceani --section "demo: hello" oceani.mdc 55 33
4648 four ::= 2 + 2 ; five ::= 10/2
4649 const pie ::= "I like Pie";
4650 cake ::= "The cake is"
4661 print "Hello World, what lovely oceans you have!"
4662 print "Are there", five, "?"
4663 print pi, pie, "but", cake
4665 A := $argv[1]; B := $argv[2]
4667 /* When a variable is defined in both branches of an 'if',
4668 * and used afterwards, the variables are merged.
4674 print "Is", A, "bigger than", B,"? ", bigger
4675 /* If a variable is not used after the 'if', no
4676 * merge happens, so types can be different
4679 double:string = "yes"
4680 print A, "is more than twice", B, "?", double
4683 print "double", B, "is", double
4688 if a > 0 and then b > 0:
4694 print "GCD of", A, "and", B,"is", a
4696 print a, "is not positive, cannot calculate GCD"
4698 print b, "is not positive, cannot calculate GCD"
4703 print "Fibonacci:", f1,f2,
4704 then togo = togo - 1
4712 /* Binary search... */
4717 mid := (lo + hi) / 2
4730 print "Yay, I found", target
4732 print "Closest I found was", lo
4737 // "middle square" PRNG. Not particularly good, but one my
4738 // Dad taught me - the first one I ever heard of.
4739 for i:=1; then i = i + 1; while i < size:
4740 n := list[i-1] * list[i-1]
4741 list[i] = (n / 100) % 10 000
4743 print "Before sort:",
4744 for i:=0; then i = i + 1; while i < size:
4748 for i := 1; then i=i+1; while i < size:
4749 for j:=i-1; then j=j-1; while j >= 0:
4750 if list[j] > list[j+1]:
4754 print " After sort:",
4755 for i:=0; then i = i + 1; while i < size:
4759 if 1 == 2 then print "yes"; else print "no"
4763 bob.alive = (bob.name == "Hello")
4764 print "bob", "is" if bob.alive else "isn't", "alive"