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;
1160 static struct variable *var_ref(struct parse_context *c, struct text s)
1162 struct binding *b = find_binding(c, s);
1163 struct variable *v = b->var;
1164 struct variable *v2;
1166 switch (v ? v->scope : OutScope) {
1169 /* Caller will report the error */
1172 /* All CondScope variables of this name need to be merged
1173 * and become InScope
1175 v->depth = v->min_depth;
1177 for (v2 = v->previous;
1178 v2 && v2->scope == CondScope;
1180 variable_merge(v, v2);
1188 static void var_block_close(struct parse_context *c, enum closetype ct,
1191 /* Close off all variables that are in_scope.
1192 * Some variables in c->scope may already be not-in-scope,
1193 * such as when a PendingScope variable is hidden by a new
1194 * variable with the same name.
1195 * So we check for v->name->var != v and drop them.
1196 * If we choose to make a variable OutScope, we drop it
1199 struct variable *v, **vp, *v2;
1202 for (vp = &c->in_scope;
1203 (v = *vp) && v->min_depth > c->scope_depth;
1204 (v->scope == OutScope || v->name->var != v)
1205 ? (*vp = v->in_scope, 0)
1206 : ( vp = &v->in_scope, 0)) {
1207 v->min_depth = c->scope_depth;
1208 if (v->name->var != v)
1209 /* This is still in scope, but we haven't just
1213 v->min_depth = c->scope_depth;
1214 if (v->scope == InScope) {
1215 /* This variable gets cleaned up when 'e' finishes */
1216 variable_unlink_exec(v);
1217 v->cleanup_exec = e;
1218 v->next_free = e->to_free;
1223 case CloseParallel: /* handle PendingScope */
1227 if (c->scope_stack->child_count == 1)
1228 /* first among parallel branches */
1229 v->scope = PendingScope;
1230 else if (v->previous &&
1231 v->previous->scope == PendingScope)
1232 /* all previous branches used name */
1233 v->scope = PendingScope;
1234 else if (v->type == Tlabel)
1235 /* Labels remain pending even when not used */
1236 v->scope = PendingScope; // UNTESTED
1238 v->scope = OutScope;
1239 if (ct == CloseElse) {
1240 /* All Pending variables with this name
1241 * are now Conditional */
1243 v2 && v2->scope == PendingScope;
1245 v2->scope = CondScope;
1249 /* Not possible as it would require
1250 * parallel scope to be nested immediately
1251 * in a parallel scope, and that never
1255 /* Not possible as we already tested for
1261 case CloseSequential:
1262 if (v->type == Tlabel)
1263 v->scope = PendingScope;
1266 v->scope = OutScope;
1269 /* There was no 'else', so we can only become
1270 * conditional if we know the cases were exhaustive,
1271 * and that doesn't mean anything yet.
1272 * So only labels become conditional..
1275 v2 && v2->scope == PendingScope;
1277 if (v2->type == Tlabel)
1278 v2->scope = CondScope;
1280 v2->scope = OutScope;
1283 case OutScope: break;
1292 The value of a variable is store separately from the variable, on an
1293 analogue of a stack frame. There are (currently) two frames that can be
1294 active. A global frame which currently only stores constants, and a
1295 stacked frame which stores local variables. Each variable knows if it
1296 is global or not, and what its index into the frame is.
1298 Values in the global frame are known immediately they are relevant, so
1299 the frame needs to be reallocated as it grows so it can store those
1300 values. The local frame doesn't get values until the interpreted phase
1301 is started, so there is no need to allocate until the size is known.
1303 We initialize the `frame_pos` to an impossible value, so that we can
1304 tell if it was set or not later.
1306 ###### variable fields
1310 ###### variable init
1313 ###### parse context
1315 short global_size, global_alloc;
1317 void *global, *local;
1319 ###### ast functions
1321 static struct value *var_value(struct parse_context *c, struct variable *v)
1324 if (!c->local || !v->type)
1325 return NULL; // NOTEST
1326 if (v->frame_pos + v->type->size > c->local_size) {
1327 printf("INVALID frame_pos\n"); // NOTEST
1330 return c->local + v->frame_pos;
1332 if (c->global_size > c->global_alloc) {
1333 int old = c->global_alloc;
1334 c->global_alloc = (c->global_size | 1023) + 1024;
1335 c->global = realloc(c->global, c->global_alloc);
1336 memset(c->global + old, 0, c->global_alloc - old);
1338 return c->global + v->frame_pos;
1341 static struct value *global_alloc(struct parse_context *c, struct type *t,
1342 struct variable *v, struct value *init)
1345 struct variable scratch;
1347 if (t->prepare_type)
1348 t->prepare_type(c, t, 1); // NOTEST
1350 if (c->global_size & (t->align - 1))
1351 c->global_size = (c->global_size + t->align) & ~(t->align-1); // UNTESTED
1356 v->frame_pos = c->global_size;
1358 c->global_size += v->type->size;
1359 ret = var_value(c, v);
1361 memcpy(ret, init, t->size);
1367 As global values are found -- struct field initializers, labels etc --
1368 `global_alloc()` is called to record the value in the global frame.
1370 When the program is fully parsed, we need to walk the list of variables
1371 to find any that weren't merged away and that aren't global, and to
1372 calculate the frame size and assign a frame position for each variable.
1373 For this we have `scope_finalize()`.
1375 ###### ast functions
1377 static void scope_finalize(struct parse_context *c)
1381 for (b = c->varlist; b; b = b->next) {
1383 for (v = b->var; v; v = v->previous) {
1384 struct type *t = v->type;
1389 if (c->local_size & (t->align - 1))
1390 c->local_size = (c->local_size + t->align) & ~(t->align-1);
1391 v->frame_pos = c->local_size;
1392 c->local_size += v->type->size;
1395 c->local = calloc(1, c->local_size);
1398 ###### free context storage
1399 free(context.global);
1400 free(context.local);
1404 Executables can be lots of different things. In many cases an
1405 executable is just an operation combined with one or two other
1406 executables. This allows for expressions and lists etc. Other times an
1407 executable is something quite specific like a constant or variable name.
1408 So we define a `struct exec` to be a general executable with a type, and
1409 a `struct binode` which is a subclass of `exec`, forms a node in a
1410 binary tree, and holds an operation. There will be other subclasses,
1411 and to access these we need to be able to `cast` the `exec` into the
1412 various other types. The first field in any `struct exec` is the type
1413 from the `exec_types` enum.
1416 #define cast(structname, pointer) ({ \
1417 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1418 if (__mptr && *__mptr != X##structname) abort(); \
1419 (struct structname *)( (char *)__mptr);})
1421 #define new(structname) ({ \
1422 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1423 __ptr->type = X##structname; \
1424 __ptr->line = -1; __ptr->column = -1; \
1427 #define new_pos(structname, token) ({ \
1428 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1429 __ptr->type = X##structname; \
1430 __ptr->line = token.line; __ptr->column = token.col; \
1439 enum exec_types type;
1448 struct exec *left, *right;
1451 ###### ast functions
1453 static int __fput_loc(struct exec *loc, FILE *f)
1457 if (loc->line >= 0) {
1458 fprintf(f, "%d:%d: ", loc->line, loc->column);
1461 if (loc->type == Xbinode)
1462 return __fput_loc(cast(binode,loc)->left, f) ||
1463 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1466 static void fput_loc(struct exec *loc, FILE *f)
1468 if (!__fput_loc(loc, f))
1469 fprintf(f, "??:??: "); // NOTEST
1472 Each different type of `exec` node needs a number of functions defined,
1473 a bit like methods. We must be able to free it, print it, analyse it
1474 and execute it. Once we have specific `exec` types we will need to
1475 parse them too. Let's take this a bit more slowly.
1479 The parser generator requires a `free_foo` function for each struct
1480 that stores attributes and they will often be `exec`s and subtypes
1481 there-of. So we need `free_exec` which can handle all the subtypes,
1482 and we need `free_binode`.
1484 ###### ast functions
1486 static void free_binode(struct binode *b)
1491 free_exec(b->right);
1495 ###### core functions
1496 static void free_exec(struct exec *e)
1505 ###### forward decls
1507 static void free_exec(struct exec *e);
1509 ###### free exec cases
1510 case Xbinode: free_binode(cast(binode, e)); break;
1514 Printing an `exec` requires that we know the current indent level for
1515 printing line-oriented components. As will become clear later, we
1516 also want to know what sort of bracketing to use.
1518 ###### ast functions
1520 static void do_indent(int i, char *str)
1527 ###### core functions
1528 static void print_binode(struct binode *b, int indent, int bracket)
1532 ## print binode cases
1536 static void print_exec(struct exec *e, int indent, int bracket)
1542 print_binode(cast(binode, e), indent, bracket); break;
1547 do_indent(indent, "/* FREE");
1548 for (v = e->to_free; v; v = v->next_free) {
1549 printf(" %.*s", v->name->name.len, v->name->name.txt);
1550 if (v->frame_pos >= 0)
1551 printf("(%d+%d)", v->frame_pos,
1552 v->type ? v->type->size:0);
1558 ###### forward decls
1560 static void print_exec(struct exec *e, int indent, int bracket);
1564 As discussed, analysis involves propagating type requirements around the
1565 program and looking for errors.
1567 So `propagate_types` is passed an expected type (being a `struct type`
1568 pointer together with some `val_rules` flags) that the `exec` is
1569 expected to return, and returns the type that it does return, either
1570 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1571 by reference. It is set to `0` when an error is found, and `2` when
1572 any change is made. If it remains unchanged at `1`, then no more
1573 propagation is needed.
1577 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1581 if (rules & Rnolabel)
1582 fputs(" (labels not permitted)", stderr);
1585 ###### core functions
1587 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1588 struct type *type, int rules);
1589 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1590 struct type *type, int rules)
1597 switch (prog->type) {
1600 struct binode *b = cast(binode, prog);
1602 ## propagate binode cases
1606 ## propagate exec cases
1611 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1612 struct type *type, int rules)
1614 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1623 Interpreting an `exec` doesn't require anything but the `exec`. State
1624 is stored in variables and each variable will be directly linked from
1625 within the `exec` tree. The exception to this is the `main` function
1626 which needs to look at command line arguments. This function will be
1627 interpreted separately.
1629 Each `exec` can return a value combined with a type in `struct lrval`.
1630 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1631 the location of a value, which can be updated, in `lval`. Others will
1632 set `lval` to NULL indicating that there is a value of appropriate type
1635 ###### core functions
1639 struct value rval, *lval;
1642 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1644 static struct value interp_exec(struct parse_context *c, struct exec *e,
1645 struct type **typeret)
1647 struct lrval ret = _interp_exec(c, e);
1649 if (!ret.type) abort();
1651 *typeret = ret.type;
1653 dup_value(ret.type, ret.lval, &ret.rval);
1657 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1658 struct type **typeret)
1660 struct lrval ret = _interp_exec(c, e);
1663 *typeret = ret.type;
1665 free_value(ret.type, &ret.rval);
1669 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1672 struct value rv = {}, *lrv = NULL;
1673 struct type *rvtype;
1675 rvtype = ret.type = Tnone;
1685 struct binode *b = cast(binode, e);
1686 struct value left, right, *lleft;
1687 struct type *ltype, *rtype;
1688 ltype = rtype = Tnone;
1690 ## interp binode cases
1692 free_value(ltype, &left);
1693 free_value(rtype, &right);
1696 ## interp exec cases
1701 ## interp exec cleanup
1707 Now that we have the shape of the interpreter in place we can add some
1708 complex types and connected them in to the data structures and the
1709 different phases of parse, analyse, print, interpret.
1711 Thus far we have arrays and structs.
1715 Arrays can be declared by giving a size and a type, as `[size]type' so
1716 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1717 size can be either a literal number, or a named constant. Some day an
1718 arbitrary expression will be supported.
1720 As a formal parameter to a function, the array can be declared with a
1721 new variable as the size: `name:[size::number]string`. The `size`
1722 variable is set to the size of the array and must be a constant. As
1723 `number` is the only supported type, it can be left out:
1724 `name:[size::]string`.
1726 Arrays cannot be assigned. When pointers are introduced we will also
1727 introduce array slices which can refer to part or all of an array -
1728 the assignment syntax will create a slice. For now, an array can only
1729 ever be referenced by the name it is declared with. It is likely that
1730 a "`copy`" primitive will eventually be define which can be used to
1731 make a copy of an array with controllable recursive depth.
1733 For now we have two sorts of array, those with fixed size either because
1734 it is given as a literal number or because it is a struct member (which
1735 cannot have a runtime-changing size), and those with a size that is
1736 determined at runtime - local variables with a const size. The former
1737 have their size calculated at parse time, the latter at run time.
1739 For the latter type, the `size` field of the type is the size of a
1740 pointer, and the array is reallocated every time it comes into scope.
1742 We differentiate struct fields with a const size from local variables
1743 with a const size by whether they are prepared at parse time or not.
1745 ###### type union fields
1748 int unspec; // size is unspecified - vsize must be set.
1751 struct variable *vsize;
1752 struct type *member;
1755 ###### value union fields
1756 void *array; // used if not static_size
1758 ###### value functions
1760 static void array_prepare_type(struct parse_context *c, struct type *type,
1763 struct value *vsize;
1765 if (!type->array.vsize || type->array.static_size)
1768 vsize = var_value(c, type->array.vsize);
1770 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1771 type->array.size = mpz_get_si(q);
1775 type->array.static_size = 1;
1776 type->size = type->array.size * type->array.member->size;
1777 type->align = type->array.member->align;
1781 static void array_init(struct type *type, struct value *val)
1784 void *ptr = val->ptr;
1788 if (!type->array.static_size) {
1789 val->array = calloc(type->array.size,
1790 type->array.member->size);
1793 for (i = 0; i < type->array.size; i++) {
1795 v = (void*)ptr + i * type->array.member->size;
1796 val_init(type->array.member, v);
1800 static void array_free(struct type *type, struct value *val)
1803 void *ptr = val->ptr;
1805 if (!type->array.static_size)
1807 for (i = 0; i < type->array.size; i++) {
1809 v = (void*)ptr + i * type->array.member->size;
1810 free_value(type->array.member, v);
1812 if (!type->array.static_size)
1816 static int array_compat(struct type *require, struct type *have)
1818 if (have->compat != require->compat)
1819 return 0; // UNTESTED
1820 /* Both are arrays, so we can look at details */
1821 if (!type_compat(require->array.member, have->array.member, 0))
1823 if (have->array.unspec && require->array.unspec) {
1824 if (have->array.vsize && require->array.vsize &&
1825 have->array.vsize != require->array.vsize) // UNTESTED
1826 /* sizes might not be the same */
1827 return 0; // UNTESTED
1830 if (have->array.unspec || require->array.unspec)
1831 return 1; // UNTESTED
1832 if (require->array.vsize == NULL && have->array.vsize == NULL)
1833 return require->array.size == have->array.size;
1835 return require->array.vsize == have->array.vsize; // UNTESTED
1838 static void array_print_type(struct type *type, FILE *f)
1841 if (type->array.vsize) {
1842 struct binding *b = type->array.vsize->name;
1843 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1844 type->array.unspec ? "::" : "");
1846 fprintf(f, "%d]", type->array.size);
1847 type_print(type->array.member, f);
1850 static struct type array_prototype = {
1852 .prepare_type = array_prepare_type,
1853 .print_type = array_print_type,
1854 .compat = array_compat,
1856 .size = sizeof(void*),
1857 .align = sizeof(void*),
1860 ###### declare terminals
1865 | [ NUMBER ] Type ${ {
1868 struct text noname = { "", 0 };
1871 $0 = t = add_type(c, noname, &array_prototype);
1872 t->array.member = $<4;
1873 t->array.vsize = NULL;
1874 if (number_parse(num, tail, $2.txt) == 0)
1875 tok_err(c, "error: unrecognised number", &$2);
1877 tok_err(c, "error: unsupported number suffix", &$2);
1879 t->array.size = mpz_get_ui(mpq_numref(num));
1880 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1881 tok_err(c, "error: array size must be an integer",
1883 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1884 tok_err(c, "error: array size is too large",
1888 t->array.static_size = 1;
1889 t->size = t->array.size * t->array.member->size;
1890 t->align = t->array.member->align;
1893 | [ IDENTIFIER ] Type ${ {
1894 struct variable *v = var_ref(c, $2.txt);
1895 struct text noname = { "", 0 };
1898 tok_err(c, "error: name undeclared", &$2);
1899 else if (!v->constant)
1900 tok_err(c, "error: array size must be a constant", &$2);
1902 $0 = add_type(c, noname, &array_prototype);
1903 $0->array.member = $<4;
1905 $0->array.vsize = v;
1910 OptType -> Type ${ $0 = $<1; }$
1913 ###### formal type grammar
1915 | [ IDENTIFIER :: OptType ] Type ${ {
1916 struct variable *v = var_decl(c, $ID.txt);
1917 struct text noname = { "", 0 };
1923 $0 = add_type(c, noname, &array_prototype);
1924 $0->array.member = $<6;
1926 $0->array.unspec = 1;
1927 $0->array.vsize = v;
1933 ###### variable grammar
1935 | Variable [ Expression ] ${ {
1936 struct binode *b = new(binode);
1943 ###### print binode cases
1945 print_exec(b->left, -1, bracket);
1947 print_exec(b->right, -1, bracket);
1951 ###### propagate binode cases
1953 /* left must be an array, right must be a number,
1954 * result is the member type of the array
1956 propagate_types(b->right, c, ok, Tnum, 0);
1957 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1958 if (!t || t->compat != array_compat) {
1959 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1962 if (!type_compat(type, t->array.member, rules)) {
1963 type_err(c, "error: have %1 but need %2", prog,
1964 t->array.member, rules, type);
1966 return t->array.member;
1970 ###### interp binode cases
1976 lleft = linterp_exec(c, b->left, <ype);
1977 right = interp_exec(c, b->right, &rtype);
1979 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1983 if (ltype->array.static_size)
1986 ptr = *(void**)lleft;
1987 rvtype = ltype->array.member;
1988 if (i >= 0 && i < ltype->array.size)
1989 lrv = ptr + i * rvtype->size;
1991 val_init(ltype->array.member, &rv);
1998 A `struct` is a data-type that contains one or more other data-types.
1999 It differs from an array in that each member can be of a different
2000 type, and they are accessed by name rather than by number. Thus you
2001 cannot choose an element by calculation, you need to know what you
2004 The language makes no promises about how a given structure will be
2005 stored in memory - it is free to rearrange fields to suit whatever
2006 criteria seems important.
2008 Structs are declared separately from program code - they cannot be
2009 declared in-line in a variable declaration like arrays can. A struct
2010 is given a name and this name is used to identify the type - the name
2011 is not prefixed by the word `struct` as it would be in C.
2013 Structs are only treated as the same if they have the same name.
2014 Simply having the same fields in the same order is not enough. This
2015 might change once we can create structure initializers from a list of
2018 Each component datum is identified much like a variable is declared,
2019 with a name, one or two colons, and a type. The type cannot be omitted
2020 as there is no opportunity to deduce the type from usage. An initial
2021 value can be given following an equals sign, so
2023 ##### Example: a struct type
2029 would declare a type called "complex" which has two number fields,
2030 each initialised to zero.
2032 Struct will need to be declared separately from the code that uses
2033 them, so we will need to be able to print out the declaration of a
2034 struct when reprinting the whole program. So a `print_type_decl` type
2035 function will be needed.
2037 ###### type union fields
2049 ###### type functions
2050 void (*print_type_decl)(struct type *type, FILE *f);
2052 ###### value functions
2054 static void structure_init(struct type *type, struct value *val)
2058 for (i = 0; i < type->structure.nfields; i++) {
2060 v = (void*) val->ptr + type->structure.fields[i].offset;
2061 if (type->structure.fields[i].init)
2062 dup_value(type->structure.fields[i].type,
2063 type->structure.fields[i].init,
2066 val_init(type->structure.fields[i].type, v);
2070 static void structure_free(struct type *type, struct value *val)
2074 for (i = 0; i < type->structure.nfields; i++) {
2076 v = (void*)val->ptr + type->structure.fields[i].offset;
2077 free_value(type->structure.fields[i].type, v);
2081 static void structure_free_type(struct type *t)
2084 for (i = 0; i < t->structure.nfields; i++)
2085 if (t->structure.fields[i].init) {
2086 free_value(t->structure.fields[i].type,
2087 t->structure.fields[i].init);
2089 free(t->structure.fields);
2092 static struct type structure_prototype = {
2093 .init = structure_init,
2094 .free = structure_free,
2095 .free_type = structure_free_type,
2096 .print_type_decl = structure_print_type,
2110 ###### free exec cases
2112 free_exec(cast(fieldref, e)->left);
2116 ###### declare terminals
2119 ###### variable grammar
2121 | Variable . IDENTIFIER ${ {
2122 struct fieldref *fr = new_pos(fieldref, $2);
2129 ###### print exec cases
2133 struct fieldref *f = cast(fieldref, e);
2134 print_exec(f->left, -1, bracket);
2135 printf(".%.*s", f->name.len, f->name.txt);
2139 ###### ast functions
2140 static int find_struct_index(struct type *type, struct text field)
2143 for (i = 0; i < type->structure.nfields; i++)
2144 if (text_cmp(type->structure.fields[i].name, field) == 0)
2149 ###### propagate exec cases
2153 struct fieldref *f = cast(fieldref, prog);
2154 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2157 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2159 else if (st->init != structure_init)
2160 type_err(c, "error: field reference attempted on %1, not a struct",
2161 f->left, st, 0, NULL);
2162 else if (f->index == -2) {
2163 f->index = find_struct_index(st, f->name);
2165 type_err(c, "error: cannot find requested field in %1",
2166 f->left, st, 0, NULL);
2168 if (f->index >= 0) {
2169 struct type *ft = st->structure.fields[f->index].type;
2170 if (!type_compat(type, ft, rules))
2171 type_err(c, "error: have %1 but need %2", prog,
2178 ###### interp exec cases
2181 struct fieldref *f = cast(fieldref, e);
2183 struct value *lleft = linterp_exec(c, f->left, <ype);
2184 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2185 rvtype = ltype->structure.fields[f->index].type;
2191 struct fieldlist *prev;
2195 ###### ast functions
2196 static void free_fieldlist(struct fieldlist *f)
2200 free_fieldlist(f->prev);
2202 free_value(f->f.type, f->f.init); // UNTESTED
2203 free(f->f.init); // UNTESTED
2208 ###### top level grammar
2209 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2211 add_type(c, $2.txt, &structure_prototype);
2213 struct fieldlist *f;
2215 for (f = $3; f; f=f->prev)
2218 t->structure.nfields = cnt;
2219 t->structure.fields = calloc(cnt, sizeof(struct field));
2222 int a = f->f.type->align;
2224 t->structure.fields[cnt] = f->f;
2225 if (t->size & (a-1))
2226 t->size = (t->size | (a-1)) + 1;
2227 t->structure.fields[cnt].offset = t->size;
2228 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2237 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2238 | { SimpleFieldList } ${ $0 = $<SFL; }$
2239 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2240 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2242 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2243 | FieldLines SimpleFieldList Newlines ${
2248 SimpleFieldList -> Field ${ $0 = $<F; }$
2249 | SimpleFieldList ; Field ${
2253 | SimpleFieldList ; ${
2256 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2258 Field -> IDENTIFIER : Type = Expression ${ {
2261 $0 = calloc(1, sizeof(struct fieldlist));
2262 $0->f.name = $1.txt;
2267 propagate_types($<5, c, &ok, $3, 0);
2270 c->parse_error = 1; // UNTESTED
2272 struct value vl = interp_exec(c, $5, NULL);
2273 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2276 | IDENTIFIER : Type ${
2277 $0 = calloc(1, sizeof(struct fieldlist));
2278 $0->f.name = $1.txt;
2280 if ($0->f.type->prepare_type)
2281 $0->f.type->prepare_type(c, $0->f.type, 1);
2284 ###### forward decls
2285 static void structure_print_type(struct type *t, FILE *f);
2287 ###### value functions
2288 static void structure_print_type(struct type *t, FILE *f) // UNTESTED
2292 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2294 for (i = 0; i < t->structure.nfields; i++) {
2295 struct field *fl = t->structure.fields + i;
2296 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2297 type_print(fl->type, f);
2298 if (fl->type->print && fl->init) {
2300 if (fl->type == Tstr)
2301 fprintf(f, "\""); // UNTESTED
2302 print_value(fl->type, fl->init);
2303 if (fl->type == Tstr)
2304 fprintf(f, "\""); // UNTESTED
2310 ###### print type decls
2312 struct type *t; // UNTESTED
2315 while (target != 0) {
2317 for (t = context.typelist; t ; t=t->next)
2318 if (t->print_type_decl) {
2327 t->print_type_decl(t, stdout);
2335 A function is a named chunk of code which can be passed parameters and
2336 can return results. Each function has an implicit type which includes
2337 the set of parameters and the return value. As yet these types cannot
2338 be declared separate from the function itself.
2340 In fact, only one function is currently possible - `main`. `main` is
2341 passed an array of strings together with the size of the array, and
2342 doesn't return anything. The strings are command line arguments.
2344 The parameters can be specified either in parentheses as a list, such as
2346 ##### Example: function 1
2348 func main(av:[ac::number]string)
2351 or as an indented list of one parameter per line
2353 ##### Example: function 2
2356 argv:[argc::number]string
2360 For constructing these lists we use a `List` binode, which will be
2361 further detailed when Expression Lists are introduced.
2371 MainFunction -> func main ( OpenScope Args ) Block Newlines ${
2374 $0->left = reorder_bilist($<Ar);
2376 var_block_close(c, CloseSequential, $0);
2377 if (c->scope_stack && !c->parse_error) abort();
2379 | func main IN OpenScope OptNL Args OUT OptNL do Block Newlines ${
2382 $0->left = reorder_bilist($<Ar);
2384 var_block_close(c, CloseSequential, $0);
2385 if (c->scope_stack && !c->parse_error) abort();
2387 | func main NEWLINE OpenScope OptNL do Block Newlines ${
2392 var_block_close(c, CloseSequential, $0);
2393 if (c->scope_stack && !c->parse_error) abort();
2396 Args -> ${ $0 = NULL; }$
2397 | Varlist ${ $0 = $<1; }$
2398 | Varlist ; ${ $0 = $<1; }$
2399 | Varlist NEWLINE ${ $0 = $<1; }$
2401 Varlist -> Varlist ; ArgDecl ${ // UNTESTED
2415 ArgDecl -> IDENTIFIER : FormalType ${ {
2416 struct variable *v = var_decl(c, $1.txt);
2422 ## Executables: the elements of code
2424 Each code element needs to be parsed, printed, analysed,
2425 interpreted, and freed. There are several, so let's just start with
2426 the easy ones and work our way up.
2430 We have already met values as separate objects. When manifest
2431 constants appear in the program text, that must result in an executable
2432 which has a constant value. So the `val` structure embeds a value in
2445 ###### ast functions
2446 struct val *new_val(struct type *T, struct token tk)
2448 struct val *v = new_pos(val, tk);
2459 $0 = new_val(Tbool, $1);
2463 $0 = new_val(Tbool, $1);
2467 $0 = new_val(Tnum, $1);
2470 if (number_parse($0->val.num, tail, $1.txt) == 0)
2471 mpq_init($0->val.num); // UNTESTED
2473 tok_err(c, "error: unsupported number suffix",
2478 $0 = new_val(Tstr, $1);
2481 string_parse(&$1, '\\', &$0->val.str, tail);
2483 tok_err(c, "error: unsupported string suffix",
2488 $0 = new_val(Tstr, $1);
2491 string_parse(&$1, '\\', &$0->val.str, tail);
2493 tok_err(c, "error: unsupported string suffix",
2498 ###### print exec cases
2501 struct val *v = cast(val, e);
2502 if (v->vtype == Tstr)
2504 print_value(v->vtype, &v->val);
2505 if (v->vtype == Tstr)
2510 ###### propagate exec cases
2513 struct val *val = cast(val, prog);
2514 if (!type_compat(type, val->vtype, rules))
2515 type_err(c, "error: expected %1%r found %2",
2516 prog, type, rules, val->vtype);
2520 ###### interp exec cases
2522 rvtype = cast(val, e)->vtype;
2523 dup_value(rvtype, &cast(val, e)->val, &rv);
2526 ###### ast functions
2527 static void free_val(struct val *v)
2530 free_value(v->vtype, &v->val);
2534 ###### free exec cases
2535 case Xval: free_val(cast(val, e)); break;
2537 ###### ast functions
2538 // Move all nodes from 'b' to 'rv', reversing their order.
2539 // In 'b' 'left' is a list, and 'right' is the last node.
2540 // In 'rv', left' is the first node and 'right' is a list.
2541 static struct binode *reorder_bilist(struct binode *b)
2543 struct binode *rv = NULL;
2546 struct exec *t = b->right;
2550 b = cast(binode, b->left);
2560 Just as we used a `val` to wrap a value into an `exec`, we similarly
2561 need a `var` to wrap a `variable` into an exec. While each `val`
2562 contained a copy of the value, each `var` holds a link to the variable
2563 because it really is the same variable no matter where it appears.
2564 When a variable is used, we need to remember to follow the `->merged`
2565 link to find the primary instance.
2573 struct variable *var;
2581 VariableDecl -> IDENTIFIER : ${ {
2582 struct variable *v = var_decl(c, $1.txt);
2583 $0 = new_pos(var, $1);
2588 v = var_ref(c, $1.txt);
2590 type_err(c, "error: variable '%v' redeclared",
2592 type_err(c, "info: this is where '%v' was first declared",
2593 v->where_decl, NULL, 0, NULL);
2596 | IDENTIFIER :: ${ {
2597 struct variable *v = var_decl(c, $1.txt);
2598 $0 = new_pos(var, $1);
2604 v = var_ref(c, $1.txt);
2606 type_err(c, "error: variable '%v' redeclared",
2608 type_err(c, "info: this is where '%v' was first declared",
2609 v->where_decl, NULL, 0, NULL);
2612 | IDENTIFIER : Type ${ {
2613 struct variable *v = var_decl(c, $1.txt);
2614 $0 = new_pos(var, $1);
2621 v = var_ref(c, $1.txt);
2623 type_err(c, "error: variable '%v' redeclared",
2625 type_err(c, "info: this is where '%v' was first declared",
2626 v->where_decl, NULL, 0, NULL);
2629 | IDENTIFIER :: Type ${ {
2630 struct variable *v = var_decl(c, $1.txt);
2631 $0 = new_pos(var, $1);
2639 v = var_ref(c, $1.txt);
2641 type_err(c, "error: variable '%v' redeclared",
2643 type_err(c, "info: this is where '%v' was first declared",
2644 v->where_decl, NULL, 0, NULL);
2649 Variable -> IDENTIFIER ${ {
2650 struct variable *v = var_ref(c, $1.txt);
2651 $0 = new_pos(var, $1);
2653 /* This might be a label - allocate a var just in case */
2654 v = var_decl(c, $1.txt);
2661 cast(var, $0)->var = v;
2665 ###### print exec cases
2668 struct var *v = cast(var, e);
2670 struct binding *b = v->var->name;
2671 printf("%.*s", b->name.len, b->name.txt);
2678 if (loc && loc->type == Xvar) {
2679 struct var *v = cast(var, loc);
2681 struct binding *b = v->var->name;
2682 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2684 fputs("???", stderr); // NOTEST
2686 fputs("NOTVAR", stderr); // NOTEST
2689 ###### propagate exec cases
2693 struct var *var = cast(var, prog);
2694 struct variable *v = var->var;
2696 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2697 return Tnone; // NOTEST
2700 if (v->constant && (rules & Rnoconstant)) {
2701 type_err(c, "error: Cannot assign to a constant: %v",
2702 prog, NULL, 0, NULL);
2703 type_err(c, "info: name was defined as a constant here",
2704 v->where_decl, NULL, 0, NULL);
2707 if (v->type == Tnone && v->where_decl == prog)
2708 type_err(c, "error: variable used but not declared: %v",
2709 prog, NULL, 0, NULL);
2710 if (v->type == NULL) {
2711 if (type && *ok != 0) {
2713 v->where_set = prog;
2718 if (!type_compat(type, v->type, rules)) {
2719 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2720 type, rules, v->type);
2721 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2722 v->type, rules, NULL);
2729 ###### interp exec cases
2732 struct var *var = cast(var, e);
2733 struct variable *v = var->var;
2736 lrv = var_value(c, v);
2741 ###### ast functions
2743 static void free_var(struct var *v)
2748 ###### free exec cases
2749 case Xvar: free_var(cast(var, e)); break;
2751 ### Expressions: Conditional
2753 Our first user of the `binode` will be conditional expressions, which
2754 is a bit odd as they actually have three components. That will be
2755 handled by having 2 binodes for each expression. The conditional
2756 expression is the lowest precedence operator which is why we define it
2757 first - to start the precedence list.
2759 Conditional expressions are of the form "value `if` condition `else`
2760 other_value". They associate to the right, so everything to the right
2761 of `else` is part of an else value, while only a higher-precedence to
2762 the left of `if` is the if values. Between `if` and `else` there is no
2763 room for ambiguity, so a full conditional expression is allowed in
2775 Expression -> Expression if Expression else Expression $$ifelse ${ {
2776 struct binode *b1 = new(binode);
2777 struct binode *b2 = new(binode);
2786 ## expression grammar
2788 ###### print binode cases
2791 b2 = cast(binode, b->right);
2792 if (bracket) printf("(");
2793 print_exec(b2->left, -1, bracket);
2795 print_exec(b->left, -1, bracket);
2797 print_exec(b2->right, -1, bracket);
2798 if (bracket) printf(")");
2801 ###### propagate binode cases
2804 /* cond must be Tbool, others must match */
2805 struct binode *b2 = cast(binode, b->right);
2808 propagate_types(b->left, c, ok, Tbool, 0);
2809 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2810 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2814 ###### interp binode cases
2817 struct binode *b2 = cast(binode, b->right);
2818 left = interp_exec(c, b->left, <ype);
2820 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
2822 rv = interp_exec(c, b2->right, &rvtype);
2828 We take a brief detour, now that we have expressions, to describe lists
2829 of expressions. These will be needed for function parameters and
2830 possibly other situations. They seem generic enough to introduce here
2831 to be used elsewhere.
2833 And ExpressionList will use the `List` type of `binode`, building up at
2834 the end. And place where they are used will probably call
2835 `reorder_bilist()` to get a more normal first/next arrangement.
2837 ###### declare terminals
2840 `List` execs have no implicit semantics, so they are never propagated or
2841 interpreted. The can be printed as a comma separate list, which is how
2842 they are parsed. Note they are also used for function formal parameter
2843 lists. In that case a separate function is used to print them.
2845 ###### print binode cases
2849 print_exec(b->left, -1, bracket);
2852 b = cast(binode, b->right);
2856 ###### propagate binode cases
2857 case List: abort(); // NOTEST
2858 ###### interp binode cases
2859 case List: abort(); // NOTEST
2864 ExpressionList -> ExpressionList , Expression ${
2877 ### Expressions: Boolean
2879 The next class of expressions to use the `binode` will be Boolean
2880 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
2881 have same corresponding precendence. The difference is that they don't
2882 evaluate the second expression if not necessary.
2891 ###### expr precedence
2896 ###### expression grammar
2897 | Expression or Expression ${ {
2898 struct binode *b = new(binode);
2904 | Expression or else Expression ${ {
2905 struct binode *b = new(binode);
2912 | Expression and Expression ${ {
2913 struct binode *b = new(binode);
2919 | Expression and then Expression ${ {
2920 struct binode *b = new(binode);
2927 | not Expression ${ {
2928 struct binode *b = new(binode);
2934 ###### print binode cases
2936 if (bracket) printf("(");
2937 print_exec(b->left, -1, bracket);
2939 print_exec(b->right, -1, bracket);
2940 if (bracket) printf(")");
2943 if (bracket) printf("(");
2944 print_exec(b->left, -1, bracket);
2945 printf(" and then ");
2946 print_exec(b->right, -1, bracket);
2947 if (bracket) printf(")");
2950 if (bracket) printf("(");
2951 print_exec(b->left, -1, bracket);
2953 print_exec(b->right, -1, bracket);
2954 if (bracket) printf(")");
2957 if (bracket) printf("(");
2958 print_exec(b->left, -1, bracket);
2959 printf(" or else ");
2960 print_exec(b->right, -1, bracket);
2961 if (bracket) printf(")");
2964 if (bracket) printf("(");
2966 print_exec(b->right, -1, bracket);
2967 if (bracket) printf(")");
2970 ###### propagate binode cases
2976 /* both must be Tbool, result is Tbool */
2977 propagate_types(b->left, c, ok, Tbool, 0);
2978 propagate_types(b->right, c, ok, Tbool, 0);
2979 if (type && type != Tbool)
2980 type_err(c, "error: %1 operation found where %2 expected", prog,
2984 ###### interp binode cases
2986 rv = interp_exec(c, b->left, &rvtype);
2987 right = interp_exec(c, b->right, &rtype);
2988 rv.bool = rv.bool && right.bool;
2991 rv = interp_exec(c, b->left, &rvtype);
2993 rv = interp_exec(c, b->right, NULL);
2996 rv = interp_exec(c, b->left, &rvtype);
2997 right = interp_exec(c, b->right, &rtype);
2998 rv.bool = rv.bool || right.bool;
3001 rv = interp_exec(c, b->left, &rvtype);
3003 rv = interp_exec(c, b->right, NULL);
3006 rv = interp_exec(c, b->right, &rvtype);
3010 ### Expressions: Comparison
3012 Of slightly higher precedence that Boolean expressions are Comparisons.
3013 A comparison takes arguments of any comparable type, but the two types
3016 To simplify the parsing we introduce an `eop` which can record an
3017 expression operator, and the `CMPop` non-terminal will match one of them.
3024 ###### ast functions
3025 static void free_eop(struct eop *e)
3039 ###### expr precedence
3040 $LEFT < > <= >= == != CMPop
3042 ###### expression grammar
3043 | Expression CMPop Expression ${ {
3044 struct binode *b = new(binode);
3054 CMPop -> < ${ $0.op = Less; }$
3055 | > ${ $0.op = Gtr; }$
3056 | <= ${ $0.op = LessEq; }$
3057 | >= ${ $0.op = GtrEq; }$
3058 | == ${ $0.op = Eql; }$
3059 | != ${ $0.op = NEql; }$
3061 ###### print binode cases
3069 if (bracket) printf("(");
3070 print_exec(b->left, -1, bracket);
3072 case Less: printf(" < "); break;
3073 case LessEq: printf(" <= "); break;
3074 case Gtr: printf(" > "); break;
3075 case GtrEq: printf(" >= "); break;
3076 case Eql: printf(" == "); break;
3077 case NEql: printf(" != "); break;
3078 default: abort(); // NOTEST
3080 print_exec(b->right, -1, bracket);
3081 if (bracket) printf(")");
3084 ###### propagate binode cases
3091 /* Both must match but not be labels, result is Tbool */
3092 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
3094 propagate_types(b->right, c, ok, t, 0);
3096 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
3098 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
3100 if (!type_compat(type, Tbool, 0))
3101 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3102 Tbool, rules, type);
3105 ###### interp binode cases
3114 left = interp_exec(c, b->left, <ype);
3115 right = interp_exec(c, b->right, &rtype);
3116 cmp = value_cmp(ltype, rtype, &left, &right);
3119 case Less: rv.bool = cmp < 0; break;
3120 case LessEq: rv.bool = cmp <= 0; break;
3121 case Gtr: rv.bool = cmp > 0; break;
3122 case GtrEq: rv.bool = cmp >= 0; break;
3123 case Eql: rv.bool = cmp == 0; break;
3124 case NEql: rv.bool = cmp != 0; break;
3125 default: rv.bool = 0; break; // NOTEST
3130 ### Expressions: The rest
3132 The remaining expressions with the highest precedence are arithmetic,
3133 string concatenation, and string conversion. String concatenation
3134 (`++`) has the same precedence as multiplication and division, but lower
3137 String conversion is a temporary feature until I get a better type
3138 system. `$` is a prefix operator which expects a string and returns
3141 `+` and `-` are both infix and prefix operations (where they are
3142 absolute value and negation). These have different operator names.
3144 We also have a 'Bracket' operator which records where parentheses were
3145 found. This makes it easy to reproduce these when printing. Possibly I
3146 should only insert brackets were needed for precedence.
3156 ###### expr precedence
3162 ###### expression grammar
3163 | Expression Eop Expression ${ {
3164 struct binode *b = new(binode);
3171 | Expression Top Expression ${ {
3172 struct binode *b = new(binode);
3179 | ( Expression ) ${ {
3180 struct binode *b = new_pos(binode, $1);
3185 | Uop Expression ${ {
3186 struct binode *b = new(binode);
3191 | Value ${ $0 = $<1; }$
3192 | Variable ${ $0 = $<1; }$
3195 Eop -> + ${ $0.op = Plus; }$
3196 | - ${ $0.op = Minus; }$
3198 Uop -> + ${ $0.op = Absolute; }$
3199 | - ${ $0.op = Negate; }$
3200 | $ ${ $0.op = StringConv; }$
3202 Top -> * ${ $0.op = Times; }$
3203 | / ${ $0.op = Divide; }$
3204 | % ${ $0.op = Rem; }$
3205 | ++ ${ $0.op = Concat; }$
3207 ###### print binode cases
3214 if (bracket) printf("(");
3215 print_exec(b->left, indent, bracket);
3217 case Plus: fputs(" + ", stdout); break;
3218 case Minus: fputs(" - ", stdout); break;
3219 case Times: fputs(" * ", stdout); break;
3220 case Divide: fputs(" / ", stdout); break;
3221 case Rem: fputs(" % ", stdout); break;
3222 case Concat: fputs(" ++ ", stdout); break;
3223 default: abort(); // NOTEST
3225 print_exec(b->right, indent, bracket);
3226 if (bracket) printf(")");
3231 if (bracket) printf("(");
3233 case Absolute: fputs("+", stdout); break;
3234 case Negate: fputs("-", stdout); break;
3235 case StringConv: fputs("$", stdout); break;
3236 default: abort(); // NOTEST
3238 print_exec(b->right, indent, bracket);
3239 if (bracket) printf(")");
3243 print_exec(b->right, indent, bracket);
3247 ###### propagate binode cases
3253 /* both must be numbers, result is Tnum */
3256 /* as propagate_types ignores a NULL,
3257 * unary ops fit here too */
3258 propagate_types(b->left, c, ok, Tnum, 0);
3259 propagate_types(b->right, c, ok, Tnum, 0);
3260 if (!type_compat(type, Tnum, 0))
3261 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3266 /* both must be Tstr, result is Tstr */
3267 propagate_types(b->left, c, ok, Tstr, 0);
3268 propagate_types(b->right, c, ok, Tstr, 0);
3269 if (!type_compat(type, Tstr, 0))
3270 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3275 /* op must be string, result is number */
3276 propagate_types(b->left, c, ok, Tstr, 0);
3277 if (!type_compat(type, Tnum, 0))
3278 type_err(c, // UNTESTED
3279 "error: Can only convert string to number, not %1",
3280 prog, type, 0, NULL);
3284 return propagate_types(b->right, c, ok, type, 0);
3286 ###### interp binode cases
3289 rv = interp_exec(c, b->left, &rvtype);
3290 right = interp_exec(c, b->right, &rtype);
3291 mpq_add(rv.num, rv.num, right.num);
3294 rv = interp_exec(c, b->left, &rvtype);
3295 right = interp_exec(c, b->right, &rtype);
3296 mpq_sub(rv.num, rv.num, right.num);
3299 rv = interp_exec(c, b->left, &rvtype);
3300 right = interp_exec(c, b->right, &rtype);
3301 mpq_mul(rv.num, rv.num, right.num);
3304 rv = interp_exec(c, b->left, &rvtype);
3305 right = interp_exec(c, b->right, &rtype);
3306 mpq_div(rv.num, rv.num, right.num);
3311 left = interp_exec(c, b->left, <ype);
3312 right = interp_exec(c, b->right, &rtype);
3313 mpz_init(l); mpz_init(r); mpz_init(rem);
3314 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3315 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3316 mpz_tdiv_r(rem, l, r);
3317 val_init(Tnum, &rv);
3318 mpq_set_z(rv.num, rem);
3319 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3324 rv = interp_exec(c, b->right, &rvtype);
3325 mpq_neg(rv.num, rv.num);
3328 rv = interp_exec(c, b->right, &rvtype);
3329 mpq_abs(rv.num, rv.num);
3332 rv = interp_exec(c, b->right, &rvtype);
3335 left = interp_exec(c, b->left, <ype);
3336 right = interp_exec(c, b->right, &rtype);
3338 rv.str = text_join(left.str, right.str);
3341 right = interp_exec(c, b->right, &rvtype);
3345 struct text tx = right.str;
3348 if (tx.txt[0] == '-') {
3349 neg = 1; // UNTESTED
3350 tx.txt++; // UNTESTED
3351 tx.len--; // UNTESTED
3353 if (number_parse(rv.num, tail, tx) == 0)
3354 mpq_init(rv.num); // UNTESTED
3356 mpq_neg(rv.num, rv.num); // UNTESTED
3358 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3362 ###### value functions
3364 static struct text text_join(struct text a, struct text b)
3367 rv.len = a.len + b.len;
3368 rv.txt = malloc(rv.len);
3369 memcpy(rv.txt, a.txt, a.len);
3370 memcpy(rv.txt+a.len, b.txt, b.len);
3374 ### Blocks, Statements, and Statement lists.
3376 Now that we have expressions out of the way we need to turn to
3377 statements. There are simple statements and more complex statements.
3378 Simple statements do not contain (syntactic) newlines, complex statements do.
3380 Statements often come in sequences and we have corresponding simple
3381 statement lists and complex statement lists.
3382 The former comprise only simple statements separated by semicolons.
3383 The later comprise complex statements and simple statement lists. They are
3384 separated by newlines. Thus the semicolon is only used to separate
3385 simple statements on the one line. This may be overly restrictive,
3386 but I'm not sure I ever want a complex statement to share a line with
3389 Note that a simple statement list can still use multiple lines if
3390 subsequent lines are indented, so
3392 ###### Example: wrapped simple statement list
3397 is a single simple statement list. This might allow room for
3398 confusion, so I'm not set on it yet.
3400 A simple statement list needs no extra syntax. A complex statement
3401 list has two syntactic forms. It can be enclosed in braces (much like
3402 C blocks), or it can be introduced by an indent and continue until an
3403 unindented newline (much like Python blocks). With this extra syntax
3404 it is referred to as a block.
3406 Note that a block does not have to include any newlines if it only
3407 contains simple statements. So both of:
3409 if condition: a=b; d=f
3411 if condition { a=b; print f }
3415 In either case the list is constructed from a `binode` list with
3416 `Block` as the operator. When parsing the list it is most convenient
3417 to append to the end, so a list is a list and a statement. When using
3418 the list it is more convenient to consider a list to be a statement
3419 and a list. So we need a function to re-order a list.
3420 `reorder_bilist` serves this purpose.
3422 The only stand-alone statement we introduce at this stage is `pass`
3423 which does nothing and is represented as a `NULL` pointer in a `Block`
3424 list. Other stand-alone statements will follow once the infrastructure
3435 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3436 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3437 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3438 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3439 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3441 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3442 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3443 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3444 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3445 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3447 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3448 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3449 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3451 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3452 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3453 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3454 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3455 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3457 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3459 ComplexStatements -> ComplexStatements ComplexStatement ${
3469 | ComplexStatement ${
3481 ComplexStatement -> SimpleStatements Newlines ${
3482 $0 = reorder_bilist($<SS);
3484 | SimpleStatements ; Newlines ${
3485 $0 = reorder_bilist($<SS);
3487 ## ComplexStatement Grammar
3490 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3496 | SimpleStatement ${
3504 SimpleStatement -> pass ${ $0 = NULL; }$
3505 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3506 ## SimpleStatement Grammar
3508 ###### print binode cases
3512 if (b->left == NULL) // UNTESTED
3513 printf("pass"); // UNTESTED
3515 print_exec(b->left, indent, bracket); // UNTESTED
3516 if (b->right) { // UNTESTED
3517 printf("; "); // UNTESTED
3518 print_exec(b->right, indent, bracket); // UNTESTED
3521 // block, one per line
3522 if (b->left == NULL)
3523 do_indent(indent, "pass\n");
3525 print_exec(b->left, indent, bracket);
3527 print_exec(b->right, indent, bracket);
3531 ###### propagate binode cases
3534 /* If any statement returns something other than Tnone
3535 * or Tbool then all such must return same type.
3536 * As each statement may be Tnone or something else,
3537 * we must always pass NULL (unknown) down, otherwise an incorrect
3538 * error might occur. We never return Tnone unless it is
3543 for (e = b; e; e = cast(binode, e->right)) {
3544 t = propagate_types(e->left, c, ok, NULL, rules);
3545 if ((rules & Rboolok) && t == Tbool)
3547 if (t && t != Tnone && t != Tbool) {
3551 type_err(c, "error: expected %1%r, found %2",
3552 e->left, type, rules, t);
3558 ###### interp binode cases
3560 while (rvtype == Tnone &&
3563 rv = interp_exec(c, b->left, &rvtype);
3564 b = cast(binode, b->right);
3568 ### The Print statement
3570 `print` is a simple statement that takes a comma-separated list of
3571 expressions and prints the values separated by spaces and terminated
3572 by a newline. No control of formatting is possible.
3574 `print` uses `ExpressionList` to collect the expressions and stores them
3575 on the left side of a `Print` binode unlessthere is a trailing comma
3576 when the list is stored on the `right` side and no trailing newline is
3582 ##### expr precedence
3585 ###### SimpleStatement Grammar
3587 | print ExpressionList ${
3591 $0->left = reorder_bilist($<EL);
3593 | print ExpressionList , ${ {
3596 $0->right = reorder_bilist($<EL);
3606 ###### print binode cases
3609 do_indent(indent, "print");
3611 print_exec(b->right, -1, bracket);
3614 print_exec(b->left, -1, bracket);
3619 ###### propagate binode cases
3622 /* don't care but all must be consistent */
3624 b = cast(binode, b->left);
3626 b = cast(binode, b->right);
3628 propagate_types(b->left, c, ok, NULL, Rnolabel);
3629 b = cast(binode, b->right);
3633 ###### interp binode cases
3637 struct binode *b2 = cast(binode, b->left);
3639 b2 = cast(binode, b->right);
3640 for (; b2; b2 = cast(binode, b2->right)) {
3641 left = interp_exec(c, b2->left, <ype);
3642 print_value(ltype, &left);
3643 free_value(ltype, &left);
3647 if (b->right == NULL)
3653 ###### Assignment statement
3655 An assignment will assign a value to a variable, providing it hasn't
3656 been declared as a constant. The analysis phase ensures that the type
3657 will be correct so the interpreter just needs to perform the
3658 calculation. There is a form of assignment which declares a new
3659 variable as well as assigning a value. If a name is assigned before
3660 it is declared, and error will be raised as the name is created as
3661 `Tlabel` and it is illegal to assign to such names.
3667 ###### declare terminals
3670 ###### SimpleStatement Grammar
3671 | Variable = Expression ${
3677 | VariableDecl = Expression ${
3685 if ($1->var->where_set == NULL) {
3687 "Variable declared with no type or value: %v",
3697 ###### print binode cases
3700 do_indent(indent, "");
3701 print_exec(b->left, indent, bracket);
3703 print_exec(b->right, indent, bracket);
3710 struct variable *v = cast(var, b->left)->var;
3711 do_indent(indent, "");
3712 print_exec(b->left, indent, bracket);
3713 if (cast(var, b->left)->var->constant) {
3715 if (v->where_decl == v->where_set) {
3716 type_print(v->type, stdout);
3721 if (v->where_decl == v->where_set) {
3722 type_print(v->type, stdout);
3728 print_exec(b->right, indent, bracket);
3735 ###### propagate binode cases
3739 /* Both must match and not be labels,
3740 * Type must support 'dup',
3741 * For Assign, left must not be constant.
3744 t = propagate_types(b->left, c, ok, NULL,
3745 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3750 if (propagate_types(b->right, c, ok, t, 0) != t)
3751 if (b->left->type == Xvar)
3752 type_err(c, "info: variable '%v' was set as %1 here.",
3753 cast(var, b->left)->var->where_set, t, rules, NULL);
3755 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3757 propagate_types(b->left, c, ok, t,
3758 (b->op == Assign ? Rnoconstant : 0));
3760 if (t && t->dup == NULL)
3761 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3766 ###### interp binode cases
3769 lleft = linterp_exec(c, b->left, <ype);
3770 right = interp_exec(c, b->right, &rtype);
3772 free_value(ltype, lleft);
3773 dup_value(ltype, &right, lleft);
3780 struct variable *v = cast(var, b->left)->var;
3783 val = var_value(c, v);
3784 if (v->type->prepare_type)
3785 v->type->prepare_type(c, v->type, 0);
3787 right = interp_exec(c, b->right, &rtype);
3788 memcpy(val, &right, rtype->size);
3791 val_init(v->type, val);
3796 ### The `use` statement
3798 The `use` statement is the last "simple" statement. It is needed when
3799 the condition in a conditional statement is a block. `use` works much
3800 like `return` in C, but only completes the `condition`, not the whole
3806 ###### expr precedence
3809 ###### SimpleStatement Grammar
3811 $0 = new_pos(binode, $1);
3814 if ($0->right->type == Xvar) {
3815 struct var *v = cast(var, $0->right);
3816 if (v->var->type == Tnone) {
3817 /* Convert this to a label */
3820 v->var->type = Tlabel;
3821 val = global_alloc(c, Tlabel, v->var, NULL);
3827 ###### print binode cases
3830 do_indent(indent, "use ");
3831 print_exec(b->right, -1, bracket);
3836 ###### propagate binode cases
3839 /* result matches value */
3840 return propagate_types(b->right, c, ok, type, 0);
3842 ###### interp binode cases
3845 rv = interp_exec(c, b->right, &rvtype);
3848 ### The Conditional Statement
3850 This is the biggy and currently the only complex statement. This
3851 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3852 It is comprised of a number of parts, all of which are optional though
3853 set combinations apply. Each part is (usually) a key word (`then` is
3854 sometimes optional) followed by either an expression or a code block,
3855 except the `casepart` which is a "key word and an expression" followed
3856 by a code block. The code-block option is valid for all parts and,
3857 where an expression is also allowed, the code block can use the `use`
3858 statement to report a value. If the code block does not report a value
3859 the effect is similar to reporting `True`.
3861 The `else` and `case` parts, as well as `then` when combined with
3862 `if`, can contain a `use` statement which will apply to some
3863 containing conditional statement. `for` parts, `do` parts and `then`
3864 parts used with `for` can never contain a `use`, except in some
3865 subordinate conditional statement.
3867 If there is a `forpart`, it is executed first, only once.
3868 If there is a `dopart`, then it is executed repeatedly providing
3869 always that the `condpart` or `cond`, if present, does not return a non-True
3870 value. `condpart` can fail to return any value if it simply executes
3871 to completion. This is treated the same as returning `True`.
3873 If there is a `thenpart` it will be executed whenever the `condpart`
3874 or `cond` returns True (or does not return any value), but this will happen
3875 *after* `dopart` (when present).
3877 If `elsepart` is present it will be executed at most once when the
3878 condition returns `False` or some value that isn't `True` and isn't
3879 matched by any `casepart`. If there are any `casepart`s, they will be
3880 executed when the condition returns a matching value.
3882 The particular sorts of values allowed in case parts has not yet been
3883 determined in the language design, so nothing is prohibited.
3885 The various blocks in this complex statement potentially provide scope
3886 for variables as described earlier. Each such block must include the
3887 "OpenScope" nonterminal before parsing the block, and must call
3888 `var_block_close()` when closing the block.
3890 The code following "`if`", "`switch`" and "`for`" does not get its own
3891 scope, but is in a scope covering the whole statement, so names
3892 declared there cannot be redeclared elsewhere. Similarly the
3893 condition following "`while`" is in a scope the covers the body
3894 ("`do`" part) of the loop, and which does not allow conditional scope
3895 extension. Code following "`then`" (both looping and non-looping),
3896 "`else`" and "`case`" each get their own local scope.
3898 The type requirements on the code block in a `whilepart` are quite
3899 unusal. It is allowed to return a value of some identifiable type, in
3900 which case the loop aborts and an appropriate `casepart` is run, or it
3901 can return a Boolean, in which case the loop either continues to the
3902 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3903 This is different both from the `ifpart` code block which is expected to
3904 return a Boolean, or the `switchpart` code block which is expected to
3905 return the same type as the casepart values. The correct analysis of
3906 the type of the `whilepart` code block is the reason for the
3907 `Rboolok` flag which is passed to `propagate_types()`.
3909 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3910 defined. As there are two scopes which cover multiple parts - one for
3911 the whole statement and one for "while" and "do" - and as we will use
3912 the 'struct exec' to track scopes, we actually need two new types of
3913 exec. One is a `binode` for the looping part, the rest is the
3914 `cond_statement`. The `cond_statement` will use an auxilliary `struct
3915 casepart` to track a list of case parts.
3926 struct exec *action;
3927 struct casepart *next;
3929 struct cond_statement {
3931 struct exec *forpart, *condpart, *thenpart, *elsepart;
3932 struct binode *looppart;
3933 struct casepart *casepart;
3936 ###### ast functions
3938 static void free_casepart(struct casepart *cp)
3942 free_exec(cp->value);
3943 free_exec(cp->action);
3950 static void free_cond_statement(struct cond_statement *s)
3954 free_exec(s->forpart);
3955 free_exec(s->condpart);
3956 free_exec(s->looppart);
3957 free_exec(s->thenpart);
3958 free_exec(s->elsepart);
3959 free_casepart(s->casepart);
3963 ###### free exec cases
3964 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3966 ###### ComplexStatement Grammar
3967 | CondStatement ${ $0 = $<1; }$
3969 ###### expr precedence
3970 $TERM for then while do
3977 // A CondStatement must end with EOL, as does CondSuffix and
3979 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
3980 // may or may not end with EOL
3981 // WhilePart and IfPart include an appropriate Suffix
3983 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
3984 // them. WhilePart opens and closes its own scope.
3985 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
3988 $0->thenpart = $<TP;
3989 $0->looppart = $<WP;
3990 var_block_close(c, CloseSequential, $0);
3992 | ForPart OptNL WhilePart CondSuffix ${
3995 $0->looppart = $<WP;
3996 var_block_close(c, CloseSequential, $0);
3998 | WhilePart CondSuffix ${
4000 $0->looppart = $<WP;
4002 | SwitchPart OptNL CasePart CondSuffix ${
4004 $0->condpart = $<SP;
4005 $CP->next = $0->casepart;
4006 $0->casepart = $<CP;
4007 var_block_close(c, CloseSequential, $0);
4009 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4011 $0->condpart = $<SP;
4012 $CP->next = $0->casepart;
4013 $0->casepart = $<CP;
4014 var_block_close(c, CloseSequential, $0);
4016 | IfPart IfSuffix ${
4018 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4019 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4020 // This is where we close an "if" statement
4021 var_block_close(c, CloseSequential, $0);
4024 CondSuffix -> IfSuffix ${
4027 | Newlines CasePart CondSuffix ${
4029 $CP->next = $0->casepart;
4030 $0->casepart = $<CP;
4032 | CasePart CondSuffix ${
4034 $CP->next = $0->casepart;
4035 $0->casepart = $<CP;
4038 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4039 | Newlines ElsePart ${ $0 = $<EP; }$
4040 | ElsePart ${$0 = $<EP; }$
4042 ElsePart -> else OpenBlock Newlines ${
4043 $0 = new(cond_statement);
4044 $0->elsepart = $<OB;
4045 var_block_close(c, CloseElse, $0->elsepart);
4047 | else OpenScope CondStatement ${
4048 $0 = new(cond_statement);
4049 $0->elsepart = $<CS;
4050 var_block_close(c, CloseElse, $0->elsepart);
4054 CasePart -> case Expression OpenScope ColonBlock ${
4055 $0 = calloc(1,sizeof(struct casepart));
4058 var_block_close(c, CloseParallel, $0->action);
4062 // These scopes are closed in CondStatement
4063 ForPart -> for OpenBlock ${
4067 ThenPart -> then OpenBlock ${
4069 var_block_close(c, CloseSequential, $0);
4073 // This scope is closed in CondStatement
4074 WhilePart -> while UseBlock OptNL do OpenBlock ${
4079 var_block_close(c, CloseSequential, $0->right);
4080 var_block_close(c, CloseSequential, $0);
4082 | while OpenScope Expression OpenScope ColonBlock ${
4087 var_block_close(c, CloseSequential, $0->right);
4088 var_block_close(c, CloseSequential, $0);
4092 IfPart -> if UseBlock OptNL then OpenBlock ${
4095 var_block_close(c, CloseParallel, $0.thenpart);
4097 | if OpenScope Expression OpenScope ColonBlock ${
4100 var_block_close(c, CloseParallel, $0.thenpart);
4102 | if OpenScope Expression OpenScope OptNL then Block ${
4105 var_block_close(c, CloseParallel, $0.thenpart);
4109 // This scope is closed in CondStatement
4110 SwitchPart -> switch OpenScope Expression ${
4113 | switch UseBlock ${
4117 ###### print binode cases
4119 if (b->left && b->left->type == Xbinode &&
4120 cast(binode, b->left)->op == Block) {
4122 do_indent(indent, "while {\n");
4124 do_indent(indent, "while\n");
4125 print_exec(b->left, indent+1, bracket);
4127 do_indent(indent, "} do {\n");
4129 do_indent(indent, "do\n");
4130 print_exec(b->right, indent+1, bracket);
4132 do_indent(indent, "}\n");
4134 do_indent(indent, "while ");
4135 print_exec(b->left, 0, bracket);
4140 print_exec(b->right, indent+1, bracket);
4142 do_indent(indent, "}\n");
4146 ###### print exec cases
4148 case Xcond_statement:
4150 struct cond_statement *cs = cast(cond_statement, e);
4151 struct casepart *cp;
4153 do_indent(indent, "for");
4154 if (bracket) printf(" {\n"); else printf("\n");
4155 print_exec(cs->forpart, indent+1, bracket);
4158 do_indent(indent, "} then {\n");
4160 do_indent(indent, "then\n");
4161 print_exec(cs->thenpart, indent+1, bracket);
4163 if (bracket) do_indent(indent, "}\n");
4166 print_exec(cs->looppart, indent, bracket);
4170 do_indent(indent, "switch");
4172 do_indent(indent, "if");
4173 if (cs->condpart && cs->condpart->type == Xbinode &&
4174 cast(binode, cs->condpart)->op == Block) {
4179 print_exec(cs->condpart, indent+1, bracket);
4181 do_indent(indent, "}\n");
4183 do_indent(indent, "then\n");
4184 print_exec(cs->thenpart, indent+1, bracket);
4188 print_exec(cs->condpart, 0, bracket);
4194 print_exec(cs->thenpart, indent+1, bracket);
4196 do_indent(indent, "}\n");
4201 for (cp = cs->casepart; cp; cp = cp->next) {
4202 do_indent(indent, "case ");
4203 print_exec(cp->value, -1, 0);
4208 print_exec(cp->action, indent+1, bracket);
4210 do_indent(indent, "}\n");
4213 do_indent(indent, "else");
4218 print_exec(cs->elsepart, indent+1, bracket);
4220 do_indent(indent, "}\n");
4225 ###### propagate binode cases
4227 t = propagate_types(b->right, c, ok, Tnone, 0);
4228 if (!type_compat(Tnone, t, 0))
4229 *ok = 0; // UNTESTED
4230 return propagate_types(b->left, c, ok, type, rules);
4232 ###### propagate exec cases
4233 case Xcond_statement:
4235 // forpart and looppart->right must return Tnone
4236 // thenpart must return Tnone if there is a loopart,
4237 // otherwise it is like elsepart.
4239 // be bool if there is no casepart
4240 // match casepart->values if there is a switchpart
4241 // either be bool or match casepart->value if there
4243 // elsepart and casepart->action must match the return type
4244 // expected of this statement.
4245 struct cond_statement *cs = cast(cond_statement, prog);
4246 struct casepart *cp;
4248 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4249 if (!type_compat(Tnone, t, 0))
4250 *ok = 0; // UNTESTED
4253 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4254 if (!type_compat(Tnone, t, 0))
4255 *ok = 0; // UNTESTED
4257 if (cs->casepart == NULL) {
4258 propagate_types(cs->condpart, c, ok, Tbool, 0);
4259 propagate_types(cs->looppart, c, ok, Tbool, 0);
4261 /* Condpart must match case values, with bool permitted */
4263 for (cp = cs->casepart;
4264 cp && !t; cp = cp->next)
4265 t = propagate_types(cp->value, c, ok, NULL, 0);
4266 if (!t && cs->condpart)
4267 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4268 if (!t && cs->looppart)
4269 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4270 // Now we have a type (I hope) push it down
4272 for (cp = cs->casepart; cp; cp = cp->next)
4273 propagate_types(cp->value, c, ok, t, 0);
4274 propagate_types(cs->condpart, c, ok, t, Rboolok);
4275 propagate_types(cs->looppart, c, ok, t, Rboolok);
4278 // (if)then, else, and case parts must return expected type.
4279 if (!cs->looppart && !type)
4280 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4282 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4283 for (cp = cs->casepart;
4285 cp = cp->next) // UNTESTED
4286 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4289 propagate_types(cs->thenpart, c, ok, type, rules);
4290 propagate_types(cs->elsepart, c, ok, type, rules);
4291 for (cp = cs->casepart; cp ; cp = cp->next)
4292 propagate_types(cp->action, c, ok, type, rules);
4298 ###### interp binode cases
4300 // This just performs one iterration of the loop
4301 rv = interp_exec(c, b->left, &rvtype);
4302 if (rvtype == Tnone ||
4303 (rvtype == Tbool && rv.bool != 0))
4304 // cnd is Tnone or Tbool, doesn't need to be freed
4305 interp_exec(c, b->right, NULL);
4308 ###### interp exec cases
4309 case Xcond_statement:
4311 struct value v, cnd;
4312 struct type *vtype, *cndtype;
4313 struct casepart *cp;
4314 struct cond_statement *cs = cast(cond_statement, e);
4317 interp_exec(c, cs->forpart, NULL);
4319 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4320 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4321 interp_exec(c, cs->thenpart, NULL);
4323 cnd = interp_exec(c, cs->condpart, &cndtype);
4324 if ((cndtype == Tnone ||
4325 (cndtype == Tbool && cnd.bool != 0))) {
4326 // cnd is Tnone or Tbool, doesn't need to be freed
4327 rv = interp_exec(c, cs->thenpart, &rvtype);
4328 // skip else (and cases)
4332 for (cp = cs->casepart; cp; cp = cp->next) {
4333 v = interp_exec(c, cp->value, &vtype);
4334 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4335 free_value(vtype, &v);
4336 free_value(cndtype, &cnd);
4337 rv = interp_exec(c, cp->action, &rvtype);
4340 free_value(vtype, &v);
4342 free_value(cndtype, &cnd);
4344 rv = interp_exec(c, cs->elsepart, &rvtype);
4351 ### Top level structure
4353 All the language elements so far can be used in various places. Now
4354 it is time to clarify what those places are.
4356 At the top level of a file there will be a number of declarations.
4357 Many of the things that can be declared haven't been described yet,
4358 such as functions, procedures, imports, and probably more.
4359 For now there are two sorts of things that can appear at the top
4360 level. They are predefined constants, `struct` types, and the `main`
4361 function. While the syntax will allow the `main` function to appear
4362 multiple times, that will trigger an error if it is actually attempted.
4364 The various declarations do not return anything. They store the
4365 various declarations in the parse context.
4367 ###### Parser: grammar
4370 Ocean -> OptNL DeclarationList
4372 ## declare terminals
4379 DeclarationList -> Declaration
4380 | DeclarationList Declaration
4382 Declaration -> ERROR Newlines ${
4383 tok_err(c, // UNTESTED
4384 "error: unhandled parse error", &$1);
4390 ## top level grammar
4394 ### The `const` section
4396 As well as being defined in with the code that uses them, constants
4397 can be declared at the top level. These have full-file scope, so they
4398 are always `InScope`. The value of a top level constant can be given
4399 as an expression, and this is evaluated immediately rather than in the
4400 later interpretation stage. Once we add functions to the language, we
4401 will need rules concern which, if any, can be used to define a top
4404 Constants are defined in a section that starts with the reserved word
4405 `const` and then has a block with a list of assignment statements.
4406 For syntactic consistency, these must use the double-colon syntax to
4407 make it clear that they are constants. Type can also be given: if
4408 not, the type will be determined during analysis, as with other
4411 As the types constants are inserted at the head of a list, printing
4412 them in the same order that they were read is not straight forward.
4413 We take a quadratic approach here and count the number of constants
4414 (variables of depth 0), then count down from there, each time
4415 searching through for the Nth constant for decreasing N.
4417 ###### top level grammar
4421 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4422 | const { SimpleConstList } Newlines
4423 | const IN OptNL ConstList OUT Newlines
4424 | const SimpleConstList Newlines
4426 ConstList -> ConstList SimpleConstLine
4428 SimpleConstList -> SimpleConstList ; Const
4431 SimpleConstLine -> SimpleConstList Newlines
4432 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4435 CType -> Type ${ $0 = $<1; }$
4438 Const -> IDENTIFIER :: CType = Expression ${ {
4442 v = var_decl(c, $1.txt);
4444 struct var *var = new_pos(var, $1);
4445 v->where_decl = var;
4450 v = var_ref(c, $1.txt);
4451 tok_err(c, "error: name already declared", &$1);
4452 type_err(c, "info: this is where '%v' was first declared",
4453 v->where_decl, NULL, 0, NULL);
4457 propagate_types($5, c, &ok, $3, 0);
4462 struct value res = interp_exec(c, $5, &v->type);
4463 global_alloc(c, v->type, v, &res);
4467 ###### print const decls
4472 while (target != 0) {
4474 for (v = context.in_scope; v; v=v->in_scope)
4475 if (v->depth == 0) {
4486 struct value *val = var_value(&context, v);
4487 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4488 type_print(v->type, stdout);
4490 if (v->type == Tstr)
4492 print_value(v->type, val);
4493 if (v->type == Tstr)
4501 ### Finally the whole `main` function.
4503 An Ocean program can currently have only one function - `main` - and
4504 that must exist. It expects an array of strings with a provided size.
4505 Following this is a `block` which is the code to execute.
4507 As this is the top level, several things are handled a bit
4509 The function is not interpreted by `interp_exec` as that isn't
4510 passed the argument list which the program requires. Similarly type
4511 analysis is a bit more interesting at this level.
4513 ###### top level grammar
4515 DeclareFunction -> MainFunction ${ {
4517 type_err(c, "\"main\" defined a second time",
4523 ###### print binode cases
4525 do_indent(indent, "func main(");
4526 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4527 struct variable *v = cast(var, b2->left)->var;
4529 print_exec(b2->left, 0, 0);
4531 type_print(v->type, stdout);
4537 print_exec(b->right, indent+1, bracket);
4539 do_indent(indent, "}\n");
4542 ###### propagate binode cases
4543 case Func: abort(); // NOTEST
4545 ###### core functions
4547 static int analyse_prog(struct exec *prog, struct parse_context *c)
4549 struct binode *bp = cast(binode, prog);
4553 struct type *argv_type;
4554 struct text argv_type_name = { " argv", 5 };
4559 argv_type = add_type(c, argv_type_name, &array_prototype);
4560 argv_type->array.member = Tstr;
4561 argv_type->array.unspec = 1;
4563 for (b = cast(binode, bp->left); b; b = cast(binode, b->right)) {
4567 propagate_types(b->left, c, &ok, argv_type, 0);
4569 default: /* invalid */ // NOTEST
4570 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4576 propagate_types(bp->right, c, &ok, Tnone, 0);
4581 /* Make sure everything is still consistent */
4582 propagate_types(bp->right, c, &ok, Tnone, 0);
4584 return 0; // UNTESTED
4589 static void interp_prog(struct parse_context *c, struct exec *prog,
4590 int argc, char **argv)
4592 struct binode *p = cast(binode, prog);
4600 al = cast(binode, p->left);
4602 struct var *v = cast(var, al->left);
4603 struct value *vl = var_value(c, v->var);
4613 mpq_set_ui(argcq, argc, 1);
4614 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4615 t->prepare_type(c, t, 0);
4616 array_init(v->var->type, vl);
4617 for (i = 0; i < argc; i++) {
4618 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4621 arg.str.txt = argv[i];
4622 arg.str.len = strlen(argv[i]);
4623 free_value(Tstr, vl2);
4624 dup_value(Tstr, &arg, vl2);
4628 al = cast(binode, al->right);
4630 v = interp_exec(c, p, &vtype);
4631 free_value(vtype, &v);
4634 ###### interp binode cases
4636 rv = interp_exec(c, b->right, &rvtype);
4639 ## And now to test it out.
4641 Having a language requires having a "hello world" program. I'll
4642 provide a little more than that: a program that prints "Hello world"
4643 finds the GCD of two numbers, prints the first few elements of
4644 Fibonacci, performs a binary search for a number, and a few other
4645 things which will likely grow as the languages grows.
4647 ###### File: oceani.mk
4650 @echo "===== DEMO ====="
4651 ./oceani --section "demo: hello" oceani.mdc 55 33
4657 four ::= 2 + 2 ; five ::= 10/2
4658 const pie ::= "I like Pie";
4659 cake ::= "The cake is"
4670 print "Hello World, what lovely oceans you have!"
4671 print "Are there", five, "?"
4672 print pi, pie, "but", cake
4674 A := $argv[1]; B := $argv[2]
4676 /* When a variable is defined in both branches of an 'if',
4677 * and used afterwards, the variables are merged.
4683 print "Is", A, "bigger than", B,"? ", bigger
4684 /* If a variable is not used after the 'if', no
4685 * merge happens, so types can be different
4688 double:string = "yes"
4689 print A, "is more than twice", B, "?", double
4692 print "double", B, "is", double
4697 if a > 0 and then b > 0:
4703 print "GCD of", A, "and", B,"is", a
4705 print a, "is not positive, cannot calculate GCD"
4707 print b, "is not positive, cannot calculate GCD"
4712 print "Fibonacci:", f1,f2,
4713 then togo = togo - 1
4721 /* Binary search... */
4726 mid := (lo + hi) / 2
4739 print "Yay, I found", target
4741 print "Closest I found was", lo
4746 // "middle square" PRNG. Not particularly good, but one my
4747 // Dad taught me - the first one I ever heard of.
4748 for i:=1; then i = i + 1; while i < size:
4749 n := list[i-1] * list[i-1]
4750 list[i] = (n / 100) % 10 000
4752 print "Before sort:",
4753 for i:=0; then i = i + 1; while i < size:
4757 for i := 1; then i=i+1; while i < size:
4758 for j:=i-1; then j=j-1; while j >= 0:
4759 if list[j] > list[j+1]:
4763 print " After sort:",
4764 for i:=0; then i = i + 1; while i < size:
4768 if 1 == 2 then print "yes"; else print "no"
4772 bob.alive = (bob.name == "Hello")
4773 print "bob", "is" if bob.alive else "isn't", "alive"