1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
113 struct parse_context {
114 struct token_config config;
122 #define container_of(ptr, type, member) ({ \
123 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
124 (type *)( (char *)__mptr - offsetof(type,member) );})
126 #define config2context(_conf) container_of(_conf, struct parse_context, \
129 ###### Parser: reduce
130 struct parse_context *c = config2context(config);
138 #include <sys/mman.h>
157 static char Usage[] =
158 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
159 static const struct option long_options[] = {
160 {"trace", 0, NULL, 't'},
161 {"print", 0, NULL, 'p'},
162 {"noexec", 0, NULL, 'n'},
163 {"brackets", 0, NULL, 'b'},
164 {"section", 1, NULL, 's'},
167 const char *options = "tpnbs";
169 static void pr_err(char *msg) // NOTEST
171 fprintf(stderr, "%s\n", msg); // NOTEST
174 int main(int argc, char *argv[])
179 struct section *s, *ss;
180 char *section = NULL;
181 struct parse_context context = {
183 .ignored = (1 << TK_mark),
184 .number_chars = ".,_+- ",
189 int doprint=0, dotrace=0, doexec=1, brackets=0;
191 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
194 case 't': dotrace=1; break;
195 case 'p': doprint=1; break;
196 case 'n': doexec=0; break;
197 case 'b': brackets=1; break;
198 case 's': section = optarg; break;
199 default: fprintf(stderr, Usage);
203 if (optind >= argc) {
204 fprintf(stderr, "oceani: no input file given\n");
207 fd = open(argv[optind], O_RDONLY);
209 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
212 context.file_name = argv[optind];
213 len = lseek(fd, 0, 2);
214 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
215 s = code_extract(file, file+len, pr_err);
217 fprintf(stderr, "oceani: could not find any code in %s\n",
222 ## context initialization
225 for (ss = s; ss; ss = ss->next) {
226 struct text sec = ss->section;
227 if (sec.len == strlen(section) &&
228 strncmp(sec.txt, section, sec.len) == 0)
232 fprintf(stderr, "oceani: cannot find section %s\n",
239 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
243 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
245 if (!context.parse_error && !analyse_funcs(&context)) {
246 fprintf(stderr, "oceani: type error in program - not running.\n");
247 context.parse_error = 1;
255 if (doexec && !context.parse_error)
256 interp_main(&context, argc - optind, argv + optind);
259 struct section *t = s->next;
264 if (!context.parse_error) {
267 ## free context types
268 ## free context storage
269 exit(context.parse_error ? 1 : 0);
274 The four requirements of parse, analyse, print, interpret apply to
275 each language element individually so that is how most of the code
278 Three of the four are fairly self explanatory. The one that requires
279 a little explanation is the analysis step.
281 The current language design does not require the types of variables to
282 be declared, but they must still have a single type. Different
283 operations impose different requirements on the variables, for example
284 addition requires both arguments to be numeric, and assignment
285 requires the variable on the left to have the same type as the
286 expression on the right.
288 Analysis involves propagating these type requirements around and
289 consequently setting the type of each variable. If any requirements
290 are violated (e.g. a string is compared with a number) or if a
291 variable needs to have two different types, then an error is raised
292 and the program will not run.
294 If the same variable is declared in both branchs of an 'if/else', or
295 in all cases of a 'switch' then the multiple instances may be merged
296 into just one variable if the variable is referenced after the
297 conditional statement. When this happens, the types must naturally be
298 consistent across all the branches. When the variable is not used
299 outside the if, the variables in the different branches are distinct
300 and can be of different types.
302 Undeclared names may only appear in "use" statements and "case" expressions.
303 These names are given a type of "label" and a unique value.
304 This allows them to fill the role of a name in an enumerated type, which
305 is useful for testing the `switch` statement.
307 As we will see, the condition part of a `while` statement can return
308 either a Boolean or some other type. This requires that the expected
309 type that gets passed around comprises a type and a flag to indicate
310 that `Tbool` is also permitted.
312 As there are, as yet, no distinct types that are compatible, there
313 isn't much subtlety in the analysis. When we have distinct number
314 types, this will become more interesting.
318 When analysis discovers an inconsistency it needs to report an error;
319 just refusing to run the code ensures that the error doesn't cascade,
320 but by itself it isn't very useful. A clear understanding of the sort
321 of error message that are useful will help guide the process of
324 At a simplistic level, the only sort of error that type analysis can
325 report is that the type of some construct doesn't match a contextual
326 requirement. For example, in `4 + "hello"` the addition provides a
327 contextual requirement for numbers, but `"hello"` is not a number. In
328 this particular example no further information is needed as the types
329 are obvious from local information. When a variable is involved that
330 isn't the case. It may be helpful to explain why the variable has a
331 particular type, by indicating the location where the type was set,
332 whether by declaration or usage.
334 Using a recursive-descent analysis we can easily detect a problem at
335 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
336 will detect that one argument is not a number and the usage of `hello`
337 will detect that a number was wanted, but not provided. In this
338 (early) version of the language, we will generate error reports at
339 multiple locations, so the use of `hello` will report an error and
340 explain were the value was set, and the addition will report an error
341 and say why numbers are needed. To be able to report locations for
342 errors, each language element will need to record a file location
343 (line and column) and each variable will need to record the language
344 element where its type was set. For now we will assume that each line
345 of an error message indicates one location in the file, and up to 2
346 types. So we provide a `printf`-like function which takes a format, a
347 location (a `struct exec` which has not yet been introduced), and 2
348 types. "`%1`" reports the first type, "`%2`" reports the second. We
349 will need a function to print the location, once we know how that is
350 stored. e As will be explained later, there are sometimes extra rules for
351 type matching and they might affect error messages, we need to pass those
354 As well as type errors, we sometimes need to report problems with
355 tokens, which might be unexpected or might name a type that has not
356 been defined. For these we have `tok_err()` which reports an error
357 with a given token. Each of the error functions sets the flag in the
358 context so indicate that parsing failed.
362 static void fput_loc(struct exec *loc, FILE *f);
363 static void type_err(struct parse_context *c,
364 char *fmt, struct exec *loc,
365 struct type *t1, int rules, struct type *t2);
367 ###### core functions
369 static void type_err(struct parse_context *c,
370 char *fmt, struct exec *loc,
371 struct type *t1, int rules, struct type *t2)
373 fprintf(stderr, "%s:", c->file_name);
374 fput_loc(loc, stderr);
375 for (; *fmt ; fmt++) {
382 case '%': fputc(*fmt, stderr); break; // NOTEST
383 default: fputc('?', stderr); break; // NOTEST
385 type_print(t1, stderr);
388 type_print(t2, stderr);
397 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
399 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
400 t->txt.len, t->txt.txt);
404 ## Entities: declared and predeclared.
406 There are various "things" that the language and/or the interpreter
407 needs to know about to parse and execute a program. These include
408 types, variables, values, and executable code. These are all lumped
409 together under the term "entities" (calling them "objects" would be
410 confusing) and introduced here. The following section will present the
411 different specific code elements which comprise or manipulate these
416 Values come in a wide range of types, with more likely to be added.
417 Each type needs to be able to print its own values (for convenience at
418 least) as well as to compare two values, at least for equality and
419 possibly for order. For now, values might need to be duplicated and
420 freed, though eventually such manipulations will be better integrated
423 Rather than requiring every numeric type to support all numeric
424 operations (add, multiple, etc), we allow types to be able to present
425 as one of a few standard types: integer, float, and fraction. The
426 existence of these conversion functions eventually enable types to
427 determine if they are compatible with other types, though such types
428 have not yet been implemented.
430 Named type are stored in a simple linked list. Objects of each type are
431 "values" which are often passed around by value.
438 ## value union fields
446 void (*init)(struct type *type, struct value *val);
447 void (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
448 void (*print)(struct type *type, struct value *val);
449 void (*print_type)(struct type *type, FILE *f);
450 int (*cmp_order)(struct type *t1, struct type *t2,
451 struct value *v1, struct value *v2);
452 int (*cmp_eq)(struct type *t1, struct type *t2,
453 struct value *v1, struct value *v2);
454 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
455 void (*free)(struct type *type, struct value *val);
456 void (*free_type)(struct type *t);
457 long long (*to_int)(struct value *v);
458 double (*to_float)(struct value *v);
459 int (*to_mpq)(mpq_t *q, struct value *v);
468 struct type *typelist;
472 static struct type *find_type(struct parse_context *c, struct text s)
474 struct type *l = c->typelist;
477 text_cmp(l->name, s) != 0)
482 static struct type *add_type(struct parse_context *c, struct text s,
487 n = calloc(1, sizeof(*n));
490 n->next = c->typelist;
495 static void free_type(struct type *t)
497 /* The type is always a reference to something in the
498 * context, so we don't need to free anything.
502 static void free_value(struct type *type, struct value *v)
506 memset(v, 0x5a, type->size);
510 static void type_print(struct type *type, FILE *f)
513 fputs("*unknown*type*", f); // NOTEST
514 else if (type->name.len)
515 fprintf(f, "%.*s", type->name.len, type->name.txt);
516 else if (type->print_type)
517 type->print_type(type, f);
519 fputs("*invalid*type*", f); // NOTEST
522 static void val_init(struct type *type, struct value *val)
524 if (type && type->init)
525 type->init(type, val);
528 static void dup_value(struct type *type,
529 struct value *vold, struct value *vnew)
531 if (type && type->dup)
532 type->dup(type, vold, vnew);
535 static int value_cmp(struct type *tl, struct type *tr,
536 struct value *left, struct value *right)
538 if (tl && tl->cmp_order)
539 return tl->cmp_order(tl, tr, left, right);
540 if (tl && tl->cmp_eq) // NOTEST
541 return tl->cmp_eq(tl, tr, left, right); // NOTEST
545 static void print_value(struct type *type, struct value *v)
547 if (type && type->print)
548 type->print(type, v);
550 printf("*Unknown*"); // NOTEST
555 static void free_value(struct type *type, struct value *v);
556 static int type_compat(struct type *require, struct type *have, int rules);
557 static void type_print(struct type *type, FILE *f);
558 static void val_init(struct type *type, struct value *v);
559 static void dup_value(struct type *type,
560 struct value *vold, struct value *vnew);
561 static int value_cmp(struct type *tl, struct type *tr,
562 struct value *left, struct value *right);
563 static void print_value(struct type *type, struct value *v);
565 ###### free context types
567 while (context.typelist) {
568 struct type *t = context.typelist;
570 context.typelist = t->next;
576 Type can be specified for local variables, for fields in a structure,
577 for formal parameters to functions, and possibly elsewhere. Different
578 rules may apply in different contexts. As a minimum, a named type may
579 always be used. Currently the type of a formal parameter can be
580 different from types in other contexts, so we have a separate grammar
586 Type -> IDENTIFIER ${
587 $0 = find_type(c, $1.txt);
590 "error: undefined type", &$1);
597 FormalType -> Type ${ $0 = $<1; }$
598 ## formal type grammar
602 Values of the base types can be numbers, which we represent as
603 multi-precision fractions, strings, Booleans and labels. When
604 analysing the program we also need to allow for places where no value
605 is meaningful (type `Tnone`) and where we don't know what type to
606 expect yet (type is `NULL`).
608 Values are never shared, they are always copied when used, and freed
609 when no longer needed.
611 When propagating type information around the program, we need to
612 determine if two types are compatible, where type `NULL` is compatible
613 with anything. There are two special cases with type compatibility,
614 both related to the Conditional Statement which will be described
615 later. In some cases a Boolean can be accepted as well as some other
616 primary type, and in others any type is acceptable except a label (`Vlabel`).
617 A separate function encoding these cases will simplify some code later.
619 ###### type functions
621 int (*compat)(struct type *this, struct type *other);
625 static int type_compat(struct type *require, struct type *have, int rules)
627 if ((rules & Rboolok) && have == Tbool)
629 if ((rules & Rnolabel) && have == Tlabel)
631 if (!require || !have)
635 return require->compat(require, have);
637 return require == have;
642 #include "parse_string.h"
643 #include "parse_number.h"
646 myLDLIBS := libnumber.o libstring.o -lgmp
647 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
649 ###### type union fields
650 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
652 ###### value union fields
659 static void _free_value(struct type *type, struct value *v)
663 switch (type->vtype) {
665 case Vstr: free(v->str.txt); break;
666 case Vnum: mpq_clear(v->num); break;
672 ###### value functions
674 static void _val_init(struct type *type, struct value *val)
676 switch(type->vtype) {
677 case Vnone: // NOTEST
680 mpq_init(val->num); break;
682 val->str.txt = malloc(1);
694 static void _dup_value(struct type *type,
695 struct value *vold, struct value *vnew)
697 switch (type->vtype) {
698 case Vnone: // NOTEST
701 vnew->label = vold->label;
704 vnew->bool = vold->bool;
708 mpq_set(vnew->num, vold->num);
711 vnew->str.len = vold->str.len;
712 vnew->str.txt = malloc(vnew->str.len);
713 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
718 static int _value_cmp(struct type *tl, struct type *tr,
719 struct value *left, struct value *right)
723 return tl - tr; // NOTEST
725 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
726 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
727 case Vstr: cmp = text_cmp(left->str, right->str); break;
728 case Vbool: cmp = left->bool - right->bool; break;
729 case Vnone: cmp = 0; // NOTEST
734 static void _print_value(struct type *type, struct value *v)
736 switch (type->vtype) {
737 case Vnone: // NOTEST
738 printf("*no-value*"); break; // NOTEST
739 case Vlabel: // NOTEST
740 printf("*label-%p*", v->label); break; // NOTEST
742 printf("%.*s", v->str.len, v->str.txt); break;
744 printf("%s", v->bool ? "True":"False"); break;
749 mpf_set_q(fl, v->num);
750 gmp_printf("%Fg", fl);
757 static void _free_value(struct type *type, struct value *v);
759 static struct type base_prototype = {
761 .print = _print_value,
762 .cmp_order = _value_cmp,
763 .cmp_eq = _value_cmp,
768 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
771 static struct type *add_base_type(struct parse_context *c, char *n,
772 enum vtype vt, int size)
774 struct text txt = { n, strlen(n) };
777 t = add_type(c, txt, &base_prototype);
780 t->align = size > sizeof(void*) ? sizeof(void*) : size;
781 if (t->size & (t->align - 1))
782 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
786 ###### context initialization
788 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
789 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
790 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
791 Tnone = add_base_type(&context, "none", Vnone, 0);
792 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
796 Variables are scoped named values. We store the names in a linked list
797 of "bindings" sorted in lexical order, and use sequential search and
804 struct binding *next; // in lexical order
808 This linked list is stored in the parse context so that "reduce"
809 functions can find or add variables, and so the analysis phase can
810 ensure that every variable gets a type.
814 struct binding *varlist; // In lexical order
818 static struct binding *find_binding(struct parse_context *c, struct text s)
820 struct binding **l = &c->varlist;
825 (cmp = text_cmp((*l)->name, s)) < 0)
829 n = calloc(1, sizeof(*n));
836 Each name can be linked to multiple variables defined in different
837 scopes. Each scope starts where the name is declared and continues
838 until the end of the containing code block. Scopes of a given name
839 cannot nest, so a declaration while a name is in-scope is an error.
841 ###### binding fields
842 struct variable *var;
846 struct variable *previous;
848 struct binding *name;
849 struct exec *where_decl;// where name was declared
850 struct exec *where_set; // where type was set
854 When a scope closes, the values of the variables might need to be freed.
855 This happens in the context of some `struct exec` and each `exec` will
856 need to know which variables need to be freed when it completes.
859 struct variable *to_free;
861 ####### variable fields
862 struct exec *cleanup_exec;
863 struct variable *next_free;
865 ####### interp exec cleanup
868 for (v = e->to_free; v; v = v->next_free) {
869 struct value *val = var_value(c, v);
870 free_value(v->type, val);
875 static void variable_unlink_exec(struct variable *v)
877 struct variable **vp;
878 if (!v->cleanup_exec)
880 for (vp = &v->cleanup_exec->to_free;
881 *vp; vp = &(*vp)->next_free) {
885 v->cleanup_exec = NULL;
890 While the naming seems strange, we include local constants in the
891 definition of variables. A name declared `var := value` can
892 subsequently be changed, but a name declared `var ::= value` cannot -
895 ###### variable fields
898 Scopes in parallel branches can be partially merged. More
899 specifically, if a given name is declared in both branches of an
900 if/else then its scope is a candidate for merging. Similarly if
901 every branch of an exhaustive switch (e.g. has an "else" clause)
902 declares a given name, then the scopes from the branches are
903 candidates for merging.
905 Note that names declared inside a loop (which is only parallel to
906 itself) are never visible after the loop. Similarly names defined in
907 scopes which are not parallel, such as those started by `for` and
908 `switch`, are never visible after the scope. Only variables defined in
909 both `then` and `else` (including the implicit then after an `if`, and
910 excluding `then` used with `for`) and in all `case`s and `else` of a
911 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
913 Labels, which are a bit like variables, follow different rules.
914 Labels are not explicitly declared, but if an undeclared name appears
915 in a context where a label is legal, that effectively declares the
916 name as a label. The declaration remains in force (or in scope) at
917 least to the end of the immediately containing block and conditionally
918 in any larger containing block which does not declare the name in some
919 other way. Importantly, the conditional scope extension happens even
920 if the label is only used in one parallel branch of a conditional --
921 when used in one branch it is treated as having been declared in all
924 Merge candidates are tentatively visible beyond the end of the
925 branching statement which creates them. If the name is used, the
926 merge is affirmed and they become a single variable visible at the
927 outer layer. If not - if it is redeclared first - the merge lapses.
929 To track scopes we have an extra stack, implemented as a linked list,
930 which roughly parallels the parse stack and which is used exclusively
931 for scoping. When a new scope is opened, a new frame is pushed and
932 the child-count of the parent frame is incremented. This child-count
933 is used to distinguish between the first of a set of parallel scopes,
934 in which declared variables must not be in scope, and subsequent
935 branches, whether they may already be conditionally scoped.
937 To push a new frame *before* any code in the frame is parsed, we need a
938 grammar reduction. This is most easily achieved with a grammar
939 element which derives the empty string, and creates the new scope when
940 it is recognised. This can be placed, for example, between a keyword
941 like "if" and the code following it.
945 struct scope *parent;
951 struct scope *scope_stack;
954 static void scope_pop(struct parse_context *c)
956 struct scope *s = c->scope_stack;
958 c->scope_stack = s->parent;
963 static void scope_push(struct parse_context *c)
965 struct scope *s = calloc(1, sizeof(*s));
967 c->scope_stack->child_count += 1;
968 s->parent = c->scope_stack;
976 OpenScope -> ${ scope_push(c); }$
978 Each variable records a scope depth and is in one of four states:
980 - "in scope". This is the case between the declaration of the
981 variable and the end of the containing block, and also between
982 the usage with affirms a merge and the end of that block.
984 The scope depth is not greater than the current parse context scope
985 nest depth. When the block of that depth closes, the state will
986 change. To achieve this, all "in scope" variables are linked
987 together as a stack in nesting order.
989 - "pending". The "in scope" block has closed, but other parallel
990 scopes are still being processed. So far, every parallel block at
991 the same level that has closed has declared the name.
993 The scope depth is the depth of the last parallel block that
994 enclosed the declaration, and that has closed.
996 - "conditionally in scope". The "in scope" block and all parallel
997 scopes have closed, and no further mention of the name has been seen.
998 This state includes a secondary nest depth (`min_depth`) which records
999 the outermost scope seen since the variable became conditionally in
1000 scope. If a use of the name is found, the variable becomes "in scope"
1001 and that secondary depth becomes the recorded scope depth. If the
1002 name is declared as a new variable, the old variable becomes "out of
1003 scope" and the recorded scope depth stays unchanged.
1005 - "out of scope". The variable is neither in scope nor conditionally
1006 in scope. It is permanently out of scope now and can be removed from
1007 the "in scope" stack.
1009 ###### variable fields
1010 int depth, min_depth;
1011 enum { OutScope, PendingScope, CondScope, InScope } scope;
1012 struct variable *in_scope;
1014 ###### parse context
1016 struct variable *in_scope;
1018 All variables with the same name are linked together using the
1019 'previous' link. Those variable that have been affirmatively merged all
1020 have a 'merged' pointer that points to one primary variable - the most
1021 recently declared instance. When merging variables, we need to also
1022 adjust the 'merged' pointer on any other variables that had previously
1023 been merged with the one that will no longer be primary.
1025 A variable that is no longer the most recent instance of a name may
1026 still have "pending" scope, if it might still be merged with most
1027 recent instance. These variables don't really belong in the
1028 "in_scope" list, but are not immediately removed when a new instance
1029 is found. Instead, they are detected and ignored when considering the
1030 list of in_scope names.
1032 The storage of the value of a variable will be described later. For now
1033 we just need to know that when a variable goes out of scope, it might
1034 need to be freed. For this we need to be able to find it, so assume that
1035 `var_value()` will provide that.
1037 ###### variable fields
1038 struct variable *merged;
1040 ###### ast functions
1042 static void variable_merge(struct variable *primary, struct variable *secondary)
1046 primary = primary->merged;
1048 for (v = primary->previous; v; v=v->previous)
1049 if (v == secondary || v == secondary->merged ||
1050 v->merged == secondary ||
1051 v->merged == secondary->merged) {
1052 v->scope = OutScope;
1053 v->merged = primary;
1054 variable_unlink_exec(v);
1058 ###### forward decls
1059 static struct value *var_value(struct parse_context *c, struct variable *v);
1061 ###### free global vars
1063 while (context.varlist) {
1064 struct binding *b = context.varlist;
1065 struct variable *v = b->var;
1066 context.varlist = b->next;
1069 struct variable *next = v->previous;
1072 free_value(v->type, var_value(&context, v));
1074 // This is a global constant
1075 free_exec(v->where_decl);
1082 #### Manipulating Bindings
1084 When a name is conditionally visible, a new declaration discards the
1085 old binding - the condition lapses. Conversely a usage of the name
1086 affirms the visibility and extends it to the end of the containing
1087 block - i.e. the block that contains both the original declaration and
1088 the latest usage. This is determined from `min_depth`. When a
1089 conditionally visible variable gets affirmed like this, it is also
1090 merged with other conditionally visible variables with the same name.
1092 When we parse a variable declaration we either report an error if the
1093 name is currently bound, or create a new variable at the current nest
1094 depth if the name is unbound or bound to a conditionally scoped or
1095 pending-scope variable. If the previous variable was conditionally
1096 scoped, it and its homonyms becomes out-of-scope.
1098 When we parse a variable reference (including non-declarative assignment
1099 "foo = bar") we report an error if the name is not bound or is bound to
1100 a pending-scope variable; update the scope if the name is bound to a
1101 conditionally scoped variable; or just proceed normally if the named
1102 variable is in scope.
1104 When we exit a scope, any variables bound at this level are either
1105 marked out of scope or pending-scoped, depending on whether the scope
1106 was sequential or parallel. Here a "parallel" scope means the "then"
1107 or "else" part of a conditional, or any "case" or "else" branch of a
1108 switch. Other scopes are "sequential".
1110 When exiting a parallel scope we check if there are any variables that
1111 were previously pending and are still visible. If there are, then
1112 they weren't redeclared in the most recent scope, so they cannot be
1113 merged and must become out-of-scope. If it is not the first of
1114 parallel scopes (based on `child_count`), we check that there was a
1115 previous binding that is still pending-scope. If there isn't, the new
1116 variable must now be out-of-scope.
1118 When exiting a sequential scope that immediately enclosed parallel
1119 scopes, we need to resolve any pending-scope variables. If there was
1120 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1121 we need to mark all pending-scope variable as out-of-scope. Otherwise
1122 all pending-scope variables become conditionally scoped.
1125 enum closetype { CloseSequential, CloseParallel, CloseElse };
1127 ###### ast functions
1129 static struct variable *var_decl(struct parse_context *c, struct text s)
1131 struct binding *b = find_binding(c, s);
1132 struct variable *v = b->var;
1134 switch (v ? v->scope : OutScope) {
1136 /* Caller will report the error */
1140 v && v->scope == CondScope;
1142 v->scope = OutScope;
1146 v = calloc(1, sizeof(*v));
1147 v->previous = b->var;
1151 v->min_depth = v->depth = c->scope_depth;
1153 v->in_scope = c->in_scope;
1159 static struct variable *var_ref(struct parse_context *c, struct text s)
1161 struct binding *b = find_binding(c, s);
1162 struct variable *v = b->var;
1163 struct variable *v2;
1165 switch (v ? v->scope : OutScope) {
1168 /* Caller will report the error */
1171 /* All CondScope variables of this name need to be merged
1172 * and become InScope
1174 v->depth = v->min_depth;
1176 for (v2 = v->previous;
1177 v2 && v2->scope == CondScope;
1179 variable_merge(v, v2);
1187 static void var_block_close(struct parse_context *c, enum closetype ct,
1190 /* Close off all variables that are in_scope.
1191 * Some variables in c->scope may already be not-in-scope,
1192 * such as when a PendingScope variable is hidden by a new
1193 * variable with the same name.
1194 * So we check for v->name->var != v and drop them.
1195 * If we choose to make a variable OutScope, we drop it
1198 struct variable *v, **vp, *v2;
1201 for (vp = &c->in_scope;
1202 (v = *vp) && v->min_depth > c->scope_depth;
1203 (v->scope == OutScope || v->name->var != v)
1204 ? (*vp = v->in_scope, 0)
1205 : ( vp = &v->in_scope, 0)) {
1206 v->min_depth = c->scope_depth;
1207 if (v->name->var != v)
1208 /* This is still in scope, but we haven't just
1212 v->min_depth = c->scope_depth;
1213 if (v->scope == InScope && e) {
1214 /* This variable gets cleaned up when 'e' finishes */
1215 variable_unlink_exec(v);
1216 v->cleanup_exec = e;
1217 v->next_free = e->to_free;
1222 case CloseParallel: /* handle PendingScope */
1226 if (c->scope_stack->child_count == 1)
1227 /* first among parallel branches */
1228 v->scope = PendingScope;
1229 else if (v->previous &&
1230 v->previous->scope == PendingScope)
1231 /* all previous branches used name */
1232 v->scope = PendingScope;
1233 else if (v->type == Tlabel)
1234 /* Labels remain pending even when not used */
1235 v->scope = PendingScope; // UNTESTED
1237 v->scope = OutScope;
1238 if (ct == CloseElse) {
1239 /* All Pending variables with this name
1240 * are now Conditional */
1242 v2 && v2->scope == PendingScope;
1244 v2->scope = CondScope;
1248 /* Not possible as it would require
1249 * parallel scope to be nested immediately
1250 * in a parallel scope, and that never
1254 /* Not possible as we already tested for
1260 case CloseSequential:
1261 if (v->type == Tlabel)
1262 v->scope = PendingScope;
1265 v->scope = OutScope;
1268 /* There was no 'else', so we can only become
1269 * conditional if we know the cases were exhaustive,
1270 * and that doesn't mean anything yet.
1271 * So only labels become conditional..
1274 v2 && v2->scope == PendingScope;
1276 if (v2->type == Tlabel)
1277 v2->scope = CondScope;
1279 v2->scope = OutScope;
1282 case OutScope: break;
1291 The value of a variable is store separately from the variable, on an
1292 analogue of a stack frame. There are (currently) two frames that can be
1293 active. A global frame which currently only stores constants, and a
1294 stacked frame which stores local variables. Each variable knows if it
1295 is global or not, and what its index into the frame is.
1297 Values in the global frame are known immediately they are relevant, so
1298 the frame needs to be reallocated as it grows so it can store those
1299 values. The local frame doesn't get values until the interpreted phase
1300 is started, so there is no need to allocate until the size is known.
1302 We initialize the `frame_pos` to an impossible value, so that we can
1303 tell if it was set or not later.
1305 ###### variable fields
1309 ###### variable init
1312 ###### parse context
1314 short global_size, global_alloc;
1316 void *global, *local;
1318 ###### ast functions
1320 static struct value *var_value(struct parse_context *c, struct variable *v)
1323 if (!c->local || !v->type)
1324 return NULL; // NOTEST
1325 if (v->frame_pos + v->type->size > c->local_size) {
1326 printf("INVALID frame_pos\n"); // NOTEST
1329 return c->local + v->frame_pos;
1331 if (c->global_size > c->global_alloc) {
1332 int old = c->global_alloc;
1333 c->global_alloc = (c->global_size | 1023) + 1024;
1334 c->global = realloc(c->global, c->global_alloc);
1335 memset(c->global + old, 0, c->global_alloc - old);
1337 return c->global + v->frame_pos;
1340 static struct value *global_alloc(struct parse_context *c, struct type *t,
1341 struct variable *v, struct value *init)
1344 struct variable scratch;
1346 if (t->prepare_type)
1347 t->prepare_type(c, t, 1); // NOTEST
1349 if (c->global_size & (t->align - 1))
1350 c->global_size = (c->global_size + t->align) & ~(t->align-1);
1355 v->frame_pos = c->global_size;
1357 c->global_size += v->type->size;
1358 ret = var_value(c, v);
1360 memcpy(ret, init, t->size);
1366 As global values are found -- struct field initializers, labels etc --
1367 `global_alloc()` is called to record the value in the global frame.
1369 When the program is fully parsed, we need to walk the list of variables
1370 to find any that weren't merged away and that aren't global, and to
1371 calculate the frame size and assign a frame position for each
1372 variable. For this we have `scope_finalize()`.
1374 ###### ast functions
1376 static int 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;
1391 if (size & (t->align - 1))
1392 size = (size + t->align) & ~(t->align-1);
1393 v->frame_pos = size;
1394 size += v->type->size;
1400 ###### free context storage
1401 free(context.global);
1405 Executables can be lots of different things. In many cases an
1406 executable is just an operation combined with one or two other
1407 executables. This allows for expressions and lists etc. Other times an
1408 executable is something quite specific like a constant or variable name.
1409 So we define a `struct exec` to be a general executable with a type, and
1410 a `struct binode` which is a subclass of `exec`, forms a node in a
1411 binary tree, and holds an operation. There will be other subclasses,
1412 and to access these we need to be able to `cast` the `exec` into the
1413 various other types. The first field in any `struct exec` is the type
1414 from the `exec_types` enum.
1417 #define cast(structname, pointer) ({ \
1418 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1419 if (__mptr && *__mptr != X##structname) abort(); \
1420 (struct structname *)( (char *)__mptr);})
1422 #define new(structname) ({ \
1423 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1424 __ptr->type = X##structname; \
1425 __ptr->line = -1; __ptr->column = -1; \
1428 #define new_pos(structname, token) ({ \
1429 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1430 __ptr->type = X##structname; \
1431 __ptr->line = token.line; __ptr->column = token.col; \
1440 enum exec_types type;
1449 struct exec *left, *right;
1452 ###### ast functions
1454 static int __fput_loc(struct exec *loc, FILE *f)
1458 if (loc->line >= 0) {
1459 fprintf(f, "%d:%d: ", loc->line, loc->column);
1462 if (loc->type == Xbinode)
1463 return __fput_loc(cast(binode,loc)->left, f) ||
1464 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1467 static void fput_loc(struct exec *loc, FILE *f)
1469 if (!__fput_loc(loc, f))
1470 fprintf(f, "??:??: ");
1473 Each different type of `exec` node needs a number of functions defined,
1474 a bit like methods. We must be able to free it, print it, analyse it
1475 and execute it. Once we have specific `exec` types we will need to
1476 parse them too. Let's take this a bit more slowly.
1480 The parser generator requires a `free_foo` function for each struct
1481 that stores attributes and they will often be `exec`s and subtypes
1482 there-of. So we need `free_exec` which can handle all the subtypes,
1483 and we need `free_binode`.
1485 ###### ast functions
1487 static void free_binode(struct binode *b)
1492 free_exec(b->right);
1496 ###### core functions
1497 static void free_exec(struct exec *e)
1506 ###### forward decls
1508 static void free_exec(struct exec *e);
1510 ###### free exec cases
1511 case Xbinode: free_binode(cast(binode, e)); break;
1515 Printing an `exec` requires that we know the current indent level for
1516 printing line-oriented components. As will become clear later, we
1517 also want to know what sort of bracketing to use.
1519 ###### ast functions
1521 static void do_indent(int i, char *str)
1528 ###### core functions
1529 static void print_binode(struct binode *b, int indent, int bracket)
1533 ## print binode cases
1537 static void print_exec(struct exec *e, int indent, int bracket)
1543 print_binode(cast(binode, e), indent, bracket); break;
1548 do_indent(indent, "/* FREE");
1549 for (v = e->to_free; v; v = v->next_free) {
1550 printf(" %.*s", v->name->name.len, v->name->name.txt);
1551 if (v->frame_pos >= 0)
1552 printf("(%d+%d)", v->frame_pos,
1553 v->type ? v->type->size:0);
1559 ###### forward decls
1561 static void print_exec(struct exec *e, int indent, int bracket);
1565 As discussed, analysis involves propagating type requirements around the
1566 program and looking for errors.
1568 So `propagate_types` is passed an expected type (being a `struct type`
1569 pointer together with some `val_rules` flags) that the `exec` is
1570 expected to return, and returns the type that it does return, either
1571 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1572 by reference. It is set to `0` when an error is found, and `2` when
1573 any change is made. If it remains unchanged at `1`, then no more
1574 propagation is needed.
1578 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1582 if (rules & Rnolabel)
1583 fputs(" (labels not permitted)", stderr);
1586 ###### forward decls
1587 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1588 struct type *type, int rules);
1589 ###### core functions
1591 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1592 struct type *type, int rules)
1599 switch (prog->type) {
1602 struct binode *b = cast(binode, prog);
1604 ## propagate binode cases
1608 ## propagate exec cases
1613 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1614 struct type *type, int rules)
1616 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1625 Interpreting an `exec` doesn't require anything but the `exec`. State
1626 is stored in variables and each variable will be directly linked from
1627 within the `exec` tree. The exception to this is the `main` function
1628 which needs to look at command line arguments. This function will be
1629 interpreted separately.
1631 Each `exec` can return a value combined with a type in `struct lrval`.
1632 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1633 the location of a value, which can be updated, in `lval`. Others will
1634 set `lval` to NULL indicating that there is a value of appropriate type
1637 ###### core functions
1641 struct value rval, *lval;
1644 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1646 static struct value interp_exec(struct parse_context *c, struct exec *e,
1647 struct type **typeret)
1649 struct lrval ret = _interp_exec(c, e);
1651 if (!ret.type) abort();
1653 *typeret = ret.type;
1655 dup_value(ret.type, ret.lval, &ret.rval);
1659 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1660 struct type **typeret)
1662 struct lrval ret = _interp_exec(c, e);
1665 *typeret = ret.type;
1667 free_value(ret.type, &ret.rval);
1671 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1674 struct value rv = {}, *lrv = NULL;
1675 struct type *rvtype;
1677 rvtype = ret.type = Tnone;
1687 struct binode *b = cast(binode, e);
1688 struct value left, right, *lleft;
1689 struct type *ltype, *rtype;
1690 ltype = rtype = Tnone;
1692 ## interp binode cases
1694 free_value(ltype, &left);
1695 free_value(rtype, &right);
1698 ## interp exec cases
1703 ## interp exec cleanup
1709 Now that we have the shape of the interpreter in place we can add some
1710 complex types and connected them in to the data structures and the
1711 different phases of parse, analyse, print, interpret.
1713 Thus far we have arrays and structs.
1717 Arrays can be declared by giving a size and a type, as `[size]type' so
1718 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1719 size can be either a literal number, or a named constant. Some day an
1720 arbitrary expression will be supported.
1722 As a formal parameter to a function, the array can be declared with a
1723 new variable as the size: `name:[size::number]string`. The `size`
1724 variable is set to the size of the array and must be a constant. As
1725 `number` is the only supported type, it can be left out:
1726 `name:[size::]string`.
1728 Arrays cannot be assigned. When pointers are introduced we will also
1729 introduce array slices which can refer to part or all of an array -
1730 the assignment syntax will create a slice. For now, an array can only
1731 ever be referenced by the name it is declared with. It is likely that
1732 a "`copy`" primitive will eventually be define which can be used to
1733 make a copy of an array with controllable recursive depth.
1735 For now we have two sorts of array, those with fixed size either because
1736 it is given as a literal number or because it is a struct member (which
1737 cannot have a runtime-changing size), and those with a size that is
1738 determined at runtime - local variables with a const size. The former
1739 have their size calculated at parse time, the latter at run time.
1741 For the latter type, the `size` field of the type is the size of a
1742 pointer, and the array is reallocated every time it comes into scope.
1744 We differentiate struct fields with a const size from local variables
1745 with a const size by whether they are prepared at parse time or not.
1747 ###### type union fields
1750 int unspec; // size is unspecified - vsize must be set.
1753 struct variable *vsize;
1754 struct type *member;
1757 ###### value union fields
1758 void *array; // used if not static_size
1760 ###### value functions
1762 static void array_prepare_type(struct parse_context *c, struct type *type,
1765 struct value *vsize;
1767 if (!type->array.vsize || type->array.static_size)
1770 vsize = var_value(c, type->array.vsize);
1772 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1773 type->array.size = mpz_get_si(q);
1777 type->array.static_size = 1;
1778 type->size = type->array.size * type->array.member->size;
1779 type->align = type->array.member->align;
1783 static void array_init(struct type *type, struct value *val)
1786 void *ptr = val->ptr;
1790 if (!type->array.static_size) {
1791 val->array = calloc(type->array.size,
1792 type->array.member->size);
1795 for (i = 0; i < type->array.size; i++) {
1797 v = (void*)ptr + i * type->array.member->size;
1798 val_init(type->array.member, v);
1802 static void array_free(struct type *type, struct value *val)
1805 void *ptr = val->ptr;
1807 if (!type->array.static_size)
1809 for (i = 0; i < type->array.size; i++) {
1811 v = (void*)ptr + i * type->array.member->size;
1812 free_value(type->array.member, v);
1814 if (!type->array.static_size)
1818 static int array_compat(struct type *require, struct type *have)
1820 if (have->compat != require->compat)
1822 /* Both are arrays, so we can look at details */
1823 if (!type_compat(require->array.member, have->array.member, 0))
1825 if (have->array.unspec && require->array.unspec) {
1826 if (have->array.vsize && require->array.vsize &&
1827 have->array.vsize != require->array.vsize) // UNTESTED
1828 /* sizes might not be the same */
1829 return 0; // UNTESTED
1832 if (have->array.unspec || require->array.unspec)
1833 return 1; // UNTESTED
1834 if (require->array.vsize == NULL && have->array.vsize == NULL)
1835 return require->array.size == have->array.size;
1837 return require->array.vsize == have->array.vsize; // UNTESTED
1840 static void array_print_type(struct type *type, FILE *f)
1843 if (type->array.vsize) {
1844 struct binding *b = type->array.vsize->name;
1845 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1846 type->array.unspec ? "::" : "");
1848 fprintf(f, "%d]", type->array.size);
1849 type_print(type->array.member, f);
1852 static struct type array_prototype = {
1854 .prepare_type = array_prepare_type,
1855 .print_type = array_print_type,
1856 .compat = array_compat,
1858 .size = sizeof(void*),
1859 .align = sizeof(void*),
1862 ###### declare terminals
1867 | [ NUMBER ] Type ${ {
1870 struct text noname = { "", 0 };
1873 $0 = t = add_type(c, noname, &array_prototype);
1874 t->array.member = $<4;
1875 t->array.vsize = NULL;
1876 if (number_parse(num, tail, $2.txt) == 0)
1877 tok_err(c, "error: unrecognised number", &$2);
1879 tok_err(c, "error: unsupported number suffix", &$2);
1881 t->array.size = mpz_get_ui(mpq_numref(num));
1882 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1883 tok_err(c, "error: array size must be an integer",
1885 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1886 tok_err(c, "error: array size is too large",
1890 t->array.static_size = 1;
1891 t->size = t->array.size * t->array.member->size;
1892 t->align = t->array.member->align;
1895 | [ IDENTIFIER ] Type ${ {
1896 struct variable *v = var_ref(c, $2.txt);
1897 struct text noname = { "", 0 };
1900 tok_err(c, "error: name undeclared", &$2);
1901 else if (!v->constant)
1902 tok_err(c, "error: array size must be a constant", &$2);
1904 $0 = add_type(c, noname, &array_prototype);
1905 $0->array.member = $<4;
1907 $0->array.vsize = v;
1912 OptType -> Type ${ $0 = $<1; }$
1915 ###### formal type grammar
1917 | [ IDENTIFIER :: OptType ] Type ${ {
1918 struct variable *v = var_decl(c, $ID.txt);
1919 struct text noname = { "", 0 };
1925 $0 = add_type(c, noname, &array_prototype);
1926 $0->array.member = $<6;
1928 $0->array.unspec = 1;
1929 $0->array.vsize = v;
1935 ###### variable grammar
1937 | Variable [ Expression ] ${ {
1938 struct binode *b = new(binode);
1945 ###### print binode cases
1947 print_exec(b->left, -1, bracket);
1949 print_exec(b->right, -1, bracket);
1953 ###### propagate binode cases
1955 /* left must be an array, right must be a number,
1956 * result is the member type of the array
1958 propagate_types(b->right, c, ok, Tnum, 0);
1959 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1960 if (!t || t->compat != array_compat) {
1961 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1964 if (!type_compat(type, t->array.member, rules)) {
1965 type_err(c, "error: have %1 but need %2", prog,
1966 t->array.member, rules, type);
1968 return t->array.member;
1972 ###### interp binode cases
1978 lleft = linterp_exec(c, b->left, <ype);
1979 right = interp_exec(c, b->right, &rtype);
1981 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1985 if (ltype->array.static_size)
1988 ptr = *(void**)lleft;
1989 rvtype = ltype->array.member;
1990 if (i >= 0 && i < ltype->array.size)
1991 lrv = ptr + i * rvtype->size;
1993 val_init(ltype->array.member, &rv);
2000 A `struct` is a data-type that contains one or more other data-types.
2001 It differs from an array in that each member can be of a different
2002 type, and they are accessed by name rather than by number. Thus you
2003 cannot choose an element by calculation, you need to know what you
2006 The language makes no promises about how a given structure will be
2007 stored in memory - it is free to rearrange fields to suit whatever
2008 criteria seems important.
2010 Structs are declared separately from program code - they cannot be
2011 declared in-line in a variable declaration like arrays can. A struct
2012 is given a name and this name is used to identify the type - the name
2013 is not prefixed by the word `struct` as it would be in C.
2015 Structs are only treated as the same if they have the same name.
2016 Simply having the same fields in the same order is not enough. This
2017 might change once we can create structure initializers from a list of
2020 Each component datum is identified much like a variable is declared,
2021 with a name, one or two colons, and a type. The type cannot be omitted
2022 as there is no opportunity to deduce the type from usage. An initial
2023 value can be given following an equals sign, so
2025 ##### Example: a struct type
2031 would declare a type called "complex" which has two number fields,
2032 each initialised to zero.
2034 Struct will need to be declared separately from the code that uses
2035 them, so we will need to be able to print out the declaration of a
2036 struct when reprinting the whole program. So a `print_type_decl` type
2037 function will be needed.
2039 ###### type union fields
2051 ###### type functions
2052 void (*print_type_decl)(struct type *type, FILE *f);
2054 ###### value functions
2056 static void structure_init(struct type *type, struct value *val)
2060 for (i = 0; i < type->structure.nfields; i++) {
2062 v = (void*) val->ptr + type->structure.fields[i].offset;
2063 if (type->structure.fields[i].init)
2064 dup_value(type->structure.fields[i].type,
2065 type->structure.fields[i].init,
2068 val_init(type->structure.fields[i].type, v);
2072 static void structure_free(struct type *type, struct value *val)
2076 for (i = 0; i < type->structure.nfields; i++) {
2078 v = (void*)val->ptr + type->structure.fields[i].offset;
2079 free_value(type->structure.fields[i].type, v);
2083 static void structure_free_type(struct type *t)
2086 for (i = 0; i < t->structure.nfields; i++)
2087 if (t->structure.fields[i].init) {
2088 free_value(t->structure.fields[i].type,
2089 t->structure.fields[i].init);
2091 free(t->structure.fields);
2094 static struct type structure_prototype = {
2095 .init = structure_init,
2096 .free = structure_free,
2097 .free_type = structure_free_type,
2098 .print_type_decl = structure_print_type,
2112 ###### free exec cases
2114 free_exec(cast(fieldref, e)->left);
2118 ###### declare terminals
2121 ###### variable grammar
2123 | Variable . IDENTIFIER ${ {
2124 struct fieldref *fr = new_pos(fieldref, $2);
2131 ###### print exec cases
2135 struct fieldref *f = cast(fieldref, e);
2136 print_exec(f->left, -1, bracket);
2137 printf(".%.*s", f->name.len, f->name.txt);
2141 ###### ast functions
2142 static int find_struct_index(struct type *type, struct text field)
2145 for (i = 0; i < type->structure.nfields; i++)
2146 if (text_cmp(type->structure.fields[i].name, field) == 0)
2151 ###### propagate exec cases
2155 struct fieldref *f = cast(fieldref, prog);
2156 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2159 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2161 else if (st->init != structure_init)
2162 type_err(c, "error: field reference attempted on %1, not a struct",
2163 f->left, st, 0, NULL);
2164 else if (f->index == -2) {
2165 f->index = find_struct_index(st, f->name);
2167 type_err(c, "error: cannot find requested field in %1",
2168 f->left, st, 0, NULL);
2170 if (f->index >= 0) {
2171 struct type *ft = st->structure.fields[f->index].type;
2172 if (!type_compat(type, ft, rules))
2173 type_err(c, "error: have %1 but need %2", prog,
2180 ###### interp exec cases
2183 struct fieldref *f = cast(fieldref, e);
2185 struct value *lleft = linterp_exec(c, f->left, <ype);
2186 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2187 rvtype = ltype->structure.fields[f->index].type;
2193 struct fieldlist *prev;
2197 ###### ast functions
2198 static void free_fieldlist(struct fieldlist *f)
2202 free_fieldlist(f->prev);
2204 free_value(f->f.type, f->f.init); // UNTESTED
2205 free(f->f.init); // UNTESTED
2210 ###### top level grammar
2211 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2213 add_type(c, $2.txt, &structure_prototype);
2215 struct fieldlist *f;
2217 for (f = $3; f; f=f->prev)
2220 t->structure.nfields = cnt;
2221 t->structure.fields = calloc(cnt, sizeof(struct field));
2224 int a = f->f.type->align;
2226 t->structure.fields[cnt] = f->f;
2227 if (t->size & (a-1))
2228 t->size = (t->size | (a-1)) + 1;
2229 t->structure.fields[cnt].offset = t->size;
2230 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2239 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2240 | { SimpleFieldList } ${ $0 = $<SFL; }$
2241 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2242 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2244 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2245 | FieldLines SimpleFieldList Newlines ${
2250 SimpleFieldList -> Field ${ $0 = $<F; }$
2251 | SimpleFieldList ; Field ${
2255 | SimpleFieldList ; ${
2258 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2260 Field -> IDENTIFIER : Type = Expression ${ {
2263 $0 = calloc(1, sizeof(struct fieldlist));
2264 $0->f.name = $1.txt;
2269 propagate_types($<5, c, &ok, $3, 0);
2272 c->parse_error = 1; // UNTESTED
2274 struct value vl = interp_exec(c, $5, NULL);
2275 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2278 | IDENTIFIER : Type ${
2279 $0 = calloc(1, sizeof(struct fieldlist));
2280 $0->f.name = $1.txt;
2282 if ($0->f.type->prepare_type)
2283 $0->f.type->prepare_type(c, $0->f.type, 1);
2286 ###### forward decls
2287 static void structure_print_type(struct type *t, FILE *f);
2289 ###### value functions
2290 static void structure_print_type(struct type *t, FILE *f) // UNTESTED
2294 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2296 for (i = 0; i < t->structure.nfields; i++) {
2297 struct field *fl = t->structure.fields + i;
2298 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2299 type_print(fl->type, f);
2300 if (fl->type->print && fl->init) {
2302 if (fl->type == Tstr)
2303 fprintf(f, "\""); // UNTESTED
2304 print_value(fl->type, fl->init);
2305 if (fl->type == Tstr)
2306 fprintf(f, "\""); // UNTESTED
2312 ###### print type decls
2314 struct type *t; // UNTESTED
2317 while (target != 0) {
2319 for (t = context.typelist; t ; t=t->next)
2320 if (t->print_type_decl && !t->check_args) {
2329 t->print_type_decl(t, stdout);
2337 A function is a chunk of code which can be passed parameters and can
2338 return results (though results are not yet implemented). Each function
2339 has a type which includes the set of parameters and the return value.
2340 As yet these types cannot be declared separately from the function
2343 The parameters can be specified either in parentheses as a list, such as
2345 ##### Example: function 1
2347 func main(av:[ac::number]string)
2350 or as an indented list of one parameter per line
2352 ##### Example: function 2
2355 argv:[argc::number]string
2359 For constructing these lists we use a `List` binode, which will be
2360 further detailed when Expression Lists are introduced.
2362 ###### type union fields
2365 struct binode *params;
2369 ###### value union fields
2370 struct exec *function;
2372 ###### type functions
2373 void (*check_args)(struct parse_context *c, int *ok,
2374 struct type *require, struct exec *args);
2376 ###### value functions
2378 static void function_free(struct type *type, struct value *val)
2380 free_exec(val->function);
2381 val->function = NULL;
2384 static int function_compat(struct type *require, struct type *have)
2386 // FIXME can I do anything here yet?
2390 static void function_check_args(struct parse_context *c, int *ok,
2391 struct type *require, struct exec *args)
2393 /* This should be 'compat', but we don't have a 'tuple' type to
2394 * hold the type of 'args'
2396 struct binode *arg = cast(binode, args);
2397 struct binode *param = require->function.params;
2400 struct var *pv = cast(var, param->left);
2402 type_err(c, "error: insufficient arguments to function.",
2403 args, NULL, 0, NULL);
2407 propagate_types(arg->left, c, ok, pv->var->type, 0);
2408 param = cast(binode, param->right);
2409 arg = cast(binode, arg->right);
2412 type_err(c, "error: too many arguments to function.",
2413 args, NULL, 0, NULL);
2416 static void function_print(struct type *type, struct value *val)
2418 print_exec(val->function, 1, 0);
2421 static void function_print_type_decl(struct type *type, FILE *f)
2425 for (b = type->function.params; b; b = cast(binode, b->right)) {
2426 struct variable *v = cast(var, b->left)->var;
2427 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
2428 v->constant ? "::" : ":");
2429 type_print(v->type, f);
2436 static void function_free_type(struct type *t)
2438 free_exec(t->function.params);
2441 static struct type function_prototype = {
2442 .size = sizeof(void*),
2443 .align = sizeof(void*),
2444 .free = function_free,
2445 .compat = function_compat,
2446 .check_args = function_check_args,
2447 .print = function_print,
2448 .print_type_decl = function_print_type_decl,
2449 .free_type = function_free_type,
2452 ###### declare terminals
2462 FuncName -> IDENTIFIER ${ {
2463 struct variable *v = var_decl(c, $1.txt);
2464 struct var *e = new_pos(var, $1);
2470 v = var_ref(c, $1.txt);
2472 type_err(c, "error: function '%v' redeclared",
2474 type_err(c, "info: this is where '%v' was first declared",
2475 v->where_decl, NULL, 0, NULL);
2482 Args -> ${ $0 = NULL; }$
2483 | Varlist ${ $0 = $<1; }$
2484 | Varlist ; ${ $0 = $<1; }$
2485 | Varlist NEWLINE ${ $0 = $<1; }$
2487 Varlist -> Varlist ; ArgDecl ${ // UNTESTED
2501 ArgDecl -> IDENTIFIER : FormalType ${ {
2502 struct variable *v = var_decl(c, $1.txt);
2508 ## Executables: the elements of code
2510 Each code element needs to be parsed, printed, analysed,
2511 interpreted, and freed. There are several, so let's just start with
2512 the easy ones and work our way up.
2516 We have already met values as separate objects. When manifest
2517 constants appear in the program text, that must result in an executable
2518 which has a constant value. So the `val` structure embeds a value in
2531 ###### ast functions
2532 struct val *new_val(struct type *T, struct token tk)
2534 struct val *v = new_pos(val, tk);
2545 $0 = new_val(Tbool, $1);
2549 $0 = new_val(Tbool, $1);
2553 $0 = new_val(Tnum, $1);
2556 if (number_parse($0->val.num, tail, $1.txt) == 0)
2557 mpq_init($0->val.num); // UNTESTED
2559 tok_err(c, "error: unsupported number suffix",
2564 $0 = new_val(Tstr, $1);
2567 string_parse(&$1, '\\', &$0->val.str, tail);
2569 tok_err(c, "error: unsupported string suffix",
2574 $0 = new_val(Tstr, $1);
2577 string_parse(&$1, '\\', &$0->val.str, tail);
2579 tok_err(c, "error: unsupported string suffix",
2584 ###### print exec cases
2587 struct val *v = cast(val, e);
2588 if (v->vtype == Tstr)
2590 print_value(v->vtype, &v->val);
2591 if (v->vtype == Tstr)
2596 ###### propagate exec cases
2599 struct val *val = cast(val, prog);
2600 if (!type_compat(type, val->vtype, rules))
2601 type_err(c, "error: expected %1%r found %2",
2602 prog, type, rules, val->vtype);
2606 ###### interp exec cases
2608 rvtype = cast(val, e)->vtype;
2609 dup_value(rvtype, &cast(val, e)->val, &rv);
2612 ###### ast functions
2613 static void free_val(struct val *v)
2616 free_value(v->vtype, &v->val);
2620 ###### free exec cases
2621 case Xval: free_val(cast(val, e)); break;
2623 ###### ast functions
2624 // Move all nodes from 'b' to 'rv', reversing their order.
2625 // In 'b' 'left' is a list, and 'right' is the last node.
2626 // In 'rv', left' is the first node and 'right' is a list.
2627 static struct binode *reorder_bilist(struct binode *b)
2629 struct binode *rv = NULL;
2632 struct exec *t = b->right;
2636 b = cast(binode, b->left);
2646 Just as we used a `val` to wrap a value into an `exec`, we similarly
2647 need a `var` to wrap a `variable` into an exec. While each `val`
2648 contained a copy of the value, each `var` holds a link to the variable
2649 because it really is the same variable no matter where it appears.
2650 When a variable is used, we need to remember to follow the `->merged`
2651 link to find the primary instance.
2659 struct variable *var;
2667 VariableDecl -> IDENTIFIER : ${ {
2668 struct variable *v = var_decl(c, $1.txt);
2669 $0 = new_pos(var, $1);
2674 v = var_ref(c, $1.txt);
2676 type_err(c, "error: variable '%v' redeclared",
2678 type_err(c, "info: this is where '%v' was first declared",
2679 v->where_decl, NULL, 0, NULL);
2682 | IDENTIFIER :: ${ {
2683 struct variable *v = var_decl(c, $1.txt);
2684 $0 = new_pos(var, $1);
2690 v = var_ref(c, $1.txt);
2692 type_err(c, "error: variable '%v' redeclared",
2694 type_err(c, "info: this is where '%v' was first declared",
2695 v->where_decl, NULL, 0, NULL);
2698 | IDENTIFIER : Type ${ {
2699 struct variable *v = var_decl(c, $1.txt);
2700 $0 = new_pos(var, $1);
2707 v = var_ref(c, $1.txt);
2709 type_err(c, "error: variable '%v' redeclared",
2711 type_err(c, "info: this is where '%v' was first declared",
2712 v->where_decl, NULL, 0, NULL);
2715 | IDENTIFIER :: Type ${ {
2716 struct variable *v = var_decl(c, $1.txt);
2717 $0 = new_pos(var, $1);
2725 v = var_ref(c, $1.txt);
2727 type_err(c, "error: variable '%v' redeclared",
2729 type_err(c, "info: this is where '%v' was first declared",
2730 v->where_decl, NULL, 0, NULL);
2735 Variable -> IDENTIFIER ${ {
2736 struct variable *v = var_ref(c, $1.txt);
2737 $0 = new_pos(var, $1);
2739 /* This might be a label - allocate a var just in case */
2740 v = var_decl(c, $1.txt);
2747 cast(var, $0)->var = v;
2751 ###### print exec cases
2754 struct var *v = cast(var, e);
2756 struct binding *b = v->var->name;
2757 printf("%.*s", b->name.len, b->name.txt);
2764 if (loc && loc->type == Xvar) {
2765 struct var *v = cast(var, loc);
2767 struct binding *b = v->var->name;
2768 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2770 fputs("???", stderr); // NOTEST
2772 fputs("NOTVAR", stderr);
2775 ###### propagate exec cases
2779 struct var *var = cast(var, prog);
2780 struct variable *v = var->var;
2782 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2783 return Tnone; // NOTEST
2786 if (v->constant && (rules & Rnoconstant)) {
2787 type_err(c, "error: Cannot assign to a constant: %v",
2788 prog, NULL, 0, NULL);
2789 type_err(c, "info: name was defined as a constant here",
2790 v->where_decl, NULL, 0, NULL);
2793 if (v->type == Tnone && v->where_decl == prog)
2794 type_err(c, "error: variable used but not declared: %v",
2795 prog, NULL, 0, NULL);
2796 if (v->type == NULL) {
2797 if (type && *ok != 0) {
2799 v->where_set = prog;
2804 if (!type_compat(type, v->type, rules)) {
2805 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2806 type, rules, v->type);
2807 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2808 v->type, rules, NULL);
2815 ###### interp exec cases
2818 struct var *var = cast(var, e);
2819 struct variable *v = var->var;
2822 lrv = var_value(c, v);
2827 ###### ast functions
2829 static void free_var(struct var *v)
2834 ###### free exec cases
2835 case Xvar: free_var(cast(var, e)); break;
2837 ### Expressions: Conditional
2839 Our first user of the `binode` will be conditional expressions, which
2840 is a bit odd as they actually have three components. That will be
2841 handled by having 2 binodes for each expression. The conditional
2842 expression is the lowest precedence operator which is why we define it
2843 first - to start the precedence list.
2845 Conditional expressions are of the form "value `if` condition `else`
2846 other_value". They associate to the right, so everything to the right
2847 of `else` is part of an else value, while only a higher-precedence to
2848 the left of `if` is the if values. Between `if` and `else` there is no
2849 room for ambiguity, so a full conditional expression is allowed in
2861 Expression -> Expression if Expression else Expression $$ifelse ${ {
2862 struct binode *b1 = new(binode);
2863 struct binode *b2 = new(binode);
2872 ## expression grammar
2874 ###### print binode cases
2877 b2 = cast(binode, b->right);
2878 if (bracket) printf("(");
2879 print_exec(b2->left, -1, bracket);
2881 print_exec(b->left, -1, bracket);
2883 print_exec(b2->right, -1, bracket);
2884 if (bracket) printf(")");
2887 ###### propagate binode cases
2890 /* cond must be Tbool, others must match */
2891 struct binode *b2 = cast(binode, b->right);
2894 propagate_types(b->left, c, ok, Tbool, 0);
2895 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2896 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2900 ###### interp binode cases
2903 struct binode *b2 = cast(binode, b->right);
2904 left = interp_exec(c, b->left, <ype);
2906 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
2908 rv = interp_exec(c, b2->right, &rvtype);
2914 We take a brief detour, now that we have expressions, to describe lists
2915 of expressions. These will be needed for function parameters and
2916 possibly other situations. They seem generic enough to introduce here
2917 to be used elsewhere.
2919 And ExpressionList will use the `List` type of `binode`, building up at
2920 the end. And place where they are used will probably call
2921 `reorder_bilist()` to get a more normal first/next arrangement.
2923 ###### declare terminals
2926 `List` execs have no implicit semantics, so they are never propagated or
2927 interpreted. The can be printed as a comma separate list, which is how
2928 they are parsed. Note they are also used for function formal parameter
2929 lists. In that case a separate function is used to print them.
2931 ###### print binode cases
2935 print_exec(b->left, -1, bracket);
2938 b = cast(binode, b->right);
2942 ###### propagate binode cases
2943 case List: abort(); // NOTEST
2944 ###### interp binode cases
2945 case List: abort(); // NOTEST
2950 ExpressionList -> ExpressionList , Expression ${
2963 ### Expressions: Boolean
2965 The next class of expressions to use the `binode` will be Boolean
2966 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
2967 have same corresponding precendence. The difference is that they don't
2968 evaluate the second expression if not necessary.
2977 ###### expr precedence
2982 ###### expression grammar
2983 | Expression or Expression ${ {
2984 struct binode *b = new(binode);
2990 | Expression or else Expression ${ {
2991 struct binode *b = new(binode);
2998 | Expression and Expression ${ {
2999 struct binode *b = new(binode);
3005 | Expression and then Expression ${ {
3006 struct binode *b = new(binode);
3013 | not Expression ${ {
3014 struct binode *b = new(binode);
3020 ###### print binode cases
3022 if (bracket) printf("(");
3023 print_exec(b->left, -1, bracket);
3025 print_exec(b->right, -1, bracket);
3026 if (bracket) printf(")");
3029 if (bracket) printf("(");
3030 print_exec(b->left, -1, bracket);
3031 printf(" and then ");
3032 print_exec(b->right, -1, bracket);
3033 if (bracket) printf(")");
3036 if (bracket) printf("(");
3037 print_exec(b->left, -1, bracket);
3039 print_exec(b->right, -1, bracket);
3040 if (bracket) printf(")");
3043 if (bracket) printf("(");
3044 print_exec(b->left, -1, bracket);
3045 printf(" or else ");
3046 print_exec(b->right, -1, bracket);
3047 if (bracket) printf(")");
3050 if (bracket) printf("(");
3052 print_exec(b->right, -1, bracket);
3053 if (bracket) printf(")");
3056 ###### propagate binode cases
3062 /* both must be Tbool, result is Tbool */
3063 propagate_types(b->left, c, ok, Tbool, 0);
3064 propagate_types(b->right, c, ok, Tbool, 0);
3065 if (type && type != Tbool)
3066 type_err(c, "error: %1 operation found where %2 expected", prog,
3070 ###### interp binode cases
3072 rv = interp_exec(c, b->left, &rvtype);
3073 right = interp_exec(c, b->right, &rtype);
3074 rv.bool = rv.bool && right.bool;
3077 rv = interp_exec(c, b->left, &rvtype);
3079 rv = interp_exec(c, b->right, NULL);
3082 rv = interp_exec(c, b->left, &rvtype);
3083 right = interp_exec(c, b->right, &rtype);
3084 rv.bool = rv.bool || right.bool;
3087 rv = interp_exec(c, b->left, &rvtype);
3089 rv = interp_exec(c, b->right, NULL);
3092 rv = interp_exec(c, b->right, &rvtype);
3096 ### Expressions: Comparison
3098 Of slightly higher precedence that Boolean expressions are Comparisons.
3099 A comparison takes arguments of any comparable type, but the two types
3102 To simplify the parsing we introduce an `eop` which can record an
3103 expression operator, and the `CMPop` non-terminal will match one of them.
3110 ###### ast functions
3111 static void free_eop(struct eop *e)
3125 ###### expr precedence
3126 $LEFT < > <= >= == != CMPop
3128 ###### expression grammar
3129 | Expression CMPop Expression ${ {
3130 struct binode *b = new(binode);
3140 CMPop -> < ${ $0.op = Less; }$
3141 | > ${ $0.op = Gtr; }$
3142 | <= ${ $0.op = LessEq; }$
3143 | >= ${ $0.op = GtrEq; }$
3144 | == ${ $0.op = Eql; }$
3145 | != ${ $0.op = NEql; }$
3147 ###### print binode cases
3155 if (bracket) printf("(");
3156 print_exec(b->left, -1, bracket);
3158 case Less: printf(" < "); break;
3159 case LessEq: printf(" <= "); break;
3160 case Gtr: printf(" > "); break;
3161 case GtrEq: printf(" >= "); break;
3162 case Eql: printf(" == "); break;
3163 case NEql: printf(" != "); break;
3164 default: abort(); // NOTEST
3166 print_exec(b->right, -1, bracket);
3167 if (bracket) printf(")");
3170 ###### propagate binode cases
3177 /* Both must match but not be labels, result is Tbool */
3178 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
3180 propagate_types(b->right, c, ok, t, 0);
3182 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
3184 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
3186 if (!type_compat(type, Tbool, 0))
3187 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3188 Tbool, rules, type);
3191 ###### interp binode cases
3200 left = interp_exec(c, b->left, <ype);
3201 right = interp_exec(c, b->right, &rtype);
3202 cmp = value_cmp(ltype, rtype, &left, &right);
3205 case Less: rv.bool = cmp < 0; break;
3206 case LessEq: rv.bool = cmp <= 0; break;
3207 case Gtr: rv.bool = cmp > 0; break;
3208 case GtrEq: rv.bool = cmp >= 0; break;
3209 case Eql: rv.bool = cmp == 0; break;
3210 case NEql: rv.bool = cmp != 0; break;
3211 default: rv.bool = 0; break; // NOTEST
3216 ### Expressions: Arithmetic etc.
3218 The remaining expressions with the highest precedence are arithmetic,
3219 string concatenation, and string conversion. String concatenation
3220 (`++`) has the same precedence as multiplication and division, but lower
3223 String conversion is a temporary feature until I get a better type
3224 system. `$` is a prefix operator which expects a string and returns
3227 `+` and `-` are both infix and prefix operations (where they are
3228 absolute value and negation). These have different operator names.
3230 We also have a 'Bracket' operator which records where parentheses were
3231 found. This makes it easy to reproduce these when printing. Possibly I
3232 should only insert brackets were needed for precedence.
3242 ###### expr precedence
3248 ###### expression grammar
3249 | Expression Eop Expression ${ {
3250 struct binode *b = new(binode);
3257 | Expression Top Expression ${ {
3258 struct binode *b = new(binode);
3265 | ( Expression ) ${ {
3266 struct binode *b = new_pos(binode, $1);
3271 | Uop Expression ${ {
3272 struct binode *b = new(binode);
3277 | Value ${ $0 = $<1; }$
3278 | Variable ${ $0 = $<1; }$
3283 Eop -> + ${ $0.op = Plus; }$
3284 | - ${ $0.op = Minus; }$
3286 Uop -> + ${ $0.op = Absolute; }$
3287 | - ${ $0.op = Negate; }$
3288 | $ ${ $0.op = StringConv; }$
3290 Top -> * ${ $0.op = Times; }$
3291 | / ${ $0.op = Divide; }$
3292 | % ${ $0.op = Rem; }$
3293 | ++ ${ $0.op = Concat; }$
3295 ###### print binode cases
3302 if (bracket) printf("(");
3303 print_exec(b->left, indent, bracket);
3305 case Plus: fputs(" + ", stdout); break;
3306 case Minus: fputs(" - ", stdout); break;
3307 case Times: fputs(" * ", stdout); break;
3308 case Divide: fputs(" / ", stdout); break;
3309 case Rem: fputs(" % ", stdout); break;
3310 case Concat: fputs(" ++ ", stdout); break;
3311 default: abort(); // NOTEST
3313 print_exec(b->right, indent, bracket);
3314 if (bracket) printf(")");
3319 if (bracket) printf("(");
3321 case Absolute: fputs("+", stdout); break;
3322 case Negate: fputs("-", stdout); break;
3323 case StringConv: fputs("$", stdout); break;
3324 default: abort(); // NOTEST
3326 print_exec(b->right, indent, bracket);
3327 if (bracket) printf(")");
3331 print_exec(b->right, indent, bracket);
3335 ###### propagate binode cases
3341 /* both must be numbers, result is Tnum */
3344 /* as propagate_types ignores a NULL,
3345 * unary ops fit here too */
3346 propagate_types(b->left, c, ok, Tnum, 0);
3347 propagate_types(b->right, c, ok, Tnum, 0);
3348 if (!type_compat(type, Tnum, 0))
3349 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3354 /* both must be Tstr, result is Tstr */
3355 propagate_types(b->left, c, ok, Tstr, 0);
3356 propagate_types(b->right, c, ok, Tstr, 0);
3357 if (!type_compat(type, Tstr, 0))
3358 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3363 /* op must be string, result is number */
3364 propagate_types(b->left, c, ok, Tstr, 0);
3365 if (!type_compat(type, Tnum, 0))
3366 type_err(c, // UNTESTED
3367 "error: Can only convert string to number, not %1",
3368 prog, type, 0, NULL);
3372 return propagate_types(b->right, c, ok, type, 0);
3374 ###### interp binode cases
3377 rv = interp_exec(c, b->left, &rvtype);
3378 right = interp_exec(c, b->right, &rtype);
3379 mpq_add(rv.num, rv.num, right.num);
3382 rv = interp_exec(c, b->left, &rvtype);
3383 right = interp_exec(c, b->right, &rtype);
3384 mpq_sub(rv.num, rv.num, right.num);
3387 rv = interp_exec(c, b->left, &rvtype);
3388 right = interp_exec(c, b->right, &rtype);
3389 mpq_mul(rv.num, rv.num, right.num);
3392 rv = interp_exec(c, b->left, &rvtype);
3393 right = interp_exec(c, b->right, &rtype);
3394 mpq_div(rv.num, rv.num, right.num);
3399 left = interp_exec(c, b->left, <ype);
3400 right = interp_exec(c, b->right, &rtype);
3401 mpz_init(l); mpz_init(r); mpz_init(rem);
3402 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3403 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3404 mpz_tdiv_r(rem, l, r);
3405 val_init(Tnum, &rv);
3406 mpq_set_z(rv.num, rem);
3407 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3412 rv = interp_exec(c, b->right, &rvtype);
3413 mpq_neg(rv.num, rv.num);
3416 rv = interp_exec(c, b->right, &rvtype);
3417 mpq_abs(rv.num, rv.num);
3420 rv = interp_exec(c, b->right, &rvtype);
3423 left = interp_exec(c, b->left, <ype);
3424 right = interp_exec(c, b->right, &rtype);
3426 rv.str = text_join(left.str, right.str);
3429 right = interp_exec(c, b->right, &rvtype);
3433 struct text tx = right.str;
3436 if (tx.txt[0] == '-') {
3437 neg = 1; // UNTESTED
3438 tx.txt++; // UNTESTED
3439 tx.len--; // UNTESTED
3441 if (number_parse(rv.num, tail, tx) == 0)
3442 mpq_init(rv.num); // UNTESTED
3444 mpq_neg(rv.num, rv.num); // UNTESTED
3446 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3450 ###### value functions
3452 static struct text text_join(struct text a, struct text b)
3455 rv.len = a.len + b.len;
3456 rv.txt = malloc(rv.len);
3457 memcpy(rv.txt, a.txt, a.len);
3458 memcpy(rv.txt+a.len, b.txt, b.len);
3464 A function call can appear either as an expression or as a statement.
3465 As functions cannot yet return values, only the statement version will work.
3466 We use a new 'Funcall' binode type to link the function with a list of
3467 arguments, form with the 'List' nodes.
3472 ###### expression grammar
3473 | Variable ( ExpressionList ) ${ {
3474 struct binode *b = new(binode);
3477 b->right = reorder_bilist($<EL);
3481 struct binode *b = new(binode);
3488 ###### SimpleStatement Grammar
3490 | Variable ( ExpressionList ) ${ {
3491 struct binode *b = new(binode);
3494 b->right = reorder_bilist($<EL);
3498 ###### print binode cases
3501 do_indent(indent, "");
3502 print_exec(b->left, -1, bracket);
3504 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3507 print_exec(b->left, -1, bracket);
3517 ###### propagate binode cases
3520 /* Every arg must match formal parameter, and result
3521 * is return type of function (currently Tnone).
3523 struct binode *args = cast(binode, b->right);
3524 struct var *v = cast(var, b->left);
3526 if (!v->var->type || v->var->type->check_args == NULL) {
3527 type_err(c, "error: attempt to call a non-function.",
3528 prog, NULL, 0, NULL);
3531 v->var->type->check_args(c, ok, v->var->type, args);
3535 ###### interp binode cases
3538 struct var *v = cast(var, b->left);
3539 struct type *t = v->var->type;
3540 void *oldlocal = c->local;
3541 int old_size = c->local_size;
3542 void *local = calloc(1, t->function.local_size);
3543 struct value *fbody = var_value(c, v->var);
3544 struct binode *arg = cast(binode, b->right);
3545 struct binode *param = t->function.params;
3548 struct var *pv = cast(var, param->left);
3549 struct type *vtype = NULL;
3550 struct value val = interp_exec(c, arg->left, &vtype);
3552 c->local = local; c->local_size = t->function.local_size;
3553 lval = var_value(c, pv->var);
3554 c->local = oldlocal; c->local_size = old_size;
3555 memcpy(lval, &val, vtype->size);
3556 param = cast(binode, param->right);
3557 arg = cast(binode, arg->right);
3559 c->local = local; c->local_size = t->function.local_size;
3560 right = interp_exec(c, fbody->function, &rtype);
3561 c->local = oldlocal; c->local_size = old_size;
3566 ### Blocks, Statements, and Statement lists.
3568 Now that we have expressions out of the way we need to turn to
3569 statements. There are simple statements and more complex statements.
3570 Simple statements do not contain (syntactic) newlines, complex statements do.
3572 Statements often come in sequences and we have corresponding simple
3573 statement lists and complex statement lists.
3574 The former comprise only simple statements separated by semicolons.
3575 The later comprise complex statements and simple statement lists. They are
3576 separated by newlines. Thus the semicolon is only used to separate
3577 simple statements on the one line. This may be overly restrictive,
3578 but I'm not sure I ever want a complex statement to share a line with
3581 Note that a simple statement list can still use multiple lines if
3582 subsequent lines are indented, so
3584 ###### Example: wrapped simple statement list
3589 is a single simple statement list. This might allow room for
3590 confusion, so I'm not set on it yet.
3592 A simple statement list needs no extra syntax. A complex statement
3593 list has two syntactic forms. It can be enclosed in braces (much like
3594 C blocks), or it can be introduced by an indent and continue until an
3595 unindented newline (much like Python blocks). With this extra syntax
3596 it is referred to as a block.
3598 Note that a block does not have to include any newlines if it only
3599 contains simple statements. So both of:
3601 if condition: a=b; d=f
3603 if condition { a=b; print f }
3607 In either case the list is constructed from a `binode` list with
3608 `Block` as the operator. When parsing the list it is most convenient
3609 to append to the end, so a list is a list and a statement. When using
3610 the list it is more convenient to consider a list to be a statement
3611 and a list. So we need a function to re-order a list.
3612 `reorder_bilist` serves this purpose.
3614 The only stand-alone statement we introduce at this stage is `pass`
3615 which does nothing and is represented as a `NULL` pointer in a `Block`
3616 list. Other stand-alone statements will follow once the infrastructure
3627 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3628 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3629 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3630 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3631 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3633 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3634 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3635 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3636 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3637 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3639 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3640 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3641 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3643 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3644 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3645 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3646 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3647 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3649 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3651 ComplexStatements -> ComplexStatements ComplexStatement ${
3661 | ComplexStatement ${
3673 ComplexStatement -> SimpleStatements Newlines ${
3674 $0 = reorder_bilist($<SS);
3676 | SimpleStatements ; Newlines ${
3677 $0 = reorder_bilist($<SS);
3679 ## ComplexStatement Grammar
3682 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3688 | SimpleStatement ${
3696 SimpleStatement -> pass ${ $0 = NULL; }$
3697 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3698 ## SimpleStatement Grammar
3700 ###### print binode cases
3704 if (b->left == NULL) // UNTESTED
3705 printf("pass"); // UNTESTED
3707 print_exec(b->left, indent, bracket); // UNTESTED
3708 if (b->right) { // UNTESTED
3709 printf("; "); // UNTESTED
3710 print_exec(b->right, indent, bracket); // UNTESTED
3713 // block, one per line
3714 if (b->left == NULL)
3715 do_indent(indent, "pass\n");
3717 print_exec(b->left, indent, bracket);
3719 print_exec(b->right, indent, bracket);
3723 ###### propagate binode cases
3726 /* If any statement returns something other than Tnone
3727 * or Tbool then all such must return same type.
3728 * As each statement may be Tnone or something else,
3729 * we must always pass NULL (unknown) down, otherwise an incorrect
3730 * error might occur. We never return Tnone unless it is
3735 for (e = b; e; e = cast(binode, e->right)) {
3736 t = propagate_types(e->left, c, ok, NULL, rules);
3737 if ((rules & Rboolok) && t == Tbool)
3739 if (t && t != Tnone && t != Tbool) {
3743 type_err(c, "error: expected %1%r, found %2",
3744 e->left, type, rules, t);
3750 ###### interp binode cases
3752 while (rvtype == Tnone &&
3755 rv = interp_exec(c, b->left, &rvtype);
3756 b = cast(binode, b->right);
3760 ### The Print statement
3762 `print` is a simple statement that takes a comma-separated list of
3763 expressions and prints the values separated by spaces and terminated
3764 by a newline. No control of formatting is possible.
3766 `print` uses `ExpressionList` to collect the expressions and stores them
3767 on the left side of a `Print` binode unlessthere is a trailing comma
3768 when the list is stored on the `right` side and no trailing newline is
3774 ##### expr precedence
3777 ###### SimpleStatement Grammar
3779 | print ExpressionList ${
3783 $0->left = reorder_bilist($<EL);
3785 | print ExpressionList , ${ {
3788 $0->right = reorder_bilist($<EL);
3798 ###### print binode cases
3801 do_indent(indent, "print");
3803 print_exec(b->right, -1, bracket);
3806 print_exec(b->left, -1, bracket);
3811 ###### propagate binode cases
3814 /* don't care but all must be consistent */
3816 b = cast(binode, b->left);
3818 b = cast(binode, b->right);
3820 propagate_types(b->left, c, ok, NULL, Rnolabel);
3821 b = cast(binode, b->right);
3825 ###### interp binode cases
3829 struct binode *b2 = cast(binode, b->left);
3831 b2 = cast(binode, b->right);
3832 for (; b2; b2 = cast(binode, b2->right)) {
3833 left = interp_exec(c, b2->left, <ype);
3834 print_value(ltype, &left);
3835 free_value(ltype, &left);
3839 if (b->right == NULL)
3845 ###### Assignment statement
3847 An assignment will assign a value to a variable, providing it hasn't
3848 been declared as a constant. The analysis phase ensures that the type
3849 will be correct so the interpreter just needs to perform the
3850 calculation. There is a form of assignment which declares a new
3851 variable as well as assigning a value. If a name is assigned before
3852 it is declared, and error will be raised as the name is created as
3853 `Tlabel` and it is illegal to assign to such names.
3859 ###### declare terminals
3862 ###### SimpleStatement Grammar
3863 | Variable = Expression ${
3869 | VariableDecl = Expression ${
3877 if ($1->var->where_set == NULL) {
3879 "Variable declared with no type or value: %v",
3889 ###### print binode cases
3892 do_indent(indent, "");
3893 print_exec(b->left, indent, bracket);
3895 print_exec(b->right, indent, bracket);
3902 struct variable *v = cast(var, b->left)->var;
3903 do_indent(indent, "");
3904 print_exec(b->left, indent, bracket);
3905 if (cast(var, b->left)->var->constant) {
3907 if (v->where_decl == v->where_set) {
3908 type_print(v->type, stdout);
3913 if (v->where_decl == v->where_set) {
3914 type_print(v->type, stdout);
3920 print_exec(b->right, indent, bracket);
3927 ###### propagate binode cases
3931 /* Both must match and not be labels,
3932 * Type must support 'dup',
3933 * For Assign, left must not be constant.
3936 t = propagate_types(b->left, c, ok, NULL,
3937 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3942 if (propagate_types(b->right, c, ok, t, 0) != t)
3943 if (b->left->type == Xvar)
3944 type_err(c, "info: variable '%v' was set as %1 here.",
3945 cast(var, b->left)->var->where_set, t, rules, NULL);
3947 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3949 propagate_types(b->left, c, ok, t,
3950 (b->op == Assign ? Rnoconstant : 0));
3952 if (t && t->dup == NULL)
3953 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3958 ###### interp binode cases
3961 lleft = linterp_exec(c, b->left, <ype);
3962 right = interp_exec(c, b->right, &rtype);
3964 free_value(ltype, lleft);
3965 dup_value(ltype, &right, lleft);
3972 struct variable *v = cast(var, b->left)->var;
3975 val = var_value(c, v);
3976 if (v->type->prepare_type)
3977 v->type->prepare_type(c, v->type, 0);
3979 right = interp_exec(c, b->right, &rtype);
3980 memcpy(val, &right, rtype->size);
3983 val_init(v->type, val);
3988 ### The `use` statement
3990 The `use` statement is the last "simple" statement. It is needed when
3991 the condition in a conditional statement is a block. `use` works much
3992 like `return` in C, but only completes the `condition`, not the whole
3998 ###### expr precedence
4001 ###### SimpleStatement Grammar
4003 $0 = new_pos(binode, $1);
4006 if ($0->right->type == Xvar) {
4007 struct var *v = cast(var, $0->right);
4008 if (v->var->type == Tnone) {
4009 /* Convert this to a label */
4012 v->var->type = Tlabel;
4013 val = global_alloc(c, Tlabel, v->var, NULL);
4019 ###### print binode cases
4022 do_indent(indent, "use ");
4023 print_exec(b->right, -1, bracket);
4028 ###### propagate binode cases
4031 /* result matches value */
4032 return propagate_types(b->right, c, ok, type, 0);
4034 ###### interp binode cases
4037 rv = interp_exec(c, b->right, &rvtype);
4040 ### The Conditional Statement
4042 This is the biggy and currently the only complex statement. This
4043 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4044 It is comprised of a number of parts, all of which are optional though
4045 set combinations apply. Each part is (usually) a key word (`then` is
4046 sometimes optional) followed by either an expression or a code block,
4047 except the `casepart` which is a "key word and an expression" followed
4048 by a code block. The code-block option is valid for all parts and,
4049 where an expression is also allowed, the code block can use the `use`
4050 statement to report a value. If the code block does not report a value
4051 the effect is similar to reporting `True`.
4053 The `else` and `case` parts, as well as `then` when combined with
4054 `if`, can contain a `use` statement which will apply to some
4055 containing conditional statement. `for` parts, `do` parts and `then`
4056 parts used with `for` can never contain a `use`, except in some
4057 subordinate conditional statement.
4059 If there is a `forpart`, it is executed first, only once.
4060 If there is a `dopart`, then it is executed repeatedly providing
4061 always that the `condpart` or `cond`, if present, does not return a non-True
4062 value. `condpart` can fail to return any value if it simply executes
4063 to completion. This is treated the same as returning `True`.
4065 If there is a `thenpart` it will be executed whenever the `condpart`
4066 or `cond` returns True (or does not return any value), but this will happen
4067 *after* `dopart` (when present).
4069 If `elsepart` is present it will be executed at most once when the
4070 condition returns `False` or some value that isn't `True` and isn't
4071 matched by any `casepart`. If there are any `casepart`s, they will be
4072 executed when the condition returns a matching value.
4074 The particular sorts of values allowed in case parts has not yet been
4075 determined in the language design, so nothing is prohibited.
4077 The various blocks in this complex statement potentially provide scope
4078 for variables as described earlier. Each such block must include the
4079 "OpenScope" nonterminal before parsing the block, and must call
4080 `var_block_close()` when closing the block.
4082 The code following "`if`", "`switch`" and "`for`" does not get its own
4083 scope, but is in a scope covering the whole statement, so names
4084 declared there cannot be redeclared elsewhere. Similarly the
4085 condition following "`while`" is in a scope the covers the body
4086 ("`do`" part) of the loop, and which does not allow conditional scope
4087 extension. Code following "`then`" (both looping and non-looping),
4088 "`else`" and "`case`" each get their own local scope.
4090 The type requirements on the code block in a `whilepart` are quite
4091 unusal. It is allowed to return a value of some identifiable type, in
4092 which case the loop aborts and an appropriate `casepart` is run, or it
4093 can return a Boolean, in which case the loop either continues to the
4094 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4095 This is different both from the `ifpart` code block which is expected to
4096 return a Boolean, or the `switchpart` code block which is expected to
4097 return the same type as the casepart values. The correct analysis of
4098 the type of the `whilepart` code block is the reason for the
4099 `Rboolok` flag which is passed to `propagate_types()`.
4101 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4102 defined. As there are two scopes which cover multiple parts - one for
4103 the whole statement and one for "while" and "do" - and as we will use
4104 the 'struct exec' to track scopes, we actually need two new types of
4105 exec. One is a `binode` for the looping part, the rest is the
4106 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4107 casepart` to track a list of case parts.
4118 struct exec *action;
4119 struct casepart *next;
4121 struct cond_statement {
4123 struct exec *forpart, *condpart, *thenpart, *elsepart;
4124 struct binode *looppart;
4125 struct casepart *casepart;
4128 ###### ast functions
4130 static void free_casepart(struct casepart *cp)
4134 free_exec(cp->value);
4135 free_exec(cp->action);
4142 static void free_cond_statement(struct cond_statement *s)
4146 free_exec(s->forpart);
4147 free_exec(s->condpart);
4148 free_exec(s->looppart);
4149 free_exec(s->thenpart);
4150 free_exec(s->elsepart);
4151 free_casepart(s->casepart);
4155 ###### free exec cases
4156 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4158 ###### ComplexStatement Grammar
4159 | CondStatement ${ $0 = $<1; }$
4161 ###### expr precedence
4162 $TERM for then while do
4169 // A CondStatement must end with EOL, as does CondSuffix and
4171 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4172 // may or may not end with EOL
4173 // WhilePart and IfPart include an appropriate Suffix
4175 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4176 // them. WhilePart opens and closes its own scope.
4177 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4180 $0->thenpart = $<TP;
4181 $0->looppart = $<WP;
4182 var_block_close(c, CloseSequential, $0);
4184 | ForPart OptNL WhilePart CondSuffix ${
4187 $0->looppart = $<WP;
4188 var_block_close(c, CloseSequential, $0);
4190 | WhilePart CondSuffix ${
4192 $0->looppart = $<WP;
4194 | SwitchPart OptNL CasePart CondSuffix ${
4196 $0->condpart = $<SP;
4197 $CP->next = $0->casepart;
4198 $0->casepart = $<CP;
4199 var_block_close(c, CloseSequential, $0);
4201 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4203 $0->condpart = $<SP;
4204 $CP->next = $0->casepart;
4205 $0->casepart = $<CP;
4206 var_block_close(c, CloseSequential, $0);
4208 | IfPart IfSuffix ${
4210 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4211 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4212 // This is where we close an "if" statement
4213 var_block_close(c, CloseSequential, $0);
4216 CondSuffix -> IfSuffix ${
4219 | Newlines CasePart CondSuffix ${
4221 $CP->next = $0->casepart;
4222 $0->casepart = $<CP;
4224 | CasePart CondSuffix ${
4226 $CP->next = $0->casepart;
4227 $0->casepart = $<CP;
4230 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4231 | Newlines ElsePart ${ $0 = $<EP; }$
4232 | ElsePart ${$0 = $<EP; }$
4234 ElsePart -> else OpenBlock Newlines ${
4235 $0 = new(cond_statement);
4236 $0->elsepart = $<OB;
4237 var_block_close(c, CloseElse, $0->elsepart);
4239 | else OpenScope CondStatement ${
4240 $0 = new(cond_statement);
4241 $0->elsepart = $<CS;
4242 var_block_close(c, CloseElse, $0->elsepart);
4246 CasePart -> case Expression OpenScope ColonBlock ${
4247 $0 = calloc(1,sizeof(struct casepart));
4250 var_block_close(c, CloseParallel, $0->action);
4254 // These scopes are closed in CondStatement
4255 ForPart -> for OpenBlock ${
4259 ThenPart -> then OpenBlock ${
4261 var_block_close(c, CloseSequential, $0);
4265 // This scope is closed in CondStatement
4266 WhilePart -> while UseBlock OptNL do OpenBlock ${
4271 var_block_close(c, CloseSequential, $0->right);
4272 var_block_close(c, CloseSequential, $0);
4274 | while OpenScope Expression OpenScope ColonBlock ${
4279 var_block_close(c, CloseSequential, $0->right);
4280 var_block_close(c, CloseSequential, $0);
4284 IfPart -> if UseBlock OptNL then OpenBlock ${
4287 var_block_close(c, CloseParallel, $0.thenpart);
4289 | if OpenScope Expression OpenScope ColonBlock ${
4292 var_block_close(c, CloseParallel, $0.thenpart);
4294 | if OpenScope Expression OpenScope OptNL then Block ${
4297 var_block_close(c, CloseParallel, $0.thenpart);
4301 // This scope is closed in CondStatement
4302 SwitchPart -> switch OpenScope Expression ${
4305 | switch UseBlock ${
4309 ###### print binode cases
4311 if (b->left && b->left->type == Xbinode &&
4312 cast(binode, b->left)->op == Block) {
4314 do_indent(indent, "while {\n");
4316 do_indent(indent, "while\n");
4317 print_exec(b->left, indent+1, bracket);
4319 do_indent(indent, "} do {\n");
4321 do_indent(indent, "do\n");
4322 print_exec(b->right, indent+1, bracket);
4324 do_indent(indent, "}\n");
4326 do_indent(indent, "while ");
4327 print_exec(b->left, 0, bracket);
4332 print_exec(b->right, indent+1, bracket);
4334 do_indent(indent, "}\n");
4338 ###### print exec cases
4340 case Xcond_statement:
4342 struct cond_statement *cs = cast(cond_statement, e);
4343 struct casepart *cp;
4345 do_indent(indent, "for");
4346 if (bracket) printf(" {\n"); else printf("\n");
4347 print_exec(cs->forpart, indent+1, bracket);
4350 do_indent(indent, "} then {\n");
4352 do_indent(indent, "then\n");
4353 print_exec(cs->thenpart, indent+1, bracket);
4355 if (bracket) do_indent(indent, "}\n");
4358 print_exec(cs->looppart, indent, bracket);
4362 do_indent(indent, "switch");
4364 do_indent(indent, "if");
4365 if (cs->condpart && cs->condpart->type == Xbinode &&
4366 cast(binode, cs->condpart)->op == Block) {
4371 print_exec(cs->condpart, indent+1, bracket);
4373 do_indent(indent, "}\n");
4375 do_indent(indent, "then\n");
4376 print_exec(cs->thenpart, indent+1, bracket);
4380 print_exec(cs->condpart, 0, bracket);
4386 print_exec(cs->thenpart, indent+1, bracket);
4388 do_indent(indent, "}\n");
4393 for (cp = cs->casepart; cp; cp = cp->next) {
4394 do_indent(indent, "case ");
4395 print_exec(cp->value, -1, 0);
4400 print_exec(cp->action, indent+1, bracket);
4402 do_indent(indent, "}\n");
4405 do_indent(indent, "else");
4410 print_exec(cs->elsepart, indent+1, bracket);
4412 do_indent(indent, "}\n");
4417 ###### propagate binode cases
4419 t = propagate_types(b->right, c, ok, Tnone, 0);
4420 if (!type_compat(Tnone, t, 0))
4421 *ok = 0; // UNTESTED
4422 return propagate_types(b->left, c, ok, type, rules);
4424 ###### propagate exec cases
4425 case Xcond_statement:
4427 // forpart and looppart->right must return Tnone
4428 // thenpart must return Tnone if there is a loopart,
4429 // otherwise it is like elsepart.
4431 // be bool if there is no casepart
4432 // match casepart->values if there is a switchpart
4433 // either be bool or match casepart->value if there
4435 // elsepart and casepart->action must match the return type
4436 // expected of this statement.
4437 struct cond_statement *cs = cast(cond_statement, prog);
4438 struct casepart *cp;
4440 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4441 if (!type_compat(Tnone, t, 0))
4442 *ok = 0; // UNTESTED
4445 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4446 if (!type_compat(Tnone, t, 0))
4447 *ok = 0; // UNTESTED
4449 if (cs->casepart == NULL) {
4450 propagate_types(cs->condpart, c, ok, Tbool, 0);
4451 propagate_types(cs->looppart, c, ok, Tbool, 0);
4453 /* Condpart must match case values, with bool permitted */
4455 for (cp = cs->casepart;
4456 cp && !t; cp = cp->next)
4457 t = propagate_types(cp->value, c, ok, NULL, 0);
4458 if (!t && cs->condpart)
4459 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4460 if (!t && cs->looppart)
4461 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4462 // Now we have a type (I hope) push it down
4464 for (cp = cs->casepart; cp; cp = cp->next)
4465 propagate_types(cp->value, c, ok, t, 0);
4466 propagate_types(cs->condpart, c, ok, t, Rboolok);
4467 propagate_types(cs->looppart, c, ok, t, Rboolok);
4470 // (if)then, else, and case parts must return expected type.
4471 if (!cs->looppart && !type)
4472 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4474 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4475 for (cp = cs->casepart;
4477 cp = cp->next) // UNTESTED
4478 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4481 propagate_types(cs->thenpart, c, ok, type, rules);
4482 propagate_types(cs->elsepart, c, ok, type, rules);
4483 for (cp = cs->casepart; cp ; cp = cp->next)
4484 propagate_types(cp->action, c, ok, type, rules);
4490 ###### interp binode cases
4492 // This just performs one iterration of the loop
4493 rv = interp_exec(c, b->left, &rvtype);
4494 if (rvtype == Tnone ||
4495 (rvtype == Tbool && rv.bool != 0))
4496 // cnd is Tnone or Tbool, doesn't need to be freed
4497 interp_exec(c, b->right, NULL);
4500 ###### interp exec cases
4501 case Xcond_statement:
4503 struct value v, cnd;
4504 struct type *vtype, *cndtype;
4505 struct casepart *cp;
4506 struct cond_statement *cs = cast(cond_statement, e);
4509 interp_exec(c, cs->forpart, NULL);
4511 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4512 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4513 interp_exec(c, cs->thenpart, NULL);
4515 cnd = interp_exec(c, cs->condpart, &cndtype);
4516 if ((cndtype == Tnone ||
4517 (cndtype == Tbool && cnd.bool != 0))) {
4518 // cnd is Tnone or Tbool, doesn't need to be freed
4519 rv = interp_exec(c, cs->thenpart, &rvtype);
4520 // skip else (and cases)
4524 for (cp = cs->casepart; cp; cp = cp->next) {
4525 v = interp_exec(c, cp->value, &vtype);
4526 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4527 free_value(vtype, &v);
4528 free_value(cndtype, &cnd);
4529 rv = interp_exec(c, cp->action, &rvtype);
4532 free_value(vtype, &v);
4534 free_value(cndtype, &cnd);
4536 rv = interp_exec(c, cs->elsepart, &rvtype);
4543 ### Top level structure
4545 All the language elements so far can be used in various places. Now
4546 it is time to clarify what those places are.
4548 At the top level of a file there will be a number of declarations.
4549 Many of the things that can be declared haven't been described yet,
4550 such as functions, procedures, imports, and probably more.
4551 For now there are two sorts of things that can appear at the top
4552 level. They are predefined constants, `struct` types, and the `main`
4553 function. While the syntax will allow the `main` function to appear
4554 multiple times, that will trigger an error if it is actually attempted.
4556 The various declarations do not return anything. They store the
4557 various declarations in the parse context.
4559 ###### Parser: grammar
4562 Ocean -> OptNL DeclarationList
4564 ## declare terminals
4571 DeclarationList -> Declaration
4572 | DeclarationList Declaration
4574 Declaration -> ERROR Newlines ${
4575 tok_err(c, // UNTESTED
4576 "error: unhandled parse error", &$1);
4582 ## top level grammar
4586 ### The `const` section
4588 As well as being defined in with the code that uses them, constants
4589 can be declared at the top level. These have full-file scope, so they
4590 are always `InScope`. The value of a top level constant can be given
4591 as an expression, and this is evaluated immediately rather than in the
4592 later interpretation stage. Once we add functions to the language, we
4593 will need rules concern which, if any, can be used to define a top
4596 Constants are defined in a section that starts with the reserved word
4597 `const` and then has a block with a list of assignment statements.
4598 For syntactic consistency, these must use the double-colon syntax to
4599 make it clear that they are constants. Type can also be given: if
4600 not, the type will be determined during analysis, as with other
4603 As the types constants are inserted at the head of a list, printing
4604 them in the same order that they were read is not straight forward.
4605 We take a quadratic approach here and count the number of constants
4606 (variables of depth 0), then count down from there, each time
4607 searching through for the Nth constant for decreasing N.
4609 ###### top level grammar
4613 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4614 | const { SimpleConstList } Newlines
4615 | const IN OptNL ConstList OUT Newlines
4616 | const SimpleConstList Newlines
4618 ConstList -> ConstList SimpleConstLine
4620 SimpleConstList -> SimpleConstList ; Const
4623 SimpleConstLine -> SimpleConstList Newlines
4624 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4627 CType -> Type ${ $0 = $<1; }$
4630 Const -> IDENTIFIER :: CType = Expression ${ {
4634 v = var_decl(c, $1.txt);
4636 struct var *var = new_pos(var, $1);
4637 v->where_decl = var;
4642 v = var_ref(c, $1.txt);
4643 tok_err(c, "error: name already declared", &$1);
4644 type_err(c, "info: this is where '%v' was first declared",
4645 v->where_decl, NULL, 0, NULL);
4649 propagate_types($5, c, &ok, $3, 0);
4654 struct value res = interp_exec(c, $5, &v->type);
4655 global_alloc(c, v->type, v, &res);
4659 ###### print const decls
4664 while (target != 0) {
4666 for (v = context.in_scope; v; v=v->in_scope)
4667 if (v->depth == 0 && v->constant) {
4678 struct value *val = var_value(&context, v);
4679 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4680 type_print(v->type, stdout);
4682 if (v->type == Tstr)
4684 print_value(v->type, val);
4685 if (v->type == Tstr)
4693 ### Function declarations
4695 The code in an Ocean program is all stored in function declarations.
4696 One of the functions must be named `main` and it must accept an array of
4697 strings as a parameter - the command line arguments.
4700 As this is the top level, several things are handled a bit
4702 The function is not interpreted by `interp_exec` as that isn't
4703 passed the argument list which the program requires. Similarly type
4704 analysis is a bit more interesting at this level.
4706 ###### top level grammar
4709 DeclareFunction -> func FuncName ( OpenScope Args ) Block Newlines ${ {
4710 struct text funcname = { " func", 5};
4713 struct value fn = {.function = $<Bl};
4714 $0->type = add_type(c, funcname, &function_prototype);
4715 $0->type->function.params = reorder_bilist($<Ar);
4716 global_alloc(c, $0->type, $0, &fn);
4717 var_block_close(c, CloseSequential, fn.function);
4719 var_block_close(c, CloseSequential, NULL);
4721 | func FuncName then IN OpenScope OptNL Args OUT OptNL do Block Newlines ${ {
4722 // FIXME that 'then' should not be there.
4723 struct text funcname = { " func", 5};
4726 struct value fn = {.function = $<Bl};
4727 $0->type = add_type(c, funcname, &function_prototype);
4728 $0->type->function.params = reorder_bilist($<Ar);
4729 global_alloc(c, $0->type, $0, &fn);
4730 var_block_close(c, CloseSequential, fn.function);
4732 var_block_close(c, CloseSequential, NULL);
4734 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${ {
4735 struct text funcname = { " func", 5};
4738 struct value fn = {.function = $<Bl};
4739 $0->type = add_type(c, funcname, &function_prototype);
4740 $0->type->function.params = NULL;
4741 global_alloc(c, $0->type, $0, &fn);
4742 var_block_close(c, CloseSequential, fn.function);
4744 var_block_close(c, CloseSequential, NULL);
4747 ###### print func decls
4752 while (target != 0) {
4754 for (v = context.in_scope; v; v=v->in_scope)
4755 if (v->depth == 0 && v->type && v->type->check_args) {
4764 struct value *val = var_value(&context, v);
4765 printf("func %.*s", v->name->name.len, v->name->name.txt);
4766 v->type->print_type_decl(v->type, stdout);
4768 print_exec(val->function, 0, brackets);
4770 print_value(v->type, val);
4771 printf("/* frame size %d */\n", v->type->function.local_size);
4777 ###### core functions
4779 static int analyse_funcs(struct parse_context *c)
4783 for (v = c->in_scope; ok && v; v = v->in_scope) {
4785 if (v->depth != 0 || !v->type || !v->type->check_args)
4787 val = var_value(c, v);
4790 propagate_types(val->function, c, &ok, Tnone, 0);
4793 /* Make sure everything is still consistent */
4794 propagate_types(val->function, c, &ok, Tnone, 0);
4795 v->type->function.local_size = scope_finalize(c);
4800 static int analyse_main(struct type *type, struct parse_context *c)
4802 struct binode *bp = type->function.params;
4806 struct type *argv_type;
4807 struct text argv_type_name = { " argv", 5 };
4809 argv_type = add_type(c, argv_type_name, &array_prototype);
4810 argv_type->array.member = Tstr;
4811 argv_type->array.unspec = 1;
4813 for (b = bp; b; b = cast(binode, b->right)) {
4817 propagate_types(b->left, c, &ok, argv_type, 0);
4819 default: /* invalid */ // NOTEST
4820 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4826 return !c->parse_error;
4829 static void interp_main(struct parse_context *c, int argc, char **argv)
4831 struct value *progp = NULL;
4832 struct text main_name = { "main", 4 };
4833 struct variable *mainv;
4839 mainv = var_ref(c, main_name);
4841 progp = var_value(c, mainv);
4842 if (!progp || !progp->function) {
4843 fprintf(stderr, "oceani: no main function found.\n");
4847 if (!analyse_main(mainv->type, c)) {
4848 fprintf(stderr, "oceani: main has wrong type.\n");
4852 al = mainv->type->function.params;
4854 c->local_size = mainv->type->function.local_size;
4855 c->local = calloc(1, c->local_size);
4857 struct var *v = cast(var, al->left);
4858 struct value *vl = var_value(c, v->var);
4868 mpq_set_ui(argcq, argc, 1);
4869 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4870 t->prepare_type(c, t, 0);
4871 array_init(v->var->type, vl);
4872 for (i = 0; i < argc; i++) {
4873 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4876 arg.str.txt = argv[i];
4877 arg.str.len = strlen(argv[i]);
4878 free_value(Tstr, vl2);
4879 dup_value(Tstr, &arg, vl2);
4883 al = cast(binode, al->right);
4885 v = interp_exec(c, progp->function, &vtype);
4886 free_value(vtype, &v);
4891 ###### ast functions
4892 void free_variable(struct variable *v)
4896 ## And now to test it out.
4898 Having a language requires having a "hello world" program. I'll
4899 provide a little more than that: a program that prints "Hello world"
4900 finds the GCD of two numbers, prints the first few elements of
4901 Fibonacci, performs a binary search for a number, and a few other
4902 things which will likely grow as the languages grows.
4904 ###### File: oceani.mk
4907 @echo "===== DEMO ====="
4908 ./oceani --section "demo: hello" oceani.mdc 55 33
4914 four ::= 2 + 2 ; five ::= 10/2
4915 const pie ::= "I like Pie";
4916 cake ::= "The cake is"
4924 func main(argv:[argc::]string)
4925 print "Hello World, what lovely oceans you have!"
4926 print "Are there", five, "?"
4927 print pi, pie, "but", cake
4929 A := $argv[1]; B := $argv[2]
4931 /* When a variable is defined in both branches of an 'if',
4932 * and used afterwards, the variables are merged.
4938 print "Is", A, "bigger than", B,"? ", bigger
4939 /* If a variable is not used after the 'if', no
4940 * merge happens, so types can be different
4943 double:string = "yes"
4944 print A, "is more than twice", B, "?", double
4947 print "double", B, "is", double
4952 if a > 0 and then b > 0:
4958 print "GCD of", A, "and", B,"is", a
4960 print a, "is not positive, cannot calculate GCD"
4962 print b, "is not positive, cannot calculate GCD"
4967 print "Fibonacci:", f1,f2,
4968 then togo = togo - 1
4976 /* Binary search... */
4981 mid := (lo + hi) / 2
4994 print "Yay, I found", target
4996 print "Closest I found was", lo
5001 // "middle square" PRNG. Not particularly good, but one my
5002 // Dad taught me - the first one I ever heard of.
5003 for i:=1; then i = i + 1; while i < size:
5004 n := list[i-1] * list[i-1]
5005 list[i] = (n / 100) % 10 000
5007 print "Before sort:",
5008 for i:=0; then i = i + 1; while i < size:
5012 for i := 1; then i=i+1; while i < size:
5013 for j:=i-1; then j=j-1; while j >= 0:
5014 if list[j] > list[j+1]:
5018 print " After sort:",
5019 for i:=0; then i = i + 1; while i < size:
5023 if 1 == 2 then print "yes"; else print "no"
5027 bob.alive = (bob.name == "Hello")
5028 print "bob", "is" if bob.alive else "isn't", "alive"