1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
113 struct parse_context {
114 struct token_config config;
123 #define container_of(ptr, type, member) ({ \
124 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
125 (type *)( (char *)__mptr - offsetof(type,member) );})
127 #define config2context(_conf) container_of(_conf, struct parse_context, \
130 ###### Parser: reduce
131 struct parse_context *c = config2context(config);
139 #include <sys/mman.h>
158 static char Usage[] =
159 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
160 static const struct option long_options[] = {
161 {"trace", 0, NULL, 't'},
162 {"print", 0, NULL, 'p'},
163 {"noexec", 0, NULL, 'n'},
164 {"brackets", 0, NULL, 'b'},
165 {"section", 1, NULL, 's'},
168 const char *options = "tpnbs";
170 static void pr_err(char *msg) // NOTEST
172 fprintf(stderr, "%s\n", msg); // NOTEST
175 int main(int argc, char *argv[])
180 struct section *s, *ss;
181 char *section = NULL;
182 struct parse_context context = {
184 .ignored = (1 << TK_mark),
185 .number_chars = ".,_+- ",
190 int doprint=0, dotrace=0, doexec=1, brackets=0;
192 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
195 case 't': dotrace=1; break;
196 case 'p': doprint=1; break;
197 case 'n': doexec=0; break;
198 case 'b': brackets=1; break;
199 case 's': section = optarg; break;
200 default: fprintf(stderr, Usage);
204 if (optind >= argc) {
205 fprintf(stderr, "oceani: no input file given\n");
208 fd = open(argv[optind], O_RDONLY);
210 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
213 context.file_name = argv[optind];
214 len = lseek(fd, 0, 2);
215 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
216 s = code_extract(file, file+len, pr_err);
218 fprintf(stderr, "oceani: could not find any code in %s\n",
223 ## context initialization
226 for (ss = s; ss; ss = ss->next) {
227 struct text sec = ss->section;
228 if (sec.len == strlen(section) &&
229 strncmp(sec.txt, section, sec.len) == 0)
233 fprintf(stderr, "oceani: cannot find section %s\n",
239 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
242 fprintf(stderr, "oceani: no main function found.\n");
243 context.parse_error = 1;
245 if (context.prog && !context.parse_error) {
246 if (!analyse_prog(context.prog, &context)) {
247 fprintf(stderr, "oceani: type error in program - not running.\n");
248 context.parse_error = 1;
251 if (context.prog && doprint) {
254 print_exec(context.prog, 0, brackets);
256 if (context.prog && doexec && !context.parse_error)
257 interp_prog(&context, context.prog, argc - optind, argv+optind);
258 free_exec(context.prog);
261 struct section *t = s->next;
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);
364 ###### core functions
366 static void type_err(struct parse_context *c,
367 char *fmt, struct exec *loc,
368 struct type *t1, int rules, struct type *t2)
370 fprintf(stderr, "%s:", c->file_name);
371 fput_loc(loc, stderr);
372 for (; *fmt ; fmt++) {
379 case '%': fputc(*fmt, stderr); break; // NOTEST
380 default: fputc('?', stderr); break; // NOTEST
382 type_print(t1, stderr);
385 type_print(t2, stderr);
394 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
396 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
397 t->txt.len, t->txt.txt);
401 ## Entities: declared and predeclared.
403 There are various "things" that the language and/or the interpreter
404 needs to know about to parse and execute a program. These include
405 types, variables, values, and executable code. These are all lumped
406 together under the term "entities" (calling them "objects" would be
407 confusing) and introduced here. The following section will present the
408 different specific code elements which comprise or manipulate these
413 Values come in a wide range of types, with more likely to be added.
414 Each type needs to be able to print its own values (for convenience at
415 least) as well as to compare two values, at least for equality and
416 possibly for order. For now, values might need to be duplicated and
417 freed, though eventually such manipulations will be better integrated
420 Rather than requiring every numeric type to support all numeric
421 operations (add, multiple, etc), we allow types to be able to present
422 as one of a few standard types: integer, float, and fraction. The
423 existence of these conversion functions eventually enable types to
424 determine if they are compatible with other types, though such types
425 have not yet been implemented.
427 Named type are stored in a simple linked list. Objects of each type are
428 "values" which are often passed around by value.
435 ## value union fields
443 void (*init)(struct type *type, struct value *val);
444 void (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
445 void (*print)(struct type *type, struct value *val);
446 void (*print_type)(struct type *type, FILE *f);
447 int (*cmp_order)(struct type *t1, struct type *t2,
448 struct value *v1, struct value *v2);
449 int (*cmp_eq)(struct type *t1, struct type *t2,
450 struct value *v1, struct value *v2);
451 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
452 void (*free)(struct type *type, struct value *val);
453 void (*free_type)(struct type *t);
454 long long (*to_int)(struct value *v);
455 double (*to_float)(struct value *v);
456 int (*to_mpq)(mpq_t *q, struct value *v);
465 struct type *typelist;
469 static struct type *find_type(struct parse_context *c, struct text s)
471 struct type *l = c->typelist;
474 text_cmp(l->name, s) != 0)
479 static struct type *add_type(struct parse_context *c, struct text s,
484 n = calloc(1, sizeof(*n));
487 n->next = c->typelist;
492 static void free_type(struct type *t)
494 /* The type is always a reference to something in the
495 * context, so we don't need to free anything.
499 static void free_value(struct type *type, struct value *v)
505 static void type_print(struct type *type, FILE *f)
508 fputs("*unknown*type*", f); // NOTEST
509 else if (type->name.len)
510 fprintf(f, "%.*s", type->name.len, type->name.txt);
511 else if (type->print_type)
512 type->print_type(type, f);
514 fputs("*invalid*type*", f); // NOTEST
517 static void val_init(struct type *type, struct value *val)
519 if (type && type->init)
520 type->init(type, val);
523 static void dup_value(struct type *type,
524 struct value *vold, struct value *vnew)
526 if (type && type->dup)
527 type->dup(type, vold, vnew);
530 static int value_cmp(struct type *tl, struct type *tr,
531 struct value *left, struct value *right)
533 if (tl && tl->cmp_order)
534 return tl->cmp_order(tl, tr, left, right);
535 if (tl && tl->cmp_eq) // NOTEST
536 return tl->cmp_eq(tl, tr, left, right); // NOTEST
540 static void print_value(struct type *type, struct value *v)
542 if (type && type->print)
543 type->print(type, v);
545 printf("*Unknown*"); // NOTEST
550 static void free_value(struct type *type, struct value *v);
551 static int type_compat(struct type *require, struct type *have, int rules);
552 static void type_print(struct type *type, FILE *f);
553 static void val_init(struct type *type, struct value *v);
554 static void dup_value(struct type *type,
555 struct value *vold, struct value *vnew);
556 static int value_cmp(struct type *tl, struct type *tr,
557 struct value *left, struct value *right);
558 static void print_value(struct type *type, struct value *v);
560 ###### free context types
562 while (context.typelist) {
563 struct type *t = context.typelist;
565 context.typelist = t->next;
571 Type can be specified for local variables, for fields in a structure,
572 for formal parameters to functions, and possibly elsewhere. Different
573 rules may apply in different contexts. As a minimum, a named type may
574 always be used. Currently the type of a formal parameter can be
575 different from types in other contexts, so we have a separate grammar
581 Type -> IDENTIFIER ${
582 $0 = find_type(c, $1.txt);
585 "error: undefined type", &$1);
592 FormalType -> Type ${ $0 = $<1; }$
593 ## formal type grammar
597 Values of the base types can be numbers, which we represent as
598 multi-precision fractions, strings, Booleans and labels. When
599 analysing the program we also need to allow for places where no value
600 is meaningful (type `Tnone`) and where we don't know what type to
601 expect yet (type is `NULL`).
603 Values are never shared, they are always copied when used, and freed
604 when no longer needed.
606 When propagating type information around the program, we need to
607 determine if two types are compatible, where type `NULL` is compatible
608 with anything. There are two special cases with type compatibility,
609 both related to the Conditional Statement which will be described
610 later. In some cases a Boolean can be accepted as well as some other
611 primary type, and in others any type is acceptable except a label (`Vlabel`).
612 A separate function encoding these cases will simplify some code later.
614 ###### type functions
616 int (*compat)(struct type *this, struct type *other);
620 static int type_compat(struct type *require, struct type *have, int rules)
622 if ((rules & Rboolok) && have == Tbool)
624 if ((rules & Rnolabel) && have == Tlabel)
626 if (!require || !have)
630 return require->compat(require, have);
632 return require == have;
637 #include "parse_string.h"
638 #include "parse_number.h"
641 myLDLIBS := libnumber.o libstring.o -lgmp
642 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
644 ###### type union fields
645 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
647 ###### value union fields
654 static void _free_value(struct type *type, struct value *v)
658 switch (type->vtype) {
660 case Vstr: free(v->str.txt); break;
661 case Vnum: mpq_clear(v->num); break;
667 ###### value functions
669 static void _val_init(struct type *type, struct value *val)
671 switch(type->vtype) {
672 case Vnone: // NOTEST
675 mpq_init(val->num); break;
677 val->str.txt = malloc(1);
689 static void _dup_value(struct type *type,
690 struct value *vold, struct value *vnew)
692 switch (type->vtype) {
693 case Vnone: // NOTEST
696 vnew->label = vold->label;
699 vnew->bool = vold->bool;
703 mpq_set(vnew->num, vold->num);
706 vnew->str.len = vold->str.len;
707 vnew->str.txt = malloc(vnew->str.len);
708 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
713 static int _value_cmp(struct type *tl, struct type *tr,
714 struct value *left, struct value *right)
718 return tl - tr; // NOTEST
720 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
721 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
722 case Vstr: cmp = text_cmp(left->str, right->str); break;
723 case Vbool: cmp = left->bool - right->bool; break;
724 case Vnone: cmp = 0; // NOTEST
729 static void _print_value(struct type *type, struct value *v)
731 switch (type->vtype) {
732 case Vnone: // NOTEST
733 printf("*no-value*"); break; // NOTEST
734 case Vlabel: // NOTEST
735 printf("*label-%p*", v->label); break; // NOTEST
737 printf("%.*s", v->str.len, v->str.txt); break;
739 printf("%s", v->bool ? "True":"False"); break;
744 mpf_set_q(fl, v->num);
745 gmp_printf("%Fg", fl);
752 static void _free_value(struct type *type, struct value *v);
754 static struct type base_prototype = {
756 .print = _print_value,
757 .cmp_order = _value_cmp,
758 .cmp_eq = _value_cmp,
763 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
766 static struct type *add_base_type(struct parse_context *c, char *n,
767 enum vtype vt, int size)
769 struct text txt = { n, strlen(n) };
772 t = add_type(c, txt, &base_prototype);
775 t->align = size > sizeof(void*) ? sizeof(void*) : size;
776 if (t->size & (t->align - 1))
777 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
781 ###### context initialization
783 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
784 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
785 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
786 Tnone = add_base_type(&context, "none", Vnone, 0);
787 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
791 Variables are scoped named values. We store the names in a linked list
792 of "bindings" sorted in lexical order, and use sequential search and
799 struct binding *next; // in lexical order
803 This linked list is stored in the parse context so that "reduce"
804 functions can find or add variables, and so the analysis phase can
805 ensure that every variable gets a type.
809 struct binding *varlist; // In lexical order
813 static struct binding *find_binding(struct parse_context *c, struct text s)
815 struct binding **l = &c->varlist;
820 (cmp = text_cmp((*l)->name, s)) < 0)
824 n = calloc(1, sizeof(*n));
831 Each name can be linked to multiple variables defined in different
832 scopes. Each scope starts where the name is declared and continues
833 until the end of the containing code block. Scopes of a given name
834 cannot nest, so a declaration while a name is in-scope is an error.
836 ###### binding fields
837 struct variable *var;
841 struct variable *previous;
843 struct binding *name;
844 struct exec *where_decl;// where name was declared
845 struct exec *where_set; // where type was set
849 While the naming seems strange, we include local constants in the
850 definition of variables. A name declared `var := value` can
851 subsequently be changed, but a name declared `var ::= value` cannot -
854 ###### variable fields
857 Scopes in parallel branches can be partially merged. More
858 specifically, if a given name is declared in both branches of an
859 if/else then its scope is a candidate for merging. Similarly if
860 every branch of an exhaustive switch (e.g. has an "else" clause)
861 declares a given name, then the scopes from the branches are
862 candidates for merging.
864 Note that names declared inside a loop (which is only parallel to
865 itself) are never visible after the loop. Similarly names defined in
866 scopes which are not parallel, such as those started by `for` and
867 `switch`, are never visible after the scope. Only variables defined in
868 both `then` and `else` (including the implicit then after an `if`, and
869 excluding `then` used with `for`) and in all `case`s and `else` of a
870 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
872 Labels, which are a bit like variables, follow different rules.
873 Labels are not explicitly declared, but if an undeclared name appears
874 in a context where a label is legal, that effectively declares the
875 name as a label. The declaration remains in force (or in scope) at
876 least to the end of the immediately containing block and conditionally
877 in any larger containing block which does not declare the name in some
878 other way. Importantly, the conditional scope extension happens even
879 if the label is only used in one parallel branch of a conditional --
880 when used in one branch it is treated as having been declared in all
883 Merge candidates are tentatively visible beyond the end of the
884 branching statement which creates them. If the name is used, the
885 merge is affirmed and they become a single variable visible at the
886 outer layer. If not - if it is redeclared first - the merge lapses.
888 To track scopes we have an extra stack, implemented as a linked list,
889 which roughly parallels the parse stack and which is used exclusively
890 for scoping. When a new scope is opened, a new frame is pushed and
891 the child-count of the parent frame is incremented. This child-count
892 is used to distinguish between the first of a set of parallel scopes,
893 in which declared variables must not be in scope, and subsequent
894 branches, whether they may already be conditionally scoped.
896 To push a new frame *before* any code in the frame is parsed, we need a
897 grammar reduction. This is most easily achieved with a grammar
898 element which derives the empty string, and creates the new scope when
899 it is recognised. This can be placed, for example, between a keyword
900 like "if" and the code following it.
904 struct scope *parent;
910 struct scope *scope_stack;
913 static void scope_pop(struct parse_context *c)
915 struct scope *s = c->scope_stack;
917 c->scope_stack = s->parent;
922 static void scope_push(struct parse_context *c)
924 struct scope *s = calloc(1, sizeof(*s));
926 c->scope_stack->child_count += 1;
927 s->parent = c->scope_stack;
935 OpenScope -> ${ scope_push(c); }$
937 Each variable records a scope depth and is in one of four states:
939 - "in scope". This is the case between the declaration of the
940 variable and the end of the containing block, and also between
941 the usage with affirms a merge and the end of that block.
943 The scope depth is not greater than the current parse context scope
944 nest depth. When the block of that depth closes, the state will
945 change. To achieve this, all "in scope" variables are linked
946 together as a stack in nesting order.
948 - "pending". The "in scope" block has closed, but other parallel
949 scopes are still being processed. So far, every parallel block at
950 the same level that has closed has declared the name.
952 The scope depth is the depth of the last parallel block that
953 enclosed the declaration, and that has closed.
955 - "conditionally in scope". The "in scope" block and all parallel
956 scopes have closed, and no further mention of the name has been seen.
957 This state includes a secondary nest depth (`min_depth`) which records
958 the outermost scope seen since the variable became conditionally in
959 scope. If a use of the name is found, the variable becomes "in scope"
960 and that secondary depth becomes the recorded scope depth. If the
961 name is declared as a new variable, the old variable becomes "out of
962 scope" and the recorded scope depth stays unchanged.
964 - "out of scope". The variable is neither in scope nor conditionally
965 in scope. It is permanently out of scope now and can be removed from
966 the "in scope" stack.
968 ###### variable fields
969 int depth, min_depth;
970 enum { OutScope, PendingScope, CondScope, InScope } scope;
971 struct variable *in_scope;
975 struct variable *in_scope;
977 All variables with the same name are linked together using the
978 'previous' link. Those variable that have been affirmatively merged all
979 have a 'merged' pointer that points to one primary variable - the most
980 recently declared instance. When merging variables, we need to also
981 adjust the 'merged' pointer on any other variables that had previously
982 been merged with the one that will no longer be primary.
984 A variable that is no longer the most recent instance of a name may
985 still have "pending" scope, if it might still be merged with most
986 recent instance. These variables don't really belong in the
987 "in_scope" list, but are not immediately removed when a new instance
988 is found. Instead, they are detected and ignored when considering the
989 list of in_scope names.
991 The storage of the value of a variable will be described later. For now
992 we just need to know that when a variable goes out of scope, it might
993 need to be freed. For this we need to be able to find it, so assume that
994 `var_value()` will provide that.
996 ###### variable fields
997 struct variable *merged;
1001 static void variable_merge(struct variable *primary, struct variable *secondary)
1005 primary = primary->merged;
1007 for (v = primary->previous; v; v=v->previous)
1008 if (v == secondary || v == secondary->merged ||
1009 v->merged == secondary ||
1010 v->merged == secondary->merged) {
1011 v->scope = OutScope;
1012 v->merged = primary;
1016 ###### forward decls
1017 static struct value *var_value(struct parse_context *c, struct variable *v);
1019 ###### free context vars
1021 while (context.varlist) {
1022 struct binding *b = context.varlist;
1023 struct variable *v = b->var;
1024 context.varlist = b->next;
1027 struct variable *t = v;
1030 free_value(t->type, var_value(&context, t));
1032 // This is a global constant
1033 free_exec(t->where_decl);
1038 #### Manipulating Bindings
1040 When a name is conditionally visible, a new declaration discards the
1041 old binding - the condition lapses. Conversely a usage of the name
1042 affirms the visibility and extends it to the end of the containing
1043 block - i.e. the block that contains both the original declaration and
1044 the latest usage. This is determined from `min_depth`. When a
1045 conditionally visible variable gets affirmed like this, it is also
1046 merged with other conditionally visible variables with the same name.
1048 When we parse a variable declaration we either report an error if the
1049 name is currently bound, or create a new variable at the current nest
1050 depth if the name is unbound or bound to a conditionally scoped or
1051 pending-scope variable. If the previous variable was conditionally
1052 scoped, it and its homonyms becomes out-of-scope.
1054 When we parse a variable reference (including non-declarative assignment
1055 "foo = bar") we report an error if the name is not bound or is bound to
1056 a pending-scope variable; update the scope if the name is bound to a
1057 conditionally scoped variable; or just proceed normally if the named
1058 variable is in scope.
1060 When we exit a scope, any variables bound at this level are either
1061 marked out of scope or pending-scoped, depending on whether the scope
1062 was sequential or parallel. Here a "parallel" scope means the "then"
1063 or "else" part of a conditional, or any "case" or "else" branch of a
1064 switch. Other scopes are "sequential".
1066 When exiting a parallel scope we check if there are any variables that
1067 were previously pending and are still visible. If there are, then
1068 there weren't redeclared in the most recent scope, so they cannot be
1069 merged and must become out-of-scope. If it is not the first of
1070 parallel scopes (based on `child_count`), we check that there was a
1071 previous binding that is still pending-scope. If there isn't, the new
1072 variable must now be out-of-scope.
1074 When exiting a sequential scope that immediately enclosed parallel
1075 scopes, we need to resolve any pending-scope variables. If there was
1076 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1077 we need to mark all pending-scope variable as out-of-scope. Otherwise
1078 all pending-scope variables become conditionally scoped.
1081 enum closetype { CloseSequential, CloseParallel, CloseElse };
1083 ###### ast functions
1085 static struct variable *var_decl(struct parse_context *c, struct text s)
1087 struct binding *b = find_binding(c, s);
1088 struct variable *v = b->var;
1090 switch (v ? v->scope : OutScope) {
1092 /* Caller will report the error */
1096 v && v->scope == CondScope;
1098 v->scope = OutScope;
1102 v = calloc(1, sizeof(*v));
1103 v->previous = b->var;
1107 v->min_depth = v->depth = c->scope_depth;
1109 v->in_scope = c->in_scope;
1114 static struct variable *var_ref(struct parse_context *c, struct text s)
1116 struct binding *b = find_binding(c, s);
1117 struct variable *v = b->var;
1118 struct variable *v2;
1120 switch (v ? v->scope : OutScope) {
1123 /* Caller will report the error */
1126 /* All CondScope variables of this name need to be merged
1127 * and become InScope
1129 v->depth = v->min_depth;
1131 for (v2 = v->previous;
1132 v2 && v2->scope == CondScope;
1134 variable_merge(v, v2);
1142 static void var_block_close(struct parse_context *c, enum closetype ct)
1144 /* Close off all variables that are in_scope.
1145 * Some variables in c->scope may already be not-in-scope,
1146 * such as when a PendingScope variable is hidden by a new
1147 * variable with the same name.
1148 * So we check for v->name->var != v and drop them.
1149 * If we choose to make a variable OutScope, we drop it
1152 struct variable *v, **vp, *v2;
1155 for (vp = &c->in_scope;
1156 (v = *vp) && v->min_depth > c->scope_depth;
1157 (v->scope == OutScope || v->name->var != v)
1158 ? (*vp = v->in_scope, 0)
1159 : ( vp = &v->in_scope, 0)) {
1160 v->min_depth = c->scope_depth;
1161 if (v->name->var != v)
1162 /* This is still in scope, but we haven't just
1168 case CloseParallel: /* handle PendingScope */
1172 if (c->scope_stack->child_count == 1)
1173 /* first among parallel branches */
1174 v->scope = PendingScope;
1175 else if (v->previous &&
1176 v->previous->scope == PendingScope)
1177 /* all previous branches used name */
1178 v->scope = PendingScope;
1179 else if (v->type == Tlabel) // UNTESTED
1180 /* Labels remain pending even when not used */
1181 v->scope = PendingScope; // UNTESTED
1182 if (ct == CloseElse) {
1183 /* All Pending variables with this name
1184 * are now Conditional */
1186 v2 && v2->scope == PendingScope;
1188 v2->scope = CondScope;
1192 /* Not possible as it would require
1193 * parallel scope to be nested immediately
1194 * in a parallel scope, and that never
1198 /* Not possible as we already tested for
1204 case CloseSequential:
1205 if (v->type == Tlabel)
1206 v->scope = PendingScope;
1209 v->scope = OutScope;
1212 /* There was no 'else', so we can only become
1213 * conditional if we know the cases were exhaustive,
1214 * and that doesn't mean anything yet.
1215 * So only labels become conditional..
1218 v2 && v2->scope == PendingScope;
1220 if (v2->type == Tlabel) {
1221 v2->scope = CondScope;
1223 v2->scope = OutScope;
1226 case OutScope: break;
1235 The value of a variable is store separately from the variable, on an
1236 analogue of a stack frame. There are (currently) two frames that can be
1237 active. A global frame which currently only stores constants, and a
1238 stacked frame which stores local variables. Each variable knows if it
1239 is global or not, and what its index into the frame is.
1241 Values in the global frame are known immediately they are relevant, so
1242 the frame needs to be reallocated as it grows so it can store those
1243 values. The local frame doesn't get values until the interpreted phase
1244 is started, so there is no need to allocate until the size is known.
1246 ###### variable fields
1250 ###### parse context
1252 short global_size, global_alloc;
1254 void *global, *local;
1256 ###### ast functions
1258 static struct value *var_value(struct parse_context *c, struct variable *v)
1261 if (!c->local || !v->type)
1263 if (v->frame_pos + v->type->size > c->local_size) {
1264 printf("INVALID frame_pos\n"); // NOTEST
1267 return c->local + v->frame_pos;
1269 if (c->global_size > c->global_alloc) {
1270 int old = c->global_alloc;
1271 c->global_alloc = (c->global_size | 1023) + 1024;
1272 c->global = realloc(c->global, c->global_alloc);
1273 memset(c->global + old, 0, c->global_alloc - old);
1275 return c->global + v->frame_pos;
1278 static struct value *global_alloc(struct parse_context *c, struct type *t,
1279 struct variable *v, struct value *init)
1282 struct variable scratch;
1284 if (t->prepare_type)
1285 t->prepare_type(c, t, 1); // NOTEST
1287 if (c->global_size & (t->align - 1))
1288 c->global_size = (c->global_size + t->align) & ~(t->align-1); // UNTESTED
1293 v->frame_pos = c->global_size;
1295 c->global_size += v->type->size;
1296 ret = var_value(c, v);
1298 memcpy(ret, init, t->size);
1304 As global values are found -- struct field initializers, labels etc --
1305 `global_alloc()` is called to record the value in the global frame.
1307 When the program is fully parsed, we need to walk the list of variables
1308 to find any that weren't merged away and that aren't global, and to
1309 calculate the frame size and assign a frame position for each variable.
1310 For this we have `scope_finalize()`.
1312 ###### ast functions
1314 static void scope_finalize(struct parse_context *c)
1318 for (b = c->varlist; b; b = b->next) {
1320 for (v = b->var; v; v = v->previous) {
1321 struct type *t = v->type;
1326 if (c->local_size & (t->align - 1))
1327 c->local_size = (c->local_size + t->align) & ~(t->align-1);
1328 v->frame_pos = c->local_size;
1329 c->local_size += v->type->size;
1332 c->local = calloc(1, c->local_size);
1335 ###### free context storage
1336 free(context.global);
1337 free(context.local);
1341 Executables can be lots of different things. In many cases an
1342 executable is just an operation combined with one or two other
1343 executables. This allows for expressions and lists etc. Other times an
1344 executable is something quite specific like a constant or variable name.
1345 So we define a `struct exec` to be a general executable with a type, and
1346 a `struct binode` which is a subclass of `exec`, forms a node in a
1347 binary tree, and holds an operation. There will be other subclasses,
1348 and to access these we need to be able to `cast` the `exec` into the
1349 various other types. The first field in any `struct exec` is the type
1350 from the `exec_types` enum.
1353 #define cast(structname, pointer) ({ \
1354 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1355 if (__mptr && *__mptr != X##structname) abort(); \
1356 (struct structname *)( (char *)__mptr);})
1358 #define new(structname) ({ \
1359 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1360 __ptr->type = X##structname; \
1361 __ptr->line = -1; __ptr->column = -1; \
1364 #define new_pos(structname, token) ({ \
1365 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1366 __ptr->type = X##structname; \
1367 __ptr->line = token.line; __ptr->column = token.col; \
1376 enum exec_types type;
1384 struct exec *left, *right;
1387 ###### ast functions
1389 static int __fput_loc(struct exec *loc, FILE *f)
1393 if (loc->line >= 0) {
1394 fprintf(f, "%d:%d: ", loc->line, loc->column);
1397 if (loc->type == Xbinode)
1398 return __fput_loc(cast(binode,loc)->left, f) ||
1399 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1402 static void fput_loc(struct exec *loc, FILE *f)
1404 if (!__fput_loc(loc, f))
1405 fprintf(f, "??:??: "); // NOTEST
1408 Each different type of `exec` node needs a number of functions defined,
1409 a bit like methods. We must be able to free it, print it, analyse it
1410 and execute it. Once we have specific `exec` types we will need to
1411 parse them too. Let's take this a bit more slowly.
1415 The parser generator requires a `free_foo` function for each struct
1416 that stores attributes and they will often be `exec`s and subtypes
1417 there-of. So we need `free_exec` which can handle all the subtypes,
1418 and we need `free_binode`.
1420 ###### ast functions
1422 static void free_binode(struct binode *b)
1427 free_exec(b->right);
1431 ###### core functions
1432 static void free_exec(struct exec *e)
1441 ###### forward decls
1443 static void free_exec(struct exec *e);
1445 ###### free exec cases
1446 case Xbinode: free_binode(cast(binode, e)); break;
1450 Printing an `exec` requires that we know the current indent level for
1451 printing line-oriented components. As will become clear later, we
1452 also want to know what sort of bracketing to use.
1454 ###### ast functions
1456 static void do_indent(int i, char *str)
1463 ###### core functions
1464 static void print_binode(struct binode *b, int indent, int bracket)
1468 ## print binode cases
1472 static void print_exec(struct exec *e, int indent, int bracket)
1478 print_binode(cast(binode, e), indent, bracket); break;
1483 ###### forward decls
1485 static void print_exec(struct exec *e, int indent, int bracket);
1489 As discussed, analysis involves propagating type requirements around the
1490 program and looking for errors.
1492 So `propagate_types` is passed an expected type (being a `struct type`
1493 pointer together with some `val_rules` flags) that the `exec` is
1494 expected to return, and returns the type that it does return, either
1495 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1496 by reference. It is set to `0` when an error is found, and `2` when
1497 any change is made. If it remains unchanged at `1`, then no more
1498 propagation is needed.
1502 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1506 if (rules & Rnolabel)
1507 fputs(" (labels not permitted)", stderr);
1510 ###### core functions
1512 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1513 struct type *type, int rules);
1514 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1515 struct type *type, int rules)
1522 switch (prog->type) {
1525 struct binode *b = cast(binode, prog);
1527 ## propagate binode cases
1531 ## propagate exec cases
1536 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1537 struct type *type, int rules)
1539 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1548 Interpreting an `exec` doesn't require anything but the `exec`. State
1549 is stored in variables and each variable will be directly linked from
1550 within the `exec` tree. The exception to this is the `main` function
1551 which needs to look at command line arguments. This function will be
1552 interpreted separately.
1554 Each `exec` can return a value combined with a type in `struct lrval`.
1555 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1556 the location of a value, which can be updated, in `lval`. Others will
1557 set `lval` to NULL indicating that there is a value of appropriate type
1560 ###### core functions
1564 struct value rval, *lval;
1567 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1569 static struct value interp_exec(struct parse_context *c, struct exec *e,
1570 struct type **typeret)
1572 struct lrval ret = _interp_exec(c, e);
1574 if (!ret.type) abort();
1576 *typeret = ret.type;
1578 dup_value(ret.type, ret.lval, &ret.rval);
1582 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1583 struct type **typeret)
1585 struct lrval ret = _interp_exec(c, e);
1588 *typeret = ret.type;
1590 free_value(ret.type, &ret.rval);
1594 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1597 struct value rv = {}, *lrv = NULL;
1598 struct type *rvtype;
1600 rvtype = ret.type = Tnone;
1610 struct binode *b = cast(binode, e);
1611 struct value left, right, *lleft;
1612 struct type *ltype, *rtype;
1613 ltype = rtype = Tnone;
1615 ## interp binode cases
1617 free_value(ltype, &left);
1618 free_value(rtype, &right);
1621 ## interp exec cases
1631 Now that we have the shape of the interpreter in place we can add some
1632 complex types and connected them in to the data structures and the
1633 different phases of parse, analyse, print, interpret.
1635 Thus far we have arrays and structs.
1639 Arrays can be declared by giving a size and a type, as `[size]type' so
1640 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1641 size can be either a literal number, or a named constant. Some day an
1642 arbitrary expression will be supported.
1644 As a formal parameter to a function, the array can be declared with a
1645 new variable as the size: `name:[size::number]string`. The `size`
1646 variable is set to the size of the array and must be a constant. As
1647 `number` is the only supported type, it can be left out:
1648 `name:[size::]string`.
1650 Arrays cannot be assigned. When pointers are introduced we will also
1651 introduce array slices which can refer to part or all of an array -
1652 the assignment syntax will create a slice. For now, an array can only
1653 ever be referenced by the name it is declared with. It is likely that
1654 a "`copy`" primitive will eventually be define which can be used to
1655 make a copy of an array with controllable recursive depth.
1657 For now we have two sorts of array, those with fixed size either because
1658 it is given as a literal number or because it is a struct member (which
1659 cannot have a runtime-changing size), and those with a size that is
1660 determined at runtime - local variables with a const size. The former
1661 have their size calculated at parse time, the latter at run time.
1663 For the latter type, the `size` field of the type is the size of a
1664 pointer, and the array is reallocated every time it comes into scope.
1666 We differentiate struct fields with a const size from local variables
1667 with a const size by whether they are prepared at parse time or not.
1669 ###### type union fields
1672 int unspec; // size is unspecified - vsize must be set.
1675 struct variable *vsize;
1676 struct type *member;
1679 ###### value union fields
1680 void *array; // used if not static_size
1682 ###### value functions
1684 static void array_prepare_type(struct parse_context *c, struct type *type,
1687 struct value *vsize;
1689 if (!type->array.vsize || type->array.static_size)
1692 vsize = var_value(c, type->array.vsize);
1694 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1695 type->array.size = mpz_get_si(q);
1699 type->array.static_size = 1;
1700 type->size = type->array.size * type->array.member->size;
1701 type->align = type->array.member->align;
1705 static void array_init(struct type *type, struct value *val)
1708 void *ptr = val->ptr;
1712 if (!type->array.static_size) {
1713 val->array = calloc(type->array.size,
1714 type->array.member->size);
1717 for (i = 0; i < type->array.size; i++) {
1719 v = (void*)ptr + i * type->array.member->size;
1720 val_init(type->array.member, v);
1724 static void array_free(struct type *type, struct value *val)
1727 void *ptr = val->ptr;
1729 if (!type->array.static_size)
1731 for (i = 0; i < type->array.size; i++) {
1733 v = (void*)ptr + i * type->array.member->size;
1734 free_value(type->array.member, v);
1736 if (!type->array.static_size)
1740 static int array_compat(struct type *require, struct type *have)
1742 if (have->compat != require->compat)
1743 return 0; // UNTESTED
1744 /* Both are arrays, so we can look at details */
1745 if (!type_compat(require->array.member, have->array.member, 0))
1747 if (have->array.unspec && require->array.unspec) {
1748 if (have->array.vsize && require->array.vsize &&
1749 have->array.vsize != require->array.vsize) // UNTESTED
1750 /* sizes might not be the same */
1751 return 0; // UNTESTED
1754 if (have->array.unspec || require->array.unspec)
1755 return 1; // UNTESTED
1756 if (require->array.vsize == NULL && have->array.vsize == NULL)
1757 return require->array.size == have->array.size;
1759 return require->array.vsize == have->array.vsize; // UNTESTED
1762 static void array_print_type(struct type *type, FILE *f)
1765 if (type->array.vsize) {
1766 struct binding *b = type->array.vsize->name;
1767 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1768 type->array.unspec ? "::" : "");
1770 fprintf(f, "%d]", type->array.size);
1771 type_print(type->array.member, f);
1774 static struct type array_prototype = {
1776 .prepare_type = array_prepare_type,
1777 .print_type = array_print_type,
1778 .compat = array_compat,
1780 .size = sizeof(void*),
1781 .align = sizeof(void*),
1784 ###### declare terminals
1789 | [ NUMBER ] Type ${ {
1792 struct text noname = { "", 0 };
1795 $0 = t = add_type(c, noname, &array_prototype);
1796 t->array.member = $<4;
1797 t->array.vsize = NULL;
1798 if (number_parse(num, tail, $2.txt) == 0)
1799 tok_err(c, "error: unrecognised number", &$2);
1801 tok_err(c, "error: unsupported number suffix", &$2);
1803 t->array.size = mpz_get_ui(mpq_numref(num));
1804 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1805 tok_err(c, "error: array size must be an integer",
1807 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1808 tok_err(c, "error: array size is too large",
1812 t->array.static_size = 1;
1813 t->size = t->array.size * t->array.member->size;
1814 t->align = t->array.member->align;
1817 | [ IDENTIFIER ] Type ${ {
1818 struct variable *v = var_ref(c, $2.txt);
1819 struct text noname = { "", 0 };
1822 tok_err(c, "error: name undeclared", &$2);
1823 else if (!v->constant)
1824 tok_err(c, "error: array size must be a constant", &$2);
1826 $0 = add_type(c, noname, &array_prototype);
1827 $0->array.member = $<4;
1829 $0->array.vsize = v;
1834 OptType -> Type ${ $0 = $<1; }$
1837 ###### formal type grammar
1839 | [ IDENTIFIER :: OptType ] Type ${ {
1840 struct variable *v = var_decl(c, $ID.txt);
1841 struct text noname = { "", 0 };
1847 $0 = add_type(c, noname, &array_prototype);
1848 $0->array.member = $<6;
1850 $0->array.unspec = 1;
1851 $0->array.vsize = v;
1857 ###### variable grammar
1859 | Variable [ Expression ] ${ {
1860 struct binode *b = new(binode);
1867 ###### print binode cases
1869 print_exec(b->left, -1, bracket);
1871 print_exec(b->right, -1, bracket);
1875 ###### propagate binode cases
1877 /* left must be an array, right must be a number,
1878 * result is the member type of the array
1880 propagate_types(b->right, c, ok, Tnum, 0);
1881 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1882 if (!t || t->compat != array_compat) {
1883 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1886 if (!type_compat(type, t->array.member, rules)) {
1887 type_err(c, "error: have %1 but need %2", prog,
1888 t->array.member, rules, type);
1890 return t->array.member;
1894 ###### interp binode cases
1900 lleft = linterp_exec(c, b->left, <ype);
1901 right = interp_exec(c, b->right, &rtype);
1903 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1907 if (ltype->array.static_size)
1910 ptr = *(void**)lleft;
1911 rvtype = ltype->array.member;
1912 if (i >= 0 && i < ltype->array.size)
1913 lrv = ptr + i * rvtype->size;
1915 val_init(ltype->array.member, &rv);
1922 A `struct` is a data-type that contains one or more other data-types.
1923 It differs from an array in that each member can be of a different
1924 type, and they are accessed by name rather than by number. Thus you
1925 cannot choose an element by calculation, you need to know what you
1928 The language makes no promises about how a given structure will be
1929 stored in memory - it is free to rearrange fields to suit whatever
1930 criteria seems important.
1932 Structs are declared separately from program code - they cannot be
1933 declared in-line in a variable declaration like arrays can. A struct
1934 is given a name and this name is used to identify the type - the name
1935 is not prefixed by the word `struct` as it would be in C.
1937 Structs are only treated as the same if they have the same name.
1938 Simply having the same fields in the same order is not enough. This
1939 might change once we can create structure initializers from a list of
1942 Each component datum is identified much like a variable is declared,
1943 with a name, one or two colons, and a type. The type cannot be omitted
1944 as there is no opportunity to deduce the type from usage. An initial
1945 value can be given following an equals sign, so
1947 ##### Example: a struct type
1953 would declare a type called "complex" which has two number fields,
1954 each initialised to zero.
1956 Struct will need to be declared separately from the code that uses
1957 them, so we will need to be able to print out the declaration of a
1958 struct when reprinting the whole program. So a `print_type_decl` type
1959 function will be needed.
1961 ###### type union fields
1973 ###### type functions
1974 void (*print_type_decl)(struct type *type, FILE *f);
1976 ###### value functions
1978 static void structure_init(struct type *type, struct value *val)
1982 for (i = 0; i < type->structure.nfields; i++) {
1984 v = (void*) val->ptr + type->structure.fields[i].offset;
1985 if (type->structure.fields[i].init)
1986 dup_value(type->structure.fields[i].type,
1987 type->structure.fields[i].init,
1990 val_init(type->structure.fields[i].type, v);
1994 static void structure_free(struct type *type, struct value *val)
1998 for (i = 0; i < type->structure.nfields; i++) {
2000 v = (void*)val->ptr + type->structure.fields[i].offset;
2001 free_value(type->structure.fields[i].type, v);
2005 static void structure_free_type(struct type *t)
2008 for (i = 0; i < t->structure.nfields; i++)
2009 if (t->structure.fields[i].init) {
2010 free_value(t->structure.fields[i].type,
2011 t->structure.fields[i].init);
2013 free(t->structure.fields);
2016 static struct type structure_prototype = {
2017 .init = structure_init,
2018 .free = structure_free,
2019 .free_type = structure_free_type,
2020 .print_type_decl = structure_print_type,
2034 ###### free exec cases
2036 free_exec(cast(fieldref, e)->left);
2040 ###### declare terminals
2043 ###### variable grammar
2045 | Variable . IDENTIFIER ${ {
2046 struct fieldref *fr = new_pos(fieldref, $2);
2053 ###### print exec cases
2057 struct fieldref *f = cast(fieldref, e);
2058 print_exec(f->left, -1, bracket);
2059 printf(".%.*s", f->name.len, f->name.txt);
2063 ###### ast functions
2064 static int find_struct_index(struct type *type, struct text field)
2067 for (i = 0; i < type->structure.nfields; i++)
2068 if (text_cmp(type->structure.fields[i].name, field) == 0)
2073 ###### propagate exec cases
2077 struct fieldref *f = cast(fieldref, prog);
2078 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2081 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2083 else if (st->init != structure_init)
2084 type_err(c, "error: field reference attempted on %1, not a struct",
2085 f->left, st, 0, NULL);
2086 else if (f->index == -2) {
2087 f->index = find_struct_index(st, f->name);
2089 type_err(c, "error: cannot find requested field in %1",
2090 f->left, st, 0, NULL);
2092 if (f->index >= 0) {
2093 struct type *ft = st->structure.fields[f->index].type;
2094 if (!type_compat(type, ft, rules))
2095 type_err(c, "error: have %1 but need %2", prog,
2102 ###### interp exec cases
2105 struct fieldref *f = cast(fieldref, e);
2107 struct value *lleft = linterp_exec(c, f->left, <ype);
2108 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2109 rvtype = ltype->structure.fields[f->index].type;
2115 struct fieldlist *prev;
2119 ###### ast functions
2120 static void free_fieldlist(struct fieldlist *f)
2124 free_fieldlist(f->prev);
2126 free_value(f->f.type, f->f.init); // UNTESTED
2127 free(f->f.init); // UNTESTED
2132 ###### top level grammar
2133 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2135 add_type(c, $2.txt, &structure_prototype);
2137 struct fieldlist *f;
2139 for (f = $3; f; f=f->prev)
2142 t->structure.nfields = cnt;
2143 t->structure.fields = calloc(cnt, sizeof(struct field));
2146 int a = f->f.type->align;
2148 t->structure.fields[cnt] = f->f;
2149 if (t->size & (a-1))
2150 t->size = (t->size | (a-1)) + 1;
2151 t->structure.fields[cnt].offset = t->size;
2152 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2161 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2162 | { SimpleFieldList } ${ $0 = $<SFL; }$
2163 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2164 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2166 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2167 | FieldLines SimpleFieldList Newlines ${
2172 SimpleFieldList -> Field ${ $0 = $<F; }$
2173 | SimpleFieldList ; Field ${
2177 | SimpleFieldList ; ${
2180 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2182 Field -> IDENTIFIER : Type = Expression ${ {
2185 $0 = calloc(1, sizeof(struct fieldlist));
2186 $0->f.name = $1.txt;
2191 propagate_types($<5, c, &ok, $3, 0);
2194 c->parse_error = 1; // UNTESTED
2196 struct value vl = interp_exec(c, $5, NULL);
2197 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2200 | IDENTIFIER : Type ${
2201 $0 = calloc(1, sizeof(struct fieldlist));
2202 $0->f.name = $1.txt;
2204 if ($0->f.type->prepare_type)
2205 $0->f.type->prepare_type(c, $0->f.type, 1);
2208 ###### forward decls
2209 static void structure_print_type(struct type *t, FILE *f);
2211 ###### value functions
2212 static void structure_print_type(struct type *t, FILE *f) // UNTESTED
2216 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2218 for (i = 0; i < t->structure.nfields; i++) {
2219 struct field *fl = t->structure.fields + i;
2220 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2221 type_print(fl->type, f);
2222 if (fl->type->print && fl->init) {
2224 if (fl->type == Tstr)
2225 fprintf(f, "\""); // UNTESTED
2226 print_value(fl->type, fl->init);
2227 if (fl->type == Tstr)
2228 fprintf(f, "\""); // UNTESTED
2234 ###### print type decls
2236 struct type *t; // UNTESTED
2239 while (target != 0) {
2241 for (t = context.typelist; t ; t=t->next)
2242 if (t->print_type_decl) {
2251 t->print_type_decl(t, stdout);
2259 A function is a named chunk of code which can be passed parameters and
2260 can return results. Each function has an implicit type which includes
2261 the set of parameters and the return value. As yet these types cannot
2262 be declared separate from the function itself.
2264 In fact, only one function is currently possible - `main`. `main` is
2265 passed an array of strings together with the size of the array, and
2266 doesn't return anything. The strings are command line arguments.
2268 The parameters can be specified either in parentheses as a list, such as
2270 ##### Example: function 1
2272 func main(av:[ac::number]string)
2275 or as an indented list of one parameter per line
2277 ##### Example: function 2
2280 argv:[argc::number]string
2292 MainFunction -> func main ( OpenScope Args ) Block Newlines ${
2295 $0->left = reorder_bilist($<Ar);
2297 var_block_close(c, CloseSequential);
2298 if (c->scope_stack && !c->parse_error) abort();
2300 | func main IN OpenScope OptNL Args OUT OptNL do Block Newlines ${
2303 $0->left = reorder_bilist($<Ar);
2305 var_block_close(c, CloseSequential);
2306 if (c->scope_stack && !c->parse_error) abort();
2308 | func main NEWLINE OpenScope OptNL do Block Newlines ${
2313 var_block_close(c, CloseSequential);
2314 if (c->scope_stack && !c->parse_error) abort();
2317 Args -> ${ $0 = NULL; }$
2318 | Varlist ${ $0 = $<1; }$
2319 | Varlist ; ${ $0 = $<1; }$
2320 | Varlist NEWLINE ${ $0 = $<1; }$
2322 Varlist -> Varlist ; ArgDecl ${ // UNTESTED
2336 ArgDecl -> IDENTIFIER : FormalType ${ {
2337 struct variable *v = var_decl(c, $1.txt);
2343 ## Executables: the elements of code
2345 Each code element needs to be parsed, printed, analysed,
2346 interpreted, and freed. There are several, so let's just start with
2347 the easy ones and work our way up.
2351 We have already met values as separate objects. When manifest
2352 constants appear in the program text, that must result in an executable
2353 which has a constant value. So the `val` structure embeds a value in
2366 ###### ast functions
2367 struct val *new_val(struct type *T, struct token tk)
2369 struct val *v = new_pos(val, tk);
2380 $0 = new_val(Tbool, $1);
2384 $0 = new_val(Tbool, $1);
2388 $0 = new_val(Tnum, $1);
2391 if (number_parse($0->val.num, tail, $1.txt) == 0)
2392 mpq_init($0->val.num); // UNTESTED
2394 tok_err(c, "error: unsupported number suffix",
2399 $0 = new_val(Tstr, $1);
2402 string_parse(&$1, '\\', &$0->val.str, tail);
2404 tok_err(c, "error: unsupported string suffix",
2409 $0 = new_val(Tstr, $1);
2412 string_parse(&$1, '\\', &$0->val.str, tail);
2414 tok_err(c, "error: unsupported string suffix",
2419 ###### print exec cases
2422 struct val *v = cast(val, e);
2423 if (v->vtype == Tstr)
2425 print_value(v->vtype, &v->val);
2426 if (v->vtype == Tstr)
2431 ###### propagate exec cases
2434 struct val *val = cast(val, prog);
2435 if (!type_compat(type, val->vtype, rules))
2436 type_err(c, "error: expected %1%r found %2",
2437 prog, type, rules, val->vtype);
2441 ###### interp exec cases
2443 rvtype = cast(val, e)->vtype;
2444 dup_value(rvtype, &cast(val, e)->val, &rv);
2447 ###### ast functions
2448 static void free_val(struct val *v)
2451 free_value(v->vtype, &v->val);
2455 ###### free exec cases
2456 case Xval: free_val(cast(val, e)); break;
2458 ###### ast functions
2459 // Move all nodes from 'b' to 'rv', reversing their order.
2460 // In 'b' 'left' is a list, and 'right' is the last node.
2461 // In 'rv', left' is the first node and 'right' is a list.
2462 static struct binode *reorder_bilist(struct binode *b)
2464 struct binode *rv = NULL;
2467 struct exec *t = b->right;
2471 b = cast(binode, b->left);
2481 Just as we used a `val` to wrap a value into an `exec`, we similarly
2482 need a `var` to wrap a `variable` into an exec. While each `val`
2483 contained a copy of the value, each `var` holds a link to the variable
2484 because it really is the same variable no matter where it appears.
2485 When a variable is used, we need to remember to follow the `->merged`
2486 link to find the primary instance.
2494 struct variable *var;
2502 VariableDecl -> IDENTIFIER : ${ {
2503 struct variable *v = var_decl(c, $1.txt);
2504 $0 = new_pos(var, $1);
2509 v = var_ref(c, $1.txt);
2511 type_err(c, "error: variable '%v' redeclared",
2513 type_err(c, "info: this is where '%v' was first declared",
2514 v->where_decl, NULL, 0, NULL);
2517 | IDENTIFIER :: ${ {
2518 struct variable *v = var_decl(c, $1.txt);
2519 $0 = new_pos(var, $1);
2525 v = var_ref(c, $1.txt);
2527 type_err(c, "error: variable '%v' redeclared",
2529 type_err(c, "info: this is where '%v' was first declared",
2530 v->where_decl, NULL, 0, NULL);
2533 | IDENTIFIER : Type ${ {
2534 struct variable *v = var_decl(c, $1.txt);
2535 $0 = new_pos(var, $1);
2542 v = var_ref(c, $1.txt);
2544 type_err(c, "error: variable '%v' redeclared",
2546 type_err(c, "info: this is where '%v' was first declared",
2547 v->where_decl, NULL, 0, NULL);
2550 | IDENTIFIER :: Type ${ {
2551 struct variable *v = var_decl(c, $1.txt);
2552 $0 = new_pos(var, $1);
2560 v = var_ref(c, $1.txt);
2562 type_err(c, "error: variable '%v' redeclared",
2564 type_err(c, "info: this is where '%v' was first declared",
2565 v->where_decl, NULL, 0, NULL);
2570 Variable -> IDENTIFIER ${ {
2571 struct variable *v = var_ref(c, $1.txt);
2572 $0 = new_pos(var, $1);
2574 /* This might be a label - allocate a var just in case */
2575 v = var_decl(c, $1.txt);
2582 cast(var, $0)->var = v;
2586 ###### print exec cases
2589 struct var *v = cast(var, e);
2591 struct binding *b = v->var->name;
2592 printf("%.*s", b->name.len, b->name.txt);
2599 if (loc && loc->type == Xvar) {
2600 struct var *v = cast(var, loc);
2602 struct binding *b = v->var->name;
2603 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2605 fputs("???", stderr); // NOTEST
2607 fputs("NOTVAR", stderr); // NOTEST
2610 ###### propagate exec cases
2614 struct var *var = cast(var, prog);
2615 struct variable *v = var->var;
2617 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2618 return Tnone; // NOTEST
2621 if (v->constant && (rules & Rnoconstant)) {
2622 type_err(c, "error: Cannot assign to a constant: %v",
2623 prog, NULL, 0, NULL);
2624 type_err(c, "info: name was defined as a constant here",
2625 v->where_decl, NULL, 0, NULL);
2628 if (v->type == Tnone && v->where_decl == prog)
2629 type_err(c, "error: variable used but not declared: %v",
2630 prog, NULL, 0, NULL);
2631 if (v->type == NULL) {
2632 if (type && *ok != 0) {
2634 v->where_set = prog;
2639 if (!type_compat(type, v->type, rules)) {
2640 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2641 type, rules, v->type);
2642 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2643 v->type, rules, NULL);
2650 ###### interp exec cases
2653 struct var *var = cast(var, e);
2654 struct variable *v = var->var;
2657 lrv = var_value(c, v);
2662 ###### ast functions
2664 static void free_var(struct var *v)
2669 ###### free exec cases
2670 case Xvar: free_var(cast(var, e)); break;
2672 ### Expressions: Conditional
2674 Our first user of the `binode` will be conditional expressions, which
2675 is a bit odd as they actually have three components. That will be
2676 handled by having 2 binodes for each expression. The conditional
2677 expression is the lowest precedence operator which is why we define it
2678 first - to start the precedence list.
2680 Conditional expressions are of the form "value `if` condition `else`
2681 other_value". They associate to the right, so everything to the right
2682 of `else` is part of an else value, while only a higher-precedence to
2683 the left of `if` is the if values. Between `if` and `else` there is no
2684 room for ambiguity, so a full conditional expression is allowed in
2696 Expression -> Expression if Expression else Expression $$ifelse ${ {
2697 struct binode *b1 = new(binode);
2698 struct binode *b2 = new(binode);
2707 ## expression grammar
2709 ###### print binode cases
2712 b2 = cast(binode, b->right);
2713 if (bracket) printf("(");
2714 print_exec(b2->left, -1, bracket);
2716 print_exec(b->left, -1, bracket);
2718 print_exec(b2->right, -1, bracket);
2719 if (bracket) printf(")");
2722 ###### propagate binode cases
2725 /* cond must be Tbool, others must match */
2726 struct binode *b2 = cast(binode, b->right);
2729 propagate_types(b->left, c, ok, Tbool, 0);
2730 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2731 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2735 ###### interp binode cases
2738 struct binode *b2 = cast(binode, b->right);
2739 left = interp_exec(c, b->left, <ype);
2741 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
2743 rv = interp_exec(c, b2->right, &rvtype);
2747 ### Expressions: Boolean
2749 The next class of expressions to use the `binode` will be Boolean
2750 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
2751 have same corresponding precendence. The difference is that they don't
2752 evaluate the second expression if not necessary.
2761 ###### expr precedence
2766 ###### expression grammar
2767 | Expression or Expression ${ {
2768 struct binode *b = new(binode);
2774 | Expression or else Expression ${ {
2775 struct binode *b = new(binode);
2782 | Expression and Expression ${ {
2783 struct binode *b = new(binode);
2789 | Expression and then Expression ${ {
2790 struct binode *b = new(binode);
2797 | not Expression ${ {
2798 struct binode *b = new(binode);
2804 ###### print binode cases
2806 if (bracket) printf("(");
2807 print_exec(b->left, -1, bracket);
2809 print_exec(b->right, -1, bracket);
2810 if (bracket) printf(")");
2813 if (bracket) printf("(");
2814 print_exec(b->left, -1, bracket);
2815 printf(" and then ");
2816 print_exec(b->right, -1, bracket);
2817 if (bracket) printf(")");
2820 if (bracket) printf("(");
2821 print_exec(b->left, -1, bracket);
2823 print_exec(b->right, -1, bracket);
2824 if (bracket) printf(")");
2827 if (bracket) printf("(");
2828 print_exec(b->left, -1, bracket);
2829 printf(" or else ");
2830 print_exec(b->right, -1, bracket);
2831 if (bracket) printf(")");
2834 if (bracket) printf("(");
2836 print_exec(b->right, -1, bracket);
2837 if (bracket) printf(")");
2840 ###### propagate binode cases
2846 /* both must be Tbool, result is Tbool */
2847 propagate_types(b->left, c, ok, Tbool, 0);
2848 propagate_types(b->right, c, ok, Tbool, 0);
2849 if (type && type != Tbool)
2850 type_err(c, "error: %1 operation found where %2 expected", prog,
2854 ###### interp binode cases
2856 rv = interp_exec(c, b->left, &rvtype);
2857 right = interp_exec(c, b->right, &rtype);
2858 rv.bool = rv.bool && right.bool;
2861 rv = interp_exec(c, b->left, &rvtype);
2863 rv = interp_exec(c, b->right, NULL);
2866 rv = interp_exec(c, b->left, &rvtype);
2867 right = interp_exec(c, b->right, &rtype);
2868 rv.bool = rv.bool || right.bool;
2871 rv = interp_exec(c, b->left, &rvtype);
2873 rv = interp_exec(c, b->right, NULL);
2876 rv = interp_exec(c, b->right, &rvtype);
2880 ### Expressions: Comparison
2882 Of slightly higher precedence that Boolean expressions are Comparisons.
2883 A comparison takes arguments of any comparable type, but the two types
2886 To simplify the parsing we introduce an `eop` which can record an
2887 expression operator, and the `CMPop` non-terminal will match one of them.
2894 ###### ast functions
2895 static void free_eop(struct eop *e)
2909 ###### expr precedence
2910 $LEFT < > <= >= == != CMPop
2912 ###### expression grammar
2913 | Expression CMPop Expression ${ {
2914 struct binode *b = new(binode);
2924 CMPop -> < ${ $0.op = Less; }$
2925 | > ${ $0.op = Gtr; }$
2926 | <= ${ $0.op = LessEq; }$
2927 | >= ${ $0.op = GtrEq; }$
2928 | == ${ $0.op = Eql; }$
2929 | != ${ $0.op = NEql; }$
2931 ###### print binode cases
2939 if (bracket) printf("(");
2940 print_exec(b->left, -1, bracket);
2942 case Less: printf(" < "); break;
2943 case LessEq: printf(" <= "); break;
2944 case Gtr: printf(" > "); break;
2945 case GtrEq: printf(" >= "); break;
2946 case Eql: printf(" == "); break;
2947 case NEql: printf(" != "); break;
2948 default: abort(); // NOTEST
2950 print_exec(b->right, -1, bracket);
2951 if (bracket) printf(")");
2954 ###### propagate binode cases
2961 /* Both must match but not be labels, result is Tbool */
2962 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2964 propagate_types(b->right, c, ok, t, 0);
2966 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
2968 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
2970 if (!type_compat(type, Tbool, 0))
2971 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2972 Tbool, rules, type);
2975 ###### interp binode cases
2984 left = interp_exec(c, b->left, <ype);
2985 right = interp_exec(c, b->right, &rtype);
2986 cmp = value_cmp(ltype, rtype, &left, &right);
2989 case Less: rv.bool = cmp < 0; break;
2990 case LessEq: rv.bool = cmp <= 0; break;
2991 case Gtr: rv.bool = cmp > 0; break;
2992 case GtrEq: rv.bool = cmp >= 0; break;
2993 case Eql: rv.bool = cmp == 0; break;
2994 case NEql: rv.bool = cmp != 0; break;
2995 default: rv.bool = 0; break; // NOTEST
3000 ### Expressions: The rest
3002 The remaining expressions with the highest precedence are arithmetic,
3003 string concatenation, and string conversion. String concatenation
3004 (`++`) has the same precedence as multiplication and division, but lower
3007 String conversion is a temporary feature until I get a better type
3008 system. `$` is a prefix operator which expects a string and returns
3011 `+` and `-` are both infix and prefix operations (where they are
3012 absolute value and negation). These have different operator names.
3014 We also have a 'Bracket' operator which records where parentheses were
3015 found. This makes it easy to reproduce these when printing. Possibly I
3016 should only insert brackets were needed for precedence.
3026 ###### expr precedence
3032 ###### expression grammar
3033 | Expression Eop Expression ${ {
3034 struct binode *b = new(binode);
3041 | Expression Top Expression ${ {
3042 struct binode *b = new(binode);
3049 | ( Expression ) ${ {
3050 struct binode *b = new_pos(binode, $1);
3055 | Uop Expression ${ {
3056 struct binode *b = new(binode);
3061 | Value ${ $0 = $<1; }$
3062 | Variable ${ $0 = $<1; }$
3065 Eop -> + ${ $0.op = Plus; }$
3066 | - ${ $0.op = Minus; }$
3068 Uop -> + ${ $0.op = Absolute; }$
3069 | - ${ $0.op = Negate; }$
3070 | $ ${ $0.op = StringConv; }$
3072 Top -> * ${ $0.op = Times; }$
3073 | / ${ $0.op = Divide; }$
3074 | % ${ $0.op = Rem; }$
3075 | ++ ${ $0.op = Concat; }$
3077 ###### print binode cases
3084 if (bracket) printf("(");
3085 print_exec(b->left, indent, bracket);
3087 case Plus: fputs(" + ", stdout); break;
3088 case Minus: fputs(" - ", stdout); break;
3089 case Times: fputs(" * ", stdout); break;
3090 case Divide: fputs(" / ", stdout); break;
3091 case Rem: fputs(" % ", stdout); break;
3092 case Concat: fputs(" ++ ", stdout); break;
3093 default: abort(); // NOTEST
3095 print_exec(b->right, indent, bracket);
3096 if (bracket) printf(")");
3101 if (bracket) printf("(");
3103 case Absolute: fputs("+", stdout); break;
3104 case Negate: fputs("-", stdout); break;
3105 case StringConv: fputs("$", stdout); break;
3106 default: abort(); // NOTEST
3108 print_exec(b->right, indent, bracket);
3109 if (bracket) printf(")");
3113 print_exec(b->right, indent, bracket);
3117 ###### propagate binode cases
3123 /* both must be numbers, result is Tnum */
3126 /* as propagate_types ignores a NULL,
3127 * unary ops fit here too */
3128 propagate_types(b->left, c, ok, Tnum, 0);
3129 propagate_types(b->right, c, ok, Tnum, 0);
3130 if (!type_compat(type, Tnum, 0))
3131 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3136 /* both must be Tstr, result is Tstr */
3137 propagate_types(b->left, c, ok, Tstr, 0);
3138 propagate_types(b->right, c, ok, Tstr, 0);
3139 if (!type_compat(type, Tstr, 0))
3140 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3145 /* op must be string, result is number */
3146 propagate_types(b->left, c, ok, Tstr, 0);
3147 if (!type_compat(type, Tnum, 0))
3148 type_err(c, // UNTESTED
3149 "error: Can only convert string to number, not %1",
3150 prog, type, 0, NULL);
3154 return propagate_types(b->right, c, ok, type, 0);
3156 ###### interp binode cases
3159 rv = interp_exec(c, b->left, &rvtype);
3160 right = interp_exec(c, b->right, &rtype);
3161 mpq_add(rv.num, rv.num, right.num);
3164 rv = interp_exec(c, b->left, &rvtype);
3165 right = interp_exec(c, b->right, &rtype);
3166 mpq_sub(rv.num, rv.num, right.num);
3169 rv = interp_exec(c, b->left, &rvtype);
3170 right = interp_exec(c, b->right, &rtype);
3171 mpq_mul(rv.num, rv.num, right.num);
3174 rv = interp_exec(c, b->left, &rvtype);
3175 right = interp_exec(c, b->right, &rtype);
3176 mpq_div(rv.num, rv.num, right.num);
3181 left = interp_exec(c, b->left, <ype);
3182 right = interp_exec(c, b->right, &rtype);
3183 mpz_init(l); mpz_init(r); mpz_init(rem);
3184 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3185 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3186 mpz_tdiv_r(rem, l, r);
3187 val_init(Tnum, &rv);
3188 mpq_set_z(rv.num, rem);
3189 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3194 rv = interp_exec(c, b->right, &rvtype);
3195 mpq_neg(rv.num, rv.num);
3198 rv = interp_exec(c, b->right, &rvtype);
3199 mpq_abs(rv.num, rv.num);
3202 rv = interp_exec(c, b->right, &rvtype);
3205 left = interp_exec(c, b->left, <ype);
3206 right = interp_exec(c, b->right, &rtype);
3208 rv.str = text_join(left.str, right.str);
3211 right = interp_exec(c, b->right, &rvtype);
3215 struct text tx = right.str;
3218 if (tx.txt[0] == '-') {
3219 neg = 1; // UNTESTED
3220 tx.txt++; // UNTESTED
3221 tx.len--; // UNTESTED
3223 if (number_parse(rv.num, tail, tx) == 0)
3224 mpq_init(rv.num); // UNTESTED
3226 mpq_neg(rv.num, rv.num); // UNTESTED
3228 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3232 ###### value functions
3234 static struct text text_join(struct text a, struct text b)
3237 rv.len = a.len + b.len;
3238 rv.txt = malloc(rv.len);
3239 memcpy(rv.txt, a.txt, a.len);
3240 memcpy(rv.txt+a.len, b.txt, b.len);
3244 ### Blocks, Statements, and Statement lists.
3246 Now that we have expressions out of the way we need to turn to
3247 statements. There are simple statements and more complex statements.
3248 Simple statements do not contain (syntactic) newlines, complex statements do.
3250 Statements often come in sequences and we have corresponding simple
3251 statement lists and complex statement lists.
3252 The former comprise only simple statements separated by semicolons.
3253 The later comprise complex statements and simple statement lists. They are
3254 separated by newlines. Thus the semicolon is only used to separate
3255 simple statements on the one line. This may be overly restrictive,
3256 but I'm not sure I ever want a complex statement to share a line with
3259 Note that a simple statement list can still use multiple lines if
3260 subsequent lines are indented, so
3262 ###### Example: wrapped simple statement list
3267 is a single simple statement list. This might allow room for
3268 confusion, so I'm not set on it yet.
3270 A simple statement list needs no extra syntax. A complex statement
3271 list has two syntactic forms. It can be enclosed in braces (much like
3272 C blocks), or it can be introduced by an indent and continue until an
3273 unindented newline (much like Python blocks). With this extra syntax
3274 it is referred to as a block.
3276 Note that a block does not have to include any newlines if it only
3277 contains simple statements. So both of:
3279 if condition: a=b; d=f
3281 if condition { a=b; print f }
3285 In either case the list is constructed from a `binode` list with
3286 `Block` as the operator. When parsing the list it is most convenient
3287 to append to the end, so a list is a list and a statement. When using
3288 the list it is more convenient to consider a list to be a statement
3289 and a list. So we need a function to re-order a list.
3290 `reorder_bilist` serves this purpose.
3292 The only stand-alone statement we introduce at this stage is `pass`
3293 which does nothing and is represented as a `NULL` pointer in a `Block`
3294 list. Other stand-alone statements will follow once the infrastructure
3305 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3306 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3307 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3308 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3309 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3311 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3312 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3313 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3314 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3315 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3317 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3318 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3319 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3321 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3322 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3323 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3324 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3325 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3327 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3329 ComplexStatements -> ComplexStatements ComplexStatement ${
3339 | ComplexStatement ${
3351 ComplexStatement -> SimpleStatements Newlines ${
3352 $0 = reorder_bilist($<SS);
3354 | SimpleStatements ; Newlines ${
3355 $0 = reorder_bilist($<SS);
3357 ## ComplexStatement Grammar
3360 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3366 | SimpleStatement ${
3374 SimpleStatement -> pass ${ $0 = NULL; }$
3375 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3376 ## SimpleStatement Grammar
3378 ###### print binode cases
3382 if (b->left == NULL) // UNTESTED
3383 printf("pass"); // UNTESTED
3385 print_exec(b->left, indent, bracket); // UNTESTED
3386 if (b->right) { // UNTESTED
3387 printf("; "); // UNTESTED
3388 print_exec(b->right, indent, bracket); // UNTESTED
3391 // block, one per line
3392 if (b->left == NULL)
3393 do_indent(indent, "pass\n");
3395 print_exec(b->left, indent, bracket);
3397 print_exec(b->right, indent, bracket);
3401 ###### propagate binode cases
3404 /* If any statement returns something other than Tnone
3405 * or Tbool then all such must return same type.
3406 * As each statement may be Tnone or something else,
3407 * we must always pass NULL (unknown) down, otherwise an incorrect
3408 * error might occur. We never return Tnone unless it is
3413 for (e = b; e; e = cast(binode, e->right)) {
3414 t = propagate_types(e->left, c, ok, NULL, rules);
3415 if ((rules & Rboolok) && t == Tbool)
3417 if (t && t != Tnone && t != Tbool) {
3421 type_err(c, "error: expected %1%r, found %2",
3422 e->left, type, rules, t);
3428 ###### interp binode cases
3430 while (rvtype == Tnone &&
3433 rv = interp_exec(c, b->left, &rvtype);
3434 b = cast(binode, b->right);
3438 ### The Print statement
3440 `print` is a simple statement that takes a comma-separated list of
3441 expressions and prints the values separated by spaces and terminated
3442 by a newline. No control of formatting is possible.
3444 `print` faces the same list-ordering issue as blocks, and uses the
3450 ##### expr precedence
3453 ###### SimpleStatement Grammar
3455 | print ExpressionList ${
3456 $0 = reorder_bilist($<2);
3458 | print ExpressionList , ${
3463 $0 = reorder_bilist($0);
3474 ExpressionList -> ExpressionList , Expression ${
3487 ###### print binode cases
3490 do_indent(indent, "print");
3494 print_exec(b->left, -1, bracket);
3498 b = cast(binode, b->right);
3504 ###### propagate binode cases
3507 /* don't care but all must be consistent */
3508 propagate_types(b->left, c, ok, NULL, Rnolabel);
3509 propagate_types(b->right, c, ok, NULL, Rnolabel);
3512 ###### interp binode cases
3518 for ( ; b; b = cast(binode, b->right))
3522 left = interp_exec(c, b->left, <ype);
3523 print_value(ltype, &left);
3524 free_value(ltype, &left);
3535 ###### Assignment statement
3537 An assignment will assign a value to a variable, providing it hasn't
3538 been declared as a constant. The analysis phase ensures that the type
3539 will be correct so the interpreter just needs to perform the
3540 calculation. There is a form of assignment which declares a new
3541 variable as well as assigning a value. If a name is assigned before
3542 it is declared, and error will be raised as the name is created as
3543 `Tlabel` and it is illegal to assign to such names.
3549 ###### declare terminals
3552 ###### SimpleStatement Grammar
3553 | Variable = Expression ${
3559 | VariableDecl = Expression ${
3567 if ($1->var->where_set == NULL) {
3569 "Variable declared with no type or value: %v",
3579 ###### print binode cases
3582 do_indent(indent, "");
3583 print_exec(b->left, indent, bracket);
3585 print_exec(b->right, indent, bracket);
3592 struct variable *v = cast(var, b->left)->var;
3593 do_indent(indent, "");
3594 print_exec(b->left, indent, bracket);
3595 if (cast(var, b->left)->var->constant) {
3597 if (v->where_decl == v->where_set) {
3598 type_print(v->type, stdout);
3603 if (v->where_decl == v->where_set) {
3604 type_print(v->type, stdout);
3610 print_exec(b->right, indent, bracket);
3617 ###### propagate binode cases
3621 /* Both must match and not be labels,
3622 * Type must support 'dup',
3623 * For Assign, left must not be constant.
3626 t = propagate_types(b->left, c, ok, NULL,
3627 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3632 if (propagate_types(b->right, c, ok, t, 0) != t)
3633 if (b->left->type == Xvar)
3634 type_err(c, "info: variable '%v' was set as %1 here.",
3635 cast(var, b->left)->var->where_set, t, rules, NULL);
3637 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3639 propagate_types(b->left, c, ok, t,
3640 (b->op == Assign ? Rnoconstant : 0));
3642 if (t && t->dup == NULL)
3643 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3648 ###### interp binode cases
3651 lleft = linterp_exec(c, b->left, <ype);
3652 right = interp_exec(c, b->right, &rtype);
3654 free_value(ltype, lleft);
3655 dup_value(ltype, &right, lleft);
3662 struct variable *v = cast(var, b->left)->var;
3665 val = var_value(c, v);
3666 free_value(v->type, val);
3667 if (v->type->prepare_type)
3668 v->type->prepare_type(c, v->type, 0);
3670 right = interp_exec(c, b->right, &rtype);
3671 memcpy(val, &right, rtype->size);
3674 val_init(v->type, val);
3679 ### The `use` statement
3681 The `use` statement is the last "simple" statement. It is needed when
3682 the condition in a conditional statement is a block. `use` works much
3683 like `return` in C, but only completes the `condition`, not the whole
3689 ###### expr precedence
3692 ###### SimpleStatement Grammar
3694 $0 = new_pos(binode, $1);
3697 if ($0->right->type == Xvar) {
3698 struct var *v = cast(var, $0->right);
3699 if (v->var->type == Tnone) {
3700 /* Convert this to a label */
3703 v->var->type = Tlabel;
3704 val = global_alloc(c, Tlabel, v->var, NULL);
3710 ###### print binode cases
3713 do_indent(indent, "use ");
3714 print_exec(b->right, -1, bracket);
3719 ###### propagate binode cases
3722 /* result matches value */
3723 return propagate_types(b->right, c, ok, type, 0);
3725 ###### interp binode cases
3728 rv = interp_exec(c, b->right, &rvtype);
3731 ### The Conditional Statement
3733 This is the biggy and currently the only complex statement. This
3734 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3735 It is comprised of a number of parts, all of which are optional though
3736 set combinations apply. Each part is (usually) a key word (`then` is
3737 sometimes optional) followed by either an expression or a code block,
3738 except the `casepart` which is a "key word and an expression" followed
3739 by a code block. The code-block option is valid for all parts and,
3740 where an expression is also allowed, the code block can use the `use`
3741 statement to report a value. If the code block does not report a value
3742 the effect is similar to reporting `True`.
3744 The `else` and `case` parts, as well as `then` when combined with
3745 `if`, can contain a `use` statement which will apply to some
3746 containing conditional statement. `for` parts, `do` parts and `then`
3747 parts used with `for` can never contain a `use`, except in some
3748 subordinate conditional statement.
3750 If there is a `forpart`, it is executed first, only once.
3751 If there is a `dopart`, then it is executed repeatedly providing
3752 always that the `condpart` or `cond`, if present, does not return a non-True
3753 value. `condpart` can fail to return any value if it simply executes
3754 to completion. This is treated the same as returning `True`.
3756 If there is a `thenpart` it will be executed whenever the `condpart`
3757 or `cond` returns True (or does not return any value), but this will happen
3758 *after* `dopart` (when present).
3760 If `elsepart` is present it will be executed at most once when the
3761 condition returns `False` or some value that isn't `True` and isn't
3762 matched by any `casepart`. If there are any `casepart`s, they will be
3763 executed when the condition returns a matching value.
3765 The particular sorts of values allowed in case parts has not yet been
3766 determined in the language design, so nothing is prohibited.
3768 The various blocks in this complex statement potentially provide scope
3769 for variables as described earlier. Each such block must include the
3770 "OpenScope" nonterminal before parsing the block, and must call
3771 `var_block_close()` when closing the block.
3773 The code following "`if`", "`switch`" and "`for`" does not get its own
3774 scope, but is in a scope covering the whole statement, so names
3775 declared there cannot be redeclared elsewhere. Similarly the
3776 condition following "`while`" is in a scope the covers the body
3777 ("`do`" part) of the loop, and which does not allow conditional scope
3778 extension. Code following "`then`" (both looping and non-looping),
3779 "`else`" and "`case`" each get their own local scope.
3781 The type requirements on the code block in a `whilepart` are quite
3782 unusal. It is allowed to return a value of some identifiable type, in
3783 which case the loop aborts and an appropriate `casepart` is run, or it
3784 can return a Boolean, in which case the loop either continues to the
3785 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3786 This is different both from the `ifpart` code block which is expected to
3787 return a Boolean, or the `switchpart` code block which is expected to
3788 return the same type as the casepart values. The correct analysis of
3789 the type of the `whilepart` code block is the reason for the
3790 `Rboolok` flag which is passed to `propagate_types()`.
3792 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3793 defined. As there are two scopes which cover multiple parts - one for
3794 the whole statement and one for "while" and "do" - and as we will use
3795 the 'struct exec' to track scopes, we actually need two new types of
3796 exec. One is a `binode` for the looping part, the rest is the
3797 `cond_statement`. The `cond_statement` will use an auxilliary `struct
3798 casepart` to track a list of case parts.
3809 struct exec *action;
3810 struct casepart *next;
3812 struct cond_statement {
3814 struct exec *forpart, *condpart, *thenpart, *elsepart;
3815 struct binode *looppart;
3816 struct casepart *casepart;
3819 ###### ast functions
3821 static void free_casepart(struct casepart *cp)
3825 free_exec(cp->value);
3826 free_exec(cp->action);
3833 static void free_cond_statement(struct cond_statement *s)
3837 free_exec(s->forpart);
3838 free_exec(s->condpart);
3839 free_exec(s->looppart);
3840 free_exec(s->thenpart);
3841 free_exec(s->elsepart);
3842 free_casepart(s->casepart);
3846 ###### free exec cases
3847 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3849 ###### ComplexStatement Grammar
3850 | CondStatement ${ $0 = $<1; }$
3852 ###### expr precedence
3853 $TERM for then while do
3860 // A CondStatement must end with EOL, as does CondSuffix and
3862 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
3863 // may or may not end with EOL
3864 // WhilePart and IfPart include an appropriate Suffix
3866 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
3867 // them. WhilePart opens and closes its own scope.
3868 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
3871 $0->thenpart = $<TP;
3872 $0->looppart = $<WP;
3873 var_block_close(c, CloseSequential);
3875 | ForPart OptNL WhilePart CondSuffix ${
3878 $0->looppart = $<WP;
3879 var_block_close(c, CloseSequential);
3881 | WhilePart CondSuffix ${
3883 $0->looppart = $<WP;
3885 | SwitchPart OptNL CasePart CondSuffix ${
3887 $0->condpart = $<SP;
3888 $CP->next = $0->casepart;
3889 $0->casepart = $<CP;
3890 var_block_close(c, CloseSequential);
3892 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
3894 $0->condpart = $<SP;
3895 $CP->next = $0->casepart;
3896 $0->casepart = $<CP;
3897 var_block_close(c, CloseSequential);
3899 | IfPart IfSuffix ${
3901 $0->condpart = $IP.condpart; $IP.condpart = NULL;
3902 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
3903 // This is where we close an "if" statement
3904 var_block_close(c, CloseSequential);
3907 CondSuffix -> IfSuffix ${
3910 | Newlines CasePart CondSuffix ${
3912 $CP->next = $0->casepart;
3913 $0->casepart = $<CP;
3915 | CasePart CondSuffix ${
3917 $CP->next = $0->casepart;
3918 $0->casepart = $<CP;
3921 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
3922 | Newlines ElsePart ${ $0 = $<EP; }$
3923 | ElsePart ${$0 = $<EP; }$
3925 ElsePart -> else OpenBlock Newlines ${
3926 $0 = new(cond_statement);
3927 $0->elsepart = $<OB;
3928 var_block_close(c, CloseElse);
3930 | else OpenScope CondStatement ${
3931 $0 = new(cond_statement);
3932 $0->elsepart = $<CS;
3933 var_block_close(c, CloseElse);
3937 CasePart -> case Expression OpenScope ColonBlock ${
3938 $0 = calloc(1,sizeof(struct casepart));
3941 var_block_close(c, CloseParallel);
3945 // These scopes are closed in CondStatement
3946 ForPart -> for OpenBlock ${
3950 ThenPart -> then OpenBlock ${
3952 var_block_close(c, CloseSequential);
3956 // This scope is closed in CondStatement
3957 WhilePart -> while UseBlock OptNL do OpenBlock ${
3962 var_block_close(c, CloseSequential);
3963 var_block_close(c, CloseSequential);
3965 | while OpenScope Expression OpenScope ColonBlock ${
3970 var_block_close(c, CloseSequential);
3971 var_block_close(c, CloseSequential);
3975 IfPart -> if UseBlock OptNL then OpenBlock ${
3978 var_block_close(c, CloseParallel);
3980 | if OpenScope Expression OpenScope ColonBlock ${
3983 var_block_close(c, CloseParallel);
3985 | if OpenScope Expression OpenScope OptNL then Block ${
3988 var_block_close(c, CloseParallel);
3992 // This scope is closed in CondStatement
3993 SwitchPart -> switch OpenScope Expression ${
3996 | switch UseBlock ${
4000 ###### print binode cases
4002 if (b->left && b->left->type == Xbinode &&
4003 cast(binode, b->left)->op == Block) {
4005 do_indent(indent, "while {\n");
4007 do_indent(indent, "while\n");
4008 print_exec(b->left, indent+1, bracket);
4010 do_indent(indent, "} do {\n");
4012 do_indent(indent, "do\n");
4013 print_exec(b->right, indent+1, bracket);
4015 do_indent(indent, "}\n");
4017 do_indent(indent, "while ");
4018 print_exec(b->left, 0, bracket);
4023 print_exec(b->right, indent+1, bracket);
4025 do_indent(indent, "}\n");
4029 ###### print exec cases
4031 case Xcond_statement:
4033 struct cond_statement *cs = cast(cond_statement, e);
4034 struct casepart *cp;
4036 do_indent(indent, "for");
4037 if (bracket) printf(" {\n"); else printf("\n");
4038 print_exec(cs->forpart, indent+1, bracket);
4041 do_indent(indent, "} then {\n");
4043 do_indent(indent, "then\n");
4044 print_exec(cs->thenpart, indent+1, bracket);
4046 if (bracket) do_indent(indent, "}\n");
4049 print_exec(cs->looppart, indent, bracket);
4053 do_indent(indent, "switch");
4055 do_indent(indent, "if");
4056 if (cs->condpart && cs->condpart->type == Xbinode &&
4057 cast(binode, cs->condpart)->op == Block) {
4062 print_exec(cs->condpart, indent+1, bracket);
4064 do_indent(indent, "}\n");
4066 do_indent(indent, "then\n");
4067 print_exec(cs->thenpart, indent+1, bracket);
4071 print_exec(cs->condpart, 0, bracket);
4077 print_exec(cs->thenpart, indent+1, bracket);
4079 do_indent(indent, "}\n");
4084 for (cp = cs->casepart; cp; cp = cp->next) {
4085 do_indent(indent, "case ");
4086 print_exec(cp->value, -1, 0);
4091 print_exec(cp->action, indent+1, bracket);
4093 do_indent(indent, "}\n");
4096 do_indent(indent, "else");
4101 print_exec(cs->elsepart, indent+1, bracket);
4103 do_indent(indent, "}\n");
4108 ###### propagate binode cases
4110 t = propagate_types(b->right, c, ok, Tnone, 0);
4111 if (!type_compat(Tnone, t, 0))
4112 *ok = 0; // UNTESTED
4113 return propagate_types(b->left, c, ok, type, rules);
4115 ###### propagate exec cases
4116 case Xcond_statement:
4118 // forpart and looppart->right must return Tnone
4119 // thenpart must return Tnone if there is a loopart,
4120 // otherwise it is like elsepart.
4122 // be bool if there is no casepart
4123 // match casepart->values if there is a switchpart
4124 // either be bool or match casepart->value if there
4126 // elsepart and casepart->action must match the return type
4127 // expected of this statement.
4128 struct cond_statement *cs = cast(cond_statement, prog);
4129 struct casepart *cp;
4131 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4132 if (!type_compat(Tnone, t, 0))
4133 *ok = 0; // UNTESTED
4136 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4137 if (!type_compat(Tnone, t, 0))
4138 *ok = 0; // UNTESTED
4140 if (cs->casepart == NULL) {
4141 propagate_types(cs->condpart, c, ok, Tbool, 0);
4142 propagate_types(cs->looppart, c, ok, Tbool, 0);
4144 /* Condpart must match case values, with bool permitted */
4146 for (cp = cs->casepart;
4147 cp && !t; cp = cp->next)
4148 t = propagate_types(cp->value, c, ok, NULL, 0);
4149 if (!t && cs->condpart)
4150 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4151 if (!t && cs->looppart)
4152 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4153 // Now we have a type (I hope) push it down
4155 for (cp = cs->casepart; cp; cp = cp->next)
4156 propagate_types(cp->value, c, ok, t, 0);
4157 propagate_types(cs->condpart, c, ok, t, Rboolok);
4158 propagate_types(cs->looppart, c, ok, t, Rboolok);
4161 // (if)then, else, and case parts must return expected type.
4162 if (!cs->looppart && !type)
4163 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4165 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4166 for (cp = cs->casepart;
4168 cp = cp->next) // UNTESTED
4169 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4172 propagate_types(cs->thenpart, c, ok, type, rules);
4173 propagate_types(cs->elsepart, c, ok, type, rules);
4174 for (cp = cs->casepart; cp ; cp = cp->next)
4175 propagate_types(cp->action, c, ok, type, rules);
4181 ###### interp binode cases
4183 // This just performs one iterration of the loop
4184 rv = interp_exec(c, b->left, &rvtype);
4185 if (rvtype == Tnone ||
4186 (rvtype == Tbool && rv.bool != 0))
4187 // cnd is Tnone or Tbool, doesn't need to be freed
4188 interp_exec(c, b->right, NULL);
4191 ###### interp exec cases
4192 case Xcond_statement:
4194 struct value v, cnd;
4195 struct type *vtype, *cndtype;
4196 struct casepart *cp;
4197 struct cond_statement *cs = cast(cond_statement, e);
4200 interp_exec(c, cs->forpart, NULL);
4202 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4203 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4204 interp_exec(c, cs->thenpart, NULL);
4206 cnd = interp_exec(c, cs->condpart, &cndtype);
4207 if ((cndtype == Tnone ||
4208 (cndtype == Tbool && cnd.bool != 0))) {
4209 // cnd is Tnone or Tbool, doesn't need to be freed
4210 rv = interp_exec(c, cs->thenpart, &rvtype);
4211 // skip else (and cases)
4215 for (cp = cs->casepart; cp; cp = cp->next) {
4216 v = interp_exec(c, cp->value, &vtype);
4217 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4218 free_value(vtype, &v);
4219 free_value(cndtype, &cnd);
4220 rv = interp_exec(c, cp->action, &rvtype);
4223 free_value(vtype, &v);
4225 free_value(cndtype, &cnd);
4227 rv = interp_exec(c, cs->elsepart, &rvtype);
4234 ### Top level structure
4236 All the language elements so far can be used in various places. Now
4237 it is time to clarify what those places are.
4239 At the top level of a file there will be a number of declarations.
4240 Many of the things that can be declared haven't been described yet,
4241 such as functions, procedures, imports, and probably more.
4242 For now there are two sorts of things that can appear at the top
4243 level. They are predefined constants, `struct` types, and the `main`
4244 function. While the syntax will allow the `main` function to appear
4245 multiple times, that will trigger an error if it is actually attempted.
4247 The various declarations do not return anything. They store the
4248 various declarations in the parse context.
4250 ###### Parser: grammar
4253 Ocean -> OptNL DeclarationList
4255 ## declare terminals
4262 DeclarationList -> Declaration
4263 | DeclarationList Declaration
4265 Declaration -> ERROR Newlines ${
4266 tok_err(c, // UNTESTED
4267 "error: unhandled parse error", &$1);
4273 ## top level grammar
4277 ### The `const` section
4279 As well as being defined in with the code that uses them, constants
4280 can be declared at the top level. These have full-file scope, so they
4281 are always `InScope`. The value of a top level constant can be given
4282 as an expression, and this is evaluated immediately rather than in the
4283 later interpretation stage. Once we add functions to the language, we
4284 will need rules concern which, if any, can be used to define a top
4287 Constants are defined in a section that starts with the reserved word
4288 `const` and then has a block with a list of assignment statements.
4289 For syntactic consistency, these must use the double-colon syntax to
4290 make it clear that they are constants. Type can also be given: if
4291 not, the type will be determined during analysis, as with other
4294 As the types constants are inserted at the head of a list, printing
4295 them in the same order that they were read is not straight forward.
4296 We take a quadratic approach here and count the number of constants
4297 (variables of depth 0), then count down from there, each time
4298 searching through for the Nth constant for decreasing N.
4300 ###### top level grammar
4304 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4305 | const { SimpleConstList } Newlines
4306 | const IN OptNL ConstList OUT Newlines
4307 | const SimpleConstList Newlines
4309 ConstList -> ConstList SimpleConstLine
4311 SimpleConstList -> SimpleConstList ; Const
4314 SimpleConstLine -> SimpleConstList Newlines
4315 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4318 CType -> Type ${ $0 = $<1; }$
4321 Const -> IDENTIFIER :: CType = Expression ${ {
4325 v = var_decl(c, $1.txt);
4327 struct var *var = new_pos(var, $1);
4328 v->where_decl = var;
4333 v = var_ref(c, $1.txt);
4334 tok_err(c, "error: name already declared", &$1);
4335 type_err(c, "info: this is where '%v' was first declared",
4336 v->where_decl, NULL, 0, NULL);
4340 propagate_types($5, c, &ok, $3, 0);
4345 struct value res = interp_exec(c, $5, &v->type);
4346 global_alloc(c, v->type, v, &res);
4350 ###### print const decls
4355 while (target != 0) {
4357 for (v = context.in_scope; v; v=v->in_scope)
4358 if (v->depth == 0) {
4369 struct value *val = var_value(&context, v);
4370 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4371 type_print(v->type, stdout);
4373 if (v->type == Tstr)
4375 print_value(v->type, val);
4376 if (v->type == Tstr)
4384 ### Finally the whole `main` function.
4386 An Ocean program can currently have only one function - `main` - and
4387 that must exist. It expects an array of strings with a provided size.
4388 Following this is a `block` which is the code to execute.
4390 As this is the top level, several things are handled a bit
4392 The function is not interpreted by `interp_exec` as that isn't
4393 passed the argument list which the program requires. Similarly type
4394 analysis is a bit more interesting at this level.
4396 ###### top level grammar
4398 DeclareFunction -> MainFunction ${ {
4400 type_err(c, "\"main\" defined a second time",
4406 ###### print binode cases
4409 do_indent(indent, "func main(");
4410 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4411 struct variable *v = cast(var, b2->left)->var;
4413 print_exec(b2->left, 0, 0);
4415 type_print(v->type, stdout);
4421 print_exec(b->right, indent+1, bracket);
4423 do_indent(indent, "}\n");
4426 ###### propagate binode cases
4428 case Func: abort(); // NOTEST
4430 ###### core functions
4432 static int analyse_prog(struct exec *prog, struct parse_context *c)
4434 struct binode *bp = cast(binode, prog);
4438 struct type *argv_type;
4439 struct text argv_type_name = { " argv", 5 };
4444 argv_type = add_type(c, argv_type_name, &array_prototype);
4445 argv_type->array.member = Tstr;
4446 argv_type->array.unspec = 1;
4448 for (b = cast(binode, bp->left); b; b = cast(binode, b->right)) {
4452 propagate_types(b->left, c, &ok, argv_type, 0);
4454 default: /* invalid */ // NOTEST
4455 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4461 propagate_types(bp->right, c, &ok, Tnone, 0);
4466 /* Make sure everything is still consistent */
4467 propagate_types(bp->right, c, &ok, Tnone, 0);
4469 return 0; // UNTESTED
4474 static void interp_prog(struct parse_context *c, struct exec *prog,
4475 int argc, char **argv)
4477 struct binode *p = cast(binode, prog);
4485 al = cast(binode, p->left);
4487 struct var *v = cast(var, al->left);
4488 struct value *vl = var_value(c, v->var);
4498 mpq_set_ui(argcq, argc, 1);
4499 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4500 t->prepare_type(c, t, 0);
4501 array_init(v->var->type, vl);
4502 for (i = 0; i < argc; i++) {
4503 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4506 arg.str.txt = argv[i];
4507 arg.str.len = strlen(argv[i]);
4508 free_value(Tstr, vl2);
4509 dup_value(Tstr, &arg, vl2);
4513 al = cast(binode, al->right);
4515 v = interp_exec(c, p, &vtype);
4516 free_value(vtype, &v);
4519 ###### interp binode cases
4520 case List: abort(); // NOTEST
4523 rv = interp_exec(c, b->right, &rvtype);
4526 ## And now to test it out.
4528 Having a language requires having a "hello world" program. I'll
4529 provide a little more than that: a program that prints "Hello world"
4530 finds the GCD of two numbers, prints the first few elements of
4531 Fibonacci, performs a binary search for a number, and a few other
4532 things which will likely grow as the languages grows.
4534 ###### File: oceani.mk
4537 @echo "===== DEMO ====="
4538 ./oceani --section "demo: hello" oceani.mdc 55 33
4544 four ::= 2 + 2 ; five ::= 10/2
4545 const pie ::= "I like Pie";
4546 cake ::= "The cake is"
4557 print "Hello World, what lovely oceans you have!"
4558 print "Are there", five, "?"
4559 print pi, pie, "but", cake
4561 A := $argv[1]; B := $argv[2]
4563 /* When a variable is defined in both branches of an 'if',
4564 * and used afterwards, the variables are merged.
4570 print "Is", A, "bigger than", B,"? ", bigger
4571 /* If a variable is not used after the 'if', no
4572 * merge happens, so types can be different
4575 double:string = "yes"
4576 print A, "is more than twice", B, "?", double
4579 print "double", B, "is", double
4584 if a > 0 and then b > 0:
4590 print "GCD of", A, "and", B,"is", a
4592 print a, "is not positive, cannot calculate GCD"
4594 print b, "is not positive, cannot calculate GCD"
4599 print "Fibonacci:", f1,f2,
4600 then togo = togo - 1
4608 /* Binary search... */
4613 mid := (lo + hi) / 2
4626 print "Yay, I found", target
4628 print "Closest I found was", lo
4633 // "middle square" PRNG. Not particularly good, but one my
4634 // Dad taught me - the first one I ever heard of.
4635 for i:=1; then i = i + 1; while i < size:
4636 n := list[i-1] * list[i-1]
4637 list[i] = (n / 100) % 10 000
4639 print "Before sort:",
4640 for i:=0; then i = i + 1; while i < size:
4644 for i := 1; then i=i+1; while i < size:
4645 for j:=i-1; then j=j-1; while j >= 0:
4646 if list[j] > list[j+1]:
4650 print " After sort:",
4651 for i:=0; then i = i + 1; while i < size:
4655 if 1 == 2 then print "yes"; else print "no"
4659 bob.alive = (bob.name == "Hello")
4660 print "bob", "is" if bob.alive else "isn't", "alive"