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";
169 int main(int argc, char *argv[])
174 struct section *s, *ss;
175 char *section = NULL;
176 struct parse_context context = {
178 .ignored = (1 << TK_mark),
179 .number_chars = ".,_+- ",
184 int doprint=0, dotrace=0, doexec=1, brackets=0;
186 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
189 case 't': dotrace=1; break;
190 case 'p': doprint=1; break;
191 case 'n': doexec=0; break;
192 case 'b': brackets=1; break;
193 case 's': section = optarg; break;
194 default: fprintf(stderr, Usage);
198 if (optind >= argc) {
199 fprintf(stderr, "oceani: no input file given\n");
202 fd = open(argv[optind], O_RDONLY);
204 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
207 context.file_name = argv[optind];
208 len = lseek(fd, 0, 2);
209 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
210 s = code_extract(file, file+len, NULL);
212 fprintf(stderr, "oceani: could not find any code in %s\n",
217 ## context initialization
220 for (ss = s; ss; ss = ss->next) {
221 struct text sec = ss->section;
222 if (sec.len == strlen(section) &&
223 strncmp(sec.txt, section, sec.len) == 0)
227 fprintf(stderr, "oceani: cannot find section %s\n",
233 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
236 fprintf(stderr, "oceani: no main function found.\n");
237 context.parse_error = 1;
239 if (context.prog && doprint) {
242 print_exec(context.prog, 0, brackets);
244 if (context.prog && doexec && !context.parse_error) {
245 if (!analyse_prog(context.prog, &context)) {
246 fprintf(stderr, "oceani: type error in program - not running.\n");
249 interp_prog(&context, context.prog, argc - optind, argv+optind);
251 free_exec(context.prog);
254 struct section *t = s->next;
260 ## free context types
261 exit(context.parse_error ? 1 : 0);
266 The four requirements of parse, analyse, print, interpret apply to
267 each language element individually so that is how most of the code
270 Three of the four are fairly self explanatory. The one that requires
271 a little explanation is the analysis step.
273 The current language design does not require the types of variables to
274 be declared, but they must still have a single type. Different
275 operations impose different requirements on the variables, for example
276 addition requires both arguments to be numeric, and assignment
277 requires the variable on the left to have the same type as the
278 expression on the right.
280 Analysis involves propagating these type requirements around and
281 consequently setting the type of each variable. If any requirements
282 are violated (e.g. a string is compared with a number) or if a
283 variable needs to have two different types, then an error is raised
284 and the program will not run.
286 If the same variable is declared in both branchs of an 'if/else', or
287 in all cases of a 'switch' then the multiple instances may be merged
288 into just one variable if the variable is referenced after the
289 conditional statement. When this happens, the types must naturally be
290 consistent across all the branches. When the variable is not used
291 outside the if, the variables in the different branches are distinct
292 and can be of different types.
294 Undeclared names may only appear in "use" statements and "case" expressions.
295 These names are given a type of "label" and a unique value.
296 This allows them to fill the role of a name in an enumerated type, which
297 is useful for testing the `switch` statement.
299 As we will see, the condition part of a `while` statement can return
300 either a Boolean or some other type. This requires that the expected
301 type that gets passed around comprises a type and a flag to indicate
302 that `Tbool` is also permitted.
304 As there are, as yet, no distinct types that are compatible, there
305 isn't much subtlety in the analysis. When we have distinct number
306 types, this will become more interesting.
310 When analysis discovers an inconsistency it needs to report an error;
311 just refusing to run the code ensures that the error doesn't cascade,
312 but by itself it isn't very useful. A clear understanding of the sort
313 of error message that are useful will help guide the process of
316 At a simplistic level, the only sort of error that type analysis can
317 report is that the type of some construct doesn't match a contextual
318 requirement. For example, in `4 + "hello"` the addition provides a
319 contextual requirement for numbers, but `"hello"` is not a number. In
320 this particular example no further information is needed as the types
321 are obvious from local information. When a variable is involved that
322 isn't the case. It may be helpful to explain why the variable has a
323 particular type, by indicating the location where the type was set,
324 whether by declaration or usage.
326 Using a recursive-descent analysis we can easily detect a problem at
327 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
328 will detect that one argument is not a number and the usage of `hello`
329 will detect that a number was wanted, but not provided. In this
330 (early) version of the language, we will generate error reports at
331 multiple locations, so the use of `hello` will report an error and
332 explain were the value was set, and the addition will report an error
333 and say why numbers are needed. To be able to report locations for
334 errors, each language element will need to record a file location
335 (line and column) and each variable will need to record the language
336 element where its type was set. For now we will assume that each line
337 of an error message indicates one location in the file, and up to 2
338 types. So we provide a `printf`-like function which takes a format, a
339 location (a `struct exec` which has not yet been introduced), and 2
340 types. "`%1`" reports the first type, "`%2`" reports the second. We
341 will need a function to print the location, once we know how that is
342 stored. e As will be explained later, there are sometimes extra rules for
343 type matching and they might affect error messages, we need to pass those
346 As well as type errors, we sometimes need to report problems with
347 tokens, which might be unexpected or might name a type that has not
348 been defined. For these we have `tok_err()` which reports an error
349 with a given token. Each of the error functions sets the flag in the
350 context so indicate that parsing failed.
354 static void fput_loc(struct exec *loc, FILE *f);
356 ###### core functions
358 static void type_err(struct parse_context *c,
359 char *fmt, struct exec *loc,
360 struct type *t1, int rules, struct type *t2)
362 fprintf(stderr, "%s:", c->file_name);
363 fput_loc(loc, stderr);
364 for (; *fmt ; fmt++) {
371 case '%': fputc(*fmt, stderr); break; // NOTEST
372 default: fputc('?', stderr); break; // NOTEST
374 type_print(t1, stderr);
377 type_print(t2, stderr);
386 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
388 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
389 t->txt.len, t->txt.txt);
393 ## Entities: declared and predeclared.
395 There are various "things" that the language and/or the interpreter
396 needs to know about to parse and execute a program. These include
397 types, variables, values, and executable code. These are all lumped
398 together under the term "entities" (calling them "objects" would be
399 confusing) and introduced here. The following section will present the
400 different specific code elements which comprise or manipulate these
405 Values come in a wide range of types, with more likely to be added.
406 Each type needs to be able to print its own values (for convenience at
407 least) as well as to compare two values, at least for equality and
408 possibly for order. For now, values might need to be duplicated and
409 freed, though eventually such manipulations will be better integrated
412 Rather than requiring every numeric type to support all numeric
413 operations (add, multiple, etc), we allow types to be able to present
414 as one of a few standard types: integer, float, and fraction. The
415 existence of these conversion functions eventually enable types to
416 determine if they are compatible with other types, though such types
417 have not yet been implemented.
419 Named type are stored in a simple linked list. Objects of each type are
420 "values" which are often passed around by value.
427 ## value union fields
435 void (*init)(struct type *type, struct value *val);
436 void (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
437 void (*print)(struct type *type, struct value *val);
438 void (*print_type)(struct type *type, FILE *f);
439 int (*cmp_order)(struct type *t1, struct type *t2,
440 struct value *v1, struct value *v2);
441 int (*cmp_eq)(struct type *t1, struct type *t2,
442 struct value *v1, struct value *v2);
443 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
444 void (*free)(struct type *type, struct value *val);
445 void (*free_type)(struct type *t);
446 long long (*to_int)(struct value *v);
447 double (*to_float)(struct value *v);
448 int (*to_mpq)(mpq_t *q, struct value *v);
457 struct type *typelist;
461 static struct type *find_type(struct parse_context *c, struct text s)
463 struct type *l = c->typelist;
466 text_cmp(l->name, s) != 0)
471 static struct type *add_type(struct parse_context *c, struct text s,
476 n = calloc(1, sizeof(*n));
479 n->next = c->typelist;
484 static void free_type(struct type *t)
486 /* The type is always a reference to something in the
487 * context, so we don't need to free anything.
491 static void free_value(struct type *type, struct value *v)
497 static void type_print(struct type *type, FILE *f)
500 fputs("*unknown*type*", f); // NOTEST
501 else if (type->name.len)
502 fprintf(f, "%.*s", type->name.len, type->name.txt);
503 else if (type->print_type)
504 type->print_type(type, f);
506 fputs("*invalid*type*", f); // NOTEST
509 static void val_init(struct type *type, struct value *val)
511 if (type && type->init)
512 type->init(type, val);
515 static void dup_value(struct type *type,
516 struct value *vold, struct value *vnew)
518 if (type && type->dup)
519 type->dup(type, vold, vnew);
522 static int value_cmp(struct type *tl, struct type *tr,
523 struct value *left, struct value *right)
525 if (tl && tl->cmp_order)
526 return tl->cmp_order(tl, tr, left, right);
527 if (tl && tl->cmp_eq)
528 return tl->cmp_eq(tl, tr, left, right);
532 static void print_value(struct type *type, struct value *v)
534 if (type && type->print)
535 type->print(type, v);
537 printf("*Unknown*"); // NOTEST
542 static void free_value(struct type *type, struct value *v);
543 static int type_compat(struct type *require, struct type *have, int rules);
544 static void type_print(struct type *type, FILE *f);
545 static void val_init(struct type *type, struct value *v);
546 static void dup_value(struct type *type,
547 struct value *vold, struct value *vnew);
548 static int value_cmp(struct type *tl, struct type *tr,
549 struct value *left, struct value *right);
550 static void print_value(struct type *type, struct value *v);
552 ###### free context types
554 while (context.typelist) {
555 struct type *t = context.typelist;
557 context.typelist = t->next;
563 Type can be specified for local variables, for fields in a structure,
564 for formal parameters to functions, and possibly elsewhere. Different
565 rules may apply in different contexts. As a minimum, a named type may
566 always be used. Currently the type of a formal parameter can be
567 different from types in other contexts, so we have a separate grammar
573 Type -> IDENTIFIER ${
574 $0 = find_type(c, $1.txt);
577 "error: undefined type", &$1);
584 FormalType -> Type ${ $0 = $<1; }$
585 ## formal type grammar
589 Values of the base types can be numbers, which we represent as
590 multi-precision fractions, strings, Booleans and labels. When
591 analysing the program we also need to allow for places where no value
592 is meaningful (type `Tnone`) and where we don't know what type to
593 expect yet (type is `NULL`).
595 Values are never shared, they are always copied when used, and freed
596 when no longer needed.
598 When propagating type information around the program, we need to
599 determine if two types are compatible, where type `NULL` is compatible
600 with anything. There are two special cases with type compatibility,
601 both related to the Conditional Statement which will be described
602 later. In some cases a Boolean can be accepted as well as some other
603 primary type, and in others any type is acceptable except a label (`Vlabel`).
604 A separate function encoding these cases will simplify some code later.
606 ###### type functions
608 int (*compat)(struct type *this, struct type *other);
612 static int type_compat(struct type *require, struct type *have, int rules)
614 if ((rules & Rboolok) && have == Tbool)
616 if ((rules & Rnolabel) && have == Tlabel)
618 if (!require || !have)
622 return require->compat(require, have);
624 return require == have;
629 #include "parse_string.h"
630 #include "parse_number.h"
633 myLDLIBS := libnumber.o libstring.o -lgmp
634 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
636 ###### type union fields
637 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
639 ###### value union fields
646 static void _free_value(struct type *type, struct value *v)
650 switch (type->vtype) {
652 case Vstr: free(v->str.txt); break;
653 case Vnum: mpq_clear(v->num); break;
659 ###### value functions
661 static void _val_init(struct type *type, struct value *val)
663 switch(type->vtype) {
664 case Vnone: // NOTEST
667 mpq_init(val->num); break;
669 val->str.txt = malloc(1);
681 static void _dup_value(struct type *type,
682 struct value *vold, struct value *vnew)
684 switch (type->vtype) {
685 case Vnone: // NOTEST
688 vnew->label = vold->label;
691 vnew->bool = vold->bool;
695 mpq_set(vnew->num, vold->num);
698 vnew->str.len = vold->str.len;
699 vnew->str.txt = malloc(vnew->str.len);
700 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
705 static int _value_cmp(struct type *tl, struct type *tr,
706 struct value *left, struct value *right)
710 return tl - tr; // NOTEST
712 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
713 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
714 case Vstr: cmp = text_cmp(left->str, right->str); break;
715 case Vbool: cmp = left->bool - right->bool; break;
716 case Vnone: cmp = 0; // NOTEST
721 static void _print_value(struct type *type, struct value *v)
723 switch (type->vtype) {
724 case Vnone: // NOTEST
725 printf("*no-value*"); break; // NOTEST
726 case Vlabel: // NOTEST
727 printf("*label-%p*", v->label); break; // NOTEST
729 printf("%.*s", v->str.len, v->str.txt); break;
731 printf("%s", v->bool ? "True":"False"); break;
736 mpf_set_q(fl, v->num);
737 gmp_printf("%Fg", fl);
744 static void _free_value(struct type *type, struct value *v);
746 static struct type base_prototype = {
748 .print = _print_value,
749 .cmp_order = _value_cmp,
750 .cmp_eq = _value_cmp,
755 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
758 static struct type *add_base_type(struct parse_context *c, char *n,
759 enum vtype vt, int size)
761 struct text txt = { n, strlen(n) };
764 t = add_type(c, txt, &base_prototype);
767 t->align = size > sizeof(void*) ? sizeof(void*) : size;
768 if (t->size & (t->align - 1))
769 t->size = (t->size | (t->align - 1)) + 1;
773 ###### context initialization
775 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
776 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
777 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
778 Tnone = add_base_type(&context, "none", Vnone, 0);
779 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
783 Variables are scoped named values. We store the names in a linked list
784 of "bindings" sorted in lexical order, and use sequential search and
791 struct binding *next; // in lexical order
795 This linked list is stored in the parse context so that "reduce"
796 functions can find or add variables, and so the analysis phase can
797 ensure that every variable gets a type.
801 struct binding *varlist; // In lexical order
805 static struct binding *find_binding(struct parse_context *c, struct text s)
807 struct binding **l = &c->varlist;
812 (cmp = text_cmp((*l)->name, s)) < 0)
816 n = calloc(1, sizeof(*n));
823 Each name can be linked to multiple variables defined in different
824 scopes. Each scope starts where the name is declared and continues
825 until the end of the containing code block. Scopes of a given name
826 cannot nest, so a declaration while a name is in-scope is an error.
828 ###### binding fields
829 struct variable *var;
833 struct variable *previous;
835 struct binding *name;
836 struct exec *where_decl;// where name was declared
837 struct exec *where_set; // where type was set
841 While the naming seems strange, we include local constants in the
842 definition of variables. A name declared `var := value` can
843 subsequently be changed, but a name declared `var ::= value` cannot -
846 ###### variable fields
849 Scopes in parallel branches can be partially merged. More
850 specifically, if a given name is declared in both branches of an
851 if/else then its scope is a candidate for merging. Similarly if
852 every branch of an exhaustive switch (e.g. has an "else" clause)
853 declares a given name, then the scopes from the branches are
854 candidates for merging.
856 Note that names declared inside a loop (which is only parallel to
857 itself) are never visible after the loop. Similarly names defined in
858 scopes which are not parallel, such as those started by `for` and
859 `switch`, are never visible after the scope. Only variables defined in
860 both `then` and `else` (including the implicit then after an `if`, and
861 excluding `then` used with `for`) and in all `case`s and `else` of a
862 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
864 Labels, which are a bit like variables, follow different rules.
865 Labels are not explicitly declared, but if an undeclared name appears
866 in a context where a label is legal, that effectively declares the
867 name as a label. The declaration remains in force (or in scope) at
868 least to the end of the immediately containing block and conditionally
869 in any larger containing block which does not declare the name in some
870 other way. Importantly, the conditional scope extension happens even
871 if the label is only used in one parallel branch of a conditional --
872 when used in one branch it is treated as having been declared in all
875 Merge candidates are tentatively visible beyond the end of the
876 branching statement which creates them. If the name is used, the
877 merge is affirmed and they become a single variable visible at the
878 outer layer. If not - if it is redeclared first - the merge lapses.
880 To track scopes we have an extra stack, implemented as a linked list,
881 which roughly parallels the parse stack and which is used exclusively
882 for scoping. When a new scope is opened, a new frame is pushed and
883 the child-count of the parent frame is incremented. This child-count
884 is used to distinguish between the first of a set of parallel scopes,
885 in which declared variables must not be in scope, and subsequent
886 branches, whether they may already be conditionally scoped.
888 To push a new frame *before* any code in the frame is parsed, we need a
889 grammar reduction. This is most easily achieved with a grammar
890 element which derives the empty string, and creates the new scope when
891 it is recognised. This can be placed, for example, between a keyword
892 like "if" and the code following it.
896 struct scope *parent;
902 struct scope *scope_stack;
905 static void scope_pop(struct parse_context *c)
907 struct scope *s = c->scope_stack;
909 c->scope_stack = s->parent;
914 static void scope_push(struct parse_context *c)
916 struct scope *s = calloc(1, sizeof(*s));
918 c->scope_stack->child_count += 1;
919 s->parent = c->scope_stack;
927 OpenScope -> ${ scope_push(c); }$
928 ClosePara -> ${ var_block_close(c, CloseParallel); }$
930 Each variable records a scope depth and is in one of four states:
932 - "in scope". This is the case between the declaration of the
933 variable and the end of the containing block, and also between
934 the usage with affirms a merge and the end of that block.
936 The scope depth is not greater than the current parse context scope
937 nest depth. When the block of that depth closes, the state will
938 change. To achieve this, all "in scope" variables are linked
939 together as a stack in nesting order.
941 - "pending". The "in scope" block has closed, but other parallel
942 scopes are still being processed. So far, every parallel block at
943 the same level that has closed has declared the name.
945 The scope depth is the depth of the last parallel block that
946 enclosed the declaration, and that has closed.
948 - "conditionally in scope". The "in scope" block and all parallel
949 scopes have closed, and no further mention of the name has been
950 seen. This state includes a secondary nest depth which records the
951 outermost scope seen since the variable became conditionally in
952 scope. If a use of the name is found, the variable becomes "in
953 scope" and that secondary depth becomes the recorded scope depth.
954 If the name is declared as a new variable, the old variable becomes
955 "out of scope" and the recorded scope depth stays unchanged.
957 - "out of scope". The variable is neither in scope nor conditionally
958 in scope. It is permanently out of scope now and can be removed from
959 the "in scope" stack.
961 ###### variable fields
962 int depth, min_depth;
963 enum { OutScope, PendingScope, CondScope, InScope } scope;
964 struct variable *in_scope;
968 struct variable *in_scope;
970 All variables with the same name are linked together using the
971 'previous' link. Those variable that have been affirmatively merged all
972 have a 'merged' pointer that points to one primary variable - the most
973 recently declared instance. When merging variables, we need to also
974 adjust the 'merged' pointer on any other variables that had previously
975 been merged with the one that will no longer be primary.
977 A variable that is no longer the most recent instance of a name may
978 still have "pending" scope, if it might still be merged with most
979 recent instance. These variables don't really belong in the
980 "in_scope" list, but are not immediately removed when a new instance
981 is found. Instead, they are detected and ignored when considering the
982 list of in_scope names.
984 The storage of the value of a variable will be described later. For now
985 we just need to know that when a variable goes out of scope, it might
986 need to be freed. For this we need to be able to find it, so assume that
987 `var_value()` will provide that.
989 ###### variable fields
990 struct variable *merged;
994 static void variable_merge(struct variable *primary, struct variable *secondary)
1000 primary = primary->merged; // NOTEST
1002 for (v = primary->previous; v; v=v->previous)
1003 if (v == secondary || v == secondary->merged ||
1004 v->merged == secondary ||
1005 (v->merged && v->merged == secondary->merged)) {
1006 v->scope = OutScope;
1007 v->merged = primary;
1011 ###### forward decls
1012 static struct value *var_value(struct parse_context *c, struct variable *v);
1014 ###### free context vars
1016 while (context.varlist) {
1017 struct binding *b = context.varlist;
1018 struct variable *v = b->var;
1019 context.varlist = b->next;
1022 struct variable *t = v;
1025 free_value(t->type, var_value(&context, t));
1027 // This is a global constant
1028 free_exec(t->where_decl);
1033 #### Manipulating Bindings
1035 When a name is conditionally visible, a new declaration discards the
1036 old binding - the condition lapses. Conversely a usage of the name
1037 affirms the visibility and extends it to the end of the containing
1038 block - i.e. the block that contains both the original declaration and
1039 the latest usage. This is determined from `min_depth`. When a
1040 conditionally visible variable gets affirmed like this, it is also
1041 merged with other conditionally visible variables with the same name.
1043 When we parse a variable declaration we either report an error if the
1044 name is currently bound, or create a new variable at the current nest
1045 depth if the name is unbound or bound to a conditionally scoped or
1046 pending-scope variable. If the previous variable was conditionally
1047 scoped, it and its homonyms becomes out-of-scope.
1049 When we parse a variable reference (including non-declarative assignment
1050 "foo = bar") we report an error if the name is not bound or is bound to
1051 a pending-scope variable; update the scope if the name is bound to a
1052 conditionally scoped variable; or just proceed normally if the named
1053 variable is in scope.
1055 When we exit a scope, any variables bound at this level are either
1056 marked out of scope or pending-scoped, depending on whether the scope
1057 was sequential or parallel. Here a "parallel" scope means the "then"
1058 or "else" part of a conditional, or any "case" or "else" branch of a
1059 switch. Other scopes are "sequential".
1061 When exiting a parallel scope we check if there are any variables that
1062 were previously pending and are still visible. If there are, then
1063 there weren't redeclared in the most recent scope, so they cannot be
1064 merged and must become out-of-scope. If it is not the first of
1065 parallel scopes (based on `child_count`), we check that there was a
1066 previous binding that is still pending-scope. If there isn't, the new
1067 variable must now be out-of-scope.
1069 When exiting a sequential scope that immediately enclosed parallel
1070 scopes, we need to resolve any pending-scope variables. If there was
1071 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1072 we need to mark all pending-scope variable as out-of-scope. Otherwise
1073 all pending-scope variables become conditionally scoped.
1076 enum closetype { CloseSequential, CloseParallel, CloseElse };
1078 ###### ast functions
1080 static struct variable *var_decl(struct parse_context *c, struct text s)
1082 struct binding *b = find_binding(c, s);
1083 struct variable *v = b->var;
1085 switch (v ? v->scope : OutScope) {
1087 /* Caller will report the error */
1091 v && v->scope == CondScope;
1093 v->scope = OutScope;
1097 v = calloc(1, sizeof(*v));
1098 v->previous = b->var;
1101 v->min_depth = v->depth = c->scope_depth;
1103 v->in_scope = c->in_scope;
1108 static struct variable *var_ref(struct parse_context *c, struct text s)
1110 struct binding *b = find_binding(c, s);
1111 struct variable *v = b->var;
1112 struct variable *v2;
1114 switch (v ? v->scope : OutScope) {
1117 /* Caller will report the error */
1120 /* All CondScope variables of this name need to be merged
1121 * and become InScope
1123 v->depth = v->min_depth;
1125 for (v2 = v->previous;
1126 v2 && v2->scope == CondScope;
1128 variable_merge(v, v2);
1136 static void var_block_close(struct parse_context *c, enum closetype ct)
1138 /* Close off all variables that are in_scope */
1139 struct variable *v, **vp, *v2;
1142 for (vp = &c->in_scope;
1143 v = *vp, v && v->depth > c->scope_depth && v->min_depth > c->scope_depth;
1145 if (v->name->var == v) switch (ct) {
1147 case CloseParallel: /* handle PendingScope */
1151 if (c->scope_stack->child_count == 1)
1152 v->scope = PendingScope;
1153 else if (v->previous &&
1154 v->previous->scope == PendingScope)
1155 v->scope = PendingScope;
1156 else if (v->type == Tlabel)
1157 v->scope = PendingScope;
1158 else if (v->name->var == v)
1159 v->scope = OutScope;
1160 if (ct == CloseElse) {
1161 /* All Pending variables with this name
1162 * are now Conditional */
1164 v2 && v2->scope == PendingScope;
1166 v2->scope = CondScope;
1171 v2 && v2->scope == PendingScope;
1173 if (v2->type != Tlabel)
1174 v2->scope = OutScope;
1176 case OutScope: break;
1179 case CloseSequential:
1180 if (v->type == Tlabel)
1181 v->scope = PendingScope;
1184 v->scope = OutScope;
1187 /* There was no 'else', so we can only become
1188 * conditional if we know the cases were exhaustive,
1189 * and that doesn't mean anything yet.
1190 * So only labels become conditional..
1193 v2 && v2->scope == PendingScope;
1195 if (v2->type == Tlabel) {
1196 v2->scope = CondScope;
1197 v2->min_depth = c->scope_depth;
1199 v2->scope = OutScope;
1202 case OutScope: break;
1206 if (v->scope == OutScope || v->name->var != v)
1215 The value of a variable is store separately from the variable, on an
1216 analogue of a stack frame. There are (currently) two frames that can be
1217 active. A global frame which currently only stores constants, and a
1218 stacked frame which stores local variables. Each variable knows if it
1219 is global or not, and what its index into the frame is.
1221 Values in the global frame are known immediately they are relevant, so
1222 the frame needs to be reallocated as it grows so it can store those
1223 values. The local frame doesn't get values until the interpreted phase
1224 is started, so there is no need to allocate until the size is known.
1226 ###### variable fields
1230 ###### parse context
1232 short global_size, global_alloc;
1234 void *global, *local;
1236 ###### ast functions
1238 static struct value *var_value(struct parse_context *c, struct variable *v)
1241 if (!c->local || !v->type)
1243 if (v->frame_pos + v->type->size > c->local_size) {
1244 printf("INVALID frame_pos\n"); // NOTEST
1247 return c->local + v->frame_pos;
1249 if (c->global_size > c->global_alloc) {
1250 int old = c->global_alloc;
1251 c->global_alloc = (c->global_size | 1023) + 1024;
1252 c->global = realloc(c->global, c->global_alloc);
1253 memset(c->global + old, 0, c->global_alloc - old);
1255 return c->global + v->frame_pos;
1258 static struct value *global_alloc(struct parse_context *c, struct type *t,
1259 struct variable *v, struct value *init)
1262 struct variable scratch;
1264 if (t->prepare_type)
1265 t->prepare_type(c, t, 1);
1267 if (c->global_size & (t->align - 1))
1268 c->global_size = (c->global_size + t->align) & ~(t->align-1);
1273 v->frame_pos = c->global_size;
1275 c->global_size += v->type->size;
1276 ret = var_value(c, v);
1278 memcpy(ret, init, t->size);
1284 As global values are found -- struct field initializers, labels etc --
1285 `global_alloc()` is called to record the value in the global frame.
1287 When the program is fully parsed, we need to walk the list of variables
1288 to find any that weren't merged away and that aren't global, and to
1289 calculate the frame size and assign a frame position for each variable.
1290 For this we have `scope_finalize()`.
1292 ###### ast functions
1294 static void scope_finalize(struct parse_context *c)
1298 for (b = c->varlist; b; b = b->next) {
1300 for (v = b->var; v; v = v->previous) {
1301 struct type *t = v->type;
1302 if (v->merged && v->merged != v)
1306 if (c->local_size & (t->align - 1))
1307 c->local_size = (c->local_size + t->align) & ~(t->align-1);
1308 v->frame_pos = c->local_size;
1309 c->local_size += v->type->size;
1312 c->local = calloc(1, c->local_size);
1315 ###### free context vars
1316 free(context.global);
1317 free(context.local);
1321 Executables can be lots of different things. In many cases an
1322 executable is just an operation combined with one or two other
1323 executables. This allows for expressions and lists etc. Other times an
1324 executable is something quite specific like a constant or variable name.
1325 So we define a `struct exec` to be a general executable with a type, and
1326 a `struct binode` which is a subclass of `exec`, forms a node in a
1327 binary tree, and holds an operation. There will be other subclasses,
1328 and to access these we need to be able to `cast` the `exec` into the
1329 various other types. The first field in any `struct exec` is the type
1330 from the `exec_types` enum.
1333 #define cast(structname, pointer) ({ \
1334 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1335 if (__mptr && *__mptr != X##structname) abort(); \
1336 (struct structname *)( (char *)__mptr);})
1338 #define new(structname) ({ \
1339 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1340 __ptr->type = X##structname; \
1341 __ptr->line = -1; __ptr->column = -1; \
1344 #define new_pos(structname, token) ({ \
1345 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1346 __ptr->type = X##structname; \
1347 __ptr->line = token.line; __ptr->column = token.col; \
1356 enum exec_types type;
1364 struct exec *left, *right;
1367 ###### ast functions
1369 static int __fput_loc(struct exec *loc, FILE *f)
1373 if (loc->line >= 0) {
1374 fprintf(f, "%d:%d: ", loc->line, loc->column);
1377 if (loc->type == Xbinode)
1378 return __fput_loc(cast(binode,loc)->left, f) ||
1379 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1382 static void fput_loc(struct exec *loc, FILE *f)
1384 if (!__fput_loc(loc, f))
1385 fprintf(f, "??:??: "); // NOTEST
1388 Each different type of `exec` node needs a number of functions defined,
1389 a bit like methods. We must be able to free it, print it, analyse it
1390 and execute it. Once we have specific `exec` types we will need to
1391 parse them too. Let's take this a bit more slowly.
1395 The parser generator requires a `free_foo` function for each struct
1396 that stores attributes and they will often be `exec`s and subtypes
1397 there-of. So we need `free_exec` which can handle all the subtypes,
1398 and we need `free_binode`.
1400 ###### ast functions
1402 static void free_binode(struct binode *b)
1407 free_exec(b->right);
1411 ###### core functions
1412 static void free_exec(struct exec *e)
1421 ###### forward decls
1423 static void free_exec(struct exec *e);
1425 ###### free exec cases
1426 case Xbinode: free_binode(cast(binode, e)); break;
1430 Printing an `exec` requires that we know the current indent level for
1431 printing line-oriented components. As will become clear later, we
1432 also want to know what sort of bracketing to use.
1434 ###### ast functions
1436 static void do_indent(int i, char *str)
1443 ###### core functions
1444 static void print_binode(struct binode *b, int indent, int bracket)
1448 ## print binode cases
1452 static void print_exec(struct exec *e, int indent, int bracket)
1458 print_binode(cast(binode, e), indent, bracket); break;
1463 ###### forward decls
1465 static void print_exec(struct exec *e, int indent, int bracket);
1469 As discussed, analysis involves propagating type requirements around the
1470 program and looking for errors.
1472 So `propagate_types` is passed an expected type (being a `struct type`
1473 pointer together with some `val_rules` flags) that the `exec` is
1474 expected to return, and returns the type that it does return, either
1475 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1476 by reference. It is set to `0` when an error is found, and `2` when
1477 any change is made. If it remains unchanged at `1`, then no more
1478 propagation is needed.
1482 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1486 if (rules & Rnolabel)
1487 fputs(" (labels not permitted)", stderr);
1490 ###### core functions
1492 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1493 struct type *type, int rules);
1494 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1495 struct type *type, int rules)
1502 switch (prog->type) {
1505 struct binode *b = cast(binode, prog);
1507 ## propagate binode cases
1511 ## propagate exec cases
1516 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1517 struct type *type, int rules)
1519 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1528 Interpreting an `exec` doesn't require anything but the `exec`. State
1529 is stored in variables and each variable will be directly linked from
1530 within the `exec` tree. The exception to this is the `main` function
1531 which needs to look at command line arguments. This function will be
1532 interpreted separately.
1534 Each `exec` can return a value combined with a type in `struct lrval`.
1535 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1536 the location of a value, which can be updated, in `lval`. Others will
1537 set `lval` to NULL indicating that there is a value of appropriate type
1540 ###### core functions
1544 struct value rval, *lval;
1547 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1549 static struct value interp_exec(struct parse_context *c, struct exec *e,
1550 struct type **typeret)
1552 struct lrval ret = _interp_exec(c, e);
1554 if (!ret.type) abort();
1556 *typeret = ret.type;
1558 dup_value(ret.type, ret.lval, &ret.rval);
1562 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1563 struct type **typeret)
1565 struct lrval ret = _interp_exec(c, e);
1568 *typeret = ret.type;
1570 free_value(ret.type, &ret.rval);
1574 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1577 struct value rv = {}, *lrv = NULL;
1578 struct type *rvtype;
1580 rvtype = ret.type = Tnone;
1590 struct binode *b = cast(binode, e);
1591 struct value left, right, *lleft;
1592 struct type *ltype, *rtype;
1593 ltype = rtype = Tnone;
1595 ## interp binode cases
1597 free_value(ltype, &left);
1598 free_value(rtype, &right);
1601 ## interp exec cases
1611 Now that we have the shape of the interpreter in place we can add some
1612 complex types and connected them in to the data structures and the
1613 different phases of parse, analyse, print, interpret.
1615 Thus far we have arrays and structs.
1619 Arrays can be declared by giving a size and a type, as `[size]type' so
1620 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1621 size can be either a literal number, or a named constant. Some day an
1622 arbitrary expression will be supported.
1624 As a formal parameter to a function, the array can be declared with a
1625 new variable as the size: `name:[size::number]string`. The `size`
1626 variable is set to the size of the array and must be a constant. As
1627 `number` is the only supported type, it can be left out:
1628 `name:[size::]string`.
1630 Arrays cannot be assigned. When pointers are introduced we will also
1631 introduce array slices which can refer to part or all of an array -
1632 the assignment syntax will create a slice. For now, an array can only
1633 ever be referenced by the name it is declared with. It is likely that
1634 a "`copy`" primitive will eventually be define which can be used to
1635 make a copy of an array with controllable recursive depth.
1637 For now we have two sorts of array, those with fixed size either because
1638 it is given as a literal number or because it is a struct member (which
1639 cannot have a runtime-changing size), and those with a size that is
1640 determined at runtime - local variables with a const size. The former
1641 have their size calculated at parse time, the latter at run time.
1643 For the latter type, the `size` field of the type is the size of a
1644 pointer, and the array is reallocated every time it comes into scope.
1646 We differentiate struct fields with a const size from local variables
1647 with a const size by whether they are prepared at parse time or not.
1649 ###### type union fields
1652 int unspec; // size is unspecified - vsize must be set.
1655 struct variable *vsize;
1656 struct type *member;
1659 ###### value union fields
1660 void *array; // used if not static_size
1662 ###### value functions
1664 static void array_prepare_type(struct parse_context *c, struct type *type,
1667 struct value *vsize;
1669 if (!type->array.vsize || type->array.static_size)
1672 vsize = var_value(c, type->array.vsize);
1674 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1675 type->array.size = mpz_get_si(q);
1679 type->array.static_size = 1;
1680 type->size = type->array.size * type->array.member->size;
1681 type->align = type->array.member->align;
1685 static void array_init(struct type *type, struct value *val)
1688 void *ptr = val->ptr;
1692 if (!type->array.static_size) {
1693 val->array = calloc(type->array.size,
1694 type->array.member->size);
1697 for (i = 0; i < type->array.size; i++) {
1699 v = (void*)ptr + i * type->array.member->size;
1700 val_init(type->array.member, v);
1704 static void array_free(struct type *type, struct value *val)
1707 void *ptr = val->ptr;
1709 if (!type->array.static_size)
1711 for (i = 0; i < type->array.size; i++) {
1713 v = (void*)ptr + i * type->array.member->size;
1714 free_value(type->array.member, v);
1716 if (!type->array.static_size)
1720 static int array_compat(struct type *require, struct type *have)
1722 if (have->compat != require->compat)
1724 /* Both are arrays, so we can look at details */
1725 if (!type_compat(require->array.member, have->array.member, 0))
1727 if (have->array.unspec && require->array.unspec) {
1728 if (have->array.vsize && require->array.vsize &&
1729 have->array.vsize != require->array.vsize)
1730 /* sizes might not be the same */
1734 if (have->array.unspec || require->array.unspec)
1736 if (require->array.vsize == NULL && have->array.vsize == NULL)
1737 return require->array.size == have->array.size;
1739 return require->array.vsize == have->array.vsize;
1742 static void array_print_type(struct type *type, FILE *f)
1745 if (type->array.vsize) {
1746 struct binding *b = type->array.vsize->name;
1747 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1748 type->array.unspec ? "::" : "");
1750 fprintf(f, "%d]", type->array.size);
1751 type_print(type->array.member, f);
1754 static struct type array_prototype = {
1756 .prepare_type = array_prepare_type,
1757 .print_type = array_print_type,
1758 .compat = array_compat,
1760 .size = sizeof(void*),
1761 .align = sizeof(void*),
1764 ###### declare terminals
1769 | [ NUMBER ] Type ${ {
1772 struct text noname = { "", 0 };
1775 $0 = t = add_type(c, noname, &array_prototype);
1776 t->array.member = $<4;
1777 t->array.vsize = NULL;
1778 if (number_parse(num, tail, $2.txt) == 0)
1779 tok_err(c, "error: unrecognised number", &$2);
1781 tok_err(c, "error: unsupported number suffix", &$2);
1783 t->array.size = mpz_get_ui(mpq_numref(num));
1784 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1785 tok_err(c, "error: array size must be an integer",
1787 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1788 tok_err(c, "error: array size is too large",
1792 t->array.static_size = 1;
1793 t->size = t->array.size * t->array.member->size;
1794 t->align = t->array.member->align;
1797 | [ IDENTIFIER ] Type ${ {
1798 struct variable *v = var_ref(c, $2.txt);
1799 struct text noname = { "", 0 };
1802 tok_err(c, "error: name undeclared", &$2);
1803 else if (!v->constant)
1804 tok_err(c, "error: array size must be a constant", &$2);
1806 $0 = add_type(c, noname, &array_prototype);
1807 $0->array.member = $<4;
1809 $0->array.vsize = v;
1814 OptType -> Type ${ $0 = $<1; }$
1817 ###### formal type grammar
1819 | [ IDENTIFIER :: OptType ] Type ${ {
1820 struct variable *v = var_decl(c, $ID.txt);
1821 struct text noname = { "", 0 };
1827 $0 = add_type(c, noname, &array_prototype);
1828 $0->array.member = $<6;
1830 $0->array.unspec = 1;
1831 $0->array.vsize = v;
1837 ###### variable grammar
1839 | Variable [ Expression ] ${ {
1840 struct binode *b = new(binode);
1847 ###### print binode cases
1849 print_exec(b->left, -1, bracket);
1851 print_exec(b->right, -1, bracket);
1855 ###### propagate binode cases
1857 /* left must be an array, right must be a number,
1858 * result is the member type of the array
1860 propagate_types(b->right, c, ok, Tnum, 0);
1861 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1862 if (!t || t->compat != array_compat) {
1863 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1866 if (!type_compat(type, t->array.member, rules)) {
1867 type_err(c, "error: have %1 but need %2", prog,
1868 t->array.member, rules, type);
1870 return t->array.member;
1874 ###### interp binode cases
1880 lleft = linterp_exec(c, b->left, <ype);
1881 right = interp_exec(c, b->right, &rtype);
1883 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1887 if (ltype->array.static_size)
1890 ptr = *(void**)lleft;
1891 rvtype = ltype->array.member;
1892 if (i >= 0 && i < ltype->array.size)
1893 lrv = ptr + i * rvtype->size;
1895 val_init(ltype->array.member, &rv);
1902 A `struct` is a data-type that contains one or more other data-types.
1903 It differs from an array in that each member can be of a different
1904 type, and they are accessed by name rather than by number. Thus you
1905 cannot choose an element by calculation, you need to know what you
1908 The language makes no promises about how a given structure will be
1909 stored in memory - it is free to rearrange fields to suit whatever
1910 criteria seems important.
1912 Structs are declared separately from program code - they cannot be
1913 declared in-line in a variable declaration like arrays can. A struct
1914 is given a name and this name is used to identify the type - the name
1915 is not prefixed by the word `struct` as it would be in C.
1917 Structs are only treated as the same if they have the same name.
1918 Simply having the same fields in the same order is not enough. This
1919 might change once we can create structure initializers from a list of
1922 Each component datum is identified much like a variable is declared,
1923 with a name, one or two colons, and a type. The type cannot be omitted
1924 as there is no opportunity to deduce the type from usage. An initial
1925 value can be given following an equals sign, so
1927 ##### Example: a struct type
1933 would declare a type called "complex" which has two number fields,
1934 each initialised to zero.
1936 Struct will need to be declared separately from the code that uses
1937 them, so we will need to be able to print out the declaration of a
1938 struct when reprinting the whole program. So a `print_type_decl` type
1939 function will be needed.
1941 ###### type union fields
1953 ###### type functions
1954 void (*print_type_decl)(struct type *type, FILE *f);
1956 ###### value functions
1958 static void structure_init(struct type *type, struct value *val)
1962 for (i = 0; i < type->structure.nfields; i++) {
1964 v = (void*) val->ptr + type->structure.fields[i].offset;
1965 if (type->structure.fields[i].init)
1966 dup_value(type->structure.fields[i].type,
1967 type->structure.fields[i].init,
1970 val_init(type->structure.fields[i].type, v);
1974 static void structure_free(struct type *type, struct value *val)
1978 for (i = 0; i < type->structure.nfields; i++) {
1980 v = (void*)val->ptr + type->structure.fields[i].offset;
1981 free_value(type->structure.fields[i].type, v);
1985 static void structure_free_type(struct type *t)
1988 for (i = 0; i < t->structure.nfields; i++)
1989 if (t->structure.fields[i].init) {
1990 free_value(t->structure.fields[i].type,
1991 t->structure.fields[i].init);
1993 free(t->structure.fields);
1996 static struct type structure_prototype = {
1997 .init = structure_init,
1998 .free = structure_free,
1999 .free_type = structure_free_type,
2000 .print_type_decl = structure_print_type,
2014 ###### free exec cases
2016 free_exec(cast(fieldref, e)->left);
2020 ###### declare terminals
2023 ###### variable grammar
2025 | Variable . IDENTIFIER ${ {
2026 struct fieldref *fr = new_pos(fieldref, $2);
2033 ###### print exec cases
2037 struct fieldref *f = cast(fieldref, e);
2038 print_exec(f->left, -1, bracket);
2039 printf(".%.*s", f->name.len, f->name.txt);
2043 ###### ast functions
2044 static int find_struct_index(struct type *type, struct text field)
2047 for (i = 0; i < type->structure.nfields; i++)
2048 if (text_cmp(type->structure.fields[i].name, field) == 0)
2053 ###### propagate exec cases
2057 struct fieldref *f = cast(fieldref, prog);
2058 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2061 type_err(c, "error: unknown type for field access", f->left,
2063 else if (st->init != structure_init)
2064 type_err(c, "error: field reference attempted on %1, not a struct",
2065 f->left, st, 0, NULL);
2066 else if (f->index == -2) {
2067 f->index = find_struct_index(st, f->name);
2069 type_err(c, "error: cannot find requested field in %1",
2070 f->left, st, 0, NULL);
2072 if (f->index >= 0) {
2073 struct type *ft = st->structure.fields[f->index].type;
2074 if (!type_compat(type, ft, rules))
2075 type_err(c, "error: have %1 but need %2", prog,
2082 ###### interp exec cases
2085 struct fieldref *f = cast(fieldref, e);
2087 struct value *lleft = linterp_exec(c, f->left, <ype);
2088 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2089 rvtype = ltype->structure.fields[f->index].type;
2095 struct fieldlist *prev;
2099 ###### ast functions
2100 static void free_fieldlist(struct fieldlist *f)
2104 free_fieldlist(f->prev);
2106 free_value(f->f.type, f->f.init);
2112 ###### top level grammar
2113 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2115 add_type(c, $2.txt, &structure_prototype);
2117 struct fieldlist *f;
2119 for (f = $3; f; f=f->prev)
2122 t->structure.nfields = cnt;
2123 t->structure.fields = calloc(cnt, sizeof(struct field));
2126 int a = f->f.type->align;
2128 t->structure.fields[cnt] = f->f;
2129 if (t->size & (a-1))
2130 t->size = (t->size | (a-1)) + 1;
2131 t->structure.fields[cnt].offset = t->size;
2132 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2141 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2142 | { SimpleFieldList } ${ $0 = $<SFL; }$
2143 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2144 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2146 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2147 | FieldLines SimpleFieldList Newlines ${
2152 SimpleFieldList -> Field ${ $0 = $<F; }$
2153 | SimpleFieldList ; Field ${
2157 | SimpleFieldList ; ${
2160 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2162 Field -> IDENTIFIER : Type = Expression ${ {
2165 $0 = calloc(1, sizeof(struct fieldlist));
2166 $0->f.name = $1.txt;
2171 propagate_types($<5, c, &ok, $3, 0);
2176 struct value vl = interp_exec(c, $5, NULL);
2177 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2180 | IDENTIFIER : Type ${
2181 $0 = calloc(1, sizeof(struct fieldlist));
2182 $0->f.name = $1.txt;
2184 if ($0->f.type->prepare_type)
2185 $0->f.type->prepare_type(c, $0->f.type, 1);
2188 ###### forward decls
2189 static void structure_print_type(struct type *t, FILE *f);
2191 ###### value functions
2192 static void structure_print_type(struct type *t, FILE *f)
2196 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2198 for (i = 0; i < t->structure.nfields; i++) {
2199 struct field *fl = t->structure.fields + i;
2200 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2201 type_print(fl->type, f);
2202 if (fl->type->print && fl->init) {
2204 if (fl->type == Tstr)
2206 print_value(fl->type, fl->init);
2207 if (fl->type == Tstr)
2214 ###### print type decls
2219 while (target != 0) {
2221 for (t = context.typelist; t ; t=t->next)
2222 if (t->print_type_decl) {
2231 t->print_type_decl(t, stdout);
2239 A function is a named chunk of code which can be passed parameters and
2240 can return results. Each function has an implicit type which includes
2241 the set of parameters and the return value. As yet these types cannot
2242 be declared separate from the function itself.
2244 In fact, only one function is currently possible - `main`. `main` is
2245 passed an array of strings together with the size of the array, and
2246 doesn't return anything. The strings are command line arguments.
2248 The parameters can be specified either in parentheses as a list, such as
2250 ##### Example: function 1
2252 func main(av:[ac::number]string)
2255 or as an indented list of one parameter per line
2257 ##### Example: function 2
2260 argv:[argc::number]string
2272 MainFunction -> func main ( OpenScope Args ) Block Newlines ${
2275 $0->left = reorder_bilist($<Ar);
2277 var_block_close(c, CloseSequential);
2278 if (c->scope_stack && !c->parse_error) abort();
2280 | func main IN OpenScope OptNL Args OUT OptNL do Block Newlines ${
2283 $0->left = reorder_bilist($<Ar);
2285 var_block_close(c, CloseSequential);
2286 if (c->scope_stack && !c->parse_error) abort();
2288 | func main NEWLINE OpenScope OptNL do Block Newlines ${
2293 var_block_close(c, CloseSequential);
2294 if (c->scope_stack && !c->parse_error) abort();
2297 Args -> ${ $0 = NULL; }$
2298 | Varlist ${ $0 = $<1; }$
2299 | Varlist ; ${ $0 = $<1; }$
2300 | Varlist NEWLINE ${ $0 = $<1; }$
2302 Varlist -> Varlist ; ArgDecl ${
2316 ArgDecl -> IDENTIFIER : FormalType ${ {
2317 struct variable *v = var_decl(c, $1.txt);
2323 ## Executables: the elements of code
2325 Each code element needs to be parsed, printed, analysed,
2326 interpreted, and freed. There are several, so let's just start with
2327 the easy ones and work our way up.
2331 We have already met values as separate objects. When manifest
2332 constants appear in the program text, that must result in an executable
2333 which has a constant value. So the `val` structure embeds a value in
2346 ###### ast functions
2347 struct val *new_val(struct type *T, struct token tk)
2349 struct val *v = new_pos(val, tk);
2360 $0 = new_val(Tbool, $1);
2364 $0 = new_val(Tbool, $1);
2368 $0 = new_val(Tnum, $1);
2371 if (number_parse($0->val.num, tail, $1.txt) == 0)
2372 mpq_init($0->val.num);
2374 tok_err(c, "error: unsupported number suffix",
2379 $0 = new_val(Tstr, $1);
2382 string_parse(&$1, '\\', &$0->val.str, tail);
2384 tok_err(c, "error: unsupported string suffix",
2389 $0 = new_val(Tstr, $1);
2392 string_parse(&$1, '\\', &$0->val.str, tail);
2394 tok_err(c, "error: unsupported string suffix",
2399 ###### print exec cases
2402 struct val *v = cast(val, e);
2403 if (v->vtype == Tstr)
2405 print_value(v->vtype, &v->val);
2406 if (v->vtype == Tstr)
2411 ###### propagate exec cases
2414 struct val *val = cast(val, prog);
2415 if (!type_compat(type, val->vtype, rules))
2416 type_err(c, "error: expected %1%r found %2",
2417 prog, type, rules, val->vtype);
2421 ###### interp exec cases
2423 rvtype = cast(val, e)->vtype;
2424 dup_value(rvtype, &cast(val, e)->val, &rv);
2427 ###### ast functions
2428 static void free_val(struct val *v)
2431 free_value(v->vtype, &v->val);
2435 ###### free exec cases
2436 case Xval: free_val(cast(val, e)); break;
2438 ###### ast functions
2439 // Move all nodes from 'b' to 'rv', reversing their order.
2440 // In 'b' 'left' is a list, and 'right' is the last node.
2441 // In 'rv', left' is the first node and 'right' is a list.
2442 static struct binode *reorder_bilist(struct binode *b)
2444 struct binode *rv = NULL;
2447 struct exec *t = b->right;
2451 b = cast(binode, b->left);
2461 Just as we used a `val` to wrap a value into an `exec`, we similarly
2462 need a `var` to wrap a `variable` into an exec. While each `val`
2463 contained a copy of the value, each `var` holds a link to the variable
2464 because it really is the same variable no matter where it appears.
2465 When a variable is used, we need to remember to follow the `->merged`
2466 link to find the primary instance.
2474 struct variable *var;
2482 VariableDecl -> IDENTIFIER : ${ {
2483 struct variable *v = var_decl(c, $1.txt);
2484 $0 = new_pos(var, $1);
2489 v = var_ref(c, $1.txt);
2491 type_err(c, "error: variable '%v' redeclared",
2493 type_err(c, "info: this is where '%v' was first declared",
2494 v->where_decl, NULL, 0, NULL);
2497 | IDENTIFIER :: ${ {
2498 struct variable *v = var_decl(c, $1.txt);
2499 $0 = new_pos(var, $1);
2505 v = var_ref(c, $1.txt);
2507 type_err(c, "error: variable '%v' redeclared",
2509 type_err(c, "info: this is where '%v' was first declared",
2510 v->where_decl, NULL, 0, NULL);
2513 | IDENTIFIER : Type ${ {
2514 struct variable *v = var_decl(c, $1.txt);
2515 $0 = new_pos(var, $1);
2522 v = var_ref(c, $1.txt);
2524 type_err(c, "error: variable '%v' redeclared",
2526 type_err(c, "info: this is where '%v' was first declared",
2527 v->where_decl, NULL, 0, NULL);
2530 | IDENTIFIER :: Type ${ {
2531 struct variable *v = var_decl(c, $1.txt);
2532 $0 = new_pos(var, $1);
2540 v = var_ref(c, $1.txt);
2542 type_err(c, "error: variable '%v' redeclared",
2544 type_err(c, "info: this is where '%v' was first declared",
2545 v->where_decl, NULL, 0, NULL);
2550 Variable -> IDENTIFIER ${ {
2551 struct variable *v = var_ref(c, $1.txt);
2552 $0 = new_pos(var, $1);
2554 /* This might be a label - allocate a var just in case */
2555 v = var_decl(c, $1.txt);
2562 cast(var, $0)->var = v;
2566 ###### print exec cases
2569 struct var *v = cast(var, e);
2571 struct binding *b = v->var->name;
2572 printf("%.*s", b->name.len, b->name.txt);
2579 if (loc && loc->type == Xvar) {
2580 struct var *v = cast(var, loc);
2582 struct binding *b = v->var->name;
2583 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2585 fputs("???", stderr); // NOTEST
2587 fputs("NOTVAR", stderr); // NOTEST
2590 ###### propagate exec cases
2594 struct var *var = cast(var, prog);
2595 struct variable *v = var->var;
2597 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2598 return Tnone; // NOTEST
2602 if (v->constant && (rules & Rnoconstant)) {
2603 type_err(c, "error: Cannot assign to a constant: %v",
2604 prog, NULL, 0, NULL);
2605 type_err(c, "info: name was defined as a constant here",
2606 v->where_decl, NULL, 0, NULL);
2609 if (v->type == Tnone && v->where_decl == prog)
2610 type_err(c, "error: variable used but not declared: %v",
2611 prog, NULL, 0, NULL);
2612 if (v->type == NULL) {
2613 if (type && *ok != 0) {
2615 v->where_set = prog;
2620 if (!type_compat(type, v->type, rules)) {
2621 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2622 type, rules, v->type);
2623 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2624 v->type, rules, NULL);
2631 ###### interp exec cases
2634 struct var *var = cast(var, e);
2635 struct variable *v = var->var;
2639 lrv = var_value(c, v);
2644 ###### ast functions
2646 static void free_var(struct var *v)
2651 ###### free exec cases
2652 case Xvar: free_var(cast(var, e)); break;
2654 ### Expressions: Conditional
2656 Our first user of the `binode` will be conditional expressions, which
2657 is a bit odd as they actually have three components. That will be
2658 handled by having 2 binodes for each expression. The conditional
2659 expression is the lowest precedence operator which is why we define it
2660 first - to start the precedence list.
2662 Conditional expressions are of the form "value `if` condition `else`
2663 other_value". They associate to the right, so everything to the right
2664 of `else` is part of an else value, while only a higher-precedence to
2665 the left of `if` is the if values. Between `if` and `else` there is no
2666 room for ambiguity, so a full conditional expression is allowed in
2678 Expression -> Expression if Expression else Expression $$ifelse ${ {
2679 struct binode *b1 = new(binode);
2680 struct binode *b2 = new(binode);
2689 ## expression grammar
2691 ###### print binode cases
2694 b2 = cast(binode, b->right);
2695 if (bracket) printf("(");
2696 print_exec(b2->left, -1, bracket);
2698 print_exec(b->left, -1, bracket);
2700 print_exec(b2->right, -1, bracket);
2701 if (bracket) printf(")");
2704 ###### propagate binode cases
2707 /* cond must be Tbool, others must match */
2708 struct binode *b2 = cast(binode, b->right);
2711 propagate_types(b->left, c, ok, Tbool, 0);
2712 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2713 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2717 ###### interp binode cases
2720 struct binode *b2 = cast(binode, b->right);
2721 left = interp_exec(c, b->left, <ype);
2723 rv = interp_exec(c, b2->left, &rvtype);
2725 rv = interp_exec(c, b2->right, &rvtype);
2729 ### Expressions: Boolean
2731 The next class of expressions to use the `binode` will be Boolean
2732 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
2733 have same corresponding precendence. The difference is that they don't
2734 evaluate the second expression if not necessary.
2743 ###### expr precedence
2748 ###### expression grammar
2749 | Expression or Expression ${ {
2750 struct binode *b = new(binode);
2756 | Expression or else Expression ${ {
2757 struct binode *b = new(binode);
2764 | Expression and Expression ${ {
2765 struct binode *b = new(binode);
2771 | Expression and then Expression ${ {
2772 struct binode *b = new(binode);
2779 | not Expression ${ {
2780 struct binode *b = new(binode);
2786 ###### print binode cases
2788 if (bracket) printf("(");
2789 print_exec(b->left, -1, bracket);
2791 print_exec(b->right, -1, bracket);
2792 if (bracket) printf(")");
2795 if (bracket) printf("(");
2796 print_exec(b->left, -1, bracket);
2797 printf(" and then ");
2798 print_exec(b->right, -1, bracket);
2799 if (bracket) printf(")");
2802 if (bracket) printf("(");
2803 print_exec(b->left, -1, bracket);
2805 print_exec(b->right, -1, bracket);
2806 if (bracket) printf(")");
2809 if (bracket) printf("(");
2810 print_exec(b->left, -1, bracket);
2811 printf(" or else ");
2812 print_exec(b->right, -1, bracket);
2813 if (bracket) printf(")");
2816 if (bracket) printf("(");
2818 print_exec(b->right, -1, bracket);
2819 if (bracket) printf(")");
2822 ###### propagate binode cases
2828 /* both must be Tbool, result is Tbool */
2829 propagate_types(b->left, c, ok, Tbool, 0);
2830 propagate_types(b->right, c, ok, Tbool, 0);
2831 if (type && type != Tbool)
2832 type_err(c, "error: %1 operation found where %2 expected", prog,
2836 ###### interp binode cases
2838 rv = interp_exec(c, b->left, &rvtype);
2839 right = interp_exec(c, b->right, &rtype);
2840 rv.bool = rv.bool && right.bool;
2843 rv = interp_exec(c, b->left, &rvtype);
2845 rv = interp_exec(c, b->right, NULL);
2848 rv = interp_exec(c, b->left, &rvtype);
2849 right = interp_exec(c, b->right, &rtype);
2850 rv.bool = rv.bool || right.bool;
2853 rv = interp_exec(c, b->left, &rvtype);
2855 rv = interp_exec(c, b->right, NULL);
2858 rv = interp_exec(c, b->right, &rvtype);
2862 ### Expressions: Comparison
2864 Of slightly higher precedence that Boolean expressions are Comparisons.
2865 A comparison takes arguments of any comparable type, but the two types
2868 To simplify the parsing we introduce an `eop` which can record an
2869 expression operator, and the `CMPop` non-terminal will match one of them.
2876 ###### ast functions
2877 static void free_eop(struct eop *e)
2891 ###### expr precedence
2892 $LEFT < > <= >= == != CMPop
2894 ###### expression grammar
2895 | Expression CMPop Expression ${ {
2896 struct binode *b = new(binode);
2906 CMPop -> < ${ $0.op = Less; }$
2907 | > ${ $0.op = Gtr; }$
2908 | <= ${ $0.op = LessEq; }$
2909 | >= ${ $0.op = GtrEq; }$
2910 | == ${ $0.op = Eql; }$
2911 | != ${ $0.op = NEql; }$
2913 ###### print binode cases
2921 if (bracket) printf("(");
2922 print_exec(b->left, -1, bracket);
2924 case Less: printf(" < "); break;
2925 case LessEq: printf(" <= "); break;
2926 case Gtr: printf(" > "); break;
2927 case GtrEq: printf(" >= "); break;
2928 case Eql: printf(" == "); break;
2929 case NEql: printf(" != "); break;
2930 default: abort(); // NOTEST
2932 print_exec(b->right, -1, bracket);
2933 if (bracket) printf(")");
2936 ###### propagate binode cases
2943 /* Both must match but not be labels, result is Tbool */
2944 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2946 propagate_types(b->right, c, ok, t, 0);
2948 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2950 t = propagate_types(b->left, c, ok, t, 0);
2952 if (!type_compat(type, Tbool, 0))
2953 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2954 Tbool, rules, type);
2957 ###### interp binode cases
2966 left = interp_exec(c, b->left, <ype);
2967 right = interp_exec(c, b->right, &rtype);
2968 cmp = value_cmp(ltype, rtype, &left, &right);
2971 case Less: rv.bool = cmp < 0; break;
2972 case LessEq: rv.bool = cmp <= 0; break;
2973 case Gtr: rv.bool = cmp > 0; break;
2974 case GtrEq: rv.bool = cmp >= 0; break;
2975 case Eql: rv.bool = cmp == 0; break;
2976 case NEql: rv.bool = cmp != 0; break;
2977 default: rv.bool = 0; break; // NOTEST
2982 ### Expressions: The rest
2984 The remaining expressions with the highest precedence are arithmetic,
2985 string concatenation, and string conversion. String concatenation
2986 (`++`) has the same precedence as multiplication and division, but lower
2989 String conversion is a temporary feature until I get a better type
2990 system. `$` is a prefix operator which expects a string and returns
2993 `+` and `-` are both infix and prefix operations (where they are
2994 absolute value and negation). These have different operator names.
2996 We also have a 'Bracket' operator which records where parentheses were
2997 found. This makes it easy to reproduce these when printing. Possibly I
2998 should only insert brackets were needed for precedence.
3008 ###### expr precedence
3014 ###### expression grammar
3015 | Expression Eop Expression ${ {
3016 struct binode *b = new(binode);
3023 | Expression Top Expression ${ {
3024 struct binode *b = new(binode);
3031 | ( Expression ) ${ {
3032 struct binode *b = new_pos(binode, $1);
3037 | Uop Expression ${ {
3038 struct binode *b = new(binode);
3043 | Value ${ $0 = $<1; }$
3044 | Variable ${ $0 = $<1; }$
3047 Eop -> + ${ $0.op = Plus; }$
3048 | - ${ $0.op = Minus; }$
3050 Uop -> + ${ $0.op = Absolute; }$
3051 | - ${ $0.op = Negate; }$
3052 | $ ${ $0.op = StringConv; }$
3054 Top -> * ${ $0.op = Times; }$
3055 | / ${ $0.op = Divide; }$
3056 | % ${ $0.op = Rem; }$
3057 | ++ ${ $0.op = Concat; }$
3059 ###### print binode cases
3066 if (bracket) printf("(");
3067 print_exec(b->left, indent, bracket);
3069 case Plus: fputs(" + ", stdout); break;
3070 case Minus: fputs(" - ", stdout); break;
3071 case Times: fputs(" * ", stdout); break;
3072 case Divide: fputs(" / ", stdout); break;
3073 case Rem: fputs(" % ", stdout); break;
3074 case Concat: fputs(" ++ ", stdout); break;
3075 default: abort(); // NOTEST
3077 print_exec(b->right, indent, bracket);
3078 if (bracket) printf(")");
3083 if (bracket) printf("(");
3085 case Absolute: fputs("+", stdout); break;
3086 case Negate: fputs("-", stdout); break;
3087 case StringConv: fputs("$", stdout); break;
3088 default: abort(); // NOTEST
3090 print_exec(b->right, indent, bracket);
3091 if (bracket) printf(")");
3095 print_exec(b->right, indent, bracket);
3099 ###### propagate binode cases
3105 /* both must be numbers, result is Tnum */
3108 /* as propagate_types ignores a NULL,
3109 * unary ops fit here too */
3110 propagate_types(b->left, c, ok, Tnum, 0);
3111 propagate_types(b->right, c, ok, Tnum, 0);
3112 if (!type_compat(type, Tnum, 0))
3113 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3118 /* both must be Tstr, result is Tstr */
3119 propagate_types(b->left, c, ok, Tstr, 0);
3120 propagate_types(b->right, c, ok, Tstr, 0);
3121 if (!type_compat(type, Tstr, 0))
3122 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3127 /* op must be string, result is number */
3128 propagate_types(b->left, c, ok, Tstr, 0);
3129 if (!type_compat(type, Tnum, 0))
3131 "error: Can only convert string to number, not %1",
3132 prog, type, 0, NULL);
3136 return propagate_types(b->right, c, ok, type, 0);
3138 ###### interp binode cases
3141 rv = interp_exec(c, b->left, &rvtype);
3142 right = interp_exec(c, b->right, &rtype);
3143 mpq_add(rv.num, rv.num, right.num);
3146 rv = interp_exec(c, b->left, &rvtype);
3147 right = interp_exec(c, b->right, &rtype);
3148 mpq_sub(rv.num, rv.num, right.num);
3151 rv = interp_exec(c, b->left, &rvtype);
3152 right = interp_exec(c, b->right, &rtype);
3153 mpq_mul(rv.num, rv.num, right.num);
3156 rv = interp_exec(c, b->left, &rvtype);
3157 right = interp_exec(c, b->right, &rtype);
3158 mpq_div(rv.num, rv.num, right.num);
3163 left = interp_exec(c, b->left, <ype);
3164 right = interp_exec(c, b->right, &rtype);
3165 mpz_init(l); mpz_init(r); mpz_init(rem);
3166 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3167 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3168 mpz_tdiv_r(rem, l, r);
3169 val_init(Tnum, &rv);
3170 mpq_set_z(rv.num, rem);
3171 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3176 rv = interp_exec(c, b->right, &rvtype);
3177 mpq_neg(rv.num, rv.num);
3180 rv = interp_exec(c, b->right, &rvtype);
3181 mpq_abs(rv.num, rv.num);
3184 rv = interp_exec(c, b->right, &rvtype);
3187 left = interp_exec(c, b->left, <ype);
3188 right = interp_exec(c, b->right, &rtype);
3190 rv.str = text_join(left.str, right.str);
3193 right = interp_exec(c, b->right, &rvtype);
3197 struct text tx = right.str;
3200 if (tx.txt[0] == '-') {
3205 if (number_parse(rv.num, tail, tx) == 0)
3208 mpq_neg(rv.num, rv.num);
3210 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt);
3214 ###### value functions
3216 static struct text text_join(struct text a, struct text b)
3219 rv.len = a.len + b.len;
3220 rv.txt = malloc(rv.len);
3221 memcpy(rv.txt, a.txt, a.len);
3222 memcpy(rv.txt+a.len, b.txt, b.len);
3226 ### Blocks, Statements, and Statement lists.
3228 Now that we have expressions out of the way we need to turn to
3229 statements. There are simple statements and more complex statements.
3230 Simple statements do not contain (syntactic) newlines, complex statements do.
3232 Statements often come in sequences and we have corresponding simple
3233 statement lists and complex statement lists.
3234 The former comprise only simple statements separated by semicolons.
3235 The later comprise complex statements and simple statement lists. They are
3236 separated by newlines. Thus the semicolon is only used to separate
3237 simple statements on the one line. This may be overly restrictive,
3238 but I'm not sure I ever want a complex statement to share a line with
3241 Note that a simple statement list can still use multiple lines if
3242 subsequent lines are indented, so
3244 ###### Example: wrapped simple statement list
3249 is a single simple statement list. This might allow room for
3250 confusion, so I'm not set on it yet.
3252 A simple statement list needs no extra syntax. A complex statement
3253 list has two syntactic forms. It can be enclosed in braces (much like
3254 C blocks), or it can be introduced by an indent and continue until an
3255 unindented newline (much like Python blocks). With this extra syntax
3256 it is referred to as a block.
3258 Note that a block does not have to include any newlines if it only
3259 contains simple statements. So both of:
3261 if condition: a=b; d=f
3263 if condition { a=b; print f }
3267 In either case the list is constructed from a `binode` list with
3268 `Block` as the operator. When parsing the list it is most convenient
3269 to append to the end, so a list is a list and a statement. When using
3270 the list it is more convenient to consider a list to be a statement
3271 and a list. So we need a function to re-order a list.
3272 `reorder_bilist` serves this purpose.
3274 The only stand-alone statement we introduce at this stage is `pass`
3275 which does nothing and is represented as a `NULL` pointer in a `Block`
3276 list. Other stand-alone statements will follow once the infrastructure
3287 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3288 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3289 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3290 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3291 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3293 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3294 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3295 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3296 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3297 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3299 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3300 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3301 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3303 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3304 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3305 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3306 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3307 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3309 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3311 ComplexStatements -> ComplexStatements ComplexStatement ${
3321 | ComplexStatement ${
3333 ComplexStatement -> SimpleStatements Newlines ${
3334 $0 = reorder_bilist($<SS);
3336 | SimpleStatements ; Newlines ${
3337 $0 = reorder_bilist($<SS);
3339 ## ComplexStatement Grammar
3342 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3348 | SimpleStatement ${
3356 SimpleStatement -> pass ${ $0 = NULL; }$
3357 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3358 ## SimpleStatement Grammar
3360 ###### print binode cases
3364 if (b->left == NULL)
3367 print_exec(b->left, indent, bracket);
3370 print_exec(b->right, indent, bracket);
3373 // block, one per line
3374 if (b->left == NULL)
3375 do_indent(indent, "pass\n");
3377 print_exec(b->left, indent, bracket);
3379 print_exec(b->right, indent, bracket);
3383 ###### propagate binode cases
3386 /* If any statement returns something other than Tnone
3387 * or Tbool then all such must return same type.
3388 * As each statement may be Tnone or something else,
3389 * we must always pass NULL (unknown) down, otherwise an incorrect
3390 * error might occur. We never return Tnone unless it is
3395 for (e = b; e; e = cast(binode, e->right)) {
3396 t = propagate_types(e->left, c, ok, NULL, rules);
3397 if ((rules & Rboolok) && t == Tbool)
3399 if (t && t != Tnone && t != Tbool) {
3403 type_err(c, "error: expected %1%r, found %2",
3404 e->left, type, rules, t);
3410 ###### interp binode cases
3412 while (rvtype == Tnone &&
3415 rv = interp_exec(c, b->left, &rvtype);
3416 b = cast(binode, b->right);
3420 ### The Print statement
3422 `print` is a simple statement that takes a comma-separated list of
3423 expressions and prints the values separated by spaces and terminated
3424 by a newline. No control of formatting is possible.
3426 `print` faces the same list-ordering issue as blocks, and uses the
3432 ##### expr precedence
3435 ###### SimpleStatement Grammar
3437 | print ExpressionList ${
3438 $0 = reorder_bilist($<2);
3440 | print ExpressionList , ${
3445 $0 = reorder_bilist($0);
3456 ExpressionList -> ExpressionList , Expression ${
3469 ###### print binode cases
3472 do_indent(indent, "print");
3476 print_exec(b->left, -1, bracket);
3480 b = cast(binode, b->right);
3486 ###### propagate binode cases
3489 /* don't care but all must be consistent */
3490 propagate_types(b->left, c, ok, NULL, Rnolabel);
3491 propagate_types(b->right, c, ok, NULL, Rnolabel);
3494 ###### interp binode cases
3500 for ( ; b; b = cast(binode, b->right))
3504 left = interp_exec(c, b->left, <ype);
3505 print_value(ltype, &left);
3506 free_value(ltype, &left);
3517 ###### Assignment statement
3519 An assignment will assign a value to a variable, providing it hasn't
3520 been declared as a constant. The analysis phase ensures that the type
3521 will be correct so the interpreter just needs to perform the
3522 calculation. There is a form of assignment which declares a new
3523 variable as well as assigning a value. If a name is assigned before
3524 it is declared, and error will be raised as the name is created as
3525 `Tlabel` and it is illegal to assign to such names.
3531 ###### declare terminals
3534 ###### SimpleStatement Grammar
3535 | Variable = Expression ${
3541 | VariableDecl = Expression ${
3549 if ($1->var->where_set == NULL) {
3551 "Variable declared with no type or value: %v",
3561 ###### print binode cases
3564 do_indent(indent, "");
3565 print_exec(b->left, indent, bracket);
3567 print_exec(b->right, indent, bracket);
3574 struct variable *v = cast(var, b->left)->var;
3575 do_indent(indent, "");
3576 print_exec(b->left, indent, bracket);
3577 if (cast(var, b->left)->var->constant) {
3578 if (v->where_decl == v->where_set) {
3580 type_print(v->type, stdout);
3585 if (v->where_decl == v->where_set) {
3587 type_print(v->type, stdout);
3594 print_exec(b->right, indent, bracket);
3601 ###### propagate binode cases
3605 /* Both must match and not be labels,
3606 * Type must support 'dup',
3607 * For Assign, left must not be constant.
3610 t = propagate_types(b->left, c, ok, NULL,
3611 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3616 if (propagate_types(b->right, c, ok, t, 0) != t)
3617 if (b->left->type == Xvar)
3618 type_err(c, "info: variable '%v' was set as %1 here.",
3619 cast(var, b->left)->var->where_set, t, rules, NULL);
3621 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3623 propagate_types(b->left, c, ok, t,
3624 (b->op == Assign ? Rnoconstant : 0));
3626 if (t && t->dup == NULL)
3627 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3632 ###### interp binode cases
3635 lleft = linterp_exec(c, b->left, <ype);
3636 right = interp_exec(c, b->right, &rtype);
3638 free_value(ltype, lleft);
3639 dup_value(ltype, &right, lleft);
3646 struct variable *v = cast(var, b->left)->var;
3650 val = var_value(c, v);
3651 free_value(v->type, val);
3652 if (v->type->prepare_type)
3653 v->type->prepare_type(c, v->type, 0);
3655 right = interp_exec(c, b->right, &rtype);
3656 memcpy(val, &right, rtype->size);
3659 val_init(v->type, val);
3664 ### The `use` statement
3666 The `use` statement is the last "simple" statement. It is needed when
3667 the condition in a conditional statement is a block. `use` works much
3668 like `return` in C, but only completes the `condition`, not the whole
3674 ###### expr precedence
3677 ###### SimpleStatement Grammar
3679 $0 = new_pos(binode, $1);
3682 if ($0->right->type == Xvar) {
3683 struct var *v = cast(var, $0->right);
3684 if (v->var->type == Tnone) {
3685 /* Convert this to a label */
3688 v->var->type = Tlabel;
3689 val = global_alloc(c, Tlabel, v->var, NULL);
3695 ###### print binode cases
3698 do_indent(indent, "use ");
3699 print_exec(b->right, -1, bracket);
3704 ###### propagate binode cases
3707 /* result matches value */
3708 return propagate_types(b->right, c, ok, type, 0);
3710 ###### interp binode cases
3713 rv = interp_exec(c, b->right, &rvtype);
3716 ### The Conditional Statement
3718 This is the biggy and currently the only complex statement. This
3719 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
3720 It is comprised of a number of parts, all of which are optional though
3721 set combinations apply. Each part is (usually) a key word (`then` is
3722 sometimes optional) followed by either an expression or a code block,
3723 except the `casepart` which is a "key word and an expression" followed
3724 by a code block. The code-block option is valid for all parts and,
3725 where an expression is also allowed, the code block can use the `use`
3726 statement to report a value. If the code block does not report a value
3727 the effect is similar to reporting `True`.
3729 The `else` and `case` parts, as well as `then` when combined with
3730 `if`, can contain a `use` statement which will apply to some
3731 containing conditional statement. `for` parts, `do` parts and `then`
3732 parts used with `for` can never contain a `use`, except in some
3733 subordinate conditional statement.
3735 If there is a `forpart`, it is executed first, only once.
3736 If there is a `dopart`, then it is executed repeatedly providing
3737 always that the `condpart` or `cond`, if present, does not return a non-True
3738 value. `condpart` can fail to return any value if it simply executes
3739 to completion. This is treated the same as returning `True`.
3741 If there is a `thenpart` it will be executed whenever the `condpart`
3742 or `cond` returns True (or does not return any value), but this will happen
3743 *after* `dopart` (when present).
3745 If `elsepart` is present it will be executed at most once when the
3746 condition returns `False` or some value that isn't `True` and isn't
3747 matched by any `casepart`. If there are any `casepart`s, they will be
3748 executed when the condition returns a matching value.
3750 The particular sorts of values allowed in case parts has not yet been
3751 determined in the language design, so nothing is prohibited.
3753 The various blocks in this complex statement potentially provide scope
3754 for variables as described earlier. Each such block must include the
3755 "OpenScope" nonterminal before parsing the block, and must call
3756 `var_block_close()` when closing the block.
3758 The code following "`if`", "`switch`" and "`for`" does not get its own
3759 scope, but is in a scope covering the whole statement, so names
3760 declared there cannot be redeclared elsewhere. Similarly the
3761 condition following "`while`" is in a scope the covers the body
3762 ("`do`" part) of the loop, and which does not allow conditional scope
3763 extension. Code following "`then`" (both looping and non-looping),
3764 "`else`" and "`case`" each get their own local scope.
3766 The type requirements on the code block in a `whilepart` are quite
3767 unusal. It is allowed to return a value of some identifiable type, in
3768 which case the loop aborts and an appropriate `casepart` is run, or it
3769 can return a Boolean, in which case the loop either continues to the
3770 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
3771 This is different both from the `ifpart` code block which is expected to
3772 return a Boolean, or the `switchpart` code block which is expected to
3773 return the same type as the casepart values. The correct analysis of
3774 the type of the `whilepart` code block is the reason for the
3775 `Rboolok` flag which is passed to `propagate_types()`.
3777 The `cond_statement` cannot fit into a `binode` so a new `exec` is
3786 struct exec *action;
3787 struct casepart *next;
3789 struct cond_statement {
3791 struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
3792 struct casepart *casepart;
3795 ###### ast functions
3797 static void free_casepart(struct casepart *cp)
3801 free_exec(cp->value);
3802 free_exec(cp->action);
3809 static void free_cond_statement(struct cond_statement *s)
3813 free_exec(s->forpart);
3814 free_exec(s->condpart);
3815 free_exec(s->dopart);
3816 free_exec(s->thenpart);
3817 free_exec(s->elsepart);
3818 free_casepart(s->casepart);
3822 ###### free exec cases
3823 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
3825 ###### ComplexStatement Grammar
3826 | CondStatement ${ $0 = $<1; }$
3828 ###### expr precedence
3829 $TERM for then while do
3836 // A CondStatement must end with EOL, as does CondSuffix and
3838 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
3839 // may or may not end with EOL
3840 // WhilePart and IfPart include an appropriate Suffix
3842 // Both ForPart and Whilepart open scopes, and CondSuffix only
3843 // closes one - so in the first branch here we have another to close.
3844 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
3847 $0->thenpart = $<TP;
3848 $0->condpart = $WP.condpart; $WP.condpart = NULL;
3849 $0->dopart = $WP.dopart; $WP.dopart = NULL;
3850 var_block_close(c, CloseSequential);
3852 | ForPart OptNL WhilePart CondSuffix ${
3855 $0->condpart = $WP.condpart; $WP.condpart = NULL;
3856 $0->dopart = $WP.dopart; $WP.dopart = NULL;
3857 var_block_close(c, CloseSequential);
3859 | WhilePart CondSuffix ${
3861 $0->condpart = $WP.condpart; $WP.condpart = NULL;
3862 $0->dopart = $WP.dopart; $WP.dopart = NULL;
3864 | SwitchPart OptNL CasePart CondSuffix ${
3866 $0->condpart = $<SP;
3867 $CP->next = $0->casepart;
3868 $0->casepart = $<CP;
3870 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
3872 $0->condpart = $<SP;
3873 $CP->next = $0->casepart;
3874 $0->casepart = $<CP;
3876 | IfPart IfSuffix ${
3878 $0->condpart = $IP.condpart; $IP.condpart = NULL;
3879 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
3880 // This is where we close an "if" statement
3881 var_block_close(c, CloseSequential);
3884 CondSuffix -> IfSuffix ${
3886 // This is where we close scope of the whole
3887 // "for" or "while" statement
3888 var_block_close(c, CloseSequential);
3890 | Newlines CasePart CondSuffix ${
3892 $CP->next = $0->casepart;
3893 $0->casepart = $<CP;
3895 | CasePart CondSuffix ${
3897 $CP->next = $0->casepart;
3898 $0->casepart = $<CP;
3901 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
3902 | Newlines ElsePart ${ $0 = $<EP; }$
3903 | ElsePart ${$0 = $<EP; }$
3905 ElsePart -> else OpenBlock Newlines ${
3906 $0 = new(cond_statement);
3907 $0->elsepart = $<OB;
3908 var_block_close(c, CloseElse);
3910 | else OpenScope CondStatement ${
3911 $0 = new(cond_statement);
3912 $0->elsepart = $<CS;
3913 var_block_close(c, CloseElse);
3917 CasePart -> case Expression OpenScope ColonBlock ${
3918 $0 = calloc(1,sizeof(struct casepart));
3921 var_block_close(c, CloseParallel);
3925 // These scopes are closed in CondSuffix
3926 ForPart -> for OpenBlock ${
3930 ThenPart -> then OpenBlock ${
3932 var_block_close(c, CloseSequential);
3936 // This scope is closed in CondSuffix
3937 WhilePart -> while UseBlock OptNL do Block ${
3941 | while OpenScope Expression ColonBlock ${
3942 $0.condpart = $<Exp;
3946 IfPart -> if UseBlock OptNL then OpenBlock ClosePara ${
3950 | if OpenScope Expression OpenScope ColonBlock ClosePara ${
3954 | if OpenScope Expression OpenScope OptNL then Block ClosePara ${
3960 // This scope is closed in CondSuffix
3961 SwitchPart -> switch OpenScope Expression ${
3964 | switch UseBlock ${
3968 ###### print exec cases
3970 case Xcond_statement:
3972 struct cond_statement *cs = cast(cond_statement, e);
3973 struct casepart *cp;
3975 do_indent(indent, "for");
3976 if (bracket) printf(" {\n"); else printf("\n");
3977 print_exec(cs->forpart, indent+1, bracket);
3980 do_indent(indent, "} then {\n");
3982 do_indent(indent, "then\n");
3983 print_exec(cs->thenpart, indent+1, bracket);
3985 if (bracket) do_indent(indent, "}\n");
3989 if (cs->condpart && cs->condpart->type == Xbinode &&
3990 cast(binode, cs->condpart)->op == Block) {
3992 do_indent(indent, "while {\n");
3994 do_indent(indent, "while\n");
3995 print_exec(cs->condpart, indent+1, bracket);
3997 do_indent(indent, "} do {\n");
3999 do_indent(indent, "do\n");
4000 print_exec(cs->dopart, indent+1, bracket);
4002 do_indent(indent, "}\n");
4004 do_indent(indent, "while ");
4005 print_exec(cs->condpart, 0, bracket);
4010 print_exec(cs->dopart, indent+1, bracket);
4012 do_indent(indent, "}\n");
4017 do_indent(indent, "switch");
4019 do_indent(indent, "if");
4020 if (cs->condpart && cs->condpart->type == Xbinode &&
4021 cast(binode, cs->condpart)->op == Block) {
4026 print_exec(cs->condpart, indent+1, bracket);
4028 do_indent(indent, "}\n");
4030 do_indent(indent, "then:\n");
4031 print_exec(cs->thenpart, indent+1, bracket);
4035 print_exec(cs->condpart, 0, bracket);
4041 print_exec(cs->thenpart, indent+1, bracket);
4043 do_indent(indent, "}\n");
4048 for (cp = cs->casepart; cp; cp = cp->next) {
4049 do_indent(indent, "case ");
4050 print_exec(cp->value, -1, 0);
4055 print_exec(cp->action, indent+1, bracket);
4057 do_indent(indent, "}\n");
4060 do_indent(indent, "else");
4065 print_exec(cs->elsepart, indent+1, bracket);
4067 do_indent(indent, "}\n");
4072 ###### propagate exec cases
4073 case Xcond_statement:
4075 // forpart and dopart must return Tnone
4076 // thenpart must return Tnone if there is a dopart,
4077 // otherwise it is like elsepart.
4079 // be bool if there is no casepart
4080 // match casepart->values if there is a switchpart
4081 // either be bool or match casepart->value if there
4083 // elsepart and casepart->action must match the return type
4084 // expected of this statement.
4085 struct cond_statement *cs = cast(cond_statement, prog);
4086 struct casepart *cp;
4088 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4089 if (!type_compat(Tnone, t, 0))
4091 t = propagate_types(cs->dopart, c, ok, Tnone, 0);
4092 if (!type_compat(Tnone, t, 0))
4095 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4096 if (!type_compat(Tnone, t, 0))
4099 if (cs->casepart == NULL)
4100 propagate_types(cs->condpart, c, ok, Tbool, 0);
4102 /* Condpart must match case values, with bool permitted */
4104 for (cp = cs->casepart;
4105 cp && !t; cp = cp->next)
4106 t = propagate_types(cp->value, c, ok, NULL, 0);
4107 if (!t && cs->condpart)
4108 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok);
4109 // Now we have a type (I hope) push it down
4111 for (cp = cs->casepart; cp; cp = cp->next)
4112 propagate_types(cp->value, c, ok, t, 0);
4113 propagate_types(cs->condpart, c, ok, t, Rboolok);
4116 // (if)then, else, and case parts must return expected type.
4117 if (!cs->dopart && !type)
4118 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4120 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4121 for (cp = cs->casepart;
4124 type = propagate_types(cp->action, c, ok, NULL, rules);
4127 propagate_types(cs->thenpart, c, ok, type, rules);
4128 propagate_types(cs->elsepart, c, ok, type, rules);
4129 for (cp = cs->casepart; cp ; cp = cp->next)
4130 propagate_types(cp->action, c, ok, type, rules);
4136 ###### interp exec cases
4137 case Xcond_statement:
4139 struct value v, cnd;
4140 struct type *vtype, *cndtype;
4141 struct casepart *cp;
4142 struct cond_statement *cs = cast(cond_statement, e);
4145 interp_exec(c, cs->forpart, NULL);
4148 cnd = interp_exec(c, cs->condpart, &cndtype);
4151 if (!(cndtype == Tnone ||
4152 (cndtype == Tbool && cnd.bool != 0)))
4154 // cnd is Tnone or Tbool, doesn't need to be freed
4156 interp_exec(c, cs->dopart, NULL);
4159 rv = interp_exec(c, cs->thenpart, &rvtype);
4160 if (rvtype != Tnone || !cs->dopart)
4162 free_value(rvtype, &rv);
4165 } while (cs->dopart);
4167 for (cp = cs->casepart; cp; cp = cp->next) {
4168 v = interp_exec(c, cp->value, &vtype);
4169 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4170 free_value(vtype, &v);
4171 free_value(cndtype, &cnd);
4172 rv = interp_exec(c, cp->action, &rvtype);
4175 free_value(vtype, &v);
4177 free_value(cndtype, &cnd);
4179 rv = interp_exec(c, cs->elsepart, &rvtype);
4186 ### Top level structure
4188 All the language elements so far can be used in various places. Now
4189 it is time to clarify what those places are.
4191 At the top level of a file there will be a number of declarations.
4192 Many of the things that can be declared haven't been described yet,
4193 such as functions, procedures, imports, and probably more.
4194 For now there are two sorts of things that can appear at the top
4195 level. They are predefined constants, `struct` types, and the `main`
4196 function. While the syntax will allow the `main` function to appear
4197 multiple times, that will trigger an error if it is actually attempted.
4199 The various declarations do not return anything. They store the
4200 various declarations in the parse context.
4202 ###### Parser: grammar
4205 Ocean -> OptNL DeclarationList
4207 ## declare terminals
4214 DeclarationList -> Declaration
4215 | DeclarationList Declaration
4217 Declaration -> ERROR Newlines ${
4219 "error: unhandled parse error", &$1);
4225 ## top level grammar
4229 ### The `const` section
4231 As well as being defined in with the code that uses them, constants
4232 can be declared at the top level. These have full-file scope, so they
4233 are always `InScope`. The value of a top level constant can be given
4234 as an expression, and this is evaluated immediately rather than in the
4235 later interpretation stage. Once we add functions to the language, we
4236 will need rules concern which, if any, can be used to define a top
4239 Constants are defined in a section that starts with the reserved word
4240 `const` and then has a block with a list of assignment statements.
4241 For syntactic consistency, these must use the double-colon syntax to
4242 make it clear that they are constants. Type can also be given: if
4243 not, the type will be determined during analysis, as with other
4246 As the types constants are inserted at the head of a list, printing
4247 them in the same order that they were read is not straight forward.
4248 We take a quadratic approach here and count the number of constants
4249 (variables of depth 0), then count down from there, each time
4250 searching through for the Nth constant for decreasing N.
4252 ###### top level grammar
4256 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4257 | const { SimpleConstList } Newlines
4258 | const IN OptNL ConstList OUT Newlines
4259 | const SimpleConstList Newlines
4261 ConstList -> ConstList SimpleConstLine
4263 SimpleConstList -> SimpleConstList ; Const
4266 SimpleConstLine -> SimpleConstList Newlines
4267 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4270 CType -> Type ${ $0 = $<1; }$
4273 Const -> IDENTIFIER :: CType = Expression ${ {
4277 v = var_decl(c, $1.txt);
4279 struct var *var = new_pos(var, $1);
4280 v->where_decl = var;
4285 v = var_ref(c, $1.txt);
4286 tok_err(c, "error: name already declared", &$1);
4287 type_err(c, "info: this is where '%v' was first declared",
4288 v->where_decl, NULL, 0, NULL);
4292 propagate_types($5, c, &ok, $3, 0);
4297 struct value res = interp_exec(c, $5, &v->type);
4298 global_alloc(c, v->type, v, &res);
4302 ###### print const decls
4307 while (target != 0) {
4309 for (v = context.in_scope; v; v=v->in_scope)
4310 if (v->depth == 0) {
4321 struct value *val = var_value(&context, v);
4322 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4323 type_print(v->type, stdout);
4325 if (v->type == Tstr)
4327 print_value(v->type, val);
4328 if (v->type == Tstr)
4336 ### Finally the whole `main` function.
4338 An Ocean program can currently have only one function - `main` - and
4339 that must exist. It expects an array of strings with a provided size.
4340 Following this is a `block` which is the code to execute.
4342 As this is the top level, several things are handled a bit
4344 The function is not interpreted by `interp_exec` as that isn't
4345 passed the argument list which the program requires. Similarly type
4346 analysis is a bit more interesting at this level.
4348 ###### top level grammar
4350 DeclareFunction -> MainFunction ${ {
4352 type_err(c, "\"main\" defined a second time",
4358 ###### print binode cases
4361 do_indent(indent, "func main(");
4362 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
4363 struct variable *v = cast(var, b2->left)->var;
4365 print_exec(b2->left, 0, 0);
4367 type_print(v->type, stdout);
4373 print_exec(b->right, indent+1, bracket);
4375 do_indent(indent, "}\n");
4378 ###### propagate binode cases
4380 case Func: abort(); // NOTEST
4382 ###### core functions
4384 static int analyse_prog(struct exec *prog, struct parse_context *c)
4386 struct binode *bp = cast(binode, prog);
4390 struct type *argv_type;
4391 struct text argv_type_name = { " argv", 5 };
4396 argv_type = add_type(c, argv_type_name, &array_prototype);
4397 argv_type->array.member = Tstr;
4398 argv_type->array.unspec = 1;
4400 for (b = cast(binode, bp->left); b; b = cast(binode, b->right)) {
4404 propagate_types(b->left, c, &ok, argv_type, 0);
4406 default: /* invalid */ // NOTEST
4407 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4413 propagate_types(bp->right, c, &ok, Tnone, 0);
4418 /* Make sure everything is still consistent */
4419 propagate_types(bp->right, c, &ok, Tnone, 0);
4426 static void interp_prog(struct parse_context *c, struct exec *prog,
4427 int argc, char **argv)
4429 struct binode *p = cast(binode, prog);
4437 al = cast(binode, p->left);
4439 struct var *v = cast(var, al->left);
4440 struct value *vl = var_value(c, v->var);
4450 mpq_set_ui(argcq, argc, 1);
4451 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4452 t->prepare_type(c, t, 0);
4453 array_init(v->var->type, vl);
4454 for (i = 0; i < argc; i++) {
4455 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4458 arg.str.txt = argv[i];
4459 arg.str.len = strlen(argv[i]);
4460 free_value(Tstr, vl2);
4461 dup_value(Tstr, &arg, vl2);
4465 al = cast(binode, al->right);
4467 v = interp_exec(c, p->right, &vtype);
4468 free_value(vtype, &v);
4471 ###### interp binode cases
4473 case Func: abort(); // NOTEST
4475 ## And now to test it out.
4477 Having a language requires having a "hello world" program. I'll
4478 provide a little more than that: a program that prints "Hello world"
4479 finds the GCD of two numbers, prints the first few elements of
4480 Fibonacci, performs a binary search for a number, and a few other
4481 things which will likely grow as the languages grows.
4483 ###### File: oceani.mk
4486 @echo "===== DEMO ====="
4487 ./oceani --section "demo: hello" oceani.mdc 55 33
4493 four ::= 2 + 2 ; five ::= 10/2
4494 const pie ::= "I like Pie";
4495 cake ::= "The cake is"
4506 print "Hello World, what lovely oceans you have!"
4507 print "Are there", five, "?"
4508 print pi, pie, "but", cake
4510 A := $argv[1]; B := $argv[2]
4512 /* When a variable is defined in both branches of an 'if',
4513 * and used afterwards, the variables are merged.
4519 print "Is", A, "bigger than", B,"? ", bigger
4520 /* If a variable is not used after the 'if', no
4521 * merge happens, so types can be different
4524 double:string = "yes"
4525 print A, "is more than twice", B, "?", double
4528 print "double", B, "is", double
4533 if a > 0 and then b > 0:
4539 print "GCD of", A, "and", B,"is", a
4541 print a, "is not positive, cannot calculate GCD"
4543 print b, "is not positive, cannot calculate GCD"
4548 print "Fibonacci:", f1,f2,
4549 then togo = togo - 1
4557 /* Binary search... */
4562 mid := (lo + hi) / 2
4574 print "Yay, I found", target
4576 print "Closest I found was", mid
4581 // "middle square" PRNG. Not particularly good, but one my
4582 // Dad taught me - the first one I ever heard of.
4583 for i:=1; then i = i + 1; while i < size:
4584 n := list[i-1] * list[i-1]
4585 list[i] = (n / 100) % 10 000
4587 print "Before sort:",
4588 for i:=0; then i = i + 1; while i < size:
4592 for i := 1; then i=i+1; while i < size:
4593 for j:=i-1; then j=j-1; while j >= 0:
4594 if list[j] > list[j+1]:
4598 print " After sort:",
4599 for i:=0; then i = i + 1; while i < size:
4603 if 1 == 2 then print "yes"; else print "no"
4607 bob.alive = (bob.name == "Hello")
4608 print "bob", "is" if bob.alive else "isn't", "alive"