1 # Ocean Interpreter - Jamison Creek version
3 Ocean is intended to be a compiled language, so this interpreter is
4 not targeted at being the final product. It is, rather, an intermediate
5 stage and fills that role in two distinct ways.
7 Firstly, it exists as a platform to experiment with the early language
8 design. An interpreter is easy to write and easy to get working, so
9 the barrier for entry is lower if I aim to start with an interpreter.
11 Secondly, the plan for the Ocean compiler is to write it in the
12 [Ocean language](http://ocean-lang.org). To achieve this we naturally
13 need some sort of boot-strap process and this interpreter - written in
14 portable C - will fill that role. It will be used to bootstrap the
17 Two features that are not needed to fill either of these roles are
18 performance and completeness. The interpreter only needs to be fast
19 enough to run small test programs and occasionally to run the compiler
20 on itself. It only needs to be complete enough to test aspects of the
21 design which are developed before the compiler is working, and to run
22 the compiler on itself. Any features not used by the compiler when
23 compiling itself are superfluous. They may be included anyway, but
26 Nonetheless, the interpreter should end up being reasonably complete,
27 and any performance bottlenecks which appear and are easily fixed, will
32 This third version of the interpreter exists to test out some initial
33 ideas relating to types. Particularly it adds arrays (indexed from
34 zero) and simple structures. Basic control flow and variable scoping
35 are already fairly well established, as are basic numerical and
38 Some operators that have only recently been added, and so have not
39 generated all that much experience yet are "and then" and "or else" as
40 short-circuit Boolean operators, and the "if ... else" trinary
41 operator which can select between two expressions based on a third
42 (which appears syntactically in the middle).
44 The "func" clause currently only allows a "main" function to be
45 declared. That will be extended when proper function support is added.
47 An element that is present purely to make a usable language, and
48 without any expectation that they will remain, is the "print" statement
49 which performs simple output.
51 The current scalar types are "number", "Boolean", and "string".
52 Boolean will likely stay in its current form, the other two might, but
53 could just as easily be changed.
57 Versions of the interpreter which obviously do not support a complete
58 language will be named after creeks and streams. This one is Jamison
61 Once we have something reasonably resembling a complete language, the
62 names of rivers will be used.
63 Early versions of the compiler will be named after seas. Major
64 releases of the compiler will be named after oceans. Hopefully I will
65 be finished once I get to the Pacific Ocean release.
69 As well as parsing and executing a program, the interpreter can print
70 out the program from the parsed internal structure. This is useful
71 for validating the parsing.
72 So the main requirements of the interpreter are:
74 - Parse the program, possibly with tracing,
75 - Analyse the parsed program to ensure consistency,
77 - Execute the "main" function in the program, if no parsing or
78 consistency errors were found.
80 This is all performed by a single C program extracted with
83 There will be two formats for printing the program: a default and one
84 that uses bracketing. So a `--bracket` command line option is needed
85 for that. Normally the first code section found is used, however an
86 alternate section can be requested so that a file (such as this one)
87 can contain multiple programs. This is effected with the `--section`
90 This code must be compiled with `-fplan9-extensions` so that anonymous
91 structures can be used.
93 ###### File: oceani.mk
95 myCFLAGS := -Wall -g -fplan9-extensions
96 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
97 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
98 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
100 all :: $(LDLIBS) oceani
101 oceani.c oceani.h : oceani.mdc parsergen
102 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
103 oceani.mk: oceani.mdc md2c
106 oceani: oceani.o $(LDLIBS)
107 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
109 ###### Parser: header
111 struct parse_context;
113 struct parse_context {
114 struct token_config config;
122 #define container_of(ptr, type, member) ({ \
123 const typeof( ((type *)0)->member ) *__mptr = (ptr); \
124 (type *)( (char *)__mptr - offsetof(type,member) );})
126 #define config2context(_conf) container_of(_conf, struct parse_context, \
129 ###### Parser: reduce
130 struct parse_context *c = config2context(config);
138 #include <sys/mman.h>
157 static char Usage[] =
158 "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
159 static const struct option long_options[] = {
160 {"trace", 0, NULL, 't'},
161 {"print", 0, NULL, 'p'},
162 {"noexec", 0, NULL, 'n'},
163 {"brackets", 0, NULL, 'b'},
164 {"section", 1, NULL, 's'},
167 const char *options = "tpnbs";
169 static void pr_err(char *msg) // NOTEST
171 fprintf(stderr, "%s\n", msg); // NOTEST
174 int main(int argc, char *argv[])
179 struct section *s = NULL, *ss;
180 char *section = NULL;
181 struct parse_context context = {
183 .ignored = (1 << TK_mark),
184 .number_chars = ".,_+- ",
189 int doprint=0, dotrace=0, doexec=1, brackets=0;
191 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
194 case 't': dotrace=1; break;
195 case 'p': doprint=1; break;
196 case 'n': doexec=0; break;
197 case 'b': brackets=1; break;
198 case 's': section = optarg; break;
199 default: fprintf(stderr, Usage);
203 if (optind >= argc) {
204 fprintf(stderr, "oceani: no input file given\n");
207 fd = open(argv[optind], O_RDONLY);
209 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
212 context.file_name = argv[optind];
213 len = lseek(fd, 0, 2);
214 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
215 s = code_extract(file, file+len, pr_err);
217 fprintf(stderr, "oceani: could not find any code in %s\n",
222 ## context initialization
225 for (ss = s; ss; ss = ss->next) {
226 struct text sec = ss->section;
227 if (sec.len == strlen(section) &&
228 strncmp(sec.txt, section, sec.len) == 0)
232 fprintf(stderr, "oceani: cannot find section %s\n",
239 fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
240 goto cleanup; // NOTEST
243 parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
245 if (!context.parse_error && !analyse_funcs(&context)) {
246 fprintf(stderr, "oceani: type error in program - not running.\n");
247 context.parse_error = 1;
255 if (doexec && !context.parse_error)
256 interp_main(&context, argc - optind, argv + optind);
259 struct section *t = s->next;
264 // FIXME parser should pop scope even on error
265 while (context.scope_depth > 0)
268 ## free context types
269 ## free context storage
270 exit(context.parse_error ? 1 : 0);
275 The four requirements of parse, analyse, print, interpret apply to
276 each language element individually so that is how most of the code
279 Three of the four are fairly self explanatory. The one that requires
280 a little explanation is the analysis step.
282 The current language design does not require the types of variables to
283 be declared, but they must still have a single type. Different
284 operations impose different requirements on the variables, for example
285 addition requires both arguments to be numeric, and assignment
286 requires the variable on the left to have the same type as the
287 expression on the right.
289 Analysis involves propagating these type requirements around and
290 consequently setting the type of each variable. If any requirements
291 are violated (e.g. a string is compared with a number) or if a
292 variable needs to have two different types, then an error is raised
293 and the program will not run.
295 If the same variable is declared in both branchs of an 'if/else', or
296 in all cases of a 'switch' then the multiple instances may be merged
297 into just one variable if the variable is referenced after the
298 conditional statement. When this happens, the types must naturally be
299 consistent across all the branches. When the variable is not used
300 outside the if, the variables in the different branches are distinct
301 and can be of different types.
303 Undeclared names may only appear in "use" statements and "case" expressions.
304 These names are given a type of "label" and a unique value.
305 This allows them to fill the role of a name in an enumerated type, which
306 is useful for testing the `switch` statement.
308 As we will see, the condition part of a `while` statement can return
309 either a Boolean or some other type. This requires that the expected
310 type that gets passed around comprises a type and a flag to indicate
311 that `Tbool` is also permitted.
313 As there are, as yet, no distinct types that are compatible, there
314 isn't much subtlety in the analysis. When we have distinct number
315 types, this will become more interesting.
319 When analysis discovers an inconsistency it needs to report an error;
320 just refusing to run the code ensures that the error doesn't cascade,
321 but by itself it isn't very useful. A clear understanding of the sort
322 of error message that are useful will help guide the process of
325 At a simplistic level, the only sort of error that type analysis can
326 report is that the type of some construct doesn't match a contextual
327 requirement. For example, in `4 + "hello"` the addition provides a
328 contextual requirement for numbers, but `"hello"` is not a number. In
329 this particular example no further information is needed as the types
330 are obvious from local information. When a variable is involved that
331 isn't the case. It may be helpful to explain why the variable has a
332 particular type, by indicating the location where the type was set,
333 whether by declaration or usage.
335 Using a recursive-descent analysis we can easily detect a problem at
336 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
337 will detect that one argument is not a number and the usage of `hello`
338 will detect that a number was wanted, but not provided. In this
339 (early) version of the language, we will generate error reports at
340 multiple locations, so the use of `hello` will report an error and
341 explain were the value was set, and the addition will report an error
342 and say why numbers are needed. To be able to report locations for
343 errors, each language element will need to record a file location
344 (line and column) and each variable will need to record the language
345 element where its type was set. For now we will assume that each line
346 of an error message indicates one location in the file, and up to 2
347 types. So we provide a `printf`-like function which takes a format, a
348 location (a `struct exec` which has not yet been introduced), and 2
349 types. "`%1`" reports the first type, "`%2`" reports the second. We
350 will need a function to print the location, once we know how that is
351 stored. e As will be explained later, there are sometimes extra rules for
352 type matching and they might affect error messages, we need to pass those
355 As well as type errors, we sometimes need to report problems with
356 tokens, which might be unexpected or might name a type that has not
357 been defined. For these we have `tok_err()` which reports an error
358 with a given token. Each of the error functions sets the flag in the
359 context so indicate that parsing failed.
363 static void fput_loc(struct exec *loc, FILE *f);
364 static void type_err(struct parse_context *c,
365 char *fmt, struct exec *loc,
366 struct type *t1, int rules, struct type *t2);
368 ###### core functions
370 static void type_err(struct parse_context *c,
371 char *fmt, struct exec *loc,
372 struct type *t1, int rules, struct type *t2)
374 fprintf(stderr, "%s:", c->file_name);
375 fput_loc(loc, stderr);
376 for (; *fmt ; fmt++) {
383 case '%': fputc(*fmt, stderr); break; // NOTEST
384 default: fputc('?', stderr); break; // NOTEST
386 type_print(t1, stderr);
389 type_print(t2, stderr);
398 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
400 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
401 t->txt.len, t->txt.txt);
405 ## Entities: declared and predeclared.
407 There are various "things" that the language and/or the interpreter
408 needs to know about to parse and execute a program. These include
409 types, variables, values, and executable code. These are all lumped
410 together under the term "entities" (calling them "objects" would be
411 confusing) and introduced here. The following section will present the
412 different specific code elements which comprise or manipulate these
417 Values come in a wide range of types, with more likely to be added.
418 Each type needs to be able to print its own values (for convenience at
419 least) as well as to compare two values, at least for equality and
420 possibly for order. For now, values might need to be duplicated and
421 freed, though eventually such manipulations will be better integrated
424 Rather than requiring every numeric type to support all numeric
425 operations (add, multiple, etc), we allow types to be able to present
426 as one of a few standard types: integer, float, and fraction. The
427 existence of these conversion functions eventually enable types to
428 determine if they are compatible with other types, though such types
429 have not yet been implemented.
431 Named type are stored in a simple linked list. Objects of each type are
432 "values" which are often passed around by value.
439 ## value union fields
447 void (*init)(struct type *type, struct value *val);
448 void (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
449 void (*print)(struct type *type, struct value *val);
450 void (*print_type)(struct type *type, FILE *f);
451 int (*cmp_order)(struct type *t1, struct type *t2,
452 struct value *v1, struct value *v2);
453 int (*cmp_eq)(struct type *t1, struct type *t2,
454 struct value *v1, struct value *v2);
455 void (*dup)(struct type *type, struct value *vold, struct value *vnew);
456 void (*free)(struct type *type, struct value *val);
457 void (*free_type)(struct type *t);
458 long long (*to_int)(struct value *v);
459 double (*to_float)(struct value *v);
460 int (*to_mpq)(mpq_t *q, struct value *v);
469 struct type *typelist;
473 static struct type *find_type(struct parse_context *c, struct text s)
475 struct type *l = c->typelist;
478 text_cmp(l->name, s) != 0)
483 static struct type *add_type(struct parse_context *c, struct text s,
488 n = calloc(1, sizeof(*n));
491 n->next = c->typelist;
496 static void free_type(struct type *t)
498 /* The type is always a reference to something in the
499 * context, so we don't need to free anything.
503 static void free_value(struct type *type, struct value *v)
507 memset(v, 0x5a, type->size);
511 static void type_print(struct type *type, FILE *f)
514 fputs("*unknown*type*", f); // NOTEST
515 else if (type->name.len)
516 fprintf(f, "%.*s", type->name.len, type->name.txt);
517 else if (type->print_type)
518 type->print_type(type, f);
520 fputs("*invalid*type*", f); // NOTEST
523 static void val_init(struct type *type, struct value *val)
525 if (type && type->init)
526 type->init(type, val);
529 static void dup_value(struct type *type,
530 struct value *vold, struct value *vnew)
532 if (type && type->dup)
533 type->dup(type, vold, vnew);
536 static int value_cmp(struct type *tl, struct type *tr,
537 struct value *left, struct value *right)
539 if (tl && tl->cmp_order)
540 return tl->cmp_order(tl, tr, left, right);
541 if (tl && tl->cmp_eq) // NOTEST
542 return tl->cmp_eq(tl, tr, left, right); // NOTEST
546 static void print_value(struct type *type, struct value *v)
548 if (type && type->print)
549 type->print(type, v);
551 printf("*Unknown*"); // NOTEST
556 static void free_value(struct type *type, struct value *v);
557 static int type_compat(struct type *require, struct type *have, int rules);
558 static void type_print(struct type *type, FILE *f);
559 static void val_init(struct type *type, struct value *v);
560 static void dup_value(struct type *type,
561 struct value *vold, struct value *vnew);
562 static int value_cmp(struct type *tl, struct type *tr,
563 struct value *left, struct value *right);
564 static void print_value(struct type *type, struct value *v);
566 ###### free context types
568 while (context.typelist) {
569 struct type *t = context.typelist;
571 context.typelist = t->next;
577 Type can be specified for local variables, for fields in a structure,
578 for formal parameters to functions, and possibly elsewhere. Different
579 rules may apply in different contexts. As a minimum, a named type may
580 always be used. Currently the type of a formal parameter can be
581 different from types in other contexts, so we have a separate grammar
587 Type -> IDENTIFIER ${
588 $0 = find_type(c, $1.txt);
591 "error: undefined type", &$1);
598 FormalType -> Type ${ $0 = $<1; }$
599 ## formal type grammar
603 Values of the base types can be numbers, which we represent as
604 multi-precision fractions, strings, Booleans and labels. When
605 analysing the program we also need to allow for places where no value
606 is meaningful (type `Tnone`) and where we don't know what type to
607 expect yet (type is `NULL`).
609 Values are never shared, they are always copied when used, and freed
610 when no longer needed.
612 When propagating type information around the program, we need to
613 determine if two types are compatible, where type `NULL` is compatible
614 with anything. There are two special cases with type compatibility,
615 both related to the Conditional Statement which will be described
616 later. In some cases a Boolean can be accepted as well as some other
617 primary type, and in others any type is acceptable except a label (`Vlabel`).
618 A separate function encoding these cases will simplify some code later.
620 ###### type functions
622 int (*compat)(struct type *this, struct type *other);
626 static int type_compat(struct type *require, struct type *have, int rules)
628 if ((rules & Rboolok) && have == Tbool)
630 if ((rules & Rnolabel) && have == Tlabel)
632 if (!require || !have)
636 return require->compat(require, have);
638 return require == have;
643 #include "parse_string.h"
644 #include "parse_number.h"
647 myLDLIBS := libnumber.o libstring.o -lgmp
648 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
650 ###### type union fields
651 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
653 ###### value union fields
660 static void _free_value(struct type *type, struct value *v)
664 switch (type->vtype) {
666 case Vstr: free(v->str.txt); break;
667 case Vnum: mpq_clear(v->num); break;
673 ###### value functions
675 static void _val_init(struct type *type, struct value *val)
677 switch(type->vtype) {
678 case Vnone: // NOTEST
681 mpq_init(val->num); break;
683 val->str.txt = malloc(1);
695 static void _dup_value(struct type *type,
696 struct value *vold, struct value *vnew)
698 switch (type->vtype) {
699 case Vnone: // NOTEST
702 vnew->label = vold->label;
705 vnew->bool = vold->bool;
709 mpq_set(vnew->num, vold->num);
712 vnew->str.len = vold->str.len;
713 vnew->str.txt = malloc(vnew->str.len);
714 memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
719 static int _value_cmp(struct type *tl, struct type *tr,
720 struct value *left, struct value *right)
724 return tl - tr; // NOTEST
726 case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
727 case Vnum: cmp = mpq_cmp(left->num, right->num); break;
728 case Vstr: cmp = text_cmp(left->str, right->str); break;
729 case Vbool: cmp = left->bool - right->bool; break;
730 case Vnone: cmp = 0; // NOTEST
735 static void _print_value(struct type *type, struct value *v)
737 switch (type->vtype) {
738 case Vnone: // NOTEST
739 printf("*no-value*"); break; // NOTEST
740 case Vlabel: // NOTEST
741 printf("*label-%p*", v->label); break; // NOTEST
743 printf("%.*s", v->str.len, v->str.txt); break;
745 printf("%s", v->bool ? "True":"False"); break;
750 mpf_set_q(fl, v->num);
751 gmp_printf("%Fg", fl);
758 static void _free_value(struct type *type, struct value *v);
760 static struct type base_prototype = {
762 .print = _print_value,
763 .cmp_order = _value_cmp,
764 .cmp_eq = _value_cmp,
769 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
772 static struct type *add_base_type(struct parse_context *c, char *n,
773 enum vtype vt, int size)
775 struct text txt = { n, strlen(n) };
778 t = add_type(c, txt, &base_prototype);
781 t->align = size > sizeof(void*) ? sizeof(void*) : size;
782 if (t->size & (t->align - 1))
783 t->size = (t->size | (t->align - 1)) + 1; // NOTEST
787 ###### context initialization
789 Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
790 Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
791 Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
792 Tnone = add_base_type(&context, "none", Vnone, 0);
793 Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
797 Variables are scoped named values. We store the names in a linked list
798 of "bindings" sorted in lexical order, and use sequential search and
805 struct binding *next; // in lexical order
809 This linked list is stored in the parse context so that "reduce"
810 functions can find or add variables, and so the analysis phase can
811 ensure that every variable gets a type.
815 struct binding *varlist; // In lexical order
819 static struct binding *find_binding(struct parse_context *c, struct text s)
821 struct binding **l = &c->varlist;
826 (cmp = text_cmp((*l)->name, s)) < 0)
830 n = calloc(1, sizeof(*n));
837 Each name can be linked to multiple variables defined in different
838 scopes. Each scope starts where the name is declared and continues
839 until the end of the containing code block. Scopes of a given name
840 cannot nest, so a declaration while a name is in-scope is an error.
842 ###### binding fields
843 struct variable *var;
847 struct variable *previous;
849 struct binding *name;
850 struct exec *where_decl;// where name was declared
851 struct exec *where_set; // where type was set
855 When a scope closes, the values of the variables might need to be freed.
856 This happens in the context of some `struct exec` and each `exec` will
857 need to know which variables need to be freed when it completes.
860 struct variable *to_free;
862 ####### variable fields
863 struct exec *cleanup_exec;
864 struct variable *next_free;
866 ####### interp exec cleanup
869 for (v = e->to_free; v; v = v->next_free) {
870 struct value *val = var_value(c, v);
871 free_value(v->type, val);
876 static void variable_unlink_exec(struct variable *v)
878 struct variable **vp;
879 if (!v->cleanup_exec)
881 for (vp = &v->cleanup_exec->to_free;
882 *vp; vp = &(*vp)->next_free) {
886 v->cleanup_exec = NULL;
891 While the naming seems strange, we include local constants in the
892 definition of variables. A name declared `var := value` can
893 subsequently be changed, but a name declared `var ::= value` cannot -
896 ###### variable fields
899 Scopes in parallel branches can be partially merged. More
900 specifically, if a given name is declared in both branches of an
901 if/else then its scope is a candidate for merging. Similarly if
902 every branch of an exhaustive switch (e.g. has an "else" clause)
903 declares a given name, then the scopes from the branches are
904 candidates for merging.
906 Note that names declared inside a loop (which is only parallel to
907 itself) are never visible after the loop. Similarly names defined in
908 scopes which are not parallel, such as those started by `for` and
909 `switch`, are never visible after the scope. Only variables defined in
910 both `then` and `else` (including the implicit then after an `if`, and
911 excluding `then` used with `for`) and in all `case`s and `else` of a
912 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
914 Labels, which are a bit like variables, follow different rules.
915 Labels are not explicitly declared, but if an undeclared name appears
916 in a context where a label is legal, that effectively declares the
917 name as a label. The declaration remains in force (or in scope) at
918 least to the end of the immediately containing block and conditionally
919 in any larger containing block which does not declare the name in some
920 other way. Importantly, the conditional scope extension happens even
921 if the label is only used in one parallel branch of a conditional --
922 when used in one branch it is treated as having been declared in all
925 Merge candidates are tentatively visible beyond the end of the
926 branching statement which creates them. If the name is used, the
927 merge is affirmed and they become a single variable visible at the
928 outer layer. If not - if it is redeclared first - the merge lapses.
930 To track scopes we have an extra stack, implemented as a linked list,
931 which roughly parallels the parse stack and which is used exclusively
932 for scoping. When a new scope is opened, a new frame is pushed and
933 the child-count of the parent frame is incremented. This child-count
934 is used to distinguish between the first of a set of parallel scopes,
935 in which declared variables must not be in scope, and subsequent
936 branches, whether they may already be conditionally scoped.
938 To push a new frame *before* any code in the frame is parsed, we need a
939 grammar reduction. This is most easily achieved with a grammar
940 element which derives the empty string, and creates the new scope when
941 it is recognised. This can be placed, for example, between a keyword
942 like "if" and the code following it.
946 struct scope *parent;
952 struct scope *scope_stack;
955 static void scope_pop(struct parse_context *c)
957 struct scope *s = c->scope_stack;
959 c->scope_stack = s->parent;
964 static void scope_push(struct parse_context *c)
966 struct scope *s = calloc(1, sizeof(*s));
968 c->scope_stack->child_count += 1;
969 s->parent = c->scope_stack;
977 OpenScope -> ${ scope_push(c); }$
979 Each variable records a scope depth and is in one of four states:
981 - "in scope". This is the case between the declaration of the
982 variable and the end of the containing block, and also between
983 the usage with affirms a merge and the end of that block.
985 The scope depth is not greater than the current parse context scope
986 nest depth. When the block of that depth closes, the state will
987 change. To achieve this, all "in scope" variables are linked
988 together as a stack in nesting order.
990 - "pending". The "in scope" block has closed, but other parallel
991 scopes are still being processed. So far, every parallel block at
992 the same level that has closed has declared the name.
994 The scope depth is the depth of the last parallel block that
995 enclosed the declaration, and that has closed.
997 - "conditionally in scope". The "in scope" block and all parallel
998 scopes have closed, and no further mention of the name has been seen.
999 This state includes a secondary nest depth (`min_depth`) which records
1000 the outermost scope seen since the variable became conditionally in
1001 scope. If a use of the name is found, the variable becomes "in scope"
1002 and that secondary depth becomes the recorded scope depth. If the
1003 name is declared as a new variable, the old variable becomes "out of
1004 scope" and the recorded scope depth stays unchanged.
1006 - "out of scope". The variable is neither in scope nor conditionally
1007 in scope. It is permanently out of scope now and can be removed from
1008 the "in scope" stack.
1010 ###### variable fields
1011 int depth, min_depth;
1012 enum { OutScope, PendingScope, CondScope, InScope } scope;
1013 struct variable *in_scope;
1015 ###### parse context
1017 struct variable *in_scope;
1019 All variables with the same name are linked together using the
1020 'previous' link. Those variable that have been affirmatively merged all
1021 have a 'merged' pointer that points to one primary variable - the most
1022 recently declared instance. When merging variables, we need to also
1023 adjust the 'merged' pointer on any other variables that had previously
1024 been merged with the one that will no longer be primary.
1026 A variable that is no longer the most recent instance of a name may
1027 still have "pending" scope, if it might still be merged with most
1028 recent instance. These variables don't really belong in the
1029 "in_scope" list, but are not immediately removed when a new instance
1030 is found. Instead, they are detected and ignored when considering the
1031 list of in_scope names.
1033 The storage of the value of a variable will be described later. For now
1034 we just need to know that when a variable goes out of scope, it might
1035 need to be freed. For this we need to be able to find it, so assume that
1036 `var_value()` will provide that.
1038 ###### variable fields
1039 struct variable *merged;
1041 ###### ast functions
1043 static void variable_merge(struct variable *primary, struct variable *secondary)
1047 primary = primary->merged;
1049 for (v = primary->previous; v; v=v->previous)
1050 if (v == secondary || v == secondary->merged ||
1051 v->merged == secondary ||
1052 v->merged == secondary->merged) {
1053 v->scope = OutScope;
1054 v->merged = primary;
1055 variable_unlink_exec(v);
1059 ###### forward decls
1060 static struct value *var_value(struct parse_context *c, struct variable *v);
1062 ###### free global vars
1064 while (context.varlist) {
1065 struct binding *b = context.varlist;
1066 struct variable *v = b->var;
1067 context.varlist = b->next;
1070 struct variable *next = v->previous;
1073 free_value(v->type, var_value(&context, v));
1075 // This is a global constant
1076 free_exec(v->where_decl);
1083 #### Manipulating Bindings
1085 When a name is conditionally visible, a new declaration discards the
1086 old binding - the condition lapses. Conversely a usage of the name
1087 affirms the visibility and extends it to the end of the containing
1088 block - i.e. the block that contains both the original declaration and
1089 the latest usage. This is determined from `min_depth`. When a
1090 conditionally visible variable gets affirmed like this, it is also
1091 merged with other conditionally visible variables with the same name.
1093 When we parse a variable declaration we either report an error if the
1094 name is currently bound, or create a new variable at the current nest
1095 depth if the name is unbound or bound to a conditionally scoped or
1096 pending-scope variable. If the previous variable was conditionally
1097 scoped, it and its homonyms becomes out-of-scope.
1099 When we parse a variable reference (including non-declarative assignment
1100 "foo = bar") we report an error if the name is not bound or is bound to
1101 a pending-scope variable; update the scope if the name is bound to a
1102 conditionally scoped variable; or just proceed normally if the named
1103 variable is in scope.
1105 When we exit a scope, any variables bound at this level are either
1106 marked out of scope or pending-scoped, depending on whether the scope
1107 was sequential or parallel. Here a "parallel" scope means the "then"
1108 or "else" part of a conditional, or any "case" or "else" branch of a
1109 switch. Other scopes are "sequential".
1111 When exiting a parallel scope we check if there are any variables that
1112 were previously pending and are still visible. If there are, then
1113 they weren't redeclared in the most recent scope, so they cannot be
1114 merged and must become out-of-scope. If it is not the first of
1115 parallel scopes (based on `child_count`), we check that there was a
1116 previous binding that is still pending-scope. If there isn't, the new
1117 variable must now be out-of-scope.
1119 When exiting a sequential scope that immediately enclosed parallel
1120 scopes, we need to resolve any pending-scope variables. If there was
1121 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1122 we need to mark all pending-scope variable as out-of-scope. Otherwise
1123 all pending-scope variables become conditionally scoped.
1126 enum closetype { CloseSequential, CloseParallel, CloseElse };
1128 ###### ast functions
1130 static struct variable *var_decl(struct parse_context *c, struct text s)
1132 struct binding *b = find_binding(c, s);
1133 struct variable *v = b->var;
1135 switch (v ? v->scope : OutScope) {
1137 /* Caller will report the error */
1141 v && v->scope == CondScope;
1143 v->scope = OutScope;
1147 v = calloc(1, sizeof(*v));
1148 v->previous = b->var;
1152 v->min_depth = v->depth = c->scope_depth;
1154 v->in_scope = c->in_scope;
1160 static struct variable *var_ref(struct parse_context *c, struct text s)
1162 struct binding *b = find_binding(c, s);
1163 struct variable *v = b->var;
1164 struct variable *v2;
1166 switch (v ? v->scope : OutScope) {
1169 /* Caller will report the error */
1172 /* All CondScope variables of this name need to be merged
1173 * and become InScope
1175 v->depth = v->min_depth;
1177 for (v2 = v->previous;
1178 v2 && v2->scope == CondScope;
1180 variable_merge(v, v2);
1188 static void var_block_close(struct parse_context *c, enum closetype ct,
1191 /* Close off all variables that are in_scope.
1192 * Some variables in c->scope may already be not-in-scope,
1193 * such as when a PendingScope variable is hidden by a new
1194 * variable with the same name.
1195 * So we check for v->name->var != v and drop them.
1196 * If we choose to make a variable OutScope, we drop it
1199 struct variable *v, **vp, *v2;
1202 for (vp = &c->in_scope;
1203 (v = *vp) && v->min_depth > c->scope_depth;
1204 (v->scope == OutScope || v->name->var != v)
1205 ? (*vp = v->in_scope, 0)
1206 : ( vp = &v->in_scope, 0)) {
1207 v->min_depth = c->scope_depth;
1208 if (v->name->var != v)
1209 /* This is still in scope, but we haven't just
1213 v->min_depth = c->scope_depth;
1214 if (v->scope == InScope && e) {
1215 /* This variable gets cleaned up when 'e' finishes */
1216 variable_unlink_exec(v);
1217 v->cleanup_exec = e;
1218 v->next_free = e->to_free;
1223 case CloseParallel: /* handle PendingScope */
1227 if (c->scope_stack->child_count == 1)
1228 /* first among parallel branches */
1229 v->scope = PendingScope;
1230 else if (v->previous &&
1231 v->previous->scope == PendingScope)
1232 /* all previous branches used name */
1233 v->scope = PendingScope;
1234 else if (v->type == Tlabel)
1235 /* Labels remain pending even when not used */
1236 v->scope = PendingScope; // UNTESTED
1238 v->scope = OutScope;
1239 if (ct == CloseElse) {
1240 /* All Pending variables with this name
1241 * are now Conditional */
1243 v2 && v2->scope == PendingScope;
1245 v2->scope = CondScope;
1249 /* Not possible as it would require
1250 * parallel scope to be nested immediately
1251 * in a parallel scope, and that never
1255 /* Not possible as we already tested for
1261 case CloseSequential:
1262 if (v->type == Tlabel)
1263 v->scope = PendingScope;
1266 v->scope = OutScope;
1269 /* There was no 'else', so we can only become
1270 * conditional if we know the cases were exhaustive,
1271 * and that doesn't mean anything yet.
1272 * So only labels become conditional..
1275 v2 && v2->scope == PendingScope;
1277 if (v2->type == Tlabel)
1278 v2->scope = CondScope;
1280 v2->scope = OutScope;
1283 case OutScope: break;
1292 The value of a variable is store separately from the variable, on an
1293 analogue of a stack frame. There are (currently) two frames that can be
1294 active. A global frame which currently only stores constants, and a
1295 stacked frame which stores local variables. Each variable knows if it
1296 is global or not, and what its index into the frame is.
1298 Values in the global frame are known immediately they are relevant, so
1299 the frame needs to be reallocated as it grows so it can store those
1300 values. The local frame doesn't get values until the interpreted phase
1301 is started, so there is no need to allocate until the size is known.
1303 We initialize the `frame_pos` to an impossible value, so that we can
1304 tell if it was set or not later.
1306 ###### variable fields
1310 ###### variable init
1313 ###### parse context
1315 short global_size, global_alloc;
1317 void *global, *local;
1319 ###### ast functions
1321 static struct value *var_value(struct parse_context *c, struct variable *v)
1324 if (!c->local || !v->type)
1325 return NULL; // NOTEST
1326 if (v->frame_pos + v->type->size > c->local_size) {
1327 printf("INVALID frame_pos\n"); // NOTEST
1330 return c->local + v->frame_pos;
1332 if (c->global_size > c->global_alloc) {
1333 int old = c->global_alloc;
1334 c->global_alloc = (c->global_size | 1023) + 1024;
1335 c->global = realloc(c->global, c->global_alloc);
1336 memset(c->global + old, 0, c->global_alloc - old);
1338 return c->global + v->frame_pos;
1341 static struct value *global_alloc(struct parse_context *c, struct type *t,
1342 struct variable *v, struct value *init)
1345 struct variable scratch;
1347 if (t->prepare_type)
1348 t->prepare_type(c, t, 1); // NOTEST
1350 if (c->global_size & (t->align - 1))
1351 c->global_size = (c->global_size + t->align) & ~(t->align-1);
1356 v->frame_pos = c->global_size;
1358 c->global_size += v->type->size;
1359 ret = var_value(c, v);
1361 memcpy(ret, init, t->size);
1367 As global values are found -- struct field initializers, labels etc --
1368 `global_alloc()` is called to record the value in the global frame.
1370 When the program is fully parsed, we need to walk the list of variables
1371 to find any that weren't merged away and that aren't global, and to
1372 calculate the frame size and assign a frame position for each
1373 variable. For this we have `scope_finalize()`.
1375 ###### ast functions
1377 static int scope_finalize(struct parse_context *c)
1382 for (b = c->varlist; b; b = b->next) {
1384 for (v = b->var; v; v = v->previous) {
1385 struct type *t = v->type;
1392 if (size & (t->align - 1))
1393 size = (size + t->align) & ~(t->align-1);
1394 v->frame_pos = size;
1395 size += v->type->size;
1401 ###### free context storage
1402 free(context.global);
1406 Executables can be lots of different things. In many cases an
1407 executable is just an operation combined with one or two other
1408 executables. This allows for expressions and lists etc. Other times an
1409 executable is something quite specific like a constant or variable name.
1410 So we define a `struct exec` to be a general executable with a type, and
1411 a `struct binode` which is a subclass of `exec`, forms a node in a
1412 binary tree, and holds an operation. There will be other subclasses,
1413 and to access these we need to be able to `cast` the `exec` into the
1414 various other types. The first field in any `struct exec` is the type
1415 from the `exec_types` enum.
1418 #define cast(structname, pointer) ({ \
1419 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1420 if (__mptr && *__mptr != X##structname) abort(); \
1421 (struct structname *)( (char *)__mptr);})
1423 #define new(structname) ({ \
1424 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1425 __ptr->type = X##structname; \
1426 __ptr->line = -1; __ptr->column = -1; \
1429 #define new_pos(structname, token) ({ \
1430 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1431 __ptr->type = X##structname; \
1432 __ptr->line = token.line; __ptr->column = token.col; \
1441 enum exec_types type;
1450 struct exec *left, *right;
1453 ###### ast functions
1455 static int __fput_loc(struct exec *loc, FILE *f)
1459 if (loc->line >= 0) {
1460 fprintf(f, "%d:%d: ", loc->line, loc->column);
1463 if (loc->type == Xbinode)
1464 return __fput_loc(cast(binode,loc)->left, f) ||
1465 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1468 static void fput_loc(struct exec *loc, FILE *f)
1470 if (!__fput_loc(loc, f))
1471 fprintf(f, "??:??: ");
1474 Each different type of `exec` node needs a number of functions defined,
1475 a bit like methods. We must be able to free it, print it, analyse it
1476 and execute it. Once we have specific `exec` types we will need to
1477 parse them too. Let's take this a bit more slowly.
1481 The parser generator requires a `free_foo` function for each struct
1482 that stores attributes and they will often be `exec`s and subtypes
1483 there-of. So we need `free_exec` which can handle all the subtypes,
1484 and we need `free_binode`.
1486 ###### ast functions
1488 static void free_binode(struct binode *b)
1493 free_exec(b->right);
1497 ###### core functions
1498 static void free_exec(struct exec *e)
1507 ###### forward decls
1509 static void free_exec(struct exec *e);
1511 ###### free exec cases
1512 case Xbinode: free_binode(cast(binode, e)); break;
1516 Printing an `exec` requires that we know the current indent level for
1517 printing line-oriented components. As will become clear later, we
1518 also want to know what sort of bracketing to use.
1520 ###### ast functions
1522 static void do_indent(int i, char *str)
1529 ###### core functions
1530 static void print_binode(struct binode *b, int indent, int bracket)
1534 ## print binode cases
1538 static void print_exec(struct exec *e, int indent, int bracket)
1544 print_binode(cast(binode, e), indent, bracket); break;
1549 do_indent(indent, "/* FREE");
1550 for (v = e->to_free; v; v = v->next_free) {
1551 printf(" %.*s", v->name->name.len, v->name->name.txt);
1552 if (v->frame_pos >= 0)
1553 printf("(%d+%d)", v->frame_pos,
1554 v->type ? v->type->size:0);
1560 ###### forward decls
1562 static void print_exec(struct exec *e, int indent, int bracket);
1566 As discussed, analysis involves propagating type requirements around the
1567 program and looking for errors.
1569 So `propagate_types` is passed an expected type (being a `struct type`
1570 pointer together with some `val_rules` flags) that the `exec` is
1571 expected to return, and returns the type that it does return, either
1572 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1573 by reference. It is set to `0` when an error is found, and `2` when
1574 any change is made. If it remains unchanged at `1`, then no more
1575 propagation is needed.
1579 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 1<<2};
1583 if (rules & Rnolabel)
1584 fputs(" (labels not permitted)", stderr);
1587 ###### forward decls
1588 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1589 struct type *type, int rules);
1590 ###### core functions
1592 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1593 struct type *type, int rules)
1600 switch (prog->type) {
1603 struct binode *b = cast(binode, prog);
1605 ## propagate binode cases
1609 ## propagate exec cases
1614 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1615 struct type *type, int rules)
1617 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1626 Interpreting an `exec` doesn't require anything but the `exec`. State
1627 is stored in variables and each variable will be directly linked from
1628 within the `exec` tree. The exception to this is the `main` function
1629 which needs to look at command line arguments. This function will be
1630 interpreted separately.
1632 Each `exec` can return a value combined with a type in `struct lrval`.
1633 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1634 the location of a value, which can be updated, in `lval`. Others will
1635 set `lval` to NULL indicating that there is a value of appropriate type
1638 ###### core functions
1642 struct value rval, *lval;
1645 static struct lrval _interp_exec(struct parse_context *c, struct exec *e);
1647 static struct value interp_exec(struct parse_context *c, struct exec *e,
1648 struct type **typeret)
1650 struct lrval ret = _interp_exec(c, e);
1652 if (!ret.type) abort();
1654 *typeret = ret.type;
1656 dup_value(ret.type, ret.lval, &ret.rval);
1660 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1661 struct type **typeret)
1663 struct lrval ret = _interp_exec(c, e);
1666 *typeret = ret.type;
1668 free_value(ret.type, &ret.rval);
1672 static struct lrval _interp_exec(struct parse_context *c, struct exec *e)
1675 struct value rv = {}, *lrv = NULL;
1676 struct type *rvtype;
1678 rvtype = ret.type = Tnone;
1688 struct binode *b = cast(binode, e);
1689 struct value left, right, *lleft;
1690 struct type *ltype, *rtype;
1691 ltype = rtype = Tnone;
1693 ## interp binode cases
1695 free_value(ltype, &left);
1696 free_value(rtype, &right);
1699 ## interp exec cases
1704 ## interp exec cleanup
1710 Now that we have the shape of the interpreter in place we can add some
1711 complex types and connected them in to the data structures and the
1712 different phases of parse, analyse, print, interpret.
1714 Thus far we have arrays and structs.
1718 Arrays can be declared by giving a size and a type, as `[size]type' so
1719 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1720 size can be either a literal number, or a named constant. Some day an
1721 arbitrary expression will be supported.
1723 As a formal parameter to a function, the array can be declared with a
1724 new variable as the size: `name:[size::number]string`. The `size`
1725 variable is set to the size of the array and must be a constant. As
1726 `number` is the only supported type, it can be left out:
1727 `name:[size::]string`.
1729 Arrays cannot be assigned. When pointers are introduced we will also
1730 introduce array slices which can refer to part or all of an array -
1731 the assignment syntax will create a slice. For now, an array can only
1732 ever be referenced by the name it is declared with. It is likely that
1733 a "`copy`" primitive will eventually be define which can be used to
1734 make a copy of an array with controllable recursive depth.
1736 For now we have two sorts of array, those with fixed size either because
1737 it is given as a literal number or because it is a struct member (which
1738 cannot have a runtime-changing size), and those with a size that is
1739 determined at runtime - local variables with a const size. The former
1740 have their size calculated at parse time, the latter at run time.
1742 For the latter type, the `size` field of the type is the size of a
1743 pointer, and the array is reallocated every time it comes into scope.
1745 We differentiate struct fields with a const size from local variables
1746 with a const size by whether they are prepared at parse time or not.
1748 ###### type union fields
1751 int unspec; // size is unspecified - vsize must be set.
1754 struct variable *vsize;
1755 struct type *member;
1758 ###### value union fields
1759 void *array; // used if not static_size
1761 ###### value functions
1763 static void array_prepare_type(struct parse_context *c, struct type *type,
1766 struct value *vsize;
1768 if (!type->array.vsize || type->array.static_size)
1771 vsize = var_value(c, type->array.vsize);
1773 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1774 type->array.size = mpz_get_si(q);
1778 type->array.static_size = 1;
1779 type->size = type->array.size * type->array.member->size;
1780 type->align = type->array.member->align;
1784 static void array_init(struct type *type, struct value *val)
1787 void *ptr = val->ptr;
1791 if (!type->array.static_size) {
1792 val->array = calloc(type->array.size,
1793 type->array.member->size);
1796 for (i = 0; i < type->array.size; i++) {
1798 v = (void*)ptr + i * type->array.member->size;
1799 val_init(type->array.member, v);
1803 static void array_free(struct type *type, struct value *val)
1806 void *ptr = val->ptr;
1808 if (!type->array.static_size)
1810 for (i = 0; i < type->array.size; i++) {
1812 v = (void*)ptr + i * type->array.member->size;
1813 free_value(type->array.member, v);
1815 if (!type->array.static_size)
1819 static int array_compat(struct type *require, struct type *have)
1821 if (have->compat != require->compat)
1823 /* Both are arrays, so we can look at details */
1824 if (!type_compat(require->array.member, have->array.member, 0))
1826 if (have->array.unspec && require->array.unspec) {
1827 if (have->array.vsize && require->array.vsize &&
1828 have->array.vsize != require->array.vsize) // UNTESTED
1829 /* sizes might not be the same */
1830 return 0; // UNTESTED
1833 if (have->array.unspec || require->array.unspec)
1834 return 1; // UNTESTED
1835 if (require->array.vsize == NULL && have->array.vsize == NULL)
1836 return require->array.size == have->array.size;
1838 return require->array.vsize == have->array.vsize; // UNTESTED
1841 static void array_print_type(struct type *type, FILE *f)
1844 if (type->array.vsize) {
1845 struct binding *b = type->array.vsize->name;
1846 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1847 type->array.unspec ? "::" : "");
1849 fprintf(f, "%d]", type->array.size);
1850 type_print(type->array.member, f);
1853 static struct type array_prototype = {
1855 .prepare_type = array_prepare_type,
1856 .print_type = array_print_type,
1857 .compat = array_compat,
1859 .size = sizeof(void*),
1860 .align = sizeof(void*),
1863 ###### declare terminals
1868 | [ NUMBER ] Type ${ {
1871 struct text noname = { "", 0 };
1874 $0 = t = add_type(c, noname, &array_prototype);
1875 t->array.member = $<4;
1876 t->array.vsize = NULL;
1877 if (number_parse(num, tail, $2.txt) == 0)
1878 tok_err(c, "error: unrecognised number", &$2);
1880 tok_err(c, "error: unsupported number suffix", &$2);
1883 t->array.size = mpz_get_ui(mpq_numref(num));
1884 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1885 tok_err(c, "error: array size must be an integer",
1887 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1888 tok_err(c, "error: array size is too large",
1892 t->array.static_size = 1;
1893 t->size = t->array.size * t->array.member->size;
1894 t->align = t->array.member->align;
1897 | [ IDENTIFIER ] Type ${ {
1898 struct variable *v = var_ref(c, $2.txt);
1899 struct text noname = { "", 0 };
1902 tok_err(c, "error: name undeclared", &$2);
1903 else if (!v->constant)
1904 tok_err(c, "error: array size must be a constant", &$2);
1906 $0 = add_type(c, noname, &array_prototype);
1907 $0->array.member = $<4;
1909 $0->array.vsize = v;
1914 OptType -> Type ${ $0 = $<1; }$
1917 ###### formal type grammar
1919 | [ IDENTIFIER :: OptType ] Type ${ {
1920 struct variable *v = var_decl(c, $ID.txt);
1921 struct text noname = { "", 0 };
1927 $0 = add_type(c, noname, &array_prototype);
1928 $0->array.member = $<6;
1930 $0->array.unspec = 1;
1931 $0->array.vsize = v;
1937 ###### variable grammar
1939 | Variable [ Expression ] ${ {
1940 struct binode *b = new(binode);
1947 ###### print binode cases
1949 print_exec(b->left, -1, bracket);
1951 print_exec(b->right, -1, bracket);
1955 ###### propagate binode cases
1957 /* left must be an array, right must be a number,
1958 * result is the member type of the array
1960 propagate_types(b->right, c, ok, Tnum, 0);
1961 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
1962 if (!t || t->compat != array_compat) {
1963 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
1966 if (!type_compat(type, t->array.member, rules)) {
1967 type_err(c, "error: have %1 but need %2", prog,
1968 t->array.member, rules, type);
1970 return t->array.member;
1974 ###### interp binode cases
1980 lleft = linterp_exec(c, b->left, <ype);
1981 right = interp_exec(c, b->right, &rtype);
1983 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
1987 if (ltype->array.static_size)
1990 ptr = *(void**)lleft;
1991 rvtype = ltype->array.member;
1992 if (i >= 0 && i < ltype->array.size)
1993 lrv = ptr + i * rvtype->size;
1995 val_init(ltype->array.member, &rv);
2002 A `struct` is a data-type that contains one or more other data-types.
2003 It differs from an array in that each member can be of a different
2004 type, and they are accessed by name rather than by number. Thus you
2005 cannot choose an element by calculation, you need to know what you
2008 The language makes no promises about how a given structure will be
2009 stored in memory - it is free to rearrange fields to suit whatever
2010 criteria seems important.
2012 Structs are declared separately from program code - they cannot be
2013 declared in-line in a variable declaration like arrays can. A struct
2014 is given a name and this name is used to identify the type - the name
2015 is not prefixed by the word `struct` as it would be in C.
2017 Structs are only treated as the same if they have the same name.
2018 Simply having the same fields in the same order is not enough. This
2019 might change once we can create structure initializers from a list of
2022 Each component datum is identified much like a variable is declared,
2023 with a name, one or two colons, and a type. The type cannot be omitted
2024 as there is no opportunity to deduce the type from usage. An initial
2025 value can be given following an equals sign, so
2027 ##### Example: a struct type
2033 would declare a type called "complex" which has two number fields,
2034 each initialised to zero.
2036 Struct will need to be declared separately from the code that uses
2037 them, so we will need to be able to print out the declaration of a
2038 struct when reprinting the whole program. So a `print_type_decl` type
2039 function will be needed.
2041 ###### type union fields
2053 ###### type functions
2054 void (*print_type_decl)(struct type *type, FILE *f);
2056 ###### value functions
2058 static void structure_init(struct type *type, struct value *val)
2062 for (i = 0; i < type->structure.nfields; i++) {
2064 v = (void*) val->ptr + type->structure.fields[i].offset;
2065 if (type->structure.fields[i].init)
2066 dup_value(type->structure.fields[i].type,
2067 type->structure.fields[i].init,
2070 val_init(type->structure.fields[i].type, v);
2074 static void structure_free(struct type *type, struct value *val)
2078 for (i = 0; i < type->structure.nfields; i++) {
2080 v = (void*)val->ptr + type->structure.fields[i].offset;
2081 free_value(type->structure.fields[i].type, v);
2085 static void structure_free_type(struct type *t)
2088 for (i = 0; i < t->structure.nfields; i++)
2089 if (t->structure.fields[i].init) {
2090 free_value(t->structure.fields[i].type,
2091 t->structure.fields[i].init);
2093 free(t->structure.fields);
2096 static struct type structure_prototype = {
2097 .init = structure_init,
2098 .free = structure_free,
2099 .free_type = structure_free_type,
2100 .print_type_decl = structure_print_type,
2114 ###### free exec cases
2116 free_exec(cast(fieldref, e)->left);
2120 ###### declare terminals
2123 ###### variable grammar
2125 | Variable . IDENTIFIER ${ {
2126 struct fieldref *fr = new_pos(fieldref, $2);
2133 ###### print exec cases
2137 struct fieldref *f = cast(fieldref, e);
2138 print_exec(f->left, -1, bracket);
2139 printf(".%.*s", f->name.len, f->name.txt);
2143 ###### ast functions
2144 static int find_struct_index(struct type *type, struct text field)
2147 for (i = 0; i < type->structure.nfields; i++)
2148 if (text_cmp(type->structure.fields[i].name, field) == 0)
2153 ###### propagate exec cases
2157 struct fieldref *f = cast(fieldref, prog);
2158 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2161 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2163 else if (st->init != structure_init)
2164 type_err(c, "error: field reference attempted on %1, not a struct",
2165 f->left, st, 0, NULL);
2166 else if (f->index == -2) {
2167 f->index = find_struct_index(st, f->name);
2169 type_err(c, "error: cannot find requested field in %1",
2170 f->left, st, 0, NULL);
2172 if (f->index >= 0) {
2173 struct type *ft = st->structure.fields[f->index].type;
2174 if (!type_compat(type, ft, rules))
2175 type_err(c, "error: have %1 but need %2", prog,
2182 ###### interp exec cases
2185 struct fieldref *f = cast(fieldref, e);
2187 struct value *lleft = linterp_exec(c, f->left, <ype);
2188 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2189 rvtype = ltype->structure.fields[f->index].type;
2195 struct fieldlist *prev;
2199 ###### ast functions
2200 static void free_fieldlist(struct fieldlist *f)
2204 free_fieldlist(f->prev);
2206 free_value(f->f.type, f->f.init); // UNTESTED
2207 free(f->f.init); // UNTESTED
2212 ###### top level grammar
2213 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2215 add_type(c, $2.txt, &structure_prototype);
2217 struct fieldlist *f;
2219 for (f = $3; f; f=f->prev)
2222 t->structure.nfields = cnt;
2223 t->structure.fields = calloc(cnt, sizeof(struct field));
2226 int a = f->f.type->align;
2228 t->structure.fields[cnt] = f->f;
2229 if (t->size & (a-1))
2230 t->size = (t->size | (a-1)) + 1;
2231 t->structure.fields[cnt].offset = t->size;
2232 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2241 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2242 | { SimpleFieldList } ${ $0 = $<SFL; }$
2243 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2244 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2246 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2247 | FieldLines SimpleFieldList Newlines ${
2252 SimpleFieldList -> Field ${ $0 = $<F; }$
2253 | SimpleFieldList ; Field ${
2257 | SimpleFieldList ; ${
2260 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2262 Field -> IDENTIFIER : Type = Expression ${ {
2265 $0 = calloc(1, sizeof(struct fieldlist));
2266 $0->f.name = $1.txt;
2271 propagate_types($<5, c, &ok, $3, 0);
2274 c->parse_error = 1; // UNTESTED
2276 struct value vl = interp_exec(c, $5, NULL);
2277 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2280 | IDENTIFIER : Type ${
2281 $0 = calloc(1, sizeof(struct fieldlist));
2282 $0->f.name = $1.txt;
2284 if ($0->f.type->prepare_type)
2285 $0->f.type->prepare_type(c, $0->f.type, 1);
2288 ###### forward decls
2289 static void structure_print_type(struct type *t, FILE *f);
2291 ###### value functions
2292 static void structure_print_type(struct type *t, FILE *f) // UNTESTED
2296 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2298 for (i = 0; i < t->structure.nfields; i++) {
2299 struct field *fl = t->structure.fields + i;
2300 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2301 type_print(fl->type, f);
2302 if (fl->type->print && fl->init) {
2304 if (fl->type == Tstr)
2305 fprintf(f, "\""); // UNTESTED
2306 print_value(fl->type, fl->init);
2307 if (fl->type == Tstr)
2308 fprintf(f, "\""); // UNTESTED
2314 ###### print type decls
2316 struct type *t; // UNTESTED
2319 while (target != 0) {
2321 for (t = context.typelist; t ; t=t->next)
2322 if (t->print_type_decl && !t->check_args) {
2331 t->print_type_decl(t, stdout);
2339 A function is a chunk of code which can be passed parameters and can
2340 return results. Each function has a type which includes the set of
2341 parameters and the return value. As yet these types cannot be declared
2342 separately from the function itself.
2344 The parameters can be specified either in parentheses as a ';' separated
2347 ##### Example: function 1
2349 func main(av:[ac::number]string; env:[envc::number]string)
2352 or as an indented list of one parameter per line (though each line can
2353 be a ';' separated list)
2355 ##### Example: function 2
2358 argv:[argc::number]string
2359 env:[envc::number]string
2363 In the first case a return type can follow the paentheses after a colon,
2364 in the second it is given on a line starting with the word `return`.
2366 ##### Example: functions that return
2368 func add(a:number; b:number): number
2379 For constructing these lists we use a `List` binode, which will be
2380 further detailed when Expression Lists are introduced.
2382 ###### type union fields
2385 struct binode *params;
2386 struct type *return_type;
2390 ###### value union fields
2391 struct exec *function;
2393 ###### type functions
2394 void (*check_args)(struct parse_context *c, int *ok,
2395 struct type *require, struct exec *args);
2397 ###### value functions
2399 static void function_free(struct type *type, struct value *val)
2401 free_exec(val->function);
2402 val->function = NULL;
2405 static int function_compat(struct type *require, struct type *have)
2407 // FIXME can I do anything here yet?
2411 static void function_check_args(struct parse_context *c, int *ok,
2412 struct type *require, struct exec *args)
2414 /* This should be 'compat', but we don't have a 'tuple' type to
2415 * hold the type of 'args'
2417 struct binode *arg = cast(binode, args);
2418 struct binode *param = require->function.params;
2421 struct var *pv = cast(var, param->left);
2423 type_err(c, "error: insufficient arguments to function.",
2424 args, NULL, 0, NULL);
2428 propagate_types(arg->left, c, ok, pv->var->type, 0);
2429 param = cast(binode, param->right);
2430 arg = cast(binode, arg->right);
2433 type_err(c, "error: too many arguments to function.",
2434 args, NULL, 0, NULL);
2437 static void function_print(struct type *type, struct value *val)
2439 print_exec(val->function, 1, 0);
2442 static void function_print_type_decl(struct type *type, FILE *f)
2446 for (b = type->function.params; b; b = cast(binode, b->right)) {
2447 struct variable *v = cast(var, b->left)->var;
2448 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
2449 v->constant ? "::" : ":");
2450 type_print(v->type, f);
2455 if (type->function.return_type != Tnone) {
2457 type_print(type->function.return_type, f);
2462 static void function_free_type(struct type *t)
2464 free_exec(t->function.params);
2467 static struct type function_prototype = {
2468 .size = sizeof(void*),
2469 .align = sizeof(void*),
2470 .free = function_free,
2471 .compat = function_compat,
2472 .check_args = function_check_args,
2473 .print = function_print,
2474 .print_type_decl = function_print_type_decl,
2475 .free_type = function_free_type,
2478 ###### declare terminals
2488 FuncName -> IDENTIFIER ${ {
2489 struct variable *v = var_decl(c, $1.txt);
2490 struct var *e = new_pos(var, $1);
2496 v = var_ref(c, $1.txt);
2498 type_err(c, "error: function '%v' redeclared",
2500 type_err(c, "info: this is where '%v' was first declared",
2501 v->where_decl, NULL, 0, NULL);
2507 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
2508 | Args ArgsLine NEWLINE ${ {
2509 struct binode *b = $<AL;
2510 struct binode **bp = &b;
2512 bp = (struct binode **)&(*bp)->left;
2517 ArgsLine -> ${ $0 = NULL; }$
2518 | Varlist ${ $0 = $<1; }$
2519 | Varlist ; ${ $0 = $<1; }$
2521 Varlist -> Varlist ; ArgDecl ${
2535 ArgDecl -> IDENTIFIER : FormalType ${ {
2536 struct variable *v = var_decl(c, $1.txt);
2542 ## Executables: the elements of code
2544 Each code element needs to be parsed, printed, analysed,
2545 interpreted, and freed. There are several, so let's just start with
2546 the easy ones and work our way up.
2550 We have already met values as separate objects. When manifest
2551 constants appear in the program text, that must result in an executable
2552 which has a constant value. So the `val` structure embeds a value in
2565 ###### ast functions
2566 struct val *new_val(struct type *T, struct token tk)
2568 struct val *v = new_pos(val, tk);
2579 $0 = new_val(Tbool, $1);
2583 $0 = new_val(Tbool, $1);
2587 $0 = new_val(Tnum, $1);
2590 if (number_parse($0->val.num, tail, $1.txt) == 0)
2591 mpq_init($0->val.num); // UNTESTED
2593 tok_err(c, "error: unsupported number suffix",
2598 $0 = new_val(Tstr, $1);
2601 string_parse(&$1, '\\', &$0->val.str, tail);
2603 tok_err(c, "error: unsupported string suffix",
2608 $0 = new_val(Tstr, $1);
2611 string_parse(&$1, '\\', &$0->val.str, tail);
2613 tok_err(c, "error: unsupported string suffix",
2618 ###### print exec cases
2621 struct val *v = cast(val, e);
2622 if (v->vtype == Tstr)
2624 print_value(v->vtype, &v->val);
2625 if (v->vtype == Tstr)
2630 ###### propagate exec cases
2633 struct val *val = cast(val, prog);
2634 if (!type_compat(type, val->vtype, rules))
2635 type_err(c, "error: expected %1%r found %2",
2636 prog, type, rules, val->vtype);
2640 ###### interp exec cases
2642 rvtype = cast(val, e)->vtype;
2643 dup_value(rvtype, &cast(val, e)->val, &rv);
2646 ###### ast functions
2647 static void free_val(struct val *v)
2650 free_value(v->vtype, &v->val);
2654 ###### free exec cases
2655 case Xval: free_val(cast(val, e)); break;
2657 ###### ast functions
2658 // Move all nodes from 'b' to 'rv', reversing their order.
2659 // In 'b' 'left' is a list, and 'right' is the last node.
2660 // In 'rv', left' is the first node and 'right' is a list.
2661 static struct binode *reorder_bilist(struct binode *b)
2663 struct binode *rv = NULL;
2666 struct exec *t = b->right;
2670 b = cast(binode, b->left);
2680 Just as we used a `val` to wrap a value into an `exec`, we similarly
2681 need a `var` to wrap a `variable` into an exec. While each `val`
2682 contained a copy of the value, each `var` holds a link to the variable
2683 because it really is the same variable no matter where it appears.
2684 When a variable is used, we need to remember to follow the `->merged`
2685 link to find the primary instance.
2693 struct variable *var;
2701 VariableDecl -> IDENTIFIER : ${ {
2702 struct variable *v = var_decl(c, $1.txt);
2703 $0 = new_pos(var, $1);
2708 v = var_ref(c, $1.txt);
2710 type_err(c, "error: variable '%v' redeclared",
2712 type_err(c, "info: this is where '%v' was first declared",
2713 v->where_decl, NULL, 0, NULL);
2716 | IDENTIFIER :: ${ {
2717 struct variable *v = var_decl(c, $1.txt);
2718 $0 = new_pos(var, $1);
2724 v = var_ref(c, $1.txt);
2726 type_err(c, "error: variable '%v' redeclared",
2728 type_err(c, "info: this is where '%v' was first declared",
2729 v->where_decl, NULL, 0, NULL);
2732 | IDENTIFIER : Type ${ {
2733 struct variable *v = var_decl(c, $1.txt);
2734 $0 = new_pos(var, $1);
2741 v = var_ref(c, $1.txt);
2743 type_err(c, "error: variable '%v' redeclared",
2745 type_err(c, "info: this is where '%v' was first declared",
2746 v->where_decl, NULL, 0, NULL);
2749 | IDENTIFIER :: Type ${ {
2750 struct variable *v = var_decl(c, $1.txt);
2751 $0 = new_pos(var, $1);
2759 v = var_ref(c, $1.txt);
2761 type_err(c, "error: variable '%v' redeclared",
2763 type_err(c, "info: this is where '%v' was first declared",
2764 v->where_decl, NULL, 0, NULL);
2769 Variable -> IDENTIFIER ${ {
2770 struct variable *v = var_ref(c, $1.txt);
2771 $0 = new_pos(var, $1);
2773 /* This might be a label - allocate a var just in case */
2774 v = var_decl(c, $1.txt);
2781 cast(var, $0)->var = v;
2785 ###### print exec cases
2788 struct var *v = cast(var, e);
2790 struct binding *b = v->var->name;
2791 printf("%.*s", b->name.len, b->name.txt);
2798 if (loc && loc->type == Xvar) {
2799 struct var *v = cast(var, loc);
2801 struct binding *b = v->var->name;
2802 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2804 fputs("???", stderr); // NOTEST
2806 fputs("NOTVAR", stderr);
2809 ###### propagate exec cases
2813 struct var *var = cast(var, prog);
2814 struct variable *v = var->var;
2816 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2817 return Tnone; // NOTEST
2820 if (v->constant && (rules & Rnoconstant)) {
2821 type_err(c, "error: Cannot assign to a constant: %v",
2822 prog, NULL, 0, NULL);
2823 type_err(c, "info: name was defined as a constant here",
2824 v->where_decl, NULL, 0, NULL);
2827 if (v->type == Tnone && v->where_decl == prog)
2828 type_err(c, "error: variable used but not declared: %v",
2829 prog, NULL, 0, NULL);
2830 if (v->type == NULL) {
2831 if (type && *ok != 0) {
2833 v->where_set = prog;
2838 if (!type_compat(type, v->type, rules)) {
2839 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2840 type, rules, v->type);
2841 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2842 v->type, rules, NULL);
2849 ###### interp exec cases
2852 struct var *var = cast(var, e);
2853 struct variable *v = var->var;
2856 lrv = var_value(c, v);
2861 ###### ast functions
2863 static void free_var(struct var *v)
2868 ###### free exec cases
2869 case Xvar: free_var(cast(var, e)); break;
2871 ### Expressions: Conditional
2873 Our first user of the `binode` will be conditional expressions, which
2874 is a bit odd as they actually have three components. That will be
2875 handled by having 2 binodes for each expression. The conditional
2876 expression is the lowest precedence operator which is why we define it
2877 first - to start the precedence list.
2879 Conditional expressions are of the form "value `if` condition `else`
2880 other_value". They associate to the right, so everything to the right
2881 of `else` is part of an else value, while only a higher-precedence to
2882 the left of `if` is the if values. Between `if` and `else` there is no
2883 room for ambiguity, so a full conditional expression is allowed in
2895 Expression -> Expression if Expression else Expression $$ifelse ${ {
2896 struct binode *b1 = new(binode);
2897 struct binode *b2 = new(binode);
2906 ## expression grammar
2908 ###### print binode cases
2911 b2 = cast(binode, b->right);
2912 if (bracket) printf("(");
2913 print_exec(b2->left, -1, bracket);
2915 print_exec(b->left, -1, bracket);
2917 print_exec(b2->right, -1, bracket);
2918 if (bracket) printf(")");
2921 ###### propagate binode cases
2924 /* cond must be Tbool, others must match */
2925 struct binode *b2 = cast(binode, b->right);
2928 propagate_types(b->left, c, ok, Tbool, 0);
2929 t = propagate_types(b2->left, c, ok, type, Rnolabel);
2930 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
2934 ###### interp binode cases
2937 struct binode *b2 = cast(binode, b->right);
2938 left = interp_exec(c, b->left, <ype);
2940 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
2942 rv = interp_exec(c, b2->right, &rvtype);
2948 We take a brief detour, now that we have expressions, to describe lists
2949 of expressions. These will be needed for function parameters and
2950 possibly other situations. They seem generic enough to introduce here
2951 to be used elsewhere.
2953 And ExpressionList will use the `List` type of `binode`, building up at
2954 the end. And place where they are used will probably call
2955 `reorder_bilist()` to get a more normal first/next arrangement.
2957 ###### declare terminals
2960 `List` execs have no implicit semantics, so they are never propagated or
2961 interpreted. The can be printed as a comma separate list, which is how
2962 they are parsed. Note they are also used for function formal parameter
2963 lists. In that case a separate function is used to print them.
2965 ###### print binode cases
2969 print_exec(b->left, -1, bracket);
2972 b = cast(binode, b->right);
2976 ###### propagate binode cases
2977 case List: abort(); // NOTEST
2978 ###### interp binode cases
2979 case List: abort(); // NOTEST
2984 ExpressionList -> ExpressionList , Expression ${
2997 ### Expressions: Boolean
2999 The next class of expressions to use the `binode` will be Boolean
3000 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3001 have same corresponding precendence. The difference is that they don't
3002 evaluate the second expression if not necessary.
3011 ###### expr precedence
3016 ###### expression grammar
3017 | Expression or Expression ${ {
3018 struct binode *b = new(binode);
3024 | Expression or else Expression ${ {
3025 struct binode *b = new(binode);
3032 | Expression and Expression ${ {
3033 struct binode *b = new(binode);
3039 | Expression and then Expression ${ {
3040 struct binode *b = new(binode);
3047 | not Expression ${ {
3048 struct binode *b = new(binode);
3054 ###### print binode cases
3056 if (bracket) printf("(");
3057 print_exec(b->left, -1, bracket);
3059 print_exec(b->right, -1, bracket);
3060 if (bracket) printf(")");
3063 if (bracket) printf("(");
3064 print_exec(b->left, -1, bracket);
3065 printf(" and then ");
3066 print_exec(b->right, -1, bracket);
3067 if (bracket) printf(")");
3070 if (bracket) printf("(");
3071 print_exec(b->left, -1, bracket);
3073 print_exec(b->right, -1, bracket);
3074 if (bracket) printf(")");
3077 if (bracket) printf("(");
3078 print_exec(b->left, -1, bracket);
3079 printf(" or else ");
3080 print_exec(b->right, -1, bracket);
3081 if (bracket) printf(")");
3084 if (bracket) printf("(");
3086 print_exec(b->right, -1, bracket);
3087 if (bracket) printf(")");
3090 ###### propagate binode cases
3096 /* both must be Tbool, result is Tbool */
3097 propagate_types(b->left, c, ok, Tbool, 0);
3098 propagate_types(b->right, c, ok, Tbool, 0);
3099 if (type && type != Tbool)
3100 type_err(c, "error: %1 operation found where %2 expected", prog,
3104 ###### interp binode cases
3106 rv = interp_exec(c, b->left, &rvtype);
3107 right = interp_exec(c, b->right, &rtype);
3108 rv.bool = rv.bool && right.bool;
3111 rv = interp_exec(c, b->left, &rvtype);
3113 rv = interp_exec(c, b->right, NULL);
3116 rv = interp_exec(c, b->left, &rvtype);
3117 right = interp_exec(c, b->right, &rtype);
3118 rv.bool = rv.bool || right.bool;
3121 rv = interp_exec(c, b->left, &rvtype);
3123 rv = interp_exec(c, b->right, NULL);
3126 rv = interp_exec(c, b->right, &rvtype);
3130 ### Expressions: Comparison
3132 Of slightly higher precedence that Boolean expressions are Comparisons.
3133 A comparison takes arguments of any comparable type, but the two types
3136 To simplify the parsing we introduce an `eop` which can record an
3137 expression operator, and the `CMPop` non-terminal will match one of them.
3144 ###### ast functions
3145 static void free_eop(struct eop *e)
3159 ###### expr precedence
3160 $LEFT < > <= >= == != CMPop
3162 ###### expression grammar
3163 | Expression CMPop Expression ${ {
3164 struct binode *b = new(binode);
3174 CMPop -> < ${ $0.op = Less; }$
3175 | > ${ $0.op = Gtr; }$
3176 | <= ${ $0.op = LessEq; }$
3177 | >= ${ $0.op = GtrEq; }$
3178 | == ${ $0.op = Eql; }$
3179 | != ${ $0.op = NEql; }$
3181 ###### print binode cases
3189 if (bracket) printf("(");
3190 print_exec(b->left, -1, bracket);
3192 case Less: printf(" < "); break;
3193 case LessEq: printf(" <= "); break;
3194 case Gtr: printf(" > "); break;
3195 case GtrEq: printf(" >= "); break;
3196 case Eql: printf(" == "); break;
3197 case NEql: printf(" != "); break;
3198 default: abort(); // NOTEST
3200 print_exec(b->right, -1, bracket);
3201 if (bracket) printf(")");
3204 ###### propagate binode cases
3211 /* Both must match but not be labels, result is Tbool */
3212 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
3214 propagate_types(b->right, c, ok, t, 0);
3216 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
3218 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
3220 if (!type_compat(type, Tbool, 0))
3221 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3222 Tbool, rules, type);
3225 ###### interp binode cases
3234 left = interp_exec(c, b->left, <ype);
3235 right = interp_exec(c, b->right, &rtype);
3236 cmp = value_cmp(ltype, rtype, &left, &right);
3239 case Less: rv.bool = cmp < 0; break;
3240 case LessEq: rv.bool = cmp <= 0; break;
3241 case Gtr: rv.bool = cmp > 0; break;
3242 case GtrEq: rv.bool = cmp >= 0; break;
3243 case Eql: rv.bool = cmp == 0; break;
3244 case NEql: rv.bool = cmp != 0; break;
3245 default: rv.bool = 0; break; // NOTEST
3250 ### Expressions: Arithmetic etc.
3252 The remaining expressions with the highest precedence are arithmetic,
3253 string concatenation, and string conversion. String concatenation
3254 (`++`) has the same precedence as multiplication and division, but lower
3257 String conversion is a temporary feature until I get a better type
3258 system. `$` is a prefix operator which expects a string and returns
3261 `+` and `-` are both infix and prefix operations (where they are
3262 absolute value and negation). These have different operator names.
3264 We also have a 'Bracket' operator which records where parentheses were
3265 found. This makes it easy to reproduce these when printing. Possibly I
3266 should only insert brackets were needed for precedence.
3276 ###### expr precedence
3282 ###### expression grammar
3283 | Expression Eop Expression ${ {
3284 struct binode *b = new(binode);
3291 | Expression Top Expression ${ {
3292 struct binode *b = new(binode);
3299 | ( Expression ) ${ {
3300 struct binode *b = new_pos(binode, $1);
3305 | Uop Expression ${ {
3306 struct binode *b = new(binode);
3311 | Value ${ $0 = $<1; }$
3312 | Variable ${ $0 = $<1; }$
3317 Eop -> + ${ $0.op = Plus; }$
3318 | - ${ $0.op = Minus; }$
3320 Uop -> + ${ $0.op = Absolute; }$
3321 | - ${ $0.op = Negate; }$
3322 | $ ${ $0.op = StringConv; }$
3324 Top -> * ${ $0.op = Times; }$
3325 | / ${ $0.op = Divide; }$
3326 | % ${ $0.op = Rem; }$
3327 | ++ ${ $0.op = Concat; }$
3329 ###### print binode cases
3336 if (bracket) printf("(");
3337 print_exec(b->left, indent, bracket);
3339 case Plus: fputs(" + ", stdout); break;
3340 case Minus: fputs(" - ", stdout); break;
3341 case Times: fputs(" * ", stdout); break;
3342 case Divide: fputs(" / ", stdout); break;
3343 case Rem: fputs(" % ", stdout); break;
3344 case Concat: fputs(" ++ ", stdout); break;
3345 default: abort(); // NOTEST
3347 print_exec(b->right, indent, bracket);
3348 if (bracket) printf(")");
3353 if (bracket) printf("(");
3355 case Absolute: fputs("+", stdout); break;
3356 case Negate: fputs("-", stdout); break;
3357 case StringConv: fputs("$", stdout); break;
3358 default: abort(); // NOTEST
3360 print_exec(b->right, indent, bracket);
3361 if (bracket) printf(")");
3365 print_exec(b->right, indent, bracket);
3369 ###### propagate binode cases
3375 /* both must be numbers, result is Tnum */
3378 /* as propagate_types ignores a NULL,
3379 * unary ops fit here too */
3380 propagate_types(b->left, c, ok, Tnum, 0);
3381 propagate_types(b->right, c, ok, Tnum, 0);
3382 if (!type_compat(type, Tnum, 0))
3383 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3388 /* both must be Tstr, result is Tstr */
3389 propagate_types(b->left, c, ok, Tstr, 0);
3390 propagate_types(b->right, c, ok, Tstr, 0);
3391 if (!type_compat(type, Tstr, 0))
3392 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3397 /* op must be string, result is number */
3398 propagate_types(b->left, c, ok, Tstr, 0);
3399 if (!type_compat(type, Tnum, 0))
3400 type_err(c, // UNTESTED
3401 "error: Can only convert string to number, not %1",
3402 prog, type, 0, NULL);
3406 return propagate_types(b->right, c, ok, type, 0);
3408 ###### interp binode cases
3411 rv = interp_exec(c, b->left, &rvtype);
3412 right = interp_exec(c, b->right, &rtype);
3413 mpq_add(rv.num, rv.num, right.num);
3416 rv = interp_exec(c, b->left, &rvtype);
3417 right = interp_exec(c, b->right, &rtype);
3418 mpq_sub(rv.num, rv.num, right.num);
3421 rv = interp_exec(c, b->left, &rvtype);
3422 right = interp_exec(c, b->right, &rtype);
3423 mpq_mul(rv.num, rv.num, right.num);
3426 rv = interp_exec(c, b->left, &rvtype);
3427 right = interp_exec(c, b->right, &rtype);
3428 mpq_div(rv.num, rv.num, right.num);
3433 left = interp_exec(c, b->left, <ype);
3434 right = interp_exec(c, b->right, &rtype);
3435 mpz_init(l); mpz_init(r); mpz_init(rem);
3436 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3437 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3438 mpz_tdiv_r(rem, l, r);
3439 val_init(Tnum, &rv);
3440 mpq_set_z(rv.num, rem);
3441 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3446 rv = interp_exec(c, b->right, &rvtype);
3447 mpq_neg(rv.num, rv.num);
3450 rv = interp_exec(c, b->right, &rvtype);
3451 mpq_abs(rv.num, rv.num);
3454 rv = interp_exec(c, b->right, &rvtype);
3457 left = interp_exec(c, b->left, <ype);
3458 right = interp_exec(c, b->right, &rtype);
3460 rv.str = text_join(left.str, right.str);
3463 right = interp_exec(c, b->right, &rvtype);
3467 struct text tx = right.str;
3470 if (tx.txt[0] == '-') {
3471 neg = 1; // UNTESTED
3472 tx.txt++; // UNTESTED
3473 tx.len--; // UNTESTED
3475 if (number_parse(rv.num, tail, tx) == 0)
3476 mpq_init(rv.num); // UNTESTED
3478 mpq_neg(rv.num, rv.num); // UNTESTED
3480 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3484 ###### value functions
3486 static struct text text_join(struct text a, struct text b)
3489 rv.len = a.len + b.len;
3490 rv.txt = malloc(rv.len);
3491 memcpy(rv.txt, a.txt, a.len);
3492 memcpy(rv.txt+a.len, b.txt, b.len);
3498 A function call can appear either as an expression or as a statement.
3499 As functions cannot yet return values, only the statement version will work.
3500 We use a new 'Funcall' binode type to link the function with a list of
3501 arguments, form with the 'List' nodes.
3506 ###### expression grammar
3507 | Variable ( ExpressionList ) ${ {
3508 struct binode *b = new(binode);
3511 b->right = reorder_bilist($<EL);
3515 struct binode *b = new(binode);
3522 ###### SimpleStatement Grammar
3524 | Variable ( ExpressionList ) ${ {
3525 struct binode *b = new(binode);
3528 b->right = reorder_bilist($<EL);
3532 ###### print binode cases
3535 do_indent(indent, "");
3536 print_exec(b->left, -1, bracket);
3538 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3541 print_exec(b->left, -1, bracket);
3551 ###### propagate binode cases
3554 /* Every arg must match formal parameter, and result
3555 * is return type of function
3557 struct binode *args = cast(binode, b->right);
3558 struct var *v = cast(var, b->left);
3560 if (!v->var->type || v->var->type->check_args == NULL) {
3561 type_err(c, "error: attempt to call a non-function.",
3562 prog, NULL, 0, NULL);
3565 v->var->type->check_args(c, ok, v->var->type, args);
3566 return v->var->type->function.return_type;
3569 ###### interp binode cases
3572 struct var *v = cast(var, b->left);
3573 struct type *t = v->var->type;
3574 void *oldlocal = c->local;
3575 int old_size = c->local_size;
3576 void *local = calloc(1, t->function.local_size);
3577 struct value *fbody = var_value(c, v->var);
3578 struct binode *arg = cast(binode, b->right);
3579 struct binode *param = t->function.params;
3582 struct var *pv = cast(var, param->left);
3583 struct type *vtype = NULL;
3584 struct value val = interp_exec(c, arg->left, &vtype);
3586 c->local = local; c->local_size = t->function.local_size;
3587 lval = var_value(c, pv->var);
3588 c->local = oldlocal; c->local_size = old_size;
3589 memcpy(lval, &val, vtype->size);
3590 param = cast(binode, param->right);
3591 arg = cast(binode, arg->right);
3593 c->local = local; c->local_size = t->function.local_size;
3594 rv = interp_exec(c, fbody->function, &rvtype);
3595 c->local = oldlocal; c->local_size = old_size;
3600 ### Blocks, Statements, and Statement lists.
3602 Now that we have expressions out of the way we need to turn to
3603 statements. There are simple statements and more complex statements.
3604 Simple statements do not contain (syntactic) newlines, complex statements do.
3606 Statements often come in sequences and we have corresponding simple
3607 statement lists and complex statement lists.
3608 The former comprise only simple statements separated by semicolons.
3609 The later comprise complex statements and simple statement lists. They are
3610 separated by newlines. Thus the semicolon is only used to separate
3611 simple statements on the one line. This may be overly restrictive,
3612 but I'm not sure I ever want a complex statement to share a line with
3615 Note that a simple statement list can still use multiple lines if
3616 subsequent lines are indented, so
3618 ###### Example: wrapped simple statement list
3623 is a single simple statement list. This might allow room for
3624 confusion, so I'm not set on it yet.
3626 A simple statement list needs no extra syntax. A complex statement
3627 list has two syntactic forms. It can be enclosed in braces (much like
3628 C blocks), or it can be introduced by an indent and continue until an
3629 unindented newline (much like Python blocks). With this extra syntax
3630 it is referred to as a block.
3632 Note that a block does not have to include any newlines if it only
3633 contains simple statements. So both of:
3635 if condition: a=b; d=f
3637 if condition { a=b; print f }
3641 In either case the list is constructed from a `binode` list with
3642 `Block` as the operator. When parsing the list it is most convenient
3643 to append to the end, so a list is a list and a statement. When using
3644 the list it is more convenient to consider a list to be a statement
3645 and a list. So we need a function to re-order a list.
3646 `reorder_bilist` serves this purpose.
3648 The only stand-alone statement we introduce at this stage is `pass`
3649 which does nothing and is represented as a `NULL` pointer in a `Block`
3650 list. Other stand-alone statements will follow once the infrastructure
3661 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3662 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3663 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3664 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3665 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3667 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3668 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3669 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3670 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3671 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3673 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3674 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3675 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3677 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3678 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3679 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3680 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3681 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3683 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3685 ComplexStatements -> ComplexStatements ComplexStatement ${
3695 | ComplexStatement ${
3707 ComplexStatement -> SimpleStatements Newlines ${
3708 $0 = reorder_bilist($<SS);
3710 | SimpleStatements ; Newlines ${
3711 $0 = reorder_bilist($<SS);
3713 ## ComplexStatement Grammar
3716 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3722 | SimpleStatement ${
3730 SimpleStatement -> pass ${ $0 = NULL; }$
3731 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3732 ## SimpleStatement Grammar
3734 ###### print binode cases
3738 if (b->left == NULL) // UNTESTED
3739 printf("pass"); // UNTESTED
3741 print_exec(b->left, indent, bracket); // UNTESTED
3742 if (b->right) { // UNTESTED
3743 printf("; "); // UNTESTED
3744 print_exec(b->right, indent, bracket); // UNTESTED
3747 // block, one per line
3748 if (b->left == NULL)
3749 do_indent(indent, "pass\n");
3751 print_exec(b->left, indent, bracket);
3753 print_exec(b->right, indent, bracket);
3757 ###### propagate binode cases
3760 /* If any statement returns something other than Tnone
3761 * or Tbool then all such must return same type.
3762 * As each statement may be Tnone or something else,
3763 * we must always pass NULL (unknown) down, otherwise an incorrect
3764 * error might occur. We never return Tnone unless it is
3769 for (e = b; e; e = cast(binode, e->right)) {
3770 t = propagate_types(e->left, c, ok, NULL, rules);
3771 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
3773 if (t == Tnone && e->right)
3774 /* Only the final statement *must* return a value
3782 type_err(c, "error: expected %1%r, found %2",
3783 e->left, type, rules, t);
3789 ###### interp binode cases
3791 while (rvtype == Tnone &&
3794 rv = interp_exec(c, b->left, &rvtype);
3795 b = cast(binode, b->right);
3799 ### The Print statement
3801 `print` is a simple statement that takes a comma-separated list of
3802 expressions and prints the values separated by spaces and terminated
3803 by a newline. No control of formatting is possible.
3805 `print` uses `ExpressionList` to collect the expressions and stores them
3806 on the left side of a `Print` binode unlessthere is a trailing comma
3807 when the list is stored on the `right` side and no trailing newline is
3813 ##### expr precedence
3816 ###### SimpleStatement Grammar
3818 | print ExpressionList ${
3822 $0->left = reorder_bilist($<EL);
3824 | print ExpressionList , ${ {
3827 $0->right = reorder_bilist($<EL);
3837 ###### print binode cases
3840 do_indent(indent, "print");
3842 print_exec(b->right, -1, bracket);
3845 print_exec(b->left, -1, bracket);
3850 ###### propagate binode cases
3853 /* don't care but all must be consistent */
3855 b = cast(binode, b->left);
3857 b = cast(binode, b->right);
3859 propagate_types(b->left, c, ok, NULL, Rnolabel);
3860 b = cast(binode, b->right);
3864 ###### interp binode cases
3868 struct binode *b2 = cast(binode, b->left);
3870 b2 = cast(binode, b->right);
3871 for (; b2; b2 = cast(binode, b2->right)) {
3872 left = interp_exec(c, b2->left, <ype);
3873 print_value(ltype, &left);
3874 free_value(ltype, &left);
3878 if (b->right == NULL)
3884 ###### Assignment statement
3886 An assignment will assign a value to a variable, providing it hasn't
3887 been declared as a constant. The analysis phase ensures that the type
3888 will be correct so the interpreter just needs to perform the
3889 calculation. There is a form of assignment which declares a new
3890 variable as well as assigning a value. If a name is assigned before
3891 it is declared, and error will be raised as the name is created as
3892 `Tlabel` and it is illegal to assign to such names.
3898 ###### declare terminals
3901 ###### SimpleStatement Grammar
3902 | Variable = Expression ${
3908 | VariableDecl = Expression ${
3916 if ($1->var->where_set == NULL) {
3918 "Variable declared with no type or value: %v",
3929 ###### print binode cases
3932 do_indent(indent, "");
3933 print_exec(b->left, indent, bracket);
3935 print_exec(b->right, indent, bracket);
3942 struct variable *v = cast(var, b->left)->var;
3943 do_indent(indent, "");
3944 print_exec(b->left, indent, bracket);
3945 if (cast(var, b->left)->var->constant) {
3947 if (v->where_decl == v->where_set) {
3948 type_print(v->type, stdout);
3953 if (v->where_decl == v->where_set) {
3954 type_print(v->type, stdout);
3960 print_exec(b->right, indent, bracket);
3967 ###### propagate binode cases
3971 /* Both must match and not be labels,
3972 * Type must support 'dup',
3973 * For Assign, left must not be constant.
3976 t = propagate_types(b->left, c, ok, NULL,
3977 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
3982 if (propagate_types(b->right, c, ok, t, 0) != t)
3983 if (b->left->type == Xvar)
3984 type_err(c, "info: variable '%v' was set as %1 here.",
3985 cast(var, b->left)->var->where_set, t, rules, NULL);
3987 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
3989 propagate_types(b->left, c, ok, t,
3990 (b->op == Assign ? Rnoconstant : 0));
3992 if (t && t->dup == NULL)
3993 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
3998 ###### interp binode cases
4001 lleft = linterp_exec(c, b->left, <ype);
4002 right = interp_exec(c, b->right, &rtype);
4004 free_value(ltype, lleft);
4005 dup_value(ltype, &right, lleft);
4012 struct variable *v = cast(var, b->left)->var;
4015 val = var_value(c, v);
4016 if (v->type->prepare_type)
4017 v->type->prepare_type(c, v->type, 0);
4019 right = interp_exec(c, b->right, &rtype);
4020 memcpy(val, &right, rtype->size);
4023 val_init(v->type, val);
4028 ### The `use` statement
4030 The `use` statement is the last "simple" statement. It is needed when a
4031 statement block can return a value. This includes the body of a
4032 function which has a return type, and the "condition" code blocks in
4033 `if`, `while`, and `switch` statements.
4038 ###### expr precedence
4041 ###### SimpleStatement Grammar
4043 $0 = new_pos(binode, $1);
4046 if ($0->right->type == Xvar) {
4047 struct var *v = cast(var, $0->right);
4048 if (v->var->type == Tnone) {
4049 /* Convert this to a label */
4052 v->var->type = Tlabel;
4053 val = global_alloc(c, Tlabel, v->var, NULL);
4059 ###### print binode cases
4062 do_indent(indent, "use ");
4063 print_exec(b->right, -1, bracket);
4068 ###### propagate binode cases
4071 /* result matches value */
4072 return propagate_types(b->right, c, ok, type, 0);
4074 ###### interp binode cases
4077 rv = interp_exec(c, b->right, &rvtype);
4080 ### The Conditional Statement
4082 This is the biggy and currently the only complex statement. This
4083 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4084 It is comprised of a number of parts, all of which are optional though
4085 set combinations apply. Each part is (usually) a key word (`then` is
4086 sometimes optional) followed by either an expression or a code block,
4087 except the `casepart` which is a "key word and an expression" followed
4088 by a code block. The code-block option is valid for all parts and,
4089 where an expression is also allowed, the code block can use the `use`
4090 statement to report a value. If the code block does not report a value
4091 the effect is similar to reporting `True`.
4093 The `else` and `case` parts, as well as `then` when combined with
4094 `if`, can contain a `use` statement which will apply to some
4095 containing conditional statement. `for` parts, `do` parts and `then`
4096 parts used with `for` can never contain a `use`, except in some
4097 subordinate conditional statement.
4099 If there is a `forpart`, it is executed first, only once.
4100 If there is a `dopart`, then it is executed repeatedly providing
4101 always that the `condpart` or `cond`, if present, does not return a non-True
4102 value. `condpart` can fail to return any value if it simply executes
4103 to completion. This is treated the same as returning `True`.
4105 If there is a `thenpart` it will be executed whenever the `condpart`
4106 or `cond` returns True (or does not return any value), but this will happen
4107 *after* `dopart` (when present).
4109 If `elsepart` is present it will be executed at most once when the
4110 condition returns `False` or some value that isn't `True` and isn't
4111 matched by any `casepart`. If there are any `casepart`s, they will be
4112 executed when the condition returns a matching value.
4114 The particular sorts of values allowed in case parts has not yet been
4115 determined in the language design, so nothing is prohibited.
4117 The various blocks in this complex statement potentially provide scope
4118 for variables as described earlier. Each such block must include the
4119 "OpenScope" nonterminal before parsing the block, and must call
4120 `var_block_close()` when closing the block.
4122 The code following "`if`", "`switch`" and "`for`" does not get its own
4123 scope, but is in a scope covering the whole statement, so names
4124 declared there cannot be redeclared elsewhere. Similarly the
4125 condition following "`while`" is in a scope the covers the body
4126 ("`do`" part) of the loop, and which does not allow conditional scope
4127 extension. Code following "`then`" (both looping and non-looping),
4128 "`else`" and "`case`" each get their own local scope.
4130 The type requirements on the code block in a `whilepart` are quite
4131 unusal. It is allowed to return a value of some identifiable type, in
4132 which case the loop aborts and an appropriate `casepart` is run, or it
4133 can return a Boolean, in which case the loop either continues to the
4134 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4135 This is different both from the `ifpart` code block which is expected to
4136 return a Boolean, or the `switchpart` code block which is expected to
4137 return the same type as the casepart values. The correct analysis of
4138 the type of the `whilepart` code block is the reason for the
4139 `Rboolok` flag which is passed to `propagate_types()`.
4141 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4142 defined. As there are two scopes which cover multiple parts - one for
4143 the whole statement and one for "while" and "do" - and as we will use
4144 the 'struct exec' to track scopes, we actually need two new types of
4145 exec. One is a `binode` for the looping part, the rest is the
4146 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4147 casepart` to track a list of case parts.
4158 struct exec *action;
4159 struct casepart *next;
4161 struct cond_statement {
4163 struct exec *forpart, *condpart, *thenpart, *elsepart;
4164 struct binode *looppart;
4165 struct casepart *casepart;
4168 ###### ast functions
4170 static void free_casepart(struct casepart *cp)
4174 free_exec(cp->value);
4175 free_exec(cp->action);
4182 static void free_cond_statement(struct cond_statement *s)
4186 free_exec(s->forpart);
4187 free_exec(s->condpart);
4188 free_exec(s->looppart);
4189 free_exec(s->thenpart);
4190 free_exec(s->elsepart);
4191 free_casepart(s->casepart);
4195 ###### free exec cases
4196 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4198 ###### ComplexStatement Grammar
4199 | CondStatement ${ $0 = $<1; }$
4201 ###### expr precedence
4202 $TERM for then while do
4209 // A CondStatement must end with EOL, as does CondSuffix and
4211 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4212 // may or may not end with EOL
4213 // WhilePart and IfPart include an appropriate Suffix
4215 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4216 // them. WhilePart opens and closes its own scope.
4217 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4220 $0->thenpart = $<TP;
4221 $0->looppart = $<WP;
4222 var_block_close(c, CloseSequential, $0);
4224 | ForPart OptNL WhilePart CondSuffix ${
4227 $0->looppart = $<WP;
4228 var_block_close(c, CloseSequential, $0);
4230 | WhilePart CondSuffix ${
4232 $0->looppart = $<WP;
4234 | SwitchPart OptNL CasePart CondSuffix ${
4236 $0->condpart = $<SP;
4237 $CP->next = $0->casepart;
4238 $0->casepart = $<CP;
4239 var_block_close(c, CloseSequential, $0);
4241 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4243 $0->condpart = $<SP;
4244 $CP->next = $0->casepart;
4245 $0->casepart = $<CP;
4246 var_block_close(c, CloseSequential, $0);
4248 | IfPart IfSuffix ${
4250 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4251 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4252 // This is where we close an "if" statement
4253 var_block_close(c, CloseSequential, $0);
4256 CondSuffix -> IfSuffix ${
4259 | Newlines CasePart CondSuffix ${
4261 $CP->next = $0->casepart;
4262 $0->casepart = $<CP;
4264 | CasePart CondSuffix ${
4266 $CP->next = $0->casepart;
4267 $0->casepart = $<CP;
4270 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4271 | Newlines ElsePart ${ $0 = $<EP; }$
4272 | ElsePart ${$0 = $<EP; }$
4274 ElsePart -> else OpenBlock Newlines ${
4275 $0 = new(cond_statement);
4276 $0->elsepart = $<OB;
4277 var_block_close(c, CloseElse, $0->elsepart);
4279 | else OpenScope CondStatement ${
4280 $0 = new(cond_statement);
4281 $0->elsepart = $<CS;
4282 var_block_close(c, CloseElse, $0->elsepart);
4286 CasePart -> case Expression OpenScope ColonBlock ${
4287 $0 = calloc(1,sizeof(struct casepart));
4290 var_block_close(c, CloseParallel, $0->action);
4294 // These scopes are closed in CondStatement
4295 ForPart -> for OpenBlock ${
4299 ThenPart -> then OpenBlock ${
4301 var_block_close(c, CloseSequential, $0);
4305 // This scope is closed in CondStatement
4306 WhilePart -> while UseBlock OptNL do OpenBlock ${
4311 var_block_close(c, CloseSequential, $0->right);
4312 var_block_close(c, CloseSequential, $0);
4314 | while OpenScope Expression OpenScope ColonBlock ${
4319 var_block_close(c, CloseSequential, $0->right);
4320 var_block_close(c, CloseSequential, $0);
4324 IfPart -> if UseBlock OptNL then OpenBlock ${
4327 var_block_close(c, CloseParallel, $0.thenpart);
4329 | if OpenScope Expression OpenScope ColonBlock ${
4332 var_block_close(c, CloseParallel, $0.thenpart);
4334 | if OpenScope Expression OpenScope OptNL then Block ${
4337 var_block_close(c, CloseParallel, $0.thenpart);
4341 // This scope is closed in CondStatement
4342 SwitchPart -> switch OpenScope Expression ${
4345 | switch UseBlock ${
4349 ###### print binode cases
4351 if (b->left && b->left->type == Xbinode &&
4352 cast(binode, b->left)->op == Block) {
4354 do_indent(indent, "while {\n");
4356 do_indent(indent, "while\n");
4357 print_exec(b->left, indent+1, bracket);
4359 do_indent(indent, "} do {\n");
4361 do_indent(indent, "do\n");
4362 print_exec(b->right, indent+1, bracket);
4364 do_indent(indent, "}\n");
4366 do_indent(indent, "while ");
4367 print_exec(b->left, 0, bracket);
4372 print_exec(b->right, indent+1, bracket);
4374 do_indent(indent, "}\n");
4378 ###### print exec cases
4380 case Xcond_statement:
4382 struct cond_statement *cs = cast(cond_statement, e);
4383 struct casepart *cp;
4385 do_indent(indent, "for");
4386 if (bracket) printf(" {\n"); else printf("\n");
4387 print_exec(cs->forpart, indent+1, bracket);
4390 do_indent(indent, "} then {\n");
4392 do_indent(indent, "then\n");
4393 print_exec(cs->thenpart, indent+1, bracket);
4395 if (bracket) do_indent(indent, "}\n");
4398 print_exec(cs->looppart, indent, bracket);
4402 do_indent(indent, "switch");
4404 do_indent(indent, "if");
4405 if (cs->condpart && cs->condpart->type == Xbinode &&
4406 cast(binode, cs->condpart)->op == Block) {
4411 print_exec(cs->condpart, indent+1, bracket);
4413 do_indent(indent, "}\n");
4415 do_indent(indent, "then\n");
4416 print_exec(cs->thenpart, indent+1, bracket);
4420 print_exec(cs->condpart, 0, bracket);
4426 print_exec(cs->thenpart, indent+1, bracket);
4428 do_indent(indent, "}\n");
4433 for (cp = cs->casepart; cp; cp = cp->next) {
4434 do_indent(indent, "case ");
4435 print_exec(cp->value, -1, 0);
4440 print_exec(cp->action, indent+1, bracket);
4442 do_indent(indent, "}\n");
4445 do_indent(indent, "else");
4450 print_exec(cs->elsepart, indent+1, bracket);
4452 do_indent(indent, "}\n");
4457 ###### propagate binode cases
4459 t = propagate_types(b->right, c, ok, Tnone, 0);
4460 if (!type_compat(Tnone, t, 0))
4461 *ok = 0; // UNTESTED
4462 return propagate_types(b->left, c, ok, type, rules);
4464 ###### propagate exec cases
4465 case Xcond_statement:
4467 // forpart and looppart->right must return Tnone
4468 // thenpart must return Tnone if there is a loopart,
4469 // otherwise it is like elsepart.
4471 // be bool if there is no casepart
4472 // match casepart->values if there is a switchpart
4473 // either be bool or match casepart->value if there
4475 // elsepart and casepart->action must match the return type
4476 // expected of this statement.
4477 struct cond_statement *cs = cast(cond_statement, prog);
4478 struct casepart *cp;
4480 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4481 if (!type_compat(Tnone, t, 0))
4482 *ok = 0; // UNTESTED
4485 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4486 if (!type_compat(Tnone, t, 0))
4487 *ok = 0; // UNTESTED
4489 if (cs->casepart == NULL) {
4490 propagate_types(cs->condpart, c, ok, Tbool, 0);
4491 propagate_types(cs->looppart, c, ok, Tbool, 0);
4493 /* Condpart must match case values, with bool permitted */
4495 for (cp = cs->casepart;
4496 cp && !t; cp = cp->next)
4497 t = propagate_types(cp->value, c, ok, NULL, 0);
4498 if (!t && cs->condpart)
4499 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4500 if (!t && cs->looppart)
4501 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4502 // Now we have a type (I hope) push it down
4504 for (cp = cs->casepart; cp; cp = cp->next)
4505 propagate_types(cp->value, c, ok, t, 0);
4506 propagate_types(cs->condpart, c, ok, t, Rboolok);
4507 propagate_types(cs->looppart, c, ok, t, Rboolok);
4510 // (if)then, else, and case parts must return expected type.
4511 if (!cs->looppart && !type)
4512 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4514 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4515 for (cp = cs->casepart;
4517 cp = cp->next) // UNTESTED
4518 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4521 propagate_types(cs->thenpart, c, ok, type, rules);
4522 propagate_types(cs->elsepart, c, ok, type, rules);
4523 for (cp = cs->casepart; cp ; cp = cp->next)
4524 propagate_types(cp->action, c, ok, type, rules);
4530 ###### interp binode cases
4532 // This just performs one iterration of the loop
4533 rv = interp_exec(c, b->left, &rvtype);
4534 if (rvtype == Tnone ||
4535 (rvtype == Tbool && rv.bool != 0))
4536 // cnd is Tnone or Tbool, doesn't need to be freed
4537 interp_exec(c, b->right, NULL);
4540 ###### interp exec cases
4541 case Xcond_statement:
4543 struct value v, cnd;
4544 struct type *vtype, *cndtype;
4545 struct casepart *cp;
4546 struct cond_statement *cs = cast(cond_statement, e);
4549 interp_exec(c, cs->forpart, NULL);
4551 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4552 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4553 interp_exec(c, cs->thenpart, NULL);
4555 cnd = interp_exec(c, cs->condpart, &cndtype);
4556 if ((cndtype == Tnone ||
4557 (cndtype == Tbool && cnd.bool != 0))) {
4558 // cnd is Tnone or Tbool, doesn't need to be freed
4559 rv = interp_exec(c, cs->thenpart, &rvtype);
4560 // skip else (and cases)
4564 for (cp = cs->casepart; cp; cp = cp->next) {
4565 v = interp_exec(c, cp->value, &vtype);
4566 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4567 free_value(vtype, &v);
4568 free_value(cndtype, &cnd);
4569 rv = interp_exec(c, cp->action, &rvtype);
4572 free_value(vtype, &v);
4574 free_value(cndtype, &cnd);
4576 rv = interp_exec(c, cs->elsepart, &rvtype);
4583 ### Top level structure
4585 All the language elements so far can be used in various places. Now
4586 it is time to clarify what those places are.
4588 At the top level of a file there will be a number of declarations.
4589 Many of the things that can be declared haven't been described yet,
4590 such as functions, procedures, imports, and probably more.
4591 For now there are two sorts of things that can appear at the top
4592 level. They are predefined constants, `struct` types, and the `main`
4593 function. While the syntax will allow the `main` function to appear
4594 multiple times, that will trigger an error if it is actually attempted.
4596 The various declarations do not return anything. They store the
4597 various declarations in the parse context.
4599 ###### Parser: grammar
4602 Ocean -> OptNL DeclarationList
4604 ## declare terminals
4611 DeclarationList -> Declaration
4612 | DeclarationList Declaration
4614 Declaration -> ERROR Newlines ${
4615 tok_err(c, // UNTESTED
4616 "error: unhandled parse error", &$1);
4622 ## top level grammar
4626 ### The `const` section
4628 As well as being defined in with the code that uses them, constants
4629 can be declared at the top level. These have full-file scope, so they
4630 are always `InScope`. The value of a top level constant can be given
4631 as an expression, and this is evaluated immediately rather than in the
4632 later interpretation stage. Once we add functions to the language, we
4633 will need rules concern which, if any, can be used to define a top
4636 Constants are defined in a section that starts with the reserved word
4637 `const` and then has a block with a list of assignment statements.
4638 For syntactic consistency, these must use the double-colon syntax to
4639 make it clear that they are constants. Type can also be given: if
4640 not, the type will be determined during analysis, as with other
4643 As the types constants are inserted at the head of a list, printing
4644 them in the same order that they were read is not straight forward.
4645 We take a quadratic approach here and count the number of constants
4646 (variables of depth 0), then count down from there, each time
4647 searching through for the Nth constant for decreasing N.
4649 ###### top level grammar
4653 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4654 | const { SimpleConstList } Newlines
4655 | const IN OptNL ConstList OUT Newlines
4656 | const SimpleConstList Newlines
4658 ConstList -> ConstList SimpleConstLine
4660 SimpleConstList -> SimpleConstList ; Const
4663 SimpleConstLine -> SimpleConstList Newlines
4664 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4667 CType -> Type ${ $0 = $<1; }$
4670 Const -> IDENTIFIER :: CType = Expression ${ {
4674 v = var_decl(c, $1.txt);
4676 struct var *var = new_pos(var, $1);
4677 v->where_decl = var;
4683 struct variable *vorig = var_ref(c, $1.txt);
4684 tok_err(c, "error: name already declared", &$1);
4685 type_err(c, "info: this is where '%v' was first declared",
4686 vorig->where_decl, NULL, 0, NULL);
4690 propagate_types($5, c, &ok, $3, 0);
4695 struct value res = interp_exec(c, $5, &v->type);
4696 global_alloc(c, v->type, v, &res);
4700 ###### print const decls
4705 while (target != 0) {
4707 for (v = context.in_scope; v; v=v->in_scope)
4708 if (v->depth == 0 && v->constant) {
4719 struct value *val = var_value(&context, v);
4720 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4721 type_print(v->type, stdout);
4723 if (v->type == Tstr)
4725 print_value(v->type, val);
4726 if (v->type == Tstr)
4734 ### Function declarations
4736 The code in an Ocean program is all stored in function declarations.
4737 One of the functions must be named `main` and it must accept an array of
4738 strings as a parameter - the command line arguments.
4740 As this is the top level, several things are handled a bit differently.
4741 The function is not interpreted by `interp_exec` as that isn't passed
4742 the argument list which the program requires. Similarly type analysis
4743 is a bit more interesting at this level.
4745 ###### ast functions
4747 static struct variable *declare_function(struct parse_context *c,
4748 struct variable *name,
4749 struct binode *args,
4753 struct text funcname = {" func", 5};
4755 struct value fn = {.function = code};
4756 name->type = add_type(c, funcname, &function_prototype);
4757 name->type->function.params = reorder_bilist(args);
4758 name->type->function.return_type = ret;
4759 global_alloc(c, name->type, name, &fn);
4760 var_block_close(c, CloseSequential, code);
4765 var_block_close(c, CloseSequential, NULL);
4770 ###### declare terminals
4773 ###### top level grammar
4776 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
4777 $0 = declare_function(c, $<FN, $<Ar, Tnone, $<Bl);
4779 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
4780 $0 = declare_function(c, $<FN, $<Ar, Tnone, $<Bl);
4782 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
4783 $0 = declare_function(c, $<FN, NULL, Tnone, $<Bl);
4785 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
4786 $0 = declare_function(c, $<FN, $<Ar, $<Ty, $<Bl);
4788 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
4789 $0 = declare_function(c, $<FN, $<Ar, $<Ty, $<Bl);
4791 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
4792 $0 = declare_function(c, $<FN, NULL, $<Ty, $<Bl);
4795 ###### print func decls
4800 while (target != 0) {
4802 for (v = context.in_scope; v; v=v->in_scope)
4803 if (v->depth == 0 && v->type && v->type->check_args) {
4812 struct value *val = var_value(&context, v);
4813 printf("func %.*s", v->name->name.len, v->name->name.txt);
4814 v->type->print_type_decl(v->type, stdout);
4816 print_exec(val->function, 0, brackets);
4818 print_value(v->type, val);
4819 printf("/* frame size %d */\n", v->type->function.local_size);
4825 ###### core functions
4827 static int analyse_funcs(struct parse_context *c)
4831 for (v = c->in_scope; v; v = v->in_scope) {
4834 if (v->depth != 0 || !v->type || !v->type->check_args)
4836 val = var_value(c, v);
4839 propagate_types(val->function, c, &ok,
4840 v->type->function.return_type, 0);
4843 /* Make sure everything is still consistent */
4844 propagate_types(val->function, c, &ok,
4845 v->type->function.return_type, 0);
4848 if (!v->type->function.return_type->dup) {
4849 type_err(c, "error: function cannot return value of type %1",
4850 v->where_decl, v->type->function.return_type, 0, NULL);
4853 v->type->function.local_size = scope_finalize(c);
4858 static int analyse_main(struct type *type, struct parse_context *c)
4860 struct binode *bp = type->function.params;
4864 struct type *argv_type;
4865 struct text argv_type_name = { " argv", 5 };
4867 argv_type = add_type(c, argv_type_name, &array_prototype);
4868 argv_type->array.member = Tstr;
4869 argv_type->array.unspec = 1;
4871 for (b = bp; b; b = cast(binode, b->right)) {
4875 propagate_types(b->left, c, &ok, argv_type, 0);
4877 default: /* invalid */ // NOTEST
4878 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4884 return !c->parse_error;
4887 static void interp_main(struct parse_context *c, int argc, char **argv)
4889 struct value *progp = NULL;
4890 struct text main_name = { "main", 4 };
4891 struct variable *mainv;
4897 mainv = var_ref(c, main_name);
4899 progp = var_value(c, mainv);
4900 if (!progp || !progp->function) {
4901 fprintf(stderr, "oceani: no main function found.\n");
4905 if (!analyse_main(mainv->type, c)) {
4906 fprintf(stderr, "oceani: main has wrong type.\n");
4910 al = mainv->type->function.params;
4912 c->local_size = mainv->type->function.local_size;
4913 c->local = calloc(1, c->local_size);
4915 struct var *v = cast(var, al->left);
4916 struct value *vl = var_value(c, v->var);
4926 mpq_set_ui(argcq, argc, 1);
4927 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
4928 t->prepare_type(c, t, 0);
4929 array_init(v->var->type, vl);
4930 for (i = 0; i < argc; i++) {
4931 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
4933 arg.str.txt = argv[i];
4934 arg.str.len = strlen(argv[i]);
4935 free_value(Tstr, vl2);
4936 dup_value(Tstr, &arg, vl2);
4940 al = cast(binode, al->right);
4942 v = interp_exec(c, progp->function, &vtype);
4943 free_value(vtype, &v);
4948 ###### ast functions
4949 void free_variable(struct variable *v)
4953 ## And now to test it out.
4955 Having a language requires having a "hello world" program. I'll
4956 provide a little more than that: a program that prints "Hello world"
4957 finds the GCD of two numbers, prints the first few elements of
4958 Fibonacci, performs a binary search for a number, and a few other
4959 things which will likely grow as the languages grows.
4961 ###### File: oceani.mk
4964 @echo "===== DEMO ====="
4965 ./oceani --section "demo: hello" oceani.mdc 55 33
4971 four ::= 2 + 2 ; five ::= 10/2
4972 const pie ::= "I like Pie";
4973 cake ::= "The cake is"
4981 func main(argv:[argc::]string)
4982 print "Hello World, what lovely oceans you have!"
4983 print "Are there", five, "?"
4984 print pi, pie, "but", cake
4986 A := $argv[1]; B := $argv[2]
4988 /* When a variable is defined in both branches of an 'if',
4989 * and used afterwards, the variables are merged.
4995 print "Is", A, "bigger than", B,"? ", bigger
4996 /* If a variable is not used after the 'if', no
4997 * merge happens, so types can be different
5000 double:string = "yes"
5001 print A, "is more than twice", B, "?", double
5004 print "double", B, "is", double
5009 if a > 0 and then b > 0:
5015 print "GCD of", A, "and", B,"is", a
5017 print a, "is not positive, cannot calculate GCD"
5019 print b, "is not positive, cannot calculate GCD"
5024 print "Fibonacci:", f1,f2,
5025 then togo = togo - 1
5033 /* Binary search... */
5038 mid := (lo + hi) / 2
5051 print "Yay, I found", target
5053 print "Closest I found was", lo
5058 // "middle square" PRNG. Not particularly good, but one my
5059 // Dad taught me - the first one I ever heard of.
5060 for i:=1; then i = i + 1; while i < size:
5061 n := list[i-1] * list[i-1]
5062 list[i] = (n / 100) % 10 000
5064 print "Before sort:",
5065 for i:=0; then i = i + 1; while i < size:
5069 for i := 1; then i=i+1; while i < size:
5070 for j:=i-1; then j=j-1; while j >= 0:
5071 if list[j] > list[j+1]:
5075 print " After sort:",
5076 for i:=0; then i = i + 1; while i < size:
5080 if 1 == 2 then print "yes"; else print "no"
5084 bob.alive = (bob.name == "Hello")
5085 print "bob", "is" if bob.alive else "isn't", "alive"