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 We need a total ordering of scopes so we can easily compare to variables
939 to see if they are concurrently in scope. To achieve this we record a
940 `scope_count` which is actually a count of both beginnings and endings
941 of scopes. Then each variable has a record of the scope count where it
942 enters scope, and where it leaves.
944 To push a new frame *before* any code in the frame is parsed, we need a
945 grammar reduction. This is most easily achieved with a grammar
946 element which derives the empty string, and creates the new scope when
947 it is recognised. This can be placed, for example, between a keyword
948 like "if" and the code following it.
952 struct scope *parent;
959 struct scope *scope_stack;
961 ###### variable fields
962 int scope_start, scope_end;
965 static void scope_pop(struct parse_context *c)
967 struct scope *s = c->scope_stack;
969 c->scope_stack = s->parent;
975 static void scope_push(struct parse_context *c)
977 struct scope *s = calloc(1, sizeof(*s));
979 c->scope_stack->child_count += 1;
980 s->parent = c->scope_stack;
989 OpenScope -> ${ scope_push(c); }$
991 Each variable records a scope depth and is in one of four states:
993 - "in scope". This is the case between the declaration of the
994 variable and the end of the containing block, and also between
995 the usage with affirms a merge and the end of that block.
997 The scope depth is not greater than the current parse context scope
998 nest depth. When the block of that depth closes, the state will
999 change. To achieve this, all "in scope" variables are linked
1000 together as a stack in nesting order.
1002 - "pending". The "in scope" block has closed, but other parallel
1003 scopes are still being processed. So far, every parallel block at
1004 the same level that has closed has declared the name.
1006 The scope depth is the depth of the last parallel block that
1007 enclosed the declaration, and that has closed.
1009 - "conditionally in scope". The "in scope" block and all parallel
1010 scopes have closed, and no further mention of the name has been seen.
1011 This state includes a secondary nest depth (`min_depth`) which records
1012 the outermost scope seen since the variable became conditionally in
1013 scope. If a use of the name is found, the variable becomes "in scope"
1014 and that secondary depth becomes the recorded scope depth. If the
1015 name is declared as a new variable, the old variable becomes "out of
1016 scope" and the recorded scope depth stays unchanged.
1018 - "out of scope". The variable is neither in scope nor conditionally
1019 in scope. It is permanently out of scope now and can be removed from
1020 the "in scope" stack. When a variable becomes out-of-scope it is
1021 moved to a separate list (`out_scope`) of variables which have fully
1022 known scope. This will be used at the end of each function to assign
1023 each variable a place in the stack frame.
1025 ###### variable fields
1026 int depth, min_depth;
1027 enum { OutScope, PendingScope, CondScope, InScope } scope;
1028 struct variable *in_scope;
1030 ###### parse context
1032 struct variable *in_scope;
1033 struct variable *out_scope;
1035 All variables with the same name are linked together using the
1036 'previous' link. Those variable that have been affirmatively merged all
1037 have a 'merged' pointer that points to one primary variable - the most
1038 recently declared instance. When merging variables, we need to also
1039 adjust the 'merged' pointer on any other variables that had previously
1040 been merged with the one that will no longer be primary.
1042 A variable that is no longer the most recent instance of a name may
1043 still have "pending" scope, if it might still be merged with most
1044 recent instance. These variables don't really belong in the
1045 "in_scope" list, but are not immediately removed when a new instance
1046 is found. Instead, they are detected and ignored when considering the
1047 list of in_scope names.
1049 The storage of the value of a variable will be described later. For now
1050 we just need to know that when a variable goes out of scope, it might
1051 need to be freed. For this we need to be able to find it, so assume that
1052 `var_value()` will provide that.
1054 ###### variable fields
1055 struct variable *merged;
1057 ###### ast functions
1059 static void variable_merge(struct variable *primary, struct variable *secondary)
1063 primary = primary->merged;
1065 for (v = primary->previous; v; v=v->previous)
1066 if (v == secondary || v == secondary->merged ||
1067 v->merged == secondary ||
1068 v->merged == secondary->merged) {
1069 v->scope = OutScope;
1070 v->merged = primary;
1071 if (v->scope_start < primary->scope_start)
1072 primary->scope_start = v->scope_start;
1073 if (v->scope_end > primary->scope_end)
1074 primary->scope_end = v->scope_end; // NOTEST
1075 variable_unlink_exec(v);
1079 ###### forward decls
1080 static struct value *var_value(struct parse_context *c, struct variable *v);
1082 ###### free global vars
1084 while (context.varlist) {
1085 struct binding *b = context.varlist;
1086 struct variable *v = b->var;
1087 context.varlist = b->next;
1090 struct variable *next = v->previous;
1093 free_value(v->type, var_value(&context, v));
1095 // This is a global constant
1096 free_exec(v->where_decl);
1103 #### Manipulating Bindings
1105 When a name is conditionally visible, a new declaration discards the old
1106 binding - the condition lapses. Similarly when we reach the end of a
1107 function (outermost non-global scope) any conditional scope must lapse.
1108 Conversely a usage of the name affirms the visibility and extends it to
1109 the end of the containing block - i.e. the block that contains both the
1110 original declaration and the latest usage. This is determined from
1111 `min_depth`. When a conditionally visible variable gets affirmed like
1112 this, it is also merged with other conditionally visible variables with
1115 When we parse a variable declaration we either report an error if the
1116 name is currently bound, or create a new variable at the current nest
1117 depth if the name is unbound or bound to a conditionally scoped or
1118 pending-scope variable. If the previous variable was conditionally
1119 scoped, it and its homonyms becomes out-of-scope.
1121 When we parse a variable reference (including non-declarative assignment
1122 "foo = bar") we report an error if the name is not bound or is bound to
1123 a pending-scope variable; update the scope if the name is bound to a
1124 conditionally scoped variable; or just proceed normally if the named
1125 variable is in scope.
1127 When we exit a scope, any variables bound at this level are either
1128 marked out of scope or pending-scoped, depending on whether the scope
1129 was sequential or parallel. Here a "parallel" scope means the "then"
1130 or "else" part of a conditional, or any "case" or "else" branch of a
1131 switch. Other scopes are "sequential".
1133 When exiting a parallel scope we check if there are any variables that
1134 were previously pending and are still visible. If there are, then
1135 they weren't redeclared in the most recent scope, so they cannot be
1136 merged and must become out-of-scope. If it is not the first of
1137 parallel scopes (based on `child_count`), we check that there was a
1138 previous binding that is still pending-scope. If there isn't, the new
1139 variable must now be out-of-scope.
1141 When exiting a sequential scope that immediately enclosed parallel
1142 scopes, we need to resolve any pending-scope variables. If there was
1143 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1144 we need to mark all pending-scope variable as out-of-scope. Otherwise
1145 all pending-scope variables become conditionally scoped.
1148 enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
1150 ###### ast functions
1152 static struct variable *var_decl(struct parse_context *c, struct text s)
1154 struct binding *b = find_binding(c, s);
1155 struct variable *v = b->var;
1157 switch (v ? v->scope : OutScope) {
1159 /* Caller will report the error */
1163 v && v->scope == CondScope;
1165 v->scope = OutScope;
1169 v = calloc(1, sizeof(*v));
1170 v->previous = b->var;
1174 v->min_depth = v->depth = c->scope_depth;
1176 v->in_scope = c->in_scope;
1177 v->scope_start = c->scope_count;
1183 static struct variable *var_ref(struct parse_context *c, struct text s)
1185 struct binding *b = find_binding(c, s);
1186 struct variable *v = b->var;
1187 struct variable *v2;
1189 switch (v ? v->scope : OutScope) {
1192 /* Caller will report the error */
1195 /* All CondScope variables of this name need to be merged
1196 * and become InScope
1198 v->depth = v->min_depth;
1200 for (v2 = v->previous;
1201 v2 && v2->scope == CondScope;
1203 variable_merge(v, v2);
1211 static int var_refile(struct parse_context *c, struct variable *v)
1213 /* Variable just went out of scope. Add it to the out_scope
1214 * list, sorted by ->scope_start
1216 struct variable **vp = &c->out_scope;
1217 while ((*vp) && (*vp)->scope_start < v->scope_start)
1218 vp = &(*vp)->in_scope;
1224 static void var_block_close(struct parse_context *c, enum closetype ct,
1227 /* Close off all variables that are in_scope.
1228 * Some variables in c->scope may already be not-in-scope,
1229 * such as when a PendingScope variable is hidden by a new
1230 * variable with the same name.
1231 * So we check for v->name->var != v and drop them.
1232 * If we choose to make a variable OutScope, we drop it
1235 struct variable *v, **vp, *v2;
1238 for (vp = &c->in_scope;
1239 (v = *vp) && v->min_depth > c->scope_depth;
1240 (v->scope == OutScope || v->name->var != v)
1241 ? (*vp = v->in_scope, var_refile(c, v))
1242 : ( vp = &v->in_scope, 0)) {
1243 v->min_depth = c->scope_depth;
1244 if (v->name->var != v)
1245 /* This is still in scope, but we haven't just
1249 v->min_depth = c->scope_depth;
1250 if (v->scope == InScope)
1251 v->scope_end = c->scope_count;
1252 if (v->scope == InScope && e && !v->global) {
1253 /* This variable gets cleaned up when 'e' finishes */
1254 variable_unlink_exec(v);
1255 v->cleanup_exec = e;
1256 v->next_free = e->to_free;
1261 case CloseParallel: /* handle PendingScope */
1265 if (c->scope_stack->child_count == 1)
1266 /* first among parallel branches */
1267 v->scope = PendingScope;
1268 else if (v->previous &&
1269 v->previous->scope == PendingScope)
1270 /* all previous branches used name */
1271 v->scope = PendingScope;
1272 else if (v->type == Tlabel)
1273 /* Labels remain pending even when not used */
1274 v->scope = PendingScope; // UNTESTED
1276 v->scope = OutScope;
1277 if (ct == CloseElse) {
1278 /* All Pending variables with this name
1279 * are now Conditional */
1281 v2 && v2->scope == PendingScope;
1283 v2->scope = CondScope;
1287 /* Not possible as it would require
1288 * parallel scope to be nested immediately
1289 * in a parallel scope, and that never
1293 /* Not possible as we already tested for
1300 if (v->scope == CondScope)
1301 /* Condition cannot continue past end of function */
1304 case CloseSequential:
1305 if (v->type == Tlabel)
1306 v->scope = PendingScope;
1309 v->scope = OutScope;
1312 /* There was no 'else', so we can only become
1313 * conditional if we know the cases were exhaustive,
1314 * and that doesn't mean anything yet.
1315 * So only labels become conditional..
1318 v2 && v2->scope == PendingScope;
1320 if (v2->type == Tlabel)
1321 v2->scope = CondScope;
1323 v2->scope = OutScope;
1326 case OutScope: break;
1335 The value of a variable is store separately from the variable, on an
1336 analogue of a stack frame. There are (currently) two frames that can be
1337 active. A global frame which currently only stores constants, and a
1338 stacked frame which stores local variables. Each variable knows if it
1339 is global or not, and what its index into the frame is.
1341 Values in the global frame are known immediately they are relevant, so
1342 the frame needs to be reallocated as it grows so it can store those
1343 values. The local frame doesn't get values until the interpreted phase
1344 is started, so there is no need to allocate until the size is known.
1346 We initialize the `frame_pos` to an impossible value, so that we can
1347 tell if it was set or not later.
1349 ###### variable fields
1353 ###### variable init
1356 ###### parse context
1358 short global_size, global_alloc;
1360 void *global, *local;
1362 ###### ast functions
1364 static struct value *var_value(struct parse_context *c, struct variable *v)
1367 if (!c->local || !v->type)
1368 return NULL; // NOTEST
1369 if (v->frame_pos + v->type->size > c->local_size) {
1370 printf("INVALID frame_pos\n"); // NOTEST
1373 return c->local + v->frame_pos;
1375 if (c->global_size > c->global_alloc) {
1376 int old = c->global_alloc;
1377 c->global_alloc = (c->global_size | 1023) + 1024;
1378 c->global = realloc(c->global, c->global_alloc);
1379 memset(c->global + old, 0, c->global_alloc - old);
1381 return c->global + v->frame_pos;
1384 static struct value *global_alloc(struct parse_context *c, struct type *t,
1385 struct variable *v, struct value *init)
1388 struct variable scratch;
1390 if (t->prepare_type)
1391 t->prepare_type(c, t, 1); // NOTEST
1393 if (c->global_size & (t->align - 1))
1394 c->global_size = (c->global_size + t->align) & ~(t->align-1);
1399 v->frame_pos = c->global_size;
1401 c->global_size += v->type->size;
1402 ret = var_value(c, v);
1404 memcpy(ret, init, t->size);
1410 As global values are found -- struct field initializers, labels etc --
1411 `global_alloc()` is called to record the value in the global frame.
1413 When the program is fully parsed, each function is analysed, we need to
1414 walk the list of variables local to that function and assign them an
1415 offset in the stack frame. For this we have `scope_finalize()`.
1417 We keep the stack from dense by re-using space for between variables
1418 that are not in scope at the same time. The `out_scope` list is sorted
1419 by `scope_start` and as we process a varible, we move it to an FIFO
1420 stack. For each variable we consider, we first discard any from the
1421 stack anything that went out of scope before the new variable came in.
1422 Then we place the new variable just after the one at the top of the
1425 ###### ast functions
1427 static void scope_finalize(struct parse_context *c, struct type *ft)
1430 struct variable *next = ft->function.scope;
1431 struct variable *done = NULL;
1433 struct variable *v = next;
1434 struct type *t = v->type;
1441 while (done && done->scope_end < v->scope_start)
1442 done = done->in_scope;
1444 pos = done->frame_pos + done->type->size;
1447 if (pos & (t->align - 1))
1448 pos = (pos + t->align) & ~(t->align-1);
1450 if (size < pos + v->type->size)
1451 size = pos + v->type->size;
1455 c->out_scope = NULL;
1456 ft->function.local_size = size;
1459 ###### free context storage
1460 free(context.global);
1464 Executables can be lots of different things. In many cases an
1465 executable is just an operation combined with one or two other
1466 executables. This allows for expressions and lists etc. Other times an
1467 executable is something quite specific like a constant or variable name.
1468 So we define a `struct exec` to be a general executable with a type, and
1469 a `struct binode` which is a subclass of `exec`, forms a node in a
1470 binary tree, and holds an operation. There will be other subclasses,
1471 and to access these we need to be able to `cast` the `exec` into the
1472 various other types. The first field in any `struct exec` is the type
1473 from the `exec_types` enum.
1476 #define cast(structname, pointer) ({ \
1477 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1478 if (__mptr && *__mptr != X##structname) abort(); \
1479 (struct structname *)( (char *)__mptr);})
1481 #define new(structname) ({ \
1482 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1483 __ptr->type = X##structname; \
1484 __ptr->line = -1; __ptr->column = -1; \
1487 #define new_pos(structname, token) ({ \
1488 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1489 __ptr->type = X##structname; \
1490 __ptr->line = token.line; __ptr->column = token.col; \
1499 enum exec_types type;
1508 struct exec *left, *right;
1511 ###### ast functions
1513 static int __fput_loc(struct exec *loc, FILE *f)
1517 if (loc->line >= 0) {
1518 fprintf(f, "%d:%d: ", loc->line, loc->column);
1521 if (loc->type == Xbinode)
1522 return __fput_loc(cast(binode,loc)->left, f) ||
1523 __fput_loc(cast(binode,loc)->right, f); // NOTEST
1526 static void fput_loc(struct exec *loc, FILE *f)
1528 if (!__fput_loc(loc, f))
1529 fprintf(f, "??:??: ");
1532 Each different type of `exec` node needs a number of functions defined,
1533 a bit like methods. We must be able to free it, print it, analyse it
1534 and execute it. Once we have specific `exec` types we will need to
1535 parse them too. Let's take this a bit more slowly.
1539 The parser generator requires a `free_foo` function for each struct
1540 that stores attributes and they will often be `exec`s and subtypes
1541 there-of. So we need `free_exec` which can handle all the subtypes,
1542 and we need `free_binode`.
1544 ###### ast functions
1546 static void free_binode(struct binode *b)
1551 free_exec(b->right);
1555 ###### core functions
1556 static void free_exec(struct exec *e)
1565 ###### forward decls
1567 static void free_exec(struct exec *e);
1569 ###### free exec cases
1570 case Xbinode: free_binode(cast(binode, e)); break;
1574 Printing an `exec` requires that we know the current indent level for
1575 printing line-oriented components. As will become clear later, we
1576 also want to know what sort of bracketing to use.
1578 ###### ast functions
1580 static void do_indent(int i, char *str)
1587 ###### core functions
1588 static void print_binode(struct binode *b, int indent, int bracket)
1592 ## print binode cases
1596 static void print_exec(struct exec *e, int indent, int bracket)
1602 print_binode(cast(binode, e), indent, bracket); break;
1607 do_indent(indent, "/* FREE");
1608 for (v = e->to_free; v; v = v->next_free) {
1609 printf(" %.*s", v->name->name.len, v->name->name.txt);
1610 printf("[%d,%d]", v->scope_start, v->scope_end);
1611 if (v->frame_pos >= 0)
1612 printf("(%d+%d)", v->frame_pos,
1613 v->type ? v->type->size:0);
1619 ###### forward decls
1621 static void print_exec(struct exec *e, int indent, int bracket);
1625 As discussed, analysis involves propagating type requirements around the
1626 program and looking for errors.
1628 So `propagate_types` is passed an expected type (being a `struct type`
1629 pointer together with some `val_rules` flags) that the `exec` is
1630 expected to return, and returns the type that it does return, either
1631 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1632 by reference. It is set to `0` when an error is found, and `2` when
1633 any change is made. If it remains unchanged at `1`, then no more
1634 propagation is needed.
1638 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 1<<2};
1642 if (rules & Rnolabel)
1643 fputs(" (labels not permitted)", stderr);
1646 ###### forward decls
1647 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1648 struct type *type, int rules);
1649 ###### core functions
1651 static struct type *__propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1652 struct type *type, int rules)
1659 switch (prog->type) {
1662 struct binode *b = cast(binode, prog);
1664 ## propagate binode cases
1668 ## propagate exec cases
1673 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1674 struct type *type, int rules)
1676 struct type *ret = __propagate_types(prog, c, ok, type, rules);
1685 Interpreting an `exec` doesn't require anything but the `exec`. State
1686 is stored in variables and each variable will be directly linked from
1687 within the `exec` tree. The exception to this is the `main` function
1688 which needs to look at command line arguments. This function will be
1689 interpreted separately.
1691 Each `exec` can return a value combined with a type in `struct lrval`.
1692 The type may be `Tnone` but must be non-NULL. Some `exec`s will return
1693 the location of a value, which can be updated, in `lval`. Others will
1694 set `lval` to NULL indicating that there is a value of appropriate type
1697 ###### core functions
1701 struct value rval, *lval;
1704 /* If dest is passed, dtype must give the expected type, and
1705 * result can go there, in which case type is returned as NULL.
1707 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
1708 struct value *dest, struct type *dtype);
1710 static struct value interp_exec(struct parse_context *c, struct exec *e,
1711 struct type **typeret)
1713 struct lrval ret = _interp_exec(c, e, NULL, NULL);
1715 if (!ret.type) abort();
1717 *typeret = ret.type;
1719 dup_value(ret.type, ret.lval, &ret.rval);
1723 static struct value *linterp_exec(struct parse_context *c, struct exec *e,
1724 struct type **typeret)
1726 struct lrval ret = _interp_exec(c, e, NULL, NULL);
1728 if (!ret.type) abort();
1730 *typeret = ret.type;
1732 free_value(ret.type, &ret.rval);
1736 /* dinterp_exec is used when the destination type is certain and
1737 * the value has a place to go.
1739 static void dinterp_exec(struct parse_context *c, struct exec *e,
1740 struct value *dest, struct type *dtype,
1743 struct lrval ret = _interp_exec(c, e, dest, dtype);
1747 free_value(dtype, dest);
1749 dup_value(dtype, ret.lval, dest);
1751 memcpy(dest, &ret.rval, dtype->size);
1754 static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
1755 struct value *dest, struct type *dtype)
1757 /* If the result is copied to dest, ret.type is set to NULL */
1759 struct value rv = {}, *lrv = NULL;
1760 struct type *rvtype;
1762 rvtype = ret.type = Tnone;
1772 struct binode *b = cast(binode, e);
1773 struct value left, right, *lleft;
1774 struct type *ltype, *rtype;
1775 ltype = rtype = Tnone;
1777 ## interp binode cases
1779 free_value(ltype, &left);
1780 free_value(rtype, &right);
1783 ## interp exec cases
1790 ## interp exec cleanup
1796 Now that we have the shape of the interpreter in place we can add some
1797 complex types and connected them in to the data structures and the
1798 different phases of parse, analyse, print, interpret.
1800 Thus far we have arrays and structs.
1804 Arrays can be declared by giving a size and a type, as `[size]type' so
1805 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
1806 size can be either a literal number, or a named constant. Some day an
1807 arbitrary expression will be supported.
1809 As a formal parameter to a function, the array can be declared with a
1810 new variable as the size: `name:[size::number]string`. The `size`
1811 variable is set to the size of the array and must be a constant. As
1812 `number` is the only supported type, it can be left out:
1813 `name:[size::]string`.
1815 Arrays cannot be assigned. When pointers are introduced we will also
1816 introduce array slices which can refer to part or all of an array -
1817 the assignment syntax will create a slice. For now, an array can only
1818 ever be referenced by the name it is declared with. It is likely that
1819 a "`copy`" primitive will eventually be define which can be used to
1820 make a copy of an array with controllable recursive depth.
1822 For now we have two sorts of array, those with fixed size either because
1823 it is given as a literal number or because it is a struct member (which
1824 cannot have a runtime-changing size), and those with a size that is
1825 determined at runtime - local variables with a const size. The former
1826 have their size calculated at parse time, the latter at run time.
1828 For the latter type, the `size` field of the type is the size of a
1829 pointer, and the array is reallocated every time it comes into scope.
1831 We differentiate struct fields with a const size from local variables
1832 with a const size by whether they are prepared at parse time or not.
1834 ###### type union fields
1837 int unspec; // size is unspecified - vsize must be set.
1840 struct variable *vsize;
1841 struct type *member;
1844 ###### value union fields
1845 void *array; // used if not static_size
1847 ###### value functions
1849 static void array_prepare_type(struct parse_context *c, struct type *type,
1852 struct value *vsize;
1854 if (!type->array.vsize || type->array.static_size)
1857 vsize = var_value(c, type->array.vsize);
1859 mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
1860 type->array.size = mpz_get_si(q);
1864 type->array.static_size = 1;
1865 type->size = type->array.size * type->array.member->size;
1866 type->align = type->array.member->align;
1870 static void array_init(struct type *type, struct value *val)
1873 void *ptr = val->ptr;
1877 if (!type->array.static_size) {
1878 val->array = calloc(type->array.size,
1879 type->array.member->size);
1882 for (i = 0; i < type->array.size; i++) {
1884 v = (void*)ptr + i * type->array.member->size;
1885 val_init(type->array.member, v);
1889 static void array_free(struct type *type, struct value *val)
1892 void *ptr = val->ptr;
1894 if (!type->array.static_size)
1896 for (i = 0; i < type->array.size; i++) {
1898 v = (void*)ptr + i * type->array.member->size;
1899 free_value(type->array.member, v);
1901 if (!type->array.static_size)
1905 static int array_compat(struct type *require, struct type *have)
1907 if (have->compat != require->compat)
1909 /* Both are arrays, so we can look at details */
1910 if (!type_compat(require->array.member, have->array.member, 0))
1912 if (have->array.unspec && require->array.unspec) {
1913 if (have->array.vsize && require->array.vsize &&
1914 have->array.vsize != require->array.vsize) // UNTESTED
1915 /* sizes might not be the same */
1916 return 0; // UNTESTED
1919 if (have->array.unspec || require->array.unspec)
1920 return 1; // UNTESTED
1921 if (require->array.vsize == NULL && have->array.vsize == NULL)
1922 return require->array.size == have->array.size;
1924 return require->array.vsize == have->array.vsize; // UNTESTED
1927 static void array_print_type(struct type *type, FILE *f)
1930 if (type->array.vsize) {
1931 struct binding *b = type->array.vsize->name;
1932 fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
1933 type->array.unspec ? "::" : "");
1935 fprintf(f, "%d]", type->array.size);
1936 type_print(type->array.member, f);
1939 static struct type array_prototype = {
1941 .prepare_type = array_prepare_type,
1942 .print_type = array_print_type,
1943 .compat = array_compat,
1945 .size = sizeof(void*),
1946 .align = sizeof(void*),
1949 ###### declare terminals
1954 | [ NUMBER ] Type ${ {
1957 struct text noname = { "", 0 };
1960 $0 = t = add_type(c, noname, &array_prototype);
1961 t->array.member = $<4;
1962 t->array.vsize = NULL;
1963 if (number_parse(num, tail, $2.txt) == 0)
1964 tok_err(c, "error: unrecognised number", &$2);
1966 tok_err(c, "error: unsupported number suffix", &$2);
1969 t->array.size = mpz_get_ui(mpq_numref(num));
1970 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
1971 tok_err(c, "error: array size must be an integer",
1973 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
1974 tok_err(c, "error: array size is too large",
1978 t->array.static_size = 1;
1979 t->size = t->array.size * t->array.member->size;
1980 t->align = t->array.member->align;
1983 | [ IDENTIFIER ] Type ${ {
1984 struct variable *v = var_ref(c, $2.txt);
1985 struct text noname = { "", 0 };
1988 tok_err(c, "error: name undeclared", &$2);
1989 else if (!v->constant)
1990 tok_err(c, "error: array size must be a constant", &$2);
1992 $0 = add_type(c, noname, &array_prototype);
1993 $0->array.member = $<4;
1995 $0->array.vsize = v;
2000 OptType -> Type ${ $0 = $<1; }$
2003 ###### formal type grammar
2005 | [ IDENTIFIER :: OptType ] Type ${ {
2006 struct variable *v = var_decl(c, $ID.txt);
2007 struct text noname = { "", 0 };
2013 $0 = add_type(c, noname, &array_prototype);
2014 $0->array.member = $<6;
2016 $0->array.unspec = 1;
2017 $0->array.vsize = v;
2023 ###### variable grammar
2025 | Variable [ Expression ] ${ {
2026 struct binode *b = new(binode);
2033 ###### print binode cases
2035 print_exec(b->left, -1, bracket);
2037 print_exec(b->right, -1, bracket);
2041 ###### propagate binode cases
2043 /* left must be an array, right must be a number,
2044 * result is the member type of the array
2046 propagate_types(b->right, c, ok, Tnum, 0);
2047 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
2048 if (!t || t->compat != array_compat) {
2049 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
2052 if (!type_compat(type, t->array.member, rules)) {
2053 type_err(c, "error: have %1 but need %2", prog,
2054 t->array.member, rules, type);
2056 return t->array.member;
2060 ###### interp binode cases
2066 lleft = linterp_exec(c, b->left, <ype);
2067 right = interp_exec(c, b->right, &rtype);
2069 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
2073 if (ltype->array.static_size)
2076 ptr = *(void**)lleft;
2077 rvtype = ltype->array.member;
2078 if (i >= 0 && i < ltype->array.size)
2079 lrv = ptr + i * rvtype->size;
2081 val_init(ltype->array.member, &rv); // UNSAFE
2088 A `struct` is a data-type that contains one or more other data-types.
2089 It differs from an array in that each member can be of a different
2090 type, and they are accessed by name rather than by number. Thus you
2091 cannot choose an element by calculation, you need to know what you
2094 The language makes no promises about how a given structure will be
2095 stored in memory - it is free to rearrange fields to suit whatever
2096 criteria seems important.
2098 Structs are declared separately from program code - they cannot be
2099 declared in-line in a variable declaration like arrays can. A struct
2100 is given a name and this name is used to identify the type - the name
2101 is not prefixed by the word `struct` as it would be in C.
2103 Structs are only treated as the same if they have the same name.
2104 Simply having the same fields in the same order is not enough. This
2105 might change once we can create structure initializers from a list of
2108 Each component datum is identified much like a variable is declared,
2109 with a name, one or two colons, and a type. The type cannot be omitted
2110 as there is no opportunity to deduce the type from usage. An initial
2111 value can be given following an equals sign, so
2113 ##### Example: a struct type
2119 would declare a type called "complex" which has two number fields,
2120 each initialised to zero.
2122 Struct will need to be declared separately from the code that uses
2123 them, so we will need to be able to print out the declaration of a
2124 struct when reprinting the whole program. So a `print_type_decl` type
2125 function will be needed.
2127 ###### type union fields
2139 ###### type functions
2140 void (*print_type_decl)(struct type *type, FILE *f);
2142 ###### value functions
2144 static void structure_init(struct type *type, struct value *val)
2148 for (i = 0; i < type->structure.nfields; i++) {
2150 v = (void*) val->ptr + type->structure.fields[i].offset;
2151 if (type->structure.fields[i].init)
2152 dup_value(type->structure.fields[i].type,
2153 type->structure.fields[i].init,
2156 val_init(type->structure.fields[i].type, v);
2160 static void structure_free(struct type *type, struct value *val)
2164 for (i = 0; i < type->structure.nfields; i++) {
2166 v = (void*)val->ptr + type->structure.fields[i].offset;
2167 free_value(type->structure.fields[i].type, v);
2171 static void structure_free_type(struct type *t)
2174 for (i = 0; i < t->structure.nfields; i++)
2175 if (t->structure.fields[i].init) {
2176 free_value(t->structure.fields[i].type,
2177 t->structure.fields[i].init);
2179 free(t->structure.fields);
2182 static struct type structure_prototype = {
2183 .init = structure_init,
2184 .free = structure_free,
2185 .free_type = structure_free_type,
2186 .print_type_decl = structure_print_type,
2200 ###### free exec cases
2202 free_exec(cast(fieldref, e)->left);
2206 ###### declare terminals
2209 ###### variable grammar
2211 | Variable . IDENTIFIER ${ {
2212 struct fieldref *fr = new_pos(fieldref, $2);
2219 ###### print exec cases
2223 struct fieldref *f = cast(fieldref, e);
2224 print_exec(f->left, -1, bracket);
2225 printf(".%.*s", f->name.len, f->name.txt);
2229 ###### ast functions
2230 static int find_struct_index(struct type *type, struct text field)
2233 for (i = 0; i < type->structure.nfields; i++)
2234 if (text_cmp(type->structure.fields[i].name, field) == 0)
2239 ###### propagate exec cases
2243 struct fieldref *f = cast(fieldref, prog);
2244 struct type *st = propagate_types(f->left, c, ok, NULL, 0);
2247 type_err(c, "error: unknown type for field access", f->left, // UNTESTED
2249 else if (st->init != structure_init)
2250 type_err(c, "error: field reference attempted on %1, not a struct",
2251 f->left, st, 0, NULL);
2252 else if (f->index == -2) {
2253 f->index = find_struct_index(st, f->name);
2255 type_err(c, "error: cannot find requested field in %1",
2256 f->left, st, 0, NULL);
2258 if (f->index >= 0) {
2259 struct type *ft = st->structure.fields[f->index].type;
2260 if (!type_compat(type, ft, rules))
2261 type_err(c, "error: have %1 but need %2", prog,
2268 ###### interp exec cases
2271 struct fieldref *f = cast(fieldref, e);
2273 struct value *lleft = linterp_exec(c, f->left, <ype);
2274 lrv = (void*)lleft->ptr + ltype->structure.fields[f->index].offset;
2275 rvtype = ltype->structure.fields[f->index].type;
2281 struct fieldlist *prev;
2285 ###### ast functions
2286 static void free_fieldlist(struct fieldlist *f)
2290 free_fieldlist(f->prev);
2292 free_value(f->f.type, f->f.init); // UNTESTED
2293 free(f->f.init); // UNTESTED
2298 ###### top level grammar
2299 DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
2301 add_type(c, $2.txt, &structure_prototype);
2303 struct fieldlist *f;
2305 for (f = $3; f; f=f->prev)
2308 t->structure.nfields = cnt;
2309 t->structure.fields = calloc(cnt, sizeof(struct field));
2312 int a = f->f.type->align;
2314 t->structure.fields[cnt] = f->f;
2315 if (t->size & (a-1))
2316 t->size = (t->size | (a-1)) + 1;
2317 t->structure.fields[cnt].offset = t->size;
2318 t->size += ((f->f.type->size - 1) | (a-1)) + 1;
2327 FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
2328 | { SimpleFieldList } ${ $0 = $<SFL; }$
2329 | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
2330 | SimpleFieldList EOL ${ $0 = $<SFL; }$
2332 FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
2333 | FieldLines SimpleFieldList Newlines ${
2338 SimpleFieldList -> Field ${ $0 = $<F; }$
2339 | SimpleFieldList ; Field ${
2343 | SimpleFieldList ; ${
2346 | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
2348 Field -> IDENTIFIER : Type = Expression ${ {
2351 $0 = calloc(1, sizeof(struct fieldlist));
2352 $0->f.name = $1.txt;
2357 propagate_types($<5, c, &ok, $3, 0);
2360 c->parse_error = 1; // UNTESTED
2362 struct value vl = interp_exec(c, $5, NULL);
2363 $0->f.init = global_alloc(c, $0->f.type, NULL, &vl);
2366 | IDENTIFIER : Type ${
2367 $0 = calloc(1, sizeof(struct fieldlist));
2368 $0->f.name = $1.txt;
2370 if ($0->f.type->prepare_type)
2371 $0->f.type->prepare_type(c, $0->f.type, 1);
2374 ###### forward decls
2375 static void structure_print_type(struct type *t, FILE *f);
2377 ###### value functions
2378 static void structure_print_type(struct type *t, FILE *f)
2382 fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
2384 for (i = 0; i < t->structure.nfields; i++) {
2385 struct field *fl = t->structure.fields + i;
2386 fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
2387 type_print(fl->type, f);
2388 if (fl->type->print && fl->init) {
2390 if (fl->type == Tstr)
2391 fprintf(f, "\""); // UNTESTED
2392 print_value(fl->type, fl->init);
2393 if (fl->type == Tstr)
2394 fprintf(f, "\""); // UNTESTED
2400 ###### print type decls
2405 while (target != 0) {
2407 for (t = context.typelist; t ; t=t->next)
2408 if (t->print_type_decl && !t->check_args) {
2417 t->print_type_decl(t, stdout);
2425 A function is a chunk of code which can be passed parameters and can
2426 return results. Each function has a type which includes the set of
2427 parameters and the return value. As yet these types cannot be declared
2428 separately from the function itself.
2430 The parameters can be specified either in parentheses as a ';' separated
2433 ##### Example: function 1
2435 func main(av:[ac::number]string; env:[envc::number]string)
2438 or as an indented list of one parameter per line (though each line can
2439 be a ';' separated list)
2441 ##### Example: function 2
2444 argv:[argc::number]string
2445 env:[envc::number]string
2449 In the first case a return type can follow the paentheses after a colon,
2450 in the second it is given on a line starting with the word `return`.
2452 ##### Example: functions that return
2454 func add(a:number; b:number): number
2465 For constructing these lists we use a `List` binode, which will be
2466 further detailed when Expression Lists are introduced.
2468 ###### type union fields
2471 struct binode *params;
2472 struct type *return_type;
2473 struct variable *scope;
2477 ###### value union fields
2478 struct exec *function;
2480 ###### type functions
2481 void (*check_args)(struct parse_context *c, int *ok,
2482 struct type *require, struct exec *args);
2484 ###### value functions
2486 static void function_free(struct type *type, struct value *val)
2488 free_exec(val->function);
2489 val->function = NULL;
2492 static int function_compat(struct type *require, struct type *have)
2494 // FIXME can I do anything here yet?
2498 static void function_check_args(struct parse_context *c, int *ok,
2499 struct type *require, struct exec *args)
2501 /* This should be 'compat', but we don't have a 'tuple' type to
2502 * hold the type of 'args'
2504 struct binode *arg = cast(binode, args);
2505 struct binode *param = require->function.params;
2508 struct var *pv = cast(var, param->left);
2510 type_err(c, "error: insufficient arguments to function.",
2511 args, NULL, 0, NULL);
2515 propagate_types(arg->left, c, ok, pv->var->type, 0);
2516 param = cast(binode, param->right);
2517 arg = cast(binode, arg->right);
2520 type_err(c, "error: too many arguments to function.",
2521 args, NULL, 0, NULL);
2524 static void function_print(struct type *type, struct value *val)
2526 print_exec(val->function, 1, 0);
2529 static void function_print_type_decl(struct type *type, FILE *f)
2533 for (b = type->function.params; b; b = cast(binode, b->right)) {
2534 struct variable *v = cast(var, b->left)->var;
2535 fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
2536 v->constant ? "::" : ":");
2537 type_print(v->type, f);
2542 if (type->function.return_type != Tnone) {
2544 type_print(type->function.return_type, f);
2549 static void function_free_type(struct type *t)
2551 free_exec(t->function.params);
2554 static struct type function_prototype = {
2555 .size = sizeof(void*),
2556 .align = sizeof(void*),
2557 .free = function_free,
2558 .compat = function_compat,
2559 .check_args = function_check_args,
2560 .print = function_print,
2561 .print_type_decl = function_print_type_decl,
2562 .free_type = function_free_type,
2565 ###### declare terminals
2575 FuncName -> IDENTIFIER ${ {
2576 struct variable *v = var_decl(c, $1.txt);
2577 struct var *e = new_pos(var, $1);
2583 v = var_ref(c, $1.txt);
2585 type_err(c, "error: function '%v' redeclared",
2587 type_err(c, "info: this is where '%v' was first declared",
2588 v->where_decl, NULL, 0, NULL);
2594 Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
2595 | Args ArgsLine NEWLINE ${ {
2596 struct binode *b = $<AL;
2597 struct binode **bp = &b;
2599 bp = (struct binode **)&(*bp)->left;
2604 ArgsLine -> ${ $0 = NULL; }$
2605 | Varlist ${ $0 = $<1; }$
2606 | Varlist ; ${ $0 = $<1; }$
2608 Varlist -> Varlist ; ArgDecl ${
2622 ArgDecl -> IDENTIFIER : FormalType ${ {
2623 struct variable *v = var_decl(c, $1.txt);
2629 ## Executables: the elements of code
2631 Each code element needs to be parsed, printed, analysed,
2632 interpreted, and freed. There are several, so let's just start with
2633 the easy ones and work our way up.
2637 We have already met values as separate objects. When manifest
2638 constants appear in the program text, that must result in an executable
2639 which has a constant value. So the `val` structure embeds a value in
2652 ###### ast functions
2653 struct val *new_val(struct type *T, struct token tk)
2655 struct val *v = new_pos(val, tk);
2666 $0 = new_val(Tbool, $1);
2670 $0 = new_val(Tbool, $1);
2674 $0 = new_val(Tnum, $1);
2677 if (number_parse($0->val.num, tail, $1.txt) == 0)
2678 mpq_init($0->val.num); // UNTESTED
2680 tok_err(c, "error: unsupported number suffix",
2685 $0 = new_val(Tstr, $1);
2688 string_parse(&$1, '\\', &$0->val.str, tail);
2690 tok_err(c, "error: unsupported string suffix",
2695 $0 = new_val(Tstr, $1);
2698 string_parse(&$1, '\\', &$0->val.str, tail);
2700 tok_err(c, "error: unsupported string suffix",
2705 ###### print exec cases
2708 struct val *v = cast(val, e);
2709 if (v->vtype == Tstr)
2711 print_value(v->vtype, &v->val);
2712 if (v->vtype == Tstr)
2717 ###### propagate exec cases
2720 struct val *val = cast(val, prog);
2721 if (!type_compat(type, val->vtype, rules))
2722 type_err(c, "error: expected %1%r found %2",
2723 prog, type, rules, val->vtype);
2727 ###### interp exec cases
2729 rvtype = cast(val, e)->vtype;
2730 dup_value(rvtype, &cast(val, e)->val, &rv);
2733 ###### ast functions
2734 static void free_val(struct val *v)
2737 free_value(v->vtype, &v->val);
2741 ###### free exec cases
2742 case Xval: free_val(cast(val, e)); break;
2744 ###### ast functions
2745 // Move all nodes from 'b' to 'rv', reversing their order.
2746 // In 'b' 'left' is a list, and 'right' is the last node.
2747 // In 'rv', left' is the first node and 'right' is a list.
2748 static struct binode *reorder_bilist(struct binode *b)
2750 struct binode *rv = NULL;
2753 struct exec *t = b->right;
2757 b = cast(binode, b->left);
2767 Just as we used a `val` to wrap a value into an `exec`, we similarly
2768 need a `var` to wrap a `variable` into an exec. While each `val`
2769 contained a copy of the value, each `var` holds a link to the variable
2770 because it really is the same variable no matter where it appears.
2771 When a variable is used, we need to remember to follow the `->merged`
2772 link to find the primary instance.
2780 struct variable *var;
2788 VariableDecl -> IDENTIFIER : ${ {
2789 struct variable *v = var_decl(c, $1.txt);
2790 $0 = new_pos(var, $1);
2795 v = var_ref(c, $1.txt);
2797 type_err(c, "error: variable '%v' redeclared",
2799 type_err(c, "info: this is where '%v' was first declared",
2800 v->where_decl, NULL, 0, NULL);
2803 | IDENTIFIER :: ${ {
2804 struct variable *v = var_decl(c, $1.txt);
2805 $0 = new_pos(var, $1);
2811 v = var_ref(c, $1.txt);
2813 type_err(c, "error: variable '%v' redeclared",
2815 type_err(c, "info: this is where '%v' was first declared",
2816 v->where_decl, NULL, 0, NULL);
2819 | IDENTIFIER : Type ${ {
2820 struct variable *v = var_decl(c, $1.txt);
2821 $0 = new_pos(var, $1);
2828 v = var_ref(c, $1.txt);
2830 type_err(c, "error: variable '%v' redeclared",
2832 type_err(c, "info: this is where '%v' was first declared",
2833 v->where_decl, NULL, 0, NULL);
2836 | IDENTIFIER :: Type ${ {
2837 struct variable *v = var_decl(c, $1.txt);
2838 $0 = new_pos(var, $1);
2846 v = var_ref(c, $1.txt);
2848 type_err(c, "error: variable '%v' redeclared",
2850 type_err(c, "info: this is where '%v' was first declared",
2851 v->where_decl, NULL, 0, NULL);
2856 Variable -> IDENTIFIER ${ {
2857 struct variable *v = var_ref(c, $1.txt);
2858 $0 = new_pos(var, $1);
2860 /* This might be a label - allocate a var just in case */
2861 v = var_decl(c, $1.txt);
2868 cast(var, $0)->var = v;
2872 ###### print exec cases
2875 struct var *v = cast(var, e);
2877 struct binding *b = v->var->name;
2878 printf("%.*s", b->name.len, b->name.txt);
2885 if (loc && loc->type == Xvar) {
2886 struct var *v = cast(var, loc);
2888 struct binding *b = v->var->name;
2889 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
2891 fputs("???", stderr); // NOTEST
2893 fputs("NOTVAR", stderr);
2896 ###### propagate exec cases
2900 struct var *var = cast(var, prog);
2901 struct variable *v = var->var;
2903 type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
2904 return Tnone; // NOTEST
2907 if (v->constant && (rules & Rnoconstant)) {
2908 type_err(c, "error: Cannot assign to a constant: %v",
2909 prog, NULL, 0, NULL);
2910 type_err(c, "info: name was defined as a constant here",
2911 v->where_decl, NULL, 0, NULL);
2914 if (v->type == Tnone && v->where_decl == prog)
2915 type_err(c, "error: variable used but not declared: %v",
2916 prog, NULL, 0, NULL);
2917 if (v->type == NULL) {
2918 if (type && *ok != 0) {
2920 v->where_set = prog;
2925 if (!type_compat(type, v->type, rules)) {
2926 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
2927 type, rules, v->type);
2928 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
2929 v->type, rules, NULL);
2936 ###### interp exec cases
2939 struct var *var = cast(var, e);
2940 struct variable *v = var->var;
2943 lrv = var_value(c, v);
2948 ###### ast functions
2950 static void free_var(struct var *v)
2955 ###### free exec cases
2956 case Xvar: free_var(cast(var, e)); break;
2958 ### Expressions: Conditional
2960 Our first user of the `binode` will be conditional expressions, which
2961 is a bit odd as they actually have three components. That will be
2962 handled by having 2 binodes for each expression. The conditional
2963 expression is the lowest precedence operator which is why we define it
2964 first - to start the precedence list.
2966 Conditional expressions are of the form "value `if` condition `else`
2967 other_value". They associate to the right, so everything to the right
2968 of `else` is part of an else value, while only a higher-precedence to
2969 the left of `if` is the if values. Between `if` and `else` there is no
2970 room for ambiguity, so a full conditional expression is allowed in
2982 Expression -> Expression if Expression else Expression $$ifelse ${ {
2983 struct binode *b1 = new(binode);
2984 struct binode *b2 = new(binode);
2993 ## expression grammar
2995 ###### print binode cases
2998 b2 = cast(binode, b->right);
2999 if (bracket) printf("(");
3000 print_exec(b2->left, -1, bracket);
3002 print_exec(b->left, -1, bracket);
3004 print_exec(b2->right, -1, bracket);
3005 if (bracket) printf(")");
3008 ###### propagate binode cases
3011 /* cond must be Tbool, others must match */
3012 struct binode *b2 = cast(binode, b->right);
3015 propagate_types(b->left, c, ok, Tbool, 0);
3016 t = propagate_types(b2->left, c, ok, type, Rnolabel);
3017 t2 = propagate_types(b2->right, c, ok, type ?: t, Rnolabel);
3021 ###### interp binode cases
3024 struct binode *b2 = cast(binode, b->right);
3025 left = interp_exec(c, b->left, <ype);
3027 rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
3029 rv = interp_exec(c, b2->right, &rvtype);
3035 We take a brief detour, now that we have expressions, to describe lists
3036 of expressions. These will be needed for function parameters and
3037 possibly other situations. They seem generic enough to introduce here
3038 to be used elsewhere.
3040 And ExpressionList will use the `List` type of `binode`, building up at
3041 the end. And place where they are used will probably call
3042 `reorder_bilist()` to get a more normal first/next arrangement.
3044 ###### declare terminals
3047 `List` execs have no implicit semantics, so they are never propagated or
3048 interpreted. The can be printed as a comma separate list, which is how
3049 they are parsed. Note they are also used for function formal parameter
3050 lists. In that case a separate function is used to print them.
3052 ###### print binode cases
3056 print_exec(b->left, -1, bracket);
3059 b = cast(binode, b->right);
3063 ###### propagate binode cases
3064 case List: abort(); // NOTEST
3065 ###### interp binode cases
3066 case List: abort(); // NOTEST
3071 ExpressionList -> ExpressionList , Expression ${
3084 ### Expressions: Boolean
3086 The next class of expressions to use the `binode` will be Boolean
3087 expressions. "`and then`" and "`or else`" are similar to `and` and `or`
3088 have same corresponding precendence. The difference is that they don't
3089 evaluate the second expression if not necessary.
3098 ###### expr precedence
3103 ###### expression grammar
3104 | Expression or Expression ${ {
3105 struct binode *b = new(binode);
3111 | Expression or else Expression ${ {
3112 struct binode *b = new(binode);
3119 | Expression and Expression ${ {
3120 struct binode *b = new(binode);
3126 | Expression and then Expression ${ {
3127 struct binode *b = new(binode);
3134 | not Expression ${ {
3135 struct binode *b = new(binode);
3141 ###### print binode cases
3143 if (bracket) printf("(");
3144 print_exec(b->left, -1, bracket);
3146 print_exec(b->right, -1, bracket);
3147 if (bracket) printf(")");
3150 if (bracket) printf("(");
3151 print_exec(b->left, -1, bracket);
3152 printf(" and then ");
3153 print_exec(b->right, -1, bracket);
3154 if (bracket) printf(")");
3157 if (bracket) printf("(");
3158 print_exec(b->left, -1, bracket);
3160 print_exec(b->right, -1, bracket);
3161 if (bracket) printf(")");
3164 if (bracket) printf("(");
3165 print_exec(b->left, -1, bracket);
3166 printf(" or else ");
3167 print_exec(b->right, -1, bracket);
3168 if (bracket) printf(")");
3171 if (bracket) printf("(");
3173 print_exec(b->right, -1, bracket);
3174 if (bracket) printf(")");
3177 ###### propagate binode cases
3183 /* both must be Tbool, result is Tbool */
3184 propagate_types(b->left, c, ok, Tbool, 0);
3185 propagate_types(b->right, c, ok, Tbool, 0);
3186 if (type && type != Tbool)
3187 type_err(c, "error: %1 operation found where %2 expected", prog,
3191 ###### interp binode cases
3193 rv = interp_exec(c, b->left, &rvtype);
3194 right = interp_exec(c, b->right, &rtype);
3195 rv.bool = rv.bool && right.bool;
3198 rv = interp_exec(c, b->left, &rvtype);
3200 rv = interp_exec(c, b->right, NULL);
3203 rv = interp_exec(c, b->left, &rvtype);
3204 right = interp_exec(c, b->right, &rtype);
3205 rv.bool = rv.bool || right.bool;
3208 rv = interp_exec(c, b->left, &rvtype);
3210 rv = interp_exec(c, b->right, NULL);
3213 rv = interp_exec(c, b->right, &rvtype);
3217 ### Expressions: Comparison
3219 Of slightly higher precedence that Boolean expressions are Comparisons.
3220 A comparison takes arguments of any comparable type, but the two types
3223 To simplify the parsing we introduce an `eop` which can record an
3224 expression operator, and the `CMPop` non-terminal will match one of them.
3231 ###### ast functions
3232 static void free_eop(struct eop *e)
3246 ###### expr precedence
3247 $LEFT < > <= >= == != CMPop
3249 ###### expression grammar
3250 | Expression CMPop Expression ${ {
3251 struct binode *b = new(binode);
3261 CMPop -> < ${ $0.op = Less; }$
3262 | > ${ $0.op = Gtr; }$
3263 | <= ${ $0.op = LessEq; }$
3264 | >= ${ $0.op = GtrEq; }$
3265 | == ${ $0.op = Eql; }$
3266 | != ${ $0.op = NEql; }$
3268 ###### print binode cases
3276 if (bracket) printf("(");
3277 print_exec(b->left, -1, bracket);
3279 case Less: printf(" < "); break;
3280 case LessEq: printf(" <= "); break;
3281 case Gtr: printf(" > "); break;
3282 case GtrEq: printf(" >= "); break;
3283 case Eql: printf(" == "); break;
3284 case NEql: printf(" != "); break;
3285 default: abort(); // NOTEST
3287 print_exec(b->right, -1, bracket);
3288 if (bracket) printf(")");
3291 ###### propagate binode cases
3298 /* Both must match but not be labels, result is Tbool */
3299 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
3301 propagate_types(b->right, c, ok, t, 0);
3303 t = propagate_types(b->right, c, ok, NULL, Rnolabel); // UNTESTED
3305 t = propagate_types(b->left, c, ok, t, 0); // UNTESTED
3307 if (!type_compat(type, Tbool, 0))
3308 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
3309 Tbool, rules, type);
3312 ###### interp binode cases
3321 left = interp_exec(c, b->left, <ype);
3322 right = interp_exec(c, b->right, &rtype);
3323 cmp = value_cmp(ltype, rtype, &left, &right);
3326 case Less: rv.bool = cmp < 0; break;
3327 case LessEq: rv.bool = cmp <= 0; break;
3328 case Gtr: rv.bool = cmp > 0; break;
3329 case GtrEq: rv.bool = cmp >= 0; break;
3330 case Eql: rv.bool = cmp == 0; break;
3331 case NEql: rv.bool = cmp != 0; break;
3332 default: rv.bool = 0; break; // NOTEST
3337 ### Expressions: Arithmetic etc.
3339 The remaining expressions with the highest precedence are arithmetic,
3340 string concatenation, and string conversion. String concatenation
3341 (`++`) has the same precedence as multiplication and division, but lower
3344 String conversion is a temporary feature until I get a better type
3345 system. `$` is a prefix operator which expects a string and returns
3348 `+` and `-` are both infix and prefix operations (where they are
3349 absolute value and negation). These have different operator names.
3351 We also have a 'Bracket' operator which records where parentheses were
3352 found. This makes it easy to reproduce these when printing. Possibly I
3353 should only insert brackets were needed for precedence.
3363 ###### expr precedence
3369 ###### expression grammar
3370 | Expression Eop Expression ${ {
3371 struct binode *b = new(binode);
3378 | Expression Top Expression ${ {
3379 struct binode *b = new(binode);
3386 | ( Expression ) ${ {
3387 struct binode *b = new_pos(binode, $1);
3392 | Uop Expression ${ {
3393 struct binode *b = new(binode);
3398 | Value ${ $0 = $<1; }$
3399 | Variable ${ $0 = $<1; }$
3404 Eop -> + ${ $0.op = Plus; }$
3405 | - ${ $0.op = Minus; }$
3407 Uop -> + ${ $0.op = Absolute; }$
3408 | - ${ $0.op = Negate; }$
3409 | $ ${ $0.op = StringConv; }$
3411 Top -> * ${ $0.op = Times; }$
3412 | / ${ $0.op = Divide; }$
3413 | % ${ $0.op = Rem; }$
3414 | ++ ${ $0.op = Concat; }$
3416 ###### print binode cases
3423 if (bracket) printf("(");
3424 print_exec(b->left, indent, bracket);
3426 case Plus: fputs(" + ", stdout); break;
3427 case Minus: fputs(" - ", stdout); break;
3428 case Times: fputs(" * ", stdout); break;
3429 case Divide: fputs(" / ", stdout); break;
3430 case Rem: fputs(" % ", stdout); break;
3431 case Concat: fputs(" ++ ", stdout); break;
3432 default: abort(); // NOTEST
3434 print_exec(b->right, indent, bracket);
3435 if (bracket) printf(")");
3440 if (bracket) printf("(");
3442 case Absolute: fputs("+", stdout); break;
3443 case Negate: fputs("-", stdout); break;
3444 case StringConv: fputs("$", stdout); break;
3445 default: abort(); // NOTEST
3447 print_exec(b->right, indent, bracket);
3448 if (bracket) printf(")");
3452 print_exec(b->right, indent, bracket);
3456 ###### propagate binode cases
3462 /* both must be numbers, result is Tnum */
3465 /* as propagate_types ignores a NULL,
3466 * unary ops fit here too */
3467 propagate_types(b->left, c, ok, Tnum, 0);
3468 propagate_types(b->right, c, ok, Tnum, 0);
3469 if (!type_compat(type, Tnum, 0))
3470 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
3475 /* both must be Tstr, result is Tstr */
3476 propagate_types(b->left, c, ok, Tstr, 0);
3477 propagate_types(b->right, c, ok, Tstr, 0);
3478 if (!type_compat(type, Tstr, 0))
3479 type_err(c, "error: Concat returns %1 but %2 expected", prog,
3484 /* op must be string, result is number */
3485 propagate_types(b->left, c, ok, Tstr, 0);
3486 if (!type_compat(type, Tnum, 0))
3487 type_err(c, // UNTESTED
3488 "error: Can only convert string to number, not %1",
3489 prog, type, 0, NULL);
3493 return propagate_types(b->right, c, ok, type, 0);
3495 ###### interp binode cases
3498 rv = interp_exec(c, b->left, &rvtype);
3499 right = interp_exec(c, b->right, &rtype);
3500 mpq_add(rv.num, rv.num, right.num);
3503 rv = interp_exec(c, b->left, &rvtype);
3504 right = interp_exec(c, b->right, &rtype);
3505 mpq_sub(rv.num, rv.num, right.num);
3508 rv = interp_exec(c, b->left, &rvtype);
3509 right = interp_exec(c, b->right, &rtype);
3510 mpq_mul(rv.num, rv.num, right.num);
3513 rv = interp_exec(c, b->left, &rvtype);
3514 right = interp_exec(c, b->right, &rtype);
3515 mpq_div(rv.num, rv.num, right.num);
3520 left = interp_exec(c, b->left, <ype);
3521 right = interp_exec(c, b->right, &rtype);
3522 mpz_init(l); mpz_init(r); mpz_init(rem);
3523 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
3524 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
3525 mpz_tdiv_r(rem, l, r);
3526 val_init(Tnum, &rv);
3527 mpq_set_z(rv.num, rem);
3528 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
3533 rv = interp_exec(c, b->right, &rvtype);
3534 mpq_neg(rv.num, rv.num);
3537 rv = interp_exec(c, b->right, &rvtype);
3538 mpq_abs(rv.num, rv.num);
3541 rv = interp_exec(c, b->right, &rvtype);
3544 left = interp_exec(c, b->left, <ype);
3545 right = interp_exec(c, b->right, &rtype);
3547 rv.str = text_join(left.str, right.str);
3550 right = interp_exec(c, b->right, &rvtype);
3554 struct text tx = right.str;
3557 if (tx.txt[0] == '-') {
3558 neg = 1; // UNTESTED
3559 tx.txt++; // UNTESTED
3560 tx.len--; // UNTESTED
3562 if (number_parse(rv.num, tail, tx) == 0)
3563 mpq_init(rv.num); // UNTESTED
3565 mpq_neg(rv.num, rv.num); // UNTESTED
3567 printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
3571 ###### value functions
3573 static struct text text_join(struct text a, struct text b)
3576 rv.len = a.len + b.len;
3577 rv.txt = malloc(rv.len);
3578 memcpy(rv.txt, a.txt, a.len);
3579 memcpy(rv.txt+a.len, b.txt, b.len);
3585 A function call can appear either as an expression or as a statement.
3586 As functions cannot yet return values, only the statement version will work.
3587 We use a new 'Funcall' binode type to link the function with a list of
3588 arguments, form with the 'List' nodes.
3593 ###### expression grammar
3594 | Variable ( ExpressionList ) ${ {
3595 struct binode *b = new(binode);
3598 b->right = reorder_bilist($<EL);
3602 struct binode *b = new(binode);
3609 ###### SimpleStatement Grammar
3611 | Variable ( ExpressionList ) ${ {
3612 struct binode *b = new(binode);
3615 b->right = reorder_bilist($<EL);
3619 ###### print binode cases
3622 do_indent(indent, "");
3623 print_exec(b->left, -1, bracket);
3625 for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
3628 print_exec(b->left, -1, bracket);
3638 ###### propagate binode cases
3641 /* Every arg must match formal parameter, and result
3642 * is return type of function
3644 struct binode *args = cast(binode, b->right);
3645 struct var *v = cast(var, b->left);
3647 if (!v->var->type || v->var->type->check_args == NULL) {
3648 type_err(c, "error: attempt to call a non-function.",
3649 prog, NULL, 0, NULL);
3652 v->var->type->check_args(c, ok, v->var->type, args);
3653 return v->var->type->function.return_type;
3656 ###### interp binode cases
3659 struct var *v = cast(var, b->left);
3660 struct type *t = v->var->type;
3661 void *oldlocal = c->local;
3662 int old_size = c->local_size;
3663 void *local = calloc(1, t->function.local_size);
3664 struct value *fbody = var_value(c, v->var);
3665 struct binode *arg = cast(binode, b->right);
3666 struct binode *param = t->function.params;
3669 struct var *pv = cast(var, param->left);
3670 struct type *vtype = NULL;
3671 struct value val = interp_exec(c, arg->left, &vtype);
3673 c->local = local; c->local_size = t->function.local_size;
3674 lval = var_value(c, pv->var);
3675 c->local = oldlocal; c->local_size = old_size;
3676 memcpy(lval, &val, vtype->size);
3677 param = cast(binode, param->right);
3678 arg = cast(binode, arg->right);
3680 c->local = local; c->local_size = t->function.local_size;
3681 rv = interp_exec(c, fbody->function, &rvtype);
3682 c->local = oldlocal; c->local_size = old_size;
3687 ### Blocks, Statements, and Statement lists.
3689 Now that we have expressions out of the way we need to turn to
3690 statements. There are simple statements and more complex statements.
3691 Simple statements do not contain (syntactic) newlines, complex statements do.
3693 Statements often come in sequences and we have corresponding simple
3694 statement lists and complex statement lists.
3695 The former comprise only simple statements separated by semicolons.
3696 The later comprise complex statements and simple statement lists. They are
3697 separated by newlines. Thus the semicolon is only used to separate
3698 simple statements on the one line. This may be overly restrictive,
3699 but I'm not sure I ever want a complex statement to share a line with
3702 Note that a simple statement list can still use multiple lines if
3703 subsequent lines are indented, so
3705 ###### Example: wrapped simple statement list
3710 is a single simple statement list. This might allow room for
3711 confusion, so I'm not set on it yet.
3713 A simple statement list needs no extra syntax. A complex statement
3714 list has two syntactic forms. It can be enclosed in braces (much like
3715 C blocks), or it can be introduced by an indent and continue until an
3716 unindented newline (much like Python blocks). With this extra syntax
3717 it is referred to as a block.
3719 Note that a block does not have to include any newlines if it only
3720 contains simple statements. So both of:
3722 if condition: a=b; d=f
3724 if condition { a=b; print f }
3728 In either case the list is constructed from a `binode` list with
3729 `Block` as the operator. When parsing the list it is most convenient
3730 to append to the end, so a list is a list and a statement. When using
3731 the list it is more convenient to consider a list to be a statement
3732 and a list. So we need a function to re-order a list.
3733 `reorder_bilist` serves this purpose.
3735 The only stand-alone statement we introduce at this stage is `pass`
3736 which does nothing and is represented as a `NULL` pointer in a `Block`
3737 list. Other stand-alone statements will follow once the infrastructure
3748 Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3749 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3750 | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3751 | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3752 | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3754 OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3755 | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3756 | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3757 | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3758 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3760 UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3761 | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3762 | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
3764 ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
3765 | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
3766 | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
3767 | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
3768 | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
3770 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
3772 ComplexStatements -> ComplexStatements ComplexStatement ${
3782 | ComplexStatement ${
3794 ComplexStatement -> SimpleStatements Newlines ${
3795 $0 = reorder_bilist($<SS);
3797 | SimpleStatements ; Newlines ${
3798 $0 = reorder_bilist($<SS);
3800 ## ComplexStatement Grammar
3803 SimpleStatements -> SimpleStatements ; SimpleStatement ${
3809 | SimpleStatement ${
3817 SimpleStatement -> pass ${ $0 = NULL; }$
3818 | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
3819 ## SimpleStatement Grammar
3821 ###### print binode cases
3825 if (b->left == NULL) // UNTESTED
3826 printf("pass"); // UNTESTED
3828 print_exec(b->left, indent, bracket); // UNTESTED
3829 if (b->right) { // UNTESTED
3830 printf("; "); // UNTESTED
3831 print_exec(b->right, indent, bracket); // UNTESTED
3834 // block, one per line
3835 if (b->left == NULL)
3836 do_indent(indent, "pass\n");
3838 print_exec(b->left, indent, bracket);
3840 print_exec(b->right, indent, bracket);
3844 ###### propagate binode cases
3847 /* If any statement returns something other than Tnone
3848 * or Tbool then all such must return same type.
3849 * As each statement may be Tnone or something else,
3850 * we must always pass NULL (unknown) down, otherwise an incorrect
3851 * error might occur. We never return Tnone unless it is
3856 for (e = b; e; e = cast(binode, e->right)) {
3857 t = propagate_types(e->left, c, ok, NULL, rules);
3858 if ((rules & Rboolok) && (t == Tbool || t == Tnone))
3860 if (t == Tnone && e->right)
3861 /* Only the final statement *must* return a value
3869 type_err(c, "error: expected %1%r, found %2",
3870 e->left, type, rules, t);
3876 ###### interp binode cases
3878 while (rvtype == Tnone &&
3881 rv = interp_exec(c, b->left, &rvtype);
3882 b = cast(binode, b->right);
3886 ### The Print statement
3888 `print` is a simple statement that takes a comma-separated list of
3889 expressions and prints the values separated by spaces and terminated
3890 by a newline. No control of formatting is possible.
3892 `print` uses `ExpressionList` to collect the expressions and stores them
3893 on the left side of a `Print` binode unlessthere is a trailing comma
3894 when the list is stored on the `right` side and no trailing newline is
3900 ##### expr precedence
3903 ###### SimpleStatement Grammar
3905 | print ExpressionList ${
3909 $0->left = reorder_bilist($<EL);
3911 | print ExpressionList , ${ {
3914 $0->right = reorder_bilist($<EL);
3924 ###### print binode cases
3927 do_indent(indent, "print");
3929 print_exec(b->right, -1, bracket);
3932 print_exec(b->left, -1, bracket);
3937 ###### propagate binode cases
3940 /* don't care but all must be consistent */
3942 b = cast(binode, b->left);
3944 b = cast(binode, b->right);
3946 propagate_types(b->left, c, ok, NULL, Rnolabel);
3947 b = cast(binode, b->right);
3951 ###### interp binode cases
3955 struct binode *b2 = cast(binode, b->left);
3957 b2 = cast(binode, b->right);
3958 for (; b2; b2 = cast(binode, b2->right)) {
3959 left = interp_exec(c, b2->left, <ype);
3960 print_value(ltype, &left);
3961 free_value(ltype, &left);
3965 if (b->right == NULL)
3971 ###### Assignment statement
3973 An assignment will assign a value to a variable, providing it hasn't
3974 been declared as a constant. The analysis phase ensures that the type
3975 will be correct so the interpreter just needs to perform the
3976 calculation. There is a form of assignment which declares a new
3977 variable as well as assigning a value. If a name is assigned before
3978 it is declared, and error will be raised as the name is created as
3979 `Tlabel` and it is illegal to assign to such names.
3985 ###### declare terminals
3988 ###### SimpleStatement Grammar
3989 | Variable = Expression ${
3995 | VariableDecl = Expression ${
4003 if ($1->var->where_set == NULL) {
4005 "Variable declared with no type or value: %v",
4016 ###### print binode cases
4019 do_indent(indent, "");
4020 print_exec(b->left, indent, bracket);
4022 print_exec(b->right, indent, bracket);
4029 struct variable *v = cast(var, b->left)->var;
4030 do_indent(indent, "");
4031 print_exec(b->left, indent, bracket);
4032 if (cast(var, b->left)->var->constant) {
4034 if (v->where_decl == v->where_set) {
4035 type_print(v->type, stdout);
4040 if (v->where_decl == v->where_set) {
4041 type_print(v->type, stdout);
4047 print_exec(b->right, indent, bracket);
4054 ###### propagate binode cases
4058 /* Both must match and not be labels,
4059 * Type must support 'dup',
4060 * For Assign, left must not be constant.
4063 t = propagate_types(b->left, c, ok, NULL,
4064 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
4069 if (propagate_types(b->right, c, ok, t, 0) != t)
4070 if (b->left->type == Xvar)
4071 type_err(c, "info: variable '%v' was set as %1 here.",
4072 cast(var, b->left)->var->where_set, t, rules, NULL);
4074 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
4076 propagate_types(b->left, c, ok, t,
4077 (b->op == Assign ? Rnoconstant : 0));
4079 if (t && t->dup == NULL)
4080 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
4085 ###### interp binode cases
4088 lleft = linterp_exec(c, b->left, <ype);
4090 dinterp_exec(c, b->right, lleft, ltype, 1);
4096 struct variable *v = cast(var, b->left)->var;
4099 val = var_value(c, v);
4100 if (v->type->prepare_type)
4101 v->type->prepare_type(c, v->type, 0);
4103 dinterp_exec(c, b->right, val, v->type, 0);
4105 val_init(v->type, val);
4109 ### The `use` statement
4111 The `use` statement is the last "simple" statement. It is needed when a
4112 statement block can return a value. This includes the body of a
4113 function which has a return type, and the "condition" code blocks in
4114 `if`, `while`, and `switch` statements.
4119 ###### expr precedence
4122 ###### SimpleStatement Grammar
4124 $0 = new_pos(binode, $1);
4127 if ($0->right->type == Xvar) {
4128 struct var *v = cast(var, $0->right);
4129 if (v->var->type == Tnone) {
4130 /* Convert this to a label */
4133 v->var->type = Tlabel;
4134 val = global_alloc(c, Tlabel, v->var, NULL);
4140 ###### print binode cases
4143 do_indent(indent, "use ");
4144 print_exec(b->right, -1, bracket);
4149 ###### propagate binode cases
4152 /* result matches value */
4153 return propagate_types(b->right, c, ok, type, 0);
4155 ###### interp binode cases
4158 rv = interp_exec(c, b->right, &rvtype);
4161 ### The Conditional Statement
4163 This is the biggy and currently the only complex statement. This
4164 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
4165 It is comprised of a number of parts, all of which are optional though
4166 set combinations apply. Each part is (usually) a key word (`then` is
4167 sometimes optional) followed by either an expression or a code block,
4168 except the `casepart` which is a "key word and an expression" followed
4169 by a code block. The code-block option is valid for all parts and,
4170 where an expression is also allowed, the code block can use the `use`
4171 statement to report a value. If the code block does not report a value
4172 the effect is similar to reporting `True`.
4174 The `else` and `case` parts, as well as `then` when combined with
4175 `if`, can contain a `use` statement which will apply to some
4176 containing conditional statement. `for` parts, `do` parts and `then`
4177 parts used with `for` can never contain a `use`, except in some
4178 subordinate conditional statement.
4180 If there is a `forpart`, it is executed first, only once.
4181 If there is a `dopart`, then it is executed repeatedly providing
4182 always that the `condpart` or `cond`, if present, does not return a non-True
4183 value. `condpart` can fail to return any value if it simply executes
4184 to completion. This is treated the same as returning `True`.
4186 If there is a `thenpart` it will be executed whenever the `condpart`
4187 or `cond` returns True (or does not return any value), but this will happen
4188 *after* `dopart` (when present).
4190 If `elsepart` is present it will be executed at most once when the
4191 condition returns `False` or some value that isn't `True` and isn't
4192 matched by any `casepart`. If there are any `casepart`s, they will be
4193 executed when the condition returns a matching value.
4195 The particular sorts of values allowed in case parts has not yet been
4196 determined in the language design, so nothing is prohibited.
4198 The various blocks in this complex statement potentially provide scope
4199 for variables as described earlier. Each such block must include the
4200 "OpenScope" nonterminal before parsing the block, and must call
4201 `var_block_close()` when closing the block.
4203 The code following "`if`", "`switch`" and "`for`" does not get its own
4204 scope, but is in a scope covering the whole statement, so names
4205 declared there cannot be redeclared elsewhere. Similarly the
4206 condition following "`while`" is in a scope the covers the body
4207 ("`do`" part) of the loop, and which does not allow conditional scope
4208 extension. Code following "`then`" (both looping and non-looping),
4209 "`else`" and "`case`" each get their own local scope.
4211 The type requirements on the code block in a `whilepart` are quite
4212 unusal. It is allowed to return a value of some identifiable type, in
4213 which case the loop aborts and an appropriate `casepart` is run, or it
4214 can return a Boolean, in which case the loop either continues to the
4215 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
4216 This is different both from the `ifpart` code block which is expected to
4217 return a Boolean, or the `switchpart` code block which is expected to
4218 return the same type as the casepart values. The correct analysis of
4219 the type of the `whilepart` code block is the reason for the
4220 `Rboolok` flag which is passed to `propagate_types()`.
4222 The `cond_statement` cannot fit into a `binode` so a new `exec` is
4223 defined. As there are two scopes which cover multiple parts - one for
4224 the whole statement and one for "while" and "do" - and as we will use
4225 the 'struct exec' to track scopes, we actually need two new types of
4226 exec. One is a `binode` for the looping part, the rest is the
4227 `cond_statement`. The `cond_statement` will use an auxilliary `struct
4228 casepart` to track a list of case parts.
4239 struct exec *action;
4240 struct casepart *next;
4242 struct cond_statement {
4244 struct exec *forpart, *condpart, *thenpart, *elsepart;
4245 struct binode *looppart;
4246 struct casepart *casepart;
4249 ###### ast functions
4251 static void free_casepart(struct casepart *cp)
4255 free_exec(cp->value);
4256 free_exec(cp->action);
4263 static void free_cond_statement(struct cond_statement *s)
4267 free_exec(s->forpart);
4268 free_exec(s->condpart);
4269 free_exec(s->looppart);
4270 free_exec(s->thenpart);
4271 free_exec(s->elsepart);
4272 free_casepart(s->casepart);
4276 ###### free exec cases
4277 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
4279 ###### ComplexStatement Grammar
4280 | CondStatement ${ $0 = $<1; }$
4282 ###### expr precedence
4283 $TERM for then while do
4290 // A CondStatement must end with EOL, as does CondSuffix and
4292 // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
4293 // may or may not end with EOL
4294 // WhilePart and IfPart include an appropriate Suffix
4296 // ForPart, SwitchPart, and IfPart open scopes, o we have to close
4297 // them. WhilePart opens and closes its own scope.
4298 CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
4301 $0->thenpart = $<TP;
4302 $0->looppart = $<WP;
4303 var_block_close(c, CloseSequential, $0);
4305 | ForPart OptNL WhilePart CondSuffix ${
4308 $0->looppart = $<WP;
4309 var_block_close(c, CloseSequential, $0);
4311 | WhilePart CondSuffix ${
4313 $0->looppart = $<WP;
4315 | SwitchPart OptNL CasePart CondSuffix ${
4317 $0->condpart = $<SP;
4318 $CP->next = $0->casepart;
4319 $0->casepart = $<CP;
4320 var_block_close(c, CloseSequential, $0);
4322 | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
4324 $0->condpart = $<SP;
4325 $CP->next = $0->casepart;
4326 $0->casepart = $<CP;
4327 var_block_close(c, CloseSequential, $0);
4329 | IfPart IfSuffix ${
4331 $0->condpart = $IP.condpart; $IP.condpart = NULL;
4332 $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
4333 // This is where we close an "if" statement
4334 var_block_close(c, CloseSequential, $0);
4337 CondSuffix -> IfSuffix ${
4340 | Newlines CasePart CondSuffix ${
4342 $CP->next = $0->casepart;
4343 $0->casepart = $<CP;
4345 | CasePart CondSuffix ${
4347 $CP->next = $0->casepart;
4348 $0->casepart = $<CP;
4351 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
4352 | Newlines ElsePart ${ $0 = $<EP; }$
4353 | ElsePart ${$0 = $<EP; }$
4355 ElsePart -> else OpenBlock Newlines ${
4356 $0 = new(cond_statement);
4357 $0->elsepart = $<OB;
4358 var_block_close(c, CloseElse, $0->elsepart);
4360 | else OpenScope CondStatement ${
4361 $0 = new(cond_statement);
4362 $0->elsepart = $<CS;
4363 var_block_close(c, CloseElse, $0->elsepart);
4367 CasePart -> case Expression OpenScope ColonBlock ${
4368 $0 = calloc(1,sizeof(struct casepart));
4371 var_block_close(c, CloseParallel, $0->action);
4375 // These scopes are closed in CondStatement
4376 ForPart -> for OpenBlock ${
4380 ThenPart -> then OpenBlock ${
4382 var_block_close(c, CloseSequential, $0);
4386 // This scope is closed in CondStatement
4387 WhilePart -> while UseBlock OptNL do OpenBlock ${
4392 var_block_close(c, CloseSequential, $0->right);
4393 var_block_close(c, CloseSequential, $0);
4395 | while OpenScope Expression OpenScope ColonBlock ${
4400 var_block_close(c, CloseSequential, $0->right);
4401 var_block_close(c, CloseSequential, $0);
4405 IfPart -> if UseBlock OptNL then OpenBlock ${
4408 var_block_close(c, CloseParallel, $0.thenpart);
4410 | if OpenScope Expression OpenScope ColonBlock ${
4413 var_block_close(c, CloseParallel, $0.thenpart);
4415 | if OpenScope Expression OpenScope OptNL then Block ${
4418 var_block_close(c, CloseParallel, $0.thenpart);
4422 // This scope is closed in CondStatement
4423 SwitchPart -> switch OpenScope Expression ${
4426 | switch UseBlock ${
4430 ###### print binode cases
4432 if (b->left && b->left->type == Xbinode &&
4433 cast(binode, b->left)->op == Block) {
4435 do_indent(indent, "while {\n");
4437 do_indent(indent, "while\n");
4438 print_exec(b->left, indent+1, bracket);
4440 do_indent(indent, "} do {\n");
4442 do_indent(indent, "do\n");
4443 print_exec(b->right, indent+1, bracket);
4445 do_indent(indent, "}\n");
4447 do_indent(indent, "while ");
4448 print_exec(b->left, 0, bracket);
4453 print_exec(b->right, indent+1, bracket);
4455 do_indent(indent, "}\n");
4459 ###### print exec cases
4461 case Xcond_statement:
4463 struct cond_statement *cs = cast(cond_statement, e);
4464 struct casepart *cp;
4466 do_indent(indent, "for");
4467 if (bracket) printf(" {\n"); else printf("\n");
4468 print_exec(cs->forpart, indent+1, bracket);
4471 do_indent(indent, "} then {\n");
4473 do_indent(indent, "then\n");
4474 print_exec(cs->thenpart, indent+1, bracket);
4476 if (bracket) do_indent(indent, "}\n");
4479 print_exec(cs->looppart, indent, bracket);
4483 do_indent(indent, "switch");
4485 do_indent(indent, "if");
4486 if (cs->condpart && cs->condpart->type == Xbinode &&
4487 cast(binode, cs->condpart)->op == Block) {
4492 print_exec(cs->condpart, indent+1, bracket);
4494 do_indent(indent, "}\n");
4496 do_indent(indent, "then\n");
4497 print_exec(cs->thenpart, indent+1, bracket);
4501 print_exec(cs->condpart, 0, bracket);
4507 print_exec(cs->thenpart, indent+1, bracket);
4509 do_indent(indent, "}\n");
4514 for (cp = cs->casepart; cp; cp = cp->next) {
4515 do_indent(indent, "case ");
4516 print_exec(cp->value, -1, 0);
4521 print_exec(cp->action, indent+1, bracket);
4523 do_indent(indent, "}\n");
4526 do_indent(indent, "else");
4531 print_exec(cs->elsepart, indent+1, bracket);
4533 do_indent(indent, "}\n");
4538 ###### propagate binode cases
4540 t = propagate_types(b->right, c, ok, Tnone, 0);
4541 if (!type_compat(Tnone, t, 0))
4542 *ok = 0; // UNTESTED
4543 return propagate_types(b->left, c, ok, type, rules);
4545 ###### propagate exec cases
4546 case Xcond_statement:
4548 // forpart and looppart->right must return Tnone
4549 // thenpart must return Tnone if there is a loopart,
4550 // otherwise it is like elsepart.
4552 // be bool if there is no casepart
4553 // match casepart->values if there is a switchpart
4554 // either be bool or match casepart->value if there
4556 // elsepart and casepart->action must match the return type
4557 // expected of this statement.
4558 struct cond_statement *cs = cast(cond_statement, prog);
4559 struct casepart *cp;
4561 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
4562 if (!type_compat(Tnone, t, 0))
4563 *ok = 0; // UNTESTED
4566 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
4567 if (!type_compat(Tnone, t, 0))
4568 *ok = 0; // UNTESTED
4570 if (cs->casepart == NULL) {
4571 propagate_types(cs->condpart, c, ok, Tbool, 0);
4572 propagate_types(cs->looppart, c, ok, Tbool, 0);
4574 /* Condpart must match case values, with bool permitted */
4576 for (cp = cs->casepart;
4577 cp && !t; cp = cp->next)
4578 t = propagate_types(cp->value, c, ok, NULL, 0);
4579 if (!t && cs->condpart)
4580 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok); // UNTESTED
4581 if (!t && cs->looppart)
4582 t = propagate_types(cs->looppart, c, ok, NULL, Rboolok); // UNTESTED
4583 // Now we have a type (I hope) push it down
4585 for (cp = cs->casepart; cp; cp = cp->next)
4586 propagate_types(cp->value, c, ok, t, 0);
4587 propagate_types(cs->condpart, c, ok, t, Rboolok);
4588 propagate_types(cs->looppart, c, ok, t, Rboolok);
4591 // (if)then, else, and case parts must return expected type.
4592 if (!cs->looppart && !type)
4593 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
4595 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
4596 for (cp = cs->casepart;
4598 cp = cp->next) // UNTESTED
4599 type = propagate_types(cp->action, c, ok, NULL, rules); // UNTESTED
4602 propagate_types(cs->thenpart, c, ok, type, rules);
4603 propagate_types(cs->elsepart, c, ok, type, rules);
4604 for (cp = cs->casepart; cp ; cp = cp->next)
4605 propagate_types(cp->action, c, ok, type, rules);
4611 ###### interp binode cases
4613 // This just performs one iterration of the loop
4614 rv = interp_exec(c, b->left, &rvtype);
4615 if (rvtype == Tnone ||
4616 (rvtype == Tbool && rv.bool != 0))
4617 // rvtype is Tnone or Tbool, doesn't need to be freed
4618 interp_exec(c, b->right, NULL);
4621 ###### interp exec cases
4622 case Xcond_statement:
4624 struct value v, cnd;
4625 struct type *vtype, *cndtype;
4626 struct casepart *cp;
4627 struct cond_statement *cs = cast(cond_statement, e);
4630 interp_exec(c, cs->forpart, NULL);
4632 while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
4633 cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
4634 interp_exec(c, cs->thenpart, NULL);
4636 cnd = interp_exec(c, cs->condpart, &cndtype);
4637 if ((cndtype == Tnone ||
4638 (cndtype == Tbool && cnd.bool != 0))) {
4639 // cnd is Tnone or Tbool, doesn't need to be freed
4640 rv = interp_exec(c, cs->thenpart, &rvtype);
4641 // skip else (and cases)
4645 for (cp = cs->casepart; cp; cp = cp->next) {
4646 v = interp_exec(c, cp->value, &vtype);
4647 if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
4648 free_value(vtype, &v);
4649 free_value(cndtype, &cnd);
4650 rv = interp_exec(c, cp->action, &rvtype);
4653 free_value(vtype, &v);
4655 free_value(cndtype, &cnd);
4657 rv = interp_exec(c, cs->elsepart, &rvtype);
4664 ### Top level structure
4666 All the language elements so far can be used in various places. Now
4667 it is time to clarify what those places are.
4669 At the top level of a file there will be a number of declarations.
4670 Many of the things that can be declared haven't been described yet,
4671 such as functions, procedures, imports, and probably more.
4672 For now there are two sorts of things that can appear at the top
4673 level. They are predefined constants, `struct` types, and the `main`
4674 function. While the syntax will allow the `main` function to appear
4675 multiple times, that will trigger an error if it is actually attempted.
4677 The various declarations do not return anything. They store the
4678 various declarations in the parse context.
4680 ###### Parser: grammar
4683 Ocean -> OptNL DeclarationList
4685 ## declare terminals
4692 DeclarationList -> Declaration
4693 | DeclarationList Declaration
4695 Declaration -> ERROR Newlines ${
4696 tok_err(c, // UNTESTED
4697 "error: unhandled parse error", &$1);
4703 ## top level grammar
4707 ### The `const` section
4709 As well as being defined in with the code that uses them, constants
4710 can be declared at the top level. These have full-file scope, so they
4711 are always `InScope`. The value of a top level constant can be given
4712 as an expression, and this is evaluated immediately rather than in the
4713 later interpretation stage. Once we add functions to the language, we
4714 will need rules concern which, if any, can be used to define a top
4717 Constants are defined in a section that starts with the reserved word
4718 `const` and then has a block with a list of assignment statements.
4719 For syntactic consistency, these must use the double-colon syntax to
4720 make it clear that they are constants. Type can also be given: if
4721 not, the type will be determined during analysis, as with other
4724 As the types constants are inserted at the head of a list, printing
4725 them in the same order that they were read is not straight forward.
4726 We take a quadratic approach here and count the number of constants
4727 (variables of depth 0), then count down from there, each time
4728 searching through for the Nth constant for decreasing N.
4730 ###### top level grammar
4734 DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
4735 | const { SimpleConstList } Newlines
4736 | const IN OptNL ConstList OUT Newlines
4737 | const SimpleConstList Newlines
4739 ConstList -> ConstList SimpleConstLine
4741 SimpleConstList -> SimpleConstList ; Const
4744 SimpleConstLine -> SimpleConstList Newlines
4745 | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
4748 CType -> Type ${ $0 = $<1; }$
4751 Const -> IDENTIFIER :: CType = Expression ${ {
4755 v = var_decl(c, $1.txt);
4757 struct var *var = new_pos(var, $1);
4758 v->where_decl = var;
4764 struct variable *vorig = var_ref(c, $1.txt);
4765 tok_err(c, "error: name already declared", &$1);
4766 type_err(c, "info: this is where '%v' was first declared",
4767 vorig->where_decl, NULL, 0, NULL);
4771 propagate_types($5, c, &ok, $3, 0);
4776 struct value res = interp_exec(c, $5, &v->type);
4777 global_alloc(c, v->type, v, &res);
4781 ###### print const decls
4786 while (target != 0) {
4788 for (v = context.in_scope; v; v=v->in_scope)
4789 if (v->depth == 0 && v->constant) {
4800 struct value *val = var_value(&context, v);
4801 printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
4802 type_print(v->type, stdout);
4804 if (v->type == Tstr)
4806 print_value(v->type, val);
4807 if (v->type == Tstr)
4815 ### Function declarations
4817 The code in an Ocean program is all stored in function declarations.
4818 One of the functions must be named `main` and it must accept an array of
4819 strings as a parameter - the command line arguments.
4821 As this is the top level, several things are handled a bit differently.
4822 The function is not interpreted by `interp_exec` as that isn't passed
4823 the argument list which the program requires. Similarly type analysis
4824 is a bit more interesting at this level.
4826 ###### ast functions
4828 static struct variable *declare_function(struct parse_context *c,
4829 struct variable *name,
4830 struct binode *args,
4834 struct text funcname = {" func", 5};
4836 struct value fn = {.function = code};
4837 name->type = add_type(c, funcname, &function_prototype);
4838 name->type->function.params = reorder_bilist(args);
4839 name->type->function.return_type = ret;
4840 global_alloc(c, name->type, name, &fn);
4841 var_block_close(c, CloseFunction, code);
4842 name->type->function.scope = c->out_scope;
4847 var_block_close(c, CloseFunction, NULL);
4849 c->out_scope = NULL;
4853 ###### declare terminals
4856 ###### top level grammar
4859 DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
4860 $0 = declare_function(c, $<FN, $<Ar, Tnone, $<Bl);
4862 | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
4863 $0 = declare_function(c, $<FN, $<Ar, Tnone, $<Bl);
4865 | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
4866 $0 = declare_function(c, $<FN, NULL, Tnone, $<Bl);
4868 | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
4869 $0 = declare_function(c, $<FN, $<Ar, $<Ty, $<Bl);
4871 | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
4872 $0 = declare_function(c, $<FN, $<Ar, $<Ty, $<Bl);
4874 | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
4875 $0 = declare_function(c, $<FN, NULL, $<Ty, $<Bl);
4878 ###### print func decls
4883 while (target != 0) {
4885 for (v = context.in_scope; v; v=v->in_scope)
4886 if (v->depth == 0 && v->type && v->type->check_args) {
4895 struct value *val = var_value(&context, v);
4896 printf("func %.*s", v->name->name.len, v->name->name.txt);
4897 v->type->print_type_decl(v->type, stdout);
4899 print_exec(val->function, 0, brackets);
4901 print_value(v->type, val);
4902 printf("/* frame size %d */\n", v->type->function.local_size);
4908 ###### core functions
4910 static int analyse_funcs(struct parse_context *c)
4914 for (v = c->in_scope; v; v = v->in_scope) {
4917 if (v->depth != 0 || !v->type || !v->type->check_args)
4919 val = var_value(c, v);
4922 propagate_types(val->function, c, &ok,
4923 v->type->function.return_type, 0);
4926 /* Make sure everything is still consistent */
4927 propagate_types(val->function, c, &ok,
4928 v->type->function.return_type, 0);
4931 if (!v->type->function.return_type->dup) {
4932 type_err(c, "error: function cannot return value of type %1",
4933 v->where_decl, v->type->function.return_type, 0, NULL);
4936 scope_finalize(c, v->type);
4941 static int analyse_main(struct type *type, struct parse_context *c)
4943 struct binode *bp = type->function.params;
4947 struct type *argv_type;
4948 struct text argv_type_name = { " argv", 5 };
4950 argv_type = add_type(c, argv_type_name, &array_prototype);
4951 argv_type->array.member = Tstr;
4952 argv_type->array.unspec = 1;
4954 for (b = bp; b; b = cast(binode, b->right)) {
4958 propagate_types(b->left, c, &ok, argv_type, 0);
4960 default: /* invalid */ // NOTEST
4961 propagate_types(b->left, c, &ok, Tnone, 0); // NOTEST
4967 return !c->parse_error;
4970 static void interp_main(struct parse_context *c, int argc, char **argv)
4972 struct value *progp = NULL;
4973 struct text main_name = { "main", 4 };
4974 struct variable *mainv;
4980 mainv = var_ref(c, main_name);
4982 progp = var_value(c, mainv);
4983 if (!progp || !progp->function) {
4984 fprintf(stderr, "oceani: no main function found.\n");
4988 if (!analyse_main(mainv->type, c)) {
4989 fprintf(stderr, "oceani: main has wrong type.\n");
4993 al = mainv->type->function.params;
4995 c->local_size = mainv->type->function.local_size;
4996 c->local = calloc(1, c->local_size);
4998 struct var *v = cast(var, al->left);
4999 struct value *vl = var_value(c, v->var);
5009 mpq_set_ui(argcq, argc, 1);
5010 memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
5011 t->prepare_type(c, t, 0);
5012 array_init(v->var->type, vl);
5013 for (i = 0; i < argc; i++) {
5014 struct value *vl2 = vl->array + i * v->var->type->array.member->size;
5016 arg.str.txt = argv[i];
5017 arg.str.len = strlen(argv[i]);
5018 free_value(Tstr, vl2);
5019 dup_value(Tstr, &arg, vl2);
5023 al = cast(binode, al->right);
5025 v = interp_exec(c, progp->function, &vtype);
5026 free_value(vtype, &v);
5031 ###### ast functions
5032 void free_variable(struct variable *v)
5036 ## And now to test it out.
5038 Having a language requires having a "hello world" program. I'll
5039 provide a little more than that: a program that prints "Hello world"
5040 finds the GCD of two numbers, prints the first few elements of
5041 Fibonacci, performs a binary search for a number, and a few other
5042 things which will likely grow as the languages grows.
5044 ###### File: oceani.mk
5047 @echo "===== DEMO ====="
5048 ./oceani --section "demo: hello" oceani.mdc 55 33
5054 four ::= 2 + 2 ; five ::= 10/2
5055 const pie ::= "I like Pie";
5056 cake ::= "The cake is"
5064 func main(argv:[argc::]string)
5065 print "Hello World, what lovely oceans you have!"
5066 print "Are there", five, "?"
5067 print pi, pie, "but", cake
5069 A := $argv[1]; B := $argv[2]
5071 /* When a variable is defined in both branches of an 'if',
5072 * and used afterwards, the variables are merged.
5078 print "Is", A, "bigger than", B,"? ", bigger
5079 /* If a variable is not used after the 'if', no
5080 * merge happens, so types can be different
5083 double:string = "yes"
5084 print A, "is more than twice", B, "?", double
5087 print "double", B, "is", double
5092 if a > 0 and then b > 0:
5098 print "GCD of", A, "and", B,"is", a
5100 print a, "is not positive, cannot calculate GCD"
5102 print b, "is not positive, cannot calculate GCD"
5107 print "Fibonacci:", f1,f2,
5108 then togo = togo - 1
5116 /* Binary search... */
5121 mid := (lo + hi) / 2
5134 print "Yay, I found", target
5136 print "Closest I found was", lo
5141 // "middle square" PRNG. Not particularly good, but one my
5142 // Dad taught me - the first one I ever heard of.
5143 for i:=1; then i = i + 1; while i < size:
5144 n := list[i-1] * list[i-1]
5145 list[i] = (n / 100) % 10 000
5147 print "Before sort:",
5148 for i:=0; then i = i + 1; while i < size:
5152 for i := 1; then i=i+1; while i < size:
5153 for j:=i-1; then j=j-1; while j >= 0:
5154 if list[j] > list[j+1]:
5158 print " After sort:",
5159 for i:=0; then i = i + 1; while i < size:
5163 if 1 == 2 then print "yes"; else print "no"
5167 bob.alive = (bob.name == "Hello")
5168 print "bob", "is" if bob.alive else "isn't", "alive"