1 # Ocean Interpreter - Stoney 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 second version of the interpreter exists to test out the
33 structured statement providing conditions and iteration, and simple
34 variable scoping. Clearly we need some minimal other functionality so
35 that values can be tested and instructions iterated over. All that
36 functionality is clearly not normative at this stage (not that
37 anything is **really** normative yet) and will change, so early test
38 code will certainly break in later versions.
40 The under-test parts of the language are:
42 - conditional/looping structured statements
43 - the `use` statement which is needed for that
44 - Variable binding using ":=" and "::=", and assignment using "=".
46 Elements which are present to make a usable language are:
48 - "blocks" of multiple statements.
49 - `pass`: a statement which does nothing.
50 - expressions: `+`, `-`, `*`, `/`, `%` can apply to numbers and `++` can
51 catenate strings. `and`, `or`, `not` manipulate Booleans, and
52 normal comparison operators can work on all three types.
53 - `print`: will print the values in a list of expressions.
54 - `program`: is given a list of identifiers to initialize from
59 Versions of the interpreter which obviously do not support a complete
60 language will be named after creeks and streams. This one is Stoney
63 Once we have something reasonably resembling a complete language, the
64 names of rivers will be used.
65 Early versions of the compiler will be named after seas. Major
66 releases of the compiler will be named after oceans. Hopefully I will
67 be finished once I get to the Pacific Ocean release.
71 As well as parsing and executing a program, the interpreter can print
72 out the program from the parsed internal structure. This is useful
73 for validating the parsing.
74 So the main requirements of the interpreter are:
76 - Parse the program, possibly with tracing,
77 - Analyse the parsed program to ensure consistency,
79 - Execute the program.
81 This is all performed by a single C program extracted with
84 There will be two formats for printing the program: a default and one
85 that uses bracketing. So a `--bracket` command line option is needed
86 for that. Normally the first code section found is used, however an
87 alternate section can be requested so that a file (such as this one)
88 can contain multiple programs This is effected with the `--section`
91 This code must be compiled with `-fplan9-extensions` so that anonymous
92 structures can be used.
94 ###### File: oceani.mk
96 myCFLAGS := -Wall -g -fplan9-extensions
97 CFLAGS := $(filter-out $(myCFLAGS),$(CFLAGS)) $(myCFLAGS)
98 myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
99 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
101 all :: $(LDLIBS) oceani
102 oceani.c oceani.h : oceani.mdc parsergen
103 ./parsergen -o oceani --LALR --tag Parser oceani.mdc
104 oceani.mk: oceani.mdc md2c
107 oceani: oceani.o $(LDLIBS)
108 $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
110 ###### Parser: header
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, \
135 #include <sys/mman.h>
154 static char Usage[] = "Usage: oceani --trace --print --noexec --brackets"
155 "--section=SectionName prog.ocn\n";
156 static const struct option long_options[] = {
157 {"trace", 0, NULL, 't'},
158 {"print", 0, NULL, 'p'},
159 {"noexec", 0, NULL, 'n'},
160 {"brackets", 0, NULL, 'b'},
161 {"section", 1, NULL, 's'},
164 const char *options = "tpnbs";
165 int main(int argc, char *argv[])
171 char *section = NULL;
172 struct parse_context context = {
174 .ignored = (1 << TK_line_comment)
175 | (1 << TK_block_comment),
176 .number_chars = ".,_+-",
181 int doprint=0, dotrace=0, doexec=1, brackets=0;
184 while ((opt = getopt_long(argc, argv, options, long_options, NULL))
187 case 't': dotrace=1; break;
188 case 'p': doprint=1; break;
189 case 'n': doexec=0; break;
190 case 'b': brackets=1; break;
191 case 's': section = optarg; break;
192 default: fprintf(stderr, Usage);
196 if (optind >= argc) {
197 fprintf(stderr, "oceani: no input file given\n");
200 fd = open(argv[optind], O_RDONLY);
202 fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
205 context.file_name = argv[optind];
206 len = lseek(fd, 0, 2);
207 file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
208 s = code_extract(file, file+len, NULL);
210 fprintf(stderr, "oceani: could not find any code in %s\n",
215 ## context initialization
219 for (ss = s; ss; ss = ss->next) {
220 struct text sec = ss->section;
221 if (sec.len == strlen(section) &&
222 strncmp(sec.txt, section, sec.len) == 0)
226 prog = parse_oceani(ss->code, &context.config,
227 dotrace ? stderr : NULL);
229 fprintf(stderr, "oceani: cannot find section %s\n",
234 prog = parse_oceani(s->code, &context.config,
235 dotrace ? stderr : NULL);
237 fprintf(stderr, "oceani: fatal parser error.\n");
238 context.parse_error = 1;
241 print_exec(*prog, 0, brackets);
242 if (prog && doexec && !context.parse_error) {
243 if (!analyse_prog(*prog, &context)) {
244 fprintf(stderr, "oceani: type error in program - not running.\n");
247 interp_prog(*prog, argv+optind+1);
254 struct section *t = s->next;
260 ## free context types
261 exit(context.parse_error ? 1 : 0);
266 The four requirements of parse, analyse, print, interpret apply to
267 each language element individually so that is how most of the code
270 Three of the four are fairly self explanatory. The one that requires
271 a little explanation is the analysis step.
273 The current language design does not require the types of variables to
274 be declared, but they must still have a single type. Different
275 operations impose different requirements on the variables, for example
276 addition requires both arguments to be numeric, and assignment
277 requires the variable on the left to have the same type as the
278 expression on the right.
280 Analysis involves propagating these type requirements around and
281 consequently setting the type of each variable. If any requirements
282 are violated (e.g. a string is compared with a number) or if a
283 variable needs to have two different types, then an error is raised
284 and the program will not run.
286 If the same variable is declared in both branchs of an 'if/else', or
287 in all cases of a 'switch' then the multiple instances may be merged
288 into just one variable if the variable is references after the
289 conditional statement. When this happens, the types must naturally be
290 consistent across all the branches. When the variable is not used
291 outside the if, the variables in the different branches are distinct
292 and can be of different types.
294 Determining the types of all variables early is important for
295 processing command line arguments. These can be assigned to any type
296 of variable, but we must first know the correct type so any required
297 conversion can happen. If a variable is associated with a command
298 line argument but no type can be interpreted (e.g. the variable is
299 only ever used in a `print` statement), then the type is set to
302 Undeclared names may only appear in "use" statements and "case" expressions.
303 These names are given a type of "label" and a unique value.
304 This allows them to fill the role of a name in an enumerated type, which
305 is useful for testing the `switch` statement.
307 As we will see, the condition part of a `while` statement can return
308 either a Boolean or some other type. This requires that the expect
309 type that gets passed around comprises a type (`enum vtype`) and a
310 flag to indicate that `Vbool` is also permitted.
312 As there are, as yet, no distinct types that are compatible, there
313 isn't much subtlety in the analysis. When we have distinct number
314 types, this will become more interesting.
318 When analysis discovers an inconsistency it needs to report an error;
319 just refusing to run the code ensures that the error doesn't cascade,
320 but by itself it isn't very useful. A clear understand of the sort of
321 error message that are useful will help guide the process of analysis.
323 At a simplistic level, the only sort of error that type analysis can
324 report is that the type of some construct doesn't match a contextual
325 requirement. For example, in `4 + "hello"` the addition provides a
326 contextual requirement for numbers, but `"hello"` is not a number. In
327 this particular example no further information is needed as the types
328 are obvious from local information. When a variable is involved that
329 isn't the case. It may be helpful to explain why the variable has a
330 particular type, by indicating the location where the type was set,
331 whether by declaration or usage.
333 Using a recursive-descent analysis we can easily detect a problem at
334 multiple locations. In "`hello:= "there"; 4 + hello`" the addition
335 will detect that one argument is not a number and the usage of `hello`
336 will detect that a number was wanted, but not provided. In this
337 (early) version of the language, we will generate error reports at
338 multiple locations, so the use of `hello` will report an error and
339 explain were the value was set, and the addition will report an error
340 and say why numbers are needed. To be able to report locations for
341 errors, each language element will need to record a file location
342 (line and column) and each variable will need to record the language
343 element where its type was set. For now we will assume that each line
344 of an error message indicates one location in the file, and up to 2
345 types. So we provide a `printf`-like function which takes a format, a
346 language (a `struct exec` which has not yet been introduced), and 2
347 types. "`%1`" reports the first type, "`%2`" reports the second. We
348 will need a function to print the location, once we know how that is
349 stored. As will be explained later, there are sometimes extra rules for
350 type matching and they might affect error messages, we need to pass those
353 As well as type errors, we sometimes need to report problems with
354 tokens, which might be unexpected or might name a type that has not
355 been defined. For these we have `tok_err()` which reports an error
356 with a given token. Each of the error functions sets the flag in the
357 context so indicate that parsing failed.
361 static void fput_loc(struct exec *loc, FILE *f);
363 ###### core functions
365 static void type_err(struct parse_context *c,
366 char *fmt, struct exec *loc,
367 struct type *t1, int rules, struct type *t2)
369 fprintf(stderr, "%s:", c->file_name);
370 fput_loc(loc, stderr);
371 for (; *fmt ; fmt++) {
378 case '%': fputc(*fmt, stderr); break;
379 default: fputc('?', stderr); break;
381 type_print(t1, stderr);
384 type_print(t2, stderr);
393 static void tok_err(struct parse_context *c, char *fmt, struct token *t)
395 fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
396 t->txt.len, t->txt.txt);
402 One last introductory step before detailing the language elements and
403 providing their four requirements is to establish the data structures
404 to store these elements.
406 There are two key objects that we need to work with: executable
407 elements which comprise the program, and values which the program
408 works with. Between these are the variables in their various scopes
409 which hold the values, and types which classify the values stored and
410 manipulatd by executables.
414 Values come in a wide range of types, with more likely to be added.
415 Each type needs to be able to parse and print its own values (for
416 convenience at least) as well as to compare two values, at least for
417 equality and possibly for order. For now, values might need to be
418 duplicated and freed, though eventually such manipulations will be
419 better integrated into the language.
421 Rather than requiring every numeric type to support all numeric
422 operations (add, multiple, etc), we allow types to be able to present
423 as one of a few standard types: integer, float, and fraction. The
424 existance of these conversion functions enable types to determine if
425 they are compatible with other types.
427 Named type are stored in a simple linked list. Objects of each type are "values"
428 which are often passed around by value.
435 ## value union fields
442 struct value (*init)(struct type *type);
443 struct value (*prepare)(struct type *type);
444 struct value (*parse)(struct type *type, char *str);
445 void (*print)(struct value val);
446 void (*print_type)(struct type *type, FILE *f);
447 int (*cmp_order)(struct value v1, struct value v2);
448 int (*cmp_eq)(struct value v1, struct value v2);
449 struct value (*dup)(struct value val);
450 void (*free)(struct value val);
451 int (*compat)(struct type *this, struct type *other);
452 long long (*to_int)(struct value *v);
453 double (*to_float)(struct value *v);
454 int (*to_mpq)(mpq_t *q, struct value *v);
462 struct type *typelist;
466 static struct type *find_type(struct parse_context *c, struct text s)
468 struct type *l = c->typelist;
471 text_cmp(l->name, s) != 0)
476 static struct type *add_type(struct parse_context *c, struct text s,
481 n = calloc(1, sizeof(*n));
484 n->next = c->typelist;
489 static void free_type(struct type *t)
491 /* The type is always a reference to something in the
492 * context, so we don't need to free anything.
496 static void free_value(struct value v)
502 static int type_compat(struct type *require, struct type *have, int rules)
504 if ((rules & Rboolok) && have == Tbool)
506 if ((rules & Rnolabel) && have == Tlabel)
508 if (!require || !have)
512 return require->compat(require, have);
514 return require == have;
517 static void type_print(struct type *type, FILE *f)
520 fputs("*unknown*type*", f);
521 else if (type->name.len)
522 fprintf(f, "%.*s", type->name.len, type->name.txt);
523 else if (type->print_type)
524 type->print_type(type, f);
526 fputs("*invalid*type*", f);
529 static struct value val_prepare(struct type *type)
534 return type->prepare(type);
539 static struct value val_init(struct type *type)
544 return type->init(type);
549 static struct value dup_value(struct value v)
552 return v.type->dup(v);
556 static int value_cmp(struct value left, struct value right)
558 if (left.type && left.type->cmp_order)
559 return left.type->cmp_order(left, right);
560 if (left.type && left.type->cmp_eq)
561 return left.type->cmp_eq(left, right);
565 static void print_value(struct value v)
567 if (v.type && v.type->print)
573 static struct value parse_value(struct type *type, char *arg)
577 if (type && type->parse)
578 return type->parse(type, arg);
585 static void free_value(struct value v);
586 static int type_compat(struct type *require, struct type *have, int rules);
587 static void type_print(struct type *type, FILE *f);
588 static struct value val_init(struct type *type);
589 static struct value dup_value(struct value v);
590 static int value_cmp(struct value left, struct value right);
591 static void print_value(struct value v);
592 static struct value parse_value(struct type *type, char *arg);
594 ###### free context types
596 while (context.typelist) {
597 struct type *t = context.typelist;
599 context.typelist = t->next;
605 Values of the base types can be numbers, which we represent as
606 multi-precision fractions, strings, Booleans and labels. When
607 analysing the program we also need to allow for places where no value
608 is meaningful (type `Tnone`) and where we don't know what type to
609 expect yet (type is `NULL`).
611 Values are never shared, they are always copied when used, and freed
612 when no longer needed.
614 When propagating type information around the program, we need to
615 determine if two types are compatible, where type `NULL` is compatible
616 with anything. There are two special cases with type compatibility,
617 both related to the Conditional Statement which will be described
618 later. In some cases a Boolean can be accepted as well as some other
619 primary type, and in others any type is acceptable except a label (`Vlabel`).
620 A separate function encode these cases will simplify some code later.
622 When assigning command line arguments to variables, we need to be able
623 to parse each type from a string.
631 myLDLIBS := libnumber.o libstring.o -lgmp
632 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
634 ###### type union fields
635 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
637 ###### value union fields
644 static void _free_value(struct value v)
646 switch (v.type->vtype) {
648 case Vstr: free(v.str.txt); break;
649 case Vnum: mpq_clear(v.num); break;
655 ###### value functions
657 static struct value _val_prepare(struct type *type)
662 switch(type->vtype) {
666 memset(&rv.num, 0, sizeof(rv.num));
682 static struct value _val_init(struct type *type)
687 switch(type->vtype) {
691 mpq_init(rv.num); break;
693 rv.str.txt = malloc(1);
706 static struct value _dup_value(struct value v)
710 switch (rv.type->vtype) {
721 mpq_set(rv.num, v.num);
724 rv.str.len = v.str.len;
725 rv.str.txt = malloc(rv.str.len);
726 memcpy(rv.str.txt, v.str.txt, v.str.len);
732 static int _value_cmp(struct value left, struct value right)
735 if (left.type != right.type)
736 return left.type - right.type;
737 switch (left.type->vtype) {
738 case Vlabel: cmp = left.label == right.label ? 0 : 1; break;
739 case Vnum: cmp = mpq_cmp(left.num, right.num); break;
740 case Vstr: cmp = text_cmp(left.str, right.str); break;
741 case Vbool: cmp = left.bool - right.bool; break;
747 static void _print_value(struct value v)
749 switch (v.type->vtype) {
751 printf("*no-value*"); break;
753 printf("*label-%p*", v.label); break;
755 printf("%.*s", v.str.len, v.str.txt); break;
757 printf("%s", v.bool ? "True":"False"); break;
762 mpf_set_q(fl, v.num);
763 gmp_printf("%Fg", fl);
770 static struct value _parse_value(struct type *type, char *arg)
778 switch(type->vtype) {
784 val.str.len = strlen(arg);
785 val.str.txt = malloc(val.str.len);
786 memcpy(val.str.txt, arg, val.str.len);
793 tx.txt = arg; tx.len = strlen(tx.txt);
794 if (number_parse(val.num, tail, tx) == 0)
797 mpq_neg(val.num, val.num);
799 printf("Unsupported suffix: %s\n", arg);
804 if (strcasecmp(arg, "true") == 0 ||
805 strcmp(arg, "1") == 0)
807 else if (strcasecmp(arg, "false") == 0 ||
808 strcmp(arg, "0") == 0)
811 printf("Bad bool: %s\n", arg);
819 static void _free_value(struct value v);
821 static struct type base_prototype = {
823 .prepare = _val_prepare,
824 .parse = _parse_value,
825 .print = _print_value,
826 .cmp_order = _value_cmp,
827 .cmp_eq = _value_cmp,
832 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
835 static struct type *add_base_type(struct parse_context *c, char *n, enum vtype vt)
837 struct text txt = { n, strlen(n) };
840 t = add_type(c, txt, &base_prototype);
845 ###### context initialization
847 Tbool = add_base_type(&context, "Boolean", Vbool);
848 Tstr = add_base_type(&context, "string", Vstr);
849 Tnum = add_base_type(&context, "number", Vnum);
850 Tnone = add_base_type(&context, "none", Vnone);
851 Tlabel = add_base_type(&context, "label", Vlabel);
855 Variables are scoped named values. We store the names in a linked
856 list of "bindings" sorted lexically, and use sequential search and
863 struct binding *next; // in lexical order
867 This linked list is stored in the parse context so that "reduce"
868 functions can find or add variables, and so the analysis phase can
869 ensure that every variable gets a type.
873 struct binding *varlist; // In lexical order
877 static struct binding *find_binding(struct parse_context *c, struct text s)
879 struct binding **l = &c->varlist;
884 (cmp = text_cmp((*l)->name, s)) < 0)
888 n = calloc(1, sizeof(*n));
895 Each name can be linked to multiple variables defined in different
896 scopes. Each scope starts where the name is declared and continues
897 until the end of the containing code block. Scopes of a given name
898 cannot nest, so a declaration while a name is in-scope is an error.
900 ###### binding fields
901 struct variable *var;
905 struct variable *previous;
907 struct binding *name;
908 struct exec *where_decl;// where name was declared
909 struct exec *where_set; // where type was set
913 While the naming seems strange, we include local constants in the
914 definition of variables. A name declared `var := value` can
915 subsequently be changed, but a name declared `var ::= value` cannot -
918 ###### variable fields
921 Scopes in parallel branches can be partially merged. More
922 specifically, if a given name is declared in both branches of an
923 if/else then its scope is a candidate for merging. Similarly if
924 every branch of an exhaustive switch (e.g. has an "else" clause)
925 declares a given name, then the scopes from the branches are
926 candidates for merging.
928 Note that names declared inside a loop (which is only parallel to
929 itself) are never visible after the loop. Similarly names defined in
930 scopes which are not parallel, such as those started by `for` and
931 `switch`, are never visible after the scope. Only variables defined in
932 both `then` and `else` (including the implicit then after an `if`, and
933 excluding `then` used with `for`) and in all `case`s and `else` of a
934 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
936 Labels, which are a bit like variables, follow different rules.
937 Labels are not explicitly declared, but if an undeclared name appears
938 in a context where a label is legal, that effectively declares the
939 name as a label. The declaration remains in force (or in scope) at
940 least to the end of the immediately containing block and conditionally
941 in any larger containing block which does not declare the name in some
942 other way. Importantly, the conditional scope extension happens even
943 if the label is only used in one parallel branch of a conditional --
944 when used in one branch it is treated as having been declared in all
947 Merge candidates are tentatively visible beyond the end of the
948 branching statement which creates them. If the name is used, the
949 merge is affirmed and they become a single variable visible at the
950 outer layer. If not - if it is redeclared first - the merge lapses.
952 To track scopes we have an extra stack, implemented as a linked list,
953 which roughly parallels the parse stack and which is used exclusively
954 for scoping. When a new scope is opened, a new frame is pushed and
955 the child-count of the parent frame is incremented. This child-count
956 is used to distinguish between the first of a set of parallel scopes,
957 in which declared variables must not be in scope, and subsequent
958 branches, whether they must already be conditionally scoped.
960 To push a new frame *before* any code in the frame is parsed, we need a
961 grammar reduction. This is most easily achieved with a grammar
962 element which derives the empty string, and creates the new scope when
963 it is recognized. This can be placed, for example, between a keyword
964 like "if" and the code following it.
968 struct scope *parent;
974 struct scope *scope_stack;
977 static void scope_pop(struct parse_context *c)
979 struct scope *s = c->scope_stack;
981 c->scope_stack = s->parent;
986 static void scope_push(struct parse_context *c)
988 struct scope *s = calloc(1, sizeof(*s));
990 c->scope_stack->child_count += 1;
991 s->parent = c->scope_stack;
999 OpenScope -> ${ scope_push(config2context(config)); }$
1002 Each variable records a scope depth and is in one of four states:
1004 - "in scope". This is the case between the declaration of the
1005 variable and the end of the containing block, and also between
1006 the usage with affirms a merge and the end of that block.
1008 The scope depth is not greater than the current parse context scope
1009 nest depth. When the block of that depth closes, the state will
1010 change. To achieve this, all "in scope" variables are linked
1011 together as a stack in nesting order.
1013 - "pending". The "in scope" block has closed, but other parallel
1014 scopes are still being processed. So far, every parallel block at
1015 the same level that has closed has declared the name.
1017 The scope depth is the depth of the last parallel block that
1018 enclosed the declaration, and that has closed.
1020 - "conditionally in scope". The "in scope" block and all parallel
1021 scopes have closed, and no further mention of the name has been
1022 seen. This state includes a secondary nest depth which records the
1023 outermost scope seen since the variable became conditionally in
1024 scope. If a use of the name is found, the variable becomes "in
1025 scope" and that secondary depth becomes the recorded scope depth.
1026 If the name is declared as a new variable, the old variable becomes
1027 "out of scope" and the recorded scope depth stays unchanged.
1029 - "out of scope". The variable is neither in scope nor conditionally
1030 in scope. It is permanently out of scope now and can be removed from
1031 the "in scope" stack.
1034 ###### variable fields
1035 int depth, min_depth;
1036 enum { OutScope, PendingScope, CondScope, InScope } scope;
1037 struct variable *in_scope;
1039 ###### parse context
1041 struct variable *in_scope;
1043 All variables with the same name are linked together using the
1044 'previous' link. Those variable that have
1045 been affirmatively merged all have a 'merged' pointer that points to
1046 one primary variable - the most recently declared instance. When
1047 merging variables, we need to also adjust the 'merged' pointer on any
1048 other variables that had previously been merged with the one that will
1049 no longer be primary.
1051 ###### variable fields
1052 struct variable *merged;
1054 ###### ast functions
1056 static void variable_merge(struct variable *primary, struct variable *secondary)
1060 if (primary->merged)
1062 primary = primary->merged;
1064 for (v = primary->previous; v; v=v->previous)
1065 if (v == secondary || v == secondary->merged ||
1066 v->merged == secondary ||
1067 (v->merged && v->merged == secondary->merged)) {
1068 v->scope = OutScope;
1069 v->merged = primary;
1073 ###### free context vars
1075 while (context.varlist) {
1076 struct binding *b = context.varlist;
1077 struct variable *v = b->var;
1078 context.varlist = b->next;
1081 struct variable *t = v;
1089 #### Manipulating Bindings
1091 When a name is conditionally visible, a new declaration discards the
1092 old binding - the condition lapses. Conversely a usage of the name
1093 affirms the visibility and extends it to the end of the containing
1094 block - i.e. the block that contains both the original declaration and
1095 the latest usage. This is determined from `min_depth`. When a
1096 conditionally visible variable gets affirmed like this, it is also
1097 merged with other conditionally visible variables with the same name.
1099 When we parse a variable declaration we either signal an error if the
1100 name is currently bound, or create a new variable at the current nest
1101 depth if the name is unbound or bound to a conditionally scoped or
1102 pending-scope variable. If the previous variable was conditionally
1103 scoped, it and its homonyms becomes out-of-scope.
1105 When we parse a variable reference (including non-declarative
1106 assignment) we signal an error if the name is not bound or is bound to
1107 a pending-scope variable; update the scope if the name is bound to a
1108 conditionally scoped variable; or just proceed normally if the named
1109 variable is in scope.
1111 When we exit a scope, any variables bound at this level are either
1112 marked out of scope or pending-scoped, depending on whether the
1113 scope was sequential or parallel.
1115 When exiting a parallel scope we check if there are any variables that
1116 were previously pending and are still visible. If there are, then
1117 there weren't redeclared in the most recent scope, so they cannot be
1118 merged and must become out-of-scope. If it is not the first of
1119 parallel scopes (based on `child_count`), we check that there was a
1120 previous binding that is still pending-scope. If there isn't, the new
1121 variable must now be out-of-scope.
1123 When exiting a sequential scope that immediately enclosed parallel
1124 scopes, we need to resolve any pending-scope variables. If there was
1125 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1126 we need to mark all pending-scope variable as out-of-scope. Otherwise
1127 all pending-scope variables become conditionally scoped.
1130 enum closetype { CloseSequential, CloseParallel, CloseElse };
1132 ###### ast functions
1134 static struct variable *var_decl(struct parse_context *c, struct text s)
1136 struct binding *b = find_binding(c, s);
1137 struct variable *v = b->var;
1139 switch (v ? v->scope : OutScope) {
1141 /* Caller will report the error */
1145 v && v->scope == CondScope;
1147 v->scope = OutScope;
1151 v = calloc(1, sizeof(*v));
1152 v->previous = b->var;
1155 v->min_depth = v->depth = c->scope_depth;
1157 v->in_scope = c->in_scope;
1159 v->val = val_prepare(NULL);
1163 static struct variable *var_ref(struct parse_context *c, struct text s)
1165 struct binding *b = find_binding(c, s);
1166 struct variable *v = b->var;
1167 struct variable *v2;
1169 switch (v ? v->scope : OutScope) {
1172 /* Signal an error - once that is possible */
1175 /* All CondScope variables of this name need to be merged
1176 * and become InScope
1178 v->depth = v->min_depth;
1180 for (v2 = v->previous;
1181 v2 && v2->scope == CondScope;
1183 variable_merge(v, v2);
1191 static void var_block_close(struct parse_context *c, enum closetype ct)
1193 /* close of all variables that are in_scope */
1194 struct variable *v, **vp, *v2;
1197 for (vp = &c->in_scope;
1198 v = *vp, v && v->depth > c->scope_depth && v->min_depth > c->scope_depth;
1202 case CloseParallel: /* handle PendingScope */
1206 if (c->scope_stack->child_count == 1)
1207 v->scope = PendingScope;
1208 else if (v->previous &&
1209 v->previous->scope == PendingScope)
1210 v->scope = PendingScope;
1211 else if (v->val.type == Tlabel)
1212 v->scope = PendingScope;
1213 else if (v->name->var == v)
1214 v->scope = OutScope;
1215 if (ct == CloseElse) {
1216 /* All Pending variables with this name
1217 * are now Conditional */
1219 v2 && v2->scope == PendingScope;
1221 v2->scope = CondScope;
1226 v2 && v2->scope == PendingScope;
1228 if (v2->val.type != Tlabel)
1229 v2->scope = OutScope;
1231 case OutScope: break;
1234 case CloseSequential:
1235 if (v->val.type == Tlabel)
1236 v->scope = PendingScope;
1239 v->scope = OutScope;
1242 /* There was no 'else', so we can only become
1243 * conditional if we know the cases were exhaustive,
1244 * and that doesn't mean anything yet.
1245 * So only labels become conditional..
1248 v2 && v2->scope == PendingScope;
1250 if (v2->val.type == Tlabel) {
1251 v2->scope = CondScope;
1252 v2->min_depth = c->scope_depth;
1254 v2->scope = OutScope;
1257 case OutScope: break;
1261 if (v->scope == OutScope)
1270 Executables can be lots of different things. In many cases an
1271 executable is just an operation combined with one or two other
1272 executables. This allows for expressions and lists etc. Other times
1273 an executable is something quite specific like a constant or variable
1274 name. So we define a `struct exec` to be a general executable with a
1275 type, and a `struct binode` which is a subclass of `exec`, forms a
1276 node in a binary tree, and holds an operation. There will be other
1277 subclasses, and to access these we need to be able to `cast` the
1278 `exec` into the various other types.
1281 #define cast(structname, pointer) ({ \
1282 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1283 if (__mptr && *__mptr != X##structname) abort(); \
1284 (struct structname *)( (char *)__mptr);})
1286 #define new(structname) ({ \
1287 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1288 __ptr->type = X##structname; \
1289 __ptr->line = -1; __ptr->column = -1; \
1292 #define new_pos(structname, token) ({ \
1293 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1294 __ptr->type = X##structname; \
1295 __ptr->line = token.line; __ptr->column = token.col; \
1304 enum exec_types type;
1312 struct exec *left, *right;
1315 ###### ast functions
1317 static int __fput_loc(struct exec *loc, FILE *f)
1319 if (loc->line >= 0) {
1320 fprintf(f, "%d:%d: ", loc->line, loc->column);
1323 if (loc->type == Xbinode)
1324 return __fput_loc(cast(binode,loc)->left, f) ||
1325 __fput_loc(cast(binode,loc)->right, f);
1328 static void fput_loc(struct exec *loc, FILE *f)
1330 if (!__fput_loc(loc, f))
1331 fprintf(f, "??:??: ");
1334 Each different type of `exec` node needs a number of functions
1335 defined, a bit like methods. We must be able to be able to free it,
1336 print it, analyse it and execute it. Once we have specific `exec`
1337 types we will need to parse them too. Let's take this a bit more
1342 The parser generator requires a `free_foo` function for each struct
1343 that stores attributes and they will be `exec`s and subtypes there-of.
1344 So we need `free_exec` which can handle all the subtypes, and we need
1347 ###### ast functions
1349 static void free_binode(struct binode *b)
1354 free_exec(b->right);
1358 ###### core functions
1359 static void free_exec(struct exec *e)
1368 ###### forward decls
1370 static void free_exec(struct exec *e);
1372 ###### free exec cases
1373 case Xbinode: free_binode(cast(binode, e)); break;
1377 Printing an `exec` requires that we know the current indent level for
1378 printing line-oriented components. As will become clear later, we
1379 also want to know what sort of bracketing to use.
1381 ###### ast functions
1383 static void do_indent(int i, char *str)
1390 ###### core functions
1391 static void print_binode(struct binode *b, int indent, int bracket)
1395 ## print binode cases
1399 static void print_exec(struct exec *e, int indent, int bracket)
1405 print_binode(cast(binode, e), indent, bracket); break;
1410 ###### forward decls
1412 static void print_exec(struct exec *e, int indent, int bracket);
1416 As discussed, analysis involves propagating type requirements around
1417 the program and looking for errors.
1419 So `propagate_types` is passed an expected type (being a `struct type`
1420 pointer together with some `val_rules` flags) that the `exec` is
1421 expected to return, and returns the type that it does return, either
1422 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1423 by reference. It is set to `0` when an error is found, and `2` when
1424 any change is made. If it remains unchanged at `1`, then no more
1425 propagation is needed.
1429 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1433 if (rules & Rnolabel)
1434 fputs(" (labels not permitted)", stderr);
1437 ###### core functions
1439 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1440 struct type *type, int rules)
1447 switch (prog->type) {
1450 struct binode *b = cast(binode, prog);
1452 ## propagate binode cases
1456 ## propagate exec cases
1463 Interpreting an `exec` doesn't require anything but the `exec`. State
1464 is stored in variables and each variable will be directly linked from
1465 within the `exec` tree. The exception to this is the whole `program`
1466 which needs to look at command line arguments. The `program` will be
1467 interpreted separately.
1469 Each `exec` can return a value, which may be `Tnone` but must be non-NULL;
1471 ###### core functions
1474 struct value val, *lval;
1477 static struct lrval _interp_exec(struct exec *e);
1479 static struct value interp_exec(struct exec *e)
1481 struct lrval ret = _interp_exec(e);
1484 return dup_value(*ret.lval);
1489 static struct value *linterp_exec(struct exec *e)
1491 struct lrval ret = _interp_exec(e);
1496 static struct lrval _interp_exec(struct exec *e)
1499 struct value rv, *lrv = NULL;
1510 struct binode *b = cast(binode, e);
1511 struct value left, right, *lleft;
1512 left.type = right.type = Tnone;
1514 ## interp binode cases
1516 free_value(left); free_value(right);
1519 ## interp exec cases
1526 ## Language elements
1528 Each language element needs to be parsed, printed, analysed,
1529 interpreted, and freed. There are several, so let's just start with
1530 the easy ones and work our way up.
1534 We have already met values as separate objects. When manifest
1535 constants appear in the program text, that must result in an executable
1536 which has a constant value. So the `val` structure embeds a value in
1552 $0 = new_pos(val, $1);
1553 $0->val.type = Tbool;
1557 $0 = new_pos(val, $1);
1558 $0->val.type = Tbool;
1562 $0 = new_pos(val, $1);
1563 $0->val.type = Tnum;
1566 if (number_parse($0->val.num, tail, $1.txt) == 0)
1567 mpq_init($0->val.num);
1569 tok_err(config2context(config), "error: unsupported number suffix",
1574 $0 = new_pos(val, $1);
1575 $0->val.type = Tstr;
1578 string_parse(&$1, '\\', &$0->val.str, tail);
1580 tok_err(config2context(config), "error: unsupported string suffix",
1585 $0 = new_pos(val, $1);
1586 $0->val.type = Tstr;
1589 string_parse(&$1, '\\', &$0->val.str, tail);
1591 tok_err(config2context(config), "error: unsupported string suffix",
1596 ###### print exec cases
1599 struct val *v = cast(val, e);
1600 if (v->val.type == Tstr)
1602 print_value(v->val);
1603 if (v->val.type == Tstr)
1608 ###### propagate exec cases
1611 struct val *val = cast(val, prog);
1612 if (!type_compat(type, val->val.type, rules)) {
1613 type_err(c, "error: expected %1%r found %2",
1614 prog, type, rules, val->val.type);
1617 return val->val.type;
1620 ###### interp exec cases
1622 rv = dup_value(cast(val, e)->val);
1625 ###### ast functions
1626 static void free_val(struct val *v)
1634 ###### free exec cases
1635 case Xval: free_val(cast(val, e)); break;
1637 ###### ast functions
1638 // Move all nodes from 'b' to 'rv', reversing the order.
1639 // In 'b' 'left' is a list, and 'right' is the last node.
1640 // In 'rv', left' is the first node and 'right' is a list.
1641 static struct binode *reorder_bilist(struct binode *b)
1643 struct binode *rv = NULL;
1646 struct exec *t = b->right;
1650 b = cast(binode, b->left);
1660 Just as we used a `val` to wrap a value into an `exec`, we similarly
1661 need a `var` to wrap a `variable` into an exec. While each `val`
1662 contained a copy of the value, each `var` hold a link to the variable
1663 because it really is the same variable no matter where it appears.
1664 When a variable is used, we need to remember to follow the `->merged`
1665 link to find the primary instance.
1673 struct variable *var;
1679 VariableDecl -> IDENTIFIER : ${ {
1680 struct variable *v = var_decl(config2context(config), $1.txt);
1681 $0 = new_pos(var, $1);
1686 v = var_ref(config2context(config), $1.txt);
1688 type_err(config2context(config), "error: variable '%v' redeclared",
1689 $0, Tnone, 0, Tnone);
1690 type_err(config2context(config), "info: this is where '%v' was first declared",
1691 v->where_decl, Tnone, 0, Tnone);
1694 | IDENTIFIER :: ${ {
1695 struct variable *v = var_decl(config2context(config), $1.txt);
1696 $0 = new_pos(var, $1);
1702 v = var_ref(config2context(config), $1.txt);
1704 type_err(config2context(config), "error: variable '%v' redeclared",
1705 $0, Tnone, 0, Tnone);
1706 type_err(config2context(config), "info: this is where '%v' was first declared",
1707 v->where_decl, Tnone, 0, Tnone);
1710 | IDENTIFIER : Type ${ {
1711 struct variable *v = var_decl(config2context(config), $1.txt);
1712 $0 = new_pos(var, $1);
1717 v->val = val_prepare($<3);
1719 v = var_ref(config2context(config), $1.txt);
1721 type_err(config2context(config), "error: variable '%v' redeclared",
1722 $0, Tnone, 0, Tnone);
1723 type_err(config2context(config), "info: this is where '%v' was first declared",
1724 v->where_decl, Tnone, 0, Tnone);
1727 | IDENTIFIER :: Type ${ {
1728 struct variable *v = var_decl(config2context(config), $1.txt);
1729 $0 = new_pos(var, $1);
1734 v->val = val_prepare($<3);
1737 v = var_ref(config2context(config), $1.txt);
1739 type_err(config2context(config), "error: variable '%v' redeclared",
1740 $0, Tnone, 0, Tnone);
1741 type_err(config2context(config), "info: this is where '%v' was first declared",
1742 v->where_decl, Tnone, 0, Tnone);
1747 Variable -> IDENTIFIER ${ {
1748 struct variable *v = var_ref(config2context(config), $1.txt);
1749 $0 = new_pos(var, $1);
1751 /* This might be a label - allocate a var just in case */
1752 v = var_decl(config2context(config), $1.txt);
1754 v->val = val_prepare(Tlabel);
1755 v->val.label = &v->val;
1759 cast(var, $0)->var = v;
1764 Type -> IDENTIFIER ${
1765 $0 = find_type(config2context(config), $1.txt);
1767 tok_err(config2context(config),
1768 "error: undefined type", &$1);
1775 ###### print exec cases
1778 struct var *v = cast(var, e);
1780 struct binding *b = v->var->name;
1781 printf("%.*s", b->name.len, b->name.txt);
1788 if (loc->type == Xvar) {
1789 struct var *v = cast(var, loc);
1791 struct binding *b = v->var->name;
1792 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
1794 fputs("???", stderr);
1796 fputs("NOTVAR", stderr);
1799 ###### propagate exec cases
1803 struct var *var = cast(var, prog);
1804 struct variable *v = var->var;
1806 type_err(c, "%d:BUG: no variable!!", prog, Tnone, 0, Tnone);
1812 if (v->constant && (rules & Rnoconstant)) {
1813 type_err(c, "error: Cannot assign to a constant: %v",
1814 prog, NULL, 0, NULL);
1815 type_err(c, "info: name was defined as a constant here",
1816 v->where_decl, NULL, 0, NULL);
1820 if (v->val.type == NULL) {
1821 if (type && *ok != 0) {
1822 v->val = val_prepare(type);
1823 v->where_set = prog;
1828 if (!type_compat(type, v->val.type, rules)) {
1829 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
1830 type, rules, v->val.type);
1831 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
1832 v->val.type, rules, Tnone);
1840 ###### interp exec cases
1843 struct var *var = cast(var, e);
1844 struct variable *v = var->var;
1852 ###### ast functions
1854 static void free_var(struct var *v)
1859 ###### free exec cases
1860 case Xvar: free_var(cast(var, e)); break;
1862 ### Expressions: Boolean
1864 Our first user of the `binode` will be expressions, and particularly
1865 Boolean expressions. As I haven't implemented precedence in the
1866 parser generator yet, we need different names for each precedence
1867 level used by expressions. The outer most or lowest level precedence
1868 are Boolean `or` `and`, and `not` which form an `Expression` out of `BTerm`s
1879 Expression -> Expression or BTerm ${ {
1880 struct binode *b = new(binode);
1886 | BTerm ${ $0 = $<1; }$
1888 BTerm -> BTerm and BFact ${ {
1889 struct binode *b = new(binode);
1895 | BFact ${ $0 = $<1; }$
1897 BFact -> not BFact ${ {
1898 struct binode *b = new(binode);
1905 ###### print binode cases
1907 print_exec(b->left, -1, 0);
1909 print_exec(b->right, -1, 0);
1912 print_exec(b->left, -1, 0);
1914 print_exec(b->right, -1, 0);
1918 print_exec(b->right, -1, 0);
1921 ###### propagate binode cases
1925 /* both must be Tbool, result is Tbool */
1926 propagate_types(b->left, c, ok, Tbool, 0);
1927 propagate_types(b->right, c, ok, Tbool, 0);
1928 if (type && type != Tbool) {
1929 type_err(c, "error: %1 operation found where %2 expected", prog,
1935 ###### interp binode cases
1937 rv = interp_exec(b->left);
1938 right = interp_exec(b->right);
1939 rv.bool = rv.bool && right.bool;
1942 rv = interp_exec(b->left);
1943 right = interp_exec(b->right);
1944 rv.bool = rv.bool || right.bool;
1947 rv = interp_exec(b->right);
1951 ### Expressions: Comparison
1953 Of slightly higher precedence that Boolean expressions are
1955 A comparison takes arguments of any type, but the two types must be
1958 To simplify the parsing we introduce an `eop` which can record an
1959 expression operator.
1966 ###### ast functions
1967 static void free_eop(struct eop *e)
1982 | Expr CMPop Expr ${ {
1983 struct binode *b = new(binode);
1989 | Expr ${ $0 = $<1; }$
1994 CMPop -> < ${ $0.op = Less; }$
1995 | > ${ $0.op = Gtr; }$
1996 | <= ${ $0.op = LessEq; }$
1997 | >= ${ $0.op = GtrEq; }$
1998 | == ${ $0.op = Eql; }$
1999 | != ${ $0.op = NEql; }$
2001 ###### print binode cases
2009 print_exec(b->left, -1, 0);
2011 case Less: printf(" < "); break;
2012 case LessEq: printf(" <= "); break;
2013 case Gtr: printf(" > "); break;
2014 case GtrEq: printf(" >= "); break;
2015 case Eql: printf(" == "); break;
2016 case NEql: printf(" != "); break;
2019 print_exec(b->right, -1, 0);
2022 ###### propagate binode cases
2029 /* Both must match but not be labels, result is Tbool */
2030 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2032 propagate_types(b->right, c, ok, t, 0);
2034 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2036 t = propagate_types(b->left, c, ok, t, 0);
2038 if (!type_compat(type, Tbool, 0)) {
2039 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2040 Tbool, rules, type);
2045 ###### interp binode cases
2054 left = interp_exec(b->left);
2055 right = interp_exec(b->right);
2056 cmp = value_cmp(left, right);
2059 case Less: rv.bool = cmp < 0; break;
2060 case LessEq: rv.bool = cmp <= 0; break;
2061 case Gtr: rv.bool = cmp > 0; break;
2062 case GtrEq: rv.bool = cmp >= 0; break;
2063 case Eql: rv.bool = cmp == 0; break;
2064 case NEql: rv.bool = cmp != 0; break;
2065 default: rv.bool = 0; break;
2070 ### Expressions: The rest
2072 The remaining expressions with the highest precedence are arithmetic
2073 and string concatenation. They are `Expr`, `Term`, and `Factor`.
2074 The `Factor` is where the `Value` and `Variable` that we already have
2077 `+` and `-` are both infix and prefix operations (where they are
2078 absolute value and negation). These have different operator names.
2080 We also have a 'Bracket' operator which records where parentheses were
2081 found. This makes it easy to reproduce these when printing. Once
2082 precedence is handled better I might be able to discard this.
2094 Expr -> Expr Eop Term ${ {
2095 struct binode *b = new(binode);
2101 | Term ${ $0 = $<1; }$
2103 Term -> Term Top Factor ${ {
2104 struct binode *b = new(binode);
2110 | Factor ${ $0 = $<1; }$
2112 Factor -> ( Expression ) ${ {
2113 struct binode *b = new_pos(binode, $1);
2119 struct binode *b = new(binode);
2124 | Value ${ $0 = $<1; }$
2125 | Variable ${ $0 = $<1; }$
2128 Eop -> + ${ $0.op = Plus; }$
2129 | - ${ $0.op = Minus; }$
2131 Uop -> + ${ $0.op = Absolute; }$
2132 | - ${ $0.op = Negate; }$
2134 Top -> * ${ $0.op = Times; }$
2135 | / ${ $0.op = Divide; }$
2136 | % ${ $0.op = Rem; }$
2137 | ++ ${ $0.op = Concat; }$
2139 ###### print binode cases
2146 print_exec(b->left, indent, 0);
2148 case Plus: fputs(" + ", stdout); break;
2149 case Minus: fputs(" - ", stdout); break;
2150 case Times: fputs(" * ", stdout); break;
2151 case Divide: fputs(" / ", stdout); break;
2152 case Rem: fputs(" % ", stdout); break;
2153 case Concat: fputs(" ++ ", stdout); break;
2156 print_exec(b->right, indent, 0);
2160 print_exec(b->right, indent, 0);
2164 print_exec(b->right, indent, 0);
2168 print_exec(b->right, indent, 0);
2172 ###### propagate binode cases
2178 /* both must be numbers, result is Tnum */
2181 /* as propagate_types ignores a NULL,
2182 * unary ops fit here too */
2183 propagate_types(b->left, c, ok, Tnum, 0);
2184 propagate_types(b->right, c, ok, Tnum, 0);
2185 if (!type_compat(type, Tnum, 0)) {
2186 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
2193 /* both must be Tstr, result is Tstr */
2194 propagate_types(b->left, c, ok, Tstr, 0);
2195 propagate_types(b->right, c, ok, Tstr, 0);
2196 if (!type_compat(type, Tstr, 0)) {
2197 type_err(c, "error: Concat returns %1 but %2 expected", prog,
2204 return propagate_types(b->right, c, ok, type, 0);
2206 ###### interp binode cases
2209 rv = interp_exec(b->left);
2210 right = interp_exec(b->right);
2211 mpq_add(rv.num, rv.num, right.num);
2214 rv = interp_exec(b->left);
2215 right = interp_exec(b->right);
2216 mpq_sub(rv.num, rv.num, right.num);
2219 rv = interp_exec(b->left);
2220 right = interp_exec(b->right);
2221 mpq_mul(rv.num, rv.num, right.num);
2224 rv = interp_exec(b->left);
2225 right = interp_exec(b->right);
2226 mpq_div(rv.num, rv.num, right.num);
2231 left = interp_exec(b->left);
2232 right = interp_exec(b->right);
2233 mpz_init(l); mpz_init(r); mpz_init(rem);
2234 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
2235 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
2236 mpz_tdiv_r(rem, l, r);
2237 rv = val_init(Tnum);
2238 mpq_set_z(rv.num, rem);
2239 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
2243 rv = interp_exec(b->right);
2244 mpq_neg(rv.num, rv.num);
2247 rv = interp_exec(b->right);
2248 mpq_abs(rv.num, rv.num);
2251 rv = interp_exec(b->right);
2254 left = interp_exec(b->left);
2255 right = interp_exec(b->right);
2257 rv.str = text_join(left.str, right.str);
2261 ###### value functions
2263 static struct text text_join(struct text a, struct text b)
2266 rv.len = a.len + b.len;
2267 rv.txt = malloc(rv.len);
2268 memcpy(rv.txt, a.txt, a.len);
2269 memcpy(rv.txt+a.len, b.txt, b.len);
2274 ### Blocks, Statements, and Statement lists.
2276 Now that we have expressions out of the way we need to turn to
2277 statements. There are simple statements and more complex statements.
2278 Simple statements do not contain newlines, complex statements do.
2280 Statements often come in sequences and we have corresponding simple
2281 statement lists and complex statement lists.
2282 The former comprise only simple statements separated by semicolons.
2283 The later comprise complex statements and simple statement lists. They are
2284 separated by newlines. Thus the semicolon is only used to separate
2285 simple statements on the one line. This may be overly restrictive,
2286 but I'm not sure I ever want a complex statement to share a line with
2289 Note that a simple statement list can still use multiple lines if
2290 subsequent lines are indented, so
2292 ###### Example: wrapped simple statement list
2297 is a single simple statement list. This might allow room for
2298 confusion, so I'm not set on it yet.
2300 A simple statement list needs no extra syntax. A complex statement
2301 list has two syntactic forms. It can be enclosed in braces (much like
2302 C blocks), or it can be introduced by a colon and continue until an
2303 unindented newline (much like Python blocks). With this extra syntax
2304 it is referred to as a block.
2306 Note that a block does not have to include any newlines if it only
2307 contains simple statements. So both of:
2309 if condition: a=b; d=f
2311 if condition { a=b; print f }
2315 In either case the list is constructed from a `binode` list with
2316 `Block` as the operator. When parsing the list it is most convenient
2317 to append to the end, so a list is a list and a statement. When using
2318 the list it is more convenient to consider a list to be a statement
2319 and a list. So we need a function to re-order a list.
2320 `reorder_bilist` serves this purpose.
2322 The only stand-alone statement we introduce at this stage is `pass`
2323 which does nothing and is represented as a `NULL` pointer in a `Block`
2324 list. Other stand-alone statements will follow once the infrastructure
2344 Block -> Open Statementlist Close ${ $0 = $<2; }$
2345 | Open Newlines Statementlist Close ${ $0 = $<3; }$
2346 | Open SimpleStatements } ${ $0 = reorder_bilist($<2); }$
2347 | Open Newlines SimpleStatements } ${ $0 = reorder_bilist($<3); }$
2348 | : Statementlist ${ $0 = $<2; }$
2349 | : SimpleStatements ${ $0 = reorder_bilist($<2); }$
2351 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<1); }$
2353 ComplexStatements -> ComplexStatements ComplexStatement ${
2359 | ComplexStatements NEWLINE ${ $0 = $<1; }$
2360 | ComplexStatement ${
2368 ComplexStatement -> SimpleStatements NEWLINE ${
2369 $0 = reorder_bilist($<1);
2371 ## ComplexStatement Grammar
2374 SimpleStatements -> SimpleStatements ; SimpleStatement ${
2380 | SimpleStatement ${
2386 | SimpleStatements ; ${ $0 = $<1; }$
2388 SimpleStatement -> pass ${ $0 = NULL; }$
2389 ## SimpleStatement Grammar
2391 ###### print binode cases
2395 if (b->left == NULL)
2398 print_exec(b->left, indent, 0);
2401 print_exec(b->right, indent, 0);
2404 // block, one per line
2405 if (b->left == NULL)
2406 do_indent(indent, "pass\n");
2408 print_exec(b->left, indent, bracket);
2410 print_exec(b->right, indent, bracket);
2414 ###### propagate binode cases
2417 /* If any statement returns something other than Tnone
2418 * or Tbool then all such must return same type.
2419 * As each statement may be Tnone or something else,
2420 * we must always pass NULL (unknown) down, otherwise an incorrect
2421 * error might occur. We never return Tnone unless it is
2426 for (e = b; e; e = cast(binode, e->right)) {
2427 t = propagate_types(e->left, c, ok, NULL, rules);
2428 if ((rules & Rboolok) && t == Tbool)
2430 if (t && t != Tnone && t != Tbool) {
2433 else if (t != type) {
2434 type_err(c, "error: expected %1%r, found %2",
2435 e->left, type, rules, t);
2443 ###### interp binode cases
2445 while (rv.type == Tnone &&
2448 rv = interp_exec(b->left);
2449 b = cast(binode, b->right);
2453 ### The Print statement
2455 `print` is a simple statement that takes a comma-separated list of
2456 expressions and prints the values separated by spaces and terminated
2457 by a newline. No control of formatting is possible.
2459 `print` faces the same list-ordering issue as blocks, and uses the
2465 ###### SimpleStatement Grammar
2467 | print ExpressionList ${
2468 $0 = reorder_bilist($<2);
2470 | print ExpressionList , ${
2475 $0 = reorder_bilist($0);
2486 ExpressionList -> ExpressionList , Expression ${
2499 ###### print binode cases
2502 do_indent(indent, "print");
2506 print_exec(b->left, -1, 0);
2510 b = cast(binode, b->right);
2516 ###### propagate binode cases
2519 /* don't care but all must be consistent */
2520 propagate_types(b->left, c, ok, NULL, Rnolabel);
2521 propagate_types(b->right, c, ok, NULL, Rnolabel);
2524 ###### interp binode cases
2530 for ( ; b; b = cast(binode, b->right))
2534 left = interp_exec(b->left);
2547 ###### Assignment statement
2549 An assignment will assign a value to a variable, providing it hasn't
2550 be declared as a constant. The analysis phase ensures that the type
2551 will be correct so the interpreter just needs to perform the
2552 calculation. There is a form of assignment which declares a new
2553 variable as well as assigning a value. If a name is assigned before
2554 it is declared, and error will be raised as the name is created as
2555 `Tlabel` and it is illegal to assign to such names.
2561 ###### SimpleStatement Grammar
2562 | Variable = Expression ${
2568 | VariableDecl = Expression ${
2576 if ($1->var->where_set == NULL) {
2577 type_err(config2context(config), "Variable declared with no type or value: %v",
2587 ###### print binode cases
2590 do_indent(indent, "");
2591 print_exec(b->left, indent, 0);
2593 print_exec(b->right, indent, 0);
2600 struct variable *v = cast(var, b->left)->var;
2601 do_indent(indent, "");
2602 print_exec(b->left, indent, 0);
2603 if (cast(var, b->left)->var->constant) {
2604 if (v->where_decl == v->where_set) {
2606 type_print(v->val.type, stdout);
2611 if (v->where_decl == v->where_set) {
2613 type_print(v->val.type, stdout);
2620 print_exec(b->right, indent, 0);
2627 ###### propagate binode cases
2631 /* Both must match and not be labels,
2632 * Type must support 'dup',
2633 * For Assign, left must not be constant.
2636 t = propagate_types(b->left, c, ok, NULL,
2637 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
2642 if (propagate_types(b->right, c, ok, t, 0) != t)
2643 if (b->left->type == Xvar)
2644 type_err(c, "info: variable '%v' was set as %1 here.",
2645 cast(var, b->left)->var->where_set, t, rules, Tnone);
2647 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2649 propagate_types(b->left, c, ok, t,
2650 (b->op == Assign ? Rnoconstant : 0));
2652 if (t && t->dup == NULL) {
2653 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
2660 ###### interp binode cases
2663 lleft = linterp_exec(b->left);
2664 right = interp_exec(b->right);
2675 struct variable *v = cast(var, b->left)->var;
2679 right = interp_exec(b->right);
2681 right = val_init(v->val.type);
2688 ### The `use` statement
2690 The `use` statement is the last "simple" statement. It is needed when
2691 the condition in a conditional statement is a block. `use` works much
2692 like `return` in C, but only completes the `condition`, not the whole
2698 ###### SimpleStatement Grammar
2700 $0 = new_pos(binode, $1);
2705 ###### print binode cases
2708 do_indent(indent, "use ");
2709 print_exec(b->right, -1, 0);
2714 ###### propagate binode cases
2717 /* result matches value */
2718 return propagate_types(b->right, c, ok, type, 0);
2720 ###### interp binode cases
2723 rv = interp_exec(b->right);
2726 ### The Conditional Statement
2728 This is the biggy and currently the only complex statement. This
2729 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
2730 It is comprised of a number of parts, all of which are optional though
2731 set combinations apply. Each part is (usually) a key word (`then` is
2732 sometimes optional) followed by either an expression or a code block,
2733 except the `casepart` which is a "key word and an expression" followed
2734 by a code block. The code-block option is valid for all parts and,
2735 where an expression is also allowed, the code block can use the `use`
2736 statement to report a value. If the code block does not report a value
2737 the effect is similar to reporting `True`.
2739 The `else` and `case` parts, as well as `then` when combined with
2740 `if`, can contain a `use` statement which will apply to some
2741 containing conditional statement. `for` parts, `do` parts and `then`
2742 parts used with `for` can never contain a `use`, except in some
2743 subordinate conditional statement.
2745 If there is a `forpart`, it is executed first, only once.
2746 If there is a `dopart`, then it is executed repeatedly providing
2747 always that the `condpart` or `cond`, if present, does not return a non-True
2748 value. `condpart` can fail to return any value if it simply executes
2749 to completion. This is treated the same as returning `True`.
2751 If there is a `thenpart` it will be executed whenever the `condpart`
2752 or `cond` returns True (or does not return any value), but this will happen
2753 *after* `dopart` (when present).
2755 If `elsepart` is present it will be executed at most once when the
2756 condition returns `False` or some value that isn't `True` and isn't
2757 matched by any `casepart`. If there are any `casepart`s, they will be
2758 executed when the condition returns a matching value.
2760 The particular sorts of values allowed in case parts has not yet been
2761 determined in the language design, so nothing is prohibited.
2763 The various blocks in this complex statement potentially provide scope
2764 for variables as described earlier. Each such block must include the
2765 "OpenScope" nonterminal before parsing the block, and must call
2766 `var_block_close()` when closing the block.
2768 The code following "`if`", "`switch`" and "`for`" does not get its own
2769 scope, but is in a scope covering the whole statement, so names
2770 declared there cannot be redeclared elsewhere. Similarly the
2771 condition following "`while`" is in a scope the covers the body
2772 ("`do`" part) of the loop, and which does not allow conditional scope
2773 extension. Code following "`then`" (both looping and non-looping),
2774 "`else`" and "`case`" each get their own local scope.
2776 The type requirements on the code block in a `whilepart` are quite
2777 unusal. It is allowed to return a value of some identifiable type, in
2778 which case the loop aborts and an appropriate `casepart` is run, or it
2779 can return a Boolean, in which case the loop either continues to the
2780 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
2781 This is different both from the `ifpart` code block which is expected to
2782 return a Boolean, or the `switchpart` code block which is expected to
2783 return the same type as the casepart values. The correct analysis of
2784 the type of the `whilepart` code block is the reason for the
2785 `Rboolok` flag which is passed to `propagate_types()`.
2787 The `cond_statement` cannot fit into a `binode` so a new `exec` is
2796 struct exec *action;
2797 struct casepart *next;
2799 struct cond_statement {
2801 struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
2802 struct casepart *casepart;
2805 ###### ast functions
2807 static void free_casepart(struct casepart *cp)
2811 free_exec(cp->value);
2812 free_exec(cp->action);
2819 static void free_cond_statement(struct cond_statement *s)
2823 free_exec(s->forpart);
2824 free_exec(s->condpart);
2825 free_exec(s->dopart);
2826 free_exec(s->thenpart);
2827 free_exec(s->elsepart);
2828 free_casepart(s->casepart);
2832 ###### free exec cases
2833 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
2835 ###### ComplexStatement Grammar
2836 | CondStatement ${ $0 = $<1; }$
2841 // both ForThen and Whilepart open scopes, and CondSuffix only
2842 // closes one - so in the first branch here we have another to close.
2843 CondStatement -> ForThen WhilePart CondSuffix ${
2845 $0->forpart = $1.forpart; $1.forpart = NULL;
2846 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
2847 $0->condpart = $2.condpart; $2.condpart = NULL;
2848 $0->dopart = $2.dopart; $2.dopart = NULL;
2849 var_block_close(config2context(config), CloseSequential);
2851 | WhilePart CondSuffix ${
2853 $0->condpart = $1.condpart; $1.condpart = NULL;
2854 $0->dopart = $1.dopart; $1.dopart = NULL;
2856 | SwitchPart CondSuffix ${
2860 | IfPart IfSuffix ${
2862 $0->condpart = $1.condpart; $1.condpart = NULL;
2863 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
2864 // This is where we close an "if" statement
2865 var_block_close(config2context(config), CloseSequential);
2868 CondSuffix -> IfSuffix ${
2870 // This is where we close scope of the whole
2871 // "for" or "while" statement
2872 var_block_close(config2context(config), CloseSequential);
2874 | CasePart CondSuffix ${
2876 $1->next = $0->casepart;
2881 CasePart -> Newlines case Expression OpenScope Block ${
2882 $0 = calloc(1,sizeof(struct casepart));
2885 var_block_close(config2context(config), CloseParallel);
2887 | case Expression OpenScope Block ${
2888 $0 = calloc(1,sizeof(struct casepart));
2891 var_block_close(config2context(config), CloseParallel);
2895 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
2896 | Newlines else OpenScope Block ${
2897 $0 = new(cond_statement);
2899 var_block_close(config2context(config), CloseElse);
2901 | else OpenScope Block ${
2902 $0 = new(cond_statement);
2904 var_block_close(config2context(config), CloseElse);
2906 | Newlines else OpenScope CondStatement ${
2907 $0 = new(cond_statement);
2909 var_block_close(config2context(config), CloseElse);
2911 | else OpenScope CondStatement ${
2912 $0 = new(cond_statement);
2914 var_block_close(config2context(config), CloseElse);
2919 // These scopes are closed in CondSuffix
2920 ForPart -> for OpenScope SimpleStatements ${
2921 $0 = reorder_bilist($<3);
2923 | for OpenScope Block ${
2927 ThenPart -> then OpenScope SimpleStatements ${
2928 $0 = reorder_bilist($<3);
2929 var_block_close(config2context(config), CloseSequential);
2931 | then OpenScope Block ${
2933 var_block_close(config2context(config), CloseSequential);
2936 ThenPartNL -> ThenPart OptNL ${
2940 // This scope is closed in CondSuffix
2941 WhileHead -> while OpenScope Block ${
2946 ForThen -> ForPart OptNL ThenPartNL ${
2954 // This scope is closed in CondSuffix
2955 WhilePart -> while OpenScope Expression Block ${
2956 $0.type = Xcond_statement;
2960 | WhileHead OptNL do Block ${
2961 $0.type = Xcond_statement;
2966 IfPart -> if OpenScope Expression OpenScope Block ${
2967 $0.type = Xcond_statement;
2970 var_block_close(config2context(config), CloseParallel);
2972 | if OpenScope Block OptNL then OpenScope Block ${
2973 $0.type = Xcond_statement;
2976 var_block_close(config2context(config), CloseParallel);
2980 // This scope is closed in CondSuffix
2981 SwitchPart -> switch OpenScope Expression ${
2984 | switch OpenScope Block ${
2988 ###### print exec cases
2990 case Xcond_statement:
2992 struct cond_statement *cs = cast(cond_statement, e);
2993 struct casepart *cp;
2995 do_indent(indent, "for");
2996 if (bracket) printf(" {\n"); else printf(":\n");
2997 print_exec(cs->forpart, indent+1, bracket);
3000 do_indent(indent, "} then {\n");
3002 do_indent(indent, "then:\n");
3003 print_exec(cs->thenpart, indent+1, bracket);
3005 if (bracket) do_indent(indent, "}\n");
3009 if (cs->condpart && cs->condpart->type == Xbinode &&
3010 cast(binode, cs->condpart)->op == Block) {
3012 do_indent(indent, "while {\n");
3014 do_indent(indent, "while:\n");
3015 print_exec(cs->condpart, indent+1, bracket);
3017 do_indent(indent, "} do {\n");
3019 do_indent(indent, "do:\n");
3020 print_exec(cs->dopart, indent+1, bracket);
3022 do_indent(indent, "}\n");
3024 do_indent(indent, "while ");
3025 print_exec(cs->condpart, 0, bracket);
3030 print_exec(cs->dopart, indent+1, bracket);
3032 do_indent(indent, "}\n");
3037 do_indent(indent, "switch");
3039 do_indent(indent, "if");
3040 if (cs->condpart && cs->condpart->type == Xbinode &&
3041 cast(binode, cs->condpart)->op == Block) {
3046 print_exec(cs->condpart, indent+1, bracket);
3048 do_indent(indent, "}\n");
3050 do_indent(indent, "then:\n");
3051 print_exec(cs->thenpart, indent+1, bracket);
3055 print_exec(cs->condpart, 0, bracket);
3061 print_exec(cs->thenpart, indent+1, bracket);
3063 do_indent(indent, "}\n");
3068 for (cp = cs->casepart; cp; cp = cp->next) {
3069 do_indent(indent, "case ");
3070 print_exec(cp->value, -1, 0);
3075 print_exec(cp->action, indent+1, bracket);
3077 do_indent(indent, "}\n");
3080 do_indent(indent, "else");
3085 print_exec(cs->elsepart, indent+1, bracket);
3087 do_indent(indent, "}\n");
3092 ###### propagate exec cases
3093 case Xcond_statement:
3095 // forpart and dopart must return Tnone
3096 // thenpart must return Tnone if there is a dopart,
3097 // otherwise it is like elsepart.
3099 // be bool if there is no casepart
3100 // match casepart->values if there is a switchpart
3101 // either be bool or match casepart->value if there
3103 // elsepart and casepart->action must match the return type
3104 // expected of this statement.
3105 struct cond_statement *cs = cast(cond_statement, prog);
3106 struct casepart *cp;
3108 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
3109 if (!type_compat(Tnone, t, 0))
3111 t = propagate_types(cs->dopart, c, ok, Tnone, 0);
3112 if (!type_compat(Tnone, t, 0))
3115 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
3116 if (!type_compat(Tnone, t, 0))
3119 if (cs->casepart == NULL)
3120 propagate_types(cs->condpart, c, ok, Tbool, 0);
3122 /* Condpart must match case values, with bool permitted */
3124 for (cp = cs->casepart;
3125 cp && !t; cp = cp->next)
3126 t = propagate_types(cp->value, c, ok, NULL, 0);
3127 if (!t && cs->condpart)
3128 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok);
3129 // Now we have a type (I hope) push it down
3131 for (cp = cs->casepart; cp; cp = cp->next)
3132 propagate_types(cp->value, c, ok, t, 0);
3133 propagate_types(cs->condpart, c, ok, t, Rboolok);
3136 // (if)then, else, and case parts must return expected type.
3137 if (!cs->dopart && !type)
3138 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
3140 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
3141 for (cp = cs->casepart;
3144 type = propagate_types(cp->action, c, ok, NULL, rules);
3147 propagate_types(cs->thenpart, c, ok, type, rules);
3148 propagate_types(cs->elsepart, c, ok, type, rules);
3149 for (cp = cs->casepart; cp ; cp = cp->next)
3150 propagate_types(cp->action, c, ok, type, rules);
3156 ###### interp exec cases
3157 case Xcond_statement:
3159 struct value v, cnd;
3160 struct casepart *cp;
3161 struct cond_statement *c = cast(cond_statement, e);
3164 interp_exec(c->forpart);
3167 cnd = interp_exec(c->condpart);
3170 if (!(cnd.type == Tnone ||
3171 (cnd.type == Tbool && cnd.bool != 0)))
3173 // cnd is Tnone or Tbool, doesn't need to be freed
3175 interp_exec(c->dopart);
3178 rv = interp_exec(c->thenpart);
3179 if (rv.type != Tnone || !c->dopart)
3183 } while (c->dopart);
3185 for (cp = c->casepart; cp; cp = cp->next) {
3186 v = interp_exec(cp->value);
3187 if (value_cmp(v, cnd) == 0) {
3190 rv = interp_exec(cp->action);
3197 rv = interp_exec(c->elsepart);
3206 Now that we have the shape of the interpreter in place we can add some
3207 complex types and connected them in to the data structures and the
3208 different phases of parse, analyse, print, interpret.
3210 For now, just arrays.
3214 Arrays can be declared by giving a size and a type, as `[size]type' so
3215 `freq:[26]number` declares `freq` to be an array of 26 numbers. The
3216 size can be an arbitrary expression which is evaluated when the name
3219 Arrays cannot be assigned. When pointers are introduced we will also
3220 introduce array slices which can refer to part or all of an array -
3221 the assignment syntax will create a slice. For now, an array can only
3222 ever be referenced by the name it is declared with. It is likely that
3223 a "`copy`" primitive will eventually be define which can be used to
3224 make a copy of an array with controllable depth.
3226 ###### type union fields
3230 struct variable *vsize;
3231 struct type *member;
3234 ###### value union fields
3236 struct value *elmnts;
3239 ###### value functions
3241 static struct value array_prepare(struct type *type)
3246 ret.array.elmnts = NULL;
3250 static struct value array_init(struct type *type)
3256 if (type->array.vsize) {
3259 mpz_tdiv_q(q, mpq_numref(type->array.vsize->val.num),
3260 mpq_denref(type->array.vsize->val.num));
3261 type->array.size = mpz_get_si(q);
3264 ret.array.elmnts = calloc(type->array.size,
3265 sizeof(ret.array.elmnts[0]));
3266 for (i = 0; ret.array.elmnts && i < type->array.size; i++)
3267 ret.array.elmnts[i] = val_init(type->array.member);
3271 static void array_free(struct value val)
3275 if (val.array.elmnts)
3276 for (i = 0; i < val.type->array.size; i++)
3277 free_value(val.array.elmnts[i]);
3278 free(val.array.elmnts);
3281 static int array_compat(struct type *require, struct type *have)
3283 if (have->compat != require->compat)
3285 /* Both are arrays, so we can look at details */
3286 if (!type_compat(require->array.member, have->array.member, 0))
3288 if (require->array.vsize == NULL && have->array.vsize == NULL)
3289 return require->array.size == have->array.size;
3291 return require->array.vsize == have->array.vsize;
3294 static void array_print_type(struct type *type, FILE *f)
3297 if (type->array.vsize) {
3298 struct binding *b = type->array.vsize->name;
3299 fprintf(f, "%.*s]", b->name.len, b->name.txt);
3301 fprintf(f, "%d]", type->array.size);
3302 type_print(type->array.member, f);
3305 static struct type array_prototype = {
3306 .prepare = array_prepare,
3308 .print_type = array_print_type,
3309 .compat = array_compat,
3315 | [ NUMBER ] Type ${
3316 $0 = calloc(1, sizeof(struct type));
3317 *($0) = array_prototype;
3318 $0->array.member = $<4;
3319 $0->array.vsize = NULL;
3323 if (number_parse(num, tail, $2.txt) == 0)
3324 tok_err(config2context(config), "error: unrecognised number", &$2);
3326 tok_err(config2context(config), "error: unsupported number suffix", &$2);
3328 $0->array.size = mpz_get_ui(mpq_numref(num));
3329 if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
3330 tok_err(config2context(config), "error: array size must be an integer",
3332 } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
3333 tok_err(config2context(config), "error: array size is too large",
3339 | [ IDENTIFIER ] Type ${ {
3340 struct variable *v = var_ref(config2context(config), $2.txt);
3343 tok_err(config2context(config), "error: name undeclared", &$2);
3344 else if (!v->constant)
3345 tok_err(config2context(config), "error: array size must be a constant", &$2);
3347 $0 = calloc(1, sizeof(struct type));
3348 *($0) = array_prototype;
3349 $0->array.member = $<4;
3351 $0->array.vsize = v;
3357 ###### variable grammar
3359 | Variable [ Expression ] ${ {
3360 struct binode *b = new(binode);
3367 ###### print binode cases
3369 print_exec(b->left, -1, 0);
3371 print_exec(b->right, -1, 0);
3375 ###### propagate binode cases
3377 /* left must be an array, right must be a number,
3378 * result is the member type of the array
3380 propagate_types(b->right, c, ok, Tnum, 0);
3381 t = propagate_types(b->left, c, ok, NULL, rules & Rnoconstant);
3382 if (!t || t->compat != array_compat) {
3383 type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
3387 if (!type_compat(type, t->array.member, rules)) {
3388 type_err(c, "error: have %1 but need %2", prog,
3389 t->array.member, rules, type);
3392 return t->array.member;
3396 ###### interp binode cases
3401 lleft = linterp_exec(b->left);
3402 right = interp_exec(b->right);
3404 mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
3408 if (i >= 0 && i < lleft->type->array.size)
3409 lrv = &lleft->array.elmnts[i];
3411 rv = val_init(lleft->type->array.member);
3415 ### Finally the whole program.
3417 Somewhat reminiscent of Pascal a (current) Ocean program starts with
3418 the keyword "program" and a list of variable names which are assigned
3419 values from command line arguments. Following this is a `block` which
3420 is the code to execute.
3422 As this is the top level, several things are handled a bit
3424 The whole program is not interpreted by `interp_exec` as that isn't
3425 passed the argument list which the program requires. Similarly type
3426 analysis is a bit more interesting at this level.
3431 ###### Parser: grammar
3434 Program -> program OpenScope Varlist Block OptNL ${
3437 $0->left = reorder_bilist($<3);
3439 var_block_close(config2context(config), CloseSequential);
3440 if (config2context(config)->scope_stack) abort();
3443 tok_err(config2context(config),
3444 "error: unhandled parse error", &$1);
3447 Varlist -> Varlist ArgDecl ${
3456 ArgDecl -> IDENTIFIER ${ {
3457 struct variable *v = var_decl(config2context(config), $1.txt);
3464 ###### print binode cases
3466 do_indent(indent, "program");
3467 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
3469 print_exec(b2->left, 0, 0);
3475 print_exec(b->right, indent+1, bracket);
3477 do_indent(indent, "}\n");
3480 ###### propagate binode cases
3481 case Program: abort();
3483 ###### core functions
3485 static int analyse_prog(struct exec *prog, struct parse_context *c)
3487 struct binode *b = cast(binode, prog);
3494 propagate_types(b->right, c, &ok, Tnone, 0);
3499 for (b = cast(binode, b->left); b; b = cast(binode, b->right)) {
3500 struct var *v = cast(var, b->left);
3501 if (!v->var->val.type) {
3502 v->var->where_set = b;
3503 v->var->val = val_prepare(Tstr);
3506 b = cast(binode, prog);
3509 propagate_types(b->right, c, &ok, Tnone, 0);
3514 /* Make sure everything is still consistent */
3515 propagate_types(b->right, c, &ok, Tnone, 0);
3519 static void interp_prog(struct exec *prog, char **argv)
3521 struct binode *p = cast(binode, prog);
3527 al = cast(binode, p->left);
3529 struct var *v = cast(var, al->left);
3530 struct value *vl = &v->var->val;
3532 if (argv[0] == NULL) {
3533 printf("Not enough args\n");
3536 al = cast(binode, al->right);
3538 *vl = parse_value(vl->type, argv[0]);
3539 if (vl->type == NULL)
3543 v = interp_exec(p->right);
3547 ###### interp binode cases
3548 case Program: abort();
3550 ## And now to test it out.
3552 Having a language requires having a "hello world" program. I'll
3553 provide a little more than that: a program that prints "Hello world"
3554 finds the GCD of two numbers, prints the first few elements of
3555 Fibonacci, and performs a binary search for a number.
3557 ###### File: oceani.mk
3560 @echo "===== TEST ====="
3561 ./oceani --section "test: hello" oceani.mdc 55 33
3566 print "Hello World, what lovely oceans you have!"
3567 /* When a variable is defined in both branches of an 'if',
3568 * and used afterwards, the variables are merged.
3574 print "Is", A, "bigger than", B,"? ", bigger
3575 /* If a variable is not used after the 'if', no
3576 * merge happens, so types can be different
3579 double:string = "yes"
3580 print A, "is more than twice", B, "?", double
3583 print "double", B, "is", double
3594 print "GCD of", A, "and", B,"is", a
3596 print a, "is not positive, cannot calculate GCD"
3598 print b, "is not positive, cannot calculate GCD"
3603 print "Fibonacci:", f1,f2,
3604 then togo = togo - 1
3612 /* Binary search... */
3617 mid := (lo + hi) / 2
3629 print "Yay, I found", target
3631 print "Closest I found was", mid
3636 for i:=1; then i = i + 1; while i < size:
3637 n := list[i-1] * list[i-1]
3638 list[i] = (n / 100) % 10000
3640 print "Before sort:"
3641 for i:=0; then i = i + 1; while i < size:
3642 print "list[",i,"]=",list[i]
3644 for i := 1; then i=i+1; while i < size:
3645 for j:=i-1; then j=j-1; while j >= 0:
3646 if list[j] > list[j+1]:
3651 for i:=0; then i = i + 1; while i < size:
3652 print "list[",i,"]=",list[i]