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);
583 ###### free context types
585 while (context.typelist) {
586 struct type *t = context.typelist;
588 context.typelist = t->next;
594 Values of the base types can be numbers, which we represent as
595 multi-precision fractions, strings, Booleans and labels. When
596 analysing the program we also need to allow for places where no value
597 is meaningful (type `Tnone`) and where we don't know what type to
598 expect yet (type is `NULL`).
600 Values are never shared, they are always copied when used, and freed
601 when no longer needed.
603 When propagating type information around the program, we need to
604 determine if two types are compatible, where type `NULL` is compatible
605 with anything. There are two special cases with type compatibility,
606 both related to the Conditional Statement which will be described
607 later. In some cases a Boolean can be accepted as well as some other
608 primary type, and in others any type is acceptable except a label (`Vlabel`).
609 A separate function encode these cases will simplify some code later.
611 When assigning command line arguments to variables, we need to be able
612 to parse each type from a string.
620 myLDLIBS := libnumber.o libstring.o -lgmp
621 LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
623 ###### type union fields
624 enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
626 ###### value union fields
633 static void _free_value(struct value v)
635 switch (v.type->vtype) {
637 case Vstr: free(v.str.txt); break;
638 case Vnum: mpq_clear(v.num); break;
644 ###### value functions
646 static struct value _val_prepare(struct type *type)
651 switch(type->vtype) {
655 memset(&rv.num, 0, sizeof(rv.num));
671 static struct value _val_init(struct type *type)
676 switch(type->vtype) {
680 mpq_init(rv.num); break;
682 rv.str.txt = malloc(1);
695 static struct value _dup_value(struct value v)
699 switch (rv.type->vtype) {
710 mpq_set(rv.num, v.num);
713 rv.str.len = v.str.len;
714 rv.str.txt = malloc(rv.str.len);
715 memcpy(rv.str.txt, v.str.txt, v.str.len);
721 static int _value_cmp(struct value left, struct value right)
724 if (left.type != right.type)
725 return left.type - right.type;
726 switch (left.type->vtype) {
727 case Vlabel: cmp = left.label == right.label ? 0 : 1; break;
728 case Vnum: cmp = mpq_cmp(left.num, right.num); break;
729 case Vstr: cmp = text_cmp(left.str, right.str); break;
730 case Vbool: cmp = left.bool - right.bool; break;
736 static void _print_value(struct value v)
738 switch (v.type->vtype) {
740 printf("*no-value*"); break;
742 printf("*label-%p*", v.label); break;
744 printf("%.*s", v.str.len, v.str.txt); break;
746 printf("%s", v.bool ? "True":"False"); break;
751 mpf_set_q(fl, v.num);
752 gmp_printf("%Fg", fl);
759 static struct value _parse_value(struct type *type, char *arg)
767 switch(type->vtype) {
773 val.str.len = strlen(arg);
774 val.str.txt = malloc(val.str.len);
775 memcpy(val.str.txt, arg, val.str.len);
782 tx.txt = arg; tx.len = strlen(tx.txt);
783 if (number_parse(val.num, tail, tx) == 0)
786 mpq_neg(val.num, val.num);
788 printf("Unsupported suffix: %s\n", arg);
793 if (strcasecmp(arg, "true") == 0 ||
794 strcmp(arg, "1") == 0)
796 else if (strcasecmp(arg, "false") == 0 ||
797 strcmp(arg, "0") == 0)
800 printf("Bad bool: %s\n", arg);
808 static void _free_value(struct value v);
810 static struct type base_prototype = {
812 .prepare = _val_prepare,
813 .parse = _parse_value,
814 .print = _print_value,
815 .cmp_order = _value_cmp,
816 .cmp_eq = _value_cmp,
821 static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
824 static struct type *add_base_type(struct parse_context *c, char *n, enum vtype vt)
826 struct text txt = { n, strlen(n) };
829 t = add_type(c, txt, &base_prototype);
834 ###### context initialization
836 Tbool = add_base_type(&context, "Boolean", Vbool);
837 Tstr = add_base_type(&context, "string", Vstr);
838 Tnum = add_base_type(&context, "number", Vnum);
839 Tnone = add_base_type(&context, "none", Vnone);
840 Tlabel = add_base_type(&context, "label", Vlabel);
844 Variables are scoped named values. We store the names in a linked
845 list of "bindings" sorted lexically, and use sequential search and
852 struct binding *next; // in lexical order
856 This linked list is stored in the parse context so that "reduce"
857 functions can find or add variables, and so the analysis phase can
858 ensure that every variable gets a type.
862 struct binding *varlist; // In lexical order
866 static struct binding *find_binding(struct parse_context *c, struct text s)
868 struct binding **l = &c->varlist;
873 (cmp = text_cmp((*l)->name, s)) < 0)
877 n = calloc(1, sizeof(*n));
884 Each name can be linked to multiple variables defined in different
885 scopes. Each scope starts where the name is declared and continues
886 until the end of the containing code block. Scopes of a given name
887 cannot nest, so a declaration while a name is in-scope is an error.
889 ###### binding fields
890 struct variable *var;
894 struct variable *previous;
896 struct binding *name;
897 struct exec *where_decl;// where name was declared
898 struct exec *where_set; // where type was set
902 While the naming seems strange, we include local constants in the
903 definition of variables. A name declared `var := value` can
904 subsequently be changed, but a name declared `var ::= value` cannot -
907 ###### variable fields
910 Scopes in parallel branches can be partially merged. More
911 specifically, if a given name is declared in both branches of an
912 if/else then its scope is a candidate for merging. Similarly if
913 every branch of an exhaustive switch (e.g. has an "else" clause)
914 declares a given name, then the scopes from the branches are
915 candidates for merging.
917 Note that names declared inside a loop (which is only parallel to
918 itself) are never visible after the loop. Similarly names defined in
919 scopes which are not parallel, such as those started by `for` and
920 `switch`, are never visible after the scope. Only variables defined in
921 both `then` and `else` (including the implicit then after an `if`, and
922 excluding `then` used with `for`) and in all `case`s and `else` of a
923 `switch` or `while` can be visible beyond the `if`/`switch`/`while`.
925 Labels, which are a bit like variables, follow different rules.
926 Labels are not explicitly declared, but if an undeclared name appears
927 in a context where a label is legal, that effectively declares the
928 name as a label. The declaration remains in force (or in scope) at
929 least to the end of the immediately containing block and conditionally
930 in any larger containing block which does not declare the name in some
931 other way. Importantly, the conditional scope extension happens even
932 if the label is only used in one parallel branch of a conditional --
933 when used in one branch it is treated as having been declared in all
936 Merge candidates are tentatively visible beyond the end of the
937 branching statement which creates them. If the name is used, the
938 merge is affirmed and they become a single variable visible at the
939 outer layer. If not - if it is redeclared first - the merge lapses.
941 To track scopes we have an extra stack, implemented as a linked list,
942 which roughly parallels the parse stack and which is used exclusively
943 for scoping. When a new scope is opened, a new frame is pushed and
944 the child-count of the parent frame is incremented. This child-count
945 is used to distinguish between the first of a set of parallel scopes,
946 in which declared variables must not be in scope, and subsequent
947 branches, whether they must already be conditionally scoped.
949 To push a new frame *before* any code in the frame is parsed, we need a
950 grammar reduction. This is most easily achieved with a grammar
951 element which derives the empty string, and creates the new scope when
952 it is recognized. This can be placed, for example, between a keyword
953 like "if" and the code following it.
957 struct scope *parent;
963 struct scope *scope_stack;
966 static void scope_pop(struct parse_context *c)
968 struct scope *s = c->scope_stack;
970 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;
988 OpenScope -> ${ scope_push(config2context(config)); }$
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
1011 seen. This state includes a secondary nest depth which records the
1012 outermost scope seen since the variable became conditionally in
1013 scope. If a use of the name is found, the variable becomes "in
1014 scope" and that secondary depth becomes the recorded scope depth.
1015 If the name is declared as a new variable, the old variable becomes
1016 "out of 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.
1023 ###### variable fields
1024 int depth, min_depth;
1025 enum { OutScope, PendingScope, CondScope, InScope } scope;
1026 struct variable *in_scope;
1028 ###### parse context
1030 struct variable *in_scope;
1032 All variables with the same name are linked together using the
1033 'previous' link. Those variable that have
1034 been affirmatively merged all have a 'merged' pointer that points to
1035 one primary variable - the most recently declared instance. When
1036 merging variables, we need to also adjust the 'merged' pointer on any
1037 other variables that had previously been merged with the one that will
1038 no longer be primary.
1040 ###### variable fields
1041 struct variable *merged;
1043 ###### ast functions
1045 static void variable_merge(struct variable *primary, struct variable *secondary)
1049 if (primary->merged)
1051 primary = primary->merged;
1053 for (v = primary->previous; v; v=v->previous)
1054 if (v == secondary || v == secondary->merged ||
1055 v->merged == secondary ||
1056 (v->merged && v->merged == secondary->merged)) {
1057 v->scope = OutScope;
1058 v->merged = primary;
1062 ###### free context vars
1064 while (context.varlist) {
1065 struct binding *b = context.varlist;
1066 struct variable *v = b->var;
1067 context.varlist = b->next;
1070 struct variable *t = v;
1078 #### Manipulating Bindings
1080 When a name is conditionally visible, a new declaration discards the
1081 old binding - the condition lapses. Conversely a usage of the name
1082 affirms the visibility and extends it to the end of the containing
1083 block - i.e. the block that contains both the original declaration and
1084 the latest usage. This is determined from `min_depth`. When a
1085 conditionally visible variable gets affirmed like this, it is also
1086 merged with other conditionally visible variables with the same name.
1088 When we parse a variable declaration we either signal an error if the
1089 name is currently bound, or create a new variable at the current nest
1090 depth if the name is unbound or bound to a conditionally scoped or
1091 pending-scope variable. If the previous variable was conditionally
1092 scoped, it and its homonyms becomes out-of-scope.
1094 When we parse a variable reference (including non-declarative
1095 assignment) we signal an error if the name is not bound or is bound to
1096 a pending-scope variable; update the scope if the name is bound to a
1097 conditionally scoped variable; or just proceed normally if the named
1098 variable is in scope.
1100 When we exit a scope, any variables bound at this level are either
1101 marked out of scope or pending-scoped, depending on whether the
1102 scope was sequential or parallel.
1104 When exiting a parallel scope we check if there are any variables that
1105 were previously pending and are still visible. If there are, then
1106 there weren't redeclared in the most recent scope, so they cannot be
1107 merged and must become out-of-scope. If it is not the first of
1108 parallel scopes (based on `child_count`), we check that there was a
1109 previous binding that is still pending-scope. If there isn't, the new
1110 variable must now be out-of-scope.
1112 When exiting a sequential scope that immediately enclosed parallel
1113 scopes, we need to resolve any pending-scope variables. If there was
1114 no `else` clause, and we cannot determine that the `switch` was exhaustive,
1115 we need to mark all pending-scope variable as out-of-scope. Otherwise
1116 all pending-scope variables become conditionally scoped.
1119 enum closetype { CloseSequential, CloseParallel, CloseElse };
1121 ###### ast functions
1123 static struct variable *var_decl(struct parse_context *c, struct text s)
1125 struct binding *b = find_binding(c, s);
1126 struct variable *v = b->var;
1128 switch (v ? v->scope : OutScope) {
1130 /* Caller will report the error */
1134 v && v->scope == CondScope;
1136 v->scope = OutScope;
1140 v = calloc(1, sizeof(*v));
1141 v->previous = b->var;
1144 v->min_depth = v->depth = c->scope_depth;
1146 v->in_scope = c->in_scope;
1148 v->val = val_prepare(NULL);
1152 static struct variable *var_ref(struct parse_context *c, struct text s)
1154 struct binding *b = find_binding(c, s);
1155 struct variable *v = b->var;
1156 struct variable *v2;
1158 switch (v ? v->scope : OutScope) {
1161 /* Signal an error - once that is possible */
1164 /* All CondScope variables of this name need to be merged
1165 * and become InScope
1167 v->depth = v->min_depth;
1169 for (v2 = v->previous;
1170 v2 && v2->scope == CondScope;
1172 variable_merge(v, v2);
1180 static void var_block_close(struct parse_context *c, enum closetype ct)
1182 /* close of all variables that are in_scope */
1183 struct variable *v, **vp, *v2;
1186 for (vp = &c->in_scope;
1187 v = *vp, v && v->depth > c->scope_depth && v->min_depth > c->scope_depth;
1191 case CloseParallel: /* handle PendingScope */
1195 if (c->scope_stack->child_count == 1)
1196 v->scope = PendingScope;
1197 else if (v->previous &&
1198 v->previous->scope == PendingScope)
1199 v->scope = PendingScope;
1200 else if (v->val.type == Tlabel)
1201 v->scope = PendingScope;
1202 else if (v->name->var == v)
1203 v->scope = OutScope;
1204 if (ct == CloseElse) {
1205 /* All Pending variables with this name
1206 * are now Conditional */
1208 v2 && v2->scope == PendingScope;
1210 v2->scope = CondScope;
1215 v2 && v2->scope == PendingScope;
1217 if (v2->val.type != Tlabel)
1218 v2->scope = OutScope;
1220 case OutScope: break;
1223 case CloseSequential:
1224 if (v->val.type == Tlabel)
1225 v->scope = PendingScope;
1228 v->scope = OutScope;
1231 /* There was no 'else', so we can only become
1232 * conditional if we know the cases were exhaustive,
1233 * and that doesn't mean anything yet.
1234 * So only labels become conditional..
1237 v2 && v2->scope == PendingScope;
1239 if (v2->val.type == Tlabel) {
1240 v2->scope = CondScope;
1241 v2->min_depth = c->scope_depth;
1243 v2->scope = OutScope;
1246 case OutScope: break;
1250 if (v->scope == OutScope)
1259 Executables can be lots of different things. In many cases an
1260 executable is just an operation combined with one or two other
1261 executables. This allows for expressions and lists etc. Other times
1262 an executable is something quite specific like a constant or variable
1263 name. So we define a `struct exec` to be a general executable with a
1264 type, and a `struct binode` which is a subclass of `exec`, forms a
1265 node in a binary tree, and holds an operation. There will be other
1266 subclasses, and to access these we need to be able to `cast` the
1267 `exec` into the various other types.
1270 #define cast(structname, pointer) ({ \
1271 const typeof( ((struct structname *)0)->type) *__mptr = &(pointer)->type; \
1272 if (__mptr && *__mptr != X##structname) abort(); \
1273 (struct structname *)( (char *)__mptr);})
1275 #define new(structname) ({ \
1276 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1277 __ptr->type = X##structname; \
1278 __ptr->line = -1; __ptr->column = -1; \
1281 #define new_pos(structname, token) ({ \
1282 struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
1283 __ptr->type = X##structname; \
1284 __ptr->line = token.line; __ptr->column = token.col; \
1293 enum exec_types type;
1301 struct exec *left, *right;
1304 ###### ast functions
1306 static int __fput_loc(struct exec *loc, FILE *f)
1308 if (loc->line >= 0) {
1309 fprintf(f, "%d:%d: ", loc->line, loc->column);
1312 if (loc->type == Xbinode)
1313 return __fput_loc(cast(binode,loc)->left, f) ||
1314 __fput_loc(cast(binode,loc)->right, f);
1317 static void fput_loc(struct exec *loc, FILE *f)
1319 if (!__fput_loc(loc, f))
1320 fprintf(f, "??:??: ");
1323 Each different type of `exec` node needs a number of functions
1324 defined, a bit like methods. We must be able to be able to free it,
1325 print it, analyse it and execute it. Once we have specific `exec`
1326 types we will need to parse them too. Let's take this a bit more
1331 The parser generator requires a `free_foo` function for each struct
1332 that stores attributes and they will be `exec`s and subtypes there-of.
1333 So we need `free_exec` which can handle all the subtypes, and we need
1336 ###### ast functions
1338 static void free_binode(struct binode *b)
1343 free_exec(b->right);
1347 ###### core functions
1348 static void free_exec(struct exec *e)
1357 ###### forward decls
1359 static void free_exec(struct exec *e);
1361 ###### free exec cases
1362 case Xbinode: free_binode(cast(binode, e)); break;
1366 Printing an `exec` requires that we know the current indent level for
1367 printing line-oriented components. As will become clear later, we
1368 also want to know what sort of bracketing to use.
1370 ###### ast functions
1372 static void do_indent(int i, char *str)
1379 ###### core functions
1380 static void print_binode(struct binode *b, int indent, int bracket)
1384 ## print binode cases
1388 static void print_exec(struct exec *e, int indent, int bracket)
1394 print_binode(cast(binode, e), indent, bracket); break;
1399 ###### forward decls
1401 static void print_exec(struct exec *e, int indent, int bracket);
1405 As discussed, analysis involves propagating type requirements around
1406 the program and looking for errors.
1408 So `propagate_types` is passed an expected type (being a `struct type`
1409 pointer together with some `val_rules` flags) that the `exec` is
1410 expected to return, and returns the type that it does return, either
1411 of which can be `NULL` signifying "unknown". An `ok` flag is passed
1412 by reference. It is set to `0` when an error is found, and `2` when
1413 any change is made. If it remains unchanged at `1`, then no more
1414 propagation is needed.
1418 enum val_rules {Rnolabel = 1<<0, Rboolok = 1<<1, Rnoconstant = 2<<1};
1422 if (rules & Rnolabel)
1423 fputs(" (labels not permitted)", stderr);
1426 ###### core functions
1428 static struct type *propagate_types(struct exec *prog, struct parse_context *c, int *ok,
1429 struct type *type, int rules)
1436 switch (prog->type) {
1439 struct binode *b = cast(binode, prog);
1441 ## propagate binode cases
1445 ## propagate exec cases
1452 Interpreting an `exec` doesn't require anything but the `exec`. State
1453 is stored in variables and each variable will be directly linked from
1454 within the `exec` tree. The exception to this is the whole `program`
1455 which needs to look at command line arguments. The `program` will be
1456 interpreted separately.
1458 Each `exec` can return a value, which may be `Tnone` but must be non-NULL;
1460 ###### core functions
1463 struct value val, *lval;
1466 static struct lrval _interp_exec(struct exec *e);
1468 static struct value interp_exec(struct exec *e)
1470 struct lrval ret = _interp_exec(e);
1473 return dup_value(*ret.lval);
1478 static struct value *linterp_exec(struct exec *e)
1480 struct lrval ret = _interp_exec(e);
1485 static struct lrval _interp_exec(struct exec *e)
1488 struct value rv, *lrv = NULL;
1499 struct binode *b = cast(binode, e);
1500 struct value left, right, *lleft;
1501 left.type = right.type = Tnone;
1503 ## interp binode cases
1505 free_value(left); free_value(right);
1508 ## interp exec cases
1515 ## Language elements
1517 Each language element needs to be parsed, printed, analysed,
1518 interpreted, and freed. There are several, so let's just start with
1519 the easy ones and work our way up.
1523 We have already met values as separate objects. When manifest
1524 constants appear in the program text, that must result in an executable
1525 which has a constant value. So the `val` structure embeds a value in
1541 $0 = new_pos(val, $1);
1542 $0->val.type = Tbool;
1546 $0 = new_pos(val, $1);
1547 $0->val.type = Tbool;
1551 $0 = new_pos(val, $1);
1552 $0->val.type = Tnum;
1555 if (number_parse($0->val.num, tail, $1.txt) == 0)
1556 mpq_init($0->val.num);
1558 tok_err(config2context(config), "error: unsupported number suffix",
1563 $0 = new_pos(val, $1);
1564 $0->val.type = Tstr;
1567 string_parse(&$1, '\\', &$0->val.str, tail);
1569 tok_err(config2context(config), "error: unsupported string 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 ###### print exec cases
1588 struct val *v = cast(val, e);
1589 if (v->val.type == Tstr)
1591 print_value(v->val);
1592 if (v->val.type == Tstr)
1597 ###### propagate exec cases
1600 struct val *val = cast(val, prog);
1601 if (!type_compat(type, val->val.type, rules)) {
1602 type_err(c, "error: expected %1%r found %2",
1603 prog, type, rules, val->val.type);
1606 return val->val.type;
1609 ###### interp exec cases
1611 rv = dup_value(cast(val, e)->val);
1614 ###### ast functions
1615 static void free_val(struct val *v)
1623 ###### free exec cases
1624 case Xval: free_val(cast(val, e)); break;
1626 ###### ast functions
1627 // Move all nodes from 'b' to 'rv', reversing the order.
1628 // In 'b' 'left' is a list, and 'right' is the last node.
1629 // In 'rv', left' is the first node and 'right' is a list.
1630 static struct binode *reorder_bilist(struct binode *b)
1632 struct binode *rv = NULL;
1635 struct exec *t = b->right;
1639 b = cast(binode, b->left);
1649 Just as we used a `val` to wrap a value into an `exec`, we similarly
1650 need a `var` to wrap a `variable` into an exec. While each `val`
1651 contained a copy of the value, each `var` hold a link to the variable
1652 because it really is the same variable no matter where it appears.
1653 When a variable is used, we need to remember to follow the `->merged`
1654 link to find the primary instance.
1662 struct variable *var;
1668 VariableDecl -> IDENTIFIER : ${ {
1669 struct variable *v = var_decl(config2context(config), $1.txt);
1670 $0 = new_pos(var, $1);
1675 v = var_ref(config2context(config), $1.txt);
1677 type_err(config2context(config), "error: variable '%v' redeclared",
1678 $0, Tnone, 0, Tnone);
1679 type_err(config2context(config), "info: this is where '%v' was first declared",
1680 v->where_decl, Tnone, 0, Tnone);
1683 | IDENTIFIER :: ${ {
1684 struct variable *v = var_decl(config2context(config), $1.txt);
1685 $0 = new_pos(var, $1);
1691 v = var_ref(config2context(config), $1.txt);
1693 type_err(config2context(config), "error: variable '%v' redeclared",
1694 $0, Tnone, 0, Tnone);
1695 type_err(config2context(config), "info: this is where '%v' was first declared",
1696 v->where_decl, Tnone, 0, Tnone);
1699 | IDENTIFIER : Type ${ {
1700 struct variable *v = var_decl(config2context(config), $1.txt);
1701 $0 = new_pos(var, $1);
1706 v->val = val_prepare($<3);
1708 v = var_ref(config2context(config), $1.txt);
1710 type_err(config2context(config), "error: variable '%v' redeclared",
1711 $0, Tnone, 0, Tnone);
1712 type_err(config2context(config), "info: this is where '%v' was first declared",
1713 v->where_decl, Tnone, 0, Tnone);
1716 | IDENTIFIER :: Type ${ {
1717 struct variable *v = var_decl(config2context(config), $1.txt);
1718 $0 = new_pos(var, $1);
1723 v->val = val_prepare($<3);
1726 v = var_ref(config2context(config), $1.txt);
1728 type_err(config2context(config), "error: variable '%v' redeclared",
1729 $0, Tnone, 0, Tnone);
1730 type_err(config2context(config), "info: this is where '%v' was first declared",
1731 v->where_decl, Tnone, 0, Tnone);
1735 Variable -> IDENTIFIER ${ {
1736 struct variable *v = var_ref(config2context(config), $1.txt);
1737 $0 = new_pos(var, $1);
1739 /* This might be a label - allocate a var just in case */
1740 v = var_decl(config2context(config), $1.txt);
1742 v->val = val_prepare(Tlabel);
1743 v->val.label = &v->val;
1751 Type -> IDENTIFIER ${
1752 $0 = find_type(config2context(config), $1.txt);
1754 tok_err(config2context(config),
1755 "error: undefined type", &$1);
1761 ###### print exec cases
1764 struct var *v = cast(var, e);
1766 struct binding *b = v->var->name;
1767 printf("%.*s", b->name.len, b->name.txt);
1774 if (loc->type == Xvar) {
1775 struct var *v = cast(var, loc);
1777 struct binding *b = v->var->name;
1778 fprintf(stderr, "%.*s", b->name.len, b->name.txt);
1780 fputs("???", stderr);
1782 fputs("NOTVAR", stderr);
1785 ###### propagate exec cases
1789 struct var *var = cast(var, prog);
1790 struct variable *v = var->var;
1792 type_err(c, "%d:BUG: no variable!!", prog, Tnone, 0, Tnone);
1798 if (v->constant && (rules & Rnoconstant)) {
1799 type_err(c, "error: Cannot assign to a constant: %v",
1800 prog, NULL, 0, NULL);
1801 type_err(c, "info: name was defined as a constant here",
1802 v->where_decl, NULL, 0, NULL);
1806 if (v->val.type == NULL) {
1807 if (type && *ok != 0) {
1808 v->val = val_prepare(type);
1809 v->where_set = prog;
1814 if (!type_compat(type, v->val.type, rules)) {
1815 type_err(c, "error: expected %1%r but variable '%v' is %2", prog,
1816 type, rules, v->val.type);
1817 type_err(c, "info: this is where '%v' was set to %1", v->where_set,
1818 v->val.type, rules, Tnone);
1826 ###### interp exec cases
1829 struct var *var = cast(var, e);
1830 struct variable *v = var->var;
1838 ###### ast functions
1840 static void free_var(struct var *v)
1845 ###### free exec cases
1846 case Xvar: free_var(cast(var, e)); break;
1848 ### Expressions: Boolean
1850 Our first user of the `binode` will be expressions, and particularly
1851 Boolean expressions. As I haven't implemented precedence in the
1852 parser generator yet, we need different names for each precedence
1853 level used by expressions. The outer most or lowest level precedence
1854 are Boolean `or` `and`, and `not` which form an `Expression` out of `BTerm`s
1865 Expression -> Expression or BTerm ${ {
1866 struct binode *b = new(binode);
1872 | BTerm ${ $0 = $<1; }$
1874 BTerm -> BTerm and BFact ${ {
1875 struct binode *b = new(binode);
1881 | BFact ${ $0 = $<1; }$
1883 BFact -> not BFact ${ {
1884 struct binode *b = new(binode);
1891 ###### print binode cases
1893 print_exec(b->left, -1, 0);
1895 print_exec(b->right, -1, 0);
1898 print_exec(b->left, -1, 0);
1900 print_exec(b->right, -1, 0);
1904 print_exec(b->right, -1, 0);
1907 ###### propagate binode cases
1911 /* both must be Tbool, result is Tbool */
1912 propagate_types(b->left, c, ok, Tbool, 0);
1913 propagate_types(b->right, c, ok, Tbool, 0);
1914 if (type && type != Tbool) {
1915 type_err(c, "error: %1 operation found where %2 expected", prog,
1921 ###### interp binode cases
1923 rv = interp_exec(b->left);
1924 right = interp_exec(b->right);
1925 rv.bool = rv.bool && right.bool;
1928 rv = interp_exec(b->left);
1929 right = interp_exec(b->right);
1930 rv.bool = rv.bool || right.bool;
1933 rv = interp_exec(b->right);
1937 ### Expressions: Comparison
1939 Of slightly higher precedence that Boolean expressions are
1941 A comparison takes arguments of any type, but the two types must be
1944 To simplify the parsing we introduce an `eop` which can record an
1945 expression operator.
1952 ###### ast functions
1953 static void free_eop(struct eop *e)
1968 | Expr CMPop Expr ${ {
1969 struct binode *b = new(binode);
1975 | Expr ${ $0 = $<1; }$
1980 CMPop -> < ${ $0.op = Less; }$
1981 | > ${ $0.op = Gtr; }$
1982 | <= ${ $0.op = LessEq; }$
1983 | >= ${ $0.op = GtrEq; }$
1984 | == ${ $0.op = Eql; }$
1985 | != ${ $0.op = NEql; }$
1987 ###### print binode cases
1995 print_exec(b->left, -1, 0);
1997 case Less: printf(" < "); break;
1998 case LessEq: printf(" <= "); break;
1999 case Gtr: printf(" > "); break;
2000 case GtrEq: printf(" >= "); break;
2001 case Eql: printf(" == "); break;
2002 case NEql: printf(" != "); break;
2005 print_exec(b->right, -1, 0);
2008 ###### propagate binode cases
2015 /* Both must match but not be labels, result is Tbool */
2016 t = propagate_types(b->left, c, ok, NULL, Rnolabel);
2018 propagate_types(b->right, c, ok, t, 0);
2020 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2022 t = propagate_types(b->left, c, ok, t, 0);
2024 if (!type_compat(type, Tbool, 0)) {
2025 type_err(c, "error: Comparison returns %1 but %2 expected", prog,
2026 Tbool, rules, type);
2031 ###### interp binode cases
2040 left = interp_exec(b->left);
2041 right = interp_exec(b->right);
2042 cmp = value_cmp(left, right);
2045 case Less: rv.bool = cmp < 0; break;
2046 case LessEq: rv.bool = cmp <= 0; break;
2047 case Gtr: rv.bool = cmp > 0; break;
2048 case GtrEq: rv.bool = cmp >= 0; break;
2049 case Eql: rv.bool = cmp == 0; break;
2050 case NEql: rv.bool = cmp != 0; break;
2051 default: rv.bool = 0; break;
2056 ### Expressions: The rest
2058 The remaining expressions with the highest precedence are arithmetic
2059 and string concatenation. They are `Expr`, `Term`, and `Factor`.
2060 The `Factor` is where the `Value` and `Variable` that we already have
2063 `+` and `-` are both infix and prefix operations (where they are
2064 absolute value and negation). These have different operator names.
2066 We also have a 'Bracket' operator which records where parentheses were
2067 found. This makes it easy to reproduce these when printing. Once
2068 precedence is handled better I might be able to discard this.
2080 Expr -> Expr Eop Term ${ {
2081 struct binode *b = new(binode);
2087 | Term ${ $0 = $<1; }$
2089 Term -> Term Top Factor ${ {
2090 struct binode *b = new(binode);
2096 | Factor ${ $0 = $<1; }$
2098 Factor -> ( Expression ) ${ {
2099 struct binode *b = new_pos(binode, $1);
2105 struct binode *b = new(binode);
2110 | Value ${ $0 = $<1; }$
2111 | Variable ${ $0 = $<1; }$
2114 Eop -> + ${ $0.op = Plus; }$
2115 | - ${ $0.op = Minus; }$
2117 Uop -> + ${ $0.op = Absolute; }$
2118 | - ${ $0.op = Negate; }$
2120 Top -> * ${ $0.op = Times; }$
2121 | / ${ $0.op = Divide; }$
2122 | % ${ $0.op = Rem; }$
2123 | ++ ${ $0.op = Concat; }$
2125 ###### print binode cases
2132 print_exec(b->left, indent, 0);
2134 case Plus: fputs(" + ", stdout); break;
2135 case Minus: fputs(" - ", stdout); break;
2136 case Times: fputs(" * ", stdout); break;
2137 case Divide: fputs(" / ", stdout); break;
2138 case Rem: fputs(" % ", stdout); break;
2139 case Concat: fputs(" ++ ", stdout); break;
2142 print_exec(b->right, indent, 0);
2146 print_exec(b->right, indent, 0);
2150 print_exec(b->right, indent, 0);
2154 print_exec(b->right, indent, 0);
2158 ###### propagate binode cases
2164 /* both must be numbers, result is Tnum */
2167 /* as propagate_types ignores a NULL,
2168 * unary ops fit here too */
2169 propagate_types(b->left, c, ok, Tnum, 0);
2170 propagate_types(b->right, c, ok, Tnum, 0);
2171 if (!type_compat(type, Tnum, 0)) {
2172 type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
2179 /* both must be Tstr, result is Tstr */
2180 propagate_types(b->left, c, ok, Tstr, 0);
2181 propagate_types(b->right, c, ok, Tstr, 0);
2182 if (!type_compat(type, Tstr, 0)) {
2183 type_err(c, "error: Concat returns %1 but %2 expected", prog,
2190 return propagate_types(b->right, c, ok, type, 0);
2192 ###### interp binode cases
2195 rv = interp_exec(b->left);
2196 right = interp_exec(b->right);
2197 mpq_add(rv.num, rv.num, right.num);
2200 rv = interp_exec(b->left);
2201 right = interp_exec(b->right);
2202 mpq_sub(rv.num, rv.num, right.num);
2205 rv = interp_exec(b->left);
2206 right = interp_exec(b->right);
2207 mpq_mul(rv.num, rv.num, right.num);
2210 rv = interp_exec(b->left);
2211 right = interp_exec(b->right);
2212 mpq_div(rv.num, rv.num, right.num);
2217 left = interp_exec(b->left);
2218 right = interp_exec(b->right);
2219 mpz_init(l); mpz_init(r); mpz_init(rem);
2220 mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
2221 mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
2222 mpz_tdiv_r(rem, l, r);
2223 rv = val_init(Tnum);
2224 mpq_set_z(rv.num, rem);
2225 mpz_clear(r); mpz_clear(l); mpz_clear(rem);
2229 rv = interp_exec(b->right);
2230 mpq_neg(rv.num, rv.num);
2233 rv = interp_exec(b->right);
2234 mpq_abs(rv.num, rv.num);
2237 rv = interp_exec(b->right);
2240 left = interp_exec(b->left);
2241 right = interp_exec(b->right);
2243 rv.str = text_join(left.str, right.str);
2247 ###### value functions
2249 static struct text text_join(struct text a, struct text b)
2252 rv.len = a.len + b.len;
2253 rv.txt = malloc(rv.len);
2254 memcpy(rv.txt, a.txt, a.len);
2255 memcpy(rv.txt+a.len, b.txt, b.len);
2260 ### Blocks, Statements, and Statement lists.
2262 Now that we have expressions out of the way we need to turn to
2263 statements. There are simple statements and more complex statements.
2264 Simple statements do not contain newlines, complex statements do.
2266 Statements often come in sequences and we have corresponding simple
2267 statement lists and complex statement lists.
2268 The former comprise only simple statements separated by semicolons.
2269 The later comprise complex statements and simple statement lists. They are
2270 separated by newlines. Thus the semicolon is only used to separate
2271 simple statements on the one line. This may be overly restrictive,
2272 but I'm not sure I ever want a complex statement to share a line with
2275 Note that a simple statement list can still use multiple lines if
2276 subsequent lines are indented, so
2278 ###### Example: wrapped simple statement list
2283 is a single simple statement list. This might allow room for
2284 confusion, so I'm not set on it yet.
2286 A simple statement list needs no extra syntax. A complex statement
2287 list has two syntactic forms. It can be enclosed in braces (much like
2288 C blocks), or it can be introduced by a colon and continue until an
2289 unindented newline (much like Python blocks). With this extra syntax
2290 it is referred to as a block.
2292 Note that a block does not have to include any newlines if it only
2293 contains simple statements. So both of:
2295 if condition: a=b; d=f
2297 if condition { a=b; print f }
2301 In either case the list is constructed from a `binode` list with
2302 `Block` as the operator. When parsing the list it is most convenient
2303 to append to the end, so a list is a list and a statement. When using
2304 the list it is more convenient to consider a list to be a statement
2305 and a list. So we need a function to re-order a list.
2306 `reorder_bilist` serves this purpose.
2308 The only stand-alone statement we introduce at this stage is `pass`
2309 which does nothing and is represented as a `NULL` pointer in a `Block`
2310 list. Other stand-alone statements will follow once the infrastructure
2330 Block -> Open Statementlist Close ${ $0 = $<2; }$
2331 | Open Newlines Statementlist Close ${ $0 = $<3; }$
2332 | Open SimpleStatements } ${ $0 = reorder_bilist($<2); }$
2333 | Open Newlines SimpleStatements } ${ $0 = reorder_bilist($<3); }$
2334 | : Statementlist ${ $0 = $<2; }$
2335 | : SimpleStatements ${ $0 = reorder_bilist($<2); }$
2337 Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<1); }$
2339 ComplexStatements -> ComplexStatements ComplexStatement ${
2345 | ComplexStatements NEWLINE ${ $0 = $<1; }$
2346 | ComplexStatement ${
2354 ComplexStatement -> SimpleStatements NEWLINE ${
2355 $0 = reorder_bilist($<1);
2357 ## ComplexStatement Grammar
2360 SimpleStatements -> SimpleStatements ; SimpleStatement ${
2366 | SimpleStatement ${
2372 | SimpleStatements ; ${ $0 = $<1; }$
2374 SimpleStatement -> pass ${ $0 = NULL; }$
2375 ## SimpleStatement Grammar
2377 ###### print binode cases
2381 if (b->left == NULL)
2384 print_exec(b->left, indent, 0);
2387 print_exec(b->right, indent, 0);
2390 // block, one per line
2391 if (b->left == NULL)
2392 do_indent(indent, "pass\n");
2394 print_exec(b->left, indent, bracket);
2396 print_exec(b->right, indent, bracket);
2400 ###### propagate binode cases
2403 /* If any statement returns something other than Tnone
2404 * or Tbool then all such must return same type.
2405 * As each statement may be Tnone or something else,
2406 * we must always pass NULL (unknown) down, otherwise an incorrect
2407 * error might occur. We never return Tnone unless it is
2412 for (e = b; e; e = cast(binode, e->right)) {
2413 t = propagate_types(e->left, c, ok, NULL, rules);
2414 if ((rules & Rboolok) && t == Tbool)
2416 if (t && t != Tnone && t != Tbool) {
2419 else if (t != type) {
2420 type_err(c, "error: expected %1%r, found %2",
2421 e->left, type, rules, t);
2429 ###### interp binode cases
2431 while (rv.type == Tnone &&
2434 rv = interp_exec(b->left);
2435 b = cast(binode, b->right);
2439 ### The Print statement
2441 `print` is a simple statement that takes a comma-separated list of
2442 expressions and prints the values separated by spaces and terminated
2443 by a newline. No control of formatting is possible.
2445 `print` faces the same list-ordering issue as blocks, and uses the
2451 ###### SimpleStatement Grammar
2453 | print ExpressionList ${
2454 $0 = reorder_bilist($<2);
2456 | print ExpressionList , ${
2461 $0 = reorder_bilist($0);
2472 ExpressionList -> ExpressionList , Expression ${
2485 ###### print binode cases
2488 do_indent(indent, "print");
2492 print_exec(b->left, -1, 0);
2496 b = cast(binode, b->right);
2502 ###### propagate binode cases
2505 /* don't care but all must be consistent */
2506 propagate_types(b->left, c, ok, NULL, Rnolabel);
2507 propagate_types(b->right, c, ok, NULL, Rnolabel);
2510 ###### interp binode cases
2516 for ( ; b; b = cast(binode, b->right))
2520 left = interp_exec(b->left);
2533 ###### Assignment statement
2535 An assignment will assign a value to a variable, providing it hasn't
2536 be declared as a constant. The analysis phase ensures that the type
2537 will be correct so the interpreter just needs to perform the
2538 calculation. There is a form of assignment which declares a new
2539 variable as well as assigning a value. If a name is assigned before
2540 it is declared, and error will be raised as the name is created as
2541 `Tlabel` and it is illegal to assign to such names.
2547 ###### SimpleStatement Grammar
2548 | Variable = Expression ${
2554 | VariableDecl = Expression ${
2562 if ($1->var->where_set == NULL) {
2563 type_err(config2context(config), "Variable declared with no type or value: %v",
2573 ###### print binode cases
2576 do_indent(indent, "");
2577 print_exec(b->left, indent, 0);
2579 print_exec(b->right, indent, 0);
2586 struct variable *v = cast(var, b->left)->var;
2587 do_indent(indent, "");
2588 print_exec(b->left, indent, 0);
2589 if (cast(var, b->left)->var->constant) {
2590 if (v->where_decl == v->where_set) {
2592 type_print(v->val.type, stdout);
2597 if (v->where_decl == v->where_set) {
2599 type_print(v->val.type, stdout);
2606 print_exec(b->right, indent, 0);
2613 ###### propagate binode cases
2617 /* Both must match and not be labels,
2618 * Type must support 'dup',
2619 * For Assign, left must not be constant.
2622 t = propagate_types(b->left, c, ok, NULL,
2623 Rnolabel | (b->op == Assign ? Rnoconstant : 0));
2628 if (propagate_types(b->right, c, ok, t, 0) != t)
2629 if (b->left->type == Xvar)
2630 type_err(c, "info: variable '%v' was set as %1 here.",
2631 cast(var, b->left)->var->where_set, t, rules, Tnone);
2633 t = propagate_types(b->right, c, ok, NULL, Rnolabel);
2635 propagate_types(b->left, c, ok, t,
2636 (b->op == Assign ? Rnoconstant : 0));
2638 if (t && t->dup == NULL) {
2639 type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
2646 ###### interp binode cases
2649 lleft = linterp_exec(b->left);
2650 right = interp_exec(b->right);
2661 struct variable *v = cast(var, b->left)->var;
2665 right = interp_exec(b->right);
2667 right = val_init(v->val.type);
2674 ### The `use` statement
2676 The `use` statement is the last "simple" statement. It is needed when
2677 the condition in a conditional statement is a block. `use` works much
2678 like `return` in C, but only completes the `condition`, not the whole
2684 ###### SimpleStatement Grammar
2686 $0 = new_pos(binode, $1);
2691 ###### print binode cases
2694 do_indent(indent, "use ");
2695 print_exec(b->right, -1, 0);
2700 ###### propagate binode cases
2703 /* result matches value */
2704 return propagate_types(b->right, c, ok, type, 0);
2706 ###### interp binode cases
2709 rv = interp_exec(b->right);
2712 ### The Conditional Statement
2714 This is the biggy and currently the only complex statement. This
2715 subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
2716 It is comprised of a number of parts, all of which are optional though
2717 set combinations apply. Each part is (usually) a key word (`then` is
2718 sometimes optional) followed by either an expression or a code block,
2719 except the `casepart` which is a "key word and an expression" followed
2720 by a code block. The code-block option is valid for all parts and,
2721 where an expression is also allowed, the code block can use the `use`
2722 statement to report a value. If the code block does not report a value
2723 the effect is similar to reporting `True`.
2725 The `else` and `case` parts, as well as `then` when combined with
2726 `if`, can contain a `use` statement which will apply to some
2727 containing conditional statement. `for` parts, `do` parts and `then`
2728 parts used with `for` can never contain a `use`, except in some
2729 subordinate conditional statement.
2731 If there is a `forpart`, it is executed first, only once.
2732 If there is a `dopart`, then it is executed repeatedly providing
2733 always that the `condpart` or `cond`, if present, does not return a non-True
2734 value. `condpart` can fail to return any value if it simply executes
2735 to completion. This is treated the same as returning `True`.
2737 If there is a `thenpart` it will be executed whenever the `condpart`
2738 or `cond` returns True (or does not return any value), but this will happen
2739 *after* `dopart` (when present).
2741 If `elsepart` is present it will be executed at most once when the
2742 condition returns `False` or some value that isn't `True` and isn't
2743 matched by any `casepart`. If there are any `casepart`s, they will be
2744 executed when the condition returns a matching value.
2746 The particular sorts of values allowed in case parts has not yet been
2747 determined in the language design, so nothing is prohibited.
2749 The various blocks in this complex statement potentially provide scope
2750 for variables as described earlier. Each such block must include the
2751 "OpenScope" nonterminal before parsing the block, and must call
2752 `var_block_close()` when closing the block.
2754 The code following "`if`", "`switch`" and "`for`" does not get its own
2755 scope, but is in a scope covering the whole statement, so names
2756 declared there cannot be redeclared elsewhere. Similarly the
2757 condition following "`while`" is in a scope the covers the body
2758 ("`do`" part) of the loop, and which does not allow conditional scope
2759 extension. Code following "`then`" (both looping and non-looping),
2760 "`else`" and "`case`" each get their own local scope.
2762 The type requirements on the code block in a `whilepart` are quite
2763 unusal. It is allowed to return a value of some identifiable type, in
2764 which case the loop aborts and an appropriate `casepart` is run, or it
2765 can return a Boolean, in which case the loop either continues to the
2766 `dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
2767 This is different both from the `ifpart` code block which is expected to
2768 return a Boolean, or the `switchpart` code block which is expected to
2769 return the same type as the casepart values. The correct analysis of
2770 the type of the `whilepart` code block is the reason for the
2771 `Rboolok` flag which is passed to `propagate_types()`.
2773 The `cond_statement` cannot fit into a `binode` so a new `exec` is
2782 struct exec *action;
2783 struct casepart *next;
2785 struct cond_statement {
2787 struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
2788 struct casepart *casepart;
2791 ###### ast functions
2793 static void free_casepart(struct casepart *cp)
2797 free_exec(cp->value);
2798 free_exec(cp->action);
2805 static void free_cond_statement(struct cond_statement *s)
2809 free_exec(s->forpart);
2810 free_exec(s->condpart);
2811 free_exec(s->dopart);
2812 free_exec(s->thenpart);
2813 free_exec(s->elsepart);
2814 free_casepart(s->casepart);
2818 ###### free exec cases
2819 case Xcond_statement: free_cond_statement(cast(cond_statement, e)); break;
2821 ###### ComplexStatement Grammar
2822 | CondStatement ${ $0 = $<1; }$
2827 // both ForThen and Whilepart open scopes, and CondSuffix only
2828 // closes one - so in the first branch here we have another to close.
2829 CondStatement -> ForThen WhilePart CondSuffix ${
2831 $0->forpart = $1.forpart; $1.forpart = NULL;
2832 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
2833 $0->condpart = $2.condpart; $2.condpart = NULL;
2834 $0->dopart = $2.dopart; $2.dopart = NULL;
2835 var_block_close(config2context(config), CloseSequential);
2837 | WhilePart CondSuffix ${
2839 $0->condpart = $1.condpart; $1.condpart = NULL;
2840 $0->dopart = $1.dopart; $1.dopart = NULL;
2842 | SwitchPart CondSuffix ${
2846 | IfPart IfSuffix ${
2848 $0->condpart = $1.condpart; $1.condpart = NULL;
2849 $0->thenpart = $1.thenpart; $1.thenpart = NULL;
2850 // This is where we close an "if" statement
2851 var_block_close(config2context(config), CloseSequential);
2854 CondSuffix -> IfSuffix ${
2856 // This is where we close scope of the whole
2857 // "for" or "while" statement
2858 var_block_close(config2context(config), CloseSequential);
2860 | CasePart CondSuffix ${
2862 $1->next = $0->casepart;
2867 CasePart -> Newlines case Expression OpenScope Block ${
2868 $0 = calloc(1,sizeof(struct casepart));
2871 var_block_close(config2context(config), CloseParallel);
2873 | case Expression OpenScope Block ${
2874 $0 = calloc(1,sizeof(struct casepart));
2877 var_block_close(config2context(config), CloseParallel);
2881 IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
2882 | Newlines else OpenScope Block ${
2883 $0 = new(cond_statement);
2885 var_block_close(config2context(config), CloseElse);
2887 | else OpenScope Block ${
2888 $0 = new(cond_statement);
2890 var_block_close(config2context(config), CloseElse);
2892 | Newlines else OpenScope CondStatement ${
2893 $0 = new(cond_statement);
2895 var_block_close(config2context(config), CloseElse);
2897 | else OpenScope CondStatement ${
2898 $0 = new(cond_statement);
2900 var_block_close(config2context(config), CloseElse);
2905 // These scopes are closed in CondSuffix
2906 ForPart -> for OpenScope SimpleStatements ${
2907 $0 = reorder_bilist($<3);
2909 | for OpenScope Block ${
2913 ThenPart -> then OpenScope SimpleStatements ${
2914 $0 = reorder_bilist($<3);
2915 var_block_close(config2context(config), CloseSequential);
2917 | then OpenScope Block ${
2919 var_block_close(config2context(config), CloseSequential);
2922 ThenPartNL -> ThenPart OptNL ${
2926 // This scope is closed in CondSuffix
2927 WhileHead -> while OpenScope Block ${
2932 ForThen -> ForPart OptNL ThenPartNL ${
2940 // This scope is closed in CondSuffix
2941 WhilePart -> while OpenScope Expression Block ${
2942 $0.type = Xcond_statement;
2946 | WhileHead OptNL do Block ${
2947 $0.type = Xcond_statement;
2952 IfPart -> if OpenScope Expression OpenScope Block ${
2953 $0.type = Xcond_statement;
2956 var_block_close(config2context(config), CloseParallel);
2958 | if OpenScope Block OptNL then OpenScope Block ${
2959 $0.type = Xcond_statement;
2962 var_block_close(config2context(config), CloseParallel);
2966 // This scope is closed in CondSuffix
2967 SwitchPart -> switch OpenScope Expression ${
2970 | switch OpenScope Block ${
2974 ###### print exec cases
2976 case Xcond_statement:
2978 struct cond_statement *cs = cast(cond_statement, e);
2979 struct casepart *cp;
2981 do_indent(indent, "for");
2982 if (bracket) printf(" {\n"); else printf(":\n");
2983 print_exec(cs->forpart, indent+1, bracket);
2986 do_indent(indent, "} then {\n");
2988 do_indent(indent, "then:\n");
2989 print_exec(cs->thenpart, indent+1, bracket);
2991 if (bracket) do_indent(indent, "}\n");
2995 if (cs->condpart && cs->condpart->type == Xbinode &&
2996 cast(binode, cs->condpart)->op == Block) {
2998 do_indent(indent, "while {\n");
3000 do_indent(indent, "while:\n");
3001 print_exec(cs->condpart, indent+1, bracket);
3003 do_indent(indent, "} do {\n");
3005 do_indent(indent, "do:\n");
3006 print_exec(cs->dopart, indent+1, bracket);
3008 do_indent(indent, "}\n");
3010 do_indent(indent, "while ");
3011 print_exec(cs->condpart, 0, bracket);
3016 print_exec(cs->dopart, indent+1, bracket);
3018 do_indent(indent, "}\n");
3023 do_indent(indent, "switch");
3025 do_indent(indent, "if");
3026 if (cs->condpart && cs->condpart->type == Xbinode &&
3027 cast(binode, cs->condpart)->op == Block) {
3032 print_exec(cs->condpart, indent+1, bracket);
3034 do_indent(indent, "}\n");
3036 do_indent(indent, "then:\n");
3037 print_exec(cs->thenpart, indent+1, bracket);
3041 print_exec(cs->condpart, 0, bracket);
3047 print_exec(cs->thenpart, indent+1, bracket);
3049 do_indent(indent, "}\n");
3054 for (cp = cs->casepart; cp; cp = cp->next) {
3055 do_indent(indent, "case ");
3056 print_exec(cp->value, -1, 0);
3061 print_exec(cp->action, indent+1, bracket);
3063 do_indent(indent, "}\n");
3066 do_indent(indent, "else");
3071 print_exec(cs->elsepart, indent+1, bracket);
3073 do_indent(indent, "}\n");
3078 ###### propagate exec cases
3079 case Xcond_statement:
3081 // forpart and dopart must return Tnone
3082 // thenpart must return Tnone if there is a dopart,
3083 // otherwise it is like elsepart.
3085 // be bool if there is no casepart
3086 // match casepart->values if there is a switchpart
3087 // either be bool or match casepart->value if there
3089 // elsepart and casepart->action must match the return type
3090 // expected of this statement.
3091 struct cond_statement *cs = cast(cond_statement, prog);
3092 struct casepart *cp;
3094 t = propagate_types(cs->forpart, c, ok, Tnone, 0);
3095 if (!type_compat(Tnone, t, 0))
3097 t = propagate_types(cs->dopart, c, ok, Tnone, 0);
3098 if (!type_compat(Tnone, t, 0))
3101 t = propagate_types(cs->thenpart, c, ok, Tnone, 0);
3102 if (!type_compat(Tnone, t, 0))
3105 if (cs->casepart == NULL)
3106 propagate_types(cs->condpart, c, ok, Tbool, 0);
3108 /* Condpart must match case values, with bool permitted */
3110 for (cp = cs->casepart;
3111 cp && !t; cp = cp->next)
3112 t = propagate_types(cp->value, c, ok, NULL, 0);
3113 if (!t && cs->condpart)
3114 t = propagate_types(cs->condpart, c, ok, NULL, Rboolok);
3115 // Now we have a type (I hope) push it down
3117 for (cp = cs->casepart; cp; cp = cp->next)
3118 propagate_types(cp->value, c, ok, t, 0);
3119 propagate_types(cs->condpart, c, ok, t, Rboolok);
3122 // (if)then, else, and case parts must return expected type.
3123 if (!cs->dopart && !type)
3124 type = propagate_types(cs->thenpart, c, ok, NULL, rules);
3126 type = propagate_types(cs->elsepart, c, ok, NULL, rules);
3127 for (cp = cs->casepart;
3130 type = propagate_types(cp->action, c, ok, NULL, rules);
3133 propagate_types(cs->thenpart, c, ok, type, rules);
3134 propagate_types(cs->elsepart, c, ok, type, rules);
3135 for (cp = cs->casepart; cp ; cp = cp->next)
3136 propagate_types(cp->action, c, ok, type, rules);
3142 ###### interp exec cases
3143 case Xcond_statement:
3145 struct value v, cnd;
3146 struct casepart *cp;
3147 struct cond_statement *c = cast(cond_statement, e);
3150 interp_exec(c->forpart);
3153 cnd = interp_exec(c->condpart);
3156 if (!(cnd.type == Tnone ||
3157 (cnd.type == Tbool && cnd.bool != 0)))
3159 // cnd is Tnone or Tbool, doesn't need to be freed
3161 interp_exec(c->dopart);
3164 rv = interp_exec(c->thenpart);
3165 if (rv.type != Tnone || !c->dopart)
3169 } while (c->dopart);
3171 for (cp = c->casepart; cp; cp = cp->next) {
3172 v = interp_exec(cp->value);
3173 if (value_cmp(v, cnd) == 0) {
3176 rv = interp_exec(cp->action);
3183 rv = interp_exec(c->elsepart);
3190 ### Finally the whole program.
3192 Somewhat reminiscent of Pascal a (current) Ocean program starts with
3193 the keyword "program" and a list of variable names which are assigned
3194 values from command line arguments. Following this is a `block` which
3195 is the code to execute.
3197 As this is the top level, several things are handled a bit
3199 The whole program is not interpreted by `interp_exec` as that isn't
3200 passed the argument list which the program requires. Similarly type
3201 analysis is a bit more interesting at this level.
3206 ###### Parser: grammar
3209 Program -> program OpenScope Varlist Block OptNL ${
3212 $0->left = reorder_bilist($<3);
3214 var_block_close(config2context(config), CloseSequential);
3215 if (config2context(config)->scope_stack) abort();
3218 tok_err(config2context(config),
3219 "error: unhandled parse error", &$1);
3222 Varlist -> Varlist ArgDecl ${
3231 ArgDecl -> IDENTIFIER ${ {
3232 struct variable *v = var_decl(config2context(config), $1.txt);
3239 ###### print binode cases
3241 do_indent(indent, "program");
3242 for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
3244 print_exec(b2->left, 0, 0);
3250 print_exec(b->right, indent+1, bracket);
3252 do_indent(indent, "}\n");
3255 ###### propagate binode cases
3256 case Program: abort();
3258 ###### core functions
3260 static int analyse_prog(struct exec *prog, struct parse_context *c)
3262 struct binode *b = cast(binode, prog);
3269 propagate_types(b->right, c, &ok, Tnone, 0);
3274 for (b = cast(binode, b->left); b; b = cast(binode, b->right)) {
3275 struct var *v = cast(var, b->left);
3276 if (!v->var->val.type) {
3277 v->var->where_set = b;
3278 v->var->val = val_prepare(Tstr);
3281 b = cast(binode, prog);
3284 propagate_types(b->right, c, &ok, Tnone, 0);
3289 /* Make sure everything is still consistent */
3290 propagate_types(b->right, c, &ok, Tnone, 0);
3294 static void interp_prog(struct exec *prog, char **argv)
3296 struct binode *p = cast(binode, prog);
3302 al = cast(binode, p->left);
3304 struct var *v = cast(var, al->left);
3305 struct value *vl = &v->var->val;
3307 if (argv[0] == NULL) {
3308 printf("Not enough args\n");
3311 al = cast(binode, al->right);
3313 *vl = parse_value(vl->type, argv[0]);
3314 if (vl->type == NULL)
3318 v = interp_exec(p->right);
3322 ###### interp binode cases
3323 case Program: abort();
3325 ## And now to test it out.
3327 Having a language requires having a "hello world" program. I'll
3328 provide a little more than that: a program that prints "Hello world"
3329 finds the GCD of two numbers, prints the first few elements of
3330 Fibonacci, and performs a binary search for a number.
3332 ###### File: oceani.mk
3335 @echo "===== TEST ====="
3336 ./oceani --section "test: hello" oceani.mdc 55 33
3341 print "Hello World, what lovely oceans you have!"
3342 /* When a variable is defined in both branches of an 'if',
3343 * and used afterwards, the variables are merged.
3349 print "Is", A, "bigger than", B,"? ", bigger
3350 /* If a variable is not used after the 'if', no
3351 * merge happens, so types can be different
3354 double:string = "yes"
3355 print A, "is more than twice", B, "?", double
3358 print "double", B, "is", double
3369 print "GCD of", A, "and", B,"is", a
3371 print a, "is not positive, cannot calculate GCD"
3373 print b, "is not positive, cannot calculate GCD"
3378 print "Fibonacci:", f1,f2,
3379 then togo = togo - 1
3387 /* Binary search... */
3392 mid := (lo + hi) / 2
3404 print "Yay, I found", target
3406 print "Closest I found was", mid