-# Ocean Interpreter - Falls Creek version
+# Ocean Interpreter - Jamison Creek version
-Ocean is intended to be an compiled language, so this interpreter is
-not targeted at being the final product. It is very much an intermediate
-stage, and fills that role in two distinct ways.
+Ocean is intended to be a compiled language, so this interpreter is
+not targeted at being the final product. It is, rather, an intermediate
+stage and fills that role in two distinct ways.
Firstly, it exists as a platform to experiment with the early language
design. An interpreter is easy to write and easy to get working, so
## Current version
-This initial version of the interpreter exists to test out the
-structured statement providing conditions and iteration. Clearly we
-need some minimal other functionality so that values can be tested and
-instructions iterated over. All that functionality is clearly not
-normative at this stage (not that anything is **really** normative
-yet) and will change, so early test code will certainly break.
-
-Beyond the structured statement and the `use` statement which is
-intimately related to it we have:
-
- - "blocks" of multiple statements.
- - `pass`: a statement which does nothing.
- - variables: any identifier is assumed to store a number, string,
- or Boolean.
- - expressions: `+`, `-`, `*`, `/` can apply to integers and `++` can
- catenate strings. `and`, `or`, `not` manipulate Booleans, and
- normal comparison operators can work on all three types.
- - assignments: can assign the value of an expression to a variable.
- - `print`: will print the values in a list of expressions.
- - `program`: is given a list of identifiers to initialize from
- arguments.
+This third version of the interpreter exists to test out some initial
+ideas relating to types. Particularly it adds arrays (indexed from
+zero) and simple structures. Basic control flow and variable scoping
+are already fairly well established, as are basic numerical and
+boolean operators.
+
+Some operators that have only recently been added, and so have not
+generated all that much experience yet are "and then" and "or else" as
+short-circuit Boolean operators, and the "if ... else" trinary
+operator which can select between two expressions based on a third
+(which appears syntactically in the middle).
+
+The "func" clause currently only allows a "main" function to be
+declared. That will be extended when proper function support is added.
+
+An element that is present purely to make a usable language, and
+without any expectation that they will remain, is the "print" statement
+which performs simple output.
+
+The current scalar types are "number", "Boolean", and "string".
+Boolean will likely stay in its current form, the other two might, but
+could just as easily be changed.
## Naming
Versions of the interpreter which obviously do not support a complete
-language will be named after creeks and streams. This one is Falls
+language will be named after creeks and streams. This one is Jamison
Creek.
Once we have something reasonably resembling a complete language, the
for validating the parsing.
So the main requirements of the interpreter are:
-- Parse the program
-- Analyse the parsed program to ensure consistency
-- print the program
-- execute the program
+- Parse the program, possibly with tracing,
+- Analyse the parsed program to ensure consistency,
+- Print the program,
+- Execute the "main" function in the program, if no parsing or
+ consistency errors were found.
This is all performed by a single C program extracted with
`parsergen`.
-There will be two formats for printing the program a default and one
-that uses bracketing. So an extra command line option is needed for
-that.
+There will be two formats for printing the program: a default and one
+that uses bracketing. So a `--bracket` command line option is needed
+for that. Normally the first code section found is used, however an
+alternate section can be requested so that a file (such as this one)
+can contain multiple programs. This is effected with the `--section`
+option.
+
+This code must be compiled with `-fplan9-extensions` so that anonymous
+structures can be used.
###### File: oceani.mk
myLDLIBS:= libparser.o libscanner.o libmdcode.o -licuuc
LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
## libs
- all :: oceani
+ all :: $(LDLIBS) oceani
oceani.c oceani.h : oceani.mdc parsergen
./parsergen -o oceani --LALR --tag Parser oceani.mdc
oceani.mk: oceani.mdc md2c
./md2c oceani.mdc
- oceani: oceani.o
+ oceani: oceani.o $(LDLIBS)
+ $(CC) $(CFLAGS) -o oceani oceani.o $(LDLIBS)
###### Parser: header
## macros
+ struct parse_context;
## ast
+ ## ast late
struct parse_context {
struct token_config config;
+ char *file_name;
+ int parse_error;
## parse context
};
-###### Parser: code
+###### macros
+
+ #define container_of(ptr, type, member) ({ \
+ const typeof( ((type *)0)->member ) *__mptr = (ptr); \
+ (type *)( (char *)__mptr - offsetof(type,member) );})
+
+ #define config2context(_conf) container_of(_conf, struct parse_context, \
+ config)
+###### Parser: reduce
+ struct parse_context *c = config2context(config);
+
+###### Parser: code
+ #define _GNU_SOURCE
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h>
## core functions
#include <getopt.h>
- static char Usage[] = "Usage: oceani --trace --print --noexec prog.ocn\n";
+ static char Usage[] =
+ "Usage: oceani --trace --print --noexec --brackets --section=SectionName prog.ocn\n";
static const struct option long_options[] = {
{"trace", 0, NULL, 't'},
{"print", 0, NULL, 'p'},
{"noexec", 0, NULL, 'n'},
{"brackets", 0, NULL, 'b'},
+ {"section", 1, NULL, 's'},
{NULL, 0, NULL, 0},
};
- const char *options = "tpnb";
+ const char *options = "tpnbs";
+
+ static void pr_err(char *msg) // NOTEST
+ {
+ fprintf(stderr, "%s\n", msg); // NOTEST
+ } // NOTEST
+
int main(int argc, char *argv[])
{
int fd;
int len;
char *file;
- struct section *s;
+ struct section *s = NULL, *ss;
+ char *section = NULL;
struct parse_context context = {
.config = {
- .ignored = (1 << TK_line_comment)
- | (1 << TK_block_comment),
- .number_chars = ".,_+-",
+ .ignored = (1 << TK_mark),
+ .number_chars = ".,_+- ",
.word_start = "_",
.word_cont = "_",
},
};
int doprint=0, dotrace=0, doexec=1, brackets=0;
- struct exec **prog;
int opt;
while ((opt = getopt_long(argc, argv, options, long_options, NULL))
!= -1) {
case 'p': doprint=1; break;
case 'n': doexec=0; break;
case 'b': brackets=1; break;
+ case 's': section = optarg; break;
default: fprintf(stderr, Usage);
exit(1);
}
fprintf(stderr, "oceani: cannot open %s\n", argv[optind]);
exit(1);
}
+ context.file_name = argv[optind];
len = lseek(fd, 0, 2);
file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
- s = code_extract(file, file+len, NULL);
+ s = code_extract(file, file+len, pr_err);
if (!s) {
fprintf(stderr, "oceani: could not find any code in %s\n",
argv[optind]);
exit(1);
}
- prog = parse_oceani(s->code, &context.config,
- dotrace ? stderr : NULL);
- if (prog && doprint)
- print_exec(*prog, 0, brackets);
- if (prog && doexec) {
- if (!analyse_prog(*prog, &context)) {
- fprintf(stderr, "oceani: type error in program\n");
- exit(1);
+
+ ## context initialization
+
+ if (section) {
+ for (ss = s; ss; ss = ss->next) {
+ struct text sec = ss->section;
+ if (sec.len == strlen(section) &&
+ strncmp(sec.txt, section, sec.len) == 0)
+ break;
}
- interp_prog(*prog, argv+optind+1);
+ if (!ss) {
+ fprintf(stderr, "oceani: cannot find section %s\n",
+ section);
+ goto cleanup;
+ }
+ } else
+ ss = s; // NOTEST
+ if (!ss->code) {
+ fprintf(stderr, "oceani: no code found in requested section\n"); // NOTEST
+ goto cleanup; // NOTEST
+ }
+
+ parse_oceani(ss->code, &context.config, dotrace ? stderr : NULL);
+
+ resolve_consts(&context);
+ prepare_types(&context);
+ if (!context.parse_error && !analyse_funcs(&context)) {
+ fprintf(stderr, "oceani: type error in program - not running.\n");
+ context.parse_error += 1;
}
- if (prog) {
- free_exec(*prog);
- free(prog);
+
+ if (doprint) {
+ ## print const decls
+ ## print type decls
+ ## print func decls
}
+ if (doexec && !context.parse_error)
+ interp_main(&context, argc - optind, argv + optind);
+ cleanup:
while (s) {
struct section *t = s->next;
code_free(s->code);
free(s);
s = t;
}
- ## free context
- exit(0);
+ // FIXME parser should pop scope even on error
+ while (context.scope_depth > 0)
+ scope_pop(&context);
+ ## free global vars
+ ## free const decls
+ ## free context types
+ ## free context storage
+ exit(context.parse_error ? 1 : 0);
}
### Analysis
-These four requirements of parse, analyse, print, interpret apply to
+The four requirements of parse, analyse, print, interpret apply to
each language element individually so that is how most of the code
will be structured.
Three of the four are fairly self explanatory. The one that requires
a little explanation is the analysis step.
-The current language design does not require variables to be declared,
-but they must have a single type. Different operations impose
-different requirements on the variables, for example addition requires
-both arguments to be numeric, and assignment requires the variable on
-the left to have the same type as the expression on the right.
+The current language design does not require the types of variables to
+be declared, but they must still have a single type. Different
+operations impose different requirements on the variables, for example
+addition requires both arguments to be numeric, and assignment
+requires the variable on the left to have the same type as the
+expression on the right.
-Analysis involves propagating these type requirements around
+Analysis involves propagating these type requirements around and
consequently setting the type of each variable. If any requirements
are violated (e.g. a string is compared with a number) or if a
variable needs to have two different types, then an error is raised
and the program will not run.
-Determining the types of all variables early is important for
-processing command line arguments. These can be assigned to any type
-of variable, but we must first know the correct type so any required
-conversion can happen. If a variable is associated with a command
-line argument but no type can be interpreted (e.g. the variable is
-only ever used in a `print` statement), then the type is set to
-'string'.
-
-If the type of a variable cannot be determined at all, then it is set
-to be a number and given a unique value. This allows it to fill the
-role of a name in an enumerated type, which is useful for testing the
-`switch` statement.
-
-## Data Structures
-
-One last introductory step before detailing the language elements and
-providing their four requirements is to establish the data structures
-to store these elements.
-
-There are two key objects that we need to work with: executable
-elements which comprise the program, and values which the program
-works with. Between these is the set of variables which hold the
-values.
-
-### Values
-
-Values can be numbers, which we represent as multi-precision
-fractions, strings and Booleans. When analysing the program we also
-need to allow for places where no value is meaningful (`Vnone`) and
-where we don't know what type to expect yet (`Vunknown`).
-A 2 character 'tail' is included in each value as the scanner wants
-to parse that from the end of numbers and we need somewhere to put
-it. It is currently ignored but one day might allow for
-e.g. "imaginary" numbers.
-
-Values are never shared, they are always copied when used, and freed
-when no longer needed.
-
-###### includes
- #include <gmp.h>
- #include "string.h"
- #include "number.h"
-
-###### libs
- myLDLIBS := libnumber.o libstring.o -lgmp
- LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
-
-###### ast
- struct value {
- enum vtype {Vunknown, Vnone, Vstr, Vnum, Vbool} vtype;
- union {
- struct text str;
- mpq_t num;
- int bool;
- };
- char tail[2];
- };
-
-###### ast functions
- void free_value(struct value v)
- {
- switch (v.vtype) {
- case Vnone:
- case Vunknown: break;
- case Vstr: free(v.str.txt); break;
- case Vnum: mpq_clear(v.num); break;
- case Vbool: break;
- }
- }
-
-###### value functions
-
- static void val_init(struct value *val, enum vtype type)
- {
- val->vtype = type;
- switch(type) {
- case Vnone:abort();
- case Vunknown: break;
- case Vnum:
- mpq_init(val->num); break;
- case Vstr:
- val->str.txt = malloc(1);
- val->str.len = 0;
- break;
- case Vbool:
- val->bool = 0;
- break;
- }
- }
-
- static struct value dup_value(struct value v)
- {
- struct value rv;
- rv.vtype = v.vtype;
- switch (rv.vtype) {
- case Vnone:
- case Vunknown: break;
- case Vbool:
- rv.bool = v.bool;
- break;
- case Vnum:
- mpq_init(rv.num);
- mpq_set(rv.num, v.num);
- break;
- case Vstr:
- rv.str.len = v.str.len;
- rv.str.txt = malloc(rv.str.len);
- memcpy(rv.str.txt, v.str.txt, v.str.len);
- break;
- }
- return rv;
- }
+If the same variable is declared in both branchs of an 'if/else', or
+in all cases of a 'switch' then the multiple instances may be merged
+into just one variable if the variable is referenced after the
+conditional statement. When this happens, the types must naturally be
+consistent across all the branches. When the variable is not used
+outside the if, the variables in the different branches are distinct
+and can be of different types.
+
+Undeclared names may only appear in "use" statements and "case" expressions.
+These names are given a type of "label" and a unique value.
+This allows them to fill the role of a name in an enumerated type, which
+is useful for testing the `switch` statement.
+
+As we will see, the condition part of a `while` statement can return
+either a Boolean or some other type. This requires that the expected
+type that gets passed around comprises a type and a flag to indicate
+that `Tbool` is also permitted.
+
+As there are, as yet, no distinct types that are compatible, there
+isn't much subtlety in the analysis. When we have distinct number
+types, this will become more interesting.
+
+#### Error reporting
+
+When analysis discovers an inconsistency it needs to report an error;
+just refusing to run the code ensures that the error doesn't cascade,
+but by itself it isn't very useful. A clear understanding of the sort
+of error message that are useful will help guide the process of
+analysis.
+
+At a simplistic level, the only sort of error that type analysis can
+report is that the type of some construct doesn't match a contextual
+requirement. For example, in `4 + "hello"` the addition provides a
+contextual requirement for numbers, but `"hello"` is not a number. In
+this particular example no further information is needed as the types
+are obvious from local information. When a variable is involved that
+isn't the case. It may be helpful to explain why the variable has a
+particular type, by indicating the location where the type was set,
+whether by declaration or usage.
+
+Using a recursive-descent analysis we can easily detect a problem at
+multiple locations. In "`hello:= "there"; 4 + hello`" the addition
+will detect that one argument is not a number and the usage of `hello`
+will detect that a number was wanted, but not provided. In this
+(early) version of the language, we will generate error reports at
+multiple locations, so the use of `hello` will report an error and
+explain were the value was set, and the addition will report an error
+and say why numbers are needed. To be able to report locations for
+errors, each language element will need to record a file location
+(line and column) and each variable will need to record the language
+element where its type was set. For now we will assume that each line
+of an error message indicates one location in the file, and up to 2
+types. So we provide a `printf`-like function which takes a format, a
+location (a `struct exec` which has not yet been introduced), and 2
+types. "`%1`" reports the first type, "`%2`" reports the second. We
+will need a function to print the location, once we know how that is
+stored. e As will be explained later, there are sometimes extra rules for
+type matching and they might affect error messages, we need to pass those
+in too.
+
+As well as type errors, we sometimes need to report problems with
+tokens, which might be unexpected or might name a type that has not
+been defined. For these we have `tok_err()` which reports an error
+with a given token. Each of the error functions sets the flag in the
+context so indicate that parsing failed.
- static int value_cmp(struct value left, struct value right)
- {
- int cmp;
- if (left.vtype != right.vtype)
- return left.vtype - right.vtype;
- switch (left.vtype) {
- case Vnum: cmp = mpq_cmp(left.num, right.num); break;
- case Vstr: cmp = text_cmp(left.str, right.str); break;
- case Vbool: cmp = left.bool - right.bool; break;
- case Vnone:
- case Vunknown: cmp = 0;
- }
- return cmp;
- }
+###### forward decls
- static struct text text_join(struct text a, struct text b)
- {
- struct text rv;
- rv.len = a.len + b.len;
- rv.txt = malloc(rv.len);
- memcpy(rv.txt, a.txt, a.len);
- memcpy(rv.txt+a.len, b.txt, b.len);
- return rv;
- }
+ static void fput_loc(struct exec *loc, FILE *f);
+ static void type_err(struct parse_context *c,
+ char *fmt, struct exec *loc,
+ struct type *t1, int rules, struct type *t2);
+ static void tok_err(struct parse_context *c, char *fmt, struct token *t);
- static void print_value(struct value v)
- {
- switch (v.vtype) {
- case Vunknown:
- printf("*Unknown*"); break;
- case Vnone:
- printf("*no-value*"); break;
- case Vstr:
- printf("%.*s", v.str.len, v.str.txt); break;
- case Vbool:
- printf("%s", v.bool ? "True":"False"); break;
- case Vnum:
- {
- mpf_t fl;
- mpf_init2(fl, 20);
- mpf_set_q(fl, v.num);
- gmp_printf("%Fg", fl);
- mpf_clear(fl);
- break;
- }
- }
- }
+###### core functions
- static int parse_value(struct value *vl, char *arg)
+ static void type_err(struct parse_context *c,
+ char *fmt, struct exec *loc,
+ struct type *t1, int rules, struct type *t2)
{
- struct text tx;
- int neg = 0;
- switch(vl->vtype) {
- case Vunknown:
- case Vnone:
- return 0;
- case Vstr:
- vl->str.len = strlen(arg);
- vl->str.txt = malloc(vl->str.len);
- memcpy(vl->str.txt, arg, vl->str.len);
- break;
- case Vnum:
- if (*arg == '-') {
- neg = 1;
- arg++;
+ fprintf(stderr, "%s:", c->file_name);
+ fput_loc(loc, stderr);
+ for (; *fmt ; fmt++) {
+ if (*fmt != '%') {
+ fputc(*fmt, stderr);
+ continue;
}
- tx.txt = arg; tx.len = strlen(tx.txt);
- if (number_parse(vl->num, vl->tail, tx) == 0)
- mpq_init(vl->num);
- else if (neg)
- mpq_neg(vl->num, vl->num);
- break;
- case Vbool:
- if (strcasecmp(arg, "true") == 0 ||
- strcmp(arg, "1") == 0)
- vl->bool = 1;
- else if (strcasecmp(arg, "false") == 0 ||
- strcmp(arg, "0") == 0)
- vl->bool = 2;
- else {
- printf("Bad bool: %s\n", arg);
- return 0;
+ fmt++;
+ switch (*fmt) {
+ case '%': fputc(*fmt, stderr); break; // NOTEST
+ default: fputc('?', stderr); break; // NOTEST
+ case '1':
+ type_print(t1, stderr);
+ break;
+ case '2':
+ type_print(t2, stderr);
+ break;
+ ## format cases
}
- break;
}
- return 1;
+ fputs("\n", stderr);
+ c->parse_error += 1;
}
-### Variables
-
-Variables are simply named values. We store them in a linked list
-sorted by name and use sequential search and insertion sort.
-
-This linked list is stored in the parse context so that reduce
-functions can find or add variables, and so the analysis phase can
-ensure that every variable gets a type.
-
-###### ast
-
- struct variable {
- struct text name;
- struct variable *next;
- struct value val;
- };
-
-###### macros
-
- #define container_of(ptr, type, member) ({ \
- const typeof( ((type *)0)->member ) *__mptr = (ptr); \
- (type *)( (char *)__mptr - offsetof(type,member) );})
-
-###### parse context
-
- struct variable *varlist;
-
-###### free context
- while (context.varlist) {
- struct variable *v = context.varlist;
- context.varlist = v->next;
- free_value(v->val);
- free(v);
+ static void tok_err(struct parse_context *c, char *fmt, struct token *t)
+ {
+ fprintf(stderr, "%s:%d:%d: %s: %.*s\n", c->file_name, t->line, t->col, fmt,
+ t->txt.len, t->txt.txt);
+ c->parse_error += 1;
}
-###### ast functions
-
- static struct variable *find_variable(struct token_config *conf, struct text s)
- {
- struct variable **l = &container_of(conf, struct parse_context,
- config)->varlist;
- struct variable *n;
- int cmp = 1;
+## Entities: declared and predeclared.
- while (*l &&
- (cmp = text_cmp((*l)->name, s)) < 0)
- l = & (*l)->next;
- if (cmp == 0)
- return *l;
- n = calloc(1, sizeof(*n));
- n->name = s;
- n->val.vtype = Vunknown;
- n->next = *l;
- *l = n;
- return n;
- }
+There are various "things" that the language and/or the interpreter
+needs to know about to parse and execute a program. These include
+types, variables, values, and executable code. These are all lumped
+together under the term "entities" (calling them "objects" would be
+confusing) and introduced here. The following section will present the
+different specific code elements which comprise or manipulate these
+various entities.
### Executables
Executables can be lots of different things. In many cases an
executable is just an operation combined with one or two other
-executables. This allows for expressions and lists etc. Other times
-an executable is something quite specific like a constant or variable
-name. So we define a `struct exec` to be a general executable with a
-type, and a `struct binode` which is a subclass of `exec` and forms a
-node in a binary tree and holding an operation. There will be other
-subclasses, and to access these we need to be able to `cast` the
-`exec` into the various other types.
+executables. This allows for expressions and lists etc. Other times an
+executable is something quite specific like a constant or variable name.
+So we define a `struct exec` to be a general executable with a type, and
+a `struct binode` which is a subclass of `exec`, forms a node in a
+binary tree, and holds an operation. There will be other subclasses,
+and to access these we need to be able to `cast` the `exec` into the
+various other types. The first field in any `struct exec` is the type
+from the `exec_types` enum.
###### macros
#define cast(structname, pointer) ({ \
if (__mptr && *__mptr != X##structname) abort(); \
(struct structname *)( (char *)__mptr);})
- #define new(structname) ({ \
+ #define new(structname) ({ \
+ struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
+ __ptr->type = X##structname; \
+ __ptr->line = -1; __ptr->column = -1; \
+ __ptr;})
+
+ #define new_pos(structname, token) ({ \
struct structname *__ptr = ((struct structname *)calloc(1,sizeof(struct structname))); \
- __ptr->type = X##structname; \
+ __ptr->type = X##structname; \
+ __ptr->line = token.line; __ptr->column = token.col; \
__ptr;})
###### ast
};
struct exec {
enum exec_types type;
+ int line, column;
+ ## exec fields
};
struct binode {
struct exec;
struct exec *left, *right;
};
-Each different type of `exec` node needs a number of functions
-defined, a bit like methods. We must be able to be able to free it,
-print it, analyse it and execute it. Once we have specific `exec`
-types we will need to parse them to. Let's take this a bit more
-slowly.
+###### ast functions
+
+ static int __fput_loc(struct exec *loc, FILE *f)
+ {
+ if (!loc)
+ return 0;
+ if (loc->line >= 0) {
+ fprintf(f, "%d:%d: ", loc->line, loc->column);
+ return 1;
+ }
+ if (loc->type == Xbinode)
+ return __fput_loc(cast(binode,loc)->left, f) ||
+ __fput_loc(cast(binode,loc)->right, f); // NOTEST
+ return 0; // NOTEST
+ }
+ static void fput_loc(struct exec *loc, FILE *f)
+ {
+ if (!__fput_loc(loc, f))
+ fprintf(f, "??:??: "); // NOTEST
+ }
+
+Each different type of `exec` node needs a number of functions defined,
+a bit like methods. We must be able to free it, print it, analyse it
+and execute it. Once we have specific `exec` types we will need to
+parse them too. Let's take this a bit more slowly.
#### Freeing
-The parser generator requires as `free_foo` function for each struct
-that stores attributes and they will be `exec`s of subtypes there-of.
-So we need `free_exec` which can handle all the subtypes, and we need
-`free_binode`.
+The parser generator requires a `free_foo` function for each struct
+that stores attributes and they will often be `exec`s and subtypes
+there-of. So we need `free_exec` which can handle all the subtypes,
+and we need `free_binode`.
###### ast functions
static void do_indent(int i, char *str)
{
- while (i--)
+ while (i-- > 0)
printf(" ");
printf("%s", str);
}
static void print_exec(struct exec *e, int indent, int bracket)
{
+ if (!e)
+ return;
switch (e->type) {
case Xbinode:
print_binode(cast(binode, e), indent, bracket); break;
## print exec cases
}
+ if (e->to_free) {
+ struct variable *v;
+ do_indent(indent, "/* FREE");
+ for (v = e->to_free; v; v = v->next_free) {
+ printf(" %.*s", v->name->name.len, v->name->name.txt);
+ printf("[%d,%d]", v->scope_start, v->scope_end);
+ if (v->frame_pos >= 0)
+ printf("(%d+%d)", v->frame_pos,
+ v->type ? v->type->size:0);
+ }
+ printf(" */\n");
+ }
}
###### forward decls
#### Analysing
-As discusses, analysis involves propagating type requirements around
-the program and looking for errors.
+As discussed, analysis involves propagating type requirements around the
+program and looking for errors.
+
+So `propagate_types` is passed an expected type (being a `struct type`
+pointer together with some `val_rules` flags) that the `exec` is
+expected to return, and returns the type that it does return, either of
+which can be `NULL` signifying "unknown". A `prop_err` flag set is
+passed by reference. It has `Efail` set when an error is found, and
+`Eretry` when the type for some element is set via propagation. If
+any expression cannot be evaluated a compile time, `Eruntime` is set.
+If the expression can be copied, `Emaycopy` is set.
+
+If it remains unchanged at `0`, then no more propagation is needed.
+
+###### ast
-So propagate_types is passed a type that the `exec` is expected to return,
-and returns the type that it does return, either of which can be `Vunknown`.
-An `ok` flag is passed by reference. It is set to `0` when an error is
-found, and `2` when any change is made. If it remains unchanged at
-`1`, then no more propagation is needed.
+ enum val_rules {Rboolok = 1<<1, Rnoconstant = 1<<2};
+ enum prop_err {Efail = 1<<0, Eretry = 1<<1, Eruntime = 1<<2,
+ Emaycopy = 1<<3};
+###### forward decls
+ static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
+ struct type *type, int rules);
###### core functions
- static enum vtype propagate_types(struct exec *prog, enum vtype type,
- int *ok)
+ static struct type *__propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
+ struct type *type, int rules)
{
- enum vtype t;
+ struct type *t;
- if (!prog) {
- if (type != Vunknown && type != Vnone)
- *ok = 0;
- return Vnone;
- }
+ if (!prog)
+ return Tnone;
switch (prog->type) {
case Xbinode:
}
## propagate exec cases
}
- return Vnone;
+ return Tnone;
+ }
+
+ static struct type *propagate_types(struct exec *prog, struct parse_context *c, enum prop_err *perr,
+ struct type *type, int rules)
+ {
+ int pre_err = c->parse_error;
+ struct type *ret = __propagate_types(prog, c, perr, type, rules);
+
+ if (c->parse_error > pre_err)
+ *perr |= Efail;
+ return ret;
}
#### Interpreting
Interpreting an `exec` doesn't require anything but the `exec`. State
is stored in variables and each variable will be directly linked from
-within the `exec` tree. The exception to this is the whole `program`
-which needs to look at command line arguments. The `program` will be
+within the `exec` tree. The exception to this is the `main` function
+which needs to look at command line arguments. This function will be
interpreted separately.
-Each `exec` can return a value, which may be `Vnone` but shouldn't be `Vunknown`.
+Each `exec` can return a value combined with a type in `struct lrval`.
+The type may be `Tnone` but must be non-NULL. Some `exec`s will return
+the location of a value, which can be updated, in `lval`. Others will
+set `lval` to NULL indicating that there is a value of appropriate type
+in `rval`.
+###### forward decls
+ static struct value interp_exec(struct parse_context *c, struct exec *e,
+ struct type **typeret);
###### core functions
- static struct value interp_exec(struct exec *e)
+ struct lrval {
+ struct type *type;
+ struct value rval, *lval;
+ };
+
+ /* If dest is passed, dtype must give the expected type, and
+ * result can go there, in which case type is returned as NULL.
+ */
+ static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
+ struct value *dest, struct type *dtype);
+
+ static struct value interp_exec(struct parse_context *c, struct exec *e,
+ struct type **typeret)
{
- struct value rv;
- rv.vtype = Vnone;
- if (!e)
- return rv;
+ struct lrval ret = _interp_exec(c, e, NULL, NULL);
+
+ if (!ret.type) abort();
+ if (typeret)
+ *typeret = ret.type;
+ if (ret.lval)
+ dup_value(ret.type, ret.lval, &ret.rval);
+ return ret.rval;
+ }
+
+ static struct value *linterp_exec(struct parse_context *c, struct exec *e,
+ struct type **typeret)
+ {
+ struct lrval ret = _interp_exec(c, e, NULL, NULL);
+
+ if (!ret.type) abort();
+ if (ret.lval)
+ *typeret = ret.type;
+ else
+ free_value(ret.type, &ret.rval);
+ return ret.lval;
+ }
+
+ /* dinterp_exec is used when the destination type is certain and
+ * the value has a place to go.
+ */
+ static void dinterp_exec(struct parse_context *c, struct exec *e,
+ struct value *dest, struct type *dtype,
+ int need_free)
+ {
+ struct lrval ret = _interp_exec(c, e, dest, dtype);
+ if (!ret.type)
+ return;
+ if (need_free)
+ free_value(dtype, dest);
+ if (ret.lval)
+ dup_value(dtype, ret.lval, dest);
+ else
+ memcpy(dest, &ret.rval, dtype->size);
+ }
+
+ static struct lrval _interp_exec(struct parse_context *c, struct exec *e,
+ struct value *dest, struct type *dtype)
+ {
+ /* If the result is copied to dest, ret.type is set to NULL */
+ struct lrval ret;
+ struct value rv = {}, *lrv = NULL;
+ struct type *rvtype;
+
+ rvtype = ret.type = Tnone;
+ if (!e) {
+ ret.lval = lrv;
+ ret.rval = rv;
+ return ret;
+ }
switch(e->type) {
case Xbinode:
{
struct binode *b = cast(binode, e);
- struct value left, right;
- left.vtype = right.vtype = Vnone;
+ struct value left, right, *lleft;
+ struct type *ltype, *rtype;
+ ltype = rtype = Tnone;
switch (b->op) {
## interp binode cases
}
- free_value(left); free_value(right);
+ free_value(ltype, &left);
+ free_value(rtype, &right);
break;
}
## interp exec cases
}
- return rv;
+ if (rvtype) {
+ ret.lval = lrv;
+ ret.rval = rv;
+ ret.type = rvtype;
+ }
+ ## interp exec cleanup
+ return ret;
}
-## Language elements
+### Types
-Each language element needs to be parsed, printed, analysed,
-interpreted, and freed. There are several, so let's just start with
-the easy ones and work our way up.
+Values come in a wide range of types, with more likely to be added.
+Each type needs to be able to print its own values (for convenience at
+least) as well as to compare two values, at least for equality and
+possibly for order. For now, values might need to be duplicated and
+freed, though eventually such manipulations will be better integrated
+into the language.
-### Values
+Rather than requiring every numeric type to support all numeric
+operations (add, multiply, etc), we allow types to be able to present
+as one of a few standard types: integer, float, and fraction. The
+existence of these conversion functions eventually enable types to
+determine if they are compatible with other types, though such types
+have not yet been implemented.
-We have already met values and separate objects. When manifest
-constants appear in the program text that must result in an executable
-which has a constant value. So the `val` structure embeds a value in
-an executable.
+Named type are stored in a simple linked list. Objects of each type are
+"values" which are often passed around by value.
-###### exec type
- Xval,
+There are both explicitly named types, and anonymous types. Anonymous
+cannot be accessed by name, but are used internally and have a name
+which might be reported in error messages.
###### ast
- struct val {
- struct exec;
- struct value val;
+
+ struct value {
+ union {
+ char ptr[1];
+ ## value union fields
+ };
};
-###### Grammar
+###### ast late
+ struct type {
+ struct text name;
+ struct type *next;
+ struct token first_use;
+ int size, align;
+ int anon;
+ void (*init)(struct type *type, struct value *val);
+ int (*prepare_type)(struct parse_context *c, struct type *type, int parse_time);
+ void (*print)(struct type *type, struct value *val, FILE *f);
+ void (*print_type)(struct type *type, FILE *f);
+ int (*cmp_order)(struct type *t1, struct type *t2,
+ struct value *v1, struct value *v2);
+ int (*cmp_eq)(struct type *t1, struct type *t2,
+ struct value *v1, struct value *v2);
+ void (*dup)(struct type *type, struct value *vold, struct value *vnew);
+ int (*test)(struct type *type, struct value *val);
+ void (*free)(struct type *type, struct value *val);
+ void (*free_type)(struct type *t);
+ long long (*to_int)(struct value *v);
+ double (*to_float)(struct value *v);
+ int (*to_mpq)(mpq_t *q, struct value *v);
+ ## type functions
+ union {
+ ## type union fields
+ };
+ };
- $*val
- Value -> True ${
- $0 = new(val);
- $0->val.vtype = Vbool;
- $0->val.bool = 1;
- }$
- | False ${
- $0 = new(val);
- $0->val.vtype = Vbool;
- $0->val.bool = 0;
- }$
- | NUMBER ${
- $0 = new(val);
- $0->val.vtype = Vnum;
- if (number_parse($0->val.num, $0->val.tail, $1.txt) == 0)
- mpq_init($0->val.num);
- }$
- | STRING ${
- $0 = new(val);
- $0->val.vtype = Vstr;
- string_parse(&$1, '\\', &$0->val.str, $0->val.tail);
- }$
+###### parse context
-###### print exec cases
- case Xval:
+ struct type *typelist;
+
+###### includes
+ #include <stdarg.h>
+
+###### ast functions
+
+ static struct type *find_type(struct parse_context *c, struct text s)
{
- struct val *v = cast(val, e);
- if (v->val.vtype == Vstr)
- printf("\"");
- print_value(v->val);
- if (v->val.vtype == Vstr)
- printf("\"");
- break;
+ struct type *t = c->typelist;
+
+ while (t && (t->anon ||
+ text_cmp(t->name, s) != 0))
+ t = t->next;
+ return t;
}
-###### propagate exec cases
- case Xval:
- {
- struct val *val = cast(val, prog);
- if (type != Vunknown &&
- type != val->val.vtype)
- *ok = 0;
- return val->val.vtype;
- }
+ static struct type *_add_type(struct parse_context *c, struct text s,
+ struct type *proto, int anon)
+ {
+ struct type *n;
-###### interp exec cases
- case Xval:
- return dup_value(cast(val, e)->val);
+ n = calloc(1, sizeof(*n));
+ if (proto)
+ *n = *proto;
+ else
+ n->size = -1;
+ n->name = s;
+ n->anon = anon;
+ n->next = c->typelist;
+ c->typelist = n;
+ return n;
+ }
-###### ast functions
- void free_val(struct val *v)
+ static struct type *add_type(struct parse_context *c, struct text s,
+ struct type *proto)
{
- if (!v)
- return;
- free_value(v->val);
- free(v);
+ return _add_type(c, s, proto, 0);
}
-###### free exec cases
- case Xval: free_val(cast(val, e)); break;
+ static struct type *add_anon_type(struct parse_context *c,
+ struct type *proto, char *name, ...)
+ {
+ struct text t;
+ va_list ap;
+
+ va_start(ap, name);
+ vasprintf(&t.txt, name, ap);
+ va_end(ap);
+ t.len = strlen(t.txt);
+ return _add_type(c, t, proto, 1);
+ }
+
+ static struct type *find_anon_type(struct parse_context *c,
+ struct type *proto, char *name, ...)
+ {
+ struct type *t = c->typelist;
+ struct text nm;
+ va_list ap;
+
+ va_start(ap, name);
+ vasprintf(&nm.txt, name, ap);
+ va_end(ap);
+ nm.len = strlen(name);
+
+ while (t && (!t->anon ||
+ text_cmp(t->name, nm) != 0))
+ t = t->next;
+ if (t) {
+ free(nm.txt);
+ return t;
+ }
+ return _add_type(c, nm, proto, 1);
+ }
+
+ static void free_type(struct type *t)
+ {
+ /* The type is always a reference to something in the
+ * context, so we don't need to free anything.
+ */
+ }
+
+ static void free_value(struct type *type, struct value *v)
+ {
+ if (type && v) {
+ type->free(type, v);
+ memset(v, 0x5a, type->size);
+ }
+ }
+
+ static void type_print(struct type *type, FILE *f)
+ {
+ if (!type)
+ fputs("*unknown*type*", f); // NOTEST
+ else if (type->name.len && !type->anon)
+ fprintf(f, "%.*s", type->name.len, type->name.txt);
+ else if (type->print_type)
+ type->print_type(type, f);
+ else if (type->name.len && type->anon)
+ fprintf(f, "\"%.*s\"", type->name.len, type->name.txt);
+ else
+ fputs("*invalid*type*", f); // NOTEST
+ }
+
+ static void val_init(struct type *type, struct value *val)
+ {
+ if (type && type->init)
+ type->init(type, val);
+ }
+
+ static void dup_value(struct type *type,
+ struct value *vold, struct value *vnew)
+ {
+ if (type && type->dup)
+ type->dup(type, vold, vnew);
+ }
+
+ static int value_cmp(struct type *tl, struct type *tr,
+ struct value *left, struct value *right)
+ {
+ if (tl && tl->cmp_order)
+ return tl->cmp_order(tl, tr, left, right);
+ if (tl && tl->cmp_eq)
+ return tl->cmp_eq(tl, tr, left, right);
+ return -1; // NOTEST
+ }
+
+ static void print_value(struct type *type, struct value *v, FILE *f)
+ {
+ if (type && type->print)
+ type->print(type, v, f);
+ else
+ fprintf(f, "*Unknown*"); // NOTEST
+ }
+
+ static void prepare_types(struct parse_context *c)
+ {
+ struct type *t;
+ int retry = 1;
+ enum { none, some, cannot } progress = none;
+
+ while (retry) {
+ retry = 0;
+
+ for (t = c->typelist; t; t = t->next) {
+ if (t->size < 0)
+ tok_err(c, "error: type used but not declared",
+ &t->first_use);
+ if (t->size == 0 && t->prepare_type) {
+ if (t->prepare_type(c, t, 1))
+ progress = some;
+ else if (progress == cannot)
+ tok_err(c, "error: type has recursive definition",
+ &t->first_use);
+ else
+ retry = 1;
+ }
+ }
+ switch (progress) {
+ case cannot:
+ retry = 0; break;
+ case none:
+ progress = cannot; break;
+ case some:
+ progress = none; break;
+ }
+ }
+ }
+
+###### forward decls
+
+ static void free_value(struct type *type, struct value *v);
+ static int type_compat(struct type *require, struct type *have, int rules);
+ static void type_print(struct type *type, FILE *f);
+ static void val_init(struct type *type, struct value *v);
+ static void dup_value(struct type *type,
+ struct value *vold, struct value *vnew);
+ static int value_cmp(struct type *tl, struct type *tr,
+ struct value *left, struct value *right);
+ static void print_value(struct type *type, struct value *v, FILE *f);
+
+###### free context types
+
+ while (context.typelist) {
+ struct type *t = context.typelist;
+
+ context.typelist = t->next;
+ if (t->free_type)
+ t->free_type(t);
+ if (t->anon)
+ free(t->name.txt);
+ free(t);
+ }
+
+Type can be specified for local variables, for fields in a structure,
+for formal parameters to functions, and possibly elsewhere. Different
+rules may apply in different contexts. As a minimum, a named type may
+always be used. Currently the type of a formal parameter can be
+different from types in other contexts, so we have a separate grammar
+symbol for those.
+
+###### Grammar
+
+ $*type
+ Type -> IDENTIFIER ${
+ $0 = find_type(c, $ID.txt);
+ if (!$0) {
+ $0 = add_type(c, $ID.txt, NULL);
+ $0->first_use = $ID;
+ }
+ }$
+ ## type grammar
+
+ FormalType -> Type ${ $0 = $<1; }$
+ ## formal type grammar
+
+#### Base Types
+
+Values of the base types can be numbers, which we represent as
+multi-precision fractions, strings, Booleans and labels. When
+analysing the program we also need to allow for places where no value
+is meaningful (type `Tnone`) and where we don't know what type to
+expect yet (type is `NULL`).
+
+Values are never shared, they are always copied when used, and freed
+when no longer needed.
+
+When propagating type information around the program, we need to
+determine if two types are compatible, where type `NULL` is compatible
+with anything. There are two special cases with type compatibility,
+both related to the Conditional Statement which will be described
+later. In some cases a Boolean can be accepted as well as some other
+primary type, and in others any type is acceptable except a label (`Vlabel`).
+A separate function encoding these cases will simplify some code later.
+
+###### type functions
+
+ int (*compat)(struct type *this, struct type *other);
+
+###### ast functions
+
+ static int type_compat(struct type *require, struct type *have, int rules)
+ {
+ if ((rules & Rboolok) && have == Tbool)
+ return 1; // NOTEST
+ if (!require || !have)
+ return 1;
+
+ if (require->compat)
+ return require->compat(require, have);
+
+ return require == have;
+ }
+
+###### includes
+ #include <gmp.h>
+ #include "parse_string.h"
+ #include "parse_number.h"
+
+###### libs
+ myLDLIBS := libnumber.o libstring.o -lgmp
+ LDLIBS := $(filter-out $(myLDLIBS),$(LDLIBS)) $(myLDLIBS)
+
+###### type union fields
+ enum vtype {Vnone, Vstr, Vnum, Vbool, Vlabel} vtype;
+
+###### value union fields
+ struct text str;
+ mpq_t num;
+ unsigned char bool;
+ int label;
+
+###### ast functions
+ static void _free_value(struct type *type, struct value *v)
+ {
+ if (!v)
+ return; // NOTEST
+ switch (type->vtype) {
+ case Vnone: break;
+ case Vstr: free(v->str.txt); break;
+ case Vnum: mpq_clear(v->num); break;
+ case Vlabel:
+ case Vbool: break;
+ }
+ }
+
+###### value functions
+
+ static void _val_init(struct type *type, struct value *val)
+ {
+ switch(type->vtype) {
+ case Vnone: // NOTEST
+ break; // NOTEST
+ case Vnum:
+ mpq_init(val->num); break;
+ case Vstr:
+ val->str.txt = malloc(1);
+ val->str.len = 0;
+ break;
+ case Vbool:
+ val->bool = 0;
+ break;
+ case Vlabel:
+ val->label = 0; // NOTEST
+ break; // NOTEST
+ }
+ }
+
+ static void _dup_value(struct type *type,
+ struct value *vold, struct value *vnew)
+ {
+ switch (type->vtype) {
+ case Vnone: // NOTEST
+ break; // NOTEST
+ case Vlabel:
+ vnew->label = vold->label; // NOTEST
+ break; // NOTEST
+ case Vbool:
+ vnew->bool = vold->bool;
+ break;
+ case Vnum:
+ mpq_init(vnew->num);
+ mpq_set(vnew->num, vold->num);
+ break;
+ case Vstr:
+ vnew->str.len = vold->str.len;
+ vnew->str.txt = malloc(vnew->str.len);
+ memcpy(vnew->str.txt, vold->str.txt, vnew->str.len);
+ break;
+ }
+ }
+
+ static int _value_cmp(struct type *tl, struct type *tr,
+ struct value *left, struct value *right)
+ {
+ int cmp;
+ if (tl != tr)
+ return tl - tr; // NOTEST
+ switch (tl->vtype) {
+ case Vlabel: cmp = left->label == right->label ? 0 : 1; break;
+ case Vnum: cmp = mpq_cmp(left->num, right->num); break;
+ case Vstr: cmp = text_cmp(left->str, right->str); break;
+ case Vbool: cmp = left->bool - right->bool; break;
+ case Vnone: cmp = 0; // NOTEST
+ }
+ return cmp;
+ }
+
+ static void _print_value(struct type *type, struct value *v, FILE *f)
+ {
+ switch (type->vtype) {
+ case Vnone: // NOTEST
+ fprintf(f, "*no-value*"); break; // NOTEST
+ case Vlabel: // NOTEST
+ fprintf(f, "*label-%d*", v->label); break; // NOTEST
+ case Vstr:
+ fprintf(f, "%.*s", v->str.len, v->str.txt); break;
+ case Vbool:
+ fprintf(f, "%s", v->bool ? "True":"False"); break;
+ case Vnum:
+ {
+ mpf_t fl;
+ mpf_init2(fl, 20);
+ mpf_set_q(fl, v->num);
+ gmp_fprintf(f, "%.10Fg", fl);
+ mpf_clear(fl);
+ break;
+ }
+ }
+ }
+
+ static void _free_value(struct type *type, struct value *v);
+
+ static int bool_test(struct type *type, struct value *v)
+ {
+ return v->bool;
+ }
+
+ static struct type base_prototype = {
+ .init = _val_init,
+ .print = _print_value,
+ .cmp_order = _value_cmp,
+ .cmp_eq = _value_cmp,
+ .dup = _dup_value,
+ .free = _free_value,
+ };
+
+ static struct type *Tbool, *Tstr, *Tnum, *Tnone, *Tlabel;
+
+###### ast functions
+ static struct type *add_base_type(struct parse_context *c, char *n,
+ enum vtype vt, int size)
+ {
+ struct text txt = { n, strlen(n) };
+ struct type *t;
+
+ t = add_type(c, txt, &base_prototype);
+ t->vtype = vt;
+ t->size = size;
+ t->align = size > sizeof(void*) ? sizeof(void*) : size;
+ if (t->size & (t->align - 1))
+ t->size = (t->size | (t->align - 1)) + 1; // NOTEST
+ return t;
+ }
+
+###### context initialization
+
+ Tbool = add_base_type(&context, "Boolean", Vbool, sizeof(char));
+ Tbool->test = bool_test;
+ Tstr = add_base_type(&context, "string", Vstr, sizeof(struct text));
+ Tnum = add_base_type(&context, "number", Vnum, sizeof(mpq_t));
+ Tnone = add_base_type(&context, "none", Vnone, 0);
+ Tlabel = add_base_type(&context, "label", Vlabel, sizeof(void*));
+
+##### Base Values
+
+We have already met values as separate objects. When manifest constants
+appear in the program text, that must result in an executable which has
+a constant value. So the `val` structure embeds a value in an
+executable.
+
+###### exec type
+ Xval,
+
+###### ast
+ struct val {
+ struct exec;
+ struct type *vtype;
+ struct value val;
+ };
+
+###### ast functions
+ struct val *new_val(struct type *T, struct token tk)
+ {
+ struct val *v = new_pos(val, tk);
+ v->vtype = T;
+ return v;
+ }
+
+###### declare terminals
+ $TERM True False
+
+###### Grammar
+
+ $*val
+ Value -> True ${
+ $0 = new_val(Tbool, $1);
+ $0->val.bool = 1;
+ }$
+ | False ${
+ $0 = new_val(Tbool, $1);
+ $0->val.bool = 0;
+ }$
+ | NUMBER ${ {
+ char tail[3];
+ $0 = new_val(Tnum, $1);
+ if (number_parse($0->val.num, tail, $1.txt) == 0)
+ mpq_init($0->val.num); // UNTESTED
+ if (tail[0])
+ tok_err(c, "error: unsupported number suffix",
+ &$1);
+ } }$
+ | STRING ${ {
+ char tail[3];
+ $0 = new_val(Tstr, $1);
+ string_parse(&$1, '\\', &$0->val.str, tail);
+ if (tail[0])
+ tok_err(c, "error: unsupported string suffix",
+ &$1);
+ } }$
+ | MULTI_STRING ${ {
+ char tail[3];
+ $0 = new_val(Tstr, $1);
+ string_parse(&$1, '\\', &$0->val.str, tail);
+ if (tail[0])
+ tok_err(c, "error: unsupported string suffix",
+ &$1);
+ } }$
+
+###### print exec cases
+ case Xval:
+ {
+ struct val *v = cast(val, e);
+ if (v->vtype == Tstr)
+ printf("\"");
+ // FIXME how to ensure numbers have same precision.
+ print_value(v->vtype, &v->val, stdout);
+ if (v->vtype == Tstr)
+ printf("\"");
+ break;
+ }
+
+###### propagate exec cases
+ case Xval:
+ {
+ struct val *val = cast(val, prog);
+ if (!type_compat(type, val->vtype, rules))
+ type_err(c, "error: expected %1 found %2",
+ prog, type, rules, val->vtype);
+ return val->vtype;
+ }
+
+###### interp exec cases
+ case Xval:
+ rvtype = cast(val, e)->vtype;
+ dup_value(rvtype, &cast(val, e)->val, &rv);
+ break;
+
+###### ast functions
+ static void free_val(struct val *v)
+ {
+ if (v)
+ free_value(v->vtype, &v->val);
+ free(v);
+ }
+
+###### free exec cases
+ case Xval: free_val(cast(val, e)); break;
###### ast functions
- // Move all nodes from 'b' to 'rv', reversing the order.
+ // Move all nodes from 'b' to 'rv', reversing their order.
// In 'b' 'left' is a list, and 'right' is the last node.
// In 'rv', left' is the first node and 'right' is a list.
- struct binode *reorder_bilist(struct binode *b)
+ static struct binode *reorder_bilist(struct binode *b)
{
struct binode *rv = NULL;
b = NULL;
rv->left = t;
}
- return rv;
+ return rv;
+ }
+
+#### Labels
+
+Labels are a temporary concept until I implement enums. There are an
+anonymous enum which is declared by usage. Thet are only allowed in
+`use` statements and corresponding `case` entries. They appear as a
+period followed by an identifier. All identifiers that are "used" must
+have a "case".
+
+For now, we have a global list of labels, and don't check that all "use"
+match "case".
+
+###### exec type
+ Xlabel,
+
+###### ast
+ struct label {
+ struct exec;
+ struct text name;
+ int value;
+ };
+###### free exec cases
+ case Xlabel:
+ free(e);
+ break;
+###### print exec cases
+ case Xlabel: {
+ struct label *l = cast(label, e);
+ printf(".%.*s", l->name.len, l->name.txt);
+ break;
+ }
+
+###### ast
+ struct labels {
+ struct labels *next;
+ struct text name;
+ int value;
+ };
+###### parse context
+ struct labels *labels;
+ int next_label;
+###### ast functions
+ static int label_lookup(struct parse_context *c, struct text name)
+ {
+ struct labels *l, **lp = &c->labels;
+ while (*lp && text_cmp((*lp)->name, name) < 0)
+ lp = &(*lp)->next;
+ if (*lp && text_cmp((*lp)->name, name) == 0)
+ return (*lp)->value;
+ l = calloc(1, sizeof(*l));
+ l->next = *lp;
+ l->name = name;
+ if (c->next_label == 0)
+ c->next_label = 2;
+ l->value = c->next_label;
+ c->next_label += 1;
+ *lp = l;
+ return l->value;
+ }
+
+###### free context storage
+ while (context.labels) {
+ struct labels *l = context.labels;
+ context.labels = l->next;
+ free(l);
+ }
+
+###### declare terminals
+ $TERM .
+###### term grammar
+ | . IDENTIFIER ${ {
+ struct label *l = new_pos(label, $ID);
+ l->name = $ID.txt;
+ $0 = l;
+ } }$
+###### propagate exec cases
+ case Xlabel: {
+ struct label *l = cast(label, prog);
+ l->value = label_lookup(c, l->name);
+ if (!type_compat(type, Tlabel, rules))
+ type_err(c, "error: expected %1 found %2",
+ prog, type, rules, Tlabel);
+ return Tlabel;
+ }
+###### interp exec cases
+ case Xlabel : {
+ struct label *l = cast(label, e);
+ rv.label = l->value;
+ rvtype = Tlabel;
+ break;
+ }
+
+
+### Variables
+
+Variables are scoped named values. We store the names in a linked list
+of "bindings" sorted in lexical order, and use sequential search and
+insertion sort.
+
+###### ast
+
+ struct binding {
+ struct text name;
+ struct binding *next; // in lexical order
+ ## binding fields
+ };
+
+This linked list is stored in the parse context so that "reduce"
+functions can find or add variables, and so the analysis phase can
+ensure that every variable gets a type.
+
+###### parse context
+
+ struct binding *varlist; // In lexical order
+
+###### ast functions
+
+ static struct binding *find_binding(struct parse_context *c, struct text s)
+ {
+ struct binding **l = &c->varlist;
+ struct binding *n;
+ int cmp = 1;
+
+ while (*l &&
+ (cmp = text_cmp((*l)->name, s)) < 0)
+ l = & (*l)->next;
+ if (cmp == 0)
+ return *l;
+ n = calloc(1, sizeof(*n));
+ n->name = s;
+ n->next = *l;
+ *l = n;
+ return n;
+ }
+
+Each name can be linked to multiple variables defined in different
+scopes. Each scope starts where the name is declared and continues
+until the end of the containing code block. Scopes of a given name
+cannot nest, so a declaration while a name is in-scope is an error.
+
+###### binding fields
+ struct variable *var;
+
+###### ast
+ struct variable {
+ struct variable *previous;
+ struct type *type;
+ struct binding *name;
+ struct exec *where_decl;// where name was declared
+ struct exec *where_set; // where type was set
+ ## variable fields
+ };
+
+When a scope closes, the values of the variables might need to be freed.
+This happens in the context of some `struct exec` and each `exec` will
+need to know which variables need to be freed when it completes.
+
+####### exec fields
+ struct variable *to_free;
+
+####### variable fields
+ struct exec *cleanup_exec;
+ struct variable *next_free;
+
+####### interp exec cleanup
+ {
+ struct variable *v;
+ for (v = e->to_free; v; v = v->next_free) {
+ struct value *val = var_value(c, v);
+ free_value(v->type, val);
+ }
+ }
+
+###### ast functions
+ static void variable_unlink_exec(struct variable *v)
+ {
+ struct variable **vp;
+ if (!v->cleanup_exec)
+ return;
+ for (vp = &v->cleanup_exec->to_free;
+ *vp; vp = &(*vp)->next_free) {
+ if (*vp != v)
+ continue;
+ *vp = v->next_free;
+ v->cleanup_exec = NULL;
+ break;
+ }
+ }
+
+While the naming seems strange, we include local constants in the
+definition of variables. A name declared `var := value` can
+subsequently be changed, but a name declared `var ::= value` cannot -
+it is constant
+
+###### variable fields
+ int constant;
+
+Scopes in parallel branches can be partially merged. More
+specifically, if a given name is declared in both branches of an
+if/else then its scope is a candidate for merging. Similarly if
+every branch of an exhaustive switch (e.g. has an "else" clause)
+declares a given name, then the scopes from the branches are
+candidates for merging.
+
+Note that names declared inside a loop (which is only parallel to
+itself) are never visible after the loop. Similarly names defined in
+scopes which are not parallel, such as those started by `for` and
+`switch`, are never visible after the scope. Only variables defined in
+both `then` and `else` (including the implicit then after an `if`, and
+excluding `then` used with `for`) and in all `case`s and `else` of a
+`switch` or `while` can be visible beyond the `if`/`switch`/`while`.
+
+Labels, which are a bit like variables, follow different rules.
+Labels are not explicitly declared, but if an undeclared name appears
+in a context where a label is legal, that effectively declares the
+name as a label. The declaration remains in force (or in scope) at
+least to the end of the immediately containing block and conditionally
+in any larger containing block which does not declare the name in some
+other way. Importantly, the conditional scope extension happens even
+if the label is only used in one parallel branch of a conditional --
+when used in one branch it is treated as having been declared in all
+branches.
+
+Merge candidates are tentatively visible beyond the end of the
+branching statement which creates them. If the name is used, the
+merge is affirmed and they become a single variable visible at the
+outer layer. If not - if it is redeclared first - the merge lapses.
+
+To track scopes we have an extra stack, implemented as a linked list,
+which roughly parallels the parse stack and which is used exclusively
+for scoping. When a new scope is opened, a new frame is pushed and
+the child-count of the parent frame is incremented. This child-count
+is used to distinguish between the first of a set of parallel scopes,
+in which declared variables must not be in scope, and subsequent
+branches, whether they may already be conditionally scoped.
+
+We need a total ordering of scopes so we can easily compare to variables
+to see if they are concurrently in scope. To achieve this we record a
+`scope_count` which is actually a count of both beginnings and endings
+of scopes. Then each variable has a record of the scope count where it
+enters scope, and where it leaves.
+
+To push a new frame *before* any code in the frame is parsed, we need a
+grammar reduction. This is most easily achieved with a grammar
+element which derives the empty string, and creates the new scope when
+it is recognised. This can be placed, for example, between a keyword
+like "if" and the code following it.
+
+###### ast
+ struct scope {
+ struct scope *parent;
+ int child_count;
+ };
+
+###### parse context
+ int scope_depth;
+ int scope_count;
+ struct scope *scope_stack;
+
+###### variable fields
+ int scope_start, scope_end;
+
+###### ast functions
+ static void scope_pop(struct parse_context *c)
+ {
+ struct scope *s = c->scope_stack;
+
+ c->scope_stack = s->parent;
+ free(s);
+ c->scope_depth -= 1;
+ c->scope_count += 1;
+ }
+
+ static void scope_push(struct parse_context *c)
+ {
+ struct scope *s = calloc(1, sizeof(*s));
+ if (c->scope_stack)
+ c->scope_stack->child_count += 1;
+ s->parent = c->scope_stack;
+ c->scope_stack = s;
+ c->scope_depth += 1;
+ c->scope_count += 1;
+ }
+
+###### Grammar
+
+ $void
+ OpenScope -> ${ scope_push(c); }$
+
+Each variable records a scope depth and is in one of four states:
+
+- "in scope". This is the case between the declaration of the
+ variable and the end of the containing block, and also between
+ the usage with affirms a merge and the end of that block.
+
+ The scope depth is not greater than the current parse context scope
+ nest depth. When the block of that depth closes, the state will
+ change. To achieve this, all "in scope" variables are linked
+ together as a stack in nesting order.
+
+- "pending". The "in scope" block has closed, but other parallel
+ scopes are still being processed. So far, every parallel block at
+ the same level that has closed has declared the name.
+
+ The scope depth is the depth of the last parallel block that
+ enclosed the declaration, and that has closed.
+
+- "conditionally in scope". The "in scope" block and all parallel
+ scopes have closed, and no further mention of the name has been seen.
+ This state includes a secondary nest depth (`min_depth`) which records
+ the outermost scope seen since the variable became conditionally in
+ scope. If a use of the name is found, the variable becomes "in scope"
+ and that secondary depth becomes the recorded scope depth. If the
+ name is declared as a new variable, the old variable becomes "out of
+ scope" and the recorded scope depth stays unchanged.
+
+- "out of scope". The variable is neither in scope nor conditionally
+ in scope. It is permanently out of scope now and can be removed from
+ the "in scope" stack. When a variable becomes out-of-scope it is
+ moved to a separate list (`out_scope`) of variables which have fully
+ known scope. This will be used at the end of each function to assign
+ each variable a place in the stack frame.
+
+###### variable fields
+ int depth, min_depth;
+ enum { OutScope, PendingScope, CondScope, InScope } scope;
+ struct variable *in_scope;
+
+###### parse context
+
+ struct variable *in_scope;
+ struct variable *out_scope;
+
+All variables with the same name are linked together using the
+'previous' link. Those variable that have been affirmatively merged all
+have a 'merged' pointer that points to one primary variable - the most
+recently declared instance. When merging variables, we need to also
+adjust the 'merged' pointer on any other variables that had previously
+been merged with the one that will no longer be primary.
+
+A variable that is no longer the most recent instance of a name may
+still have "pending" scope, if it might still be merged with most
+recent instance. These variables don't really belong in the
+"in_scope" list, but are not immediately removed when a new instance
+is found. Instead, they are detected and ignored when considering the
+list of in_scope names.
+
+The storage of the value of a variable will be described later. For now
+we just need to know that when a variable goes out of scope, it might
+need to be freed. For this we need to be able to find it, so assume that
+`var_value()` will provide that.
+
+###### variable fields
+ struct variable *merged;
+
+###### ast functions
+
+ static void variable_merge(struct variable *primary, struct variable *secondary)
+ {
+ struct variable *v;
+
+ primary = primary->merged;
+
+ for (v = primary->previous; v; v=v->previous)
+ if (v == secondary || v == secondary->merged ||
+ v->merged == secondary ||
+ v->merged == secondary->merged) {
+ v->scope = OutScope;
+ v->merged = primary;
+ if (v->scope_start < primary->scope_start)
+ primary->scope_start = v->scope_start;
+ if (v->scope_end > primary->scope_end)
+ primary->scope_end = v->scope_end; // NOTEST
+ variable_unlink_exec(v);
+ }
+ }
+
+###### forward decls
+ static struct value *var_value(struct parse_context *c, struct variable *v);
+
+###### free global vars
+
+ while (context.varlist) {
+ struct binding *b = context.varlist;
+ struct variable *v = b->var;
+ context.varlist = b->next;
+ free(b);
+ while (v) {
+ struct variable *next = v->previous;
+
+ if (v->global && v->frame_pos >= 0) {
+ free_value(v->type, var_value(&context, v));
+ if (v->depth == 0 && v->type->free == function_free)
+ // This is a function constant
+ free_exec(v->where_decl);
+ }
+ free(v);
+ v = next;
+ }
+ }
+
+#### Manipulating Bindings
+
+When a name is conditionally visible, a new declaration discards the old
+binding - the condition lapses. Similarly when we reach the end of a
+function (outermost non-global scope) any conditional scope must lapse.
+Conversely a usage of the name affirms the visibility and extends it to
+the end of the containing block - i.e. the block that contains both the
+original declaration and the latest usage. This is determined from
+`min_depth`. When a conditionally visible variable gets affirmed like
+this, it is also merged with other conditionally visible variables with
+the same name.
+
+When we parse a variable declaration we either report an error if the
+name is currently bound, or create a new variable at the current nest
+depth if the name is unbound or bound to a conditionally scoped or
+pending-scope variable. If the previous variable was conditionally
+scoped, it and its homonyms becomes out-of-scope.
+
+When we parse a variable reference (including non-declarative assignment
+"foo = bar") we report an error if the name is not bound or is bound to
+a pending-scope variable; update the scope if the name is bound to a
+conditionally scoped variable; or just proceed normally if the named
+variable is in scope.
+
+When we exit a scope, any variables bound at this level are either
+marked out of scope or pending-scoped, depending on whether the scope
+was sequential or parallel. Here a "parallel" scope means the "then"
+or "else" part of a conditional, or any "case" or "else" branch of a
+switch. Other scopes are "sequential".
+
+When exiting a parallel scope we check if there are any variables that
+were previously pending and are still visible. If there are, then
+they weren't redeclared in the most recent scope, so they cannot be
+merged and must become out-of-scope. If it is not the first of
+parallel scopes (based on `child_count`), we check that there was a
+previous binding that is still pending-scope. If there isn't, the new
+variable must now be out-of-scope.
+
+When exiting a sequential scope that immediately enclosed parallel
+scopes, we need to resolve any pending-scope variables. If there was
+no `else` clause, and we cannot determine that the `switch` was exhaustive,
+we need to mark all pending-scope variable as out-of-scope. Otherwise
+all pending-scope variables become conditionally scoped.
+
+###### ast
+ enum closetype { CloseSequential, CloseFunction, CloseParallel, CloseElse };
+
+###### ast functions
+
+ static struct variable *var_decl(struct parse_context *c, struct text s)
+ {
+ struct binding *b = find_binding(c, s);
+ struct variable *v = b->var;
+
+ switch (v ? v->scope : OutScope) {
+ case InScope:
+ /* Caller will report the error */
+ return NULL;
+ case CondScope:
+ for (;
+ v && v->scope == CondScope;
+ v = v->previous)
+ v->scope = OutScope;
+ break;
+ default: break;
+ }
+ v = calloc(1, sizeof(*v));
+ v->previous = b->var;
+ b->var = v;
+ v->name = b;
+ v->merged = v;
+ v->min_depth = v->depth = c->scope_depth;
+ v->scope = InScope;
+ v->in_scope = c->in_scope;
+ v->scope_start = c->scope_count;
+ c->in_scope = v;
+ ## variable init
+ return v;
+ }
+
+ static struct variable *var_ref(struct parse_context *c, struct text s)
+ {
+ struct binding *b = find_binding(c, s);
+ struct variable *v = b->var;
+ struct variable *v2;
+
+ switch (v ? v->scope : OutScope) {
+ case OutScope:
+ case PendingScope:
+ /* Caller will report the error */
+ return NULL;
+ case CondScope:
+ /* All CondScope variables of this name need to be merged
+ * and become InScope
+ */
+ v->depth = v->min_depth;
+ v->scope = InScope;
+ for (v2 = v->previous;
+ v2 && v2->scope == CondScope;
+ v2 = v2->previous)
+ variable_merge(v, v2);
+ break;
+ case InScope:
+ break;
+ }
+ return v;
+ }
+
+ static int var_refile(struct parse_context *c, struct variable *v)
+ {
+ /* Variable just went out of scope. Add it to the out_scope
+ * list, sorted by ->scope_start
+ */
+ struct variable **vp = &c->out_scope;
+ while ((*vp) && (*vp)->scope_start < v->scope_start)
+ vp = &(*vp)->in_scope;
+ v->in_scope = *vp;
+ *vp = v;
+ return 0;
+ }
+
+ static void var_block_close(struct parse_context *c, enum closetype ct,
+ struct exec *e)
+ {
+ /* Close off all variables that are in_scope.
+ * Some variables in c->scope may already be not-in-scope,
+ * such as when a PendingScope variable is hidden by a new
+ * variable with the same name.
+ * So we check for v->name->var != v and drop them.
+ * If we choose to make a variable OutScope, we drop it
+ * immediately too.
+ */
+ struct variable *v, **vp, *v2;
+
+ scope_pop(c);
+ for (vp = &c->in_scope;
+ (v = *vp) && v->min_depth > c->scope_depth;
+ (v->scope == OutScope || v->name->var != v)
+ ? (*vp = v->in_scope, var_refile(c, v))
+ : ( vp = &v->in_scope, 0)) {
+ v->min_depth = c->scope_depth;
+ if (v->name->var != v)
+ /* This is still in scope, but we haven't just
+ * closed the scope.
+ */
+ continue;
+ v->min_depth = c->scope_depth;
+ if (v->scope == InScope)
+ v->scope_end = c->scope_count;
+ if (v->scope == InScope && e && !v->global) {
+ /* This variable gets cleaned up when 'e' finishes */
+ variable_unlink_exec(v);
+ v->cleanup_exec = e;
+ v->next_free = e->to_free;
+ e->to_free = v;
+ }
+ switch (ct) {
+ case CloseElse:
+ case CloseParallel: /* handle PendingScope */
+ switch(v->scope) {
+ case InScope:
+ case CondScope:
+ if (c->scope_stack->child_count == 1)
+ /* first among parallel branches */
+ v->scope = PendingScope;
+ else if (v->previous &&
+ v->previous->scope == PendingScope)
+ /* all previous branches used name */
+ v->scope = PendingScope;
+ else
+ v->scope = OutScope;
+ if (ct == CloseElse) {
+ /* All Pending variables with this name
+ * are now Conditional */
+ for (v2 = v;
+ v2 && v2->scope == PendingScope;
+ v2 = v2->previous)
+ v2->scope = CondScope;
+ }
+ break;
+ case PendingScope:
+ /* Not possible as it would require
+ * parallel scope to be nested immediately
+ * in a parallel scope, and that never
+ * happens.
+ */ // NOTEST
+ case OutScope:
+ /* Not possible as we already tested for
+ * OutScope
+ */
+ abort(); // NOTEST
+ }
+ break;
+ case CloseFunction:
+ if (v->scope == CondScope)
+ /* Condition cannot continue past end of function */
+ v->scope = InScope;
+ /* fallthrough */
+ case CloseSequential:
+ switch (v->scope) {
+ case InScope:
+ v->scope = OutScope;
+ break;
+ case PendingScope:
+ /* There was no 'else', so we can only become
+ * conditional if we know the cases were exhaustive,
+ * and that doesn't mean anything yet.
+ * So only labels become conditional..
+ */
+ for (v2 = v;
+ v2 && v2->scope == PendingScope;
+ v2 = v2->previous)
+ v2->scope = OutScope;
+ break;
+ case CondScope:
+ case OutScope: break;
+ }
+ break;
+ }
+ }
+ }
+
+#### Storing Values
+
+The value of a variable is store separately from the variable, on an
+analogue of a stack frame. There are (currently) two frames that can be
+active. A global frame which currently only stores constants, and a
+stacked frame which stores local variables. Each variable knows if it
+is global or not, and what its index into the frame is.
+
+Values in the global frame are known immediately they are relevant, so
+the frame needs to be reallocated as it grows so it can store those
+values. The local frame doesn't get values until the interpreted phase
+is started, so there is no need to allocate until the size is known.
+
+We initialize the `frame_pos` to an impossible value, so that we can
+tell if it was set or not later.
+
+###### variable fields
+ short frame_pos;
+ short global;
+
+###### variable init
+ v->frame_pos = -1;
+
+###### parse context
+
+ short global_size, global_alloc;
+ short local_size;
+ void *global, *local;
+
+###### forward decls
+ static struct value *global_alloc(struct parse_context *c, struct type *t,
+ struct variable *v, struct value *init);
+
+###### ast functions
+
+ static struct value *var_value(struct parse_context *c, struct variable *v)
+ {
+ if (!v->global) {
+ if (!c->local || !v->type)
+ return NULL; // UNTESTED
+ if (v->frame_pos + v->type->size > c->local_size) {
+ printf("INVALID frame_pos\n"); // NOTEST
+ exit(2); // NOTEST
+ }
+ return c->local + v->frame_pos;
+ }
+ if (c->global_size > c->global_alloc) {
+ int old = c->global_alloc;
+ c->global_alloc = (c->global_size | 1023) + 1024;
+ c->global = realloc(c->global, c->global_alloc);
+ memset(c->global + old, 0, c->global_alloc - old);
+ }
+ return c->global + v->frame_pos;
+ }
+
+ static struct value *global_alloc(struct parse_context *c, struct type *t,
+ struct variable *v, struct value *init)
+ {
+ struct value *ret;
+ struct variable scratch;
+
+ if (t->prepare_type)
+ t->prepare_type(c, t, 1); // NOTEST
+
+ if (c->global_size & (t->align - 1))
+ c->global_size = (c->global_size + t->align) & ~(t->align-1); // NOTEST
+ if (!v) {
+ v = &scratch;
+ v->type = t;
+ }
+ v->frame_pos = c->global_size;
+ v->global = 1;
+ c->global_size += v->type->size;
+ ret = var_value(c, v);
+ if (init)
+ memcpy(ret, init, t->size);
+ else
+ val_init(t, ret); // NOTEST
+ return ret;
+ }
+
+As global values are found -- struct field initializers, labels etc --
+`global_alloc()` is called to record the value in the global frame.
+
+When the program is fully parsed, each function is analysed, we need to
+walk the list of variables local to that function and assign them an
+offset in the stack frame. For this we have `scope_finalize()`.
+
+We keep the stack from dense by re-using space for between variables
+that are not in scope at the same time. The `out_scope` list is sorted
+by `scope_start` and as we process a varible, we move it to an FIFO
+stack. For each variable we consider, we first discard any from the
+stack anything that went out of scope before the new variable came in.
+Then we place the new variable just after the one at the top of the
+stack.
+
+###### ast functions
+
+ static void scope_finalize(struct parse_context *c, struct type *ft)
+ {
+ int size = ft->function.local_size;
+ struct variable *next = ft->function.scope;
+ struct variable *done = NULL;
+
+ while (next) {
+ struct variable *v = next;
+ struct type *t = v->type;
+ int pos;
+ next = v->in_scope;
+ if (v->merged != v)
+ continue;
+ if (!t)
+ continue;
+ if (v->frame_pos >= 0)
+ continue;
+ while (done && done->scope_end < v->scope_start)
+ done = done->in_scope;
+ if (done)
+ pos = done->frame_pos + done->type->size;
+ else
+ pos = ft->function.local_size;
+ if (pos & (t->align - 1))
+ pos = (pos + t->align) & ~(t->align-1);
+ v->frame_pos = pos;
+ if (size < pos + v->type->size)
+ size = pos + v->type->size;
+ v->in_scope = done;
+ done = v;
+ }
+ c->out_scope = NULL;
+ ft->function.local_size = size;
+ }
+
+###### free context storage
+ free(context.global);
+
+#### Variables as executables
+
+Just as we used a `val` to wrap a value into an `exec`, we similarly
+need a `var` to wrap a `variable` into an exec. While each `val`
+contained a copy of the value, each `var` holds a link to the variable
+because it really is the same variable no matter where it appears.
+When a variable is used, we need to remember to follow the `->merged`
+link to find the primary instance.
+
+When a variable is declared, it may or may not be given an explicit
+type. We need to record which so that we can report the parsed code
+correctly.
+
+###### exec type
+ Xvar,
+
+###### ast
+ struct var {
+ struct exec;
+ struct variable *var;
+ };
+
+###### variable fields
+ int explicit_type;
+
+###### Grammar
+
+ $TERM : ::
+
+ $*var
+ VariableDecl -> IDENTIFIER : ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v)
+ v->where_decl = $0;
+ else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER :: ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->constant = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER : Type ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->where_set = $0;
+ v->type = $<Type;
+ v->explicit_type = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+ | IDENTIFIER :: Type ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ $0 = new_pos(var, $1);
+ $0->var = v;
+ if (v) {
+ v->where_decl = $0;
+ v->where_set = $0;
+ v->type = $<Type;
+ v->constant = 1;
+ v->explicit_type = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ $0->var = v;
+ type_err(c, "error: variable '%v' redeclared",
+ $0, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
+ } }$
+
+ $*exec
+ Variable -> IDENTIFIER ${ {
+ struct variable *v = var_ref(c, $1.txt);
+ $0 = new_pos(var, $1);
+ if (v == NULL) {
+ /* This might be a global const or a label
+ * Allocate a var with impossible type Tnone,
+ * which will be adjusted when we find out what it is,
+ * or will trigger an error.
+ */
+ v = var_decl(c, $1.txt);
+ if (v) {
+ v->type = Tnone;
+ v->where_decl = $0;
+ v->where_set = $0;
+ }
+ }
+ cast(var, $0)->var = v;
+ } }$
+
+###### print exec cases
+ case Xvar:
+ {
+ struct var *v = cast(var, e);
+ if (v->var) {
+ struct binding *b = v->var->name;
+ printf("%.*s", b->name.len, b->name.txt);
+ }
+ break;
+ }
+
+###### format cases
+ case 'v':
+ if (loc && loc->type == Xvar) {
+ struct var *v = cast(var, loc);
+ if (v->var) {
+ struct binding *b = v->var->name;
+ fprintf(stderr, "%.*s", b->name.len, b->name.txt);
+ } else
+ fputs("???", stderr); // NOTEST
+ } else
+ fputs("NOTVAR", stderr); // NOTEST
+ break;
+
+###### propagate exec cases
+
+ case Xvar:
+ {
+ struct var *var = cast(var, prog);
+ struct variable *v = var->var;
+ if (!v) {
+ type_err(c, "%d:BUG: no variable!!", prog, NULL, 0, NULL); // NOTEST
+ return Tnone; // NOTEST
+ }
+ v = v->merged;
+ if (v->constant && (rules & Rnoconstant)) {
+ type_err(c, "error: Cannot assign to a constant: %v",
+ prog, NULL, 0, NULL);
+ type_err(c, "info: name was defined as a constant here",
+ v->where_decl, NULL, 0, NULL);
+ return v->type;
+ }
+ if (v->type == Tnone && v->where_decl == prog)
+ type_err(c, "error: variable used but not declared: %v",
+ prog, NULL, 0, NULL);
+ if (v->type == NULL) {
+ if (type && !(*perr & Efail)) {
+ v->type = type;
+ v->where_set = prog;
+ *perr |= Eretry;
+ }
+ } else if (!type_compat(type, v->type, rules)) {
+ type_err(c, "error: expected %1 but variable '%v' is %2", prog,
+ type, rules, v->type);
+ type_err(c, "info: this is where '%v' was set to %1", v->where_set,
+ v->type, rules, NULL);
+ }
+ if (!v->global || v->frame_pos < 0)
+ *perr |= Eruntime;
+ if (!type)
+ return v->type;
+ return type;
+ }
+
+###### interp exec cases
+ case Xvar:
+ {
+ struct var *var = cast(var, e);
+ struct variable *v = var->var;
+
+ v = v->merged;
+ lrv = var_value(c, v);
+ rvtype = v->type;
+ break;
+ }
+
+###### ast functions
+
+ static void free_var(struct var *v)
+ {
+ free(v);
+ }
+
+###### free exec cases
+ case Xvar: free_var(cast(var, e)); break;
+
+
+### Complex types
+
+Now that we have the shape of the interpreter in place we can add some
+complex types and connected them in to the data structures and the
+different phases of parse, analyse, print, interpret.
+
+Being "complex" the language will naturally have syntax to access
+specifics of objects of these types. These will fit into the grammar as
+"Terms" which are the things that are combined with various operators to
+form "Expression". Where a Term is formed by some operation on another
+Term, the subordinate Term will always come first, so for example a
+member of an array will be expressed as the Term for the array followed
+by an index in square brackets. The strict rule of using postfix
+operations makes precedence irrelevant within terms. To provide a place
+to put the grammar for each terms of each type, we will start out by
+introducing the "Term" grammar production, with contains at least a
+simple "Value" (to be explained later).
+
+###### Grammar
+ $*exec
+ Term -> Value ${ $0 = $<1; }$
+ | Variable ${ $0 = $<1; }$
+ ## term grammar
+
+Thus far the complex types we have are arrays and structs.
+
+#### Arrays
+
+Arrays can be declared by giving a size and a type, as `[size]type' so
+`freq:[26]number` declares `freq` to be an array of 26 numbers. The
+size can be either a literal number, or a named constant. Some day an
+arbitrary expression will be supported.
+
+As a formal parameter to a function, the array can be declared with a
+new variable as the size: `name:[size::number]string`. The `size`
+variable is set to the size of the array and must be a constant. As
+`number` is the only supported type, it can be left out:
+`name:[size::]string`.
+
+Arrays cannot be assigned. When pointers are introduced we will also
+introduce array slices which can refer to part or all of an array -
+the assignment syntax will create a slice. For now, an array can only
+ever be referenced by the name it is declared with. It is likely that
+a "`copy`" primitive will eventually be define which can be used to
+make a copy of an array with controllable recursive depth.
+
+For now we have two sorts of array, those with fixed size either because
+it is given as a literal number or because it is a struct member (which
+cannot have a runtime-changing size), and those with a size that is
+determined at runtime - local variables with a const size. The former
+have their size calculated at parse time, the latter at run time.
+
+For the latter type, the `size` field of the type is the size of a
+pointer, and the array is reallocated every time it comes into scope.
+
+We differentiate struct fields with a const size from local variables
+with a const size by whether they are prepared at parse time or not.
+
+###### type union fields
+
+ struct {
+ int unspec; // size is unspecified - vsize must be set.
+ short size;
+ short static_size;
+ struct variable *vsize;
+ struct type *member;
+ } array;
+
+###### value union fields
+ void *array; // used if not static_size
+
+###### value functions
+
+ static int array_prepare_type(struct parse_context *c, struct type *type,
+ int parse_time)
+ {
+ struct value *vsize;
+ mpz_t q;
+ if (type->array.static_size)
+ return 1; // UNTESTED
+ if (type->array.unspec && parse_time)
+ return 1; // UNTESTED
+ if (parse_time && type->array.vsize && !type->array.vsize->global)
+ return 1; // UNTESTED
+
+ if (type->array.vsize) {
+ vsize = var_value(c, type->array.vsize);
+ if (!vsize)
+ return 1; // UNTESTED
+ mpz_init(q);
+ mpz_tdiv_q(q, mpq_numref(vsize->num), mpq_denref(vsize->num));
+ type->array.size = mpz_get_si(q);
+ mpz_clear(q);
+ }
+ if (!parse_time)
+ return 1;
+ if (type->array.member->size <= 0)
+ return 0; // UNTESTED
+
+ type->array.static_size = 1;
+ type->size = type->array.size * type->array.member->size;
+ type->align = type->array.member->align;
+
+ return 1;
+ }
+
+ static void array_init(struct type *type, struct value *val)
+ {
+ int i;
+ void *ptr = val->ptr;
+
+ if (!val)
+ return; // NOTEST
+ if (!type->array.static_size) {
+ val->array = calloc(type->array.size,
+ type->array.member->size);
+ ptr = val->array;
+ }
+ for (i = 0; i < type->array.size; i++) {
+ struct value *v;
+ v = (void*)ptr + i * type->array.member->size;
+ val_init(type->array.member, v);
+ }
+ }
+
+ static void array_free(struct type *type, struct value *val)
+ {
+ int i;
+ void *ptr = val->ptr;
+
+ if (!type->array.static_size)
+ ptr = val->array;
+ for (i = 0; i < type->array.size; i++) {
+ struct value *v;
+ v = (void*)ptr + i * type->array.member->size;
+ free_value(type->array.member, v);
+ }
+ if (!type->array.static_size)
+ free(ptr);
+ }
+
+ static int array_compat(struct type *require, struct type *have)
+ {
+ if (have->compat != require->compat)
+ return 0;
+ /* Both are arrays, so we can look at details */
+ if (!type_compat(require->array.member, have->array.member, 0))
+ return 0;
+ if (have->array.unspec && require->array.unspec) {
+ if (have->array.vsize && require->array.vsize &&
+ have->array.vsize != require->array.vsize) // UNTESTED
+ /* sizes might not be the same */
+ return 0; // UNTESTED
+ return 1;
+ }
+ if (have->array.unspec || require->array.unspec)
+ return 1; // UNTESTED
+ if (require->array.vsize == NULL && have->array.vsize == NULL)
+ return require->array.size == have->array.size;
+
+ return require->array.vsize == have->array.vsize; // UNTESTED
+ }
+
+ static void array_print_type(struct type *type, FILE *f)
+ {
+ fputs("[", f);
+ if (type->array.vsize) {
+ struct binding *b = type->array.vsize->name;
+ fprintf(f, "%.*s%s]", b->name.len, b->name.txt,
+ type->array.unspec ? "::" : "");
+ } else if (type->array.size)
+ fprintf(f, "%d]", type->array.size);
+ else
+ fprintf(f, "]");
+ type_print(type->array.member, f);
+ }
+
+ static struct type array_prototype = {
+ .init = array_init,
+ .prepare_type = array_prepare_type,
+ .print_type = array_print_type,
+ .compat = array_compat,
+ .free = array_free,
+ .size = sizeof(void*),
+ .align = sizeof(void*),
+ };
+
+###### declare terminals
+ $TERM [ ]
+
+###### type grammar
+
+ | [ NUMBER ] Type ${ {
+ char tail[3];
+ mpq_t num;
+ struct type *t;
+ int elements = 0;
+
+ if (number_parse(num, tail, $2.txt) == 0)
+ tok_err(c, "error: unrecognised number", &$2);
+ else if (tail[0]) {
+ tok_err(c, "error: unsupported number suffix", &$2);
+ mpq_clear(num);
+ } else {
+ elements = mpz_get_ui(mpq_numref(num));
+ if (mpz_cmp_ui(mpq_denref(num), 1) != 0) {
+ tok_err(c, "error: array size must be an integer",
+ &$2);
+ } else if (mpz_cmp_ui(mpq_numref(num), 1UL << 30) >= 0)
+ tok_err(c, "error: array size is too large",
+ &$2);
+ mpq_clear(num);
+ }
+
+ $0 = t = add_anon_type(c, &array_prototype, "array[%d]", elements );
+ t->array.size = elements;
+ t->array.member = $<4;
+ t->array.vsize = NULL;
+ } }$
+
+ | [ IDENTIFIER ] Type ${ {
+ struct variable *v = var_ref(c, $2.txt);
+
+ if (!v)
+ tok_err(c, "error: name undeclared", &$2);
+ else if (!v->constant)
+ tok_err(c, "error: array size must be a constant", &$2);
+
+ $0 = add_anon_type(c, &array_prototype, "array[%.*s]", $2.txt.len, $2.txt.txt);
+ $0->array.member = $<4;
+ $0->array.size = 0;
+ $0->array.vsize = v;
+ } }$
+
+###### Grammar
+ $*type
+ OptType -> Type ${ $0 = $<1; }$
+ | ${ $0 = NULL; }$
+
+###### formal type grammar
+
+ | [ IDENTIFIER :: OptType ] Type ${ {
+ struct variable *v = var_decl(c, $ID.txt);
+
+ v->type = $<OT;
+ v->constant = 1;
+ if (!v->type)
+ v->type = Tnum;
+ $0 = add_anon_type(c, &array_prototype, "array[var]");
+ $0->array.member = $<6;
+ $0->array.size = 0;
+ $0->array.unspec = 1;
+ $0->array.vsize = v;
+ } }$
+
+###### Binode types
+ Index,
+
+###### term grammar
+
+ | Term [ Expression ] ${ {
+ struct binode *b = new(binode);
+ b->op = Index;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+
+###### print binode cases
+ case Index:
+ print_exec(b->left, -1, bracket);
+ printf("[");
+ print_exec(b->right, -1, bracket);
+ printf("]");
+ break;
+
+###### propagate binode cases
+ case Index:
+ /* left must be an array, right must be a number,
+ * result is the member type of the array
+ */
+ propagate_types(b->right, c, perr, Tnum, 0);
+ t = propagate_types(b->left, c, perr, NULL, rules & Rnoconstant);
+ if (!t || t->compat != array_compat) {
+ type_err(c, "error: %1 cannot be indexed", prog, t, 0, NULL);
+ return NULL;
+ } else {
+ if (!type_compat(type, t->array.member, rules)) {
+ type_err(c, "error: have %1 but need %2", prog,
+ t->array.member, rules, type);
+ }
+ return t->array.member;
+ }
+ break;
+
+###### interp binode cases
+ case Index: {
+ mpz_t q;
+ long i;
+ void *ptr;
+
+ lleft = linterp_exec(c, b->left, <ype);
+ right = interp_exec(c, b->right, &rtype);
+ mpz_init(q);
+ mpz_tdiv_q(q, mpq_numref(right.num), mpq_denref(right.num));
+ i = mpz_get_si(q);
+ mpz_clear(q);
+
+ if (ltype->array.static_size)
+ ptr = lleft;
+ else
+ ptr = *(void**)lleft;
+ rvtype = ltype->array.member;
+ if (i >= 0 && i < ltype->array.size)
+ lrv = ptr + i * rvtype->size;
+ else
+ val_init(ltype->array.member, &rv); // UNSAFE
+ ltype = NULL;
+ break;
+ }
+
+#### Structs
+
+A `struct` is a data-type that contains one or more other data-types.
+It differs from an array in that each member can be of a different
+type, and they are accessed by name rather than by number. Thus you
+cannot choose an element by calculation, you need to know what you
+want up-front.
+
+The language makes no promises about how a given structure will be
+stored in memory - it is free to rearrange fields to suit whatever
+criteria seems important.
+
+Structs are declared separately from program code - they cannot be
+declared in-line in a variable declaration like arrays can. A struct
+is given a name and this name is used to identify the type - the name
+is not prefixed by the word `struct` as it would be in C.
+
+Structs are only treated as the same if they have the same name.
+Simply having the same fields in the same order is not enough. This
+might change once we can create structure initializers from a list of
+values.
+
+Each component datum is identified much like a variable is declared,
+with a name, one or two colons, and a type. The type cannot be omitted
+as there is no opportunity to deduce the type from usage. An initial
+value can be given following an equals sign, so
+
+##### Example: a struct type
+
+ struct complex:
+ x:number = 0
+ y:number = 0
+
+would declare a type called "complex" which has two number fields,
+each initialised to zero.
+
+Struct will need to be declared separately from the code that uses
+them, so we will need to be able to print out the declaration of a
+struct when reprinting the whole program. So a `print_type_decl` type
+function will be needed.
+
+###### type union fields
+
+ struct {
+ int nfields;
+ struct field {
+ struct text name;
+ struct type *type;
+ struct value *init;
+ int offset;
+ } *fields; // This is created when field_list is analysed.
+ struct fieldlist {
+ struct fieldlist *prev;
+ struct field f;
+ struct exec *init;
+ } *field_list; // This is created during parsing
+ } structure;
+
+###### type functions
+ void (*print_type_decl)(struct type *type, FILE *f);
+ struct type *(*fieldref)(struct type *t, struct parse_context *c,
+ struct fieldref *f, struct value **vp);
+
+###### value functions
+
+ static void structure_init(struct type *type, struct value *val)
+ {
+ int i;
+
+ for (i = 0; i < type->structure.nfields; i++) {
+ struct value *v;
+ v = (void*) val->ptr + type->structure.fields[i].offset;
+ if (type->structure.fields[i].init)
+ dup_value(type->structure.fields[i].type,
+ type->structure.fields[i].init,
+ v);
+ else
+ val_init(type->structure.fields[i].type, v);
+ }
+ }
+
+ static void structure_free(struct type *type, struct value *val)
+ {
+ int i;
+
+ for (i = 0; i < type->structure.nfields; i++) {
+ struct value *v;
+ v = (void*)val->ptr + type->structure.fields[i].offset;
+ free_value(type->structure.fields[i].type, v);
+ }
+ }
+
+ static void free_fieldlist(struct fieldlist *f)
+ {
+ if (!f)
+ return;
+ free_fieldlist(f->prev);
+ free_exec(f->init);
+ free(f);
+ }
+
+ static void structure_free_type(struct type *t)
+ {
+ int i;
+ for (i = 0; i < t->structure.nfields; i++)
+ if (t->structure.fields[i].init) {
+ free_value(t->structure.fields[i].type,
+ t->structure.fields[i].init);
+ }
+ free(t->structure.fields);
+ free_fieldlist(t->structure.field_list);
+ }
+
+ static int structure_prepare_type(struct parse_context *c,
+ struct type *t, int parse_time)
+ {
+ int cnt = 0;
+ struct fieldlist *f;
+
+ if (!parse_time || t->structure.fields)
+ return 1;
+
+ for (f = t->structure.field_list; f; f=f->prev) {
+ enum prop_err perr;
+ cnt += 1;
+
+ if (f->f.type->size <= 0)
+ return 0;
+ if (f->f.type->prepare_type)
+ f->f.type->prepare_type(c, f->f.type, parse_time);
+
+ if (f->init == NULL)
+ continue;
+ do {
+ perr = 0;
+ propagate_types(f->init, c, &perr, f->f.type, 0);
+ } while (perr & Eretry);
+ if (perr & Efail)
+ c->parse_error += 1; // NOTEST
+ }
+
+ t->structure.nfields = cnt;
+ t->structure.fields = calloc(cnt, sizeof(struct field));
+ f = t->structure.field_list;
+ while (cnt > 0) {
+ int a = f->f.type->align;
+ cnt -= 1;
+ t->structure.fields[cnt] = f->f;
+ if (t->size & (a-1))
+ t->size = (t->size | (a-1)) + 1;
+ t->structure.fields[cnt].offset = t->size;
+ t->size += ((f->f.type->size - 1) | (a-1)) + 1;
+ if (a > t->align)
+ t->align = a;
+
+ if (f->init && !c->parse_error) {
+ struct value vl = interp_exec(c, f->init, NULL);
+ t->structure.fields[cnt].init =
+ global_alloc(c, f->f.type, NULL, &vl);
+ }
+
+ f = f->prev;
+ }
+ return 1;
+ }
+
+ static int find_struct_index(struct type *type, struct text field)
+ {
+ int i;
+ for (i = 0; i < type->structure.nfields; i++)
+ if (text_cmp(type->structure.fields[i].name, field) == 0)
+ return i;
+ return IndexInvalid;
+ }
+
+ static struct type *structure_fieldref(struct type *t, struct parse_context *c,
+ struct fieldref *f, struct value **vp)
+ {
+ if (f->index == IndexUnknown) {
+ f->index = find_struct_index(t, f->name);
+ if (f->index < 0)
+ type_err(c, "error: cannot find requested field in %1",
+ f->left, t, 0, NULL);
+ }
+ if (f->index < 0)
+ return NULL;
+ if (vp) {
+ struct value *v = *vp;
+ v = (void*)v->ptr + t->structure.fields[f->index].offset;
+ *vp = v;
+ }
+ return t->structure.fields[f->index].type;
+ }
+
+ static struct type structure_prototype = {
+ .init = structure_init,
+ .free = structure_free,
+ .free_type = structure_free_type,
+ .print_type_decl = structure_print_type,
+ .prepare_type = structure_prepare_type,
+ .fieldref = structure_fieldref,
+ };
+
+###### exec type
+ Xfieldref,
+
+###### ast
+ struct fieldref {
+ struct exec;
+ struct exec *left;
+ int index;
+ struct text name;
+ };
+ enum { IndexUnknown = -1, IndexInvalid = -2 };
+
+###### free exec cases
+ case Xfieldref:
+ free_exec(cast(fieldref, e)->left);
+ free(e);
+ break;
+
+###### declare terminals
+ $TERM struct
+
+###### term grammar
+
+ | Term . IDENTIFIER ${ {
+ struct fieldref *fr = new_pos(fieldref, $2);
+ fr->left = $<1;
+ fr->name = $3.txt;
+ fr->index = IndexUnknown;
+ $0 = fr;
+ } }$
+
+###### print exec cases
+
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, e);
+ print_exec(f->left, -1, bracket);
+ printf(".%.*s", f->name.len, f->name.txt);
+ break;
+ }
+
+###### propagate exec cases
+
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, prog);
+ struct type *st = propagate_types(f->left, c, perr, NULL, 0);
+
+ if (!st || !st->fieldref)
+ type_err(c, "error: field reference on %1 is not supported",
+ f->left, st, 0, NULL);
+ else {
+ t = st->fieldref(st, c, f, NULL);
+ if (t && !type_compat(type, t, rules))
+ type_err(c, "error: have %1 but need %2", prog,
+ t, rules, type);
+ return t;
+ }
+ break;
+ }
+
+###### interp exec cases
+ case Xfieldref:
+ {
+ struct fieldref *f = cast(fieldref, e);
+ struct type *ltype;
+ struct value *lleft = linterp_exec(c, f->left, <ype);
+ lrv = lleft;
+ rvtype = ltype->fieldref(ltype, c, f, &lrv);
+ break;
+ }
+
+###### top level grammar
+ DeclareStruct -> struct IDENTIFIER FieldBlock Newlines ${ {
+ struct type *t;
+ t = find_type(c, $ID.txt);
+ if (!t)
+ t = add_type(c, $ID.txt, &structure_prototype);
+ else if (t->size >= 0) {
+ tok_err(c, "error: type already declared", &$ID);
+ tok_err(c, "info: this is location of declartion", &t->first_use);
+ /* Create a new one - duplicate */
+ t = add_type(c, $ID.txt, &structure_prototype);
+ } else {
+ struct type tmp = *t;
+ *t = structure_prototype;
+ t->name = tmp.name;
+ t->next = tmp.next;
+ }
+ t->structure.field_list = $<FB;
+ t->first_use = $ID;
+ } }$
+
+ $*fieldlist
+ FieldBlock -> { IN OptNL FieldLines OUT OptNL } ${ $0 = $<FL; }$
+ | { SimpleFieldList } ${ $0 = $<SFL; }$
+ | IN OptNL FieldLines OUT ${ $0 = $<FL; }$
+ | SimpleFieldList EOL ${ $0 = $<SFL; }$
+
+ FieldLines -> SimpleFieldList Newlines ${ $0 = $<SFL; }$
+ | FieldLines SimpleFieldList Newlines ${
+ $SFL->prev = $<FL;
+ $0 = $<SFL;
+ }$
+
+ SimpleFieldList -> Field ${ $0 = $<F; }$
+ | SimpleFieldList ; Field ${
+ $F->prev = $<SFL;
+ $0 = $<F;
+ }$
+ | SimpleFieldList ; ${
+ $0 = $<SFL;
+ }$
+ | ERROR ${ tok_err(c, "Syntax error in struct field", &$1); }$
+
+ Field -> IDENTIFIER : Type = Expression ${ {
+ $0 = calloc(1, sizeof(struct fieldlist));
+ $0->f.name = $ID.txt;
+ $0->f.type = $<Type;
+ $0->f.init = NULL;
+ $0->init = $<Expr;
+ } }$
+ | IDENTIFIER : Type ${
+ $0 = calloc(1, sizeof(struct fieldlist));
+ $0->f.name = $ID.txt;
+ $0->f.type = $<Type;
+ }$
+
+###### forward decls
+ static void structure_print_type(struct type *t, FILE *f);
+
+###### value functions
+ static void structure_print_type(struct type *t, FILE *f)
+ {
+ int i;
+
+ fprintf(f, "struct %.*s\n", t->name.len, t->name.txt);
+
+ for (i = 0; i < t->structure.nfields; i++) {
+ struct field *fl = t->structure.fields + i;
+ fprintf(f, " %.*s : ", fl->name.len, fl->name.txt);
+ type_print(fl->type, f);
+ if (fl->type->print && fl->init) {
+ fprintf(f, " = ");
+ if (fl->type == Tstr)
+ fprintf(f, "\""); // UNTESTED
+ print_value(fl->type, fl->init, f);
+ if (fl->type == Tstr)
+ fprintf(f, "\""); // UNTESTED
+ }
+ fprintf(f, "\n");
+ }
+ }
+
+###### print type decls
+ {
+ struct type *t;
+ int target = -1;
+
+ while (target != 0) {
+ int i = 0;
+ for (t = context.typelist; t ; t=t->next)
+ if (!t->anon && t->print_type_decl &&
+ !t->check_args) {
+ i += 1;
+ if (i == target)
+ break;
+ }
+
+ if (target == -1) {
+ target = i;
+ } else {
+ t->print_type_decl(t, stdout);
+ target -= 1;
+ }
+ }
+ }
+
+#### References
+
+References, or pointers, are values that refer to another value. They
+can only refer to a `struct`, though as a struct can embed anything they
+can effectively refer to anything.
+
+References are potentially dangerous as they might refer to some
+variable which no longer exists - either because a stack frame
+containing it has been discarded or because the value was allocated on
+the heap and has now been free. Ocean does not yet provide any
+protection against these problems. It will in due course.
+
+With references comes the opportunity and the need to explicitly
+allocate values on the "heap" and to free them. We currently provide
+fairly basic support for this.
+
+Reference make use of the `@` symbol in various ways. A type that starts
+with `@` is a reference to whatever follows. A reference value
+followed by an `@` acts as the referred value, though the `@` is often
+not needed. Finally, an expression that starts with `@` is a special
+reference related expression. Some examples might help.
+
+##### Example: Reference examples
+
+ struct foo
+ a: number
+ b: string
+ ref: @foo
+ bar: foo
+ bar.number = 23; bar.string = "hello"
+ baz: foo
+ ref = bar
+ baz = @ref
+ baz.a = ref.a * 2
+
+ ref = @new()
+ ref@ = baz
+ @free = ref
+ ref = @nil
+
+Obviously this is very contrived. `ref` is a reference to a `foo` which
+is initially set to refer to the value stored in `bar` - no extra syntax
+is needed to "Take the address of" `bar` - the fact that `ref` is a
+reference means that only the address make sense.
+
+When `ref.a` is accessed, that is whatever value is stored in `bar.a`.
+The same syntax is used for accessing fields both in structs and in
+references to structs. It would be correct to use `ref@.a`, but not
+necessary.
+
+`@new()` creates an object of whatever type is needed for the program
+to by type-correct. In future iterations of Ocean, arguments a
+constructor will access arguments, so the the syntax now looks like a
+function call. `@free` can be assigned any reference that was returned
+by `@new()`, and it will be freed. `@nil` is a value of whatever
+reference type is appropriate, and is stable and never the address of
+anything in the heap or on the stack. A reference can be assigned
+`@nil` or compared against that value.
+
+###### declare terminals
+ $TERM @
+
+###### type union fields
+
+ struct {
+ struct type *referent;
+ } reference;
+
+###### value union fields
+ struct value *ref;
+
+###### value functions
+
+ static void reference_print_type(struct type *t, FILE *f)
+ {
+ fprintf(f, "@");
+ type_print(t->reference.referent, f);
+ }
+
+ static int reference_cmp(struct type *tl, struct type *tr,
+ struct value *left, struct value *right)
+ {
+ return left->ref == right->ref ? 0 : 1;
+ }
+
+ static void reference_dup(struct type *t,
+ struct value *vold, struct value *vnew)
+ {
+ vnew->ref = vold->ref;
+ }
+
+ static void reference_free(struct type *t, struct value *v)
+ {
+ /* Nothing to do here */
+ }
+
+ static int reference_compat(struct type *require, struct type *have)
+ {
+ if (have->compat != require->compat)
+ return 0;
+ if (have->reference.referent != require->reference.referent)
+ return 0;
+ return 1;
+ }
+
+ static int reference_test(struct type *type, struct value *val)
+ {
+ return val->ref != NULL;
+ }
+
+ static struct type *reference_fieldref(struct type *t, struct parse_context *c,
+ struct fieldref *f, struct value **vp)
+ {
+ struct type *rt = t->reference.referent;
+
+ if (rt->fieldref) {
+ if (vp)
+ *vp = (*vp)->ref;
+ return rt->fieldref(rt, c, f, vp);
+ }
+ type_err(c, "error: field reference on %1 is not supported",
+ f->left, rt, 0, NULL);
+ return Tnone;
+ }
+
+
+ static struct type reference_prototype = {
+ .print_type = reference_print_type,
+ .cmp_eq = reference_cmp,
+ .dup = reference_dup,
+ .test = reference_test,
+ .free = reference_free,
+ .compat = reference_compat,
+ .fieldref = reference_fieldref,
+ .size = sizeof(void*),
+ .align = sizeof(void*),
+ };
+
+###### type grammar
+
+ | @ IDENTIFIER ${ {
+ struct type *t = find_type(c, $ID.txt);
+ if (!t) {
+ t = add_type(c, $ID.txt, NULL);
+ t->first_use = $ID;
+ }
+ $0 = find_anon_type(c, &reference_prototype, "@%.*s",
+ $ID.txt.len, $ID.txt.txt);
+ $0->reference.referent = t;
+ } }$
+
+###### core functions
+ static int text_is(struct text t, char *s)
+ {
+ return (strlen(s) == t.len &&
+ strncmp(s, t.txt, t.len) == 0);
+ }
+
+###### exec type
+ Xref,
+
+###### ast
+ struct ref {
+ struct exec;
+ enum ref_func { RefNew, RefFree, RefNil } action;
+ struct type *reftype;
+ struct exec *right;
+ };
+
+###### SimpleStatement Grammar
+
+ | @ IDENTIFIER = Expression ${ {
+ struct ref *r = new_pos(ref, $ID);
+ // Must be "free"
+ if (!text_is($ID.txt, "free"))
+ tok_err(c, "error: only \"@free\" makes sense here",
+ &$ID);
+
+ $0 = r;
+ r->action = RefFree;
+ r->right = $<Exp;
+ } }$
+
+###### expression grammar
+ | @ IDENTIFIER ( ) ${
+ // Only 'new' valid here
+ if (!text_is($ID.txt, "new")) {
+ tok_err(c, "error: Only reference function is \"@new()\"",
+ &$ID);
+ } else {
+ struct ref *r = new_pos(ref,$ID);
+ $0 = r;
+ r->action = RefNew;
+ }
+ }$
+ | @ IDENTIFIER ${
+ // Only 'nil' valid here
+ if (!text_is($ID.txt, "nil")) {
+ tok_err(c, "error: Only reference value is \"@nil\"",
+ &$ID);
+ } else {
+ struct ref *r = new_pos(ref,$ID);
+ $0 = r;
+ r->action = RefNil;
+ }
+ }$
+
+###### print exec cases
+ case Xref: {
+ struct ref *r = cast(ref, e);
+ switch (r->action) {
+ case RefNew:
+ printf("@new()"); break;
+ case RefNil:
+ printf("@nil"); break;
+ case RefFree:
+ do_indent(indent, "@free = ");
+ print_exec(r->right, indent, bracket);
+ break;
+ }
+ break;
+ }
+
+###### propagate exec cases
+ case Xref: {
+ struct ref *r = cast(ref, prog);
+ switch (r->action) {
+ case RefNew:
+ if (type && type->free != reference_free) {
+ type_err(c, "error: @new() can only be used with references, not %1",
+ prog, type, 0, NULL);
+ return NULL;
+ }
+ if (type && !r->reftype) {
+ r->reftype = type;
+ *perr |= Eretry;
+ }
+ return type;
+ case RefNil:
+ if (type && type->free != reference_free)
+ type_err(c, "error: @nil can only be used with reference, not %1",
+ prog, type, 0, NULL);
+ if (type && !r->reftype) {
+ r->reftype = type;
+ *perr |= Eretry;
+ }
+ return type;
+ case RefFree:
+ t = propagate_types(r->right, c, perr, NULL, 0);
+ if (t && t->free != reference_free)
+ type_err(c, "error: @free can only be assigned a reference, not %1",
+ prog, t, 0, NULL);
+ r->reftype = Tnone;
+ return Tnone;
+ }
+ break; // NOTEST
+ }
+
+
+###### interp exec cases
+ case Xref: {
+ struct ref *r = cast(ref, e);
+ switch (r->action) {
+ case RefNew:
+ if (r->reftype)
+ rv.ref = calloc(1, r->reftype->reference.referent->size);
+ rvtype = r->reftype;
+ break;
+ case RefNil:
+ rv.ref = NULL;
+ rvtype = r->reftype;
+ break;
+ case RefFree:
+ rv = interp_exec(c, r->right, &rvtype);
+ free_value(rvtype->reference.referent, rv.ref);
+ free(rv.ref);
+ rvtype = Tnone;
+ break;
+ }
+ break;
+ }
+
+###### free exec cases
+ case Xref: {
+ struct ref *r = cast(ref, e);
+ free_exec(r->right);
+ free(r);
+ break;
+ }
+
+###### Expressions: dereference
+
+###### Binode types
+ Deref,
+
+###### term grammar
+
+ | Term @ ${ {
+ struct binode *b = new(binode);
+ b->op = Deref;
+ b->left = $<Trm;
+ $0 = b;
+ } }$
+
+###### print binode cases
+ case Deref:
+ print_exec(b->left, -1, bracket);
+ printf("@");
+ break;
+
+###### propagate binode cases
+ case Deref:
+ /* left must be a reference, and we return what it refers to */
+ /* FIXME how can I pass the expected type down? */
+ t = propagate_types(b->left, c, perr, NULL, 0);
+ if (!t || t->free != reference_free)
+ type_err(c, "error: Cannot dereference %1", b, t, 0, NULL);
+ else
+ return t->reference.referent;
+ break;
+
+###### interp binode cases
+ case Deref: {
+ left = interp_exec(c, b->left, <ype);
+ lrv = left.ref;
+ rvtype = ltype->reference.referent;
+ break;
+ }
+
+
+#### Functions
+
+A function is a chunk of code which can be passed parameters and can
+return results. Each function has a type which includes the set of
+parameters and the return value. As yet these types cannot be declared
+separately from the function itself.
+
+The parameters can be specified either in parentheses as a ';' separated
+list, such as
+
+##### Example: function 1
+
+ func main(av:[ac::number]string; env:[envc::number]string)
+ code block
+
+or as an indented list of one parameter per line (though each line can
+be a ';' separated list)
+
+##### Example: function 2
+
+ func main
+ argv:[argc::number]string
+ env:[envc::number]string
+ do
+ code block
+
+In the first case a return type can follow the parentheses after a colon,
+in the second it is given on a line starting with the word `return`.
+
+##### Example: functions that return
+
+ func add(a:number; b:number): number
+ code block
+
+ func catenate
+ a: string
+ b: string
+ return string
+ do
+ code block
+
+Rather than returning a type, the function can specify a set of local
+variables to return as a struct. The values of these variables when the
+function exits will be provided to the caller. For this the return type
+is replaced with a block of result declarations, either in parentheses
+or bracketed by `return` and `do`.
+
+##### Example: functions returning multiple variables
+
+ func to_cartesian(rho:number; theta:number):(x:number; y:number)
+ x = .....
+ y = .....
+
+ func to_polar
+ x:number; y:number
+ return
+ rho:number
+ theta:number
+ do
+ rho = ....
+ theta = ....
+
+For constructing the lists we use a `List` binode, which will be
+further detailed when Expression Lists are introduced.
+
+###### type union fields
+
+ struct {
+ struct binode *params;
+ struct type *return_type;
+ struct variable *scope;
+ int inline_result; // return value is at start of 'local'
+ int local_size;
+ } function;
+
+###### value union fields
+ struct exec *function;
+
+###### type functions
+ void (*check_args)(struct parse_context *c, enum prop_err *perr,
+ struct type *require, struct exec *args);
+
+###### value functions
+
+ static void function_free(struct type *type, struct value *val)
+ {
+ free_exec(val->function);
+ val->function = NULL;
+ }
+
+ static int function_compat(struct type *require, struct type *have)
+ {
+ // FIXME can I do anything here yet?
+ return 0;
+ }
+
+ static void function_check_args(struct parse_context *c, enum prop_err *perr,
+ struct type *require, struct exec *args)
+ {
+ /* This should be 'compat', but we don't have a 'tuple' type to
+ * hold the type of 'args'
+ */
+ struct binode *arg = cast(binode, args);
+ struct binode *param = require->function.params;
+
+ while (param) {
+ struct var *pv = cast(var, param->left);
+ if (!arg) {
+ type_err(c, "error: insufficient arguments to function.",
+ args, NULL, 0, NULL);
+ break;
+ }
+ *perr = 0;
+ propagate_types(arg->left, c, perr, pv->var->type, 0);
+ param = cast(binode, param->right);
+ arg = cast(binode, arg->right);
+ }
+ if (arg)
+ type_err(c, "error: too many arguments to function.",
+ args, NULL, 0, NULL);
+ }
+
+ static void function_print(struct type *type, struct value *val, FILE *f)
+ {
+ print_exec(val->function, 1, 0);
+ }
+
+ static void function_print_type_decl(struct type *type, FILE *f)
+ {
+ struct binode *b;
+ fprintf(f, "(");
+ for (b = type->function.params; b; b = cast(binode, b->right)) {
+ struct variable *v = cast(var, b->left)->var;
+ fprintf(f, "%.*s%s", v->name->name.len, v->name->name.txt,
+ v->constant ? "::" : ":");
+ type_print(v->type, f);
+ if (b->right)
+ fprintf(f, "; ");
+ }
+ fprintf(f, ")");
+ if (type->function.return_type != Tnone) {
+ fprintf(f, ":");
+ if (type->function.inline_result) {
+ int i;
+ struct type *t = type->function.return_type;
+ fprintf(f, " (");
+ for (i = 0; i < t->structure.nfields; i++) {
+ struct field *fl = t->structure.fields + i;
+ if (i)
+ fprintf(f, "; ");
+ fprintf(f, "%.*s:", fl->name.len, fl->name.txt);
+ type_print(fl->type, f);
+ }
+ fprintf(f, ")");
+ } else
+ type_print(type->function.return_type, f);
+ }
+ fprintf(f, "\n");
}
-### Variables
+ static void function_free_type(struct type *t)
+ {
+ free_exec(t->function.params);
+ }
-Just as we used as `val` to wrap a value into an `exec`, we similarly
-need a `var` to wrap a `variable` into an exec. While each `val`
-contained a copy of the value, each `var` hold a link to the variable
-because it really is the same variable no matter where it appears.
+ static struct type function_prototype = {
+ .size = sizeof(void*),
+ .align = sizeof(void*),
+ .free = function_free,
+ .compat = function_compat,
+ .check_args = function_check_args,
+ .print = function_print,
+ .print_type_decl = function_print_type_decl,
+ .free_type = function_free_type,
+ };
-###### exec type
- Xvar,
+###### declare terminals
-###### ast
- struct var {
- struct exec;
- struct variable *var;
- };
+ $TERM func
+
+###### Binode types
+ List,
###### Grammar
- $*var
- Variable -> IDENTIFIER ${
- $0 = new(var);
- $0->var = find_variable(config, $1.txt);
+
+ $*variable
+ FuncName -> IDENTIFIER ${ {
+ struct variable *v = var_decl(c, $1.txt);
+ struct var *e = new_pos(var, $1);
+ e->var = v;
+ if (v) {
+ v->where_decl = e;
+ v->where_set = e;
+ $0 = v;
+ } else {
+ v = var_ref(c, $1.txt);
+ e->var = v;
+ type_err(c, "error: function '%v' redeclared",
+ e, NULL, 0, NULL);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ free_exec(e);
+ }
+ } }$
+
+ $*binode
+ Args -> ArgsLine NEWLINE ${ $0 = $<AL; }$
+ | Args ArgsLine NEWLINE ${ {
+ struct binode *b = $<AL;
+ struct binode **bp = &b;
+ while (*bp)
+ bp = (struct binode **)&(*bp)->left;
+ *bp = $<A;
+ $0 = b;
+ } }$
+
+ ArgsLine -> ${ $0 = NULL; }$
+ | Varlist ${ $0 = $<1; }$
+ | Varlist ; ${ $0 = $<1; }$
+
+ Varlist -> Varlist ; ArgDecl ${
+ $0 = new_pos(binode, $2);
+ $0->op = List;
+ $0->left = $<Vl;
+ $0->right = $<AD;
+ }$
+ | ArgDecl ${
+ $0 = new(binode);
+ $0->op = List;
+ $0->left = NULL;
+ $0->right = $<AD;
}$
-###### print exec cases
- case Xvar:
- {
- struct var *v = cast(var, e);
- printf("%.*s", v->var->name.len, v->var->name.txt);
+ $*var
+ ArgDecl -> IDENTIFIER : FormalType ${ {
+ struct variable *v = var_decl(c, $ID.txt);
+ $0 = new_pos(var, $ID);
+ $0->var = v;
+ v->where_decl = $0;
+ v->where_set = $0;
+ v->type = $<FT;
+ } }$
+
+##### Function calls
+
+A function call can appear either as an expression or as a statement.
+We use a new 'Funcall' binode type to link the function with a list of
+arguments, form with the 'List' nodes.
+
+We have already seen the "Term" which is how a function call can appear
+in an expression. To parse a function call into a statement we include
+it in the "SimpleStatement Grammar" which will be described later.
+
+###### Binode types
+ Funcall,
+
+###### term grammar
+ | Term ( ExpressionList ) ${ {
+ struct binode *b = new(binode);
+ b->op = Funcall;
+ b->left = $<T;
+ b->right = reorder_bilist($<EL);
+ $0 = b;
+ } }$
+ | Term ( ) ${ {
+ struct binode *b = new(binode);
+ b->op = Funcall;
+ b->left = $<T;
+ b->right = NULL;
+ $0 = b;
+ } }$
+
+###### SimpleStatement Grammar
+
+ | Term ( ExpressionList ) ${ {
+ struct binode *b = new(binode);
+ b->op = Funcall;
+ b->left = $<T;
+ b->right = reorder_bilist($<EL);
+ $0 = b;
+ } }$
+
+###### print binode cases
+
+ case Funcall:
+ do_indent(indent, "");
+ print_exec(b->left, -1, bracket);
+ printf("(");
+ for (b = cast(binode, b->right); b; b = cast(binode, b->right)) {
+ if (b->left) {
+ printf(" ");
+ print_exec(b->left, -1, bracket);
+ if (b->right)
+ printf(",");
+ }
+ }
+ printf(")");
+ if (indent >= 0)
+ printf("\n");
break;
+
+###### propagate binode cases
+
+ case Funcall: {
+ /* Every arg must match formal parameter, and result
+ * is return type of function
+ */
+ struct binode *args = cast(binode, b->right);
+ struct var *v = cast(var, b->left);
+
+ if (!v->var->type || v->var->type->check_args == NULL) {
+ type_err(c, "error: attempt to call a non-function.",
+ prog, NULL, 0, NULL);
+ return NULL;
+ }
+ *perr |= Eruntime;
+ v->var->type->check_args(c, perr, v->var->type, args);
+ if (v->var->type->function.inline_result)
+ *perr |= Emaycopy;
+ return v->var->type->function.return_type;
}
-###### propagate exec cases
+###### interp binode cases
- case Xvar:
- {
- struct var *var = cast(var, prog);
- if (var->var->val.vtype == Vunknown) {
- if (type != Vunknown && *ok != 0) {
- val_init(&var->var->val, type);
- *ok = 2;
- }
- return type;
+ case Funcall: {
+ struct var *v = cast(var, b->left);
+ struct type *t = v->var->type;
+ void *oldlocal = c->local;
+ int old_size = c->local_size;
+ void *local = calloc(1, t->function.local_size);
+ struct value *fbody = var_value(c, v->var);
+ struct binode *arg = cast(binode, b->right);
+ struct binode *param = t->function.params;
+
+ while (param) {
+ struct var *pv = cast(var, param->left);
+ struct type *vtype = NULL;
+ struct value val = interp_exec(c, arg->left, &vtype);
+ struct value *lval;
+ c->local = local; c->local_size = t->function.local_size;
+ lval = var_value(c, pv->var);
+ c->local = oldlocal; c->local_size = old_size;
+ memcpy(lval, &val, vtype->size);
+ param = cast(binode, param->right);
+ arg = cast(binode, arg->right);
}
- if (type == Vunknown)
- return var->var->val.vtype;
- if (type != var->var->val.vtype)
- *ok = 0;
- return type;
+ c->local = local; c->local_size = t->function.local_size;
+ if (t->function.inline_result && dtype) {
+ _interp_exec(c, fbody->function, NULL, NULL);
+ memcpy(dest, local, dtype->size);
+ rvtype = ret.type = NULL;
+ } else
+ rv = interp_exec(c, fbody->function, &rvtype);
+ c->local = oldlocal; c->local_size = old_size;
+ free(local);
+ break;
}
-###### interp exec cases
- case Xvar:
- return dup_value(cast(var, e)->var->val);
+## Complex executables: statements and expressions
-###### ast functions
+Now that we have types and values and variables and most of the basic
+Terms which provide access to these, we can explore the more complex
+code that combine all of these to get useful work done. Specifically
+statements and expressions.
- void free_var(struct var *v)
- {
- free(v);
+Expressions are various combinations of Terms. We will use operator
+precedence to ensure correct parsing. The simplest Expression is just a
+Term - others will follow.
+
+###### Grammar
+
+ $*exec
+ Expression -> Term ${ $0 = $<Term; }$
+ ## expression grammar
+
+### Expressions: Conditional
+
+Our first user of the `binode` will be conditional expressions, which
+is a bit odd as they actually have three components. That will be
+handled by having 2 binodes for each expression. The conditional
+expression is the lowest precedence operator which is why we define it
+first - to start the precedence list.
+
+Conditional expressions are of the form "value `if` condition `else`
+other_value". They associate to the right, so everything to the right
+of `else` is part of an else value, while only a higher-precedence to
+the left of `if` is the if values. Between `if` and `else` there is no
+room for ambiguity, so a full conditional expression is allowed in
+there.
+
+###### Binode types
+ CondExpr,
+
+###### declare terminals
+
+ $LEFT if $$ifelse
+
+###### expression grammar
+
+ | Expression if Expression else Expression $$ifelse ${ {
+ struct binode *b1 = new(binode);
+ struct binode *b2 = new(binode);
+ b1->op = CondExpr;
+ b1->left = $<3;
+ b1->right = b2;
+ b2->op = CondExpr;
+ b2->left = $<1;
+ b2->right = $<5;
+ $0 = b1;
+ } }$
+
+###### print binode cases
+
+ case CondExpr:
+ b2 = cast(binode, b->right);
+ if (bracket) printf("(");
+ print_exec(b2->left, -1, bracket);
+ printf(" if ");
+ print_exec(b->left, -1, bracket);
+ printf(" else ");
+ print_exec(b2->right, -1, bracket);
+ if (bracket) printf(")");
+ break;
+
+###### propagate binode cases
+
+ case CondExpr: {
+ /* cond must be Tbool, others must match */
+ struct binode *b2 = cast(binode, b->right);
+ struct type *t2;
+
+ propagate_types(b->left, c, perr, Tbool, 0);
+ t = propagate_types(b2->left, c, perr, type, 0);
+ t2 = propagate_types(b2->right, c, perr, type ?: t, 0);
+ return t ?: t2;
}
-###### free exec cases
- case Xvar: free_var(cast(var, e)); break;
+###### interp binode cases
+
+ case CondExpr: {
+ struct binode *b2 = cast(binode, b->right);
+ left = interp_exec(c, b->left, <ype);
+ if (left.bool)
+ rv = interp_exec(c, b2->left, &rvtype); // UNTESTED
+ else
+ rv = interp_exec(c, b2->right, &rvtype);
+ }
+ break;
+
+### Expression list
+
+We take a brief detour, now that we have expressions, to describe lists
+of expressions. These will be needed for function parameters and
+possibly other situations. They seem generic enough to introduce here
+to be used elsewhere.
+
+And ExpressionList will use the `List` type of `binode`, building up at
+the end. And place where they are used will probably call
+`reorder_bilist()` to get a more normal first/next arrangement.
+
+###### declare terminals
+ $TERM ,
+
+`List` execs have no implicit semantics, so they are never propagated or
+interpreted. The can be printed as a comma separate list, which is how
+they are parsed. Note they are also used for function formal parameter
+lists. In that case a separate function is used to print them.
+
+###### print binode cases
+ case List:
+ while (b) {
+ printf(" ");
+ print_exec(b->left, -1, bracket);
+ if (b->right)
+ printf(",");
+ b = cast(binode, b->right);
+ }
+ break;
+
+###### propagate binode cases
+ case List: abort(); // NOTEST
+###### interp binode cases
+ case List: abort(); // NOTEST
+
+###### Grammar
+
+ $*binode
+ ExpressionList -> ExpressionList , Expression ${
+ $0 = new(binode);
+ $0->op = List;
+ $0->left = $<1;
+ $0->right = $<3;
+ }$
+ | Expression ${
+ $0 = new(binode);
+ $0->op = List;
+ $0->left = NULL;
+ $0->right = $<1;
+ }$
### Expressions: Boolean
-Our first user of the `binode` will be expressions, and particularly
-Boolean expressions. As I haven't implemented precedence in the
-parser generator yet, we need different names from each precedence
-level used by expressions. The outer most or lowest level precedence
-are Boolean `or` `and`, and `not` which form and `Expression` our of `BTerm`s
-and `BFact`s.
+The next class of expressions to use the `binode` will be Boolean
+expressions. "`and then`" and "`or else`" are similar to `and` and `or`
+have same corresponding precendence. The difference is that they don't
+evaluate the second expression if not necessary.
###### Binode types
And,
+ AndThen,
Or,
+ OrElse,
Not,
-####### Grammar
-
- $*binode
- Expression -> Expression or BTerm ${
- $0 = new(binode);
- $0->op = Or;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | BTerm ${ $0 = $<1; }$
-
- BTerm -> BTerm and BFact ${
- $0 = new(binode);
- $0->op = And;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | BFact ${ $0 = $<1; }$
-
- BFact -> not BFact ${
- $0 = new(binode);
- $0->op = Not;
- $0->right = $<2;
- }$
- ## other BFact
+###### declare terminals
+ $LEFT or
+ $LEFT and
+ $LEFT not
+
+###### expression grammar
+ | Expression or Expression ${ {
+ struct binode *b = new(binode);
+ b->op = Or;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+ | Expression or else Expression ${ {
+ struct binode *b = new(binode);
+ b->op = OrElse;
+ b->left = $<1;
+ b->right = $<4;
+ $0 = b;
+ } }$
+
+ | Expression and Expression ${ {
+ struct binode *b = new(binode);
+ b->op = And;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+ | Expression and then Expression ${ {
+ struct binode *b = new(binode);
+ b->op = AndThen;
+ b->left = $<1;
+ b->right = $<4;
+ $0 = b;
+ } }$
+
+ | not Expression ${ {
+ struct binode *b = new(binode);
+ b->op = Not;
+ b->right = $<2;
+ $0 = b;
+ } }$
###### print binode cases
case And:
- print_exec(b->left, -1, 0);
+ if (bracket) printf("(");
+ print_exec(b->left, -1, bracket);
printf(" and ");
- print_exec(b->right, -1, 0);
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
+ break;
+ case AndThen:
+ if (bracket) printf("(");
+ print_exec(b->left, -1, bracket);
+ printf(" and then ");
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
break;
case Or:
- print_exec(b->left, -1, 0);
+ if (bracket) printf("(");
+ print_exec(b->left, -1, bracket);
printf(" or ");
- print_exec(b->right, -1, 0);
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
+ break;
+ case OrElse:
+ if (bracket) printf("(");
+ print_exec(b->left, -1, bracket);
+ printf(" or else ");
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
break;
case Not:
+ if (bracket) printf("(");
printf("not ");
- print_exec(b->right, -1, 0);
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
break;
###### propagate binode cases
case And:
+ case AndThen:
case Or:
+ case OrElse:
case Not:
- /* both must be Vbool, result is Vbool */
- propagate_types(b->left, Vbool, ok);
- propagate_types(b->right, Vbool, ok);
- if (type != Vbool && type != Vunknown)
- *ok = 0;
- return Vbool;
+ /* both must be Tbool, result is Tbool */
+ propagate_types(b->left, c, perr, Tbool, 0);
+ propagate_types(b->right, c, perr, Tbool, 0);
+ if (type && type != Tbool)
+ type_err(c, "error: %1 operation found where %2 expected", prog,
+ Tbool, 0, type);
+ return Tbool;
###### interp binode cases
case And:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
rv.bool = rv.bool && right.bool;
break;
+ case AndThen:
+ rv = interp_exec(c, b->left, &rvtype);
+ if (rv.bool)
+ rv = interp_exec(c, b->right, NULL);
+ break;
case Or:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
rv.bool = rv.bool || right.bool;
break;
+ case OrElse:
+ rv = interp_exec(c, b->left, &rvtype);
+ if (!rv.bool)
+ rv = interp_exec(c, b->right, NULL);
+ break;
case Not:
- rv = interp_exec(b->right);
+ rv = interp_exec(c, b->right, &rvtype);
rv.bool = !rv.bool;
break;
### Expressions: Comparison
-Of slightly higher precedence that Boolean expressions are
-Comparisons.
-A comparison takes arguments of any type, but the two types must be
-the same.
+Of slightly higher precedence that Boolean expressions are Comparisons.
+A comparison takes arguments of any comparable type, but the two types
+must be the same.
-To simplify the parsing we introduce an `eop` which can return an
-expression operator.
+To simplify the parsing we introduce an `eop` which can record an
+expression operator, and the `CMPop` non-terminal will match one of them.
###### ast
struct eop {
Eql,
NEql,
-###### other BFact
- | Expr CMPop Expr ${
- $0 = new(binode);
- $0->op = $2.op;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | Expr ${ $0 = $<1; }$
+###### declare terminals
+ $LEFT < > <= >= == != CMPop
+
+###### expression grammar
+ | Expression CMPop Expression ${ {
+ struct binode *b = new(binode);
+ b->op = $2.op;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
###### Grammar
$eop
- CMPop -> < ${ $0.op = Less; }$
- | > ${ $0.op = Gtr; }$
- | <= ${ $0.op = LessEq; }$
- | >= ${ $0.op = GtrEq; }$
- | == ${ $0.op = Eql; }$
- | != ${ $0.op = NEql; }$
+ CMPop -> < ${ $0.op = Less; }$
+ | > ${ $0.op = Gtr; }$
+ | <= ${ $0.op = LessEq; }$
+ | >= ${ $0.op = GtrEq; }$
+ | == ${ $0.op = Eql; }$
+ | != ${ $0.op = NEql; }$
###### print binode cases
case GtrEq:
case Eql:
case NEql:
- print_exec(b->left, -1, 0);
+ if (bracket) printf("(");
+ print_exec(b->left, -1, bracket);
switch(b->op) {
case Less: printf(" < "); break;
case LessEq: printf(" <= "); break;
case GtrEq: printf(" >= "); break;
case Eql: printf(" == "); break;
case NEql: printf(" != "); break;
- default: abort();
+ default: abort(); // NOTEST
}
- print_exec(b->right, -1, 0);
+ print_exec(b->right, -1, bracket);
+ if (bracket) printf(")");
break;
###### propagate binode cases
case GtrEq:
case Eql:
case NEql:
- /* Both must match, result is Vbool */
- t = propagate_types(b->left, Vunknown, ok);
- if (t != Vunknown)
- propagate_types(b->right, t, ok);
+ /* Both must match but not be labels, result is Tbool */
+ t = propagate_types(b->left, c, perr, NULL, 0);
+ if (t)
+ propagate_types(b->right, c, perr, t, 0);
else {
- t = propagate_types(b->right, Vunknown, ok);
- if (t != Vunknown)
- t = propagate_types(b->left, t, ok);
+ t = propagate_types(b->right, c, perr, NULL, 0); // UNTESTED
+ if (t) // UNTESTED
+ t = propagate_types(b->left, c, perr, t, 0); // UNTESTED
}
- if (type != Vbool && type != Vunknown)
- *ok = 0;
- return Vbool;
+ if (!type_compat(type, Tbool, 0))
+ type_err(c, "error: Comparison returns %1 but %2 expected", prog,
+ Tbool, rules, type);
+ return Tbool;
###### interp binode cases
case Less:
case NEql:
{
int cmp;
- left = interp_exec(b->left);
- right = interp_exec(b->right);
- cmp = value_cmp(left, right);
- rv.vtype = Vbool;
+ left = interp_exec(c, b->left, <ype);
+ right = interp_exec(c, b->right, &rtype);
+ cmp = value_cmp(ltype, rtype, &left, &right);
+ rvtype = Tbool;
switch (b->op) {
case Less: rv.bool = cmp < 0; break;
case LessEq: rv.bool = cmp <= 0; break;
case GtrEq: rv.bool = cmp >= 0; break;
case Eql: rv.bool = cmp == 0; break;
case NEql: rv.bool = cmp != 0; break;
- default: rv.bool = 0; break;
+ default: rv.bool = 0; break; // NOTEST
}
break;
}
-### Expressions: The rest
+### Expressions: Arithmetic etc.
+
+The remaining expressions with the highest precedence are arithmetic,
+string concatenation, string conversion, and testing. String concatenation
+(`++`) has the same precedence as multiplication and division, but lower
+than the uniary.
-The remaining expressions with the highest precedence are arithmetic
-and string concatenation. There are `Expr`, `Term`, and `Factor`.
-The `Factor` is where the `Value` and `Variable` that we already have
-are included.
+Testing comes in two forms. A single question mark (`?`) is a uniary
+operator which converts come types into Boolean. The general meaning is
+"is this a value value" and there will be more uses as the language
+develops. A double questionmark (`??`) is a binary operator (Choose),
+with same precedence as multiplication, which returns the LHS if it
+tests successfully, else returns the RHS.
+
+String conversion is a temporary feature until I get a better type
+system. `$` is a prefix operator which expects a string and returns
+a number.
`+` and `-` are both infix and prefix operations (where they are
absolute value and negation). These have different operator names.
We also have a 'Bracket' operator which records where parentheses were
-found. This make it easy to reproduce these when printing. Once
-precedence is handled better I might be able to discard this.
+found. This makes it easy to reproduce these when printing. Possibly I
+should only insert brackets were needed for precedence. Putting
+parentheses around an expression converts it into a Term,
###### Binode types
Plus, Minus,
- Times, Divide,
- Concat,
- Absolute, Negate,
+ Times, Divide, Rem,
+ Concat, Choose,
+ Absolute, Negate, Test,
+ StringConv,
Bracket,
-###### Grammar
-
- $*binode
- Expr -> Expr Eop Term ${
- $0 = new(binode);
- $0->op = $2.op;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | Term ${ $0 = $<1; }$
-
- Term -> Term Top Factor ${
- $0 = new(binode);
- $0->op = $2.op;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | Factor ${ $0 = $<1; }$
+###### declare terminals
+ $LEFT + - Eop
+ $LEFT * / % ++ ?? Top
+ $LEFT Uop $ ?
+ $TERM ( )
+
+###### expression grammar
+ | Expression Eop Expression ${ {
+ struct binode *b = new(binode);
+ b->op = $2.op;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+
+ | Expression Top Expression ${ {
+ struct binode *b = new(binode);
+ b->op = $2.op;
+ b->left = $<1;
+ b->right = $<3;
+ $0 = b;
+ } }$
+
+ | Uop Expression ${ {
+ struct binode *b = new(binode);
+ b->op = $1.op;
+ b->right = $<2;
+ $0 = b;
+ } }$
+
+###### term grammar
+
+ | ( Expression ) ${ {
+ struct binode *b = new_pos(binode, $1);
+ b->op = Bracket;
+ b->right = $<2;
+ $0 = b;
+ } }$
- Factor -> ( Expression ) ${
- $0 = new(binode);
- $0->op = Bracket;
- $0->right = $<2;
- }$
- | Uop Factor ${
- $0 = new(binode);
- $0->op = $1.op;
- $0->right = $<2;
- }$
- | Value ${ $0 = (struct binode *)$<1; }$
- | Variable ${ $0 = (struct binode *)$<1; }$
+###### Grammar
$eop
- Eop -> + ${ $0.op = Plus; }$
- | - ${ $0.op = Minus; }$
+ Eop -> + ${ $0.op = Plus; }$
+ | - ${ $0.op = Minus; }$
- Uop -> + ${ $0.op = Absolute; }$
- | - ${ $0.op = Negate; }$
+ Uop -> + ${ $0.op = Absolute; }$
+ | - ${ $0.op = Negate; }$
+ | $ ${ $0.op = StringConv; }$
+ | ? ${ $0.op = Test; }$
- Top -> * ${ $0.op = Times; }$
- | / ${ $0.op = Divide; }$
- | ++ ${ $0.op = Concat; }$
+ Top -> * ${ $0.op = Times; }$
+ | / ${ $0.op = Divide; }$
+ | % ${ $0.op = Rem; }$
+ | ++ ${ $0.op = Concat; }$
+ | ?? ${ $0.op = Choose; }$
###### print binode cases
case Plus:
case Times:
case Divide:
case Concat:
- print_exec(b->left, indent, 0);
+ case Rem:
+ case Choose:
+ if (bracket) printf("(");
+ print_exec(b->left, indent, bracket);
switch(b->op) {
- case Plus: printf(" + "); break;
- case Minus: printf(" - "); break;
- case Times: printf(" * "); break;
- case Divide: printf(" / "); break;
- case Concat: printf(" ++ "); break;
- default: abort();
- }
- print_exec(b->right, indent, 0);
+ case Plus: fputs(" + ", stdout); break;
+ case Minus: fputs(" - ", stdout); break;
+ case Times: fputs(" * ", stdout); break;
+ case Divide: fputs(" / ", stdout); break;
+ case Rem: fputs(" % ", stdout); break;
+ case Concat: fputs(" ++ ", stdout); break;
+ case Choose: fputs(" ?? ", stdout); break;
+ default: abort(); // NOTEST
+ } // NOTEST
+ print_exec(b->right, indent, bracket);
+ if (bracket) printf(")");
break;
case Absolute:
- printf("+");
- print_exec(b->right, indent, 0);
- break;
case Negate:
- printf("-");
- print_exec(b->right, indent, 0);
+ case StringConv:
+ case Test:
+ if (bracket) printf("(");
+ switch (b->op) {
+ case Absolute: fputs("+", stdout); break;
+ case Negate: fputs("-", stdout); break;
+ case StringConv: fputs("$", stdout); break;
+ case Test: fputs("?", stdout); break;
+ default: abort(); // NOTEST
+ } // NOTEST
+ print_exec(b->right, indent, bracket);
+ if (bracket) printf(")");
break;
case Bracket:
printf("(");
- print_exec(b->right, indent, 0);
+ print_exec(b->right, indent, bracket);
printf(")");
break;
case Plus:
case Minus:
case Times:
+ case Rem:
case Divide:
- /* both must be numbers, result is Vnum */
+ /* both must be numbers, result is Tnum */
case Absolute:
case Negate:
/* as propagate_types ignores a NULL,
* unary ops fit here too */
- propagate_types(b->left, Vnum, ok);
- propagate_types(b->right, Vnum, ok);
- if (type != Vnum && type != Vunknown)
- *ok = 0;
- return Vnum;
+ propagate_types(b->left, c, perr, Tnum, 0);
+ propagate_types(b->right, c, perr, Tnum, 0);
+ if (!type_compat(type, Tnum, 0))
+ type_err(c, "error: Arithmetic returns %1 but %2 expected", prog,
+ Tnum, rules, type);
+ return Tnum;
case Concat:
- /* both must be Vstr, result is Vstr */
- propagate_types(b->left, Vstr, ok);
- propagate_types(b->right, Vstr, ok);
- if (type != Vstr && type != Vunknown)
- *ok = 0;
- return Vstr;
+ /* both must be Tstr, result is Tstr */
+ propagate_types(b->left, c, perr, Tstr, 0);
+ propagate_types(b->right, c, perr, Tstr, 0);
+ if (!type_compat(type, Tstr, 0))
+ type_err(c, "error: Concat returns %1 but %2 expected", prog,
+ Tstr, rules, type);
+ return Tstr;
+
+ case StringConv:
+ /* op must be string, result is number */
+ propagate_types(b->left, c, perr, Tstr, 0);
+ if (!type_compat(type, Tnum, 0))
+ type_err(c, // UNTESTED
+ "error: Can only convert string to number, not %1",
+ prog, type, 0, NULL);
+ return Tnum;
+
+ case Test:
+ /* LHS must support ->test, result is Tbool */
+ t = propagate_types(b->right, c, perr, NULL, 0);
+ if (!t || !t->test)
+ type_err(c, "error: '?' requires a testable value, not %1",
+ prog, t, 0, NULL);
+ return Tbool;
+
+ case Choose:
+ /* LHS and RHS must match and are returned. Must support
+ * ->test
+ */
+ t = propagate_types(b->left, c, perr, type, rules);
+ t = propagate_types(b->right, c, perr, t, rules);
+ if (t && t->test == NULL)
+ type_err(c, "error: \"??\" requires a testable value, not %1",
+ prog, t, 0, NULL);
+ return t;
case Bracket:
- return propagate_types(b->right, type, ok);
+ return propagate_types(b->right, c, perr, type, 0);
###### interp binode cases
case Plus:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
mpq_add(rv.num, rv.num, right.num);
break;
case Minus:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
mpq_sub(rv.num, rv.num, right.num);
break;
case Times:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
mpq_mul(rv.num, rv.num, right.num);
break;
case Divide:
- rv = interp_exec(b->left);
- right = interp_exec(b->right);
+ rv = interp_exec(c, b->left, &rvtype);
+ right = interp_exec(c, b->right, &rtype);
mpq_div(rv.num, rv.num, right.num);
break;
+ case Rem: {
+ mpz_t l, r, rem;
+
+ left = interp_exec(c, b->left, <ype);
+ right = interp_exec(c, b->right, &rtype);
+ mpz_init(l); mpz_init(r); mpz_init(rem);
+ mpz_tdiv_q(l, mpq_numref(left.num), mpq_denref(left.num));
+ mpz_tdiv_q(r, mpq_numref(right.num), mpq_denref(right.num));
+ mpz_tdiv_r(rem, l, r);
+ val_init(Tnum, &rv);
+ mpq_set_z(rv.num, rem);
+ mpz_clear(r); mpz_clear(l); mpz_clear(rem);
+ rvtype = ltype;
+ break;
+ }
case Negate:
- rv = interp_exec(b->right);
+ rv = interp_exec(c, b->right, &rvtype);
mpq_neg(rv.num, rv.num);
break;
case Absolute:
- rv = interp_exec(b->right);
+ rv = interp_exec(c, b->right, &rvtype);
mpq_abs(rv.num, rv.num);
break;
case Bracket:
- rv = interp_exec(b->right);
+ rv = interp_exec(c, b->right, &rvtype);
break;
case Concat:
- left = interp_exec(b->left);
- right = interp_exec(b->right);
- rv.vtype = Vstr;
+ left = interp_exec(c, b->left, <ype);
+ right = interp_exec(c, b->right, &rtype);
+ rvtype = Tstr;
rv.str = text_join(left.str, right.str);
break;
+ case StringConv:
+ right = interp_exec(c, b->right, &rvtype);
+ rtype = Tstr;
+ rvtype = Tnum;
+
+ struct text tx = right.str;
+ char tail[3];
+ int neg = 0;
+ if (tx.txt[0] == '-') {
+ neg = 1; // UNTESTED
+ tx.txt++; // UNTESTED
+ tx.len--; // UNTESTED
+ }
+ if (number_parse(rv.num, tail, tx) == 0)
+ mpq_init(rv.num); // UNTESTED
+ else if (neg)
+ mpq_neg(rv.num, rv.num); // UNTESTED
+ if (tail[0])
+ printf("Unsupported suffix: %.*s\n", tx.len, tx.txt); // UNTESTED
+
+ break;
+ case Test:
+ right = interp_exec(c, b->right, &rtype);
+ rvtype = Tbool;
+ rv.bool = !!rtype->test(rtype, &right);
+ break;
+ case Choose:
+ left = interp_exec(c, b->left, <ype);
+ if (ltype->test(ltype, &left)) {
+ rv = left;
+ rvtype = ltype;
+ ltype = NULL;
+ } else
+ rv = interp_exec(c, b->right, &rvtype);
+ break;
+
+###### value functions
+
+ static struct text text_join(struct text a, struct text b)
+ {
+ struct text rv;
+ rv.len = a.len + b.len;
+ rv.txt = malloc(rv.len);
+ memcpy(rv.txt, a.txt, a.len);
+ memcpy(rv.txt+a.len, b.txt, b.len);
+ return rv;
+ }
### Blocks, Statements, and Statement lists.
Now that we have expressions out of the way we need to turn to
statements. There are simple statements and more complex statements.
-Simple statements do not contain newlines, complex statements do.
+Simple statements do not contain (syntactic) newlines, complex statements do.
Statements often come in sequences and we have corresponding simple
statement lists and complex statement lists.
The later comprise complex statements and simple statement lists. They are
separated by newlines. Thus the semicolon is only used to separate
simple statements on the one line. This may be overly restrictive,
-but I'm not sure I every want a complex statement to share a line with
+but I'm not sure I ever want a complex statement to share a line with
anything else.
Note that a simple statement list can still use multiple lines if
A simple statement list needs no extra syntax. A complex statement
list has two syntactic forms. It can be enclosed in braces (much like
-C blocks), or it can be introduced by a colon and continue until an
+C blocks), or it can be introduced by an indent and continue until an
unindented newline (much like Python blocks). With this extra syntax
it is referred to as a block.
The only stand-alone statement we introduce at this stage is `pass`
which does nothing and is represented as a `NULL` pointer in a `Block`
-list.
+list. Other stand-alone statements will follow once the infrastructure
+is in-place.
+
+As many statements will use binodes, we declare a binode pointer 'b' in
+the common header for all reductions to use.
+
+###### Parser: reduce
+ struct binode *b;
###### Binode types
Block,
###### Grammar
- $void
- OptNL -> Newlines
- |
-
- Newlines -> NEWLINE
- | Newlines NEWLINE
+ $TERM { } ;
$*binode
- Open -> {
- | NEWLINE {
- Close -> }
- | NEWLINE }
- Block -> Open Statementlist Close ${ $0 = $<2; }$
- | Open SimpleStatements } ${ $0 = reorder_bilist($<2); }$
- | : Statementlist ${ $0 = $<2; }$
- | : SimpleStatements ${ $0 = reorder_bilist($<2); }$
-
- Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<1); }$
+ Block -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
+ | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
+ | SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
+ | SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
+ | IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
+
+ OpenBlock -> OpenScope { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
+ | OpenScope { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
+ | OpenScope SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
+ | OpenScope SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
+ | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
+
+ UseBlock -> { OpenScope IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
+ | { OpenScope SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
+ | IN OpenScope OptNL Statementlist OUT ${ $0 = $<Sl; }$
+
+ ColonBlock -> { IN OptNL Statementlist OUT OptNL } ${ $0 = $<Sl; }$
+ | { SimpleStatements } ${ $0 = reorder_bilist($<SS); }$
+ | : SimpleStatements ; ${ $0 = reorder_bilist($<SS); }$
+ | : SimpleStatements EOL ${ $0 = reorder_bilist($<SS); }$
+ | : IN OptNL Statementlist OUT ${ $0 = $<Sl; }$
+
+ Statementlist -> ComplexStatements ${ $0 = reorder_bilist($<CS); }$
ComplexStatements -> ComplexStatements ComplexStatement ${
+ if ($2 == NULL) {
+ $0 = $<1;
+ } else {
+ $0 = new(binode);
+ $0->op = Block;
+ $0->left = $<1;
+ $0->right = $<2;
+ }
+ }$
+ | ComplexStatement ${
+ if ($1 == NULL) {
+ $0 = NULL;
+ } else {
+ $0 = new(binode);
+ $0->op = Block;
+ $0->left = NULL;
+ $0->right = $<1;
+ }
+ }$
+
+ $*exec
+ ComplexStatement -> SimpleStatements Newlines ${
+ $0 = reorder_bilist($<SS);
+ }$
+ | SimpleStatements ; Newlines ${
+ $0 = reorder_bilist($<SS);
+ }$
+ ## ComplexStatement Grammar
+
+ $*binode
+ SimpleStatements -> SimpleStatements ; SimpleStatement ${
$0 = new(binode);
$0->op = Block;
$0->left = $<1;
- $0->right = $<2;
- }$
- | ComplexStatements NEWLINE ${ $0 = $<1; }$
- | ComplexStatement ${
+ $0->right = $<3;
+ }$
+ | SimpleStatement ${
$0 = new(binode);
$0->op = Block;
$0->left = NULL;
$0->right = $<1;
- }$
+ }$
+ $TERM pass
$*exec
- ComplexStatement -> SimpleStatements NEWLINE ${
- $0 = reorder_bilist($<1);
- }$
- ## ComplexStatement Grammar
-
- $*binode
- SimpleStatements -> SimpleStatements ; SimpleStatement ${
- $0 = new(binode);
- $0->op = Block;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | SimpleStatement ${
- $0 = new(binode);
- $0->op = Block;
- $0->left = NULL;
- $0->right = $<1;
- }$
- | SimpleStatements ; ${ $0 = $<1; }$
-
SimpleStatement -> pass ${ $0 = NULL; }$
- ## SimpleStatement Grammar
+ | ERROR ${ tok_err(c, "Syntax error in statement", &$1); }$
+ ## SimpleStatement Grammar
###### print binode cases
case Block:
if (indent < 0) {
// simple statement
- if (b->left == NULL)
- printf("pass");
+ if (b->left == NULL) // UNTESTED
+ printf("pass"); // UNTESTED
else
- print_exec(b->left, indent, 0);
- if (b->right) {
- printf("; ");
- print_exec(b->right, indent, 0);
+ print_exec(b->left, indent, bracket); // UNTESTED
+ if (b->right) { // UNTESTED
+ printf("; "); // UNTESTED
+ print_exec(b->right, indent, bracket); // UNTESTED
}
} else {
// block, one per line
###### propagate binode cases
case Block:
{
- /* If any statement returns something other then Vnone
- * then all such must return same type.
- * As each statement may be Vnone or something else,
- * we must always pass Vunknown down, otherwise an incorrect
- * error might occur.
+ /* If any statement returns something other than Tnone
+ * or Tbool then all such must return same type.
+ * As each statement may be Tnone or something else,
+ * we must always pass NULL (unknown) down, otherwise an incorrect
+ * error might occur. We never return Tnone unless it is
+ * passed in.
*/
struct binode *e;
for (e = b; e; e = cast(binode, e->right)) {
- t = propagate_types(e->left, Vunknown, ok);
- if (t != Vunknown && t != Vnone) {
- if (type == Vunknown)
+ t = propagate_types(e->left, c, perr, NULL, rules);
+ if ((rules & Rboolok) && (t == Tbool || t == Tnone))
+ t = NULL;
+ if (t == Tnone && e->right)
+ /* Only the final statement *must* return a value
+ * when not Rboolok
+ */
+ t = NULL;
+ if (t) {
+ if (!type)
type = t;
else if (t != type)
- *ok = 0;
+ type_err(c, "error: expected %1, found %2",
+ e->left, type, rules, t);
}
}
return type;
###### interp binode cases
case Block:
- while (rv.vtype == Vnone &&
+ while (rvtype == Tnone &&
b) {
if (b->left)
- rv = interp_exec(b->left);
+ rv = interp_exec(c, b->left, &rvtype);
b = cast(binode, b->right);
}
break;
expressions and prints the values separated by spaces and terminated
by a newline. No control of formatting is possible.
-`print` faces the same list-ordering issue as blocks, and uses the
-same solution.
+`print` uses `ExpressionList` to collect the expressions and stores them
+on the left side of a `Print` binode unlessthere is a trailing comma
+when the list is stored on the `right` side and no trailing newline is
+printed.
###### Binode types
Print,
+##### declare terminals
+ $TERM print
+
###### SimpleStatement Grammar
| print ExpressionList ${
- $0 = reorder_bilist($<2);
- }$
- | print ExpressionList , ${
- $0 = new(binode);
- $0->op = Print;
- $0->right = NULL;
- $0->left = $<2;
- $0 = reorder_bilist($0);
+ $0 = b = new_pos(binode, $1);
+ b->op = Print;
+ b->right = NULL;
+ b->left = reorder_bilist($<EL);
}$
+ | print ExpressionList , ${ {
+ $0 = b = new_pos(binode, $1);
+ b->op = Print;
+ b->right = reorder_bilist($<EL);
+ b->left = NULL;
+ } }$
| print ${
- $0 = new(binode);
- $0->op = Print;
- $0->right = NULL;
+ $0 = b = new_pos(binode, $1);
+ b->op = Print;
+ b->left = NULL;
+ b->right = NULL;
}$
-###### Grammar
-
- $*binode
- ExpressionList -> ExpressionList , Expression ${
- $0 = new(binode);
- $0->op = Print;
- $0->left = $<1;
- $0->right = $<3;
- }$
- | Expression ${
- $0 = new(binode);
- $0->op = Print;
- $0->left = NULL;
- $0->right = $<1;
- }$
-
###### print binode cases
case Print:
do_indent(indent, "print");
- while (b) {
- if (b->left) {
- printf(" ");
- print_exec(b->left, -1, 0);
- if (b->right)
- printf(",");
- }
- b = cast(binode, b->right);
- }
+ if (b->right) {
+ print_exec(b->right, -1, bracket);
+ printf(",");
+ } else
+ print_exec(b->left, -1, bracket);
if (indent >= 0)
printf("\n");
break;
case Print:
/* don't care but all must be consistent */
- propagate_types(b->left, Vunknown, ok);
- propagate_types(b->right, Vunknown, ok);
+ if (b->left)
+ b = cast(binode, b->left);
+ else
+ b = cast(binode, b->right);
+ while (b) {
+ propagate_types(b->left, c, perr, NULL, 0);
+ b = cast(binode, b->right);
+ }
break;
###### interp binode cases
case Print:
{
- char sep = 0;
- int eol = 1;
- for ( ; b; b = cast(binode, b->right))
- if (b->left) {
- if (sep)
- putchar(sep);
- left = interp_exec(b->left);
- print_value(left);
- free_value(left);
- if (b->right)
- sep = ' ';
- } else if (sep)
- eol = 0;
- left.vtype = Vnone;
- if (eol)
+ struct binode *b2 = cast(binode, b->left);
+ if (!b2)
+ b2 = cast(binode, b->right);
+ for (; b2; b2 = cast(binode, b2->right)) {
+ left = interp_exec(c, b2->left, <ype);
+ print_value(ltype, &left, stdout);
+ free_value(ltype, &left);
+ if (b2->right)
+ putchar(' ');
+ }
+ if (b->right == NULL)
printf("\n");
+ ltype = Tnone;
break;
}
###### Assignment statement
-An assignment will assign a value to a variable. The analysis phase
-ensures that the type will be correct so the interpreted just needs to
-perform the calculation.
+An assignment will assign a value to a variable, providing it hasn't
+been declared as a constant. The analysis phase ensures that the type
+will be correct so the interpreter just needs to perform the
+calculation. There is a form of assignment which declares a new
+variable as well as assigning a value. If a name is used before
+it is declared, it is assumed to be a global constant which are allowed to
+be declared at any time.
###### Binode types
Assign,
+ Declare,
+
+###### declare terminals
+ $TERM =
###### SimpleStatement Grammar
- | Variable = Expression ${
- $0 = new(binode);
- $0->op = Assign;
- $0->left = $<1;
- $0->right =$<3;
- }$
+ | Term = Expression ${
+ $0 = b= new(binode);
+ b->op = Assign;
+ b->left = $<1;
+ b->right = $<3;
+ }$
+ | VariableDecl = Expression ${
+ $0 = b= new(binode);
+ b->op = Declare;
+ b->left = $<1;
+ b->right =$<3;
+ }$
+
+ | VariableDecl ${
+ if ($1->var->where_set == NULL) {
+ type_err(c,
+ "Variable declared with no type or value: %v",
+ $1, NULL, 0, NULL);
+ free_var($1);
+ } else {
+ $0 = b = new(binode);
+ b->op = Declare;
+ b->left = $<1;
+ b->right = NULL;
+ }
+ }$
###### print binode cases
case Assign:
do_indent(indent, "");
- print_exec(b->left, indent, 0);
+ print_exec(b->left, -1, bracket);
printf(" = ");
- print_exec(b->right, indent, 0);
+ print_exec(b->right, -1, bracket);
if (indent >= 0)
printf("\n");
break;
+ case Declare:
+ {
+ struct variable *v = cast(var, b->left)->var;
+ do_indent(indent, "");
+ print_exec(b->left, -1, bracket);
+ if (cast(var, b->left)->var->constant) {
+ printf("::");
+ if (v->explicit_type) {
+ type_print(v->type, stdout);
+ printf(" ");
+ }
+ } else {
+ printf(":");
+ if (v->explicit_type) {
+ type_print(v->type, stdout);
+ printf(" ");
+ }
+ }
+ if (b->right) {
+ printf("= ");
+ print_exec(b->right, -1, bracket);
+ }
+ if (indent >= 0)
+ printf("\n");
+ }
+ break;
+
###### propagate binode cases
case Assign:
- /* Both must match, result is Vnone */
- t = propagate_types(b->left, Vunknown, ok);
- if (t != Vunknown)
- propagate_types(b->right, t, ok);
- else {
- t = propagate_types(b->right, Vunknown, ok);
- if (t != Vunknown)
- t = propagate_types(b->left, t, ok);
+ case Declare:
+ /* Both must match and not be labels,
+ * Type must support 'dup',
+ * For Assign, left must not be constant.
+ * result is Tnone
+ */
+ t = propagate_types(b->left, c, perr, NULL,
+ (b->op == Assign ? Rnoconstant : 0));
+ if (!b->right)
+ return Tnone;
+
+ if (t) {
+ if (propagate_types(b->right, c, perr, t, 0) != t)
+ if (b->left->type == Xvar)
+ type_err(c, "info: variable '%v' was set as %1 here.",
+ cast(var, b->left)->var->where_set, t, rules, NULL);
+ } else {
+ t = propagate_types(b->right, c, perr, NULL, 0);
+ if (t)
+ propagate_types(b->left, c, perr, t,
+ (b->op == Assign ? Rnoconstant : 0));
}
- return Vnone;
+ if (t && t->dup == NULL && !(*perr & Emaycopy))
+ type_err(c, "error: cannot assign value of type %1", b, t, 0, NULL);
+ return Tnone;
+
+ break;
###### interp binode cases
case Assign:
+ lleft = linterp_exec(c, b->left, <ype);
+ if (lleft)
+ dinterp_exec(c, b->right, lleft, ltype, 1);
+ ltype = Tnone;
+ break;
+
+ case Declare:
{
struct variable *v = cast(var, b->left)->var;
- right = interp_exec(b->right);
- free_value(v->val);
- v->val = right;
- right.vtype = Vunknown;
+ struct value *val;
+ v = v->merged;
+ val = var_value(c, v);
+ if (v->type->prepare_type)
+ v->type->prepare_type(c, v->type, 0);
+ if (b->right)
+ dinterp_exec(c, b->right, val, v->type, 0);
+ else
+ val_init(v->type, val);
break;
}
### The `use` statement
-The `use` statement is the last "simple" statement. It is needed when
-the condition in a conditional statement is a block. `use` works much
-like `return` in C, but only completes the `condition`, not the whole
-function.
+The `use` statement is the last "simple" statement. It is needed when a
+statement block can return a value. This includes the body of a
+function which has a return type, and the "condition" code blocks in
+`if`, `while`, and `switch` statements.
###### Binode types
Use,
+###### declare terminals
+ $TERM use
+
###### SimpleStatement Grammar
| use Expression ${
- $0 = new(binode);
- $0->op = Use;
- $0->right = $<2;
+ $0 = b = new_pos(binode, $1);
+ b->op = Use;
+ b->right = $<2;
}$
###### print binode cases
case Use:
do_indent(indent, "use ");
- print_exec(b->right, -1, 0);
+ print_exec(b->right, -1, bracket);
if (indent >= 0)
printf("\n");
break;
case Use:
/* result matches value */
- return propagate_types(b->right, type, ok);
+ return propagate_types(b->right, c, perr, type, 0);
###### interp binode cases
case Use:
- rv = interp_exec(b->right);
+ rv = interp_exec(c, b->right, &rvtype);
break;
### The Conditional Statement
-This is the biggy and currently the only complex statement.
-This subsumes `if`, `while`, `do/while`, `switch`, and some part of
-`for`. It is comprised of a number of parts, all of which are
-optional though set combinations apply.
+This is the biggy and currently the only complex statement. This
+subsumes `if`, `while`, `do/while`, `switch`, and some parts of `for`.
+It is comprised of a number of parts, all of which are optional though
+set combinations apply. Each part is (usually) a key word (`then` is
+sometimes optional) followed by either an expression or a code block,
+except the `casepart` which is a "key word and an expression" followed
+by a code block. The code-block option is valid for all parts and,
+where an expression is also allowed, the code block can use the `use`
+statement to report a value. If the code block does not report a value
+the effect is similar to reporting `True`.
+
+The `else` and `case` parts, as well as `then` when combined with
+`if`, can contain a `use` statement which will apply to some
+containing conditional statement. `for` parts, `do` parts and `then`
+parts used with `for` can never contain a `use`, except in some
+subordinate conditional statement.
If there is a `forpart`, it is executed first, only once.
If there is a `dopart`, then it is executed repeatedly providing
always that the `condpart` or `cond`, if present, does not return a non-True
value. `condpart` can fail to return any value if it simply executes
-to completion. This is treated the same as returning True.
+to completion. This is treated the same as returning `True`.
If there is a `thenpart` it will be executed whenever the `condpart`
-or `cond` returns True (or does not return), but this will happen
+or `cond` returns True (or does not return any value), but this will happen
*after* `dopart` (when present).
If `elsepart` is present it will be executed at most once when the
-condition returns False. If there are any `casepart`s, they will be
+condition returns `False` or some value that isn't `True` and isn't
+matched by any `casepart`. If there are any `casepart`s, they will be
executed when the condition returns a matching value.
The particular sorts of values allowed in case parts has not yet been
-determined in the language design.
+determined in the language design, so nothing is prohibited.
+
+The various blocks in this complex statement potentially provide scope
+for variables as described earlier. Each such block must include the
+"OpenScope" nonterminal before parsing the block, and must call
+`var_block_close()` when closing the block.
+
+The code following "`if`", "`switch`" and "`for`" does not get its own
+scope, but is in a scope covering the whole statement, so names
+declared there cannot be redeclared elsewhere. Similarly the
+condition following "`while`" is in a scope the covers the body
+("`do`" part) of the loop, and which does not allow conditional scope
+extension. Code following "`then`" (both looping and non-looping),
+"`else`" and "`case`" each get their own local scope.
+
+The type requirements on the code block in a `whilepart` are quite
+unusal. It is allowed to return a value of some identifiable type, in
+which case the loop aborts and an appropriate `casepart` is run, or it
+can return a Boolean, in which case the loop either continues to the
+`dopart` (on `True`) or aborts and runs the `elsepart` (on `False`).
+This is different both from the `ifpart` code block which is expected to
+return a Boolean, or the `switchpart` code block which is expected to
+return the same type as the casepart values. The correct analysis of
+the type of the `whilepart` code block is the reason for the
+`Rboolok` flag which is passed to `propagate_types()`.
+
+The `cond_statement` cannot fit into a `binode` so a new `exec` is
+defined. As there are two scopes which cover multiple parts - one for
+the whole statement and one for "while" and "do" - and as we will use
+the 'struct exec' to track scopes, we actually need two new types of
+exec. One is a `binode` for the looping part, the rest is the
+`cond_statement`. The `cond_statement` will use an auxilliary `struct
+casepart` to track a list of case parts.
-The cond_statement cannot fit into a `binode` so a new `exec` is
-defined.
+###### Binode types
+ Loop
###### exec type
Xcond_statement,
};
struct cond_statement {
struct exec;
- struct exec *forpart, *condpart, *dopart, *thenpart, *elsepart;
+ struct exec *forpart, *condpart, *thenpart, *elsepart;
+ struct binode *looppart;
struct casepart *casepart;
};
}
}
- void free_cond_statement(struct cond_statement *s)
+ static void free_cond_statement(struct cond_statement *s)
{
if (!s)
return;
free_exec(s->forpart);
free_exec(s->condpart);
- free_exec(s->dopart);
+ free_exec(s->looppart);
free_exec(s->thenpart);
free_exec(s->elsepart);
free_casepart(s->casepart);
###### ComplexStatement Grammar
| CondStatement ${ $0 = $<1; }$
+###### declare terminals
+ $TERM for then while do
+ $TERM else
+ $TERM switch case
+
###### Grammar
$*cond_statement
- CondStatement -> ForThen WhilePart CondSuffix ${
- $0 = $<3;
- $0->forpart = $1.forpart; $1.forpart = NULL;
- $0->thenpart = $1.thenpart; $1.thenpart = NULL;
- $0->condpart = $2.condpart; $2.condpart = NULL;
- $0->dopart = $2.dopart; $2.dopart = NULL;
- }$
- | WhilePart CondSuffix ${
- $0 = $<2;
- $0->condpart = $1.condpart; $1.condpart = NULL;
- $0->dopart = $1.dopart; $1.dopart = NULL;
- }$
- | SwitchPart CondSuffix ${
- $0 = $<2;
- $0->condpart = $<1;
- }$
- | IfPart IfSuffix ${
- $0 = $<2;
- $0->condpart = $1.condpart; $1.condpart = NULL;
- $0->thenpart = $1.thenpart; $1.thenpart = NULL;
- }$
-
- CondSuffix -> IfSuffix ${ $0 = $<1; }$
- | Newlines case Expression Block CondSuffix ${ {
- struct casepart *cp = calloc(1, sizeof(*cp));
- $0 = $<5;
- cp->value = $<3;
- cp->action = $<4;
- cp->next = $0->casepart;
- $0->casepart = cp;
- } }$
- | case Expression Block CondSuffix ${ {
- struct casepart *cp = calloc(1, sizeof(*cp));
- $0 = $<4;
- cp->value = $<2;
- cp->action = $<3;
- cp->next = $0->casepart;
- $0->casepart = cp;
- } }$
+ // A CondStatement must end with EOL, as does CondSuffix and
+ // IfSuffix.
+ // ForPart, ThenPart, SwitchPart, CasePart are non-empty and
+ // may or may not end with EOL
+ // WhilePart and IfPart include an appropriate Suffix
+
+ // ForPart, SwitchPart, and IfPart open scopes, o we have to close
+ // them. WhilePart opens and closes its own scope.
+ CondStatement -> ForPart OptNL ThenPart OptNL WhilePart CondSuffix ${
+ $0 = $<CS;
+ $0->forpart = $<FP;
+ $0->thenpart = $<TP;
+ $0->looppart = $<WP;
+ var_block_close(c, CloseSequential, $0);
+ }$
+ | ForPart OptNL WhilePart CondSuffix ${
+ $0 = $<CS;
+ $0->forpart = $<FP;
+ $0->looppart = $<WP;
+ var_block_close(c, CloseSequential, $0);
+ }$
+ | WhilePart CondSuffix ${
+ $0 = $<CS;
+ $0->looppart = $<WP;
+ }$
+ | SwitchPart OptNL CasePart CondSuffix ${
+ $0 = $<CS;
+ $0->condpart = $<SP;
+ $CP->next = $0->casepart;
+ $0->casepart = $<CP;
+ var_block_close(c, CloseSequential, $0);
+ }$
+ | SwitchPart : IN OptNL CasePart CondSuffix OUT Newlines ${
+ $0 = $<CS;
+ $0->condpart = $<SP;
+ $CP->next = $0->casepart;
+ $0->casepart = $<CP;
+ var_block_close(c, CloseSequential, $0);
+ }$
+ | IfPart IfSuffix ${
+ $0 = $<IS;
+ $0->condpart = $IP.condpart; $IP.condpart = NULL;
+ $0->thenpart = $IP.thenpart; $IP.thenpart = NULL;
+ // This is where we close an "if" statement
+ var_block_close(c, CloseSequential, $0);
+ }$
+
+ CondSuffix -> IfSuffix ${
+ $0 = $<1;
+ }$
+ | Newlines CasePart CondSuffix ${
+ $0 = $<CS;
+ $CP->next = $0->casepart;
+ $0->casepart = $<CP;
+ }$
+ | CasePart CondSuffix ${
+ $0 = $<CS;
+ $CP->next = $0->casepart;
+ $0->casepart = $<CP;
+ }$
IfSuffix -> Newlines ${ $0 = new(cond_statement); }$
- | Newlines else Block ${
- $0 = new(cond_statement);
- $0->elsepart = $<3;
- }$
- | else Block ${
- $0 = new(cond_statement);
- $0->elsepart = $<2;
- }$
- | Newlines else CondStatement ${
- $0 = new(cond_statement);
- $0->elsepart = $<3;
- }$
- | else CondStatement ${
- $0 = new(cond_statement);
- $0->elsepart = $<2;
- }$
+ | Newlines ElsePart ${ $0 = $<EP; }$
+ | ElsePart ${$0 = $<EP; }$
+
+ ElsePart -> else OpenBlock Newlines ${
+ $0 = new(cond_statement);
+ $0->elsepart = $<OB;
+ var_block_close(c, CloseElse, $0->elsepart);
+ }$
+ | else OpenScope CondStatement ${
+ $0 = new(cond_statement);
+ $0->elsepart = $<CS;
+ var_block_close(c, CloseElse, $0->elsepart);
+ }$
+ $*casepart
+ CasePart -> case Expression OpenScope ColonBlock ${
+ $0 = calloc(1,sizeof(struct casepart));
+ $0->value = $<Ex;
+ $0->action = $<Bl;
+ var_block_close(c, CloseParallel, $0->action);
+ }$
$*exec
- ForPart -> for SimpleStatements ${
- $0 = reorder_bilist($<2);
- }$
- | for Block ${
- $0 = $<2;
- }$
-
- ThenPart -> then SimpleStatements ${
- $0 = reorder_bilist($<2);
- }$
- | then Block ${
- $0 = $<2;
- }$
-
- ThenPartNL -> ThenPart OptNL ${
- $0 = $<1;
- }$
+ // These scopes are closed in CondStatement
+ ForPart -> for OpenBlock ${
+ $0 = $<Bl;
+ }$
+
+ ThenPart -> then OpenBlock ${
+ $0 = $<OB;
+ var_block_close(c, CloseSequential, $0);
+ }$
- WhileHead -> while Block ${
- $0 = $<2;
- }$
+ $*binode
+ // This scope is closed in CondStatement
+ WhilePart -> while UseBlock OptNL do OpenBlock ${
+ $0 = new(binode);
+ $0->op = Loop;
+ $0->left = $<UB;
+ $0->right = $<OB;
+ var_block_close(c, CloseSequential, $0->right);
+ var_block_close(c, CloseSequential, $0);
+ }$
+ | while OpenScope Expression OpenScope ColonBlock ${
+ $0 = new(binode);
+ $0->op = Loop;
+ $0->left = $<Exp;
+ $0->right = $<CB;
+ var_block_close(c, CloseSequential, $0->right);
+ var_block_close(c, CloseSequential, $0);
+ }$
$cond_statement
- ForThen -> ForPart OptNL ThenPartNL ${
- $0.forpart = $<1;
- $0.thenpart = $<3;
- }$
- | ForPart OptNL ${
- $0.forpart = $<1;
- }$
-
- WhilePart -> while Expression Block ${
- $0.type = Xcond_statement;
- $0.condpart = $<2;
- $0.dopart = $<3;
- }$
- | WhileHead OptNL do Block ${
- $0.type = Xcond_statement;
- $0.condpart = $<1;
- $0.dopart = $<4;
- }$
-
- IfPart -> if Expression Block ${
- $0.type = Xcond_statement;
- $0.condpart = $<2;
- $0.thenpart = $<3;
- }$
- | if Block OptNL then Block ${
- $0.type = Xcond_statement;
- $0.condpart = $<2;
- $0.thenpart = $<5;
- }$
+ IfPart -> if UseBlock OptNL then OpenBlock ${
+ $0.condpart = $<UB;
+ $0.thenpart = $<OB;
+ var_block_close(c, CloseParallel, $0.thenpart);
+ }$
+ | if OpenScope Expression OpenScope ColonBlock ${
+ $0.condpart = $<Ex;
+ $0.thenpart = $<CB;
+ var_block_close(c, CloseParallel, $0.thenpart);
+ }$
+ | if OpenScope Expression OpenScope OptNL then Block ${
+ $0.condpart = $<Ex;
+ $0.thenpart = $<Bl;
+ var_block_close(c, CloseParallel, $0.thenpart);
+ }$
$*exec
- SwitchPart -> switch Expression ${
- $0 = $<2;
- }$
- | switch Block ${
- $0 = $<2;
- }$
+ // This scope is closed in CondStatement
+ SwitchPart -> switch OpenScope Expression ${
+ $0 = $<Ex;
+ }$
+ | switch UseBlock ${
+ $0 = $<Bl;
+ }$
+
+###### print binode cases
+ case Loop:
+ if (b->left && b->left->type == Xbinode &&
+ cast(binode, b->left)->op == Block) {
+ if (bracket)
+ do_indent(indent, "while {\n");
+ else
+ do_indent(indent, "while\n");
+ print_exec(b->left, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "} do {\n");
+ else
+ do_indent(indent, "do\n");
+ print_exec(b->right, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
+ } else {
+ do_indent(indent, "while ");
+ print_exec(b->left, 0, bracket);
+ if (bracket)
+ printf(" {\n");
+ else
+ printf(":\n");
+ print_exec(b->right, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
+ }
+ break;
###### print exec cases
struct casepart *cp;
if (cs->forpart) {
do_indent(indent, "for");
- if (bracket) printf(" {\n"); else printf(":\n");
+ if (bracket) printf(" {\n"); else printf("\n");
print_exec(cs->forpart, indent+1, bracket);
if (cs->thenpart) {
if (bracket)
do_indent(indent, "} then {\n");
else
- do_indent(indent, "then:\n");
+ do_indent(indent, "then\n");
print_exec(cs->thenpart, indent+1, bracket);
}
if (bracket) do_indent(indent, "}\n");
}
- if (cs->dopart) {
- // a loop
- if (cs->condpart && cs->condpart->type == Xbinode &&
- cast(binode, cs->condpart)->op == Block) {
- if (bracket)
- do_indent(indent, "while {\n");
- else
- do_indent(indent, "while:\n");
- print_exec(cs->condpart, indent+1, bracket);
- if (bracket)
- do_indent(indent, "} do {\n");
- else
- do_indent(indent, "do:\n");
- print_exec(cs->dopart, indent+1, bracket);
- if (bracket)
- do_indent(indent, "}\n");
- } else {
- do_indent(indent, "while ");
- print_exec(cs->condpart, 0, bracket);
- if (bracket)
- printf(" {\n");
- else
- printf(":\n");
- print_exec(cs->dopart, indent+1, bracket);
- if (bracket)
- do_indent(indent, "}\n");
- }
+ if (cs->looppart) {
+ print_exec(cs->looppart, indent, bracket);
} else {
// a condition
if (cs->casepart)
do_indent(indent, "if");
if (cs->condpart && cs->condpart->type == Xbinode &&
cast(binode, cs->condpart)->op == Block) {
- printf(":\n");
+ if (bracket)
+ printf(" {\n");
+ else
+ printf("\n");
print_exec(cs->condpart, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
if (cs->thenpart) {
- do_indent(indent, "then:\n");
+ do_indent(indent, "then\n");
print_exec(cs->thenpart, indent+1, bracket);
}
} else {
printf(" ");
print_exec(cs->condpart, 0, bracket);
if (cs->thenpart) {
- printf(":\n");
+ if (bracket)
+ printf(" {\n");
+ else
+ printf(":\n");
print_exec(cs->thenpart, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
} else
printf("\n");
}
for (cp = cs->casepart; cp; cp = cp->next) {
do_indent(indent, "case ");
print_exec(cp->value, -1, 0);
- printf(":\n");
+ if (bracket)
+ printf(" {\n");
+ else
+ printf(":\n");
print_exec(cp->action, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
}
if (cs->elsepart) {
- do_indent(indent, "else:\n");
+ do_indent(indent, "else");
+ if (bracket)
+ printf(" {\n");
+ else
+ printf("\n");
print_exec(cs->elsepart, indent+1, bracket);
+ if (bracket)
+ do_indent(indent, "}\n");
}
break;
}
+###### propagate binode cases
+ case Loop:
+ t = propagate_types(b->right, c, perr, Tnone, 0);
+ if (!type_compat(Tnone, t, 0))
+ *perr |= Efail; // UNTESTED
+ return propagate_types(b->left, c, perr, type, rules);
+
###### propagate exec cases
case Xcond_statement:
{
- // forpart and dopart must return Vnone
- // condpart must be bool or match casepart->values
- // thenpart, elsepart, casepart->action must match
- // or be Vnone
+ // forpart and looppart->right must return Tnone
+ // thenpart must return Tnone if there is a loopart,
+ // otherwise it is like elsepart.
+ // condpart must:
+ // be bool if there is no casepart
+ // match casepart->values if there is a switchpart
+ // either be bool or match casepart->value if there
+ // is a whilepart
+ // elsepart and casepart->action must match the return type
+ // expected of this statement.
struct cond_statement *cs = cast(cond_statement, prog);
- struct casepart *c;
-
- t = propagate_types(cs->forpart, Vnone, ok);
- if (t != Vunknown && t != Vnone)
- *ok = 0;
- t = propagate_types(cs->dopart, Vnone, ok);
- if (t != Vunknown && t != Vnone)
- *ok = 0;
- if (cs->casepart == NULL)
- propagate_types(cs->condpart, Vbool, ok);
- else {
- t = Vunknown;
- for (c = cs->casepart;
- c && (t == Vunknown); c = c->next)
- t = propagate_types(c->value, Vunknown, ok);
- if (t == Vunknown && cs->condpart)
- t = propagate_types(cs->condpart, Vunknown, ok);
+ struct casepart *cp;
+
+ t = propagate_types(cs->forpart, c, perr, Tnone, 0);
+ if (!type_compat(Tnone, t, 0))
+ *perr |= Efail; // UNTESTED
+
+ if (cs->looppart) {
+ t = propagate_types(cs->thenpart, c, perr, Tnone, 0);
+ if (!type_compat(Tnone, t, 0))
+ *perr |= Efail; // UNTESTED
+ }
+ if (cs->casepart == NULL) {
+ propagate_types(cs->condpart, c, perr, Tbool, 0);
+ propagate_types(cs->looppart, c, perr, Tbool, 0);
+ } else {
+ /* Condpart must match case values, with bool permitted */
+ t = NULL;
+ for (cp = cs->casepart;
+ cp && !t; cp = cp->next)
+ t = propagate_types(cp->value, c, perr, NULL, 0);
+ if (!t && cs->condpart)
+ t = propagate_types(cs->condpart, c, perr, NULL, Rboolok); // UNTESTED
+ if (!t && cs->looppart)
+ t = propagate_types(cs->looppart, c, perr, NULL, Rboolok); // UNTESTED
// Now we have a type (I hope) push it down
- if (t != Vunknown) {
- for (c = cs->casepart; c; c = c->next)
- propagate_types(c->value, t, ok);
- propagate_types(cs->condpart, t, ok);
+ if (t) {
+ for (cp = cs->casepart; cp; cp = cp->next)
+ propagate_types(cp->value, c, perr, t, 0);
+ propagate_types(cs->condpart, c, perr, t, Rboolok);
+ propagate_types(cs->looppart, c, perr, t, Rboolok);
}
}
- if (type == Vunknown || type == Vnone)
- type = propagate_types(cs->thenpart, Vunknown, ok);
- if (type == Vunknown || type == Vnone)
- type = propagate_types(cs->elsepart, Vunknown, ok);
- for (c = cs->casepart;
- c && (type == Vunknown || type == Vnone);
- c = c->next)
- type = propagate_types(c->action, Vunknown, ok);
- if (type != Vunknown && type != Vnone) {
- propagate_types(cs->thenpart, type, ok);
- propagate_types(cs->elsepart, type, ok);
- for (c = cs->casepart; c ; c = c->next)
- propagate_types(c->action, type, ok);
+ // (if)then, else, and case parts must return expected type.
+ if (!cs->looppart && !type)
+ type = propagate_types(cs->thenpart, c, perr, NULL, rules);
+ if (!type)
+ type = propagate_types(cs->elsepart, c, perr, NULL, rules);
+ for (cp = cs->casepart;
+ cp && !type;
+ cp = cp->next) // UNTESTED
+ type = propagate_types(cp->action, c, perr, NULL, rules); // UNTESTED
+ if (type) {
+ if (!cs->looppart)
+ propagate_types(cs->thenpart, c, perr, type, rules);
+ propagate_types(cs->elsepart, c, perr, type, rules);
+ for (cp = cs->casepart; cp ; cp = cp->next)
+ propagate_types(cp->action, c, perr, type, rules);
return type;
} else
- return Vunknown;
+ return NULL;
}
+###### interp binode cases
+ case Loop:
+ // This just performs one iterration of the loop
+ rv = interp_exec(c, b->left, &rvtype);
+ if (rvtype == Tnone ||
+ (rvtype == Tbool && rv.bool != 0))
+ // rvtype is Tnone or Tbool, doesn't need to be freed
+ interp_exec(c, b->right, NULL);
+ break;
+
###### interp exec cases
case Xcond_statement:
{
struct value v, cnd;
+ struct type *vtype, *cndtype;
struct casepart *cp;
- struct cond_statement *c = cast(cond_statement, e);
- if (c->forpart)
- interp_exec(c->forpart);
- do {
- if (c->condpart)
- cnd = interp_exec(c->condpart);
- else
- cnd.vtype = Vnone;
- if (!(cnd.vtype == Vnone ||
- (cnd.vtype == Vbool && cnd.bool != 0)))
- break;
- if (c->dopart) {
- free_value(cnd);
- interp_exec(c->dopart);
- }
- if (c->thenpart) {
- v = interp_exec(c->thenpart);
- if (v.vtype != Vnone || !c->dopart)
- return v;
- free_value(v);
+ struct cond_statement *cs = cast(cond_statement, e);
+
+ if (cs->forpart)
+ interp_exec(c, cs->forpart, NULL);
+ if (cs->looppart) {
+ while ((cnd = interp_exec(c, cs->looppart, &cndtype)),
+ cndtype == Tnone || (cndtype == Tbool && cnd.bool != 0))
+ interp_exec(c, cs->thenpart, NULL);
+ } else {
+ cnd = interp_exec(c, cs->condpart, &cndtype);
+ if ((cndtype == Tnone ||
+ (cndtype == Tbool && cnd.bool != 0))) {
+ // cnd is Tnone or Tbool, doesn't need to be freed
+ rv = interp_exec(c, cs->thenpart, &rvtype);
+ // skip else (and cases)
+ goto Xcond_done;
}
- } while (c->dopart);
-
- for (cp = c->casepart; cp; cp = cp->next) {
- v = interp_exec(cp->value);
- if (value_cmp(v, cnd) == 0) {
- free_value(v);
- free_value(cnd);
- return interp_exec(cp->action);
+ }
+ for (cp = cs->casepart; cp; cp = cp->next) {
+ v = interp_exec(c, cp->value, &vtype);
+ if (value_cmp(cndtype, vtype, &v, &cnd) == 0) {
+ free_value(vtype, &v);
+ free_value(cndtype, &cnd);
+ rv = interp_exec(c, cp->action, &rvtype);
+ goto Xcond_done;
}
- free_value(v);
+ free_value(vtype, &v);
}
- free_value(cnd);
- if (c->elsepart)
- return interp_exec(c->elsepart);
- v.vtype = Vnone;
- return v;
+ free_value(cndtype, &cnd);
+ if (cs->elsepart)
+ rv = interp_exec(c, cs->elsepart, &rvtype);
+ else
+ rvtype = Tnone;
+ Xcond_done:
+ break;
}
-### Finally the whole program.
+### Top level structure
-Somewhat reminiscent of Pascal a (current) Ocean program starts with
-the keyword "program" and list of variable names which are assigned
-values from command line arguments. Following this is a `block` which
-is the code to execute.
+All the language elements so far can be used in various places. Now
+it is time to clarify what those places are.
-As this is the top level, several things are handled a bit
-differently.
-The whole program is not interpreted by `interp_exec` as that isn't
-passed the argument list which the program requires. Similarly type
-analysis is a bit more interesting at this level.
+At the top level of a file there will be a number of declarations.
+Many of the things that can be declared haven't been described yet,
+such as functions, procedures, imports, and probably more.
+For now there are two sorts of things that can appear at the top
+level. They are predefined constants, `struct` types, and the `main`
+function. While the syntax will allow the `main` function to appear
+multiple times, that will trigger an error if it is actually attempted.
-###### Binode types
- Program,
+The various declarations do not return anything. They store the
+various declarations in the parse context.
###### Parser: grammar
- $*binode
- Program -> program Varlist Block OptNL ${
- $0 = new(binode);
- $0->op = Program;
- $0->left = reorder_bilist($<2);
- $0->right = $<3;
+ $void
+ Ocean -> OptNL DeclarationList
+
+ ## declare terminals
+
+ OptNL ->
+ | OptNL NEWLINE
+
+ Newlines -> NEWLINE
+ | Newlines NEWLINE
+
+ DeclarationList -> Declaration
+ | DeclarationList Declaration
+
+ Declaration -> ERROR Newlines ${
+ tok_err(c, // UNTESTED
+ "error: unhandled parse error", &$1);
}$
+ | DeclareConstant
+ | DeclareFunction
+ | DeclareStruct
+
+ ## top level grammar
- Varlist -> Varlist Variable ${
- $0 = new(binode);
- $0->op = Program;
- $0->left = $<1;
- $0->right = $<2;
- }$
- | ${ $0 = NULL; }$
## Grammar
-###### print binode cases
- case Program:
- do_indent(indent, "program");
- for (b2 = cast(binode, b->left); b2; b2 = cast(binode, b2->right)) {
- printf(" ");
- print_exec(b2->left, 0, 0);
+### The `const` section
+
+As well as being defined in with the code that uses them, constants can
+be declared at the top level. These have full-file scope, so they are
+always `InScope`, even before(!) they have been declared. The value of
+a top level constant can be given as an expression, and this is
+evaluated after parsing and before execution.
+
+A function call can be used to evaluate a constant, but it will not have
+access to any program state, once such statement becomes meaningful.
+e.g. arguments and filesystem will not be visible.
+
+Constants are defined in a section that starts with the reserved word
+`const` and then has a block with a list of assignment statements.
+For syntactic consistency, these must use the double-colon syntax to
+make it clear that they are constants. Type can also be given: if
+not, the type will be determined during analysis, as with other
+constants.
+
+###### parse context
+ struct binode *constlist;
+
+###### top level grammar
+
+ $TERM const
+
+ DeclareConstant -> const { IN OptNL ConstList OUT OptNL } Newlines
+ | const { SimpleConstList } Newlines
+ | const IN OptNL ConstList OUT Newlines
+ | const SimpleConstList Newlines
+
+ ConstList -> ConstList SimpleConstLine
+ | SimpleConstLine
+
+ SimpleConstList -> SimpleConstList ; Const
+ | Const
+ | SimpleConstList ;
+
+ SimpleConstLine -> SimpleConstList Newlines
+ | ERROR Newlines ${ tok_err(c, "Syntax error in constant", &$1); }$
+
+ $*type
+ CType -> Type ${ $0 = $<1; }$
+ | ${ $0 = NULL; }$
+
+ $void
+ Const -> IDENTIFIER :: CType = Expression ${ {
+ struct variable *v;
+ struct binode *bl, *bv;
+ struct var *var = new_pos(var, $ID);
+
+ v = var_decl(c, $ID.txt);
+ if (v) {
+ v->where_decl = var;
+ v->where_set = var;
+ v->type = $<CT;
+ v->constant = 1;
+ v->global = 1;
+ } else {
+ v = var_ref(c, $1.txt);
+ if (v->type == Tnone) {
+ v->where_decl = var;
+ v->where_set = var;
+ v->type = $<CT;
+ v->constant = 1;
+ v->global = 1;
+ } else {
+ tok_err(c, "error: name already declared", &$1);
+ type_err(c, "info: this is where '%v' was first declared",
+ v->where_decl, NULL, 0, NULL);
+ }
}
- if (bracket)
- printf(" {\n");
- else
- printf(":\n");
- print_exec(b->right, indent+1, bracket);
- if (bracket)
- do_indent(indent, "}\n");
- break;
+ var->var = v;
-###### propagate binode cases
- case Program: abort();
+ bv = new(binode);
+ bv->op = Declare;
+ bv->left = var;
+ bv->right= $<Exp;
+
+ bl = new(binode);
+ bl->op = List;
+ bl->left = c->constlist;
+ bl->right = bv;
+ c->constlist = bl;
+ } }$
###### core functions
+ static void resolve_consts(struct parse_context *c)
+ {
+ struct binode *b;
+ int retry = 1;
+ enum { none, some, cannot } progress = none;
+
+ c->constlist = reorder_bilist(c->constlist);
+ while (retry) {
+ retry = 0;
+ for (b = cast(binode, c->constlist); b;
+ b = cast(binode, b->right)) {
+ enum prop_err perr;
+ struct binode *vb = cast(binode, b->left);
+ struct var *v = cast(var, vb->left);
+ if (v->var->frame_pos >= 0)
+ continue;
+ do {
+ perr = 0;
+ propagate_types(vb->right, c, &perr,
+ v->var->type, 0);
+ } while (perr & Eretry);
+ if (perr & Efail)
+ c->parse_error += 1;
+ else if (!(perr & Eruntime)) {
+ progress = some;
+ struct value res = interp_exec(
+ c, vb->right, &v->var->type);
+ global_alloc(c, v->var->type, v->var, &res);
+ } else {
+ if (progress == cannot)
+ type_err(c, "error: const %v cannot be resolved.",
+ v, NULL, 0, NULL);
+ else
+ retry = 1;
+ }
+ }
+ switch (progress) {
+ case cannot:
+ retry = 0; break;
+ case none:
+ progress = cannot; break;
+ case some:
+ progress = none; break;
+ }
+ }
+ }
- static int analyse_prog(struct exec *prog, struct parse_context *c)
+###### print const decls
{
- struct binode *b = cast(binode, prog);
- struct variable *v;
- int ok = 1;
- int uniq = 314159;
- do {
- ok = 1;
- propagate_types(b->right, Vnone, &ok);
- } while (ok == 2);
- if (!ok)
- return 0;
+ struct binode *b;
+ int first = 1;
+
+ for (b = cast(binode, context.constlist); b;
+ b = cast(binode, b->right)) {
+ struct binode *vb = cast(binode, b->left);
+ struct var *vr = cast(var, vb->left);
+ struct variable *v = vr->var;
+
+ if (first)
+ printf("const\n");
+ first = 0;
+
+ printf(" %.*s :: ", v->name->name.len, v->name->name.txt);
+ type_print(v->type, stdout);
+ printf(" = ");
+ print_exec(vb->right, -1, 0);
+ printf("\n");
+ }
+ }
+
+###### free const decls
+ free_binode(context.constlist);
+
+### Function declarations
+
+The code in an Ocean program is all stored in function declarations.
+One of the functions must be named `main` and it must accept an array of
+strings as a parameter - the command line arguments.
+
+As this is the top level, several things are handled a bit differently.
+The function is not interpreted by `interp_exec` as that isn't passed
+the argument list which the program requires. Similarly type analysis
+is a bit more interesting at this level.
- for (b = cast(binode, b->left); b; b = cast(binode, b->right)) {
+###### ast functions
+
+ static struct type *handle_results(struct parse_context *c,
+ struct binode *results)
+ {
+ /* Create a 'struct' type from the results list, which
+ * is a list for 'struct var'
+ */
+ struct type *t = add_anon_type(c, &structure_prototype,
+ "function result");
+ int cnt = 0;
+ struct binode *b;
+
+ for (b = results; b; b = cast(binode, b->right))
+ cnt += 1;
+ t->structure.nfields = cnt;
+ t->structure.fields = calloc(cnt, sizeof(struct field));
+ cnt = 0;
+ for (b = results; b; b = cast(binode, b->right)) {
struct var *v = cast(var, b->left);
- if (v->var->val.vtype == Vunknown)
- val_init(&v->var->val, Vstr);
- }
- b = cast(binode, prog);
- do {
- ok = 1;
- propagate_types(b->right, Vnone, &ok);
- } while (ok == 2);
- if (!ok)
- return 0;
+ struct field *f = &t->structure.fields[cnt++];
+ int a = v->var->type->align;
+ f->name = v->var->name->name;
+ f->type = v->var->type;
+ f->init = NULL;
+ f->offset = t->size;
+ v->var->frame_pos = f->offset;
+ t->size += ((f->type->size - 1) | (a-1)) + 1;
+ if (a > t->align)
+ t->align = a;
+ variable_unlink_exec(v->var);
+ }
+ free_binode(results);
+ return t;
+ }
+
+ static struct variable *declare_function(struct parse_context *c,
+ struct variable *name,
+ struct binode *args,
+ struct type *ret,
+ struct binode *results,
+ struct exec *code)
+ {
+ if (name) {
+ struct value fn = {.function = code};
+ struct type *t;
+ var_block_close(c, CloseFunction, code);
+ t = add_anon_type(c, &function_prototype,
+ "func %.*s", name->name->name.len,
+ name->name->name.txt);
+ name->type = t;
+ t->function.params = reorder_bilist(args);
+ if (!ret) {
+ ret = handle_results(c, reorder_bilist(results));
+ t->function.inline_result = 1;
+ t->function.local_size = ret->size;
+ }
+ t->function.return_type = ret;
+ global_alloc(c, t, name, &fn);
+ name->type->function.scope = c->out_scope;
+ } else {
+ free_binode(args);
+ free_type(ret);
+ free_exec(code);
+ var_block_close(c, CloseFunction, NULL);
+ }
+ c->out_scope = NULL;
+ return name;
+ }
+
+###### declare terminals
+ $TERM return
+
+###### top level grammar
+
+ $*variable
+ DeclareFunction -> func FuncName ( OpenScope ArgsLine ) Block Newlines ${
+ $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
+ }$
+ | func FuncName IN OpenScope Args OUT OptNL do Block Newlines ${
+ $0 = declare_function(c, $<FN, $<Ar, Tnone, NULL, $<Bl);
+ }$
+ | func FuncName NEWLINE OpenScope OptNL do Block Newlines ${
+ $0 = declare_function(c, $<FN, NULL, Tnone, NULL, $<Bl);
+ }$
+ | func FuncName ( OpenScope ArgsLine ) : Type Block Newlines ${
+ $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
+ }$
+ | func FuncName ( OpenScope ArgsLine ) : ( ArgsLine ) Block Newlines ${
+ $0 = declare_function(c, $<FN, $<AL, NULL, $<AL2, $<Bl);
+ }$
+ | func FuncName IN OpenScope Args OUT OptNL return Type Newlines do Block Newlines ${
+ $0 = declare_function(c, $<FN, $<Ar, $<Ty, NULL, $<Bl);
+ }$
+ | func FuncName NEWLINE OpenScope return Type Newlines do Block Newlines ${
+ $0 = declare_function(c, $<FN, NULL, $<Ty, NULL, $<Bl);
+ }$
+ | func FuncName IN OpenScope Args OUT OptNL return IN Args OUT OptNL do Block Newlines ${
+ $0 = declare_function(c, $<FN, $<Ar, NULL, $<Ar2, $<Bl);
+ }$
+ | func FuncName NEWLINE OpenScope return IN Args OUT OptNL do Block Newlines ${
+ $0 = declare_function(c, $<FN, NULL, NULL, $<Ar, $<Bl);
+ }$
+
+###### print func decls
+ {
+ struct variable *v;
+ int target = -1;
+
+ while (target != 0) {
+ int i = 0;
+ for (v = context.in_scope; v; v=v->in_scope)
+ if (v->depth == 0 && v->type && v->type->check_args) {
+ i += 1;
+ if (i == target)
+ break;
+ }
+
+ if (target == -1) {
+ target = i;
+ } else {
+ struct value *val = var_value(&context, v);
+ printf("func %.*s", v->name->name.len, v->name->name.txt);
+ v->type->print_type_decl(v->type, stdout);
+ if (brackets)
+ print_exec(val->function, 0, brackets);
+ else
+ print_value(v->type, val, stdout);
+ printf("/* frame size %d */\n", v->type->function.local_size);
+ target -= 1;
+ }
+ }
+ }
+
+###### core functions
+
+ static int analyse_funcs(struct parse_context *c)
+ {
+ struct variable *v;
+ int all_ok = 1;
+ for (v = c->in_scope; v; v = v->in_scope) {
+ struct value *val;
+ struct type *ret;
+ enum prop_err perr;
+ if (v->depth != 0 || !v->type || !v->type->check_args)
+ continue;
+ ret = v->type->function.inline_result ?
+ Tnone : v->type->function.return_type;
+ val = var_value(c, v);
+ do {
+ perr = 0;
+ propagate_types(val->function, c, &perr, ret, 0);
+ } while (!(perr & Efail) && (perr & Eretry));
+ if (!(perr & Efail))
+ /* Make sure everything is still consistent */
+ propagate_types(val->function, c, &perr, ret, 0);
+ if (perr & Efail)
+ all_ok = 0;
+ if (!v->type->function.inline_result &&
+ !v->type->function.return_type->dup) {
+ type_err(c, "error: function cannot return value of type %1",
+ v->where_decl, v->type->function.return_type, 0, NULL);
+ }
+
+ scope_finalize(c, v->type);
+ }
+ return all_ok;
+ }
- for (v = c->varlist; v; v = v->next)
- if (v->val.vtype == Vunknown) {
- v->val.vtype = Vnum;
- mpq_init(v->val.num);
- mpq_set_ui(v->val.num, uniq, 1);
- uniq++;
+ static int analyse_main(struct type *type, struct parse_context *c)
+ {
+ struct binode *bp = type->function.params;
+ struct binode *b;
+ enum prop_err perr;
+ int arg = 0;
+ struct type *argv_type;
+
+ argv_type = add_anon_type(c, &array_prototype, "argv");
+ argv_type->array.member = Tstr;
+ argv_type->array.unspec = 1;
+
+ for (b = bp; b; b = cast(binode, b->right)) {
+ perr = 0;
+ switch (arg++) {
+ case 0: /* argv */
+ propagate_types(b->left, c, &perr, argv_type, 0);
+ break;
+ default: /* invalid */ // NOTEST
+ propagate_types(b->left, c, &perr, Tnone, 0); // NOTEST
}
- /* Make sure everything is still consistent */
- propagate_types(b->right, Vnone, &ok);
- return !!ok;
+ if (perr & Efail)
+ c->parse_error += 1;
+ }
+
+ return !c->parse_error;
}
- static void interp_prog(struct exec *prog, char **argv)
+ static void interp_main(struct parse_context *c, int argc, char **argv)
{
- struct binode *p = cast(binode, prog);
- struct binode *al = cast(binode, p->left);
+ struct value *progp = NULL;
+ struct text main_name = { "main", 4 };
+ struct variable *mainv;
+ struct binode *al;
+ int anum = 0;
struct value v;
+ struct type *vtype;
+
+ mainv = var_ref(c, main_name);
+ if (mainv)
+ progp = var_value(c, mainv);
+ if (!progp || !progp->function) {
+ fprintf(stderr, "oceani: no main function found.\n");
+ c->parse_error += 1;
+ return;
+ }
+ if (!analyse_main(mainv->type, c)) {
+ fprintf(stderr, "oceani: main has wrong type.\n");
+ c->parse_error += 1;
+ return;
+ }
+ al = mainv->type->function.params;
+ c->local_size = mainv->type->function.local_size;
+ c->local = calloc(1, c->local_size);
while (al) {
struct var *v = cast(var, al->left);
- struct value *vl = &v->var->val;
-
- if (argv[0] == NULL) {
- printf("Not enough args\n");
- exit(1);
+ struct value *vl = var_value(c, v->var);
+ struct value arg;
+ struct type *t;
+ mpq_t argcq;
+ int i;
+
+ switch (anum++) {
+ case 0: /* argv */
+ t = v->var->type;
+ mpq_init(argcq);
+ mpq_set_ui(argcq, argc, 1);
+ memcpy(var_value(c, t->array.vsize), &argcq, sizeof(argcq));
+ t->prepare_type(c, t, 0);
+ array_init(v->var->type, vl);
+ for (i = 0; i < argc; i++) {
+ struct value *vl2 = vl->array + i * v->var->type->array.member->size;
+
+ arg.str.txt = argv[i];
+ arg.str.len = strlen(argv[i]);
+ free_value(Tstr, vl2);
+ dup_value(Tstr, &arg, vl2);
+ }
+ break;
}
al = cast(binode, al->right);
- free_value(*vl);
- if (!parse_value(vl, argv[0]))
- exit(1);
- argv++;
}
- v = interp_exec(p->right);
- free_value(v);
+ v = interp_exec(c, progp->function, &vtype);
+ free_value(vtype, &v);
+ free(c->local);
+ c->local = NULL;
}
-###### interp binode cases
- case Program: abort();
+###### ast functions
+ void free_variable(struct variable *v)
+ {
+ }
+
+## And now to test it out.
+
+Having a language requires having a "hello world" program. I'll
+provide a little more than that: a program that prints "Hello world"
+finds the GCD of two numbers, prints the first few elements of
+Fibonacci, performs a binary search for a number, and a few other
+things which will likely grow as the languages grows.
+
+###### File: oceani.mk
+ demos :: sayhello
+ sayhello : oceani
+ @echo "===== DEMO ====="
+ ./oceani --section "demo: hello" oceani.mdc 55 33
+
+###### demo: hello
+
+ const
+ pi ::= 3.141_592_6
+ four ::= 2 + 2 ; five ::= 10/2
+ const pie ::= "I like Pie";
+ cake ::= "The cake is"
+ ++ " a lie"
+
+ struct fred
+ size:[four]number
+ name:string
+ alive:Boolean
+
+ func main(argv:[argc::]string)
+ print "Hello World, what lovely oceans you have!"
+ print "Are there", five, "?"
+ print pi, pie, "but", cake
+
+ A := $argv[1]; B := $argv[2]
+
+ /* When a variable is defined in both branches of an 'if',
+ * and used afterwards, the variables are merged.
+ */
+ if A > B:
+ bigger := "yes"
+ else
+ bigger := "no"
+ print "Is", A, "bigger than", B,"? ", bigger
+ /* If a variable is not used after the 'if', no
+ * merge happens, so types can be different
+ */
+ if A > B * 2:
+ double:string = "yes"
+ print A, "is more than twice", B, "?", double
+ else
+ double := B*2
+ print "double", B, "is", double
+
+ a : number
+ a = A;
+ b:number = B
+ if a > 0 and then b > 0:
+ while a != b:
+ if a < b:
+ b = b - a
+ else
+ a = a - b
+ print "GCD of", A, "and", B,"is", a
+ else if a <= 0:
+ print a, "is not positive, cannot calculate GCD"
+ else
+ print b, "is not positive, cannot calculate GCD"
+
+ for
+ togo := 10
+ f1 := 1; f2 := 1
+ print "Fibonacci:", f1,f2,
+ then togo = togo - 1
+ while togo > 0:
+ f3 := f1 + f2
+ print "", f3,
+ f1 = f2
+ f2 = f3
+ print ""
+
+ /* Binary search... */
+ for
+ lo:= 0; hi := 100
+ target := 77
+ while
+ mid := (lo + hi) / 2
+ if mid == target:
+ use .Found
+ if mid < target:
+ lo = mid
+ else
+ hi = mid
+ if hi - lo < 1:
+ lo = mid
+ use .GiveUp
+ use True
+ do pass
+ case .Found:
+ print "Yay, I found", target
+ case .GiveUp:
+ print "Closest I found was", lo
+
+ size::= 10
+ list:[size]number
+ list[0] = 1234
+ // "middle square" PRNG. Not particularly good, but one my
+ // Dad taught me - the first one I ever heard of.
+ for i:=1; then i = i + 1; while i < size:
+ n := list[i-1] * list[i-1]
+ list[i] = (n / 100) % 10 000
+
+ print "Before sort:",
+ for i:=0; then i = i + 1; while i < size:
+ print "", list[i],
+ print
+
+ for i := 1; then i=i+1; while i < size:
+ for j:=i-1; then j=j-1; while j >= 0:
+ if list[j] > list[j+1]:
+ t:= list[j]
+ list[j] = list[j+1]
+ list[j+1] = t
+ print " After sort:",
+ for i:=0; then i = i + 1; while i < size:
+ print "", list[i],
+ print
+
+ if 1 == 2 then print "yes"; else print "no"
+
+ bob:fred
+ bob.name = "Hello"
+ bob.alive = (bob.name == "Hello")
+ print "bob", "is" if bob.alive else "isn't", "alive"
code / ocean / blobdiff