`parsergen` program built from the C code in this file can extract
that grammar directly from this file and process it.
-
###### File: parsergen.c
#include <unistd.h>
#include <stdlib.h>
struct production {
unsigned short precedence;
enum assoc assoc;
+ char line_like;
## production fields
};
struct grammar {
listed and may be inherited by any production which uses the symbol. A
production inherits from the last symbol which has a precedence.
+The symbols on the first precedence line have the lowest precedence.
+Subsequent lines introduce symbols with higher precedence.
+
###### grammar fields
struct text current_type;
int type_isref;
goto abort;
}
vs = sym_find(g, tk.txt);
- if (vs->type != Virtual) {
- err = "symbol after $$ must be virtual";
+ if (vs->num == TK_newline)
+ p.line_like = 1;
+ else if (vs->precedence == 0) {
+ err = "symbol after $$ must have precedence";
goto abort;
+ } else {
+ p.precedence = vs->precedence;
+ p.assoc = vs->assoc;
}
- p.precedence = vs->precedence;
- p.assoc = vs->assoc;
tk = token_next(state);
}
if (tk.num == TK_open) {
return sl->ss;
}
-
### Setting `nullable`
We set `nullable` on the head symbol for any production for which all
}
}
-### Setting `can_eol` and `line_like`
+### Setting `line_like`
In order to be able to ignore newline tokens when not relevant, but
still include them in the parse when needed, we will need to know
newlines when there is an indent since the most recent start of a
line-like symbol.
-To know which symbols are line-like, we first need to know which
-symbols start with a NEWLINE token. Any symbol which is followed by a
-NEWLINE, or anything that starts with a NEWLINE, is deemed to be a line-like symbol.
-Certainly when trying to parse one of these we must take note of NEWLINEs.
-
-Clearly the `TK_newline` token can start with a NEWLINE. Any symbol
-which is the head of a production that contains a starts-with-NEWLINE
-symbol preceeded only by nullable symbols is also a
-starts-with-NEWLINE symbol. We use a new field `can_eol` to record
-this attribute of symbols, and compute it in a repetitive manner
-similar to `set_nullable`.
+A "line_like" symbol is simply any symbol that can derive a NEWLINE.
+If a symbol cannot derive a NEWLINE, then it is only part of a line -
+so is word-like. If it can derive a NEWLINE, then we consider it to
+be like a line.
-Once we have that, we can determine which symbols are `line_like` by
-seeing which are followed by a `can_eol` symbol in any production.
+Clearly the `TK_newline` token can derive a NEWLINE. Any symbol which
+is the head of a production that contains a line_like symbol is also a
+line-like symbol. We use a new field `line_like` to record this
+attribute of symbols, and compute it in a repetitive manner similar to
+`set_nullable`.
###### symbol fields
- int can_eol;
int line_like;
###### functions
- static void set_can_eol(struct grammar *g)
+ static void set_line_like(struct grammar *g)
{
int check_again = 1;
- g->symtab[TK_newline]->can_eol = 1;
+ g->symtab[TK_newline]->line_like = 1;
while (check_again) {
int p;
check_again = 0;
struct production *pr = g->productions[p];
int s;
- if (pr->head->can_eol)
+ if (pr->head->line_like)
continue;
for (s = 0 ; s < pr->body_size; s++) {
- if (pr->body[s]->can_eol) {
- pr->head->can_eol = 1;
+ if (pr->body[s]->line_like) {
+ pr->head->line_like = 1;
check_again = 1;
break;
}
- if (!pr->body[s]->nullable)
- break;
}
}
}
}
- static void set_line_like(struct grammar *g)
- {
- int p;
- for (p = 0; p < g->production_count; p++) {
- struct production *pr = g->productions[p];
- int s;
-
- for (s = 1; s < pr->body_size; s++)
- if (pr->body[s]->can_eol)
- pr->body[s-1]->line_like = 1;
- }
- }
-
### Building the `first` sets
When calculating what can follow a particular non-terminal, we will need to
a.data[i] - b.data[i];
}
+It will be helpful to know if an itemset has been "completed" or not,
+particularly for LALR where itemsets get merged, at which point they
+need to be consider for completion again. So a `completed` flag is needed.
+
+For correct handling of `TK_newline` when parsing, we will need to
+know which states (itemsets) can occur at the start of a line, so we
+will record a `starts_line` flag too whenever DOT is at the start of a
+`line_like` symbol.
+
+Finally, for handling `TK_out` we need to know whether productions in the
+current state started *before* the most recent indent. A state
+doesn't usually keep details of individual productions, so we need to
+add one extra detail. `min_prefix` is the smallest non-zero number of
+symbols *before* DOT in any production in an itemset. This will allow
+us to determine if the the most recent indent is sufficiently recent
+to cancel it against a `TK_out`. If it was seen longer ago than the
+`min_prefix`, and if the current state cannot be reduced, then the
+indented section must have ended in the middle of a syntactic unit, so
+an error must be signaled.
+
And now we can build the list of itemsets. The lookup routine returns
both a success flag and a pointer to where in the list an insert
should happen, so we don't need to search a second time.
-FIXME: document min_prefix
-
###### declarations
struct itemset {
struct itemset *next;
We also collect a set of all symbols which follow "DOT" (in `done`) as this
is used in the next stage.
-If any of these symbols are flagged as starting a line, then this
+If any of these symbols are flagged as `line_like`, then this
state must be a `starts_line` state so now is a good time to record that.
When itemsets are created we assign a precedence to the itemset from
the complete item, if there is one. We ignore the possibility of
there being two and don't (currently) handle precedence in such
grammars. When completing a grammar we ignore any item where DOT is
-followed by a terminal with a precedence lower (numerically higher)
-than that for the itemset. Unless the terminal has right
-associativity, we also ignore items where the terminal has the same
-precedence. The result is that unwanted items are still in the
-itemset, but the terminal doesn't get into the go to set, so the item
-is ineffective.
+followed by a terminal with a precedence lower than that for the
+itemset. Unless the terminal has right associativity, we also ignore
+items where the terminal has the same precedence. The result is that
+unwanted items are still in the itemset, but the terminal doesn't get
+into the go to set, so the item is ineffective.
###### complete itemset
for (i = 0; i < is->items.cnt; i++) {
struct symbol *s;
struct symset LA = INIT_SYMSET;
unsigned short sn = 0;
+ struct symset LAnl = INIT_SYMSET;
+ unsigned short snnl = 0;
if (is->min_prefix == 0 ||
(bs > 0 && bs < is->min_prefix))
continue;
s = pr->body[bs];
if (s->precedence && is->precedence &&
- is->precedence < s->precedence)
+ is->precedence > s->precedence)
/* This terminal has a low precedence and
* shouldn't be shifted
*/
}
sn = save_set(g, LA);
LA = set_find(g, sn);
+ if (symset_find(&LA, TK_newline))
+ symset_add(&LAnl, TK_newline, 0);
+ snnl = save_set(g, LAnl);
+ LAnl = set_find(g, snnl);
}
/* Add productions for this symbol */
int itm = item_num(p2, 0);
int pos = symset_find(&is->items, itm);
if (pos < 0) {
- symset_add(&is->items, itm, sn);
+ if (g->productions[p2]->line_like)
+ symset_add(&is->items, itm, snnl);
+ else
+ symset_add(&is->items, itm, sn);
/* Will have re-ordered, so start
* from beginning again */
i = -1;
} else if (type >= LALR) {
struct symset ss = set_find(g, is->items.data[pos]);
struct symset tmp = INIT_SYMSET;
+ struct symset *la = &LA;
+ if (g->productions[p2]->line_like)
+ la = &LAnl;
symset_union(&tmp, &ss);
- if (symset_union(&tmp, &LA)) {
+ if (symset_union(&tmp, la)) {
is->items.data[pos] = save_set(g, tmp);
i = -1;
- }else
+ } else
symset_free(tmp);
}
}
pos = symset_find(&newitemset, pr->head->num);
if (bp + 1 == pr->body_size &&
pr->precedence > 0 &&
- (precedence == 0 ||
- pr->precedence < precedence)) {
+ pr->precedence > precedence) {
// new itemset is reducible and has a precedence.
precedence = pr->precedence;
assoc = pr->assoc;
g->symtab[s->num] = s;
set_nullable(g);
- set_can_eol(g);
set_line_like(g);
if (type >= SLR)
build_first(g);
if (!s)
continue;
- printf(" %c%c%c%3d%c: ",
+ printf(" %c%c%3d%c: ",
s->nullable ? '.':' ',
- s->can_eol ? '>':' ',
s->line_like ? '<':' ',
s->num, symtypes[s->type]);
prtxt(s->name);
printf(" [%d%s]", s->precedence,
assoc_names[s->assoc]);
}
+ if (pr->line_like)
+ printf(" $$NEWLINE");
printf("\n");
}
Then the go to sets:
-
static void report_goto(struct grammar *g, struct symset gt)
{
int i;
which maps terminals to items that could be reduced when the terminal
is in look-ahead. We report when we get conflicts between the two.
+As a special case, if we find a SHIFT/REDUCE conflict, on the NEWLINE
+terminal, we ignore it. NEWLINES are handled specially with its own
+rules for when to shift and when to reduce. Conflicts are expected,
+but handled internally.
+
static int conflicts_slr(struct grammar *g, enum grammar_type type)
{
int i;
int p = item_prod(itm);
int bp = item_index(itm);
struct production *pr = g->productions[p];
+ struct symbol *s;
- if (bp < pr->body_size &&
- pr->body[bp]->type == Terminal) {
- /* shiftable */
- int sym = pr->body[bp]->num;
- if (symset_find(&shifts, sym) < 0)
- symset_add(&shifts, sym, itm);
- }
+ if (bp >= pr->body_size ||
+ pr->body[bp]->type != Terminal)
+ /* not shiftable */
+ continue;
+
+ s = pr->body[bp];
+ if (s->precedence && is->precedence)
+ /* Precedence resolves this, so no conflict */
+ continue;
+
+ if (symset_find(&shifts, s->num) < 0)
+ symset_add(&shifts, s->num, itm);
}
- /* Now look for reduction and conflicts */
+ /* Now look for reductions and conflicts */
for (j = 0; j < is->items.cnt; j++) {
unsigned short itm = is->items.syms[j];
int p = item_prod(itm);
int k;
for (k = 0; k < la.cnt; k++) {
int pos = symset_find(&shifts, la.syms[k]);
- if (pos >= 0) {
+ if (pos >= 0 && la.syms[k] != TK_newline) {
printf(" State %d has SHIFT/REDUCE conflict on ", i);
- prtxt(g->symtab[la.syms[k]]->name);
+ cnt++;
+ prtxt(g->symtab[la.syms[k]]->name);
printf(":\n");
report_item(g, shifts.data[pos]);
report_item(g, itm);
- cnt++;
}
pos = symset_find(&reduce, la.syms[k]);
if (pos < 0) {
return cnt;
}
-
## Generating the parser
The exported part of the parser is the `parse_XX` function, where the name
###### parser_generate
- static void gen_parser(FILE *f, struct grammar *g, char *file, char *name)
+ static void gen_parser(FILE *f, struct grammar *g, char *file, char *name,
+ struct code_node *pre_reduce)
{
gen_known(f, g);
gen_non_term(f, g);
gen_goto(f, g);
gen_states(f, g);
- gen_reduce(f, g, file);
+ gen_reduce(f, g, file, pre_reduce);
gen_free(f, g);
fprintf(f, "#line 0 \"gen_parser\"\n");
### Known words table
The known words table is simply an array of terminal symbols.
-The table of nonterminals used for tracing is a similar array.
+The table of nonterminals used for tracing is a similar array. We
+include virtual symbols in the table of non_terminals to keep the
+numbers right.
###### functions
for (i = TK_reserved;
i < g->num_syms;
i++)
- if (g->symtab[i]->type == Nonterminal)
+ if (g->symtab[i]->type != Terminal)
fprintf(f, "\t\"%.*s\",\n", g->symtab[i]->name.len,
g->symtab[i]->name.txt);
fprintf(f, "};\n\n");
short reduce_prod;
short reduce_size;
short reduce_sym;
- short starts_line;
+ char starts_line;
+ char newline_only;
short min_prefix;
};
-
###### functions
static void gen_goto(FILE *f, struct grammar *g)
}
if (prod >= 0)
- fprintf(f, "\t[%d] = { %d, goto_%d, %d, %d, %d, %d, %d },\n",
+ fprintf(f, "\t[%d] = { %d, goto_%d, %d, %d, %d, %d, %d, %d },\n",
i, is->go_to.cnt, i, prod,
g->productions[prod]->body_size,
g->productions[prod]->head->num,
- is->starts_line, is->min_prefix);
+ is->starts_line,
+ g->productions[prod]->line_like,
+ is->min_prefix);
else
- fprintf(f, "\t[%d] = { %d, goto_%d, -1, -1, -1, %d, %d },\n",
+ fprintf(f, "\t[%d] = { %d, goto_%d, -1, -1, -1, %d, 0, %d },\n",
i, is->go_to.cnt, i,
is->starts_line, is->min_prefix);
}
fputs("\n", f);
for (i = 0; i < p->body_size; i++) {
if (p->body[i]->struct_name.txt &&
- p->body[i]->isref &&
- used[i])
+ used[i]) {
// assume this has been copied out
- fprintf(f, "\t\t*(void**)body[%d] = NULL;\n", i);
+ if (p->body[i]->isref)
+ fprintf(f, "\t\t*(void**)body[%d] = NULL;\n", i);
+ else
+ fprintf(f, "\t\tmemset(body[%d], 0, sizeof(struct %.*s));\n", i, p->body[i]->struct_name.len, p->body[i]->struct_name.txt);
+ }
}
free(used);
}
###### functions
- static void gen_reduce(FILE *f, struct grammar *g, char *file)
+ static void gen_reduce(FILE *f, struct grammar *g, char *file,
+ struct code_node *code)
{
int i;
- fprintf(f, "#line 0 \"gen_reduce\"\n");
+ fprintf(f, "#line 1 \"gen_reduce\"\n");
fprintf(f, "static int do_reduce(int prod, void **body, struct token_config *config, void *ret)\n");
fprintf(f, "{\n");
fprintf(f, "\tint ret_size = 0;\n");
+ if (code)
+ code_node_print(f, code, file);
+ fprintf(f, "#line 4 \"gen_reduce\"\n");
fprintf(f, "\tswitch(prod) {\n");
for (i = 0; i < g->production_count; i++) {
struct production *p = g->productions[i];
struct code_node *hdr = NULL;
struct code_node *code = NULL;
struct code_node *gram = NULL;
+ struct code_node *pre_reduce = NULL;
for (s = table; s; s = s->next) {
struct text sec = s->section;
if (tag && !strip_tag(&sec, tag))
code = s->code;
else if (text_is(sec, "grammar"))
gram = s->code;
+ else if (text_is(sec, "reduce"))
+ pre_reduce = s->code;
else {
fprintf(stderr, "Unknown content section: %.*s\n",
s->section.len, s->section.txt);
if (f) {
if (code)
code_node_print(f, code, infile);
- gen_parser(f, g, infile, name);
+ gen_parser(f, g, infile, name, pre_reduce);
fclose(f);
} else {
fprintf(stderr, "Cannot create %s.c\n",
the most recent indent, then we do that first. If the minimum prefix
of the current state then extends back before the most recent indent,
that indent can be cancelled. If the minimum prefix is shorter then
-the indent is premature and we must start error handling, which
-currently doesn't work at all.
+the indent had ended prematurely and we must start error handling, which
+is still a work-in-progress.
`TK_newline` tokens are ignored unless the top stack frame records
that they are permitted. In that case they will not be considered for
shifting if it is possible to reduce some symbols that are all since
-the most recent start of line. This is how a newline forcible
+the most recent start of line. This is how a newline forcibly
terminates any line-like structure - we try to reduce down to at most
one symbol for each line where newlines are allowed.
+A consequence of this is that a rule like
+
+###### Example: newlines - broken
+
+ Newlines ->
+ | NEWLINE Newlines
+ IfStatement -> Newlines if ....
+
+cannot work, as the NEWLINE will never be shifted as the empty string
+will be reduced first. Optional sets of newlines need to be include
+in the thing that preceed:
+
+###### Example: newlines - works
+
+ If -> if
+ | NEWLINE If
+ IfStatement -> If ....
+
+Here the NEWLINE will be shifted because nothing can be reduced until
+the `if` is seen.
+
+When, during error handling, we discard token read in, we want to keep
+discarding until we see one that is recognised. If we had a full set
+of LR(1) grammar states, this will mean looking in the look-ahead set,
+but we don't keep a full look-ahead set. We only record the subset
+that leads to SHIFT. We can, however, deduce the look-ahead set but
+looking at the SHIFT subsets for all states that we can get to by
+reducing zero or more times. So we need a little function which
+checks if a given token is in any of these look-ahead sets.
###### parser includes
#include "parser.h"
+
###### parser_run
+
+ static int in_lookahead(struct token *tk, const struct state *states, int state)
+ {
+ while (state >= 0) {
+ if (search(&states[state], tk->num) >= 0)
+ return 1;
+ if (states[state].reduce_prod < 0)
+ return 0;
+ state = search(&states[state], states[state].reduce_sym);
+ }
+ return 0;
+ }
+
void *parser_run(struct token_state *tokens,
const struct state states[],
int (*do_reduce)(int, void**, struct token_config*, void*),
parser_trace_action(trace, "Cancel");
continue;
}
- // fall through and force a REDUCE (as 'shift'
- // will fail).
+ // fall through to error handling as both SHIFT and REDUCE
+ // will fail.
}
if (tk->num == TK_newline) {
if (!tos->newline_permitted) {
continue;
}
force_reduce:
- if (states[tos->state].reduce_prod >= 0) {
+ if (states[tos->state].reduce_prod >= 0 &&
+ states[tos->state].newline_only &&
+ tk->num != TK_newline && tk->num != TK_eof && tk->num != TK_out) {
+ /* Anything other than newline in an error as this
+ * production must end at EOL
+ */
+ } else if (states[tos->state].reduce_prod >= 0) {
void **body;
void *res;
const struct state *nextstate = &states[tos->state];
short indents = 0, start_of_line;
err_tk = tok_copy(*tk);
- while (shift(&p, TK_error, 0, 0,
- err_tk, states) == 0
- && p.tos > 0)
+ while (p.tos > 0 &&
+ shift(&p, TK_error, 0, 0,
+ err_tk, states) == 0)
// discard this state
indents += pop(&p, 1, &start_of_line, do_free);
if (p.tos == 0) {
break;
}
tos = &p.stack[p.tos-1];
- while (search(&states[tos->state], tk->num) < 0 &&
+ while (!in_lookahead(tk, states, tos->state) &&
tk->num != TK_eof) {
free(tk);
tk = tok_copy(token_next(tokens));
// FIXME update since_indent here
}
}
- if (p.tos == 0 && tk->num == TK_eof)
- break;
- tos = &p.stack[p.tos-1];
tos->indents += indents;
- exit(1);
}
free(tk);
pop(&p, p.tos, NULL, do_free);
./parsergen --tag calc -o calc parsergen.mdc
calc : calc.o libparser.o libscanner.o libmdcode.o libnumber.o
$(CC) $(CFLAGS) -o calc calc.o libparser.o libscanner.o libmdcode.o libnumber.o -licuuc -lgmp
+ calctest : calc
+ ./calc parsergen.mdc
+ demos :: calctest
# calc: header
#include <stdio.h>
#include <malloc.h>
#include <gmp.h>
+ #include <string.h>
#include "mdcode.h"
#include "scanner.h"
#include "number.h"
free(n);
}
+ static int text_is(struct text t, char *s)
+ {
+ return (strlen(s) == t.len &&
+ strncmp(s, t.txt, t.len) == 0);
+ }
+
int main(int argc, char *argv[])
{
int fd = open(argv[1], O_RDONLY);
int len = lseek(fd, 0, 2);
char *file = mmap(NULL, len, PROT_READ, MAP_SHARED, fd, 0);
- struct section *s = code_extract(file, file+len, NULL);
+ struct section *table = code_extract(file, file+len, NULL);
+ struct section *s;
struct token_config config = {
.ignored = (1 << TK_line_comment)
| (1 << TK_block_comment)
.word_start = "",
.word_cont = "",
};
- parse_calc(s->code, &config, argc > 2 ? stderr : NULL);
- while (s) {
- struct section *t = s->next;
- code_free(s->code);
- free(s);
- s = t;
+ for (s = table; s; s = s->next)
+ if (text_is(s->section, "example: input"))
+ parse_calc(s->code, &config, argc > 2 ? stderr : NULL);
+ while (table) {
+ struct section *t = table->next;
+ code_free(table->code);
+ free(table);
+ table = t;
}
exit(0);
}
# calc: grammar
- $LEFT * /
$LEFT + -
+ $LEFT * /
Session -> Session Line
| Line
| Expression / Expression ${ mpq_init($0.val); mpq_div($0.val, $1.val, $3.val); }$
| NUMBER ${ if (number_parse($0.val, $0.tail, $1.txt) == 0) mpq_init($0.val); }$
| ( Expression ) ${ mpq_init($0.val); mpq_set($0.val, $2.val); }$
+
+# example: input
+
+ 355/113
+ 3.1415926535 - 355/113
+ 2 + 4 * 5
+ 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9
+ 10 * 9 / 2
+ 1 * 1000 + 2 * 100 + 3 * 10 + 4 * 1
+
+ error
code / ocean / blobdiff