s->precedence = g->prec_levels;
s->assoc = assoc;
found += 1;
+ t = token_next(ts);
}
if (found == 0)
err = "No symbols given on precedence line";
struct symbol **body;
int body_size;
struct text code;
+ int code_line;
###### symbol fields
int first_production;
tk = token_next(state);
}
if (tk.num == TK_open) {
+ p.code_line = tk.line;
p.code = collect_code(state, tk);
if (p.code.txt == NULL) {
err = "code fragment not closed properly";
short state;
struct symset items;
struct symset go_to;
+ enum assoc assoc;
+ unsigned short precedence;
char completed;
char starts_line;
int min_prefix;
them to a data structure, of freeing them.
static int add_itemset(struct grammar *g, struct symset ss,
+ enum assoc assoc, unsigned short precedence,
enum grammar_type type)
{
struct itemset **where, *is;
is->state = g->states;
g->states += 1;
is->items = ss;
+ is->assoc = assoc;
+ is->precedence = precedence;
is->next = *where;
is->go_to = INIT_DATASET;
*where = is;
If any of these symbols are flagged as starting a line, then this
state must be a `starts_line` state so now is a good time to record that.
-NOTE: precedence handling should happen here - I haven't written this yet
-though.
+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.
###### complete itemset
for (i = 0; i < is->items.cnt; i++) {
if (bs == pr->body_size)
continue;
s = pr->body[bs];
+ if (s->precedence && is->precedence &&
+ is->precedence < s->precedence)
+ /* This terminal has a low precedence and
+ * shouldn't be shifted
+ */
+ continue;
+ if (s->precedence && is->precedence &&
+ is->precedence == s->precedence && s->assoc != Right)
+ /* This terminal has a matching precedence and is
+ * not Right-associative, so we mustn't shift it.
+ */
+ continue;
if (symset_find(&done, s->num) < 0) {
symset_add(&done, s->num, 0);
if (s->line_like)
int j;
unsigned short state;
struct symbol *sym = g->symtab[done.syms[i]];
+ enum assoc assoc = Non;
+ unsigned short precedence = 0;
struct symset newitemset = INIT_SYMSET;
if (type >= LALR)
newitemset = INIT_DATASET;
if (type >= LALR)
la = is->items.data[j];
pos = symset_find(&newitemset, pr->head->num);
+ if (bp + 1 == pr->body_size &&
+ pr->precedence > 0 &&
+ (precedence == 0 ||
+ pr->precedence < precedence)) {
+ // new itemset is reducible and has a precedence.
+ precedence = pr->precedence;
+ assoc = pr->assoc;
+ }
if (pos < 0)
symset_add(&newitemset, item_num(p, bp+1), la);
else if (type >= LALR) {
}
}
}
- state = add_itemset(g, newitemset, type);
+ state = add_itemset(g, newitemset, assoc, precedence, type);
if (symset_find(&is->go_to, done.syms[i]) < 0)
symset_add(&is->go_to, done.syms[i], state);
}
-All that is left is to crate the initial itemset from production zero, and
+All that is left is to create the initial itemset from production zero, and
with `TK_eof` as the LA set.
###### functions
}
// production 0, offset 0 (with no data)
symset_add(&first, item_num(0, 0), la);
- add_itemset(g, first, type);
+ add_itemset(g, first, Non, 0, type);
for (again = 0, is = g->items;
is;
is = is->next ?: again ? (again = 0, g->items) : NULL) {
if (dot == pr->body_size)
printf(" .");
printf(" [%d]", p);
- if (pr->precedence)
+ if (pr->precedence && dot == pr->body_size)
printf(" (%d%s)", pr->precedence,
assoc_names[pr->assoc]);
+ if (dot < pr->body_size &&
+ pr->body[dot]->precedence) {
+ struct symbol *s = pr->body[dot];
+ printf(" [%d%s]", s->precedence,
+ assoc_names[s->assoc]);
+ }
printf("\n");
}
for (s = 0; s < g->states; s++) {
int j;
struct itemset *is = g->statetab[s];
- printf(" Itemset %d:%s min prefix=%d\n",
+ printf(" Itemset %d:%s min prefix=%d",
s, is->starts_line?" (startsline)":"", is->min_prefix);
+ if (is->precedence)
+ printf(" %d%s", is->precedence, assoc_names[is->assoc]);
+ printf("\n");
for (j = 0; j < is->items.cnt; j++) {
report_item(g, is->items.syms[j]);
if (is->items.data != NO_DATA)
short reduce_prod;
short reduce_size;
short reduce_sym;
- short shift_sym;
short starts_line;
short min_prefix;
};
for (i = 0; i < g->states; i++) {
struct itemset *is = g->statetab[i];
int j, prod = -1, prod_len;
- int shift_sym = -1;
- int shift_len = 0, shift_remain = 0;
+
for (j = 0; j < is->items.cnt; j++) {
int itm = is->items.syms[j];
int p = item_prod(itm);
int bp = item_index(itm);
struct production *pr = g->productions[p];
- if (bp < pr->body_size) {
- if (shift_sym < 0 ||
- (shift_len == bp && shift_remain > pr->body_size - bp)) {
- shift_sym = pr->body[bp]->num;
- shift_len = bp;
- shift_remain = pr->body_size - bp;
- }
+ if (bp < pr->body_size)
continue;
- }
/* This is what we reduce */
if (prod < 0 || prod_len < pr->body_size) {
prod = p;
}
if (prod >= 0)
- fprintf(f, "\t[%d] = { %d, goto_%d, %d, %d, %d, 0, %d, %d },\n",
+ fprintf(f, "\t[%d] = { %d, goto_%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);
else
- fprintf(f, "\t[%d] = { %d, goto_%d, -1, -1, -1, %d, %d, %d },\n",
- i, is->go_to.cnt, i, shift_sym,
+ fprintf(f, "\t[%d] = { %d, goto_%d, -1, -1, -1, %d, %d },\n",
+ i, is->go_to.cnt, i,
is->starts_line, is->min_prefix);
}
fprintf(f, "};\n\n");
struct production *p = g->productions[i];
fprintf(f, "\tcase %d:\n", i);
- if (p->code.txt)
+ if (p->code.txt) {
+ fprintf(f, "#line %d \"%s\"\n", p->code_line, file);
gen_code(p, f, g);
+ }
if (p->head->struct_name.txt)
fprintf(f, "\t\tret_size = sizeof(struct %.*s%s);\n",
parser_trace_action(trace, "Reduce");
continue;
}
- if (tk->num == TK_out) {
- // Indent problem - synthesise tokens to get us
- // out of here.
- fprintf(stderr, "Synthesize %d to handle indent problem\n", states[tos->state].shift_sym);
- shift(&p, states[tos->state].shift_sym,
- 0, 1, tok_copy(*tk), states);
- // FIXME need to report this error somehow
- parser_trace_action(trace, "Synthesize");
- continue;
- }
/* Error. We walk up the stack until we
* find a state which will accept TK_error.
* We then shift in TK_error and see what state
# calc: grammar
+ $LEFT * /
+ $LEFT + -
+
Session -> Session Line
| Line
| ERROR NEWLINE ${ printf("Skipped a bad line\n"); }$
$number
- Expression -> Expression + Term ${ mpq_init($0.val); mpq_add($0.val, $1.val, $3.val); }$
- | Expression - Term ${ mpq_init($0.val); mpq_sub($0.val, $1.val, $3.val); }$
- | Term ${ mpq_init($0.val); mpq_set($0.val, $1.val); }$
-
- Term -> Term * Factor ${ mpq_init($0.val); mpq_mul($0.val, $1.val, $3.val); }$
- | Term / Factor ${ mpq_init($0.val); mpq_div($0.val, $1.val, $3.val); }$
- | Factor ${ mpq_init($0.val); mpq_set($0.val, $1.val); }$
-
- Factor -> NUMBER ${ if (number_parse($0.val, $0.tail, $1.txt) == 0) mpq_init($0.val); }$
+ Expression -> Expression + Expression ${ mpq_init($0.val); mpq_add($0.val, $1.val, $3.val); }$
+ | Expression - Expression ${ mpq_init($0.val); mpq_sub($0.val, $1.val, $3.val); }$
+ | Expression * Expression ${ mpq_init($0.val); mpq_mul($0.val, $1.val, $3.val); }$
+ | 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, $2.val); }$
code / ocean / blobdiff