]> ocean-lang.org Git - ocean/blobdiff - csrc/parsergen.mdc
parsergen: allow $$OUT as well as $$NEWLINE
[ocean] / csrc / parsergen.mdc
index d12a4000d3712f6a44eaf486a284793073e1abf8..76fec3a70a779cefe4589a9d79e861a61acd726e 100644 (file)
@@ -100,6 +100,7 @@ symbol.
        struct production {
                unsigned short precedence;
                enum assoc assoc;
+               char line_like;
                ## production fields
        };
        struct grammar {
@@ -496,12 +497,17 @@ Now we have all the bits we need to parse a full production.
                                goto abort;
                        }
                        vs = sym_find(g, tk.txt);
-                       if (vs->type != Virtual) {
-                               err = "symbol after $$ must be virtual";
+                       if (vs->num == TK_newline)
+                               p.line_like = 1;
+                       else if (vs->num == TK_out)
+                               p.line_like = 2;
+                       else if (vs->precedence == 0) {
+                               err = "symbol after $$ 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) {
@@ -1305,6 +1311,8 @@ into the go to set, so the item is ineffective.
                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))
@@ -1342,6 +1350,10 @@ into the go to set, so the item is ineffective.
                        }
                        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 */
@@ -1352,19 +1364,25 @@ into the go to set, so the item is ineffective.
                        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);
                        }
                }
@@ -1635,6 +1653,10 @@ it up a bit.  First the items, with production number and associativity.
                        printf(" [%d%s]", s->precedence,
                               assoc_names[s->assoc]);
                }
+               if (pr->line_like == 1)
+                       printf(" $$NEWLINE");
+               else if (pr->line_like)
+                       printf(" $$OUT");
                printf("\n");
        }
 
@@ -1773,14 +1795,10 @@ terminals to items where that terminal could be shifted and another
 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, where a
-terminal that could be shifted is in the lookahead set of some
-reducable item, then set check if the reducable item also have
-`TK_newline` in its lookahead set.  If it does, then a newline will
-force the reduction, but anything else can reasonably be shifted, so
-that isn't really a conflict.  Such apparent conflicts do not get
-counted, and are reported as non-critical.  This will not affect a
-"traditional" grammar that does not include newlines as token.
+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)
        {
@@ -1800,14 +1818,20 @@ counted, and are reported as non-critical.  This will not affect a
                                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 reductions and conflicts */
                        for (j = 0; j < is->items.cnt; j++) {
@@ -1827,13 +1851,10 @@ counted, and are reported as non-critical.  This will not affect a
                                int k;
                                for (k = 0; k < la.cnt; k++) {
                                        int pos = symset_find(&shifts, la.syms[k]);
-                                       if (pos >= 0) {
-                                               if (symset_find(&la, TK_newline) < 0) {
-                                                       printf("  State %d has SHIFT/REDUCE conflict on ", i);
-                                                       cnt++;
-                                               } else
-                                                       printf("  State %d has non-critical SHIFT/REDUCE conflict on ", i);
-                                               prtxt(g->symtab[la.syms[k]]->name);
+                                       if (pos >= 0 && la.syms[k] != TK_newline) {
+                                               printf("  State %d has SHIFT/REDUCE conflict on ", i);
+                                               cnt++;
+                                                       prtxt(g->symtab[la.syms[k]]->name);
                                                printf(":\n");
                                                report_item(g, shifts.data[pos]);
                                                report_item(g, itm);
@@ -1958,7 +1979,8 @@ The go to table is stored in a simple array of `sym` and corresponding
                short reduce_prod;
                short reduce_size;
                short reduce_sym;
-               short starts_line;
+               char starts_line;
+               char newline_only;
                short min_prefix;
        };
 
@@ -2007,13 +2029,15 @@ The go to table is stored in a simple array of `sym` and corresponding
                        }
 
                        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);
                }
@@ -2106,10 +2130,13 @@ automatically freed.  This is equivalent to assigning `NULL` to the pointer.
                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);
        }
@@ -2842,7 +2869,13 @@ checks if a given token is in any of these look-ahead sets.
                                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];