]> ocean-lang.org Git - ocean/blobdiff - csrc/parsergen.mdc
parsergen: add token location to tracing
[ocean] / csrc / parsergen.mdc
index 04974c3c6a7d749908ba3e672179e253082fca4e..c167e0967cddac00f2fe937027ff3165b45f37bd 100644 (file)
@@ -499,6 +499,8 @@ Now we have all the bits we need to parse a full production.
                        vs = sym_find(g, tk.txt);
                        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;
@@ -1309,6 +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))
@@ -1330,26 +1334,26 @@ into the go to set, so the item is ineffective.
                        continue;
                if (symset_find(&done, s->num) < 0) {
                        symset_add(&done, s->num, 0);
-                       if (s->line_like)
-                               is->starts_line = 1;
                }
                if (s->type != Nonterminal)
                        continue;
+               if (s->line_like)
+                       is->starts_line = 1;
                again = 1;
                if (type >= LALR) {
                        // Need the LA set.
                        int to_end;
                        add_first(pr, bs+1, &LA, g, &to_end);
                        if (to_end) {
-                               if (pr->line_like)
-                                       symset_add(&LA, TK_newline, 0);
-                               else {
-                                       struct symset ss = set_find(g, is->items.data[i]);
-                                       symset_union(&LA, &ss);
-                               }
+                               struct symset ss = set_find(g, is->items.data[i]);
+                               symset_union(&LA, &ss);
                        }
                        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 */
@@ -1360,19 +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);
                        }
                }
@@ -1643,8 +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)
+               if (pr->line_like == 1)
                        printf(" $$NEWLINE");
+               else if (pr->line_like)
+                       printf(" $$OUT");
                printf("\n");
        }
 
@@ -1783,14 +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)
        {
@@ -1844,12 +1852,9 @@ counted, and are reported as non-critical.  This will not affect a
                                for (k = 0; k < la.cnt; k++) {
                                        int pos = symset_find(&shifts, la.syms[k]);
                                        if (pos >= 0 && la.syms[k] != TK_newline) {
-                                               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);
+                                               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);
@@ -2866,9 +2871,13 @@ checks if a given token is in any of these look-ahead sets.
                force_reduce:
                        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
+                           !(tk->num == TK_newline ||
+                             tk->num == TK_eof ||
+                             tk->num == TK_out ||
+                             (tos->indents == 0 && tos->since_newline == 0))) {
+                               /* Anything other than newline or out or eof
+                                * in an error unless we are already at start
+                                * of line, as this production must end at EOL.
                                 */
                        } else if (states[tos->state].reduce_prod >= 0) {
                                void **body;
@@ -3020,7 +3029,7 @@ end inside square brackets.
                        fputs(reserved_words[tk->num], trace);
                else
                        text_dump(trace, tk->txt, 20);
-               fputs("]", trace);
+               fprintf(trace, ":%d:%d]", tk->line, tk->col);
        }
 
        void parser_trace_action(FILE *trace, char *action)