]> ocean-lang.org Git - ocean/blobdiff - csrc/parsergen.mdc
parsergen: add handling for TK_IN and TK_OUT
[ocean] / csrc / parsergen.mdc
index 252b3d6f6c88724c90b93163f392b0b4bfd23279..d2d70d531d163a67d1de05a1682d71573b37b404 100644 (file)
@@ -2260,13 +2260,17 @@ We keep the stack as two separate allocations.  One, `asn_stack` stores the
 production, and by keeping a separate `asn` stack, we can just pass a
 pointer into this stack.
 
-The other allocation stores all other stack fields of which there are two.
+The other allocation stores all other stack fields of which there are four.
 The `state` is the most important one and guides the parsing process.  The
 `sym` is nearly unnecessary.  However when we want to free entries from the
 `asn_stack`, it helps to know what type they are so we can call the right
 freeing function.  The symbol leads us to the right free function through
 `do_free`.
 
+The `indents` count and the `starts_indented` flag track the line
+indents in the symbol.  These are used to allow indent information to
+guide parsing and error recovery.
+
 As well as the stack of frames we have a `next` frame which is
 assembled from the incoming token and other information prior to
 pushing it onto the stack.
@@ -2277,6 +2281,8 @@ pushing it onto the stack.
                struct frame {
                        short state;
                        short sym;
+                       short starts_indented;
+                       short indents;
                } *stack, next;
                void **asn_stack;
                int stack_size;
@@ -2322,6 +2328,8 @@ if needed and pushed all the information onto the stacks.
                p->asn_stack[p->tos] = asn;
                p->tos++;
                p->next.state = newstate;
+               p->next.indents = 0;
+               p->next.starts_indented = 0;
                return 1;
        }
 
@@ -2336,12 +2344,16 @@ reduce a production, just before we `shift` the nonterminal in.
        {
                int i;
                p->tos -= num;
-               for (i = 0; i < num; i++)
+               for (i = 0; i < num; i++) {
+                       p->next.indents += p->stack[p->tos+i].indents;
                        do_free(p->stack[p->tos+i].sym,
                                p->asn_stack[p->tos+i]);
+               }
 
-               if (num)
+               if (num) {
                        p->next.state = p->stack[p->tos].state;
+                       p->next.starts_indented = p->stack[p->tos].starts_indented;
+               }
        }
 
 ### Memory allocation
@@ -2385,6 +2397,17 @@ If we can neither shift nor reduce we have an error to handle.  We pop
 single entries off the stack until we can shift the `TK_error` symbol, then
 drop input tokens until we find one we can shift into the new error state.
 
+When we find `TK_in` and `TK_out` tokens which report indents we need
+to handle them directly as the grammar cannot express what we want to
+do with them.
+
+`TK_in` tokens are easy: we simply update the `next` stack frame to
+record how many indents there are and that the next token started with
+an indent.
+
+`TK_out` tokens must either be counted off against any pending indent,
+or must force reductions until there is a pending indent which isn't
+at the start of a production.
 
 ###### parser includes
        #include "parser.h"
@@ -2396,18 +2419,39 @@ drop input tokens until we find one we can shift into the new error state.
                         FILE *trace, const char *non_term[], int knowns)
        {
                struct parser p = { 0 };
-               struct token *tk;
+               struct token *tk = NULL;
                int accepted = 0;
                void *ret;
 
-               tk = tok_copy(token_next(tokens));
                while (!accepted) {
+                       struct token *err_tk;
+                       if (!tk)
+                               tk = tok_copy(token_next(tokens));
                        p.next.sym = tk->num;
                        if (trace)
                                parser_trace(trace, &p, tk, states, non_term, knowns);
 
+                       if (p.next.sym == TK_in) {
+                               p.next.starts_indented = 1;
+                               p.next.indents = 1;
+                               free(tk);
+                               tk = NULL;
+                               continue;
+                       }
+                       if (p.next.sym == TK_out) {
+                               if (p.stack[p.tos-1].indents > p.stack[p.tos-1].starts_indented ||
+                                   (p.stack[p.tos-1].indents == 1 &&
+                                    states[p.next.state].reduce_size > 1)) {
+                                       p.stack[p.tos-1].indents -= 1;
+                                       free(tk);
+                                       tk = NULL;
+                                       continue;
+                               }
+                               // fall through and force a REDUCE (as 'shift'
+                               // will fail).
+                       }
                        if (shift(&p, tk, states)) {
-                               tk = tok_copy(token_next(tokens));
+                               tk = NULL;
                                continue;
                        }
                        if (states[p.next.state].reduce_prod >= 0) {
@@ -2429,6 +2473,15 @@ drop input tokens until we find one we can shift into the new error state.
                                        accepted = 1;
                                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[p.next.state].shift_sym);
+                               p.next.sym = states[p.next.state].shift_sym;
+                               shift(&p, tok_copy(*tk), states);
+                               // FIXME need to report this error somehow
+                               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
@@ -2436,16 +2489,29 @@ drop input tokens until we find one we can shift into the new error state.
                         * Then we discard input tokens until
                         * we find one that is acceptable.
                         */
+
+                       err_tk = tok_copy(*tk);
                        p.next.sym = TK_error;
-                       while (shift(&p, tk, states) == 0
+                       while (shift(&p, err_tk, states) == 0
                               && p.tos > 0)
                                // discard this state
                                pop(&p, 1, do_free);
-                       tk = tok_copy(*tk);
+                       if (p.tos == 0) {
+                               free(err_tk);
+                               // no state accepted TK_error
+                               break;
+                       }
                        while (search(&states[p.next.state], tk->num) < 0 &&
                               tk->num != TK_eof) {
                                free(tk);
                                tk = tok_copy(token_next(tokens));
+                               if (tk->num == TK_in)
+                                       p.next.indents += 1;
+                               if (tk->num == TK_out) {
+                                       if (p.next.indents == 0)
+                                               break;
+                                       p.next.indents -= 1;
+                               }
                        }
                        if (p.tos == 0 && tk->num == TK_eof)
                                break;