Handling for the initial token being TOK_UNKNOWN.

My intended idea was to have all syntax errors go through the same actions driven by parser.sid. But in this case SID's generated parser bails out immediately on TOK_UNKNOWN, and we never reach an ## alt. So we never reach <make-ast>, and thus never <err-syntax> about it.

That's what the assertion was about; the generated parser should never return with a NULL AST.

My workaround for this is just to recreate the equivalent syntax error at the top-level, which isn't very satisfying at all. But it does what we need here.

Spotted by @cinco-de-mayonnaise, thank you.

src/lx/parser.act

@@ -869,7 +869,19 @@
 
  lx->free(lx->buf_opaque);
 
- assert(ast != NULL);
+ /*
+ * This can happen when the first token is TOK_UNKNOWN.
+ * SID's generated parser bails out immediately.
+ * So we never reach <make-ast>, and never <err-syntax> about it.
+ *
+ * Really I wanted this handled along with the usual syntax error
+ * case, from ## alts inside parser.sid, instead of reproducing
+ * the same error here.
+ */
+ if (ast == NULL) {
+ err(lex_state, "Syntax error");
+ exit(EXIT_FAILURE);
+ }
 
  return ast;
  }