FYI: alias undeclared or declared after use

[Joel E. Denny]

Bison incorrectly assumed that string literals would not be declared as 
aliases after use.  Doing so caused an assertion failure.  Below is a 
committed patch that checks and fixes this bug.

As part of the fix, Bison now reports an error for any string literal that 
is never declared as an alias.  I just don't see the need for it in a real 
parser: there's no token number to return from the scanner.  I figure this 
was just an accident... although a few test cases exploited it.  If 
necessary, I believe that making undeclared string literals possible again 
would only require a little more work.

Joel

Index: ChangeLog
===================================================================
RCS file: /sources/bison/bison/ChangeLog,v
retrieving revision 1.1550
diff -p -u -r1.1550 ChangeLog
--- ChangeLog   14 Aug 2006 22:40:33 -0000      1.1550
+++ ChangeLog   18 Aug 2006 10:05:17 -0000
@@ -1,3 +1,28 @@
+2006-08-18  Joel E. Denny  <address@hidden>
+
+       Don't allow an undeclared string literal, but allow a string literal to
+       be used before its declaration.
+       * src/reader.c (check_and_convert_grammar): Don't invoke packgram if
+       symbols_pack complained.
+       * src/symtab.c (symbol_new): Don't count a string literal as a new
+       symbol.
+       (symbol_class_set): Don't count a string literal as a new token, and
+       don't assign it a symbol number since symbol_make_alias does that.
+       (symbol_make_alias): It's not necessary to decrement the symbol and
+       token counts anymore.  Don't assume that an alias declaration occurs
+       before any uses of the identifier or string, and thus don't assert that
+       one of them has the highest symbol number so far.
+       (symbol_check_alias_consistency): Complain if there's a string literal
+       that wasn't declared as an alias.
+       (symbols_pack): Bail if symbol_check_alias_consistency failed since
+       symbol_pack asserts that every token has been assigned a symbol number
+       although undeclared string literals have not.
+       * tests/regression.at (String alias declared after use, Undeclared
+       string literal): New test case.
+       (Characters Escapes, Web2c Actions): Declare string literals as
+       aliases.
+       * tests/sets.at (Firsts): Likewise.
+
 2006-08-14  Joel E. Denny  <address@hidden>
 
        In the grammar scanner, STRING_FINISH unclosed constructs and return
Index: src/reader.c
===================================================================
RCS file: /sources/bison/bison/src/reader.c,v
retrieving revision 1.267
diff -p -u -r1.267 reader.c
--- src/reader.c        29 Jul 2006 05:53:41 -0000      1.267
+++ src/reader.c        18 Aug 2006 10:05:17 -0000
@@ -630,7 +630,8 @@ check_and_convert_grammar (void)
   symbols_pack ();
 
   /* Convert the grammar into the format described in gram.h.  */
-  packgram ();
+  if (!complaint_issued)
+    packgram ();
 
   /* The grammar as a symbol_list is no longer needed. */
   LIST_FREE (symbol_list, grammar);
Index: src/symtab.c
===================================================================
RCS file: /sources/bison/bison/src/symtab.c,v
retrieving revision 1.75
diff -p -u -r1.75 symtab.c
--- src/symtab.c        14 Aug 2006 00:34:17 -0000      1.75
+++ src/symtab.c        18 Aug 2006 10:05:17 -0000
@@ -79,7 +79,8 @@ symbol_new (uniqstr tag, location loc)
   if (nsyms == SYMBOL_NUMBER_MAXIMUM)
     fatal (_("too many symbols in input grammar (limit is %d)"),
           SYMBOL_NUMBER_MAXIMUM);
-  nsyms++;
+  if (tag[0] != '"')
+    nsyms++;
   return res;
 }
 
@@ -266,7 +267,8 @@ symbol_class_set (symbol *sym, symbol_cl
 
   if (class == nterm_sym && sym->class != nterm_sym)
     sym->number = nvars++;
-  else if (class == token_sym && sym->number == NUMBER_UNDEFINED)
+  else if (class == token_sym && sym->number == NUMBER_UNDEFINED
+           && sym->tag[0] != '"')
     sym->number = ntokens++;
 
   sym->class = class;
@@ -361,12 +363,7 @@ symbol_make_alias (symbol *sym, symbol *
       sym->user_token_number = USER_NUMBER_ALIAS;
       symval->alias = sym;
       sym->alias = symval;
-      /* sym and symval combined are only one symbol.  */
-      nsyms--;
-      ntokens--;
-      assert (ntokens == sym->number || ntokens == symval->number);
-      sym->number = symval->number =
-       (symval->number < sym->number) ? symval->number : sym->number;
+      symval->number = sym->number;
       symbol_type_set (symval, sym->type_name, loc);
     }
 }
@@ -383,6 +380,9 @@ symbol_check_alias_consistency (symbol *
   symbol *alias = this;
   symbol *orig  = this->alias;
 
+  if (this->tag[0] == '"' && !this->alias)
+    complain_at (this->location, _("%s undeclared"), this->tag);
+
   /* Check only those that _are_ the aliases.  */
   if (!(this->alias && this->user_token_number == USER_NUMBER_ALIAS))
     return;
@@ -723,6 +723,8 @@ symbols_pack (void)
   symbols = xcalloc (nsyms, sizeof *symbols);
 
   symbols_do (symbol_check_alias_consistency_processor, NULL);
+  if (complaint_issued)
+    return;
   symbols_do (symbol_pack_processor, NULL);
 
   symbols_token_translations_init ();
Index: tests/regression.at
===================================================================
RCS file: /sources/bison/bison/tests/regression.at,v
retrieving revision 1.105
diff -p -u -r1.105 regression.at
--- tests/regression.at 14 Aug 2006 20:51:33 -0000      1.105
+++ tests/regression.at 18 Aug 2006 10:05:18 -0000
@@ -489,7 +489,9 @@ AT_DATA_GRAMMAR([input.y],
 void yyerror (const char *s);
 int yylex (void);
 %}
-[%%
+[%token QUOTES "\""
+%token TICK "'"
+%%
 exp:
   '\'' "\'"
 | '\"' "\""
@@ -700,6 +702,10 @@ statement:  struct_stat;
 struct_stat:  /* empty. */ | if else;
 if: "if" "const" "then" statement;
 else: "else" statement;
+%token IF "if";
+%token CONST "const";
+%token THEN "then";
+%token ELSE "else";
 %%
 ]])
 
@@ -1108,3 +1114,48 @@ Stack now 0
 ]])
 
 AT_CLEANUP
+
+
+
+## --------------------------------- ##
+## String alias declared after use.  ##
+## --------------------------------- ##
+
+AT_SETUP([String alias declared after use])
+
+# Bison once incorrectly asserted that the symbol number for either a token or
+# its alias was the highest symbol number so far at the point of the alias
+# declaration.  That was true unless the declaration appeared after their first
+# uses.
+
+AT_DATA([input.y],
+[[%%
+start: 'a' "A" 'b';
+%token 'a' "A";
+]])
+
+AT_CHECK([bison -t -o input.c input.y])
+
+AT_CLEANUP
+
+
+
+## --------------------------- ##
+## Undeclared string literal.  ##
+## --------------------------- ##
+
+AT_SETUP([Undeclared string literal])
+
+# Bison once allowed a string literal to be used in the grammar without any
+# declaration assigning it as an alias of another token.
+
+AT_DATA([input.y],
+[[%%
+start: "abc";
+]])
+
+AT_CHECK([bison -t -o input.c input.y], [1], [],
+[[input.y:2.8-12: "abc" undeclared
+]])
+
+AT_CLEANUP
Index: tests/sets.at
===================================================================
RCS file: /sources/bison/bison/tests/sets.at,v
retrieving revision 1.21
diff -p -u -r1.21 sets.at
--- tests/sets.at       10 Dec 2005 00:25:27 -0000      1.21
+++ tests/sets.at       18 Aug 2006 10:05:18 -0000
@@ -196,6 +196,7 @@ AT_DATA([input.y],
 [[%nonassoc '<' '>'
 %left '+' '-'
 %right '^' '='
+%token EXP "exp"
 %%
 exp:
    exp '<' exp