Bison bug fix for non-ascii unibyte char tokens on signed-char hosts

[Paul Eggert]

Just for fun I tried Bison on this grammar on a Solaris 8 sparc host:

%%
start: '¿';
%%

(That character token is an upside-down question mark in Latin-1 encoding.)

Bison reported the following bogus message:

  tokens $ and '\277' both assigned number -65

I installed the following patch, which fixes a bug for non-ASCII
unibyte char tokens on hosts that have signed char.  Perhaps this
issue should be documented in bison.texinfo, since it means that on
signed-char hosts '¿' == 191 in Bison even though '¿' == -65 in C.
However, I'm a bit leery about saying too much about this issue given
our aversion to supporting non-ASCII character literal tokens.

2002-07-01  Paul Eggert  <address@hidden>

        * src/scan-gram.l (<SC_ESCAPED_CHARACTER>): Convert to unsigned
        char, so that negative chars don't collide with $.

Index: scan-gram.l
===================================================================
RCS file: /cvsroot/bison/bison/src/scan-gram.l,v
retrieving revision 1.18
retrieving revision 1.19
diff -p -u -r1.18 -r1.19
--- scan-gram.l 30 Jun 2002 17:34:52 -0000      1.18
+++ scan-gram.l 1 Jul 2002 08:36:37 -0000       1.19
@@ -294,7 +294,8 @@ blanks   [ \t\f]+
       YY_OBS_FINISH;
       yylval->symbol = symbol_get (last_string, *yylloc);
       symbol_class_set (yylval->symbol, token_sym, *yylloc);
-      symbol_user_token_number_set (yylval->symbol, last_string[1], *yylloc);
+      symbol_user_token_number_set (yylval->symbol,
+                                   (unsigned char) last_string[1], *yylloc);
       YY_OBS_FREE;
       yy_pop_state ();
       return ID;