This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.
| Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
|---|---|---|
| Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |
| Other format: | [Raw text] | |
The following patch supersedes my previous patch: it addresses two
parser related issues: one is that:
| | The parser declares token TYPENAME and decl.h declares an enum
| | including TYPENAME. Several files include both, what results in a
| | dangerous clash.
| |
| | Given the risks, I suggest that the parser slowly moves to using the
| | `tFOO' convention, so, amongst the various possible changes, I
| | preferred this one: instead of token TYPENAME, using token tTYPENAME.
| | If you have another preference, please tell me, I'll adjust the patch
| | accordingly.
|
| Without this patch, expect next Bison to create a failure of the
| compilation: it nows outputs the tokens as enums (when __STDC__), and
| therefore, the compilation will fail (as is to be expected from id
| clashes).
And the second issue is a typing issue. On a rule such as
foo: bar {} baz;
because of the default action $$ = $1, you have to ensure %typeof (foo)
== %typeof (bar), which was not the case of these rules:
explicit_instantiation:
[...]
| SCSPEC TEMPLATE begin_explicit_instantiation typespec ';'
{ do_type_instantiation ($4.t, $1, 1);
yyungetc (';', 1); }
end_explicit_instantiation
{}
[...]
Because %type <ttype> SCSPEC.
So you have to write
foo: bar {} baz {};
The test suite exhibits exactly the same behavior before and after
this patch, i.e.:
Test Run By akim on Wed Jun 12 12:34:58 2002
Native configuration is i686-pc-linux-gnu
=== g++ tests ===
[..]
=== g++ Summary ===
# of expected passes 6808
# of unexpected failures 568
# of unexpected successes 1
# of expected failures 86
# of untested testcases 9
# of unsupported tests 3
/home/akim/src/gcc/gcc/testsuite/../g++ version gcc 3.2 20020611 (experimental)
I would like to emphasize that this patch is a simple bug fix. As is,
this grammar does not compile with CVS Bison.
Note: I have my editor systematically remove the trailing spaces, and
there appears to be many in these files. So the patch comes in two
``flavors'': the first one is `diff -w' so that you can see the real
content of the patch, but the real patch, which includes the trailing
blank (automated) removal, is below.
Index: ChangeLog
from Akim Demaille <akim@epita.fr>
* parse.y (TYPENAME): Rename as tTYPENAME to avoid the clash with
decl.h's TYPENAME.
* spew.c, lex.c: Adjust.
* parse.y (explicit_instantiation): Add empty action to override
the default $$ = $1 where it introduces a type clash.
Index: lex.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cp/lex.c,v
retrieving revision 1.282
diff -u -u -w -r1.282 lex.c
--- lex.c 4 Jun 2002 07:10:10 -0000 1.282
+++ lex.c 12 Jun 2002 11:14:43 -0000
@@ -86,10 +86,11 @@
int warn_traditional = 0;
int flag_digraphs = 1;
-/* the declaration found for the last IDENTIFIER token read in.
- yylex must look this up to detect typedefs, which get token type TYPENAME,
- so it is left around in case the identifier is not a typedef but is
- used in a context which makes it a reference to a variable. */
+/* the declaration found for the last IDENTIFIER token read in. yylex
+ must look this up to detect typedefs, which get token type
+ tTYPENAME, so it is left around in case the identifier is not a
+ typedef but is used in a context which makes it a reference to a
+ variable. */
tree lastiddecl;
/* Array for holding counts of the numbers of tokens seen. */
@@ -692,7 +693,7 @@
switch (yychar)
{
case IDENTIFIER:
- case TYPENAME:
+ case tTYPENAME:
case TYPESPEC:
case PTYPENAME:
case PFUNCNAME:
@@ -914,7 +915,7 @@
if ((yychar > 255
&& yychar != SCSPEC
&& yychar != IDENTIFIER
- && yychar != TYPENAME
+ && yychar != tTYPENAME
&& yychar != CV_QUALIFIER
&& yychar != SELFNAME)
|| yychar == 0 /* EOF */)
Index: parse.y
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cp/parse.y,v
retrieving revision 1.262
diff -u -u -w -r1.262 parse.y
--- parse.y 4 Jun 2002 07:10:14 -0000 1.262
+++ parse.y 12 Jun 2002 11:14:44 -0000
@@ -283,7 +283,7 @@
/* All identifiers that are declared typedefs in the current block.
In some contexts, they are treated just like IDENTIFIER,
but they can also serve as typespecs in declarations. */
-%token TYPENAME
+%token tTYPENAME
%token SELFNAME
/* A template function. */
@@ -345,7 +345,7 @@
%nonassoc IF
%nonassoc ELSE
-%left IDENTIFIER PFUNCNAME TYPENAME SELFNAME PTYPENAME SCSPEC TYPESPEC CV_QUALIFIER ENUM AGGR ELLIPSIS TYPEOF SIGOF OPERATOR NSNAME TYPENAME_KEYWORD ATTRIBUTE
+%left IDENTIFIER PFUNCNAME tTYPENAME SELFNAME PTYPENAME SCSPEC TYPESPEC CV_QUALIFIER ENUM AGGR ELLIPSIS TYPEOF SIGOF OPERATOR NSNAME TYPENAME_KEYWORD ATTRIBUTE
%left '{' ',' ';'
@@ -375,7 +375,7 @@
%type <code> unop
-%type <ttype> identifier IDENTIFIER TYPENAME CONSTANT expr nonnull_exprlist
+%type <ttype> identifier IDENTIFIER tTYPENAME CONSTANT expr nonnull_exprlist
%type <ttype> PFUNCNAME maybe_identifier
%type <ttype> paren_expr_or_null nontrivial_exprlist SELFNAME
%type <ttype> expr_no_commas expr_no_comma_rangle
@@ -1023,7 +1023,7 @@
identifier:
IDENTIFIER
- | TYPENAME
+ | tTYPENAME
| SELFNAME
| PTYPENAME
| NSNAME
@@ -1060,17 +1060,21 @@
{ do_type_instantiation ($4.t, $1, 1);
yyungetc (';', 1); }
end_explicit_instantiation
+ {}
| SCSPEC TEMPLATE begin_explicit_instantiation typed_declspecs
declarator
{ tree specs = strip_attrs ($4.t);
do_decl_instantiation (specs, $5, $1); }
end_explicit_instantiation
+ {}
| SCSPEC TEMPLATE begin_explicit_instantiation notype_declarator
{ do_decl_instantiation (NULL_TREE, $4, $1); }
end_explicit_instantiation
+ {}
| SCSPEC TEMPLATE begin_explicit_instantiation constructor_declarator
{ do_decl_instantiation (NULL_TREE, $4, $1); }
end_explicit_instantiation
+ {}
;
begin_explicit_instantiation:
@@ -1089,7 +1093,7 @@
PTYPENAME '<' template_arg_list_opt template_close_bracket
finish_template_type_
{ $$ = $5; }
- | TYPENAME '<' template_arg_list_opt template_close_bracket
+ | tTYPENAME '<' template_arg_list_opt template_close_bracket
finish_template_type_
{ $$ = $5; }
| self_template_type
@@ -1557,7 +1561,7 @@
unqualified_id:
notype_unqualified_id
- | TYPENAME
+ | tTYPENAME
| SELFNAME
;
@@ -2770,7 +2774,7 @@
after_type_declarator maybeasm maybe_attribute maybe_init
{ $$ = parse_field0 ($1, $<ftype>0.t, $<ftype>0.lookups,
$3, $2, $4); }
- | TYPENAME ':' expr_no_commas maybe_attribute
+ | tTYPENAME ':' expr_no_commas maybe_attribute
{ $$ = parse_bitfield0 ($1, $<ftype>0.t, $<ftype>0.lookups,
$4, $3); }
;
@@ -2793,7 +2797,7 @@
after_type_component_declarator:
after_type_declarator maybeasm maybe_attribute maybe_init
{ $$ = parse_field ($1, $3, $2, $4); }
- | TYPENAME ':' expr_no_commas maybe_attribute
+ | tTYPENAME ':' expr_no_commas maybe_attribute
{ $$ = parse_bitfield ($1, $4, $3); }
;
@@ -3065,7 +3069,7 @@
;
type_name:
- TYPENAME
+ tTYPENAME
| SELFNAME
| template_type %prec EMPTY
;
@@ -3089,7 +3093,7 @@
/* Why the @#$%^& do type_name and notype_identifier need to be expanded
inline here?!? (jason) */
nested_name_specifier_1:
- TYPENAME SCOPE
+ tTYPENAME SCOPE
{
if (TREE_CODE ($1) == IDENTIFIER_NODE)
{
@@ -3175,7 +3179,7 @@
/* This needs to return a TYPE_DECL for simple names so that we don't
forget what name was used. */
typename_sub2:
- TYPENAME SCOPE
+ tTYPENAME SCOPE
{
if (TREE_CODE ($1) != TYPE_DECL)
$$ = lastiddecl;
@@ -3613,7 +3617,7 @@
{ finish_label_stmt ($1); }
| PTYPENAME ':'
{ finish_label_stmt ($1); }
- | TYPENAME ':'
+ | tTYPENAME ':'
{ finish_label_stmt ($1); }
| SELFNAME ':'
{ finish_label_stmt ($1); }
Index: spew.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/cp/spew.c,v
retrieving revision 1.65
diff -u -u -w -r1.65 spew.c
--- spew.c 4 Jun 2002 07:10:23 -0000 1.65
+++ spew.c 12 Jun 2002 11:14:44 -0000
@@ -161,10 +161,11 @@
static tree last_token_id;
/* From lex.c: */
-/* the declaration found for the last IDENTIFIER token read in.
- yylex must look this up to detect typedefs, which get token type TYPENAME,
- so it is left around in case the identifier is not a typedef but is
- used in a context which makes it a reference to a variable. */
+/* the declaration found for the last IDENTIFIER token read in. yylex
+ must look this up to detect typedefs, which get token type
+ tTYPENAME, so it is left around in case the identifier is not a
+ typedef but is used in a context which makes it a reference to a
+ variable. */
extern tree lastiddecl; /* let our brains leak out here too */
extern int yychar; /* the lookahead symbol */
extern YYSTYPE yylval; /* the semantic value of the */
@@ -636,11 +637,11 @@
if (t && t == decl)
return SELFNAME;
- return TYPENAME;
+ return tTYPENAME;
}
/* token[0] == AGGR (struct/union/enum)
- Thus, token[1] is either a TYPENAME or a TYPENAME_DEFN.
+ Thus, token[1] is either a tTYPENAME or a TYPENAME_DEFN.
If token[2] == '{' or ':' then it's TYPENAME_DEFN.
It's also a definition if it's a forward declaration (as in 'struct Foo;')
which we can tell if token[2] == ';' *and* token[-1] != FRIEND or NEW. */
@@ -652,7 +653,7 @@
scan_tokens (2);
yc1 = nth_token (1)->yychar;
- if (yc1 != TYPENAME && yc1 != IDENTIFIER && yc1 != PTYPENAME)
+ if (yc1 != tTYPENAME && yc1 != IDENTIFIER && yc1 != PTYPENAME)
return;
yc2 = nth_token (2)->yychar;
if (yc2 == ';')
@@ -667,7 +668,7 @@
switch (yc1)
{
- case TYPENAME:
+ case tTYPENAME:
nth_token (1)->yychar = TYPENAME_DEFN;
break;
case PTYPENAME:
@@ -765,7 +766,7 @@
break;
}
case IDENTIFIER_DEFN:
- case TYPENAME:
+ case tTYPENAME:
case TYPENAME_DEFN:
case PTYPENAME:
case PTYPENAME_DEFN:
@@ -898,7 +899,7 @@
yyc = identifier_type (trrr);
switch(yyc)
{
- case TYPENAME:
+ case tTYPENAME:
case SELFNAME:
case NSNAME:
case PTYPENAME:
@@ -1469,7 +1470,7 @@
{
if (yy<256)
fprintf (stderr, "->%d < %c >\n", lineno, yy);
- else if (yy == IDENTIFIER || yy == TYPENAME)
+ else if (yy == IDENTIFIER || yy == tTYPENAME)
{
const char *id;
if (TREE_CODE (yylval.ttype) == IDENTIFIER_NODE)
Now the real patch, including trailing spaces nuking.
Attachment:
fix-TYPENAME-conflict.patch.gz
Description: Binary data
| Index Nav: | [Date Index] [Subject Index] [Author Index] [Thread Index] | |
|---|---|---|
| Message Nav: | [Date Prev] [Date Next] | [Thread Prev] [Thread Next] |