[Bug c++/29054] [4.0/4.1 Regression] ICE on friend template specialization

["jakub at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #4 from jakub at gcc dot gnu dot org  2007-01-02 15:58 -------
I believe this was fixed by the PR26912 (4.2+ version) fix.
Before those changes, the this argument was added by grokclassfn, after that
change it is being added to all METHOD_TYPE type decls in grokfndecl.
grokfndecl is here called from the friend decl, so grokdeclarator's:
8388                /* All method decls are public, so tell grokfndecl to set
8389                   TREE_PUBLIC, also.  */
8390                decl = grokfndecl (ctype, type,
8391                                   TREE_CODE (unqualified_id) !=
TEMPLATE_ID_EXPR
8392                                   ? unqualified_id : dname,
8393                                   parms,
8394                                   unqualified_id,
8395                                   virtualp, flags, quals, raises,
8396                                   friendp ? -1 : 0, friendp, 1, 0, sfk,
8397                                   funcdef_flag, template_count,
in_namespace,
friendp is non-zero and therefore grokclassfn is never called.
Later on revert_static_member_fn assumes the first argument is "this" without
checking it and eats the first argument.

Mark, do you think it is better for 4.1 to backport another part of your
changes
(i.e. the build_this_parm function addition and related changes), or e.g.
just copy over the "this" addition code after this grokfndecl call if
(friendp)?
Or change revert_static_method to only remove "this" argument?


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mark at codesourcery dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=29054