URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [gcc/] [cp/] [ChangeLog-1994] - Rev 720
Go to most recent revision | Compare with Previous | Blame | View Log
Fri Dec 30 17:57:30 1994 Mike Stump <mrs@cygnus.com>* gc.c (build_bltn_desc): Handle bool as a built-in type.Fri Dec 30 14:20:21 1994 Mike Stump <mrs@cygnus.com>* tree.c (layout_vbasetypes): Ensure that we don't loose alignmenton the complete type because of small virtual bases.Fri Dec 30 12:22:29 1994 Mike Stump <mrs@cygnus.com>* decl.c (n_incomplete): Bump n_incomplete up to int to match Cfront end.(pushdecl): Also count decls pushed that are of a type being definedas incomplete things.* class.c (finish_struct): Move hack_incomplete_structures up tojust after we set it as not being defined, so that the decls webuild for RTTI don't count as incomplete.Thu Dec 29 18:20:57 1994 Mike Stump <mrs@cygnus.com>* pt.c (tsubst): Fix problem with defining constructors in templatedclasses with virtual bases.Wed Dec 28 08:31:00 1994 Mike Stump <mrs@cygnus.com>* parse.y (TYPEID): Strip top-level cv-qualifiers on typeidexpressions.* gc.c (build_typeid): Likewise.Thu Dec 22 17:26:33 1994 Mike Stump <mrs@cygnus.com>* cvt.c (build_up_reference): Fix breakage introduced on Nov 29,don't assert on complex AGGR inits.Thu Dec 22 14:32:31 1994 Mike Stump <mrs@cygnus.com>* method.c (build_overload_value): Handle pointer to members astemplate arguments.Thu Dec 22 13:09:07 1994 Mike Stump <mrs@cygnus.com>* typeck.c (unary_complex_lvalue): Don't call sorry if we know howto do take the address of a data member for a pointer to datamember.Thu Dec 22 10:04:19 1994 Mike Stump <mrs@cygnus.com>* decl.c (grokdeclarator): Use the typedef name for linkage if thetype doesn't otherwise have a name.* decl2.c (grokfield): Likewise.* class.c (finish_struct): Since we reuse the TYPE_DECL for theDECL_NAME of enums, structs and classes, we have to avoid trying toput it in the TYPE_FIELDS again.Wed Dec 21 11:07:05 1994 Mike Stump <mrs@cygnus.com>* decl2.c (check_classfn): Ignore this parameter on static functionswhen checking to see if we match.Tue Dec 20 17:47:02 1994 Mike Stump <mrs@cygnus.com>* typeck.c (unary_complex_lvalue): Handle address of non-left mostpointers to members by calling get_delta_difference.Mon Dec 19 22:40:53 1994 Mike Stump <mrs@cygnus.com>* decl2.c (check_classfn): Don't use decls_match yet, as it modifiesstatic functions to early.Thu Dec 19 22:37:48 1994 Mike Stump <mrs@cygnus.com>* method.c (make_thunk): Handle encoding of positive thunk offsets.Sat Dec 17 13:29:50 1994 Doug Evans <dje@canuck.cygnus.com>* Make-lang.in (.PHONY): Tell GNU make C++ and c++ are phony targets.Thu Dec 15 16:32:12 1994 Mike Stump <mrs@cygnus.com>* decl2.c (check_classfn): Use decls_match to check if this hasalready been declared, as the DECL_ASSEMBLER_NAME may have beenchanged via asm("new_name").* decl.c (decls_match): Make public.Thu Dec 15 15:17:55 1994 Mike Stump <mrs@cygnus.com>* *.[chy] (expand_aggr_init): Add fourth argument to handledistinction between = init and (init) style of initializations.* *.[chy] (finish_decl): Add fifth argument to handledistinction between = init and (init) style of initializations.Tue Dec 13 19:16:05 1994 Mike Stump <mrs@cygnus.com>Fix some random `explicit' bugs.* cvt.c (convert_to_reference): Add third parameter toconvert_force.(convert_force): Likewise.* call.c (build_method_call): Likewise.* decl2.c (setup_vtbl_ptr): Likewise.* init.c (expand_virtual_init): Likewise.(build_member_call): Likewise.(build_delete): Likewise.(build_vbase_delete): Likewise.* typeck.c (build_component_addr): Likewise.(build_c_cast): Likewise.(build_modify_expr): Likewise.* cp-tree.h (CONV_NONCONVERTING): Likewise. Add so that we candistinguish the context in which the conversion appears. Add thridargument to build_c_cast.* cvt.c (cp_convert): Pass whether or not we want to considernon-converting constructors down to build_method_call.* decl2.c (reparse_absdcl_as_casts): Add third argument tobuild_c_cast.* gc.c (build_m_desc): Likewise.* init.c (build_new): Likewise.* parse.y (expr_no_commas): Likewise.(primary): Likewise.* typeck.c (build_x_function_call): Likewise.(build_static_cast): Likewise.(build_reinterpret_cast): Likewise.(build_const_cast): Likewise.(build_c_cast): Likewise.(build_ptrmemfunc): Likewise.* typeck2.c (build_functional_cast): Likewise.* init.c (expand_aggr_init): Added LOOKUP_ONLYCONVERTING toexpand_aggr_init_1 as inits are converted to the destination type.Tue Dec 13 16:18:57 1994 Jason Merrill <jason@phydeaux.cygnus.com>* Make-lang.in (cc1plus): Depends on c-pragma.o.* Makefile.in (OBJ{DEP,}S): Add ../c-pragma.o.* lex.c (check_newline): If the #pragma is not recognized by g++,try machine-specific ones too.(handle_sysv_pragma): Copied from c-lex.c.Mon Dec 12 23:53:06 1994 Mike Stump <mrs@cygnus.com>* except.c (expand_throw): Fix Dec 6th change, build_new likes areference better.Mon Dec 12 18:01:00 1994 Jason Merrill <jason@phydeaux.cygnus.com>* typeck.c (build_binary_op): Lose checks on TYPE_PTRMEMFUNC_P withIS_AGGR_TYPE, since now they will not both be set on the same type.* pt.c (do_pending_expansions): Don't clear TREE_PUBLIC oninstantiations controlled by -fexternal-templates.* decl.c (duplicate_decls): Don't complain about different values of__attribute__ ((const)) and ((noreturn)).Fri Dec 9 18:17:37 1994 Doug Evans <dje@cygnus.com>* Makefile.in (BISONFLAGS): Delete --yacc.(PARSE_H): Depend on $(PARSE_C), for parallel makes.(PARSE_C): Undo last patch.Fri Dec 2 10:44:36 1994 Mike Stump <mrs@cygnus.com>* Makefile.in (BISONFLAGS): Add --yacc so that output winds up iny.tab.c.Thu Dec 8 17:39:46 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (finish_decl): Don't call obscure_complex_init for declsof indeterminate size.Wed Dec 7 16:49:22 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (obscure_complex_init): Function to tweak the decl toprevent expand_decl from tring to initialize it.(finish_decl): Use it rather than writing the same code in threedifferent places.* parse.y (bad_parm): Stop trying to support parms without types.Wed Dec 7 12:06:56 1994 Mike Stump <mrs@cygnus.com>* decl2.c (grokfield): Make asm specs on static member functionswork.Tue Dec 6 15:43:20 1994 Mike Stump <mrs@cygnus.com>* except.c (expand_throw): Make a copy of the thrown object.Tue Dec 6 14:16:34 1994 Jason Merrill <jason@phydeaux.cygnus.com>* parse.y: : Has lower precedence than =.Tue Dec 6 12:46:17 1994 Mike Stump <mrs@cygnus.com>* decl.c (pushdecl): Use DECL_NAME of VAR_DECLs to avoid namespacemanglings.(grokvardecl): Add namespace into variable name.Tue Dec 6 11:26:55 1994 Mike Stump <mrs@cygnus.com>* decl2.c (current_namespace_id): New routine to transform a simplename into a name in a namespace.* decl.c (grokdeclarator): Use it.* decl2.c (get_namespace_id): Find the name of the currentnamespace.(push_namespace, pop_namespace): Complete out missingfunctionality.Mon Dec 5 17:11:51 1994 Jason Merrill <jason@phydeaux.cygnus.com>* class.c (finish_struct): Don't use LONG_LONG_TYPE_SIZE, as it maynot be defined. Fix warning message for enums and restore warningfor non-enums.* decl2.c (push_namespace): Dummy function.(pop_namespace): Likewise.(do_namespace_alias): Likewise.(do_using_decl): Likewise.(do_using_directive): Likewise.* parse.y: New token NSNAME for namespace names.(extdef): Add namespace, using definitions.(using_decl): New rule for using declarations.(any_id): New rule for identifiers with any degree of scoping.(identifier): Add NSNAME.(notype_identifier): Likewise.(component_decl): Add using_decl.(nested_name_specifier): Add NSNAME SCOPE.* typeck.c (convert_for_assignment): Handle conversions betweenenums and bool.* decl.c (duplicate_decls): Only propagate DECL_MAIN_VARIANT onFUNCTION_DECLs.Mon Dec 5 13:03:16 1994 Mike Stump <mrs@cygnus.com>* class.c (finish_struct): Give an error if one tries to declare abit-field's size greater than a long long, as the backend will dump.It is not an error to declare an enum bit-field greater than itsprecision. Warn if an enum bit-field is too small to hold allits values.Mon Dec 5 11:41:50 1994 Mike Stump <mrs@cygnus.com>* typeck.c (convert_for_assignment): Use cp_convert instead ofconvert so that we don't get static casts.Sun Dec 4 11:59:01 1994 Mike Stump <mrs@cygnus.com>* cvt.c (cp_convert): Don't complain about int->enum conversion ifwe are doing static casts.Fri Dec 2 18:32:41 1994 Mike Stump <mrs@cygnus.com>* error.c (dump_expr): Do something more intelligent with SAVE_EXPRswhen dumping expressions in error messages.Fri Dec 2 17:04:27 1994 Mike Stump <mrs@cygnus.com>* gc.c (build_dynamic_cast): Change interface to libg++, ensure thatthe return type is the right type, and make references work.Fri Dec 2 16:36:43 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (poplevel): Don't be confused by function-scopedeclarations of non-nested functions.(duplicate_decls): Propagate DECL_MAIN_VARIANT.(pushdecl): Use duplicate_decls to copy info from old decl into newfunction-scope one rather than doing it here.* decl2.c (mark_inline_for_output): Deal with the DECL_MAIN_VARIANTof this decl, in case this is a function-scope declaration.* decl.c (finish_enum): Make sure that the type has the rightprecision when we call fixup_*_type.Tue Nov 29 19:12:07 1994 Jason Merrill <jason@phydeaux.cygnus.com>* cvt.c (build_up_reference): Strip superfluous NOP_EXPRs; we dowant to build up references to rvalues if possible.(cp_convert): Stick on a NOP_EXPR when converting to the same type.Tue Nov 29 11:28:59 1994 Mike Stump <mrs@cygnus.com>* parse.y (maybe_raises): Handle throw ().* parse.y (ansi_raise_identifier): Grok type-ids in exceptionspecifications.* tree.c (build_exception_variant): Use list compare to check iftwo exception specifications match.* decl.c (duplicate_decls, bad_specifiers): Enhance wording on errormessages.* call.c (build_method_call): Remove TREE_RAISES.* cvt.c (convert_to_aggr): Likewise.* typeck.c (build_function_call_real, convert_arguments): Likewise.* init.c (expand_aggr_init_1): Likewise.Tue Nov 29 09:50:39 1994 Mike Stump <mrs@cygnus.com>* except.c: Add support for m68k and mips exception handlingsupport.Tue Nov 29 08:48:33 1994 Mike Stump <mrs@cygnus.com>* except.c (expand_end_all_catch): Throw into outer context, if wefall off end of catch handlers.Mon Nov 28 16:44:41 1994 Mike Stump <mrs@cygnus.com>* Makefile.in: Make is easier to decide where parse.[ch] will bebuilt.Thu Nov 17 20:11:24 1994 Doug Evans <dje@cygnus.com>* cp/Make-lang.in (CXX_INSTALL_NAME): Use program_transform_name.(GXX_INSTALL_NAME): Likewise.(CXX_CROSS_NAME): Use program_transform_cross_name.(GXX_CROSS_NAME): Likewise.(c++.install-man): Use program_transform_name on g++.1.(c++.uninstall): Likewise.Mon Nov 28 13:53:03 1994 Mike Stump <mrs@cygnus.com>* parse.y (THROW): Fix precedence of throw expressions.Mon Nov 28 13:15:16 1994 Mike Stump <mrs@cygnus.com>* typeck.c (build_unary_op): Allow promotions from bool to int onunary ~.Sun Nov 27 00:16:21 1994 Jason Merrill <jason@phydeaux.cygnus.com>* method.c (build_overload_name): Use DECL_ASSEMBLER_NAME forclasses when appropriate.(build_overload_nested_name): When dealing with a function context,use ASM_FORMAT_PRIVATE_NAME to tweak the name of the function toavoid conflicts between local classes of the same name.Wed Nov 23 17:59:42 1994 Mike Stump <mrs@cygnus.com>* gxx.gperf, parse.y, lex.h, hash.h, lex.c (init_lex), delc.c(duplicate_decls, grokdeclarator), cp-tree.h: Add support for`explicit'.* cvt.c (convert_to_reference, cp_convert, build_type_conversion_1,build_type_conversion): Use LOOKUP_ONLYCONVERTING inbuild_method_calls so that non-converting constructors are not used.* call.c (build_method_call): If we shouldn't use a non-convertingconstructor, then don't.Wed Nov 23 14:46:56 1994 Jason Merrill <jason@phydeaux.cygnus.com>* call.c (build_method_call): Don't try to synthesize methods yet.Tue Nov 22 12:45:21 1994 Jason Merrill <jason@phydeaux.cygnus.com>* pt.c (push_template_decls): Create CONST_DECLs for templateconstant parameters, not VAR_DECLs.Sat Nov 19 15:28:31 1994 Jim Wilson <wilson@chestnut.cygnus.com>* typeck.c (build_binary_op_nodefault): Can shorten shift only ifshift count is less than size in bits of arg0.Thu Nov 17 15:30:50 1994 Mike Stump <mrs@cygnus.com>* gxx.gperf, hash.h, lex.c (init_lex, real_yylex), parse.y: Add newANSI keywords and, and_eq, bitand, bitor, explicit, namespace, not,not_eq, or, or_eq, typename, using, xor, xor_eq to g++. Still needto add support for explicit, namespace, typename, and using, supportfor the rest is already in.Fri Nov 4 19:04:18 1994 Mike Stump <mrs@cygnus.com>* gc.c (get_bad_cast_node): New routine to support compile timethrows of bad_cast.* gc.c (build_dynamic_cast): Support throwing of bad_cast at compiletime.Fri Nov 4 11:12:00 1994 Mike Stump <mrs@cygnus.com>* except.c: Add hppa support.Fri Nov 4 10:50:50 1994 Mike Stump <mrs@cygnus.com>* except.c: Add rs6000 support.Thu Nov 3 14:24:23 1994 Mike Stump <mrs@cygnus.com>* except.c (do_unwind): Add i[34]86 support.Thu Nov 3 00:10:46 1994 Jason Merrill <jason@phydeaux.cygnus.com>* pt.c (do_pending_expansions): Unset TREE_PUBLIC on implicitinstantiations.Wed Nov 2 15:08:24 1994 Kung Hsu <kung@mexican.cygnus.com>* decl.c (finish_function): Emit types used in method parametersinto symbol table.Wed Nov 2 15:05:47 1994 Jason Merrill <jason@phydeaux.cygnus.com>* pt.c (process_template_parm): Allow pointer to member functiontemplate parameter types.(uses_template_parms): Handle pointer to member functionCONSTRUCTORs.* g++.c (main): Cast first argument of bzero to (char *).Pass -lstdc++ instead of -lg++ unless we are invoked as 'g++'.Mon Oct 31 14:50:48 1994 Kung Hsu <kung@mexican.cygnus.com>* gc.c (build_dynamic_cast): Rewrite to make it work.* class.c (finish_vtbls): Build more vtables if flag_rtti is on.* class.c (modify_all_direct_vtables): Likewise.* init.c (expand_direct_vtbls_init): Expand more vtables ifflag_rtti is on.* decl.c (init_type_desc): Add default return.Tue Oct 25 17:13:09 1994 Kung Hsu <kung@mexican.cygnus.com>* tree.c (debug_binfo): Get rid of the initial size entry ofvtable.* cp-tree.h: Change flag_dossier to flag rtti, define typedescriptor type nodes.* decl.c (init_type_desc): New function to initialize typedescriptor type nodes.* decl.c (record_builtin_type): Change flag_dossier to flag_rtti.* lex.c (init_lex): Likewise.* decl.c: Change variable flag_dossier to flag_rtti.* decl.c (duplicate_decls): Get rid initial size entry of vtable.* decl.c (hack_incomplete_structures): Take out assert 164.* search.c (get_abstract_virtuals_1): Likewise.* search.c (dfs_init_vbase_pointers): Change CLASSTYPE_DOSSIER toCLASSTYPE_RTTI.* parse.y: Likewise.* class.c (prepare_fresh_vtable): For virtual bases, get rightoffset.* class.c (add_virtual_function): Change flag_dossier toflag_rtti.* class.c (modify_one_vtable): Modify the right rtti entry.* class.c (override_one_vtable): Get rid of size entry.* class.c (finish_struct): Change flag_dossier to flag_rtti, andbuild extra vtables, build type descriptors for polymorphicclasses.* gc.c (build_headof): Make headof() works correctly with newrtti.* gc.c (build_typeid): Make this function work with new rtti.* gc.c (get_typeid): Make this function work with new rtti.* gc.c (build_bltn_desc): New function for new rtti.* gc.c (build_user_desc): Likewise.* gc.c (build_class_desc): Ditto.* gc.c (build_ptr_desc): Ditto.* gc.c (build_attr_desc): Ditto.* gc.c (build_func_desc): Ditto.* gc.c (build_ptmf_desc): Ditto.* gc.c (build_ptmd_desc): Ditto.* gc.c (build_t_desc): Ditto.* gc.c: Comment out old build_t_desc, build_i_desc, build_m_desc.Tue Oct 25 13:37:41 1994 Jason Merrill <jason@phydeaux.cygnus.com>* call.c (convert_harshness): Check for TREE_UNSIGNED differencesafter checking for integral conversions.Wed Nov 30 19:13:50 1994 Mike Stump <mrs@cygnus.com>* Version 2.6.3 released.Thu Nov 17 10:56:50 1994 Jason Merrill <jason@phydeaux.cygnus.com>* typeck2.c (build_m_component_ref): Check the basetype of themember pointer against the main variant of the object type.Mon Nov 14 14:21:52 1994 Jason Merrill <jason@phydeaux.cygnus.com>* cvt.c (convert_to_reference): Make sure that the original exprgets its type back when converting a reference.* method.c (build_overload_name): Clear numeric_outputed_need_bar here.(build_decl_overload): Instead of here.Tue Nov 8 17:11:24 1994 Jason Merrill <jason@phydeaux.cygnus.com>* cvt.c (cp_convert): Don't build a TARGET_EXPR if we're not in afunction.* typeck.c (convert_for_initialization): Handle initialization froma TARGET_EXPR.Sun Nov 6 01:34:24 1994 Jason Merrill <jason@phydeaux.cygnus.com>* pt.c (lookup_nested_type_by_name): Fix list-walking logic.(tsubst): When replacing a TEMPLATE_TYPE_PARM, propagateTYPE_READONLY and TYPE_VOLATILE from the argument.(unify): When unifying with a TEMPLATE_TYPE_PARM, remove cv-qualspresent in parm from arg.(type_unification): Strip REFERENCE_TYPE from the argument type.(unify): Don't strip REFERENCE_TYPE from the argument type.Sat Nov 5 22:42:15 1994 Greg McGary <gkm@magilla.cichlid.com>* pt.c (do_type_instantiation): Check to see if there's aIDENTIFIER_TEMPLATE on a class before callinginstantiate_member_templates().Sat Nov 12 06:35:42 1994 Mike Stump <mrs@cygnus.com>* Version 2.6.2 released.Thu Nov 3 18:48:19 1994 Paul Eggert <eggert@twinsun.com>* Makefile.in (spew.o, lex.o, pt.o):Depend on $(srcdir)/parse.h, not parse.h.Tue Nov 1 19:19:41 1994 Mike Stump <mrs@cygnus.com>* Version 2.6.1 released.Sun Oct 23 13:19:55 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl2.c: Declare flag_access_control.(struct lang_f_options): Add access-control.* expr.c (cplus_expand_expr, NEW_EXPR): Unset flag_access_controlfor the call to expand_aggr_init to copy the object out of thepcc_struct_return slot.* search.c (compute_access): if (!flag_access_control) returnaccess_public.Fri Oct 21 00:32:54 1994 Jason Merrill <jason@phydeaux.cygnus.com>* lex.c (cons_up_default_function): Don't try to defer methodsynthesis now.* decl.c (init_decl_processing): Use __pure_virtual for abort_fndeclinstead of abort, since the OSF/1 dynamic linker doesn't like to seerelocation entries for abort.* tree.c (array_type_nelts_total): Use sizetype, notinteger_type_node.(array_type_nelts_top): Likewise.Thu Oct 20 15:48:27 1994 Mike Stump <mrs@cygnus.com>* decl.c (grokdeclarator): Added handling for catch parameters(CATCHPARM).* except.c (expand_start_catch_block): Use the new CATCHPARM contextinstead of NORMAL.* except.c (expand_throw): Don't let convert_to_reference complainabout what we are doing.Thu Oct 20 12:55:24 1994 Jim Wilson <wilson@cygnus.com>* method.c (emit_thunk): Call instantiate_virtual_regs.Wed Oct 19 14:15:33 1994 Mike Stump <mrs@cygnus.com>* except.c (expand_exception_blocks): Make sure throw code doesn'tget put in function that won't be output.Mon Oct 17 18:03:15 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (init_decl_processing): Make alloca a builtin.Thu Oct 27 21:10:25 1994 Craig Burley <craig@burley>* g++.c (main): Only decrement "added" and set "library" toNULL when "library" != NULL (just like 940829 fix).Mon Oct 17 15:56:11 1994 Mike Stump <mrs@cygnus.com>* except.c (expand_start_catch_block): Make sure the false labelgets onto the permanent obstack, as it is used for the exceptiontable.Fri Oct 14 18:54:48 1994 Mike Stump <mrs@cygnus.com>* class.c (modify_one_vtable): Since the DECL_CONTEXT of fndecl canbe set just below, use current_fndecl instead.Fri Oct 14 15:12:22 1994 Jason Merrill <jason@phydeaux.cygnus.com>* init.c (expand_aggr_vbase_init_1): Don't call expand_aggr_init_1with LOOKUP_SPECULATIVELY.(expand_default_init): Abort if build_method_call returns NULL_TREE.* typeck.c (build_modify_expr): Don't just build a MODIFY_EXPR ifthe rhs is a TARGET_EXPR.* parse.y (left_curly): Anonymous types are not affected by #pragmainterface/implementation.* method.c (synthesize_method): Don't call setup_vtbl_ptr for thedefault constructor if it isn't needed.* lex.c (cons_up_default_function): Do synthesize methods foranonymous types if necessary.Thu Oct 13 17:44:55 1994 Jason Merrill <jason@phydeaux.cygnus.com>* method.c (build_decl_overload): Set numeric_outputed_need_bar to 0.Wed Oct 12 13:27:57 1994 Jason Merrill <jason@phydeaux.cygnus.com>* typeck.c (build_modify_expr): Understand how to copy an aggregate.* init.c (expand_default_init): Likewise. Also remove some of thecrufty code that assumes methods will not be synthesized properly.* lex.c (cons_up_default_function): If the containing type has noname, these functions should never need to be called, so justdeclare them.* lex.c (real_yylex): Use HOST_BITS_PER_WIDE_INT to determine thebitmask for lexing character constants.* call.c (build_method_call): Disable code that tries to do trickystuff with a default parameter that is a constructor call, butactually does other tricky stuff that breaks things.Wed Oct 12 16:14:01 1994 Benoit Belley <belley@cae.ca>* decl.c (finish_enum): Disable code which forces enums to be signed,since this conflicts with their use as bitfields. type_promotes_tohandles promotion of enums of underlying unsigned types to signedinteger types.Wed Oct 12 13:24:03 1994 Jason Merrill <jason@phydeaux.cygnus.com>* cvt.c (type_promotes_to): Also promote enums to long ifappropriate.* typeck.c (default_conversion): Don't expect type_promotes_to toreturn a main variant.Wed Oct 12 12:19:45 1994 Jason Merrill <jason@phydeaux.cygnus.com>* call.c (build_scoped_method_call): Don't lose side effects in theobject expression when calling a non-existent destructor.Fri Sep 2 19:05:21 1994 Rohan Lenard <rjl@iassf.easams.com.au>* call.c (build_scoped_method_call): Remove erroneous error messagewhen destructor call is written as a scoped call.Tue Oct 11 23:48:31 1994 Jason Merrill <jason@phydeaux.cygnus.com>* various: Cast pointer arguments to bzero and bcopy to char *.Tue Oct 11 19:34:32 1994 Mike Stump <mrs@cygnus.com>* class.c (get_derived_offset): Added a type parameter to limit howfar up the CLASSTYPE_VFIELD_PARENT chain we search.* class.c (modify_one_vtable, fixup_vtable_deltas): When forming theoffset to put into the vtable for the this parameter, make sure wedon't offset from a parent of the DECL_CONTEXT of the function.Tue Oct 11 16:10:52 1994 Jason Merrill <jason@phydeaux.cygnus.com>* pt.c (do_function_instantiation): Set DECL_EXTERNAL andTREE_STATIC when setting DECL_INTERFACE_KNOWN.(do_type_instantiation): Likewise.* lex.c (cons_up_default_function): Set DECL_INTERFACE_KNOWN,DECL_EXTERNAL and TREE_STATIC as appropriate.* decl2.c (finish_file): Also synthesize methods that don't haveDECL_EXTERNAL set. Set interface_unknown before doing so.* decl.c (start_function): If DECL_INTERFACE_KNOWN is set on thefunction decl, don't muck with TREE_PUBLIC and DECL_EXTERNAL.Mon Oct 10 00:56:53 1994 Jason Merrill <jason@phydeaux.cygnus.com>* lex.c (cons_up_default_function): Mark methods in a template classas template instances. Store the values of interface_unknown andinterface_only for do_pending_inlines.(do_pending_inlines): Use them.* decl2.c (finish_file): If we haven't seen a definition of afunction declared static, make the decl non-PUBLIC so compile_filecan give an error.Sun Oct 9 02:42:29 1994 Jason Merrill <jason@phydeaux.cygnus.com>* method.c (do_build_copy_constructor): Handle anonymous unions.(do_build_assign_ref): Likewise.(largest_union_member): Move from lex.c.Sat Oct 8 14:59:43 1994 Jason Merrill <jason@phydeaux.cygnus.com>Re-implement g++'s vague linkage independent of TREE_PUBLIC.* pt.c (instantiate_member_templates): Lose redundant-fexternal-templates handling.(tsubst): Set TREE_PUBLIC and DECL_EXTERNAL on new decls. Don't setTREE_STATIC or DECL_INTERFACE_KNOWN.(do_pending_expansions): Predicate on DECL_INTERFACE_KNOWN insteadof DECL_EXTERNAL for explicit instantiations.(do_function_instantiation): Do the new thing.(do_type_instantiation): Likewise.(instantiate_template): Deal with member templates defined in a .ccfile with -fexternal-templates.* except.c (expand_exception_blocks): Use DECL_LINKAGE_KNOWN todecide whether to stick builtin_throw here.* decl2.c (import_export_inline): Predicate on DECL_INTERFACE_KNOWNrather than TREE_PUBLIC. Generally fix rules.(finish_file): Use DECL_INITIAL to determine whether or not a methodhas been synthesized, rather than TREE_ASM_WRITTEN.* decl.c (warn_extern_redeclared_static): Use DECL_PUBLIC instead ofTREE_PUBLIC.(pushdecl): Likewise.(duplicate_decls): Likewise. Deal with DECL_DECLARED_STATIC andDECL_INTERFACE_KNOWN.(redeclaration_error_message): Fix checking for conflicting linkage.(define_function): Set DECL_INTERFACE_KNOWN.(grokfndecl): Function decls are PUBLIC until we are sure abouttheir linkage. Set DECL_DECLARED_STATIC as needed.(start_function): Deal with linkage. Move pushdecl after linkagemagic.(finish_function): Don't set TREE_ASM_WRITTEN on discarded inlines.* cp-tree.h (lang_decl_flags): Add interface_known anddeclared_static.(DECL_INTERFACE_KNOWN): New macro.(DECL_DECLARED_STATIC): New macro.(DECL_PUBLIC): New macro.Clean up bogus use of TREE_PUBLIC.* class.c (alter_access): Fix mistaken use of TREE_PUBLIC (itdoesn't correspond to TREE_PROTECTED and TREE_PRIVATE).* init.c (do_friend): Don't arbitrarily set TREE_PUBLIC.Wed Oct 5 13:44:41 1994 Jason Merrill <jason@phydeaux.cygnus.com>* call.c (build_overload_call_real): Don't immediately doarray->pointer conversion.* pt.c (type_unification): If not passing to a reference, stripcv-quals. Also handle array->pointer conversion.Tue Oct 4 17:45:37 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (grokdeclarator): Don't warn about applying const to aconst typedef or template type parameter.* decl2.c (finish_file): Also synthesize methods after walking thevtables. Ugly ugly ugly.Mon Oct 3 15:02:41 1994 Jason Merrill <jason@phydeaux.cygnus.com>* various: Remove lingering remnants of old exception handling code.* decl2.c (finish_file): Synthesize methods before walking thevtables, so that the vtables get emitted as needed.* decl.c (shadow_tag): Remove obsolete code for pushing tags anddealing with exceptions.Mon Oct 3 13:05:27 1994 Ian Lance Taylor <ian@sanguine.cygnus.com>* Make-lang.in (g++-cross): Depend upon version.o and $(LIBDEPS).Mon Oct 3 02:59:28 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl2.c (finish_file): Fix inline handling.Sun Oct 2 00:21:56 1994 Jason Merrill <jason@phydeaux.cygnus.com>* decl.c (grokdeclarator): Handle redundant scope even better.({push,pop}_cp_function_context): Take toplev parameter.* method.c (synthesize_method): Pass toplev parameter to{push,pop}_cp_function_context depending on decl_function_context(fndecl).* typeck.c (build_x_unary_op): Unary & on OFFSET_REFs is always thebuilt-in version.* method.c (synthesize_method): Don't be confused by __in_chrgparameter.* class.c (popclass): Set C_C_D like start_function does.* decl.c (grokdeclarator): Handle redundant scope better.* parse.y (expr_or_declarator): Add '(' expr_or_declarator ')' rule.(direct_notype_declarator): Likewise.(complex_direct_notype_declarator): Remove it here.Sat Oct 1 21:42:18 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (resolve_offset_ref): Fix types used in resolving .*expressions.Sat Oct 1 15:18:49 1994 Jason Merrill <jason@phydeaux.cygnus.com>Beginnings of work to synthesize methods only when needed.* call.c (build_method_call): Synthesize methods as necessary(currently never necessary).* class.c (popclass): Don't try to set C_C_D here, as it'll end upon the wrong obstack.* decl.c (push_cp_function_context): Mostly copied frompush_c_function_context.(pop_cp_function_context): Similarly.(finish_function): Reverse order of poplevel and pop_nested_class sothat current_class_decl is restored properly.(start_function): Likewise.(finish_function): Add parameter 'nested'. Don't callpermanent_allocation if (nested).* various: Pass extra parameter to finish_function.* decl2.c (finish_file): Reorganize end-of-file inline handling,synthesizing methods as necessary.* lex.c (cons_up_default_function): Call mark_inline_for_output.Only synthesize methods immediately if #pragma implementation(currently disabled).(do_pending_inlines): Call synthesize_method.* method.c (synthesize_method): New function; all method synthesisgoes through here. Calls do_build_assign_ref anddo_build_copy_constructor.(build_default_constructor): Remove.(build_dtor): Likewise.(build_assign_ref): Rename to do_build_assign_ref and remove stuffdone by synthesize_method.(build_copy_constructor): Similarly.Thu Sep 29 16:58:52 1994 Mike Stump <mrs@cygnus.com>* typeck.c (c_expand_return): Use magic so the backend can fixup theassignment into the return register, so cleanups won't clobber it.Thu Sep 29 13:08:50 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (hack_identifier): Don't call assemble_external fortemplate decls.* decl.c (finish_decl): Also end temporary allocation if the decl inquestion has a type of error_mark_node.Wed Sep 28 21:45:00 1994 Mike Stump <mrs@cygnus.com>* typeck.c (build_modify_expr): When optimizing ?: on lhs, make surethat if the ?: was a reference type, that the subparts will be also.Wed Sep 28 16:14:04 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* except.c (register_exception_table): Use Pmode, not PTRmode.Fri Sep 23 13:54:27 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (do_pending_inlines): Do method synthesis after thepending_inlines have been reversed.Thu Sep 22 12:53:03 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl2.c (finish_file): Fix Brendan's fix: Only callregister_exception_table if there is a non-empty exception table.Thu Sep 22 12:03:46 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* decl2.c (finish_file): Only do register_exception_table if-fhandle-exceptions is being used.Wed Sep 21 19:01:51 1994 Per Bothner <bothner@kalessin.cygnus.com>* except.c (output_exception_table_entry): Simplifyby using assemble_integer.(build_exception_table): Change to return a count.Cleanup to use standard macros, instead of hard-wiredsparc asm format. Don't make __EXCEPTION_TABLE__ global.(register_exception_table): New function. Generate call to builtin.* decl2.c (finish_file): Call register_exception_table.* cp-tree.h (build_exception_table): Fix prototype.Wed Sep 21 13:20:42 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* tree.c (break_out_calls): Don't try to duplicate the DECL_INITIAL.* decl2.c (delete_sanity): Give an error at trying to delete afunction.Wed Sep 21 11:47:10 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (cons_up_default_function): Mark synthesized destructorsinline.* decl.c (duplicate_decls): Ignore redeclarations of wchar_t assomething other than __wchar_t, complaining if -pedantic and not ina system header.Tue Sep 20 09:43:28 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (xref_tag): Set up BINFO_INHERITANCE_CHAIN on base binfoshere.* typeck.c (build_modify_expr): Require complete type after checkingfor error_mark_node.* call.c (build_method_call): Print parmtypes when complaining ofambiguous call.* typeck.c (build_modify_expr): Handle assignment to array fromnon-array.* decl.c (lookup_name_real): Deal with got_scope == error_mark_node.* call.c (build_method_call): Don't bother with the exact match.Mon Sep 19 00:51:39 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (expand_aggr_init): If we munge the type of the variable,also munge the type of the initializer.* decl.c (grokdeclarator): Use <= when comparing to RID_LAST_MODIFIER.(init_decl_processing): Push artificial declaration of wchar_t sopeople don't have to declare it before they can use it.* error.c (cp_line_of): Return lineno in lieu of 0.* typeck.c (convert_for_assignment): Handle conversion of pmfs toint and bool.(build_component_ref): Fold the COMPONENT_REF in case it can bereduced.* typeck2.c (store_init_value): Don't pedwarn about non-constantbracketed initializers for automatic variables.Sun Sep 18 10:12:12 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_decl): Don't say `typedef enum foo foo'.* decl.c (start_decl): Don't set TREE_PUBLIC on template decls justbecause they're affected by #pragma i/i. We'll deal with that whenthey get instantiated.* typeck.c (build_unary_op): Clean up cruft in ADDR_EXPR case.* class.c (instantiate_type): Set TREE_CONSTANT on instantiatedADDR_EXPRs if appropriate.* decl.c (build_ptrmemfunc_type): Unset IS_AGGR_TYPE on pmf types.* typeck.c (build_ptrmemfunc): Handle &overloaded_method as aninitializer properly.* typeck2.c (digest_init): Likewise.* tree.c (cp_build_type_variant): Like c_build_type_variant, exceptit uses build_cplus_array_type.* *.c: Use cp_build_type_variant instead of c_build_type_variant.* pt.c (do_type_instantiation): Don't try to instantiate nestedenums.Tue Sep 13 10:56:58 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (build_up_reference): Handle preincrement and predecrementproperly.Tue Sep 13 09:51:59 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* decl.c (finish_decl): Only lay out the rtl for DECL if it is, infact, static.Mon Sep 12 14:40:30 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* decl.c (finish_decl): Lay out the rtl for DECL before doinggrok_reference_init, in case it's static.Mon Sep 12 12:45:38 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Don't synthesize constructors if theclass has a field with the same name as the class. Don't die onclasses with no constructors or destructors. Don't die if the headand tail of the class are in different files.* decl.c (grokdeclarator): Don't treat a function pointer fieldwith the same name as the class as a constructor.Fri Sep 9 13:17:00 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_c_cast): Pull constant values out of theirvariables here.* decl.c (duplicate_decls): Only propagate DECL_CHAIN inFUNCTION_DECLs and TEMPLATE_DECLs.Thu Sep 8 10:07:48 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (duplicate_decls): Propagate DECL_CHAIN in all DECLs thathave it.* pt.c (unify): REALs and INTEGERs only unify with their own genus.(instantiate_member_templates): Don't muck with DECL_EXTERNAL andTREE_PUBLIC unless -fexternal-templates.Wed Sep 7 13:17:10 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (do_type_instantiation): Call instantiate_member_templates.Deal with specializations.(tsubst): Don't stick the mangled name in DECL_NAME for functioninstantiations. Don't push them, either.* decl2.c (grokfield): Move code for generating theDECL_ASSEMBLER_NAME for static members from here.* method.c (build_static_name): To here.* decl.c (grokvardecl): Call build_static_name.(duplicate_decls): Keep old DECL_ASSEMBLER_NAME.Mon Sep 5 12:49:18 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): If -Wsynth, warn when selectingsynthesized op= over user-supplied one cfront would select.* decl2.c (lang_decode_option): Handle -Wsynth.Fri Sep 2 15:11:59 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (finish_enum): Overhaul to fix several bugs.(start_enum): Disable useless code.Thu Sep 1 16:04:54 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (c_expand_return): Warn about returning a reference to atemporary.(convert_arguments): Increment argument counter when using defaultarguments, too.Wed Aug 31 14:29:22 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (finish_decl): If the type of decl is error_mark_node,don't bother trying to do anything.* typeck.c (convert_for_initialization): If the rhs contains aconstructor call, pretend the lhs type needs to be constructed.* init.c (expand_default_init): If we stick the object inside theinitializer, mark the initializer used.Tue Aug 30 13:50:18 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (build_assign_ref): Return *this;(build_assign_ref): Fix base assignment order.(build_copy_constructor): Fix member init order.Mon Aug 29 13:54:39 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* g++.c (main): Remember to clear out SAW_SPECLANG after we seeits argument.Sat Aug 27 09:36:03 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (build_copy_constructor): Also copy virtual bases.Fri Aug 26 17:05:15 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (do_pending_inlines): Clear out pending_inlines before doingany synthesis. Also first set deja_vu on all pending_inlines.* method.c (build_assign_ref): Use build_member_call to invoke baseoperator=, rather than build_modify_expr. And usebuild_reference_type instead of TYPE_REFERENCE_TO.(build_copy_constructor): Use TYPE_NESTED_NAME to identify thebasetype.* decl2.c (grokfield): Don't complain about undefined local classmethods.* class.c (finish_struct): Don't try to synthesize methods here.* lex.c (do_pending_inlines): Instead, synthesize them here.(init_lex): Initialize synth_obstack.(cons_up_default_function): Stick synthesis request onpending_inlines.Fri Aug 26 12:24:14 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* call.c (build_method_call) [PCC_STATIC_STRUCT_RETURN]: Alsoaccept an RTL_EXPR in what we're about to use for the instance,since anything which would end up with pcc_struct_return setinside cplus_expand_expr.* cp-tree.h (cons_up_default_function): Note change of prototype.Thu Aug 25 23:05:30 1994 Gerald Baumgartner <gb@cs.purdue.edu>* class.c (finish_struct): Undid change from Aug 21 testingCLASSTYPE_INTERFACE and CLASSTYPE_VTABLE_NEEDS_WRITING.* parse.y (left_curly): Likewise, undid change from Aug 21.* decl.c (xref_tag): Undid change from Aug 21, setCLASSTYPE_INTERFACE correctly, and added comments.Thu Aug 25 00:36:31 1994 Jason Merrill <jason@deneb.cygnus.com>Rework approach to synthesized methods; don't go through the parseranymore.* class.c (finish_struct): Use new synthesis approach.* lex.c (cons_up_default_function): Now just creates declaration,not code.(largest_union_member): #if 0 out.(default_assign_ref_body): Likewise.(default_copy_constructor_body): Likewise.* method.c (build_default_constructor): New function to synthesize X().(build_copy_constructor): Synthesize X(X&).(build_assign_ref): Synthesize X::operator=(X&).(build_dtor): Synthesize ~X().* error.c (cp_line_of): If we're dealing with an artificialTYPE_DECL, look at the type instead.Wed Aug 24 11:11:50 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (sort_member_init): Check warn_reorder.* decl2.c (lang_decode_option): Handle -W{no-,}reorder.* cp-tree.h (CLASSTYPE_SOURCE_LINE): New macro.* error.c (cp_line_of): Use CLASSTYPE_SOURCE_LINE for aggregates.* class.c (finish_struct): Set CLASSTYPE_SOURCE_LINE.Tue Aug 23 09:28:35 1994 Mike Stump <mrs@cygnus.com>* error.c (dump_decl): Improve wording, so that error messagesdont't read template<, class foo>...Mon Aug 22 15:30:51 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* parse.y (label_colon): Also match a TYPENAME as a label name,since they may have declared a class by that name but have alsotried to have a local label under the same name.* pt.c (coerce_template_parms): Call cp_error, not cp_error_at,for the message so they know at what point it was instantiated.Sun Aug 21 23:07:35 1994 Gerald Baumgartner <gb@cs.purdue.edu>* class.c (finish_struct): Move setting of CLASSTYPE_INTERFACE andCLASSTYPE_VTABLE_NEEDS_WRITING for signatures up to left_curly time.* decl.c (xref_tag): Move setting of CLASSTYPE_INTERFACE andCLASSTYPE_VTABLE_NEEDS_WRITING for signatures down to left_curly time.* parse.y (left_curly): New final resting place for settingCLASSTYPE_INTERFACE and CLASSTYPE_VTABLE_NEEDS_WRITING for signatures.* class.c (finish_struct): Don't test for function/field nameconflicts in signatures, since all the fields are compiler-constructed.Fri Aug 19 14:04:47 1994 Kung Hsu <kung@mexican.cygnus.com>* method.c (build_overload_nested_name): In qualified namemangling, the template with value instantiation will have numericat end and may mixed with the name length of next nested level.Add a '_' in between.* method.c (build_overload_name): Ditto.* method.c (build_overload_identifier): Ditto.Thu Aug 18 16:24:43 1994 Mike Stump <mrs@cygnus.com>* error.c (dump_decl): Handle NULL args.Thu Sep 29 16:15:36 1994 Michael I Bushnell <mib@churchy.gnu.ai.mit.edu>* g++.c: Rework last change so it's done like collect.c (andgcc.c).Wed Sep 14 10:17:27 1994 Michael I Bushnell <mib@churchy.gnu.ai.mit.edu>* g++.c: Include <sys/errno.h> in case `errno' is a macroas permitted by ANSI C.Thu Aug 18 12:48:09 1994 Mike Stump <mrs@cygnus.com>* class.c (finish_struct): Move setting of CLASSTYPE_INTERFACE andCLASSTYPE_VTABLE_NEEDS_WRITING up to left_curly time.* decl.c (xref_tag): Move setting of CLASSTYPE_INTERFACE andCLASSTYPE_VTABLE_NEEDS_WRITING down to left_curly time.* parse.y (left_curly): New final resting place for settingCLASSTYPE_INTERFACE and CLASSTYPE_VTABLE_NEEDS_WRITING.Thu Aug 11 11:32:42 1994 H.J. Lu <hjl@nynexst.com>* g++.c (main): Only decrement "added" and set "library" toNULL when "library" != NULL.Sat Aug 13 00:14:52 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Don't set TREE_PUBLIC on a function decljust because its class has a known interface.(decls_match): Deal with new format of template parms.* lex.c (cons_up_default_function): Don't play with TREE_PUBLIC andDECL_EXTERNAL here.Fri Aug 12 01:55:15 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (pushtag): SET_DECL_ARTIFICIAL on gratuitous typedefs.(xref_defn_tag): Likewise.(pushdecl): Only allow artificial typedefs to be shadowed.* init.c (emit_base_init): Pass the right binfos toexpand_aggr_init_1.* class.c (delete_duplicate_fields_1): Make it work right.(finish_struct): Catch function/field name conflict.* decl2.c (check_classfn): Pass the function to cp_error, not justthe name.* init.c (sort_member_init): Warn when order of member initializersdoes not match order of member declarations.(emit_base_init): Call expand_aggr_init_1 with LOOKUP_PROTECT.* error.c (dump_expr): Handle lists of functions.* decl.c (start_function): #pragma interface only affects functionsthat would otherwise be static.(finish_decl): Don't warn about an unused variable if it has bothconstructor and destructor, since the 'resource allocation isinitialization' idiom is relatively common.* typeck.c (comp_target_types): Don't handle TEMPLATE_TYPE_PARMs.(comp_target_parms): Likewise.(compparms): Never consider default parms.(common_base_type): Don't choose a virtual baseclass if there is amore derived class in common.(build_conditional_expr): If pedantic, pedwarn about conversion tocommon base in conditional expr.* class.c (instantiate_type): Handle template instantiation better.* typeck.c (convert_arguments): Don't try to get tricky and convertto int directly when PROMOTE_PROTOTYPES is set, as it breaksuser-defined conversions.* lex.c (check_for_missing_semicolon): Also give error at end offile.* call.c (build_method_call): Don't promote arrays to pointers here.* typeck.c (convert_arguments): Don't require the actual parameterto be of a complete type if the formal parameter is a reference.Thu Aug 11 15:21:40 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Soften 'static' on member function errorto pedwarn.* init.c (build_new): Don't automatically save rval.(build_offset_ref): Do field lookup with proper basetype_path.Thu Aug 11 12:46:54 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* errfn.c (cp_silent): Declare to mark when we should avoidemitting warnings and errors.(cp_error): Check it.(cp_warning): Likewise.(cp_pedwarn): Likewise.(cp_compiler_error): Likewise.(cp_error_at): Likewise.(cp_warning_at): Likewise.(cp_pedwarn_at): Likewise.* call.c (compute_conversion_costs): Set CP_SILENT when we startout, and make sure we turn it off before we leave.Thu Aug 11 00:02:54 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (grok_array_decl): Try computing *(A+B) if neitherargument is obviously an array.Wed Aug 10 15:32:04 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (c_expand_start_case): Do cleanups here.* parse.y (xcond): Do bool conversion here, too.(simple_stmt, SWITCH case): Don't do cleanups here.* decl.c (duplicate_decls): Don't treat builtins that have beenexplicitly declared specially.Tue Aug 9 01:16:09 1994 Jason Merrill <jason@deneb.cygnus.com>* tree.c (make_deep_copy): Support copying pointer, reference,function, array, offset and method types.* decl.c (init_decl_processing): Mark exit and abort asBUILT_IN_NONANSI so that duplicate_decls is kinder aboutredeclaration.(duplicate_decls): Don't give two errors for redeclaring a Cfunction with the same parms but a different return type.* parse.y (paren_cond_or_null): Do cleanup and bool conversion here.(condition): Instead of here.(simple_stmt, SWITCH case): Also do cleanup here.* decl2.c (finish_anon_union): Only break out FIELD_DECLs.* call.c (build_method_call): Don't throw away the side effects ofthe object in a call to a non-existent constructor.* parse.y (primary): Likewise.* method.c (build_decl_overload): Oop.* decl2.c (lang_decode_option): Deal with flag_no_nonansi_builtin,warn about uselessness of specifying -fansi-overloading.* method.c (build_decl_overload): Treat any non-member new with oneparameter as __builtin_new.* decl.c (init_decl_processing): Setup built-in meanings of exit,_exit and abort.Mon Aug 8 15:03:30 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_readonly_or_volatile): Put a space between const andvolatile if both apply.* init.c (perform_member_init): Clean up after this initialization.(emit_base_init): Clean up after each base init, not after all havebeen done.(expand_aggr_vbase_init_1): Clean up after this init.Sun Aug 7 14:55:05 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Deal with destroying references.* parse.y (condition): Do bool_truthvalue_conversion here.(paren_expr_or_null): And here.(simple_if): Not here.(simple_stmt): Or here.Sat Aug 6 22:29:45 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (paren_expr_or_null): Wrap the expression in aCLEANUP_POINT_EXPR.(condition): Likewise.Sat Aug 6 19:46:37 1994 Rohan Lenard <rjl@easams.com.au>* call.c (build_scoped_method_call): Fix error message whendestructor call refers to a nonexistent type.Sat Apr 16 22:43:30 1993 Gerald Baumgartner <gb@cs.purdue.edu>* lex.h (rid): Deleted RID_RAISES, it's never used.Moved RID_PUBLIC, RID_PRIVATE, RID_PROTECTED, RID_EXCEPTION,RID_TEMPLATE and RID_SIGNATURE to the end of the enumeration,they don't need to be touched in `grokdeclarator.'(RID_LAST_MODIFIER): Defined macro to be RID_MUTABLE.* decl.c (grokdeclarator): Use RID_LAST_MODIFIER instead ofRID_MAX as loop limit for finding declaration specifiers.Sat Apr 3 21:59:07 1993 Gerald Baumgartner <gb@cs.purdue.edu>* lex.c (debug_yytranslate): Moved to parse.y since it needs toaccess `yytname,' which is static in parse.c.Fri Apr 2 23:36:57 1993 Gerald Baumgarnter <gb@cs.purdue.edu>* cp-tree.h (GNU_xref_ref): Fixed typo in extern declaration, itwas `GNU_xref_def' instead of `GNU_xref_ref.'Fri Aug 5 14:20:16 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (do_function_instantiation): Don't set TREE_PUBLIC andDECL_EXTERNAL on 'extern' instantiations; wait until EOF to do that.(do_type_instantiation): Likewise.* decl2.c (import_export_inline): Decides at EOF what an inline'slinkage should be.(finish_file): Call it.* decl.c (start_function): Don't rely on the settings of TREE_PUBLICand DECL_EXTERNAL from do_*_instantiation. Only setDECL_DEFER_OUTPUT on inlines whose linkage might actually change.(finish_function): Use DECL_DEFER_OUTPUT to decide which inlines tomark for later consideration, rather than DECL_FUNCTION_MEMBER_P.Fri Aug 5 01:12:20 1994 Mike Stump <mrs@cygnus.com>* class.c (get_class_offset_1, get_class_offset): New routine tofind the offset of the class where a virtual function is defined,from the complete type.* class.c (modify_one_vtable, fixup_vtable_deltas): Useget_class_offset instead of virtual_offset as get_class_offset willalways provide the right answer.* tree.c (virtual_offset): Remove. It only ever worked some of thetime.Tue Aug 2 12:44:21 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Put back unary_complex_lvalue callthat I thought was redundant.* typeck.c (c_expand_return): Fix a case I missed before.Sun Jul 31 17:54:02 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (unify): Strip cv-quals from template type arguments (when'const T*' is matched to 'const char*', that does not mean that T is'const char').Fri Jul 29 01:03:06 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (do_type_instantiation): Instantiate nested TAGS, nottypedefs. Third time's the charm?* parse.y (template_parm): Support default template parms.* pt.c (process_template_parm): Likewise.(end_template_parm_list): Likewise.(coerce_template_parms): Likewise.(mangle_class_name_for_template): Likewise.(push_template_decls): Likewise.(unify): Likewise.* method.c (build_overload_identifier): Likewise.* error.c (dump_decl): Likewise.Wed Jul 27 17:47:00 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (do_type_instantiation): Only instantiate nested *classes*.Tue Jul 26 13:22:40 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (note_debug_info_needed): Also emit debugging informationfor the types of fields.Mon Jul 25 00:34:44 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (lookup_template_class): Pass 'template' tocoerce_template_parms instead of 'in_decl', since it's a moremeaningful context.* typeck.c (c_expand_return): Make sure any cleanups for the returnexpression get run.(build_c_cast): Use CONVERT_EXPR for conversion to void.* pt.c (do_type_instantiation): Also instantiate nested types.* typeck.c (convert_for_assignment): Don't die when comparingpointers with different levels of indirection.* decl.c (grokdeclarator): The sub-call to grokdeclarator forclass-local typedefs sets DECL_ARGUMENTS, so we need to clear itout.* decl2.c (finish_anon_union): Don't die if the union has nomembers.* decl.c (grokdeclarator): Undo changes to declspecs when we're doneso that 'typedef int foo, bar;' will work.* decl2.c (finish_file): Don't call expand_aggr_init fornon-aggregates.Mon Jul 25 00:03:10 1994 Teemu Torma <tot@trema.fi>* decl.c (finish_function): We can't inline constructors anddestructors under some conditions with -fpic, but don't unsetDECL_INLINE.Mon Jul 25 00:03:10 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_object_ref): Make sure 'datum' is a valid object.Sun Jul 24 14:19:31 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Don't set DECL_FIELD_BITPOS onnon-fields.(finish_struct_methods): Use copy_assignment_arg_p.* cvt.c (cp_convert): If expr is an OFFSET_REF, resolve it insteadof giving an error.* typeck.c (build_binary_op_nodefault): Don't set result_type if wedon't know how to compare the operands.* decl.c (grokdeclarator): Avoid seg fault when someone uses '__op'as a declarator-id in their program. Like the Linux headers do.Arrgh.* tree.c (lvalue_p): Treat calls to functions returning objects byvalue as lvalues again.* typeck.c (build_component_addr): Use convert_force to convert thepointer in case the component type is also a private base class.* search.c (get_matching_virtual): Fix bogus warning of overloadedvirtual.* pt.c (overload_template_name): Set DECL_ARTIFICIAL on the createdTYPE_DECL to fix bogus shadowing warnings.Fri Jul 22 01:15:32 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (expand_aggr_init_1): const and volatile mismatches do notprevent a TARGET_EXPR from initializing an object directly.Tue Jul 19 17:55:37 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (build_up_reference): Allow building up references to`this', don't warn about making references to artificial variables(like `this').* tree.c (lvalue_p): `this' is not an lvalue.* call.c (build_method_call): Accept using a typedef name (ortemplate type parameter) for explicit destructor calls.Thu Jul 14 09:42:23 1994 Mike Stump <mrs@cygnus.com>* Version 2.6.0 released.Wed Jul 13 03:57:54 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (hack_identifier): Put back old code so lists ofnon-functions will be handled properly.* cp-tree.h (TYPE_NEEDS_CONSTRUCTING): #if 0 out; this macro is nowdefined in the language-independent tree.h.* tree.c (count_functions): Avoid bogus warning when compiling thisfunction.Mon Jul 11 18:37:20 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_reference_init): Always save the initializer of areference.Fri Jul 8 17:41:46 1994 Mike Stump <mrs@cygnus.com>* decl.c (cplus_expand_expr_stmt): Wrap statement expressions insideCLEANUP_POINT_EXPRs so that the stack slots can be reused.(disabled for now)Fri Jul 8 12:59:38 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (hack_identifier): Fix for new overloading.* typeck.c (build_binary_op_nodefault): Don't mess with division byzero.Fri Jul 8 13:20:28 1994 Gerald Baumgartner <gb@cs.purdue.edu>* decl2.c (finish_file): Only call walk_sigtables, ifflag_handle_signatures is turned on, don't waste time otherwise.Fri Jul 8 02:27:41 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (push_overloaded_decl): Don't create overloads of one whenshadowing a class type.* typeck.c (build_x_function_call): Complain about overloads of one.* decl.c (grokdeclarator): Don't try to treat a char* as a tree.(grokdeclarator): Fix setting of TREE_STATIC.(start_decl): Clear DECL_IN_AGGR_P after calling duplicate_decls.Thu Jul 7 22:20:46 1994 Gerald Baumgartner <gb@andros.cygnus.com>* cp-tree.h (walk_sigtables): Created extern declaration.* decl2.c (walk_sigtables): Created function, patterned afterwalk_vtables, even though we only need it to write out sigtables.(finish_sigtable_vardecl): Created function.(finish_vtable_vardecl): Changed 0 to NULL_PTR.(finish_file): Call walk_sigtables.* sig.c (build_signature_table_constructor): Mark class memberfunction pointed to from signature table entry as addressable.Thu Jul 7 13:39:37 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (start_decl): Check new decl of static member variableagainst the declaration in the class here.(grokvardecl): Instead of here.* class.c (prepare_fresh_vtable): Call import_export_vtable if not-fvtable-thunks.(build_vtable): Likewise.* decl2.c (import_export_vtable): Move logic for deciding theinterface of a template class from here.(import_export_template): To here.(finish_vtable_vardecl): Call import_export_template beforeimport_export_vtable.Wed Jul 6 20:25:48 1994 Mike Stump <mrs@cygnus.com>* except.c (init_exception_processing): Setup interim_eh_hook tocall lang_interim_eh.* except.c (do_unwind): Propagate throw object value acrossstack unwinding.* except.c (saved_throw_value): Used to hold the value of the objectbeing thrown. It is always a reference to the real value.* except.c (expand_start_catch_block): Add handling for thevalue of the exception object.* except.c (expand_start_catch_block): Add handler for the handler,so that throws inside the handler go to the outer block.* except.c (expand_end_catch_block): Likewise.* parse.y (handler_args): Use parm instead, as the other doesn't yethandle references correctly.Wed Jul 6 17:55:32 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl2.c (mark_vtable_entries): If -ftable-thunks, set thevtable entry properly to abort.Tue Jul 5 14:07:54 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_binary_op_nodefault): Downgrade division by zeroerrors to warnings.* call.c (build_overload_call_real): Handle fnname being a list offunctions.* typeck.c (build_x_function_call): Pass list of functions tobuild_overload_call, not just the name.* tree.c (count_functions): Complain when called for invalidargument.* decl.c (grokdeclarator): Fix settings of TREE_STATIC, TREE_PUBLICand DECL_EXTERNAL on static members and initialized const members.* decl2.c (grokfield): Reflect this change.Fri Jul 1 09:35:51 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (init): ANSI C++ does not forbid { }.Thu Jun 30 00:35:22 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (lang_decode_option): Set warn_nonvdtor along with -Wall.warn_nonvdtor defaults to off.* class.c (instantiate_type): Use comptypes rather than relying ontypes to satisfy ==.* decl.c (start_function): Set DECL_DEFER_OUTPUT on all inlines thatmight be static.* tree.c (build_cplus_new): Never build WITH_CLEANUP_EXPRs.* decl.c (grok_reference_init): Deal with ADDR_EXPRs of TARGET_EXPRs.* cvt.c (cp_convert): Pass 0 to with_cleanup_p arg ofbuild_cplus_new.Wed Jun 29 22:31:09 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (finish_file): Maybe consider static inlines multipletimes, in case they reference each other.Tue Jun 28 11:58:38 1994 Gerald Baumgartner <gb@cs.purdue.edu>* class.c (finish_struct): Don't `cons_up_default_function'sfor signatures.(finish_struct): Handle an empty method_vec correctly.* decl.c (grokdeclarator): Don't warn about a signature beingempty in a signature pointer declaration if we only saw aforward declaration of the signature. Changed `warning's into`cp_warning's.* sig.c (build_sigtable): Don't die if a null signature tableconstructor is returned.(build_signature_pointer_constructor): If the signature tableconstructor is null, the _sptr field is set to a null pointerand cast to the appropriate type. Make copies of all nullpointers so that the type null_pointer_node doesn't get changed.(build_signature_table_constructor): Added comments.* sig.c (build_signature_pointer_constructor): Complain if wetry to assign to/initialize a signature pointer/reference ofan undefined signature.Mon Jun 27 14:05:16 1994 Gerald Baumgartner <gb@cs.purdue.edu>* typeck2.c (store_init_value): Don't be pedantic aboutnon-constant initializers of signature tables/pointers/references.Fri Jun 24 16:49:41 1994 Gerald Baumgartner <gb@cs.purdue.edu>* decl.c (grokdeclarator): If we are grokking an opaque typedefin a signature, don't complain about it begin static.Wed Jun 29 16:44:45 1994 Mike Stump <mrs@cygnus.com>Fixes a problem of the this pointer being wrong in virtual calls tomethods that are not overridden in more derived classes.* class.c (fixup_vtable_delta): New routine. It will fixup thedelta entries in vtables, wheever they need updating.* class.c (finish_struct): Call the new routine for all virtualbases, as they can have different offsets, than those used in baseclasses that we derive our vtable from.Tue Jun 28 23:49:28 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_binary_op): Use the types before defaultconversions in the error message.* *.c: Use c_build_type_variant instead of build_type_variant wherethe type might be an array.* call.c (build_method_call): Call build_type_variant andbuild_reference_type in the right order.* decl.c (record_builtin_type): Likewise.Wed Jun 29 16:58:53 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Call build_type_variant andbuild_reference_type in the right order.* decl.c (record_builtin_type): Likewise.Tue Jun 28 23:49:28 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_binary_op): Use the types before defaultconversions in the error message.* *.c: Use c_build_type_variant instead of build_type_variant wherethe type might be an array.Sat Jun 25 11:50:54 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): Try UDC's before doing thereinterpret_cast thang, though.Fri Jun 24 01:24:01 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (c_expand_return): Don't USE the return value locationafter we've expanded the jump.* decl2.c (finish_file): Make sure DECL_SAVED_INSNS is not 0 beforetrying to write out an inline.* cvt.c (build_up_reference): Also do address adjustment when thetarget type uses MI.(convert_to_reference): Try UDCs only after built-in conversions.(build_type_conversion_1): Don't play games with the argument to themethod.(build_type_conversion): #if 0 out code for binding to reference.Thu Jun 23 00:22:28 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (finish_file): Use TREE_SYMBOL_REFERENCED to decidewhether to emit inlines.* decl.c (grokdeclarator): Set explicit_int for decls that justspecify, say, 'long'.* init.c (do_friend): Do overload C functions (or call pushdecl,anyaway).Wed Jun 22 13:40:49 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (build_up_reference): Don't call readonly_error.(convert_to_reference): Propagate const and volatile from expr toits type.* tree.c (lvalue_p): Random CALL_EXPRs are not lvalues.* cvt.c (build_up_reference): Break out WITH_CLEANUP_EXPR whencreating a temporary.(convert_to_reference): Lose excessive and incorrect trickiness.(cp_convert): Call build_cplus_new with with_cleanup_p set.* typeck2.c (build_functional_cast): Likewise.Tue Jun 21 17:38:38 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): signed, unsigned, long and short allimply 'int'.* decl.c (grokdeclarator): Allow "this is a type" syntax.(grok_reference_init): Simplify and fix.Sun Jun 19 17:08:48 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): pedwarn about a typedef that specifies notype.Sat Jun 18 04:16:50 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (start_function): Move TREE_PUBLIC and DECL_EXTERNALtinkering to after call to pushdecl.Fri Jun 17 14:48:28 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Handle destructors for non-aggregatetypes properly.Thu Jun 16 16:48:05 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Make sure that the name given for thedestructor matches the constructor_name of the instance.* pt.c (do_function_instantiation): A non-extern instantiationoverrides a later extern one.(do_type_instantiation): Likewise.Wed Jun 15 19:34:54 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (expand_aggr_init): Use TYPE_MAIN_VARIANT to get theunqualified array type.* cp-tree.h (EMPTY_CONSTRUCTOR_P): Tests whether NODE is aCONSTRUCTOR with no elements.* decl.c (various): Lose empty_init_node.(finish_decl): Use EMPTY_CONSTRUCTOR_P, do the empty CONSTRUCTORthing depending on the value of DECL_COMMON instead offlag_conserve_space, do the empty CONSTRUCTOR thing for types thatdon't have constructors, don't treat a real empty CONSTRUCTORspecially.* typeck2.c (process_init_constructor): Don't treat empty_init_nodespecially.Wed Jun 15 19:05:25 1994 Mike Stump <mrs@cygnus.com>* class.c (override_one_vtable): Don't forget to merge in an oldoverrider when we wanted to reuse a vtable, but couldn't.Wed Jun 15 15:03:16 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (start_decl): Put statics in common again.* decl.c (grokdeclarator): Return NULL_TREE for an error rather thansetting the type to error_mark_node.* typeck.c (build_modify_expr): Build up a COMPOUND_EXPR for enumbitfield assignments.Tue Jun 14 12:23:38 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_op_properties): Const objects can be passed by value.Mon Jun 13 03:10:59 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (import_export_vtable): Force implicit instantiations tobe interface_only when -fno-implicit-templates.* decl.c (duplicate_decls): Redeclaring a class template name is anerror.* pt.c (end_template_decl): Call GNU_xref_decl for class templates.* xref.c (GNU_xref_decl): Support templates.Sat Jun 11 17:09:05 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_op_properties): Split out checking for whether thisfunction should suppress the default assignment operator.* decl2.c (grok_function_init): Likewise.(copy_assignment_arg_p): New function to do just that.Now considers virtual assignment operators that take a base as anargument to count as copy assignment operators.* search.c (dfs_debug_mark): Lose checks for DWARF_DEBUG andTREE_ASM_WRITTEN, as they are redundant.* pt.c (end_template_decl): Don't try to set DECL_CLASS_CONTEXT on adecl that has no LANG_SPECIFIC part.(do_type_instantiation): Force the debugging information for thistype to be emitted.* decl.c (start_decl): Clear up uses of various types of templates(say sorry for static data members, rather than "invalid template").(expand_static_init): Fix initialization of static data members oftemplate classes.Fri Jun 10 00:41:19 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Set DECL_CONTEXT on static data members.* g++.c (main): Use -xc++-cpp-output for .i files.* pt.c (tsubst): Give meaningful error about declaring template fora copy constructor which was not declared in the class template.(do_type_instantiation): Explicit instantiation before the classtemplate is an error.(instantiate_template): Don't die if tsubst returns error_mark_node.Thu Jun 9 19:04:59 1994 Jason Merrill <jason@deneb.cygnus.com>Don't synthesize the copy assignment operator if the one in a baseclass is pure virtual.* cp-tree.h (TYPE_HAS_ABSTRACT_ASSIGN_REF): New macro to indicatewhether the type has a pure virtual copy assignment operator.* class.c (finish_base_struct): Don't generate the copy assignmentoperator if a base class has a pure virtual one.* decl.c (grok_op_properties): Add disabled code to setTYPE_HAS_ABSTRACT_ASSIGN_REF with comment pointing to where it isactually set.* decl2.c (grok_function_init): Set TYPE_HAS_ABSTRACT_ASSIGN_REF.* decl2.c (import_export_vtable): Always treat templateinstantiations as if write_virtuals >= 2, and treat implicitinstantiations as external if -fno-implicit-templates.(finish_file): Output all pending inlines ifflag_keep_inline_functions.Wed Jun 8 20:48:02 1994 Mike Stump <mrs@cygnus.com>* tree.c (layout_vbasetypes): Align virtual base classes insidecomplete objects, so that we don't core dump on machines such asSPARCs when we access members that require larger than normalalignments, such as a double. Also, we bump up the total alignmenton the complete type, as necessary.Wed Jun 8 16:18:14 1994 Jason Merrill <jason@deneb.cygnus.com>* gxxint.texi (Free Store): New section with code for examiningcookie.(Limitations of g++): Remove operator delete entry, since it is nolonger accurate. Fix access control entry.* typeck.c (build_unary_op): Pedwarn about taking the address of orincrementing a cast to non-reference type.(build_modify_expr): Use convert instead of convert_force again.* search.c (get_base_distance): Use IS_AGGR_TYPE_CODE to check forclass type, not == RECORD_TYPE.* decl.c (grokdeclarator): Cope with grokfndecl returning NULL_TREE.* typeck2.c (report_case_error): #if 0 out.* lex.c (real_yylex): Lose RANGE.* parse.y: Likewise.Tue Jun 7 18:17:35 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (simple_stmt, case ranges): Use ELLIPSIS instead of RANGE.Mon Jun 6 19:39:57 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_c_cast): Don't shortcut conversions to the sametype. Don't replace consts with their values here, since that's nowdone in cp_convert.* cvt.c (cp_convert): When converting to bool, takeinteger_zero_node to false_node and all other INTEGER_CSTs totrue_node.(build_type_conversion): Don't complain about multiple conversionsto float if we're not really converting.Fri Jun 3 02:10:56 1994 Jason Merrill <jason@deneb.cygnus.com>Implement 'extern template class A<int>;' syntax for suppressingspecific implicit instantiations.* cp-tree.h: Update prototypes for do_*_instantiation.* pt.c (do_pending_expansions): Don't compile 'extern' explicitinstantiations.(do_function_instantiation): Set DECL_EXTERNAL on 'extern' explicitinstantiations.(do_type_instantiation): Likewise.* parse.y (explicit_instantiation): Support 'extern template classA<int>;' syntax.* decl.c (start_function): Don't modify the settings of TREE_PUBLICand DECL_EXTERNAL on explicit instantiations.* cvt.c (cp_convert): Replace constants with their values beforeconverting.(cp_convert): Consistently use 'e' instead of 'expr'.Thu Jun 2 03:53:30 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck2.c (build_x_arrow): Resolve OFFSET_REFs first.Wed Jun 1 18:57:35 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck2.c (digest_init): Handle initializing a pmf with anoverloaded method.* typeck.c (build_ptrmemfunc): Handle overloaded methods.* decl.c (pushtag): Use build_decl to make TYPE_DECLs.(xref_defn_tag): Likewise.* pt.c (process_template_parm): Likewise.(lookup_template_class): Likewise.(push_template_decls): Likewise.(instantiate_class_template): Likewise.(create_nested_upt): Likewise.* class.c (finish_struct): Don't try to set DECL_CLASS_CONTEXT onTYPE_DECLs.* typeck.c (convert_arguments): Make sure type is not NULL beforechecking its TREE_CODE.Wed Jun 1 17:40:39 1994 Mike Stump <mrs@cygnus.com>* class.c (get_derived_offset): New routine.* class.c (finish_base_struct): Make sure we set BINFO_VTABLE andBINFO_VIRTUALS when we choose a new base class to inherit from.* class.c (modify_one_vtable): Use get_derived_offset to get theoffset to the most base class subobject that we derived this binfofrom.* class.c (finish_struct): Move code to calculate theDECL_FIELD_BITPOS of the vfield up, as we need might need it fornew calls to get_derived_offset in modify_one_vtable.Wed Jun 1 16:50:59 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_member_call): Use build_pointer_type instead ofTYPE_POINTER_TO.Wed Jun 1 11:11:15 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* decl.c (grokdeclarator): Make sure we have a DNAME set before wetry to use it in an error.Wed Jun 1 09:48:49 1994 Mike Stump <mrs@cygnus.com>* typeck.c (convert_arguments, convert_for_initialization): Don'tstrip NOP_EXPRs, when we are converting to a reference.Wed Jun 1 01:11:38 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_modify_expr): Don't dereference references wheninitializing them.* decl2.c (grokfield): Don't check for grokdeclarator returningerror_mark_node any more.* decl.c (grokfndecl): Return NULL_TREE instead of error_mark_node.(start_method): Return void_type_node instead of error_mark_node.* typeck.c (build_modify_expr): Resolve offset refs earlier.Tue May 31 16:06:58 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Resolve OFFSET_REFs in the object.* typeck.c (build_modify_expr): Dereference references before tryingto assign to them.* call.c (build_method_call): Don't confuse type conversionoperators with constructors.* typeck2.c (build_functional_cast): Just call build_c_cast if therewas only one parameter.* method.c (build_typename_overload): Don't setIDENTIFIER_GLOBAL_VALUE on these identifiers.* decl.c (grok_op_properties): Warn about defining a type conversionoperator that converts to a base class (or reference to it).* cvt.c (cp_convert): Don't try to use a type conversion operatorwhen converting to a base class.(build_type_conversion_1): Don't call constructor_name_full on anidentifier.* cp-tree.h (DERIVED_FROM_P): Should be self-explanatory.* decl.c (start_decl): Don't complain that error_mark_node is anincomplete type.(finish_decl): Check for type == error_mark_node.Mon May 30 23:38:55 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (start_function): Set DECL_DEFER_OUTPUT on implicitinstantiations and inline members.* spew.c (yylex): Set looking_for_template if the next token is a '<'.* lex.h: Declare looking_for_template.* decl.c (lookup_name_real): Use looking_for_template to arbitratebetween type and template interpretations of an identifier.Sat May 28 04:07:40 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (instantiate_template): Zero out p if we found aspecialization.* decl.c (grokdeclarator): Elucidate warning.(grokdeclarator): If pedantic AND -ansi, complain about long long.Make explicit instantiation work reasonably. It is now appropriateto deprecate the use of -fexternal-templates.* pt.c (instantiate_template): Set DECL_TEMPLATE_SPECIALIZATION orDECL_IMPLICIT_INSTANTIATION on fndecl as appropriate.(end_template_instantiation): Reflect changes in USE_TEMPLATEsemantics.(do_pending_expansions): if (!flag_implicit_templates) DECIDE(0);(do_function_instantiation): Don't set EXPLICIT_INST ifflag_external_templates is set. Do set TREE_PUBLIC and DECL_EXTERNappropriately otherwise.(do_type_instantiation): Set interface info for class. SetTREE_PUBLIC and DECL_EXTERN for methods. Do none of this ifflag_external_templates is set.* parse.y: Reflect changes in USE_TEMPLATE semantics.* decl2.c: New flag flag_implicit_templates determines whether ornot implicit instantiations get emitted. This flag currentlydefaults to true, and must be true for -fexternal-templates to work.(finish_file): Consider flag_implement_inlines whensetting DECL_EXTERNAL. Consider flag_implicit_templates whendeciding whether or not to emit a static copy.* decl.c (start_function): Set TREE_PUBLIC and DECL_EXTERNALproperly for template instantiations.(start_method): Set DECL_IMPLICIT_INSTANTIATION on methods of atemplate class.* cp-tree.h (CLASSTYPE_USE_TEMPLATE): Change semantics.(DECL_USE_TEMPLATE): Parallel macro for FUNCTION and VAR_DECLs.(various others): Accessor macros for the above.Fri May 27 13:57:40 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_binary_op_nodefault): Division by constant zero isan error.Fri May 27 13:50:15 1994 Mike Stump <mrs@cygnus.com>* class.c (override_one_vtable): Don't modify things we don't own.Fri May 27 01:42:58 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (finish_decl): Don't postpone processing the initializer ofa decl with DECL_EXTERNAL set, and do call rest_of_compilation for aPUBLIC const at toplevel.(grokdeclarator): pedwarn about initializing non-const ornon-integral statics in the class body.* decl.c (pushtag): Don't try to set DECL_CLASS_CONTEXT on aTYPE_DECL.* call.c (convert_harshness): Dereference reference on rhs beforeproceeding, properly grok passing const things to non-constreferences.* typeck.c (build_unary_op): Soften error about taking the addressof main() to a pedwarn.* lex.c (default_copy_constructor_body): Unambiguously specify baseclasses (i.e. A((const class ::A&)_ctor_arg) ).(default_assign_ref_body): Likewise.Thu May 26 13:13:55 1994 Gerald Baumgartner <gb@mexican.cygnus.com>* decl2.c (grokfield): Don't complain about local signaturemethod declaration without definition.* call.c (convert_harshness): If `type' is a signature pointerand `parmtype' is a pointer to a signature, just return 0. Wedon't really convert in this case; it's a result of making the`this' parameter of a signature method a signature pointer.* call.c (build_method_call): Distinguish calling the default copyconstructor of a signature pointer/reference from a signaturemember function call.Thu May 26 12:56:25 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (grokfield): Don't set TREE_PUBLIC on member functiondeclarations.* decl.c (duplicate_decls): A previous function declaration asstatic overrides a subsequent non-static definition.(grokdeclarator): Don't set TREE_PUBLIC on inline methoddeclarations.Wed May 25 14:36:38 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Handle initialization of static constmembers.(finish_decl): Likewise.* decl2.c (grokfield): Allow initialization of static const memberseven when pedantic.* decl2.c (grokfield): Deal with grokdeclarator returningerror_mark_node.* decl.c (grok_ctor_properties): Return 0 for A(A) constructor.(grokfndecl): Check the return value of grok_ctor_properties.(start_method): Likewise.* parse.y (absdcl): Expand type_quals inline.Tue May 24 19:10:32 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (pushtag): Use IS_AGGR_TYPE rather than checking for aRECORD_TYPE.Tue May 24 18:09:16 1994 Per Bothner <bothner@kalessin.cygnus.com>* cp-tree.h (VTABLE_NAME_FORMAT): If flag_vtable_thunks,always use "__vt_%s".* decl2.c (finish_vtable_vardecl): Don't consider abstract virtualswhen looking for a "sentinal" method (to decide on emitting vtables).* decl2.c (finish_file): Scan all decls for thunks that needto be emitted.* decl2.c (finish_vtable_vardecl): Don't bother calling emit_thunk.* method.c (make_thunk): Use a more meaningful label. If thereexists a matching top-level THUNK_DECL re-use it; otherwisecreate a new THUNK_DECL (and declare it).* method.c (emit_thunk): Make thunk external/public dependingon the underlying method.Tue May 24 00:22:04 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (tsubst): Use lookup_name_nonclass to find guiding decls, notlookup_name.* call.c (build_overload_call_real): Don't immediately pick afunction which matches perfectly.* decl.c (grokdeclarator): Use c_build_type_variant for arrays.(grokdeclarator): Warn about, and throw away, cv-quals attached to areference (like 'int &const j').* typeck.c (convert_arguments): Don't mess with i for methods.* call.c (build_method_call): Pass the function decl toconvert_arguments.* typeck.c (comp_ptr_ttypes_real): New function. Implements thechecking for which multi-level pointer conversions are allowed.(comp_target_types): Call it.(convert_for_assignment): Check const parity on the ultimate targettype, too. And make those warnings pedwarns.Mon May 23 14:11:24 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_char): Use TARGET_* for character constants.Mon May 23 13:03:03 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* tree.c (debug_no_list_hash): Make static.* decl.c (decls_match): Say the types don't match if newdecl ends upwith a null type, after we've checked if olddecl does.(pushdecl): Check if the decls themselves match before looking foran extern redeclared as static, to avoid inappropriate and incorrectwarnings.Fri May 20 14:04:34 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Make warning about duplicate short, etc.a pedwarn.* typeck.c (build_c_cast): Casting to function or method type is anerror.* class.c (finish_struct): Make warning for anonymous class with noinstances a pedwarn.* Makefile.in (stamp-parse): Expect a s/r conflict.* typeck.c (build_modify_expr): pedwarn about using a non-lvaluecast as an lvalue.Thu May 19 12:08:48 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (type_promotes_to): Make sure bool promotes to int ratherthan unsigned on platforms where sizeof(char)==sizeof(int).Wed May 18 14:27:06 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_c_cast): Tack on a NOP_EXPR when casting toanother variant.(build_modify_expr): Don't strip NOP_EXPRs, and don't get trickyand treat them as lvalues.* decl.c (shadow_tag): Do complain about forward declarations ofenums and empty declarations.* parse.y: Don't complain about forward declarations of enums andempty declarations.* typeck.c (convert_for_assignment): Complain about changingthe signedness of a pointer's target type.* parse.y (stmt): Move duplicated code for checking case values fromhere.* decl2.c (check_cp_case_value): To here. And add a call toconstant_expression_warning.* typeck.c (convert_for_assignment): Don't complain about assigninga negative value to bool.* decl.c (init_decl_processing): Make bool unsigned.* class.c (finish_struct): Allow bool bitfields.Wed May 18 12:35:27 1994 Ian Lance Taylor <ian@tweedledumb.cygnus.com>* Make-lang.in (c++.install-man): Get g++.1 from $(srcdir)/cp.Wed May 18 03:28:35 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (build_type_conversion): Lose special handling oftruthvalues.* search.c (dfs_pushdecls): Improve shadowing warning.Tue May 17 13:34:46 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_delete): Throw away const and volatile on `this'.* decl.c (finish_enum): Put the constants in TYPE_VALUES again,rather than the enumerators.(pushtag): s/cdecl/c_decl/gMon May 16 23:04:01 1994 Stephen R. van den Berg <berg@pool.informatik.rwth-aachen.de>* cp/typeck.c (common_type): Attribute merging.(comp_types): Utilize COMP_TYPE_ATTRIBUTES macro.* cp/parse.y: Revamp attribute parsing.Mon May 16 01:40:34 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (shadow_tag): Also check for inappropriate use of auto andregister.* method.c (build_overload_name): Clarify that the illegal case is apointer or reference to array of unknown bound.* error.c (dump_type_prefix): Print references to arrays properly.* typeck.c (various): Be more helpful in pointercomparison diagnostics.* tree.c (lvalue_p): MODIFY_EXPRs are lvalues again. Isn't thisfun?* parse.y: Also catch an error after valid stmts.* search.c (dfs_init_vbase_pointers): Don't abort because `this' isconst.* typeck.c (convert_for_initialization): If call toconvert_to_reference generated a diagnostic, print out the parmnumber and function decl if any.* errfn.c (cp_thing): Check atarg1 to determine whether or not we'respecifying a line, not atarg.* tree.c (build_cplus_method_type): Always make `this' const.* decl2.c (grokclassfn): If -fthis-is-variable and this function isa constructor or destructor, make `this' non-const.* typeck.c (build_modify_expr): Don't warn specially aboutassignment to `this' here anymore, since it will be caught by theusual machinery.* various: Disallow specific GNU extensions (variable-size arrays,etc.) when flag_ansi is set, not necessarily when pedantic is set,so that people can compile with -pedantic-errors for tighter constchecking and such without losing desirable extensions.* typeck2.c (build_functional_cast): Call build_method_call withLOOKUP_PROTECT.(process_init_constructor): Only process FIELD_DECLs.* decl.c (finish_decl): Also force static consts with no explicitinitializer that need constructing into the data segment.* init.c (build_delete): Undo last patch, as it interferes withautomatic cleanups.Sat May 14 01:59:31 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c, class.h, cp-tree.h, cvt.c, decl2.c: Lose old overloadingcode.* init.c (build_delete): pedwarn about using plain delete to deletean array.Fri May 13 16:45:07 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (comp_target_types): Be more helpful in contravariancewarnings, and make them pedwarns.* decl.c (grokdeclarator): Use decl_context to decide whether or notthis is an access declaration.* class.c (finish_struct_bits): Set TYPE_HAS_INT_CONVERSION if ithas a conversion to enum or bool, too.Fri May 13 16:31:27 1994 Mike Stump <mrs@cygnus.com>* method.c (emit_thunk): Make declaration forcurrent_call_is_indirect local (needed for hppa).Fri May 13 16:16:37 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (uses_template_parms): Grok BOOLEAN_TYPE.(tsubst): Likewise.Fri May 13 16:23:32 1994 Mike Stump <mrs@cygnus.com>* pt.c (tsubst): If there is already a function for this expansion,use it.* pt.c (instantiate_template): Likewise.Fri May 13 10:30:42 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* parse.y (implicitly_scoped_stmt, simple_stmt case): Usekept_level_p for MARK_ENDS argument to expand_end_bindings, to avoidgenerating debug info for unemitted symbols on some systems.* cp-tree.h (build_static_cast, build_reinterpret_cast,build_const_cast): Add declarations.Fri May 13 09:50:31 1994 Mike Stump <mrs@cygnus.com>* search.c (expand_indirect_vtbls_init): Fix breakage from Apr 27fix. We now try get_binfo, and if that doesn't find what we want,we go back to the old method, which still sometimes fails.Fri May 13 01:43:18 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (initdcl): Call cplus_decl_attributes on the rightvariable.* decl2.c (cplus_decl_attributes): Don't call decl_attributes forvoid_type_node.* typeck.c (build_binary_op_nodefault): Change result_type forcomparison ops to bool.(build_binary_op): Convert args of && and || to bool.* cvt.c (build_default_binary_type_conversion): Convert args of &&and || to bool.(build_default_unary_type_conversion): Convert arg of ! to bool.(type_promotes_to): bool promotes to int.Fri May 13 01:43:18 1994 Mike Stump <mrs@cygnus.com>Implement the new builtin `bool' type.* typeck.c (build_binary_op_nodefault): Convert args of && and || tobool.(build_unary_op): Convert arg of ! to bool.* parse.y: Know true and false. Use bool_truthvalue_conversion.* method.c (build_overload_value): Know bool.(build_overload_name): Likewise.* lex.c (init_lex): Set up RID_BOOL.* gxx.gperf: Add bool, true, false.* error.c (*): Know bool.* decl.c (init_decl_processing): Set up bool, true, false.* cvt.c (cp_convert): Handle conversion to bool.(build_type_conversion): Likewise.* *.c: Accept bool where integers and enums are accepted (useINTEGRAL_CODE_P macro).Thu May 12 19:13:54 1994 Richard Earnshaw <rwe11@cl.cam.ac.uk>* g++.c: Use #ifdef for __MSDOS__, not #if.Thu May 12 18:05:18 1994 Mike Stump <mrs@cygnus.com>* decl2.c (lang_f_options): Handle -fshort-temps. -fshort-tempsgives old behavior , and destroys temporaries earlier. Defaultbehavior now conforms to the ANSI working paper.Thu May 12 14:45:35 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_modify_expr): Understand MODIFY_EXPR as an lvalue.Use convert_force to convert the result of a recursive call when weare dealing with a NOP_EXPR. Don't automatically wrap MODIFY_EXPRsin COMPOUND_EXPRs any more.(various): Lose pedantic_lvalue_warning.(unary_complex_lvalue): Understand MODIFY_EXPR.* cvt.c (convert_to_reference): Allow DECL to be error_mark_node ifwe don't know what we're initializing.Wed May 11 01:59:36 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): Modify to use convtype parameter.Only create temporaries when initializing a reference, not whencasting.(cp_convert): New main function.(convert): Call cp_convert.* cvt.c, decl.c, typeck.c: Fix calls to convert_to_reference.* cp-tree.h (CONV_*): New constants used by conversion code forselecting conversions to perform.* tree.c (lvalue_p): MODIFY_EXPRs are no longer lvalues.* typeck.c (build_{static,reinterpret,const_cast): Stubs that justcall build_c_cast.* parse.y: Add {static,reinterpret,const}_cast.* gxx.gperf: Likewise.* typeck.c (common_type): Allow methods with basetypes of differentUPTs.(comptypes): Deal with UPTs.(build_modify_expr): Wrap all MODIFY_EXPRs in a COMPOUND_EXPR.* pt.c (end_template_decl): Check for multiple definitions of membertemplates.* call.c (build_method_call): Complain about calling an abstractvirtual from a constructor.* typeck.c (pointer_int_sum): Check for the integer operand being 0after checking the validity of the pointer operand.* typeck2.c (digest_init): Pedwarn about string initializer beingtoo long.Tue May 10 12:10:28 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (push_overloaded_decl): Only throw away a builtin if thedecl in question is the artificial one.* parse.y (simple_stmt, switch): Use implicitly_scoped_stmt becauseexpand_{start,end}_case cannot happen in the middle of a block.* cvt.c (build_type_conversion_1): Use convert again.Tue May 10 11:52:04 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* typeck2.c (digest_init): Make sure we check for signed andunsigned chars as well when warning about string initializers.* init.c (emit_base_init): Check if there's a DECL_NAME on themember before trying to do an initialization for it.Tue May 10 11:34:37 1994 Mike Stump <mrs@cygnus.com>* except.c: Don't do anything useful when cross compiling.Tue May 10 03:04:13 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (duplicate_decls): Fix up handling of builtins yet again.(push_overloaded_decl): Likewise.* cvt.c (convert): Don't look for void type conversion.Mon May 9 18:05:41 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (do_friend): Only do a pushdecl for friends, notpushdecl_top_level.Mon May 9 13:36:34 1994 Jim Wilson <wilson@sphagnum.cygnus.com>* decl.c (lookup_name_current_level): Put empty statement afterthe label OUT to make the code valid C.Mon May 9 12:20:57 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_binary_op_nodefault): Only complain aboutcomparing void * and a function pointer if void * is smaller.Sun May 8 01:29:13 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (lookup_name_current_level): Move through temporary bindinglevels.* parse.y (already_scoped_stmt): Revive.(simple_stmt): Use it again.* decl.c (poplevel): Always call poplevel recursively if we'redealing with a temporary binding level.Sat May 7 10:52:28 1994 Mike Stump <mrs@cygnus.com>* decl.c (finish_decl): Make sure we run cleanups for initial valuesof decls. Cures memory leak.* decl.c (expand_static_init): Likewise for static variables.* decl2.c (finish_file): Likewise for globals.Sat May 7 03:57:44 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (commonparms): Don't complain about redefining defaultargs.* decl.c (duplicate_decls): Don't complain twice about conflictingfunction decls.(decls_match): Don't look at default args.(redeclaration_error_message): Complain about redefining defaultargs.* call.c (build_overload_call_real): Also deal with guidingdeclarations coming BEFORE the template decl.* pt.c (unify): Allow different parms to have differentcv-qualifiers.(unify): Allow trivial conversions on non-template parms.Fri May 6 03:53:23 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (tsubst): Support OFFSET_TYPEs.(unify): Likewise.* decl2.c (finish_decl_parsing): Call push_nested_class with a type.* init.c (build_offset_ref): Fix error message.* search.c (lookup_field): Likewise.* call.c (build_scoped_method_call): Pass binfo tobuild_method_call.* typeck.c (build_object_ref): Likewise.* typeck2.c (binfo_or_else): Don't return a _TYPE.* class.c (finish_struct): Don't complain about re-use of inheritednames or shadowing of type decls.* decl.c (pushdecl_class_level): Likewise.* decl.c (finish_enum): Set the type of all the enums.* class.c (finish_struct): Don't get confused by access decls.* cp-tree.h (TYPE_MAIN_DECL): New macro to get the _DECL for a_TYPE. You can stop using TYPE_NAME for that now.* parse.y: Lose doing_explicit (check $0 instead).* gxx.gperf: 'template' now has a RID.* lex.h (rid): Likewise.* lex.c (init_lex): Set up the RID for 'template'.* parse.y (type_specifier_seq): typed_typespecs ornonempty_type_quals. Use it.(handler_args): Fix bogus syntax.(raise_identifier{,s}, optional_identifier): Lose.* except.c (expand_start_catch_block): Use grokdeclarator to parsethe catch variable.(init_exception_processing): The second argument to__throw_type_match is ptr_type_node.Fri May 6 07:18:54 1994 Chip Salzenberg <chip@fin.uucp>[ change propagated from c-decl.c of snapshot 940429 ]* cp/decl.c (finish_decl): Setting asmspec_tree should notzero out the old RTL.Fri May 6 01:25:38 1994 Mike Stump <mrs@cygnus.com>Add alpha exception handling support to the compiler.Quick and dirty backend in except.c.* cp/*: Remove most remnants of old exception handling support.* decl.c (finish_function): Call expand_exception_blocks to putthe exception hanlding blocks at the end of the function.* dec.c (hack_incomplete_structures): Make sure expand_decl_cleanupcomes after expand_decl_init.* except.c: Reimplementation.* expr.c (cplus_expand_expr): Handle THROW_EXPRs.* lex.c (init_lex): Always have catch, try and throw be reservedwords, so that we may always parse exception handling.* parse.y: Cleanup to support new interface into exception handling.* tree.def (THROW_EXPR): Add.Thu May 5 17:35:37 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (simple_stmt, for loops): Use implicitly_scoped_stmt.(various): Lose .kindof_pushlevel and partially_scoped_stmt.Thu May 5 16:17:27 1994 Kung Hsu <kung@mexican.cygnus.com>* parse.y (already_scoped_stmt): Move expand_end_binding() tofix the unmatched LBB/LBE in stabs.Thu May 5 14:36:17 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (set_nested_typename): Set TREE_MANGLED on the newidentifiers.(pushdecl): Check TREE_MANGLED.(xref_tag): Likewise.* cp-tree.h (TREE_MANGLED): This identifier is aDECL_NESTED_TYPENAME (named to allow for future use to denotemangled function names as well).Implement inconsistency checking specified in [class.scope0].* decl.c (lookup_name_real): Don't set ICV here after all.(finish_enum): Also set the type of the enumerators themselves.(build_enumerator): Put the CONST_DECL in the list instead of itsinitial value.(pushdecl_class_level): Check inconsistent use of a name in theclass body.* class.c (finish_struct): Check inconsistent use of a name in theclass body. Don't set DECL_CONTEXT on types here anymore.* parse.y (qualified_type_name): Note that the identifier has nowbeen used (as a type) in the class body.* lex.c (do_identifier): Note that the identifier has now been used(as a constant) in the class body.* error.c (dump_decl): Print type and enum decls better.Thu May 5 09:35:35 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* typeck.c (build_modify_expr): Warn about assignment to `this'.Wed May 4 15:55:49 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_delete): Use the global operator delete whenrequested.* decl.c (lookup_name_real): If we find the type we're looking in abase class while defining a class, set IDENTIFIER_CLASS_VALUE forthe type.* class.c (finish_struct): Remove a couple of dependencies onlanguage linkage.* decl.c (pushtag): Classes do nest in extern "C" blocks.(pushdecl): Only set DECL_NESTED_TYPENAME on the canonical one forthe type.(pushtag): Remove another dependency on the language linkage.* lex.c (cons_up_default_function): Don't set DECL_CLASS_CONTEXT toa const-qualified type.* decl.c (push_overloaded_decl): Throw away built-in decls here.(duplicate_decls): Instead of here.Wed May 4 15:27:40 1994 Per Bothner <bothner@kalessin.cygnus.com>* typeck.c (get_member_function_from_ptrfunc): Do The RightThing (I hope) if we're using thunks.Wed May 4 13:52:38 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (specialization): aggr template_type_name ';'.(named_class_head_sans_basetype): Use it.(explicit_instantiation): Likewise.(tmpl.2): Revert.* cvt.c (build_type_conversion_1): Use convert_for_initialization,rather than convert, to do conversions after the UDC.* cp-tree.h (SHARED_MEMBER_P): This member is shared between allinstances of the class.* search.c (lookup_field): If the entity found by two routes is thesame, it's not ambiguous.Wed May 4 12:10:00 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (lookup_name_real): Check for a NULL TREE_VALUE,to prevent the compiler from crashing ...Wed May 4 11:19:45 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): If we don't have an object, checkbasetype_path to figure out where to look up the function.* typeck.c (convert_for_initialization): Pass TYPE_BINFO (type) tobuild_method_call in case exp is NULL_TREE.Tue May 3 16:02:53 1994 Per Bothner <bothner@kalessin.cygnus.com>Give a vtable entries a unique named type, for the sake of gdb.* class.c (build_vtable_entry): The address of a thunk now hastype vtable_entry_type, not ptr_type_node.* method.c (make_thunk): Fix type of THUNK_DECL.* class.c (add_virtual_function, override_one_vtable): Usevfunc_ptr_type_node, instead of ptr_type_node.* cp-tree.h (vfunc_ptr_type_node): New macro.* decl.c (init_decl_processing): Make vtable_entry_typebe a unique type of pointer to a unique function type.Tue May 3 09:20:44 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (do_explicit): Sets doing_explicit to 1.(explicit_instantiation): Use do_explicit rather than TEMPLATEdirectly, add "do_explicit error" rule.(datadef): Set doing_explicit to 0 after an explicit instantiation.(tmpl.2): Don't instantiate if we see a ';' unless we're doing anexplicit instantiation.(named_class_head_sans_basetype): Remove aggr template_type_name';' again.Mon May 2 23:17:21 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (lookup_nested_tag): Lose.* decl2.c (grokfield): Set DECL_CONTEXT on TYPE_DECLs.(lookup_name_nonclass): Lose.* decl.c (poplevel_class): Add force parameter.(lookup_name_real): Fix handling of explicit scoping which specifiesa class currently being defined. Add 'nonclass' argument.(lookup_name, lookup_name_nonclass): Shells for lookup_name_real.* class.c (finish_struct): Don't unset IDENTIFIER_CLASS_VALUEs here.(popclass): Force clearing of IDENTIFIER_CLASS_VALUEs if we're beingcalled from finish_struct.Mon May 2 19:06:21 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (init_decl_processing), cp-tree.h: Removed memptr_type.(It seeems redundant, given build_ptrmemfunc_type.)* typeck.c (get_member_function_from_ptrfunc), gc.c (build_headof,build_classof): Use vtable_entry_type instead of memptr_type.* method.c (emit_thunk): Call poplevel with functionbody==0to prevent DECL_INITIAL being set to a BLOCK.Mon May 2 15:02:11 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (named_class_head_sans_basetype): Add "aggrtemplate_type_name ';'" rule for forward declaration ofspecializations.Mon May 2 15:02:11 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (instantiate_type): Deal with pmf's.* Make-lang.in (cc1plus): Don't depend on OBJS or BC_OBJS, sincestamp-objlist does.* Makefile.in (../cc1plus): Depend on OBJDEPS.(OBJDEPS): Dependency version of OBJS.Mon May 2 12:51:31 1994 Kung Hsu <kung@mexican.cygnus.com>* search.c (dfs_debug_mark): Unmark TYPE_DECL_SUPPRESS_DEBUG, notDECL_IGNORED_P.Fri Apr 29 12:29:56 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Clear out memory of local tags. Andtypedefs.* decl2.c (grokclassfn): Don't set DECL_CONTEXT to a cv-qualifiedtype.* search.c (get_matching_virtual): Be more helpful in error message.* *: Use DECL_ARTIFICIAL (renamed from DECL_SYNTHESIZED).* lex.c (default_assign_ref_body): Expect TYPE_NESTED_NAME to work.(default_copy_constructor_body): Likewise.* class.c (finish_struct): Don't gratuitously create multiple declsfor nested classes.Thu Apr 28 23:39:38 1994 Jason Merrill <jason@deneb.cygnus.com>Avoid clobbering the arg types of other functions when revertingstatic member functions.* decl.c (revert_static_member_fn): Rearrange arguments, don'trequire values for 'fn' and 'argtypes', add warning to commentabove.(decls_match): Rearrange arguments in call to rsmf.(grok_op_properties): Don't pass values for fn and argtypes.* pt.c (instantiate_template): Don't pass values for fn and argtypes.Thu Apr 28 16:29:11 1994 Doug Evans <dje@canuck.cygnus.com>* Make-lang.in (cc1plus): Depend on stamp-objlist.* Makefile.in (BC_OBJS): Delete.(OBJS): Cat ../stamp-objlist to get language independent files.Include ../c-common.o.(../cc1plus): Delete reference to BC_OBJS.Thu Apr 28 02:12:08 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (compute_access): No really, deal with static membersproperly. Would I lie to you?Implement lexical hiding of function declarations.* pt.c (tsubst): Use lookup_name to look for function decls to guideinstantiation.* method.c (build_opfncall): Use lookup_name_nonclass to look fornon-member functions.* init.c (do_friend): Use lookup_name_nonclass to look forfunctions.* error.c (ident_fndecl): Use lookup_name to look for functions.* decl2.c (lookup_name_nonclass): New function, skips overCLASS_VALUE.* decl.c (struct binding_level): Lose overloads_shadowed field.(poplevel): Don't deal with overloads_shadowed.(push_overloaded_decl): Do lexical hiding for functions.* class.c (instantiate_type): Don't check non-members if we havemembers with the same name.* call.c (build_method_call): Use lookup_name_nonclass instead ofIDENTIFIER_GLOBAL_VALUE to check for non-member functions.(build_overload_call_real): Likewise.* decl.c (duplicate_decls): Check for ambiguous overloads here.(push_overloaded_decl): Instead of here.* decl.c (pushdecl): Back out Chip's last change.* decl.c (grok_op_properties): Operators cannot be static members.* cp-tree.h (DECL_SYNTHESIZED): DECL_SOURCE_LINE == 0(SET_DECL_SYNTHESIZED): DECL_SOURCE_LINE = 0* lex.c (cons_up_default_function): Use SET_DECL_SYNTHESIZED.* method.c (do_inline_function_hair): Don't put friends of localclasses into global scope, either.* typeck2.c (build_functional_cast): Don't look for a function callinterpretation.Thu Apr 28 15:19:46 1994 Mike Stump <mrs@cygnus.com>* cp-tree.h: Disable use of backend EH.Wed Apr 27 21:01:24 1994 Doug Evans <dje@canuck.cygnus.com>* Make-lang.in (c++.distdir): mkdir tmp/cp first.* Makefile.in (INCLUDES): Move definition to same place asparent makefile.(ALLOCA): Define.(OLDAR_FLAGS): Delete.(OLDCC): Define.(DIR): Delete.(CLIB): Define.(####site): Delete.(SUBDIR_USE_ALLOCA): Don't use ALLOCA if compiling with gcc.Wed Apr 27 19:10:04 1994 Kung Hsu <kung@mexican.cygnus.com>* decl.c (xref_tag): Not to use strstr(), it's not available onall platforms.Wed Apr 27 18:10:12 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Resolve yet another class/pmf confusion.* call.c (build_overload_call_real): Don't take the single-functionshortcut if we're dealing with an overloaded operator.Wed Apr 27 17:35:37 1994 Mike Stump <mrs@cygnus.com>* search.c (get_base_distance): Search the virtual base classbinfos, incase someone wants to convert to a real virtual baseclass.* search.c (expand_indirect_vtbls_init): Use convert_pointer_to_realinstead of convert_pointer_to, as it now will work.Wed Apr 27 15:36:49 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): Don't complain about casting awayconst and volatile.* typeck.c (build_unary_op): References are too lvalues.Wed Apr 27 13:58:05 1994 Mike Stump <mrs@cygnus.com>* class.c (override_one_vtable): We have to prepare_fresh_vtablebefore we modify it, not after, also, we cannot reuse an old vtable,once we commit to a new vtable. Implement ambiguous overrides invirtual bases as abstract. Hack until we make the classill-formed.Wed Apr 27 01:17:08 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (unary_expr): Expand new_placement[opt] andnew_initializer[opt] inline.* search.c (lookup_fnfields): Don't throw away the inheritanceinformation here, either.(compute_access): Handle static members properly.* init.c (build_member_call): Always set basetype_path, and pass itto lookup_fnfields.* search.c (lookup_field): Deal properly with the case wherexbasetype is a chain of binfos; don't throw away the inheritanceinformation.(compute_access): protected_ok always starts out at 0.* init.c (resolve_offset_ref): Don't cast `this' to the base typeuntil we've got our basetype_path.* cp-tree.h (IS_OVERLOAD_TYPE): aggregate or enum.* cvt.c (build_up_reference): Use build_pointer_type rather thanTYPE_POINTER_TO.* call.c (convert_harshness_ansi): Call type_promotes_to for realsas well.* cvt.c (type_promotes_to): Retain const and volatile, addfloat->double promotion.* decl.c (grokdeclarator): Don't bash references to arrays intoreferences to pointers in function parms. Use type_promotes_to.Tue Apr 26 23:44:36 1994 Mike Stump <mrs@cygnus.com>Finish off Apr 19th work.* class.c (finish_struct_bits): Rename has_abstract_virtuals tomight_have_abstract_virtuals.* class.c (strictly_overrides, override_one_vtable,merge_overrides): New routines to handle virtual base overrides.* class.c (finish_struct): Call merge_overrides to handle overridesin virtual bases.Tue Apr 26 12:45:53 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_function_call): Call build_function_call_real withLOOKUP_NORMAL.* *: Don't deal with TYPE_EXPRs.* tree.c (lvalue_p): If the type of the expression is a reference,it's an lvalue.* cvt.c (convert_to_reference): Complain about passing constlvalues to non-const references.(convert_from_reference): Don't arbitrarily throw away const andvolatile on the target type.* parse.y: Simplify and fix rules for `new'.* decl.c (grok_op_properties): operator void is illegal.Mon Apr 25 02:36:28 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (components): Anonymous bitfields can still have declspecs.* decl.c (pushdecl): Postpone handling of function templates like wedo C functions.* search.c (expand_indirect_vtbls_init): Fix infinite loop whenconvert_pointer_to fails.* call.c (compute_conversion_costs_ansi): A user-defined conversionby itself is better than that UDC followed by standard conversions.Don't treat integers and reals specially.* cp-tree.h: Declare flag_ansi.* typeck.c (c_expand_return): pedwarn on return in void functioneven if the expression is of type void.(build_c_cast): Don't do as much checking for casts to void.(build_modify_expr): pedwarn about array assignment if this codewasn't generated by the compiler.* tree.c (lvalue_p): A comma expression is an lvalue if its secondoperand is.* typeck.c (default_conversion): Move code for promoting enums andints from here.* cvt.c (type_promotes_to): To here.* call.c (convert_harshness_ansi): Use type_promotes_to. Also fixpromotion semantics for reals.Sun Apr 24 16:52:51 1994 Doug Evans <dje@canuck.cygnus.com>* Make-lang.in (c++.install-common): Check for g++-cross.* Makefile.in: Remove Cygnus cruft.(config.status): Delete.(RTL_H): Define.(TREE_H): Use complete pathname, some native makes have minimalVPATH support.(*.o): Use complete pathname to headers in parent dir.(doc, info, dvi): Delete.Sun Apr 24 16:52:51 1994 Doug Evans <dje@canuck.cygnus.com>* Make-lang.in (c++.install-common): Check for g++-cross.* Makefile.in: Remove Cygnus cruft.(config.status): Delete.(RTL_H): Define.(TREE_H): Use complete pathname, some native makes have minimalVPATH support.(*.o): Use complete pathname to headers in parent dir.(doc, info, dvi): Delete.Sun Apr 24 00:47:49 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (pushdecl): Avoid redundant warning on redeclaring functionwith different return type.(decls_match): Compare return types strictly.Fri Apr 22 12:55:42 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (build_type_conversion): Do try to convert through otherpointers. This will fail if the class defines multiple pointerconversions.* error.c (dump_type_prefix): Print out pointers to arrays properly.(dump_type_suffix): Likewise. (was 'int *[]', now 'int (*)[]')* typeck.c (build_unary_op): Disallow ++/-- on pointers toincomplete type.* decl.c (duplicate_decls): Check mismatched TREE_CODES afterchecking for shadowing a builtin. If we're redeclaring a builtinfunction, bash the old decl to avoid an ambiguous overload.* cvt.c (convert_to_reference): Don't force arrays to decay here.* tree.c (lvalue_p): A MODIFY_EXPR is an lvalue.* decl.c (duplicate_decls): Don't assume that the decls will havetypes.Mon Apr 18 11:35:32 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940318 snapshot ]* c-decl.c (pushdecl): Warn if type mismatch with another external declin a global scope.Fri Apr 22 06:38:56 1994 Chip Salzenberg <chip@fin.uucp>* cp/typeck2.c (signature_error): Use cp_error for "%T".Mon Apr 18 11:59:59 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940415 snapshot ]* cp/decl.c (duplicate_decls, pushdecl, builtin_function):Use DECL_FUNCTION_CODE instead of DECL_SET_FUNCTION_CODE.Mon Apr 18 11:55:18 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940409 snapshot ]* cp/decl.c (duplicate_decls): Put new type in same obstack asold ones, or permanent if old ones in different obstacks.Mon Apr 18 11:48:49 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940401 snapshot ]* cp/parse.y (attrib): Handle string args as expressions,merging the two rules. `mode' attribute now takes a string arg.Delete the rule for an identifier as arg.Mon Apr 18 11:24:00 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940312 snapshot ]* cp/typeck.c (pointer_int_sum): Multiplication should be done signed.(pointer_diff): Likewise the division.Sun Mar 6 19:43:39 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940304 snapshot ]* cp/decl.c (finish_decl): Issue warning for large objects,if requested.Sat Feb 19 22:20:32 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940218 snapshot ]* cp/parse.y (attrib): Handle attribute ((section ("string"))).* cp/decl.c (duplicate_decls): Merge section name into new decl.Tue Feb 8 09:49:17 1994 Chip Salzenberg <chip@fin.uucp>[ cp/* changes propagated from c-* changes in 940206 snapshot ]* cp/typeck.c (signed_or_unsigned_type): Check for anyINTEGRAL_TYPE_P not just INTEGER_TYPE.Mon Dec 6 13:35:31 1993 Norbert Kiesel <norbert@i3.INformatik.rwth-aachen.DE>* cp/decl.c (finish_enum): Start from 0 when determining precisionfor short enums.Fri Dec 3 17:07:58 1993 Ralph Campbell <ralphc@pyramid.COM>* cp/parse.y (unary_expr): Look at $1 for tree_code rather thancasting $$.Wed Nov 17 19:22:09 1993 Chip Salzenberg <chip@fin.uucp>* cp/typeck.c (build_binary_op_nodefault): Propagate codefrom C front-end to optimize unsigned short division.(build_conditional_expr): Fix bug in "1 ? 42 : (void *) 8".Wed Nov 17 19:17:18 1993 Chip Salzenberg <chip@fin.uucp>* cp/call.c (convert_harshness_ansi): Given an (e.g.) charconstant, prefer 'const char &' to 'int'.Wed Feb 3 13:11:48 1993 Chip Salzenberg <chip@fin.uucp>* cp/class.c (finish_struct_methods): Handle multipleconstructors in fn_fields list.Fri Apr 22 12:48:10 1994 Kung Hsu <kung@mexican.cygnus.com>* class.c (finish_struct): Use TYPE_DECL_SUPPRESS_DEBUG to flagtypes not to be dumped in stabs, like types in #pragma interface.* decl.c (init_decl_processing): Use TYPE_DECL_SUPPRESS_DEBUG tomark unknown type.Fri Apr 22 03:27:26 1994 Doug Evans <dje@cygnus.com>* Language directory reorganization.See parent makefile.Thu Apr 21 18:27:57 1994 Per Bothner <bothner@kalessin.cygnus.com>* cp-tree.h (THUNK_DELTA): It is normally negative, souse signed .i variant of frame_size rather than unsigned .u.* cp-tree.h (VTABLE_NAME_FORMAT): If flag_vtable_thunks,use "VT" rather than "vt" due to binary incompatibility.* class.c (get_vtable_name): Use strlen of VTABLE_NAME_FORMAT,rather than sizeof, since it is now an expression.* class.c (modify_one_vtable): Modify to skip initial elementcontaining a count of the vtable.Thu Apr 21 00:09:02 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (check_newline): Force interface_unknown on main input file.* pt.c (do_pending_expansions): Always emit functions that have beenexplicitly instantiated.(do_function_instantiation): Set DECL_EXPLICITLY_INSTANTIATED.(do_type_instantiation): Set CLASSTYPE_VTABLE_NEEDS_WRITING andDECL_EXPLICITLY_INSTANTIATED on all my methods.* parse.y (explicit_instantiation): Call do_type_instantiation fortypes.* decl2.c (finish_vtable_vardecl): Call import_export_vtable.* decl.c (start_function): Don't set DECL_EXTERNAL on a functionthat has been explicitly instantiated.* cp-tree.h (DECL_EXPLICITLY_INSTANTIATED): Alias forDECL_LANG_FLAG_4.* class.c: Move import_export_vtable to decl2.c, and comment out alluses.Wed Apr 20 16:51:06 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (process_next_inline): Don't muck with DECL_INLINE.(do_pending_inlines): Likewise.Tue Apr 19 22:25:41 1994 Mike Stump <mrs@cygnus.com>Reimplement vtable building, and most vtable pointer setting.Allows for earier maintenance, easier understandability, and mostimportantly, correct semantics.* class.c (build_vtable): Removed unneededSET_BINFO_VTABLE_PATH_MARKED.* class.c (prepare_fresh_vtable): Likewise. Added argument.* class.c (modify_vtable_entry): General cleanup.* class.c (related_vslot, is_normal, modify_other_vtable_entries,modify_vtable_entries): Removed.* class.c (add_virtual_function): General cleanup.* class.c (finish_base_struct): Setup BINFO_VTABLE andBINFO_VIRTUALS as early as we can, so that modify_all_vtables canwork.* class.c (finish_vtbls): New routine, mostly fromunmark_finished_struct.* class.c (overrides): New routine.* class.c (modify_one_vtable): New routine, mostly frommodify_other_vtable_entries and modify_vtable_entries.* class.c (modify_all_direct_vtables, modify_all_indirect_vtables,modify_all_vtables): New routines.* class.c (finish_struct): Added arguemnt to prepare_fresh_vtablecall. General cleanup on how pending_hard_virtuals are handled.General cleanup on modifying vtables. Use finish_vtbls, instead ofunmark_finished_struct.* cp-tree.h (init_vtbl_ptrs, expand_direct_vtbls_init,get_first_matching_virtual, get_matching_virtual,expand_vbase_vtables_init, expand_indirect_vtbls_init): Update.* cvt.c (convert_pointer_to_real): Cleanup error message.* decl.c (grokfndecl): General cleanup.* decl.c (finish_function): Change init_vtbl_ptrs call toexpand_direct_vtbls_init. Change expand_vbase_vtables_init call toexpand_indirect_vtbls_init.* init.c (expand_virtual_init): Remove unneeded argument.* init.c (init_vtbl_ptrs): Rename to expand_direct_vtbls_init, addedtwo arguments to make more general. Made more general. Now can beused for vtable pointer initialization from virtual bases.* init.c (emit_base_init): Change expand_vbase_vtables_init call toexpand_indirect_vtbls_init. Change init_vtbl_ptrs call toexpand_direct_vtbls_init.* init.c (expand_virtual_init): General cleanup.* init.c (expand_default_init): Change expand_vbase_vtables_initcall to expand_indirect_vtbls_init.* init.c (expand_recursive_init_1): Change expand_vbase_vtables_initcall to expand_indirect_vtbls_init.* init.c (expand_recursive_init): Change expand_vbase_vtables_initcall to expand_indirect_vtbls_init.* search.c (get_first_matching_virtual): Rename toget_matching_virtual. General cleanup and remove setting ofDECL_CONTEXT. That is now done in a cleaner way inmodify_vtable_entry and add_virtual_function.* search.c (expand_vbase_vtables_init): Rename toexpand_indirect_vtbls_init. General cleanup. Useexpand_direct_vtbls_init to do hard work. Ensures that _all_ vtablepointers from virtual bases are set up.* search.c (bfs_unmark_finished_struct, unmark_finished_struct):Removed.* *.[chy]: Remove support for VTABLE_USES_MASK.Tue Apr 19 12:51:59 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): Use NOP_EXPRs to switch betweenreference and pointer types instead of bashing the types directly.* call.c (build_overload_call_real): Use the TREE_CODE to determinewhether the function is overloaded or not, rather thanTREE_OVERLOADED.* *: Remove all uses of TREE_OVERLOADED.* decl.c (grokdeclarator): Only complain about initializing constfields when -ansi or -pedantic.Tue Apr 19 12:42:42 1994 Doug Evans <dje@canuck.cygnus.com>* cp-tree.h (THUNK_DELTA): frame_size is now a union.Mon Apr 18 00:17:13 1994 Jason Merrill <jason@deneb.cygnus.com>Do overloading on a block-by-block basis, not function-by-function.* decl.c: Lose overloads_to_forget.(struct binding_level): Add overloads_shadowed field.(poplevel): Restore overloads_shadowed.(push_overloaded_decl): Use overloads_shadowed instead ofoverloads_to_forget.(finish_function): Don't look at overloads_to_forget.Copy enum_overflow logic from c-decl.c.* decl.c (start_enum): Initialize enum_overflow.(build_enumerator): Use enum_overflow. Also use current_scope().* search.c (current_scope): Move Brendan's comment frombuild_enumerator here.* typeck.c (convert_for_assignment): Change warnings to pedwarns fordiscarding const/volatile.Sat Apr 16 01:18:21 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (comp_target_parms): Accept TEMPLATE_TYPE_PARMs on the rhs.(comp_target_types): Likewise.* decl.c (lookup_name): Don't unset got_scope here.* spew.c (yylex): Only replace yylval with the TYPE_NESTED_NAME ifgot_scope != NULL_TREE.Fri Apr 15 16:36:33 1994 Jason Merrill <jason@deneb.cygnus.com>Horrible kludge to prevent templates from being instantiated bytheir base classes.* parse.y (template_instantiate_once): Unset TYPE_BEING_DEFINEDbefore we get to left_curly.* pt.c (instantiate_class_template): Set TYPE_BEING_DEFINED.* error.c (dump_decl): If it's a typedef, print out the name of thedecl, not just the underlying type.* decl.c (pushdecl): If the old duplicate decl was a TYPE_DECL,update the IDENTIFIER_TYPE_VALUE of its name.* decl2.c (finish_file): When processing the initializer for astatic member, pretend that the dummy function is a member of thesame class.Fri Apr 15 15:56:35 1994 Kung Hsu <kung@mexican.cygnus.com>* class.c (build_vtable_entry): Revert Apr 4 change.* decl2.c (mark_vtable_entries): Replace pure virtual functiondecl with abort's.Fri Apr 15 13:49:33 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_conditional_expr): Pedwarn on pointer/integermismatch, and don't pedwarn on 0/function pointer mismatch.* typeck2.c (digest_init): Lose code for special handling of unions.(process_init_constructor): Since they're handled just fine here.Pedwarn on excess elements.* decl2.c (grokfield): Complain about local class method declarationwithout definition.Fri Apr 15 13:19:40 1994 Per Bothner <bothner@kalessin.cygnus.com>* method.c (emit_thunk): Add extern declaration forcurrent_call_is_indirect (needed for hppa).Thu Apr 14 16:12:31 1994 Jason Merrill <jason@deneb.cygnus.com>Improve local class support; allow classes in different blocks tohave the same name.* decl.c (pushtag): Support local classes better.(pushdecl_nonclass_level): New function for pushing mangled decls ofnested types into the appropriate scope.(xref_defn_tag): Use pushdecl_nonclass_level instead ofpushdecl_top_level.(grokfndecl): Don't mess with IDENTIFIER_GLOBAL_VALUE for localclass methods.* method.c (do_inline_function_hair): Likewise.* class.c (finish_struct): It is legal for a class with noconstructors to have nonstatic const and reference members.Thu Apr 14 07:15:11 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* decl.c (push_overloaded_decl): Avoid giving errors aboutbuilt-ins, since duplicate_decls will have given warnings/errorsfor them.Thu Apr 14 03:45:12 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): Warn about casting pointer type toreference type when this is probably not what they wanted.Wed Apr 13 13:12:35 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (finish_decl): Don't mindlessly set TREE_USED forstatic consts any more (toplev.c has now been modified tonot emit warnings if they are unused).Wed Apr 13 00:22:35 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_op_properties): If op new/delete get here withMETHOD_TYPEs, do a revert_static_member_fn.* cp-tree.h (IDENTIFIER_CLASS_TYPE_VALUE): Lose.* init.c (is_aggr_typedef): Don't look atIDENTIFIER_CLASS_TYPE_VALUE.(get_aggr_from_typedef): Likewise.(get_type_value): Likewise.* call.c (build_scoped_method_call): Don't rely on overloadedtemplate names having IDENTIFIER_CLASS_VALUE set.* parse.y (component_decl_1, fn.def2): Revert rules forconstructors.(component_decl_1, fn.def2): Use $1 instead of $$, since $$ is beingclobbered.* decl.c (start_function): Only warn about `void main()' if pedantic|| warn_return_type.Tue Apr 12 02:14:17 1994 Jason Merrill <jason@deneb.cygnus.com>Clean up overloading of the template name.* class.c (pushclass): Overload the template name whenever pushinginto the scope of a template class, not just if it isuninstantiated.(popclass): Correspondingly.* search.c (push_class_decls): Don't overload_template_name.* pt.c (overload_template_name): Don't set IDENTIFIER_LOCAL_VALUE orDECL_CONTEXT on things.* parse.y (left_curly): Don't overload_template_name.* class.c (finish_struct): Don't undo_template_name_overload.* method.c (build_opfncall): Only pass one argument to global opdelete.* call.c (build_method_call): Use TYPE_VEC_DELETE_TAKES_SIZE todecide how many arguments to use for vec delete.* decl.c (grok_op_properties): Be consistent in modifyingcurrent_class_type.(grokdeclarator): Only complain about function decls with no returntype if we're being pedantic.Mon Apr 11 00:10:53 1994 Jason Merrill <jason@deneb.cygnus.com>Add support for operator new [] and operator delete [].* tree.def: Add VEC_NEW_EXPR and VEC_DELETE_EXPR.* ptree.c (print_lang_type): Indicate vec new/delete.* parse.y: Support vec new/delete.* method.c (build_decl_overload): Deal with vec new/delete.(build_opfncall): Likewise.* lex.c (init_lex): Set up values of ansi_opname and opname_tab forvec new/delete. vec new uses "__vn", and vec delete uses "__vd".* init.c (init_init_processing): Set up BIVN and BIVD.(do_friend): Don't clean up after mistaken setting of TREE_GETS_NEW,since it doesn't happen any more.(build_new): Support vec new. Always call something.(build_x_delete): Support vec delete.(build_vec_delete): Lose dtor_dummy argument, add use_global_delete,and pass it to build_x_delete.* decl2.c (delete_sanity): Don't change behavior by whether or notthe type has a destructor. Pass use_global_delete tobuild_vec_delete.(coerce_delete_type): Make sure that the type returned has a firstargument of ptr_type_node.* decl.c (init_decl_processing): Also declare the global vecnew/delete.(grokdeclarator): Also force vec new/delete to be static.(grok_op_properties): Note presence of vec new/delete, and play withtheir args. If vec delete takes the optional size_t argument, setTYPE_VEC_DELETE_TAKES_SIZE.* cp-tree.h (TYPE_GETS_{REG,VEC}_DELETE): New macros to simplifychecking for one delete or the other.(lang_type): gets_new and gets_delete are now two bits long. Thelow bit is for the non-array version. Lose gets_placed_new.(TYPE_VEC_DELETE_TAKES_SIZE): New macro indicating that the vecdelete defined by this class wants to know how much space it isdeleting.(TYPE_VEC_NEW_USES_COOKIE): New macro to indicate when vec new mustadd a header containing the number of elements in the vector; i.e.when the elements need to be destroyed or vec delete wants to knowthe size.* class.c (finish_struct_methods): Also check for overloading vecdelete.* call.c (build_method_call): Also delete second argument for vecdelete.* decl.c (grokdeclarator): Correct complaints again.(grokdeclarator): Fix segfault on null declarator.(decls_match): Also accept redeclaration with no arguments if bothdeclarations were in C context. Bash TREE_TYPE (newdecl) here.(duplicate_decls): Instead of here.* parse.y (nested_name_specifier_1): Lose rules for dealing withsyntax errors nicely, since they break parsing of 'const i;'.* decl.c (lookup_name): if (got_scope == current_class_type)val = IDENTIFIER_CLASS_VALUE (name).* search.c (lookup_nested_tag): Look in enclosing classes, too.* spew.c (yylex): Only look one character ahead when checking for aSCOPE.* lex.c (check_newline): Read first nonwhite char beforeincrementing lineno.* decl.c (grokdeclarator): Don't claim that typedefs are variablesin warning.* parse.y: Divide up uses of unqualified_id intonotype_unqualified_id and unqualified_id, so that TYPENAME can beused as an identifier after an object.* class.c (push_nested_class): Don't push into non-class scope.* decl.c (grokdeclarator): If an identifier could be a typeconversion operator, but has no associated type, it's not a typeconversion operator.* pt.c (unify): Check for equality of constants better.* decl.c (grokdeclarator): Don't complain about access decls.Sun Apr 10 02:39:55 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): pedwarn about data definitions withouttypes here.* parse.y (datadef): Don't pedwarn about decls without types here,since that is valid for functions.(fn.def2, component_decl): Support constructors with declmods again.(nomods_initdecls): For decls without any mods, so that we don't tryto get declspecs from some arbitrary $0.* search.c (lookup_field): Use cp_error.* parse.y (nested_name_specifier_1): Don't check aggr/non-aggr typehere; it breaks destructors for non-aggr types.* decl.c (lookup_name): Only look for TYPE_DECLs in base classes ofa type being defined, like the comment says.If got_scope is not an aggregate, just return NULL_TREE.* pt.c (create_nested_upt): Kung's code for creating types nestedwithin uninstantiated templates now lives here (it used to live inhack_more_ids). It needs to be expanded.* parse.y: Stop calling see_typename so much.* decl.c (lookup_name): Deal with TTPs and UPTs.* lex.c (real_yylex): Don't set looking_for_typename just because wesaw a 'new'.(dont_see_typename): #if 0 out.* spew.c (yylex): Increment looking_for_typename if the nextcharacter is SCOPE, rather than setting it to 1; this way, the valuefrom seeing an aggr specifier will not be lost. This kinda relieson looking_for_typename never being < 0, which is now true.* parse.y (nested_name_specifier_1): Accept TEMPLATE_TYPE_PARMs,too.(named_class_head_sans_basetype): Accept template types, too. Oops.Fri Apr 8 16:39:35 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (reparse_decl_as_expr1): Handle SCOPE_REFs.* parse.y: Lose START_DECLARATOR.* search.c (lookup_nested_tag): New function to scan CLASSTYPE_TAGSfor a class.* parse.y: Simplify fn.def2 and component_decl. Support 'enumA::foo' syntax. Catch invalid scopes better.* parse.y, lex.c: Lose TYPENAME_COLON.* decl2.c (groktypefield): #if 0 out.* decl.c (lookup_name): If the type denoted by got_scope iscurrently being defined, look in CLASSTYPE_TAGS rather than FIELDS.* class.c (push_nested_class): Don't try to push intoerror_mark_node.Fri Apr 8 07:26:36 1994 Brendan Kehoe <brendan@lisa.cygnus.com>* Makefile.in (stamp-parse): Update count of conflicts to 33.Thu Apr 7 17:47:53 1994 Jason Merrill <jason@deneb.cygnus.com>A saner implementation of nested types that treats template typesno differently from non-template types. There are still someshortcomings of our system; most notably, it is difficult to lookfor a nested type that is hidden by another name, because of the waywe keep track of hidden types. But this shouldn't be a problem forjust about anyone. Perhaps lookup_field should be fixed up a bit.* spew.c: Moved handling of nested types/scoping from the lexerinto the parser. Removed variable template_type_seen_before_scope.Removed functions frob_identifier, hack_more_ids, and various cruftthat was #if 0'd out in the past, reducing the size of the file from1146 lines to 450 lines. We can't quite do away with spew.c yet,though; we still need it for do_aggr () and checking for SCOPE afterthe current identifier. And setting lastiddecl.* parse.y: Moved handling of nested types/scoping from the lexerinto the parser, using a new global variable `got_scope'. Reducedthe number of states by 53. Implemented all uses of explicit globalscope. Removed terminals SCOPED_TYPENAME and SCOPED_NAME. Removednonterminals tmpl.1, scoped_base_class, id_scope, typename_scope,scoped_typename. Added nonterminals nested_type,qualified_type_name, complete_type_name, qualified_id, ptr_to_mem,nested_name_specifier, global_scope, overqualified_id, type_name.Changed many others. Added 9 new reduce/reduce conflicts, which arenested type parallels of 9 that were already in the grammar fornon-nested types. Eight of the now 33 conflicts should be removedin the process of resolving the late binding between variable andfunction decls.* gxxint.texi (Parser): Update.* cp-tree.h (IS_AGGR_TYPE_CODE): Add UNINSTANTIATED_P_TYPE.* lex.h: Add decl for got_scope.* lex.c (see_typename): Claim to be the lexer when callinglookup_name.* decl.c (lookup_name): When called from the lexer, look atgot_scope and looking_at_typename; otherwise don't.Thu Apr 7 22:05:47 1994 Mike Stump <mrs@cygnus.com>31th Cygnus<->FSF merge.Thu Apr 7 17:47:53 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (mark_vtable_entries): Call this to mark all theentries in the vtable addressable.(finish_decl_parsing): Handle SCOPE_REFs.* decl.c (decls_match): Always call compparms with strict == 1.Handle the special case of C function redecl here.(duplicate_decls): Only keep the old type if the new decl takes noarguments.* typeck.c (compparms): Also allow t1 to be ... if strict == 0.Thu Apr 7 16:17:50 1994 Mike Stump <mrs@cygnus.com>* class.c (build_vtable_entry): Fix breakage introduced Apr 517:48:41.Wed Apr 6 16:05:10 1994 Per Bothner <bothner@kalessin.cygnus.com>* init.c (build_virtual_init), search.c (build_vbase_vtables_init),ch-tree.h: Every place these functions were called, the result wasimmediately passed to expand_expr_stmt. Reduce redundancy bycalling expand_expr_init *inside* these functions. Thesemakes for a simpler interface, and we don't have to buildcompound expressions. Hence, rename these function to:expand_virtual_init and expand_vbase_vtables_init respectively.* init.c, decl.c: Change callers of these functions.* init.c, cp-tree.h (expand_virtual_init): Make static.* decl2.c (finish_file): Check TREE_PUBLIC||TREE_ADDRESSABLErather than DECL_SAVED_INSNS before emitting inlines.Wed Apr 6 13:06:39 1994 Jason Merrill <jason@deneb.cygnus.com>* spew.c (init_spew): #if 0 out stuff used by arbitrate_lookup.* decl.c (duplicate_decls): If this is a new declaration of anextern "C" function, keep the type (for the argtypes).(redeclaration_error_message): Don't check DECL_LANGUAGE here.(decls_match): Call compparms with a value of strict dependent onthe value of strict_prototypes for DECL_LANGUAGE (oldecl).* typeck.c (compparms): ... is only equivalent to non-promotingparms if we're not being strict.* parse.y (empty_parms): Don't check flag_ansi || pedantic here.* decl.c (init_decl_processing): if (flag_ansi || pedantic)strict_prototypes_lang_c = strict_prototypes_lang_cplusplus;* decl2.c (grok_function_init): Don't set DECL_INITIAL on purevirtuals.Tue Apr 5 17:48:41 1994 Per Bothner <bothner@kalessin.cygnus.com>Support for implementing vtables with thunks.* tree.def (THUNK_DECL): New TREE_CODE.* cp-tree.h (FNADDR_FROM_VTABLE_ENTRY), tree.c(fnaddr_from_vtable_entry): Handle flag_vtable_thunks case.* cp-tree.h (memptr_type): New variable.* class.c (build_vtable_entry): Build thunk if necessary.* class.c (build_vfn_ref): If using thunks, don't needto add delta field from vtable (there is none!).* decl.c: Add memptr_type as well as vtable_entry_type.If using thunks, the latter is just ptr_type_node.* gc.c, typeck.c: Use memptr_typeChange, not vtable_entry_type.* decl2.c (finish_vtable_vardecl): Handle thunks.* expr.c (cplus_expand_expr): Support THUNK_DECL.* decl.c (grokdeclarator): Set DECL_THIS_EXTERN if "extern".* decl.c (start_function): Set current_extern_inline based onDECL_THIS_EXTERN, not TREE_PUBLIC.* decl.c (finish_function): Call mark_inline_for_output if needed,Improve intelligence about when to emit inlines.* cp-tree.h (lang_decl_flags): New field saved_inline.* cp-tree.h (DECL_SAVED_INLINE): New macro.* class.c (add_virtual_function): Don't set TREE_ADDRESSABLE.* decl.h, decl.c (pending_addressable_inlines): Removed.* decl2.c (pending_addressable_inlines): Renamed to saved_inlines.* decl2.c (mark_inline_for_output): Do nothing ifDECL_SAVED_INLINE; otherwise set it (and add to saved_inlines list).* decl2.c (finish_vtable_vardecl): SET_CLASSTYPE_INTERFACE_KNOWNand set CLASSTYPE_INTERFACE_ONLY if there is a non-inline virtual.* decl2.c (finish_file): Writing out inlines later, so we canalso handle the ones needed for vtbales.* decl2.c (write_vtable_entries, finish_vtable_typedecl): Removed.* cp-tree.h, class.c, decl2.c, search.c: Remove -fvtable-hackand flag_vtable_hack. Use -fvtable-thunks and flag_vtable_thunksinstead. (The rationale is that these optimizations both break binarycompatibility, but should become the default in a future release.)Wed Apr 6 10:53:56 1994 Mike Stump <mrs@cygnus.com>* class.c (modify_vtable_entries): Never reset the DECL_CONTEXTof a fndecl, as we might not be from that vfield.Tue Apr 5 17:43:35 1994 Kung Hsu <kung@mexican.cygnus.com>* class.c (add_virtual_function): Fix bug for pure virtual, sothat DECL_VINDEX of the dummy decl copied won't be error.(see also Apr 4 change)Tue Apr 5 17:23:45 1994 Per Bothner <bothner@kalessin.cygnus.com>* typeck.c (c_expand_return): Before checking that we're notreturning the address of a local, make sure it's a VAR_DECL.(And don't worry about it being a TREE_LIST.)Tue Apr 5 13:26:42 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (YYDEBUG): Always define.* lex.c (YYDEBUG): Likewise.Mon Apr 4 11:28:17 1994 Kung Hsu <kung@mexican.cygnus.com>* class.c (finish_struct): Backup out the change below, put thenew change for the same purpose. The change below breaks code.* class.c (finish_struct): If pure virtual, copy node and makeRTL point to abort, then put in virtual table.* decl2.c (grok_function_iit): Reinstate Mar 31 change.Sat Apr 2 03:12:58 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_new): pedwarn about newing const and volatiletypes.* tree.c (get_identifier_list): Only do the special handlingthing if we're dealing with the main variant of the record type.* cvt.c (convert_to_reference): When converting betweencompatible reference types, use the pointer conversion machinery.Don't just blindly overwrite the old type.Fri Apr 1 17:14:42 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): When looking at global functions,be sure to use instance_ptr for the first argument, not some versionof it that has been cast to a base class. Also do this beforecomparing candidates.Thu Mar 31 19:50:35 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Constructors can be called forconst objects.Thu Mar 31 16:20:16 1994 Kung Hsu <kung@mexican.cygnus.com>* decl2.c (grok_func_init): Do not abort as rtl for pur virtualfunctions. They can be defined somewhere else.Sat Jan 23 23:23:26 1994 Stephen R. van den Berg <berg@pool.informatik.rwth-aachen.de>* decl.c (init_decl_processing): Declare __builtin_return_addressand __builtin_frame_address for C++ as well.Thu Mar 31 12:35:49 1994 Mike Stump <mrs@cygnus.com>* typeck2.c (store_init_value): Integral constant variables arealways constant, even when doing -fpic.Sat Jan 23 23:23:26 1994 Stephen R. van den Berg <berg@pool.informatik.rwth-aachen.de>* decl.c (redeclaration_error_message): Pass the types tocomptypes.Wed Mar 30 21:29:25 1994 Mike Stump <mrs@cygnus.com>Cures incorrect errors about pure virtuals in a class, when theyhave been overridden in a derived class.* search.c (get_abstract_virtuals): Reimplement.* search.c (get_abstract_virtuals_1): New routine.Wed Mar 30 14:10:04 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (push_template_decls): Make the pushed level pseudoglobal.* parse.y (extdefs): Don't pop everything if the current bindinglevel is pseudo_global.* decl.c (pop_everything): Stop on reaching a pseudo-globalbinding level.* cp-tree.h (DECL_FUNCTION_MEMBER_P): Change to more reliable test.* decl.c (duplicate_decls): Only copy DECL_SOURCE_{FILE_LINE} ifthe old decl actually had an initializer.* {various}: Clean up gcc -W complaints.* cp-tree.h (DECL_FUNCTION_MEMBER_P): Currently defined to be(DECL_CONTEXT (NODE) != NULL_TREE).* parse.y (lang_extdef): Call pop_everything if necessary.* decl.c (pop_everything): New function for popping bindinglevels left over after a syntax error.(pushdecl): Use DECL_FUNCTION_MEMBER_P to decide whether or nota function is a member.Wed Mar 30 14:20:50 1994 Mike Stump <mrs@cygnus.com>Cures calling a more base base class function, when a more derivedbase class member should be called in some MI situations.* search.c (make_binfo): Use more the more specialized basebinfos from the binfo given as the second argument to make_binfo,instead of the unspecialized ones from the TYPE_BINFO.* class.c (finish_base_struct): Likewise, update callers.* search.c (dfs_get_vbase_types): Likewise.* tree.c (propagate_binfo_offsets, layout_vbasetypes): Likewise.* decl.c (xref_tag): Use NULL_TREE instead of 0.* lex.c (make_lang_type): Likewise.Wed Mar 30 14:10:04 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (pushdecl): If pushing a C-linkage function, only do apush_overloaded_decl.(duplicate_decls): Standard overloading does not shadow built-ins.Tue Mar 29 00:54:18 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (end_template_decl): Don't call push_overloaded_decl.* init.c (do_friend): Don't call push_overloaded_decl.* decl.c (pushdecl): Call push_overloaded_decl for functions andfunction templates.(duplicate_decls): Functions and function templates are notduplicates, but don't complain about calling this function tocompare them.(push_overloaded_decl): Don't deal with linkage. Callduplicate_decls.(redeclaration_error_message): Deal with linkage.* decl.c (start_function): If push_overloaded_decl returns anolder version of the function, deal with it.* decl.c (start_function): Be sure only to push_overloaded_declfor non-members.* decl.c (grokfndecl): Put back clearing of DECL_CHAIN formethods.(start_function): Lose broken and redundant code for checking olddecl.* init.c (add_friend): Give line numbers of both friend declswhen warning about re-friending.* pt.c (tsubst): Use comptypes rather than == to compare thetypes of the method as declared and as defined, since defaultparameters may be different.* call.c (build_method_call): Use brendan's candidate printingroutine.* decl.c (start_method): Methods defined in the class body areinline whether or not it's a template class.Mon Mar 28 16:39:26 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (initdcl0): Add "extern" to current_declspecs ifhave_extern_spec && ! used_extern_spcec.* tree.c (really_overloaded_fn): A fn with more than oneoverload.* pt.c (end_template_decl): Use really_overloaded_fn.* decl.c (duplicate_decls): When smashing a decl into a previousdefinition, keep the old file and line.Don't deal with overloaded functions.Lose old code for checking arg types of functions.Check for overloaded C functions.(pushdecl): Deal with overloaded functions.(start_decl): Expect pushdecl to return an appropriate function decl.(start_function): Likewise.(push_overloaded_decl): Don't check for overloaded C functions.* *.c: Stop using DECL_OVERLOADED, it being archaic.TREE_OVERLOADED should probably go, too.Mon Mar 28 14:00:45 1994 Ron Guilmette <rfg@netcom.com>* typeck.c (comp_target_types): Call comp_target_parms withstrict == 1.Sun Mar 27 00:07:45 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (empty_parms): Don't parse () as (...) in extern "C"sections if we're compiling with -ansi or -pedantic.* decl.c (decls_match): Don't treat (int) and (int&) as matching.* decl2.c (grokfield): Don't pedwarn twice about initializingfield.* decl.c (push_overloaded_decl): Warn about shadowingconstructor.(redeclaration_error_message): Don't allow 'int a; int a;'* cvt.c (build_up_reference): Only check for valid upcast ifLOOKUP_PROTECT is set, not just any flag.Fri Mar 25 01:22:31 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (check_newline): When we see a #pragma implementation,also set it for the main input file.* init.c (build_new): Convert array size argument to size_t.* parse.y (primary): If we're doing a parenthesized type-id, callgroktypename before passing it to build_new.* call.c (build_method_call): Deal properly with const andvolatile for instances of reference type.* decl.c (store_return_init): Change 'if (pedantic) error' to 'if(pedantic) pedwarn'.* decl.c (grokdeclarator): Don't complain about putting `static'and `inline' on template function decls.Thu Mar 24 23:18:19 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Preserve const & volatile on`this'.Thu Mar 24 16:21:52 1994 Mike Stump <mrs@cygnus.com>* init.c (build_new, build_vec_delete): Use global new and deletefor arrays.* decl2.c (delete_sanity): Likewise.Thu Mar 24 02:10:46 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): If i is an lvalue,(int &)i -> *(int*)&i, as per 5.2.8p9 of the latest WP.(convert_force): Call convert_to_reference with LOOKUP_COMPLAIN.Wed Mar 23 17:45:37 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (duplicate_decls): Also propagate DECL_TEMPLATE_MEMBERSand DECL_TEMPLATE_INSTANTIATIONS.* init.c (build_new): Handle array typedefs properly.Wed Mar 23 18:23:33 1994 Mike Stump <mrs@cygnus.com>30th Cygnus<->FSF merge.Wed Mar 23 00:46:24 1994 Mike Stump <mrs@cygnus.com>* class.c (modify_vtable_entries): Avoid running off the end of thevirtuals list when processing a virtual destructor.* class.c (get_vtable_entry): Likewise.Wed Mar 23 00:23:59 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (duplicate_decls): If two template decls don't match,just return 0.Tue Mar 22 23:49:41 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (convert_for_assignment): Don't pedwarn aboutconverting function pointer to void *.Tue Mar 22 22:23:19 1994 Mike Stump <mrs@cygnus.com>Major revamp of pointer to member functions. Cures majornonfunctionality when used in casts, and MI situations.* cvt.c (convert_force): Update call site of build_ptrmemfunc.* typeck.c (convert_for_assignment): Likewise.* typeck2.c (digest_init): Likewise.* typeck2.c (process_init_constructor): Simplify by moving code intodigest_init.* typeck2.c (digest_init): Do default_conversions on init value, ifwe are processing pointer to member functions.* class.c (get_vfield_offset): Now non-static. Convert bit offsetinto byte offset.* cp-tree.h (get_vfield_offset): Likewise.* typeck.c (get_member_function_from_ptrfunc): Convert down to rightinstance, before fetching vtable pointer.* typeck.c (get_delta_difference): New routine.* typeck.c (build_ptrmemfunc): Revamp to handle casting better, alsoget vtable pointer out of right subobject.Tue Mar 22 17:56:48 1994 Mike Stump <mrs@cygnus.com>* search.c (get_binfo): Return NULL instead of aborting, whenpassed a UNION_TYPE.Tue Mar 22 12:44:54 1994 Jason Merrill <jason@deneb.cygnus.com>These patches implement handling of redefinition/redeclaration oftemplates.* typeck.c (comptypes): Simplify. All TEMPLATE_TYPE_PARMs areconsidered compatible.* parse.y (template_def): Pass defn argument to end_template_decl.* pt.c (end_template_decl): Add defn argument. Check forredefinition. Simplify.* error.c (OB_UNPUT): New macro, to remove mistakes.(aggr_variety): Subroutine of dump_aggr_type.* decl.c (decls_match): Support templates.(duplicate_decls): No longer static. Don't try to lay out templatedecls.(pushdecl): Simplify.* cp-tree.h (DECL_TEMPLATE_MEMBERS): Use DECL_SIZE instead ofDECL_INITIAL.Mon Mar 21 11:46:55 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_decl): Support class template decls.(dump_type): Don't adorn template type parms.* decl.c (duplicate_decls): Save DECL_TEMPLATE_INFO from old declif it was a definition.(redeclaration_error_message): Do the cp_error thang, and rejectredefinition of templates.Mon Mar 21 19:36:06 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (grokdeclarator): Set TREE_PUBLIC for METHOD_TYPEin FIELD context, when appropriate. Also,CLASSTYPE_INTERFACE_ONLY is irrelevant to setting TREE_PUBLIC.Also, simplify check for bogus return specifiers.Mon Mar 21 11:46:55 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (after_type_declarator1): Expand type_quals.(notype_declarator1): Likewise.(absdcl1): Likewise.Sat Mar 19 01:05:17 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Treat class-local typedefs like staticmembers; i.e. 'typedef int f();' means that f is a function type,not a method type.* parse.y (decl): Change direct_* back to *.(type_id): Change direct_abstract_declarator to absdcl.(direct_declarator, direct_initdecls, direct_initdcl0): Remove again.Fri Mar 18 12:47:59 1994 Jason Merrill <jason@deneb.cygnus.com>These two patches fix crashes on instantiating a template inside afunction with C linkage or containing labels.* class.c (current_lang_stacksize): No longer static.* decl.c (struct saved_scope): Add lang_base, lang_stack,lang_name, lang_stacksize, and named_labels.(push_to_top_level): Save them.(pop_from_top_level): Restore them.* gxxint.texi (Parser): Update.These two patches finish moving the task of expr/declaratorambiguity resolution from the lexer to the parser, and add one morer/r conflict. START_DECLARATOR can now be nuked.* parse.y (decl): Add "direct_" in typespec X rules.(direct_declarator): New nonterminal fordirect_after_type_declarator and direct_notype_declarator.(direct_initdecls): Like initdecls, but uses direct_initdcl0.(direct_initdcl0): Like initdcl0, but uses direct_declarator.(named_parm): Add typespec direct_declarator rule.* spew.c (yylex): #if 0 out START_DECLARATOR insertion.These two patches disable some excessive cleverness on the part ofg++; a non-class declaration always hides a class declaration in thesame scope, and g++ was trying to unhide it depending on theenclosing expression.* spew.c (arbitrate_lookup): #if 0 out.* decl.c (lookup_name): Never call arbitrate_lookup.* parse.y (complex_notype_declarator1): Add '*'complex_notype_declarator1 and '&' complex_notype_declarator1 rules.* parse.y (complex_direct_notype_declarator): Restore id_scopesee_typename TYPENAME rule, remove all other rules beginning withthose tokens.(notype_unqualified_id): Add '~' see_typename IDENTIFIER rule.Thu Mar 17 17:30:01 1994 Jason Merrill <jason@deneb.cygnus.com>These changes fix the compiler's handling of the functional cast/object declaration ambiguities in section 6.8 of the ARM. They alsoadd 11 reduce/reduce conflicts. Sigh.* parse.y: Add precedence decls for OPERATOR and '~'.(notype_unqualified_id): New nonterminal, encompasses all of theANSI unqualified-id nonterminal except TYPENAMEs.(expr_or_declarator): New nonterminal to delay parsing of code like`int (*a)'.(primary): Use notype_unqualified_id.(decl): Add typespec initdecls ';' and typespec declarator ';'rules.(initdcl0): Deal with the above.(complex_notype_declarator1): A notype_declarator that is not alsoan expr_or_declarator.(complex_direct_notype_declarator): A direct_notype_declarator thatdoesn't conflict with expr_or_declarator. Usenotype_unqualified_id. Remove id_scope see_typename TYPENAME rule.(functional_cast): New nonterminal, for the three functional castrules. So that they can be moved aftercomplex_direct_notype_declarator.(see_typename): Don't accept type_quals any more.* decl2.c (reparse_decl_as_expr): New function to deal with parsenodes for code like `int (*a)++;'.(reparse_decl_as_expr1): Recursive subroutine of the above.(finish_decl_parsing): New function to deal with parse nodes forcode like `int (*a);'. See the difference?Thu Mar 17 12:16:10 1994 Mike Stump <mrs@cygnus.com>These changes break binary compatibility in code with classesthat use virtual bases.* search.c (dfs_get_vbase_types): Simplify and correct to makesure virtual bases are initialized in dfs ordering.* search.c (get_vbase_types): Simplify and make readable.Thu Mar 17 12:01:10 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y: s/ typename / type_id /gWed Mar 16 17:42:52 1994 Kung Hsu <kung@mexican.cygnus.com>* parse.y (typespec): Add SCOPE TYPENAME for global scopedtype. e.g. ::B x.* decl.c (complete_array_type): Fix a bug that in -pendanticmode even there's no initializer, it will continue to builddefault index.Wed Mar 16 17:43:07 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (direct_notype_declarator): Add PTYPENAME rule, removeall of the scoped PTYPENAME rules.Wed Mar 16 16:39:02 1994 Mike Stump <mrs@cygnus.com>* init.c (build_offset_ref): The value of A::typedef_name isalways the TYPE_DECL, and never an error.Tue Mar 15 20:02:35 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (get_base_distance_recursive): Two binfos can onlyrepresent the same object if they are both via_virtual.* class.c (finish_base_struct): Check vbases for ambiguity, too.* search.c (get_vbase_types): Accept binfo argument, too.Tue Mar 15 19:22:05 1994 Kung Hsu <kung@mexican.cygnus.com>* decl.c (complete_array_type): Complete TYPE_DOMAIN of theinitializer also, because back-end requires it.Tue Mar 15 15:33:31 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_expr): Support member functions (which show up asOFFSET_REFs).Mon Mar 14 16:24:36 1994 Mike Stump <mrs@cygnus.com>* init.c (build_new): Set the return type of multidimensionalnews correctly.Fri Mar 11 15:35:39 1994 Kung Hsu <kung@mexican.cygnus.com>* call.c (build_method_call): If basetype not equal to typeof the instance, use the type of the instance in buildingdestructor.Thu Mar 10 17:07:10 1994 Kung Hsu <kung@mexican.cygnus.com>* parse.y (direct_notype_declarator): Add push_nested_type for'template_type SCOPED_NAME' rule.Tue Mar 8 00:19:58 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (parm): Add typed_declspec1 {absdcl, epsilon} rules.Sat Mar 5 04:47:48 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (regcast_or_absdcl): New nonterminal to implement latereduction of constructs like `int ((int)(int)(int))'.(cast_expr): Use it.(sub_cast_expr): Everything that can come after a cast.(typed_declspecs1): typed_declspecs that are not typed_typespecs.(direct_after_type_declarator): Lose PAREN_STAR_PAREN rule.(direct_abstract_declarator): Replace '(' parmlist ')' rule with'(' complex_parmlist ')' and regcast_or_absdcl.(parmlist): Split(complex_parmlist): Parmlists that are not also typenames.(parms_comma): Enabler.(named_parm): A parm that is not also a typename. Use declaratorrather than dont_see_typename abs_or_notype_decl. Expandtyped_declspecs inline.(abs_or_notype_decl): Lose.(dont_see_typename): Comment out.(bad_parm): Break out abs_or_notype_decl into two rules.Fri Mar 4 18:22:39 1994 Jason Merrill <jason@deneb.cygnus.com>* decl2.c (reparse_decl_as_casts): New function to change parsenodes for `(int)(int)(int)' from "function taking int and returningfunction taking int and returning function taking int" to "... castto int, cast to int, cast to int".* decl2.c (reparse_decl_as_expr): Recursive function to changeparse nodes for `A()()' from "function returning function returningA" to "A().operator()".* parse.y (primary): Replace `typespec LEFT_RIGHT' rule with`typespec fcast_or_absdcl' rule.(fcast_or_absdcl): New nonterminal to implement late reduction ofconstructs like `A()()()()'.(typename): Replace `typespec absdcl1' rule with`typespec direct_abstract_declarator' rule.(direct_abstract_declarator): Replace `LEFT_RIGHT type_quals' rulewith `fcast_or_absdcl type_quals' rule.Fri Mar 4 16:18:03 1994 Mike Stump <mrs@cygnus.com>* tree.c (lvalue_p): Improve OFFSET_REF handling, so that itmatches Section 5.5.Fri Mar 4 14:01:59 1994 Jason Merrill <jason@deneb.cygnus.com>* error.c (dump_type_prefix): Don't print basetype twice forpmfs.Fri Mar 4 13:24:33 1994 Mike Stump <mrs@cygnus.com>* typeck.c (convert_arguments): Handle setHandler(A::handlerFn)so that it is like setHandler(&A::handlerFn). Cures an `invalidlvalue in unary `&''.Fri Mar 4 11:15:59 1994 Jason Merrill <jason@deneb.cygnus.com>* gxxint.texi (Copying Objects): New section discussing defaultop= problems with virtual inheritance.* decl2.c (grokoptypename): Just does grokdeclarator andbuild_typename_overload, since the parser can't call grokdeclaratordirectly.* method.c (build_typename_overload): Set IDENTIFIER_GLOBAL_VALUEand TREE_TYPE on generated identifiers.* decl.c (grokdeclarator): Don't deal with TYPE_EXPRs anymore.* parse.y (parm): Convert `const char *' to `__opPCc' here.* error.c (dump_decl): Say sorry rather than my_friendly_abortingif we can't figure out what to do.(dump_type*): Likewise.* typeck2.c (build_m_component_ref): 'component' is an expr, nota decl. Also move the IS_AGGR_TYPE check after the stripping ofREFERENCE_TYPE.Fri Mar 4 04:46:05 1994 Mike Stump <mrs@cygnus.com>* call.c (build_method_call): Handle b->setHandler(A::handlerFn)so that it is like b->setHandler(&A::handlerFn). Cures an `invalidlvalue in unary `&''.Thu Mar 3 12:38:15 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y: Add precedence specification for START_DECLARATOR.(type_quals): Move before primary.(typename): Move before typed_declspecs, add 'typespec absdcl1' rule.* decl2.c (grokoptypename): Lose.* decl.c (grokdeclarator): Parse TYPE_EXPRs in the initial scan,rather than waiting until later.Wed Mar 2 14:12:23 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (unary_expr): Use 'typename' in 'new' rules, ratherthan expanding it inline.(typename): Expand empty option of (former) absdcl inline.(abs_or_notype_decl): Likewise.(absdcl): Lose empty rule.(conversion_declarator): New nonterminal for 'typename' of 'operatortypename'.(operator_name): Use it instead of absdcl.* parse.y: Add precedence declarations for SCOPED_TYPENAME,TYPEOF, and SIGOF.(typed_declspecs): Accept typed_typespecs, rather than typespecdirectly. Add rules with reserved_typespecquals.(reserved_declspecs): Don't accept typespecqual_reserved at thebeginning of the list. The typed_declspecs rule will deal with thisomission.(declmods): Accept nonempty_type_quals, rather than TYPE_QUALdirectly.* parse.y (direct_notype_declarator,direct_after_type_declarator, direct_abstract_declarator): Split upthe declarator1 nonterminals to match the draft standard and avoidambiguities.(new_type_id, new_declarator, direct_new_declarator,new_member_declarator): New nonterminals to implement the subset of'typename' allowed in new expressions.(unary_expr): Use new_type_id instead of typename.(after_type_declarator1, absdcl1): Fix semantics of member pointers.(abs_member_declarator, after_type_member_declarator): Lose.* parse.y (absdcl1): Don't require parens aroundabs_member_declarator.(abs_member_declarator): Lose see_typename from rules.(after_type_member_declarator): Likewise.* tree.c (get_identifier_list): New function, containing codepreviously duplicated in get_decl_list and list_hash_lookup_or_cons.(get_decl_list): Use it.(list_hash_lookup_or_cons): Likewise.* parse.y (typed_declspecs, declmods): It's not necessary to hashthe declspecs on class_obstack, so don't. This way typed_typespecscan reduce to typed_declspecs.Wed Mar 2 14:29:18 1994 Jason Merrill <jason@cygnus.com>* cvt.c (build_up_reference): If we aren't checking visibility,also allow base->derived conversions.Mon Feb 28 15:14:29 1994 Per Bothner <bothner@kalessin.cygnus.com>* typeck.c (build_c_cast): Remove bogus hack when convertingto a reference type.* cp-tree.h (lang_decl::vbase_init_list, DECL_VBASE_INIT_LIST):Removed, not used.(lang_stype::methods, lang_decl::next_method): New fields.(CLASSTYPE_METHODS, DECL_NEXT_METHOD): New macros.* decl.c (duplicate_decls): Preserve DECL_NEXT_METHOD.* cp-tree.h, decl2.c (flag_vtable_hack): New flag.* decl2.c (finish_vtable_vardecl): If flag_vtable_hack,and !CLASSTYPE_INTERFACE_KNOWN, try to use the presence ofa non-inline virtual function to control emitting of vtables.* class.c (finish_struct): Build CLASSTYPE_METHODS list.* search.c (build_vbase_vtables_init): Don't assemble_external(yet) if flag_vtable_hack.* class.c (build_vfn_ref): Likewise.Mon Feb 28 14:54:13 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (component_decl): Don't include "typed_declspecsdeclarator ';'" speedup, since it breaks enums.Fri Feb 25 15:43:44 1994 Per Bothner <bothner@kalessin.cygnus.com>* class.c (finish_struct): Minor optimization for buildingfn_fields list.Fri Feb 25 15:23:42 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (start_function): Fix detection of function overloading.Thu Feb 24 22:26:19 1994 Mike Stump <mrs@cygnus.com>* lex.c (check_newline): #pragma interface can take a stringargument, just like #pragma implementation. #pragma implementationchecks for garbage on the line, line #pragma interface does. Maininput files do not auto implement like named files, #pragmaimplementation must be used explicitly.Thu Feb 24 17:09:01 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y (components): Handle list of one again.(notype_components): Likewise.(after_type_declarator1): Take maybe_raises out again.* gxxint.texi (Parser): Document additional r/r conflict.Wed Feb 23 14:42:55 1994 Jason Merrill <jason@deneb.cygnus.com>* gxxint.texi (Parser): Add node.* Makefile.in (stamp-parse): Update expected conflict count.* parse.y (various): Replace "declmods declarator" with "declmodsnotype_declarator". The comment saying that "declmods declarator ';'"corresponds to "int i;" was wrong; it corresponds to "const i;".(component_decl): Add "typed_declspecs declarator ';'" rule; this*does* correspond to "int i;". Change "declmods components" to"declmods notype_components".(components): Don't deal with a list of one anymore.(notype_components): New nonterminal, corresponds to notype_declarator.({after_,no}type_component_decl{,0}): More new nonterminals.({after_,no}type_declarator): Fold in START_DECLARATOR token.Eliminates four reduce/reduce conflicts.(expr): Depend on nontrivial_exprlist instead of nonnull_exprlist.(nontrivial_exprlist): New nonterminal: A list of at least twoexpr_no_commas's.(nonnull_exprlist): Depend on nontrival_exprlist.Eliminates four reduce/reduce conflicts.(named_class_head): Move intermediate code block into separatenonterminal so that we can stick %prec EMPTY on it.Add more %prec EMPTY's to eliminate remaining shift/reduceconflicts.(after_type_declarator): Add maybe_raises to fndecl rules.(after_type_declarator_no_typename): Remove.For correctness.Document remaining reduce/reduce conflicts.Tue Feb 22 12:10:32 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (get_base_distance): Only bash BINFO_INHERITANCE_CHAIN(TYPE_BINFO (type)) if we care about the path.* tree.c (lvalue_p): A COND_EXPR is an lvalue if both of theoptions are.Mon Feb 21 19:59:40 1994 Mike Stump <mrs@cygnus.com>* Makefile.in (mostlyclean): lex.c is a source file, don'tremove.Sat Feb 19 01:27:14 1994 Jason Merrill <jason@deneb.cygnus.com>* parse.y: Eliminate 20 shift/reduce conflicts.Fri Feb 18 11:49:42 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (type_unification): Add subr argument; if set, it meansthat we are calling ourselves recursively, so a partial match is OK.(unify): Support pointers to methods and functions.(tsubst): Support method pointers.* decl.c (build_ptrmemfunc_type): No longer static, so thattsubst can get at it.* init.c (is_aggr_typedef): Pretend template type parms areaggregates.* decl2.c (build_push_scope): If cname refers to a template typeparm, just grin and nod.* call.c (build_overload_call_real): Pass subr argument totype_unification.* pt.c (do_function_instantiation): Likewise.* class.c (instantiate_type): Likewise.* search.c (get_base_distance): If BINFO is a binfo, use it anddon't mess with its BINFO_INHERITANCE_CHAIN.* cvt.c (convert_to_reference): Fix temporary generation.If ambiguous, return error_mark_node.* init.c (build_new): Put back some necessary code.Thu Feb 17 15:39:47 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_new): Deal with array types properly.* search.c (get_binfo): Become a shell for get_base_distance.(get_binfo_recursive): Lose.(get_base_distance_recursive): Find the path to the via_virtual basethat provides the most access.(get_base_distance): Likewise.* parse.y (explicit_instantiation): Syntax is 'template classA<int>', not 'template A<int>'.* typeck.c (convert_for_initialization): Remove bogus warning.* parse.y (datadef): Revert patch of Oct 27.Thu Feb 17 15:12:29 1994 Per Bothner <bothner@kalessin.cygnus.com>* class.c (build_vfn_ref): Cast delta field to ptrdiff_type_node,rather than integer_type_node. Does wonders for the Alpha.Thu Feb 17 13:36:21 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (build_ptrmemfunc_type): Make sure that the pmf typegoes onto the same obstack as its target type.Wed Feb 16 00:34:46 1994 Jason Merrill <jason@deneb.cygnus.com>* cvt.c (convert_to_reference): If converting via constructoron local level, go back to build_cplus_new approach.* tree.c (build_cplus_new): If with_cleanup_p, set cleanup slotto error_mark_node to prevent expand_expr from building a cleanupfor this variable.* lex.c (default_assign_ref_body): Return *this from the memcpyversion, too.* decl.c (grok_reference_init): Just return if called witherror_mark_node, don't worry about initializing non-const referencewith temporary.* cvt.c (convert_to_reference): Do the right thing fornon-aggregate reference conversions, pedwarn when generating anon-const reference to a temporary.* class.c (finish_struct): TYPE_HAS_COMPLEX_{INIT,ASSIGN}_REF andTYPE_NEEDS_CONSTRUCTING all depend on TYPE_USES_VIRTUAL_BASECLASSESagain.Tue Feb 15 19:47:19 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_reference_init): Pawn off a lot of the work onconvert_to_reference. Generally do the right thing.* cvt.c (convert_to_reference): Conform to the initial comment;i.e. don't create temps if decl != error_mark_node. Handlecleanups better for temps that do get created. Don't pretendthat we can use an 'A' to initialize a 'const double &' just bytacking on a NOP_EXPR. Support LOOKUP_SPECULATIVELY.* call.c (build_method_call): Set TREE_HAS_CONSTRUCTOR onconstructor calls.Mon Feb 14 14:50:17 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grok_reference_init): Make a temporary for initializingconst reference from constant expression.Mon Feb 14 11:31:31 1994 Per Bothner <bothner@kalessin.cygnus.com>* cp-tree.h, decl.c (set_identifier_local_value): Deleted function.* decl.c (pushdecl): Define decl in correct binding_level(which isn't always the inner_binding_level).* cvt.c (build_up_reference): Don't ever call expand_aggr_init.It's ugly, and I don't think it's the right thing to do.* cp-tree.h, class.c, decl.c, decl2.c, sp/search.c:Remove NEW_CLASS_SCOPING, assuming it is always 1.* decl.c (pop_decl_level): Removed; manually inlined.Sun Feb 13 19:04:56 1994 Jason Merrill <jason@deneb.cygnus.com>* class.h (candidate): Add basetypes field.* call.c (build_method_call): Do access checking after choosing afunction, not before.* Makefile.in (cvt.o, call.o, method.o): Depend on class.h.(mostlyclean): Remove ../cc1plus.Fri Feb 11 11:52:26 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Don't allow adjusting access to a fieldof a base class if a local field has the same name.* error.c (dump_type_prefix): Output basetype for METHOD_TYPEs.Thu Jan 13 17:55:51 1994 Gnanasekaran Swaminathan <gs4t@virginia.edu>* cp-tree.h (DESTRUCTOR_NAME_P): Do not confuse AUTO_TEMP nameswith destructor names when either NO_DOLLAR_IN_LABEL orNO_DOT_IN_LABEL are not defined.Now `template <class T, T f(T&), const T*> class A {...}' works.* pt.c (grok_template_type): Substitute template parm typeswith actual types in complex type as well.(coerce_template_parms): Update the grok_template_type ()function call.* pt.c (tsubst): Traverse method list using DECL_CHAIN.* decl.c (grok_op_properties): Allow operator++/-- to havedefault arguments.* typeck2.c (store_init_value): Don't abort when called toinitialize a type that needs constructing with a CONSTRUCTOR.* init.c (expand_aggr_init_1, CONSTRUCTOR case): Ifstore_init_value fails, build and expand an INIT_EXPR. Ifstore_init_value succeeds, call expand_decl_init.Fri Feb 11 02:49:23 1994 Mike Stump <mrs@cygnus.com>* class.c (build_vbase_path): Use complete_type_p instead ofresolves_to_fixed_type_p to determine if the virtual bases are intheir right place for the type of expr. Cures problem of thinking avirtual base class is one place, when it is in fact someplace else.Fri Feb 11 00:26:46 1994 Mike Stump <mrs@cygnus.com>* init.c (resolve_offset_ref): Make sure we first convert tointermediate type, if given, when dealing with members off `this'.Solves an incorrrect `type `foo' is not a base type for type`multiple'' when it is infact, a base type.Thu Feb 10 21:49:35 1994 Mike Stump <mrs@cygnus.com>* class.c (modify_other_vtable_entries): Use get_binfo, insteadof binfo_value. Solves problem with compiler giving a `base class`B' ambiguous in binfo_value (compiler error)' on complex MIherarchies, when a virtual function is first defied in a virtualbase class.Thu Feb 10 17:19:32 1994 Mike Stump <mrs@cygnus.com>* class.c (build_vbase_path): Don't complain about ambiguousintermediate conversion when converting down to a virtual baseclass, even if they might seem to be ambiguous.Thu Feb 10 12:18:26 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck2.c (build_functional_cast): #if 0 out constructorinheritance code, improve error messages.* class.c (finish_base_struct): Complain about base with onlynon-default constructors in derived class with no constructors.* decl.c (grokdeclarator): Fix detection of virtual new/delete.Wed Feb 9 22:02:32 1994 Mike Stump <mrs@cygnus.com>* search.c (build_mi_virtuals, add_mi_virtuals,report_ambiguous_mi_virtuals): Removed unneeded code.* class.c (finish_struct_bits): Likewise.Wed Feb 9 11:27:17 1994 Jason Merrill <jason@deneb.cygnus.com>* pt.c (end_template_instantiation): Push decl beforepop_from_top_level.* typeck2.c (build_m_component_ref): Make sure datum is ofaggregate type.* init.c (get_type_value): New function, returnsIDENTIFIER_TYPE_VALUE or IDENTIFIER_CLASS_TYPE_VALUE or NULL_TREE.* call.c (build_method_call): Don't die on call to destructor fornon-type.* decl.c (grokdeclarator): Complain about virtual op new and opdelete, make static virtuals unvirtual instead of unstatic.* typeck.c (build_c_cast): Also call default_conversion onmethods.* decl.c (grokdeclarator): Don't complain about anonymousbitfields.* parse.y (simple_stmt, for loops): Move the continue point afterthe cleanups.* class.c (finish_struct): Fix setting ofTYPE_HAS_COMPLEX_INIT_REF.Tue Feb 8 13:21:40 1994 Jason Merrill <jason@deneb.cygnus.com>* init.c (build_new): Deal with `new double (1)'.* class.c (finish_struct): TYPE_HAS_COMPLEX_*_REF are supersets ofTYPE_HAS_REAL_*_REF, but TYPE_HAS_COMPLEX_INIT_REF is independent ofTYPE_NEEDS_CONSTRUCTING.* decl.c (duplicate_decls): Propagate access decls.* typeck2.c (process_init_constructor): Accept empty_init_nodefor initializing unions.* class.c, lex.c, cp-tree.h: UseTYPE_HAS_COMPLEX_ASSIGN_REF where TYPE_HAS_REAL_ASSIGN_REF was usedbefore, use TYPE_HAS_COMPLEX_INIT_REF for TYPE_NEEDS_CONSTRUCTING insome places.* decl.c (finish_decl): Don't complain about uninitialized constif it was initialized before.Mon Feb 7 18:12:34 1994 Jason Merrill <jason@deneb.cygnus.com>* lex.c (default_assign_ref_body): Don't deal with vbases fornow.* decl.c (finish_decl): Fix reversed logic for objects and otherthings that need to be constructed but have no initializer.* class.c (finish_struct): Don't set TYPE_HAS_* flags that areset by grok_op_properties or finish_decl.* decl.c: Don't warn about extern redeclared inline unless-Wextern-inline is given.* decl2.c (lang_decode_option): Likewise.* cp-tree.h: Likewise.Mon Feb 7 17:29:24 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (pushdecl_with_scope): Fix thinko. Add forwarddeclaration.* decl.c (pushdecl_with_scope): New function.* decl.c (pushdecl_top_level): Use new function.* decl.c (pushtag): Initialize newdecl.* decl.c (pushtag): Push new type decl into correct scope.Mon Feb 7 14:42:03 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c, cvt.c, init.c, search.c, cp-tree.h:Eradicate LOOKUP_PROTECTED_OK.Mon Feb 7 13:57:19 1994 Per Bothner <bothner@kalessin.cygnus.com>* decl.c (pushtag, xref_tag), cp-tree.h: Add extra parameter'globalize' to signify implicit declarations.* decl.c (globalize_nested_type, maybe_globalize_type): Removed.* decl.c (set_identifier_type_value_with_scope): New function.* decl.c (set_identifier_local_value): Simplify.* spew.c (yylex, do_addr): Modify to return a _DEFN if aforward declaration (followed by ';' and not preceded by 'friend').* class.c, decl.c, except.c, init.c, parse.y,pt.c, search.c: Add new argument to calls to xref_tag andpushtag.Mon Feb 7 00:22:59 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-tree.h (ACCESSIBLY_UNIQUELY_DERIVED_P): New macro, means whatACCESSIBLY_DERIVED_FROM_P meant before.(ACCESSIBLY_DERIVED_FROM_P): Now disregards ambiguity.* cvt.c (build_up_reference): Call get_binfo with PROTECT == 1.* search.c (get_base_distance_recursive): Members and friends ofa class X can implicitly convert an X* to a pointer to a private orprotected immediate base class of X.(get_binfo_recursive): Likewise.(get_base_distance): Ignore ambiguity if PROTECT < 0.(get_binfo): Lose multiple values of PROTECT.(compute_access): Protected is OK if the start of thesearch is an accessible base class of current_class_type.* method.c (build_opfncall): Do check access on operator new here.* decl.c (finish_function): Don't check access on operator newhere.Sun Feb 6 14:06:58 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (xref_tag): The base of a derived struct is NOT alwayspublic. Duh.* pt.c (do_explicit_instantiation): New function, called fromparser to do explicit function instantiation.(type_unification): Allow the args list to be terminated withvoid_list_node.(do_pending_expansions): Look at i->interface for non-membertemplates.* parse.y (datadef): Move explicit_instantiation here.(structsp): From here.(datadef): Complain about `int;'.Sun Feb 6 12:33:18 1994 Per Bothner <bothner@kalessin.cygnus.com>* pt.c (end_template_instantiation), cp-tree.h: Remove unusedsecond parameter, and simplify first from a TREE_LIST wherewe only care about its TREE_VALUE to just the value (an IDENTIFIER).* pt.c (instantiate_member_templates): Simplify argument listfrom a TREE_LIST to just an IDENTIFIER.* lex.c (yyprint): PRE_PARSED_CLASS_DECL is now just an IDENTIFIER.* parse.y (template_instantiate_once): Simplify accordingly.* decl.c (inner_binding_level): New. Use various places tosimplify.Sun Feb 6 02:49:37 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck2.c (build_functional_cast): int() -> int(0).Sat Feb 5 00:53:21 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Don't do a bitwise copy for op= if theclass has a virtual function table.* typeck.c (convert_for_initialization): Restore warnings aboutnot using defined op=. Should really be my_friendly_aborts, Is'pose.Fri Feb 4 14:21:00 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Tidy up conditions for doing bitwisecopies of objects.* decl.c (build_default_constructor): #if 0 out.* *: Eradicate TYPE_GETS_{ASSIGNMENT,ASSIGN_REF,CONST_ASSIGN_REF,CONST_INIT_REF}, TYPE_HAS_REAL_CONSTRUCTOR.* decl.c (grokdeclarator): Don't return void_type_node forfriends being defined here.* init.c (perform_member_init): Only do the init if it's useful.* lex.c (default_copy_constructor_body): If we don't need to domemberwise init, just call __builtin_memcpy.(default_assign_ref_body): Likewise.* decl.c (grokdeclarator): If friendp && virtualp, friendp = 0.Fri Feb 4 13:02:56 1994 Mike Stump <mrs@cygnus.com>* lex.c (reinit_parse_for_method, cons_up_default_function):Don't give warn_if_unknown_interface warning when it came from asystem header file.* pt.c (end_template_decl, instantiate_template): Likewise.* decl.c (start_decl): Likewise.Fri Feb 4 00:41:21 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Don't try to set TYPE_WAS_ANONYMOUS onenums.* decl2.c (constructor_name_full): Use IS_AGGR_TYPE_CODE instead ofIS_AGGR_TYPE, since we don't know it's a type.Thu Feb 3 11:36:46 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokdeclarator): Don't complain about anonymous unions.* cp-tree.h (TYPE_WAS_ANONYMOUS): This struct was originallyanonymous, but had a name given to it by a typedef.* decl.c (grokdeclarator): When renaming an anonymous struct, setTYPE_WAS_ANONYMOUS.* decl2.c (constructor_name_full): Use TYPE_WAS_ANONYMOUS.* cp-tree.h (DECL_UNDEFINED_FRIENDS): #if 0 out.* init.c (xref_friend): Don't set up DECL_UNDEFINED_FRIENDS.(embrace_waiting_friends): Don't use DECL_UNDEFINED_FRIENDS.* decl.c (grokdeclarator): Set TYPE_NESTED_NAME properly on nestedanonymous structs that get typedef'd.* decl.c (grokdeclarator): Always return void_type_node forfriends.* error.c (dump_function_decl): Don't use DECL_CLASS_CONTEXT forfriends.(dump_function_decl): Don't print out default args fora function used in an expression.* decl.c (grokdeclarator): Give error on abstract declarator usedin an invalid context (i.e. `void (*)();').* error.c (cp_line_of): Support _TYPE nodes.(cp_file_of): Likewise.* cvt.c (build_up_reference): Don't abort if passed a SAVE_EXPR;it can happen for the RHS of an assignment stmt where the LHS isa COND_EXPR.* init.c (expand_aggr_init_1): Deal with bracketed initializerlists properly.* class.c (finish_struct): Deal with enumerators and typedefsagain.Wed Feb 2 11:30:22 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Tidy up loop over fields.* errfn.c (cp_thing): Don't advance twice after a format.* class.c (finish_struct): Complain about needing a constructorif a member has only non-default constructors, and don't try togenerate a default constructor.* decl.c (finish_decl): Also do the constructor thing ifTYPE_NEEDS_CONSTRUCTING is set (for arrays).* search.c (unuse_fields): New function: mark all fields in thistype unused.(dfs_unuse_fields): Helper function.* class.c (pushclass): If the new class is the same as the oldclass, still unuse the fields.(unuse_fields): Move to search.c.* decl.c (grok_op_properties): Add friendp argument.(grokfndecl): Pass it.(start_method): Likewise.* decl2.c (delete_sanity): Add use_global_delete parameter to catch::delete calls.* parse.y (unary_expr): Pass new parameter to delete_sanity.* lex.c (default_copy_constructor_body): Don't choke if the unionhas no fields.(default_assign_ref_body): Likewise.* call.c (compute_conversion_costs_ansi): Do the right thing forellipsis matches.* decl.c (push_to_top_level): Optimize.* decl.c (start_function): Look for the lexical scope of a friendin DECL_CLASS_CONTEXT.* init.c (do_friend): Set DECL_CLASS_CONTEXT on global friends.Tue Feb 1 15:59:24 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-tree.h (TREE_GETS_PLACED_NEW): New macro.* init.c (init_init_processing): Don't assign BIN/BID to theIDENTIFIER_GLOBAL_VALUEs of their respective operators.(build_new): Check TREE_GETS_PLACED_NEW.* decl.c (grok_op_properties): Don't set TREE_GETS_NEW for a decl ofop new with placement, set TREE_GETS_PLACED_NEW.* cp-tree.h (ANON_UNION_P): New macro. Applies to decls.* class.c (finish_struct): Don't treat anonymous unions likeother aggregate members. Do synthesize methods for unions withouta name, since they may or may not be "anonymous unions".* decl2.c (grok_x_components): Wipe out memory of synthesized methodsin anonymous unions.* lex.c (default_copy_constructor_body): Support unions.(default_assign_ref_body): Likewise.Mon Jan 31 12:07:30 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-tree.h: Fix documentation of LOOKUP_GLOBAL, add prototypes.* error.c (args_as_string): New function (%A), like type_as_stringexcept NULL_TREE -> "..."* call.c (build_overload_call_real): Fix for new overloading.* decl.c (grok_op_properties): Set all of the TYPE_OVERLOADS_* flagshere.* parse.y (operator_name): Instead of here.* typeck2.c (build_functional_cast): Treat a TREE_LIST as a listof functions.* call.c (build_overload_call_real): Support LOOKUP_SPECULATIVELY.* method.c (build_opfncall): Don't need to massage return valueany more, call build_overload_call with all flags.* typeck.c (build_x_binary_op): Put back speculative call tobuild_opfncall.(build_x_unary_op): Likewise.(build_x_conditional_expr): Likewise.Mon Jan 31 10:00:30 1994 Mike Stump <mrs@cygnus.com>* cvt.c (build_type_conversion_1): Change call to pedwarn intowarning, and conditionalize upon warn_cast_qual.Fri Jan 28 11:48:15 1994 Jason Merrill <jason@deneb.cygnus.com>* search.c (lookup_field): If xbasetype is a binfo, copy it toavoid clobbering its inheritance info.* call.c (build_method_call): Don't overwrite basetype_path withTYPE_BINFO (inst_ptr_basetype) if they have the same type.* search.c (compute_access): Fix handling of protected inheritanceand friendship with the enclosing class.* typeck2.c (store_init_value): Allow passing of TREE_CHAIN forinitialization of arbitrary variable.* typeck2.c (build_functional_cast): Only try calling a method ifone exists.* decl.c (grokdeclarator): Move handling of constructor syntaxinitialization into first loop for generality.(parmlist_is_random): Lose.* lex.c (cons_up_default_function): Set TREE_PARMLIST on argumentsto default function.Thu Jan 27 19:26:51 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (grokparms): Abort if we get called with something we don'texpect.Thu Jan 27 17:37:25 1994 Mike Stump <mrs@cygnus.com>* call.c (build_overload_call_real): Change argument complain toflags to match style of rest of code. Pass it down tobuild_function_call_real as necessary.* call.c (build_overload_call, build_overload_call_maybe): Changeargument complain to flags to match style of rest of code.* cp-tree.h (build_function_call_real): Added fourth flagsargument.* cvt.c (convert_to_reference): Only give warning messages, ifLOOKUP_COMPLAIN is set.* typeck.c (build_x_function_call): Change simple complainargument to build_overload_call_maybe and build_overload_call, toLOOKUP_COMPLAIN to match style of rest of code.* typeck2.c (build_functional_cast): Likewise.* typeck.c (build_function_call_real): Add flags, so that we cannot complain, if we don't want to complain. Complain aboutarguments, if we are complaining, otherwise don't.* typeck.c (build_function_call, build_function_call_maybe):Stick in flags argument.* typeck.c (build_x_binary_op, build_x_unary_op,build_x_conditional_expr, build_x_compound_expr): Follow style ofbuild_x_indirect_ref, as it is more correct and more common.Thu Jan 27 14:36:20 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (build_method_call): Don't check for being called witha pointer.* decl2.c (finish_file): Don't play with DECL_CLASS_CONTEXT for thestatic initializer function.* init.c (build_member_call): Use convert_force here, too.* search.c (compute_access): Only treat static members speciallyif they are referenced directly.Wed Jan 26 18:28:14 1994 Jason Merrill <jason@deneb.cygnus.com>* gxxint.texi (Access Control): New node.* search.c (current_scope): New function; returns whichever ofcurrent_class_type and current_function_decl is the most nested.(compute_access): Total overhaul to make it clearer and morecorrect. Don't use the cache for now; in the only situation whereit was used before, it gained nothing. This frees up three of theDECL_LANG_FLAGs for possible other use!* cp-tree.h: #if 0 out DECL_PUBLIC & friends.* typeck.c (build_component_ref_1): Don't check DECL_PUBLIC.* call.c (build_method_call): Use convert_force to cast `this' --rely on the access checking for the method itself.* init.c (is_friend): Do the nesting thing, handle types. I ammy own friend.(is_friend_type): Become a shell for is_friend.(add_friend): Never stick in ctype.Why are the friendship functions in init.c, anyway?Wed Jan 26 17:50:00 1994 Mike Stump <mrs@cygnus.com>* cvt.c (build_type_conversion_1): Don't conditionalize call topedwarn upon pedantic.Wed Jan 26 17:20:46 1994 Mike Stump <mrs@cygnus.com>* cvt.c (convert_to_reference): Add 8.4.3 checking so that onegets a warning if one tries to initialize a non-const & from anon-lvalue.* cvt.c (convert_to_reference): Use %P format for argumentnumbers in warnings.Wed Jan 26 14:35:06 1994 Mike Stump <mrs@cygnus.com>* init.c (build_delete): Follow style in call.c to construct thevirtual call to the desctructor, as that code is right. Fixes aproblem of the compiler saying a pointer conversion is ambiguous.Wed Jan 26 11:28:14 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-tree.h (VTABLE_NAME_P): Change other occurrence ofVTABLE_NAME_FORMAT to VTABLE_NAME.* *: s/visibility/access/gTue Jan 25 18:39:12 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_modify_expr): Don't smash references if INIT_EXPR.Tue Jan 25 13:54:29 1994 Mike Stump <mrs@cygnus.com>* init.c (build_delete): Back out Jan 17th & 18th pacthes, asthey break libg++.Tue Jan 25 13:11:45 1994 Jason Merrill <jason@deneb.cygnus.com>* decl.c (duplicate_decls): Fix pointer arithmetic.Mon Jan 24 15:50:06 1994 Chip Salzenberg <chip@fin.uucp>[ cp-* changes propagated from c-* changes in 940114 snapshot ]* cp-parse.y (maybe_attribute): Allow multiple __attribute__clauses on a declaration.Mon Jan 24 17:06:23 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Do synthesize methods for anonstructs, just not unions.Mon Jan 24 13:50:13 1994 Kung Hsu <kung@mexican.cygnus.com>* decl.c (xref_tag): Handle anonymous nested type.* decl.c (globalize_nested_type): Add no globalize bit check.* spew.c (hack_more_ids): Templated nested decl not push toplevel.* parse.y: Get rid of 'goto do_components'. It is much betterfor debugging.* decl.c (is_anon_name): Get rid of the function and use themacro ANON_AGGRNAME_P.* pt.c: Ditto.Fri Jan 21 14:06:02 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct): Don't synthesize any methods foranonymous structs/unions.* typeck.c (build_modify_expr): Don't treat pmf's as class objects.Thu Jan 20 18:56:46 1994 Jason Merrill <jason@deneb.cygnus.com>* method.c (build_opfncall): Call build_indirect_ref onsynthesized instance for operator delete.* pt.c (type_unification): Don't abort if called with a list oftypes in ARGS.* class.c (instantiate_type): Deal with function templates.Thu Jan 20 16:55:35 1994 Jim Wilson <wilson@sphagnum.cygnus.com>* Makefile.in (CC): Default to cc not gcc.Thu Jan 20 13:47:54 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_modify_expr): Call constructor if appropriate.* decl.c (push_to_top_level): Clear out class-level bindings cache.Wed Jan 19 13:51:22 1994 Jason Merrill <jason@deneb.cygnus.com>* call.c (resolve_scope_to_name): Work recursively (previously onlylooked down one level).* lex.c (do_pending_inlines): If we're still dealing with the lastbatch of inlines, don't start working on a new one.* Makefile.in (stamp-parse): Update conflict count.(TAGS): Fix.* parse.y (explicit_instantiation): New rule; implements'template A<int>' syntax (though not 'template foo(int)' yet).(structsp): Add explicit_instantiation.Tue Jan 18 13:53:05 1994 Jason Merrill <jason@deneb.cygnus.com>* class.c (finish_struct, etc.): Simplify decision to synthesizea destructor.* call.c, class.c, cp-tree.h, decl.c, init.c,ptree.c, search.c, typeck.c, typeck2.c: NukeTYPE_NEEDS_CONSTRUCTOR (change all calls to TYPE_NEEDS_CONSTRUCTING).* init.c (expand_aggr_init_1): Don't try non-constructor methodsof initializing objects.(build_new): Don't try other methods if the constructor lookup fails.* class.c (finish_base_struct): Set cant_have_default_ctor andcant_synth_copy_ctor properly.(finish_struct): Likewise.Mon Jan 17 13:58:18 1994 Jason Merrill <jason@deneb.cygnus.com>* typeck.c (build_modify_expr_1): #if 0 out again.(build_modify_expr): #if 0 out memberwise init code again.* lex.c (default_copy_constructor_body): Be const-correct.(default_assign_ref_body): Likewise.* init.c (perform_member_init): Use TYPE_HAS_CONSTRUCTOR to decidewhether or not to use it, rather than TYPE_NEEDS_CONSTRUCTING.(expand_aggr_init): Disable silent conversion from initializer listto list of args for a constructor.* class.c (base_info): Lose needs_default_ctor.(finish_base_struct): Likewise.(finish_struct): Likewise.* decl.c (init_decl_processing): Don't turn off flag_default_inlinejust because flag_no_inline is on.(finish_decl): Use TYPE_HAS_CONSTRUCTOR to decide to useconstructor.* class.c (finish_struct): Synthesize default ctor wheneverallowed.* Makefile.in (TAGS): Don't try to run etags on cp-parse.y.Sat Jan 15 18:34:33 1994 Mike Stump <mrs@cygnus.com>* Makefile.in, configure: Handle the C++ front-end in asubdirectory.* cp-*: Move C++ front-end to cp/*.Fri Jan 14 14:09:37 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-typeck.c (build_function_call_real): Modify to match otherinstances of taking the address of the function.* cp-class.c (finish_struct): Set TYPE_HAS_REAL_CONSTRUCTOR to 1 ifthere are non-synthesized constructors.Only set TYPE_NEEDS_CONSTRUCTOR if TYPE_HAS_REAL_CONSTRUCTOR.Always generate copy constructor if possible.* cp-tree.h (lang_type): Add has_real_constructor bitfield.(TYPE_HAS_REAL_CONSTRUCTOR): Define.* cp-lex.c (default_copy_constructor_body): Use init syntaxfor all bases.* cp-type2.c (store_init_value): Only give error for initializer listif TYPE_HAS_REAL_CONSTRUCTOR.Thu Jan 13 15:38:29 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-tree.h (DECL_SYNTHESIZED): Add defn.(lang_decl): Add synthesized bitfield to decl_flags.* cp-lex.c (cons_up_default_function): Use DECL_SYNTHESIZED to markartificial methods, rather than a line # of 0.Fri Jan 14 18:25:29 1994 Kung Hsu <kung@mexican.cygnus.com>* cp-decl (xref_tag): Fix a bug in conflict type.* cp-parse.y: Add SCOPED_NAME for uninstantiated template nestedtype reference.* cp-spew.c (yylex): Generated SCOPED_NAME token.* cp-lex.c (yyprint): Handle SCOPED_NAME.Fri Jan 14 17:00:29 1994 Mike Stump <mrs@cygnus.com>* cp-decl.c (pushdecl): Revert patch from Jan 11 19:33:03, as it isnot right.Thu Jan 13 14:00:35 1994 Kung Hsu <kung@mexican.cygnus.com>* cp-decl2.c (grok_x_components): Fix a bug that enum type does nothave type_flags.Thu Jan 13 11:39:34 1994 Mike Stump <mrs@cygnus.com>Ensure that all vtable pointers are initialized with all the rightvalues.* cp-class.c (is_normal): Changed to reflect new meaning ofCLASSTYPE_VFIELD_PARENT.* cp-class.c (maybe_fixup_vptrs): Use ofCLASSTYPE_NEEDS_VIRTUAL_REINIT here is misguided. UseBINFO_MODIFIED instead.* cp-class.c (finish_struct): Changed to reflect new meaning ofCLASSTYPE_VFIELD_PARENT.* cp-decl.c (get_binfo_from_vfield): Removed, unneeded now.* cp-decl.c (finish_function): Use init_vtbl_ptrs, instead of opencoding it here.* cp-init.c (init_vfields): Changed name to init_vtbl_ptrs, andre-implement.* cp-init.c (emit_base_init): Use new name init_vtbl_ptrs.* cp-tree.h (vfield_parent): Changed to integer.* cp-tree.h (CLASSTYPE_VFIELD_PARENT): Changed docs to reflect newmeaning.* cp-tree.h (init_vtbl_ptrs): Added init_vtbl_ptrs.Wed Jan 12 18:24:16 1994 Kung Hsu <kung@mexican.cygnus.com>* cp-decl.c (xref_tag): Re-implement globalize nested type.* cp-decl2.c (grok_x_components): Ditto.* cp-parse.y: Ditto.* cp-tree.h (lang_type): Add no_globalize bit in type_flags.Wed Jan 12 14:08:09 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-decl.c (grokdeclarator): Don't set TREE_PUBLIC on frienddecls with a definition attached.* cp-typeck.c (build_modify_expr): Undo previous change in the caseof INIT_EXPRs.Tue Jan 11 19:33:03 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-typeck.c (build_modify_expr): Replace code for generatingassignment semantics for classes with an error.(build_modify_expr_1): #if 0 out.* cp-decl.c (pushdecl): Patch bogus design of pushdeclbehavior for overloaded functions (it doesn't push anything).* cp-class.c (finish_struct): When generating default op=,set TYPE_HAS_ASSIGNMENT.Mon Jan 10 18:48:06 1994 Mike Stump <mrs@cygnus.com>* cp-cvt.c (convert): Make {double, clashing enum} -> enuminvalid.* cp-typeck.c (convert_for_assignment): Simplify.* cp-decl2.c (warn_enum_clash): Removed.* invoke.texi (-Wenum-clash): Removed.* toplev.c (-Wenum-clash): Removed.Mon Jan 10 17:48:37 1994 Kung Hsu <kung@mexican.cygnus.com>* cp-decl.c (finish_decl): Fix incorrect popclass call.* cp-decl.c (is_anon_name): New function, check whether the nameis anonymous name generated by compiler.* cp-decl.c (grokdeclarator): Allow nested SCOPE_REF* cp-spew.c (hack_more_ids): Handle nested type in template.* cp-parse.y: Handle nested type reference in uninstantiatedtemplate.* cp-call.c (build_method_call): Handle uninstantiated templatecase.* cp-pt.c (search_nested_type_in_tmpl): New function, search nestedtype in template.* cp-pt.c (lookup_nested_type_by_name): New function, lookup nestedtype by name.* cp-pt.c (tsubst): Handle nested type search by name.Mon Jan 10 14:32:18 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-init.c (build_member_call): Propagate qualifiers to new type.* cp-call.c (build_method_call): Count functions the new way.Fri Jan 7 19:03:26 1994 Jason Merrill <jason@deneb.cygnus.com>* cp-decl.c (pushtag): Set DECL_ASSEMBLER_NAME for nested classes,too.Tue Jan 4 16:45:51 1994 Kung Hsu <kung@cirdan.cygnus.com>* cp-parse.y: Change to handle whether to globalize nested class.* cp-decl.c (xref_tag, maybe_globalize_type): Likewise.Mon Jan 3 22:22:32 1994 Gerald Baumgartner <gb@cygnus.com>* Makefile.in cp-call.c cp-class.c cp-cvt.c cp-decl.c cp-decl2.ccp-error.c cp-init.c cp-lex.c cp-lex.h cp-method.c cp-parse.ycp-spew.c cp-tree.c cp-tree.h cp-type2.c cp-typeck.c cp-xref.cgplus.gperf toplev.c: Incorporated C++ signature extension.* cp-sig.c: New file, contains most of signature processing.* cp-hash.h: Regenerated from gplus.gperf.* gcc.1 g++.1: Added explanation for the `-fhandle-signatures'and `-fno-handle-signatures' command line flags.* gcc.texi: Changed the last-modification date.* invoke.texi: Added `-fhandle-signatures' in the list ofC++ language options. Added explanation for this option.Copyright (C) 1994 Free Software Foundation, Inc.Copying and distribution of this file, with or without modification,are permitted in any medium without royalty provided the copyrightnotice and this notice are preserved.
Go to most recent revision | Compare with Previous | Blame | View Log
