symbol_completion_function, takes the line buffer and the
point in the line buffer as additional arguments.
(readline_line_completion_function): New function, interface
between readline and line_completion_function.
(init_main): Use it.
(complete_command): Use line_completion_function instead of
abusing rl_line_buffer. Free completion strings after printing
them.
* symtab.c (completion_list_add_name): Recheck for duplicates
if we intend to add a modified symbol.
* gdbtypes.h (cplus_struct_type): nfn_fields_total no longer
includes the number of methods from the baseclasses.
* stabsread.c (attach_fn_fields_to_type): No longer add the
number of methods from the baseclasses to TYPE_NFN_FIELDS_TOTAL,
the baseclass type might not have been completely filled in yet.
* symtab.c (total_number_of_methods): New function to compute
the total number of methods for a type, including the methods
from baseclasses.
(decode_line_1): Use it instead of TYPE_NFN_FIELDS_TOTAL to
allocate the symbol array for find_methods.
* stabsread.c (scan_file_globals): Add default case to minimal
symbol type switch, to avoid gcc -Wall warnings.
* config/rs6000/tm-rs6000.h (INIT_EXTRA_FRAME_INFO):
Don't test for zero backchain pointer to recognize a signal
handler frame, if read() gets interrupted by a signal, the
backchain will be non zero.
(SIG_FRAME_FP_OFFSET): Move to here from rs6000-tdep.c,
improve comment.
(SIG_FRAME_PC_OFFSET): New definition.
(FRAME_SAVED_PC): Return saved pc from sigcontext if this
is a signal handler frame.
* rs6000-tdep.c (function_frame_info): Do not error out
if we can't access the instructions.
* config/rs6000/tm-rs6000.h (CONVERT_FROM_FUNC_PTR_ADDR):
New definition to get the function address from a function pointer.
* valops.c (find_function_addr): Use it when calling a user
function through a function pointer.
* ch-typeprint.c, ch-valprint.c: Use TYPE_INDEX_TYPE.
* ch-valprint.c (chill_val_print): Pass index type directly
(instead of its TYPE_TARGET_TYPE) to print_type_scalar.
* stabsread.c (read_type): Don't set TYPE_FLAG_TARGET_STUB
if the index type is a stub.
f77_create_literal_complex_type.
* valops.c (f77_value_literal_string, f77_value_substring,
f77_value_literal_complex): Use xmalloc not malloc.
* valops.c (f77_value_literal_string, f77_value_substring):
Make addr char * not CORE_ADDR.
* value.h (struct value): Add new field literal_data of aligner union.
(VALUE_LITERAL_DATA): Use it.
* f-lang.h: Declare find_common_for_function.
* value.h, valops.c: Split VALUE_SUBSTRING_START into memaddr and
myaddr fields of a union. Don't overload it with the frame field
(not sure this is necessary; I'm not sure what lval_* codes
VALUE_SUBSTRING_* can be used with).
(fmbutt@engage.sps.mot.com).
* Makefile.in: Add Fortran-related files and dependencies.
* defs.h (language_fortran): New language enum.
* language.h (_LANG_fortran): Define.
(MAX_FORTRAN_DIMS): Define.
* expression.h: Reformat to standard.
(MULTI_F77_SUBSCRIPT, OP_F77_UNDETERMINED_ARGLIST,
OP_F77_LITERAL_COMPLEX, OP_F77_SUBSTR): New expression opcodes.
* gdbtypes.h (TYPE_CODE_COMPLEX, TYPE_CODE_LITERAL_COMPLEX,
TYPE_CODE_LITERAL_STRING): New type codes.
(type): New fields upper_bound_type and lower_bound_type.
(TYPE_ARRAY_UPPER_BOUND_TYPE, TYPE_ARRAY_LOWER_BOUND_TYPE,
TYPE_ARRAY_UPPER_BOUND_VALUE, TYPE_ARRAY_LOWER_BOUND_VALUE): New
macros.
(builtin_type_f_character, etc): Declare.
* value.h (VALUE_LITERAL_DATA, VALUE_SUBSTRING_START): Define.
* f-exp.y: New file, Fortran expression grammar.
* f-lang.c: New file, Fortran language support functions.
* f-lang.h: New file, Fortran language support declarations.
* f-typeprint.c: New file, Fortran type printing.
* f-valprint.c: New file, Fortran value printing.
* eval.c (evaluate_subexp): Add code for new expression opcodes,
fix wording of error message.
* gdbtypes.c (f77_create_literal_complex_type,
f77_create_literal_string_type): New functions.
* language.c (set_language_command): Add Fortran info.
(calc_f77_array_dims): New function.
* parse.c (length_of_subexp, prefixify_subexp): Add cases for new
expression opcodes.
* symfile.c (deduce_language_from_filename): Recognize .f and .F
as Fortran source files.
* valops.c (f77_value_literal_string, f77_value_substring,
f77_value_literal_complex): New functions.
fputs_unfiltered and exit directly, rather than fatal. The latter
calls vfprintf_unfiltered!
* gdbtypes.h, gdbtypes.c (can_dereference): New function.
* value.h, printcmd.c (print_value_flags): Move from here...
* annotate.c: ...to here, and make it use can_dereference.
an address to 'w'. Make it 'g' or 'h', depending on TARGET_PTR_BIT.
* defs.h: Just typedef CORE_ADDR to bfd_vma. Include bfd.h.
If BFD64, make a LONGEST a BFD_HOST_64_BIT.
* defs.h (longest_to_int): Don't depend on CC_HAS_LONG_LONG; instead
always just check against INT_MIN and INT_MAX (this also fixes things
if sizeof (long) > sizeof (int), e.g. Alpha).
* config/pa/tm-hppa.h, config/i386/sun386.h, config/rs6000/tm-rs6000.h:
Don't define LONGEST or BUILTIN_TYPE_LONGEST.
* gdbtypes.h: Remove BUILTIN_TYPE_LONGEST and
BUILTIN_TYPE_UNSIGNED_LONGEST.
* language.h, c-lang.c, ch-lang.c, m2-lang.c: Remove longest_int
and longest_unsigned_int.
* value.h (struct value): Just align to LONGEST, rather than worrying
about CC_HAS_LONG_LONG.
* valarith.c (value_binop, _initialize_valarith): Figure out type
ourself based on sizeof (LONGEST) rather than relying on
BUILTIN_TYPE_LONGEST. The point is that we don't depend on
CC_HAS_LONG_LONG anymore.
* valprint.c (val_print_type_code_int): Just call
extract_unsigned_integer directly, rather than going through
unpack_long.
* printcmd.c (decode_format): Remove code which would sometimes
change 'g' size to 'w' for integers. print_scalar_formatted handles
printing huge integers well enough, thank you.
Change comments regarding TYPE_CODE_BOOL.
* language.c (boolean_type): Always return 1 for TYPE_CODE_BOOL,
regardless of the language.
(value_true): Just call value_logical_not regardless of language.
* coffread.c (coff_read_enum_type), stabsread.c (read_enum_type):
Remove #if 0'd code which makes some enums TYPE_CODE_BOOL.
* language.h: Improve comment for la_builtin_type_vector.
* m2-lang.c (_initialize_m2_language): Don't add any fields to
builtin_type_m2_bool.
* ch-lang.c (chill_is_varying_struct), ch-lang.h: New function.
* ch-lang.c (chill_printstr): Use double quotes, not single quotes.
* ch-typeprint.c (chill_type_print_base): Handle TYPE_CODE_BITSTRING.
Improve printing of TYPE_CODE_STRING, TYPE_CODE_SET, and
TYPE_CODE_STRUCT (including checking chill_is_varying_struct).
Print TYPE_DUMMY_RANGE by printing its TYPE_TARGET_TYPE.
Handle TYPE_CODE_ENUM.
* ch-valprint.c (chill_val_print): Handle TYPE_CODE_BITSTRING.
For TYPE_CODE_STRING, never print address. Handle VARYING strings.
* gdbtypes.c (force_to_range_type): New.
* gdbtypes.c (create_set_type): Make work, following Chill layout.
* gdbtypes.h (TYPE_LOW_BOUND, TYPE_HIGH_BOUND, TYPE_DUMMY_RANGE): New.
* stabsread.c (read_type): Distinguish string and bitstring from
char-array and set.
* valarith.c (value_subscript), valops.c (value_coerce_array):
Handle STRINGs as well as ARRAYs.
* valarith.c (value_bit_index): Fix think. Use new macros.
* gdbtypes.c (check_stub_type): On TYPE_FLAG_TARGET_STUB, do
what cleanup_undefined_types does for arrays, except we clear
TYPE_FLAG_TARGET_STUB if we fix up the type.
* stabsread.c (cleanup_undefined_types): Add comments about how
doing arrays here is no longer the clean way to do it.
(read_array_type): Set TYPE_FLAG_TARGET_STUB as well as calling
add_undefined_type.
* c-typeprint.c, ch-typeprint.c: Move call to check_stub_type
outside switch so it happens for all type codes.
* cp-valprint.c (cp_print_value_fields): Recurse to val_print,
instead of c_val_print, so that check_stub_type gets called.
* gdbtypes.h, gdbtypes.c, m2-lang.c, ch-lang.c, c-lang.c: Remove
TYPE_FLAG_SIGNED. It was inconsistently set, never checked
(except in recursive_dump_type), and is pointless.
gdbtypes.h stabsread.c cp-valprint.c ChangeLog
* gdbtypes.h : add a field 'ignore_field_bits in cplus_specific,
and macros to handle the bits.
* stabsread.c (read_one_struct_field): add VISIBILITY_IGNORE, and
for field of length 0, set this bit on.
* cp-valprint.c (cp_print_value_fields): for VISIBILITY_IGNORE
field, print <no value>.
Change all references to stdout/stderr to gdb_stdout/gdb_stderr.
Replace all calls to stdio output functions with calls to
corresponding _unfiltered functions (`fprintf_unfiltered')
Replaced calls to fopen for output to gdb_fopen.
Added sufficient goo to utils.c and defs.h to make the above work.
The net effect is that stdio output functions are only directly used
in utils.c. Elsewhere, the _unfiltered and _filtered functions and
GDB_FILE type are used.
In the near future, GDB_FILE will stop being equivalant to FILE.
The semantics of some commands has changed in a very subtle way:
called in the right context, they may cause new occurences of
prompt_for_continue() behavior. The testsuite doesn't notice anything
like this, though.
Please respect this change by not reintroducing stdio output
dependencies in the main body of gdb code. All output from commands
should go to a GDB_FILE.
Target-specific code can still use stdio directly to communicate with
targets.
* symmisc.c (print_symbol): Use %02x not %2x for LOC_CONST_BYTES.
Clean up problems with targets and hosts that have 64 bit longs
and pointers and 32 bit ints.
* breakpoint.c, buildsym.c, c-lang.c, c-valprint.c, ch-lang.c,
ch-valprint.c, core.c, cp-valprint.c, dbxread.c, exec.c,
expprint.c, gdbtypes.c, infcmd.c, language.c, language.h,
m2-lang.c, maint.c, mips-tdep.c, mipsread.c, partial-stab.h,
printcmd.c, remote-vx.c, solib.c, source.c, stack.c, symfile.c,
symmisc.c, symtab.c, valops.c, valprint.c, xcoffexec.c:
Change all printf formats from %x to %lx if outputting an address.
Change la_*_format to use long format.
local_hex_string, local_hex_string_custom now take an unsigned long
argument, change all callers.
* coffread.c (read_coff_symtab): Remove superfluous cast for
complaint output.
* dbxread.c (end_psymtab): Cast MSYMBOL_INFO to long, not int.
* findvar.c, value.h (write_register): Change register value to long.
* gdbtypes.h (struct type): Change `bitsize' to long as
TYPE_FIELD_STATIC_PHYSNAME uses this field as a pointer.
* inferior.h (struct inferior_status): Change type of stop_pc to
CORE_ADDR.
* language.h (local_octal_string, local_octal_string_custom):
Remove prototype, the functions are neither defined nor used.
* mipsread.c (parse_symbol): Use temporary variable for bitsize as
f->bitsize is a long now.
* objfiles.c (add_to_objfile_sections, build_objfile_section_table):
Use unsigned long casts instead of int for abusing sections_end
pointer as integer.
* stack.c (parse_frame_specification): Change type of `args' to
CORE_ADDR for SETUP_ARBITRARY_FRAME.
* printcmd.c (make_vasize): Allow redefinition via MAKEVA_SIZE.
* mipsread.c (parse_type): Alpha cc now supports the t->continued
bit, update algorithm to match the way the compiler uses it.
read_type_number static.
(define_symbol): Don't try to deal with a missing symbol
descriptor which isn't followed by digit, '(', or '-'.
* stabsread.h: Don't declare read_type_number here.
* gdbtypes.h: Don't declare error_type here.
* xcoffread.c: Remove NO_TYPEDEFS code.
(define_symbol): Use read_type for type of enum constant,
not just read_type_number. Also don't call error().
(define_symbol): For unrecognized constant type, one complaint (the
one from error_type) is enough. Don't make our own in addition.
(define_symbol): Don't treat an N_FUN 'R' as a prototype.
* gdbtypes.h: Doc fixes.
when compiling with gcc, but disable it for now. See comment.
* defs.h (LONGEST): Define as either "long" or "long long"
based on CC_HAS_LONG_LONG.
* defs.h (longest_to_int): Use CC_HAS_LONG_LONG to control
how longest_to_int is defined.
* c-valprint.c (c_val_print): Call print_longest.
* expprint.c (dump_expression): Use PRINTF_HAS_LONG_LONG
instead of LONG_LONG.
* {printcmd.c, gdbtypes.h} (LONG_LONG): Replace usages with
CC_HAS_LONG_LONG.
* printcmd.c (print_scalar_formatted): Call print_longest
and let it figure out what to do for PRINTF_HAS_LONG_LONG.
* typeprint.c (print_type_scalar): Call print_longest and let
it figure out what to do for PRINTF_HAS_LONG_LONG.
* valprint.c (val_print_type_code_int): Call print_longest
and let it figure out what to do for PRINTF_HAS_LONG_LONG.
* stabsread.c (LONG_LONG): Replace usages with CC_HAS_LONG_LONG.
* value.h (struct value): Replace usage of LONG_LONG with
CC_HAS_LONG_LONG.
* value.h (print_longest): Add prototype.
* values.c (LONG_LONG): Replace usages with CC_HAS_LONG_LONG.
* values.c (unpack_double): Collapse code that was unnecessarily
dependent on CC_HAS_LONG_LONG. Use LONGEST instead of direct types.
* values.c (value_from_longest): Remove dependency on
CC_HAS_LONG_LONG and just use LONGEST.
* solib.c (solib_map_sections): Use bfd_get_filename
to access filename field.
* solib.c (clear_solib): Save filename and free it later, after
bfd_close, since bfd_close may reference it. Use bfd_get_filename
to access the field.
* config/convex/xm-convex.h (LONG_LONG): Replace with
CC_HAS_LONG_LONG. Add define for PRINTF_HAS_LONG_LONG.
* doc/gdbint.texinfo (LONG_LONG): Replace with CC_HAS_LONG_LONG.
Add PRINTF_HAS_LONG_LONG references.
implied that "no side effects" was nonfunctional.
* eval.c (evaluate_subexp): Add BINOP_CONCAT case to deal with
character string and bitstring concatenation.
* expprint.c (dump_expression): Add case for BINOP_CONCAT.
* expression.h (exp_opcode): Add BINOP_CONCAT.
* gdbtypes.h (type_code): Add TYPE_CODE_BITSTRING.
* language.c (string_type): Add function to determine if a type
is a string type.
* language.c (binop_type_check): Add case for BINOP_CONCAT.
* valarith.c (value_concat): New function to concatenate two
values, such as character strings or bitstrings.
* valops.c (value_string): Remove error stub and implement
function body.
* value.h (value_concat): Add prototype.
**** start-sanitize-chill ****
* ch-exp.y (operand_3): Add actions for SLASH_SLASH (//).
* ch-exp.y (yylex): Recognize SLASH_SLASH.
* ch-lang.c (chill_op_print_tab): Add SLASH_SLASH (//) as
BINOP_CONCAT.
**** end-sanitize-chill ****
constants with an explicit null byte terminator. OP_STRING is
now used for real string types.
* c-lang.c (builtin_type_*): Move declarations to lang.c since
they are used by all languages.
* c-lang.c (_initialize_c_language): Move initializations of
builtin_type_* to lang.c.
* c-typeprint.c (c_type_print_varspec_prefix,
c_type_print_varspec_suffix): TYPE_CODE_PASCAL_ARRAY renamed
to TYPE_CODE_STRING.
* c-valprint.c (c_val_print): Change the way character arrays
are printed as strings to be consistent with the way strings
are printed when pointer-to-char types are dereferenced.
Remove test of print_max before calling val_print_string, which
now does it's own test.
* eval.c (evaluate_subexp): Add case for OP_ARRAY.
* expprint.c (print_subexp, dump_expression): Add case for OP_ARRAY.
* expression.h (enum exp_opcode): Add OP_ARRAY and document.
* gdbtypes.c (builtin_type_*): Add declarations moved from
c-lang.c.
* gdbtypes.c (create_string_type): New function to create real
string types.
* gdbtypes.c (recursive_dump_type): TYPE_CODE_PASCAL_ARRAY
renamed to TYPE_CODE_STRING.
* gdbtypes.c (_initialize_gdbtypes): Add initializations of
builtin_type_* types moved from c-lang.c.
* gdbtypes.h (enum type_code): TYPE_CODE_PASCAL_ARRAY renamed
to TYPE_CODE_STRING.
* gdbtypes.h (builtin_type_string): Add extern declaration.
* gdbtypes.h (create_string_type): Add prototype.
* m2-lang.c (m2_create_fundamental_type): TYPE_CODE_PASCAL_ARRAY
renamed to TYPE_CODE_STRING.
* m88k-tdep.c (pushed_size): TYPE_CODE_PASCAL_ARRAY renamed to
TYPE_CODE_STRING.
* mipsread.c (_initialize_mipsread): TYPE_CODE_PASCAL_ARRAY
renamed to TYPE_CODE_STRING.
* parse.c (length_of_subexp, prefixify_subexp): Add case for
OP_ARRAY.
* printcmd.c (print_formatted): Recognize TYPE_CODE_STRING.
* typeprint.c (print_type_scalar): TYPE_CODE_PASCAL_ARRAY renamed
to TYPE_CODE_STRING.
* valops.c (allocate_space_in_inferior): New function and
prototype, using code ripped out of value_string.
* valops.c (value_string): Rewritten to use new function
allocate_space_in_inferior, but temporarily disabled until some
other support is in place.
* valops.c (value_array): New function to create array constants.
* valprint.c (val_print_string): Add comment to document use,
complete rewrite to fix several small buglets.
* value.h (value_array): Add prototype.
* value.h (val_print_string): Change prototype to match rewrite.
**** start-sanitize-chill ****
* ch-valprint.c (chill_val_print): Add case for TYPE_CODE_STRING.
* ch-exp.y (match_character_literal): Disable recognition of
control sequence form of character literals and document why.
**** end-sanitize-chill ****
comment.
* gdbtypes.h, gdbtypes.c: Use const in decl of
cplus_struct_default, now that pa-gas assembler has been fixed.
* hppah-nat.c: Formatting.
* hppah-tdep.c: Remove lots of useless externs for variables we
don't use.
* (find_unwind_entry): Speed up by using binary search, and a one
entry cache.
* (rp_saved): New routine to see what unwind info says about RP
being saved on the stack frame.
* (frame_saved_pc): Look for prologue to see if we need to
examine the stack for the saved RP or not.
* (init_extra_frame_info): Check for prologue, instead of
framesize to determine if we are frameless or not.
* (frame_chain_valid): Stop backtraces when we run into _start.
* (push_dummy_frame): Reformat to make more readable.
* (find_dummy_frame_regs): ditto.
* (hp_pop_frame): ditto.
* (hp_restore_pc_queue): small cleanup.
* (hp_push_arguments): ditto.
* (pa_do_registers_info): ditto.
* (skip_prologue): New routine created from SKIP_PROLOGUE macro.
* tm-hppa.h: Move contents of SKIP_PROLOGUE into hppah-tdep.c.
* Define FRAME_CHAIN_VALID.
* Turn on BELIEVE_PCC_PROMOTION so that we can access char args
passed to functions.
* paread.c (pa_symtab_read): Use new bfd conventions for
accessing linker symbol table.
* (pa_symfile_init): Access embedded STAB info via BFD section
mechanism and related macros.
* defs.h (demangle_and_match): Remove prototype.
* dwarfread.c (STREQ, STREQN): Remove macros, replaced with STREQ
and STREQN defined in defs.h.
* dwarfread.c (set_cu_language): For completely unknown languages,
try to deduce the language from the filename. Retain behavior
that for known languages we don't know how to handle, we use
language_unknown.
* dwarfread.c (enum_type, symthesize_typedef): Initialize language
and demangled name fields in symbol.
* dwarfread.c, mipsread.c, partial-stab.h: For all usages of
ADD_PSYMBOL_TO_LIST, add language and objfile parameters.
* dwarfread.c (new_symbol): Attempt to demangle C++ symbol names
and cache the results in SYMBOL_DEMANGLED_NAME for the symbol.
* elfread.c (STREQ): Remove macro, use STREQ defined in defs.h.
Replace usages throughout.
* elfread.c (demangle.h): Include.
* elfread.c (record_minimal_symbol): Remove prototype and function.
* gdbtypes.h, symtab.h (B_SET, B_CLR, B_TST, B_TYPE, B_BYTES,
B_CLRALL): Moved from symtab.h to gdbtypes.h.
* infcmd.c (jump_command): Remove code to demangle name and add
it to a cleanup list. Now just use SYMBOL_DEMANGLED_NAME.
* minsyms.c (demangle.h): Include.
* minsyms.c (lookup_minimal_symbol): Indent comment to match code.
* minsyms.c (install_minimal_symbols): Attempt to demangle symbol
names as C++ names, and cache them in SYMBOL_DEMANGLED_NAME.
* mipsread.c (psymtab_language): Add static variable.
* stabsread.c (demangle.h): Include.
* stabsread.c (define_symbol): Attempt to demangle C++ symbol
names and cache them in the SYMBOL_DEMANGLED_NAME field.
* stack.c (return_command): Remove explicit demangling of name
and use of cleanups. Just use SYMBOL_DEMANGLED_NAME.
* symfile.c (demangle.h): Include.
* symfile.c (add_psymbol_to_list, add_psymbol_addr_to_list): Fix
to match macros in symfile.h and allow them to be compiled
if INLINE_ADD_PSYMBOL is not true.
* symfile.h (INLINE_ADD_PSYMBOL): Default to true if not set.
* symfile.h (ADD_PSYMBOL_*): Add language and objfile parameters.
Add code to demangle and cache C++ symbol names. Use macro form
if INLINE_ADD_PSYMBOL is true, otherwise use C function form.
* symmisc.c (add_psymbol_to_list, add_psymbol_addr_to_list):
Remove, also defined in symfile.c, which we already fixed.
* symtab.c (expensive_mangler): Remove prototype and function.
* symtab.c (find_methods): Remove physnames parameter and fix
prototype to match.
* symtab.c (completion_list_add_symbol): Name changed to
completion_list_add_name.
* symtab.c (COMPLETION_LIST_ADD_SYMBOL): New macro, adds both
the normal symbol name and the cached C++ demangled name.
* symtab.c (lookup_demangled_partial_symbol,
lookup_demangled_block_symbol): Remove prototypes and functions.
* symtab.c (lookup_symbol): Remove use of expensive_mangler,
use lookup_block_symbol instead of lookup_demangled_block_symbol.
Remove code to try demangling names and matching them.
* symtab.c (lookup_partial_symbol, lookup_block_symbol):
Fix to try matching the cached demangled name if no match is
found using the regular symbol name.
* symtab.c (find_methods): Remove unused physnames array.
* symtab.c (name_match, NAME_MATCH): Remove function and macro,
replaced with SYMBOL_MATCHES_REGEXP from symtab.h.
* symtab.c (completion_list_add_symbol): Rewrite to use cached
C++ demangled symbol names.
* symtab.h: Much reformatting of structures and such to add
whitespace to make them more readable, and make them more
consistent with other gdb structure definitions.
* symtab.h (general_symbol_info): New struct containing fields
common to all symbols.
* symtab.h (SYMBOL_LANGUAGE, SYMBOL_DEMANGLED_NAME,
SYMBOL_SOURCE_NAME, SYMBOL_LINKAGE_NAME, SYMBOL_MATCHES_NAME,
SYMBOL_MATCHES_REGEXP, MSYMBOL_INFO, MSYMBOL_TYPE): New macros.
* symtab. (struct minimal_symbol, struct partial_symbol, struct
symbol): Use general_symbol_info struct.
* utils.c (demangle_and_match): Remove, no longer used.
* valops.c (demangle.h): Include.
* xcoffexec.c (eq): Remove macro, replace usages with STREQ.
* blockframe.c, breakpoint.c, c-exp.y, c-valprint.c, dbxread.c,
infcmd.c, m2-exp.y, minsyms.c, objfiles.h, solib.c, stack.c,
symmisc.c, symtab.c, valops.c: Replace references to minimal
symbol fields with appropriate macros.
* breakpoint.c, buildsym.c, c-exp.y, c-typeprint.c, c-valprint.c,
coffread.c, command.c, convex-tdep.c, cp-valprint.c, dbxread.c,
demangle.c, elfread.c, energize.c, environ.c, exec.c,
gdbtypes.c, i960-tdep.c, infrun.c, infrun-hacked.c, language.c,
main.c, minsyms.c, mipsread.c, partial-stab.h, remote-es1800.c,
remote-nindy.c, remote-udi.c, rs6000-tdep.c, solib.c, source.c,
sparc-pinsn.c, stabsread.c, standalone.c, state.c, stuff.c,
symfile.c, symmisc.c, symtab.c, symtab.h, tm-sysv4.h,
tm-ultra3.h, values.c, xcoffexec.c, xcoffread.c: Replace strcmp
and strncmp usages with STREQ, STREQN, or STRCMP as appropriate.
* breakpoint.c, buildsym.c, c-typeprint.c, expprint.c, findvar.c,
mipsread.c, printcmd.c, source.c, stabsread.c, stack.c,
symmisc.c, tm-29k.h, valops.c, values.c: Replace SYMBOL_NAME
references with SYMBOL_SOURCE_NAME or SYMBOL_LINKAGE_NAME as
appropriate.
* buildsym.c (start_subfile, patch_subfile_names): Default the
source language to what can be deduced from the filename.
* buildsym.c (end_symtab): Update the source language in the
allocated symtab to match what we have been using.
* buildsym.h (struct subfile): Add a language field.
* c-typeprint.c (c_print_type): Remove code to do explicit
demangling.
* dbxread.c (psymtab_language): Add static variable.
* dbxread.c (start_psymtab): Initialize psymtab_language using
deduce_language_from_filename.
* Makefile.in (SFILES_MAINDIR): Add typeprint.c, c-typeprint.c,
m2-typeprint.c, c-valprint.c cp-valprint.c m2-valprint.c.
* Makefile.in (HFILES): Add valprint.h.
* Makefile.in (OBS): Add typeprint.o, c-typeprint.o,
m2-typeprint.o, c-valprint.o, cp-valprint.o m2-valprint.o.
* typeprint.c, typeprint.h: New files for language independent
type printing functions.
* c-typeprint.c, m2-typeprint.c: New files for language dependent
type printing functions and definitions.
* valprint.h: New include file for language independent value
printing definitions.
* c-valprint.c, cp-valprint.c, m2-valprint.c: New files for language
dependent value printing functions.
* c-exp.y (production ptype): Add range_type variable and use new
create_range_type function.
* c-exp.y (tokentab2, tokentab3), c-lang.c (c_op_print_tab),
infcmd.c (path_var_name), language.c (unk_op_print_tab),
m2-lang.c (m2_op_print_tab): Change from ANSI-obsolescent
"const static" to ANSI-conformant "static const".
* c-exp.y (c_create_fundamental_type): Remove unused nbytes.
* c-exp.y (c_language_defn, cplus_language_defn): Add c_print_type,
and c_val_print.
* c-lang.h (c_print_type, c_val_print): Add prototypes.
* coffread.c (decode_type): Add range_type variable and call to
new create_range_type function.
* complaints.c (complain): Remove unused val variable.
* complaints.c (_initialize_complaints): Make it void.
* convex-tdep.c (value_of_trapped_internalvar): Add range_type
variable and call new create_range_type function.
* defs.h (enum val_prettyprint): Move enum from value.h to here
so we can avoid having to include value.h just for prototypes that
need the enum (thanks ANSI).
* dwarfread.c (struct_type): Local anonymous_size variable is
only used if !BITS_BIG_ENDIAN.
* dwarfread.c (decode_subscript_data_item): Add rangetype
variable and call new create_range_type function.
* elfread.c (elf_symfile_read): Remove unused dbx and text_sect
variables.
* eval.c (evaluate_subexp): Remove unused local variable name
and the statement with no side effects that initializes it.
* expprint.c (print_subexp): Change local_printstr to
LA_PRINT_STRING.
* gdbtypes.c (create_range_type): New function that creates
a range type using code fragments from object file readers as
an example of what has to be initialized.
* gdbtypes.c (create_array_type): Removed index_type, low_bound,
and high_bound parameters, replaced with a single range_type
parameter. Change function body to use passed in range_type
rather than handcrafting one.
* gdbtypes.h (create_range_type): Add prototype.
* gdbtypes.h (create_array_type): Change prototype parameters.
* infrun.c (normal_stop): Remove unused local variables tem and c.
* infrun.c (hook_stop_stub): Return 0 rather than random value.
* language.c (unk_lang_print_type, unk_lang_val_print): Add
stub functions that call error if called.
* language.c (unknown_language_defn, auto_language_defn,
local_language_defn): Add initializers unk_lang_print_type and
unk_lang_val_print.
* language.h (struct language_defn): Reformat for larger
comments, add la_print_type and la_val_print members. Add
LA_PRINT_TYPE and LA_VAL_PRINT macros. Change local_printchar
to LA_PRINT_CHAR and local_printstr to LA_PRINT_STRING.
* m2-lang.c (m2_create_fundamental_type): Remove unused local
variable nbytes.
* m2-lang.c (m2_language_defn): Add initializers m2_print_type
and m2_val_print.
* m2-lang.h (m2_print_type, m2_val_print): Add prototypes.
* main.c (execute_command): Remove unused local variable cmdlines.
* main.c (echo_command), stabsread.c (read_type), printcmd.c
(clear_displays), symmisc.c (block_depth), values.c
(clear_value_history):
Make testing of truth value of assignment result explicit.
* mipsread.c (upgrade_type): Update FIXME to include future use
of create_range_type.
* printcmd.c (ptype_command, ptype_eval, whatis_command,
whatis_exp, maintenance_print_type): Move prototypes and functions
to new typeprint.c.
* printcmd.c (_initialize_printcmd): Move add_com calls for
ptype_command and whatis_command to new typeprint.c.
* ser-bsd.c (serial_open): Remove unused variable sgttyb.
* source.c (find_source_lines): Local variable c only used
when LSEEK_NOT_LINEAR is defined.
* stabsread.c (read_array_type): Use new create_range_type
function.
* stabsread.c (read_range_type): Add new index_type variable and
call new create_range_type function rather than handcrafting
range types.
* symmisc.c (type_print_1): Change usages to LA_PRINT_TYPE.
* symtab.c (typedef_print usages): Use c_typedef_print, renamed.
* symtab.c (type_print_base usages): Use c_type_print_base.
* symtab.c (type_print_varspec_prefix usages): Use
c_type_print_varspec_prefix.
* symtab.c (type_print_method_args usages): Use
cp_type_print_method_args.
* valprint.c: Completely ripped apart and the fragments used
to create c-valprint.c, cp-valprint.c, m2-valprint.c, and
valprint.h. Remaining stuff is language independent.
* value.h (struct fn_field): Forward declare for prototypes.
* value.h (type_print_1): Remove prototype.
* value.h (enum val_prettyprint): Moved to defs.h.
* value.h (typedef_print): Prototype renamed to c_typedef_print.
* value.h (baseclass_offset): Add prototype.
**** start-sanitize-chill ****
* Makefile.in (SFILES_MAINDIR): Add ch-typeprint.c, ch-valprint.c.
* Makefile.in (OBS): Add ch-typeprint.o, ch-valprint.o.
* ch-typeprint.c: New file for language dependent type printing.
* ch-valprint.c: New file for language dependent value printing.
* ch-exp.y (parse_number): Remove prototype and stub function.
* ch-exp.y (decode_integer_literal): Removed unused digits and
temp variables.
* ch-exp.y (convert_float): Completely ifdef out for now.
* ch-exp.y (tokentab2, tokentab3, tokentab4, tokentab5),
ch-lang.c (chill_op_print_tab):
Change from ANSI-obsolescent "const static" to ANSI-conformant
"static const".
* ch-exp.y (yylex): Add unhandled storage class enumeration
literals to switch statement for completeness.
* ch-lang.c (chill_create_fundamental_types): Remove unused
nbytes variable. Change dummy type to 2 bytes to match int.
Handle FT_VOID types gratuituously added to chill DWARF by
compiler. Change FT_CHAR case to generate an TYPE_CODE_CHAR
type rather than a one byte TYPE_CODE_INT type.
* ch-lang.c (chill_language_defn): Add chill_print_type and
chill_val_print.
* ch-lang.h (chill_print_type, chill_val_print): Add prototypes.
**** end-sanitize-chill ****
a optional type to decorate as an array type, the type of the
index, and the bounds of the array. Records this additional info
in the array type for use with languages with nonzero array
bounds.
* gdbtypes.h (enum type_code): Update comment for TYPE_CODE_ARRAY
to note that arrays may have bounds.
* gdbtypes.h (create_array_type): Update prototype.
* c-exp.y (ptype production): Adjust for new create_array_type
calling conventions.
* coffread.c (decode_type): Call create_array_type rather than
handcrafting array types.
* convex-tdep.c (value_type): Remove, now use create_array_type.
* convex-tdep.c (value_of_trapped_internalvar): Convert calls to
vector_type into calls to create_array_type.
* dwarfread.c (decode_subscr_data): Name changed to
decode_subscript_data_item throughout.
* dwarfread.c (decode_subscript_data_item): Rewrite to use
create_array_type. Now records index type and range as well.
* dwarfread.c (dwarf_read_array_type): Rewrite as part of
change to use create_array_type.
* dwarfread.c (read_subroutine_type): Test existing user defined
types before decorating them, to ensure they are blank, and
complain about it if they are not.
* dwarfread.c (decode_fund_type): For unrecognized types, always
return some valid type (type integer). If the unrecognized type
cannot be an implementation defined type, complain as well.
* m88k-tdep.c (pushed_size): Update comment for TYPE_CODE_ARRAY.
* m88k-tdep.c (store_param): Update comment for TYPE_CODE_ARRAY.
* mipsread.c (upgrade_type): Add FIXME comment that code to
handcraft arrays should be replaced with call to create_array_type.
* stabsread.c (read_array_type): Replace code to handcraft
array types with call to create_array_type.
* valprint.c (type_print_varspec_prefix): Minor formatting
change, join lines that don't need to be split.
decl for attach_flag, it now lives in inferior.h.
* hppa-pinsn.c: Reformat opcode tables. Add function prototypes.
Make most functions static.
* hppah-nat.c: General cleanups, remove BSD specific code (since
that all lives in hppab-nat.c).
* hppah-tdep.c (frame_chain_valid), tm-hppa.h (FRAME_CHAIN):
Change sense of test against inside_entry_file(). This fix is
from U. of Utah.
* tm-hppa.h (PUSH_DUMMY_FRAME, POP_FRAME): Use char * for 2nd arg
to read/write_register_bytes().
language specific fundamental types for C.
* m2-exp.y (m2_create_fundamental_type): New function to create
language specific fundamental types for Modula 2.
* c-exp.y (c_language_defn, cplus_language_defn): Add
c_create_fundamental_type to language struct initializers.
* m2-exp.y (m2_language_defn): Add m2_create_fundamental_type
to language struct initializers.
* dwarfread.c (expression.h, language.h): Include.
* dwarfread.c (ftypes): New array to hold fundamental types
for current compilation unit.
* dwarfread.c (cu_language_defn): New pointer to language
struct for language of current compilation unit.
* dwarfread.c (dwarf_fundamental_type): New function to
create/lookup fundamental types.
* dwarfread.c (set_cu_language): Initialize cu_language_defn.
* dwarfread.c (throughout): Replace lookup_fundamental_type
with dwarf_fundamental_type.
* dwarfread.c (read_file_scope): Zero out ftypes for each new
compilation unit (may be different language or different objfile).
* gdbtypes.c (lookup_fundamental_type): Move actual type
creations into language specific fundamental type creation
functions and call via create_fundamental_type. Add comment
about this function being obsolescent.
* gdbtypes.h (FT_BYTE, FT_UNSIGNED_BYTE): New types, true byte
sized signed and unsigned integers.
* gdbtypes.h (FT_NUM_MEMBERS): Increment, new types added.
* language.c (language_def): New function to lookup a language
struct given it's enumeration.
* language.h (struct language_defn): Add la_fund_type, a pointer
to a function that creates fundamental types for this language.
* language.h (create_fundamental_type): New macro to create
fundamental types based on the current language.
* language.h (language_def): Add prototype.
* language.c (unk_lang_create_fundamental_type): New function
for initializing language structs, calls error if called.
* language.c (unk_language_defn, auto_language_defn,
local_language_defn): Use unk_lang_create_fundamental_type.
**** start-sanitize-chill ****
ch-exp.y (chill_create_fundamental_type): New function.
ch-exp.y (chill_language_defn): Add chill_create_fundamental_type.
ch-exp.y (_initialize_chill_exp): BOOL types are only one byte.
**** end-sanitize-chill ****
local formatting information.
* language.h (language_defn): Replace individual format info
with language_format_info structs.
* language.h (local_*_format, local_*_format_prefix,
local_*_format_specifier, local_*_format_suffix): New macros
for binary/octal/decimal/hex formats to access info elements.
* c-exp.y (c_language_defn): Update for new format handling.
* m2-exp.y (m2_language_defn): Update for new format handling.
* dbxread.c (language.h): Include for partial-stab.h use.
* defs.h (local_hex_format, local_hex_format_custom,
local_hex_string, local_hex_string_custom): Move to language.h.
* language.c (local_hex_format_custom, local_hex_string,
local_hex_string_custom, local_octal_format_custom): Use new
format handling.
* language.c (unknown_language_defn, auto_language_defn,
local_language_defn): Update for new format handling.
* printcmd.c (print_scalar_formatted): Use new macros
to access decimal and binary format info for printing.
**** start-sanitize-chill ****
* c-exp.y (chill_language_defn): Update for new format handling.
* ch-exp.y (CHARACTER_LITERAL): Add support to yylex.
* ch-exp.y (match_integer_literal): Add function.
* ch-exp.y (builtin_type_chill_char): Add definition.
* gdbtypes.h (builtin_type_chill_char): Add declaration.
**** end-sanitize-chill ****
language for current compilation unit being processed.
* dwarfread.c (set_cu_language): New local function to decode
and record language for current compilation unit being processed.
* dwarfread.c (synthesize_typedef): Add local function to
synthesize a typedef for C++ classes, structs, unions, and
enumerations.
* dwarfread.c (read_structure_scope): Synthesize typedefs
for C++ classes, structs and unions.
* dwarfread.c (read_enumeration): Synthesize typedefs for
C++ enumerations.
* dwarfread.c (read_file_scope): Call set_cu_language to
record language for current compilation unit. After symtab
is built, save this recorded language.
* dwarfread.c (process_dies, add_partial_symbol, struct_type,
scan_partial_symbols, new_symbol): Recognize TAG_class_type.
* dwarfread.c (add_partial_symbol): Synthesize partial symbol
typedefs for C++ classes, structs, unions, and enumerations.
* dwarfread.c (scan_compilation_units): Call set_cu_language
to record language for current compilation unit.
* dwarfread.c (scan_partial_symbols): Call add_enum_psymbol here
for TAG_enumeration_types, rather than in add_partial_symbol.
* dwarfread.c (add_partial_symbol): Combine TAG_enumeration_type
case with class, struct, and union type cases, now that they are
the same. Remove tests for non-NULL name attributes, now done
by callers.
* gdbtypes.h (TYPE_CODE_CLASS): Add type for C++ classes, but
treat as alias for TYPE_CODE_STRUCT for now.
types.
* stabsread.c (read_array_type, read_range_type, define_symbol):
Call alloc_type to alloc new types.
* stabsread.c (define_symbol): Move dbl_valu symbol field data
from type_obstack to symbol_obstack.
* stabsread.c (define_symbol): Move typedef_sym from type_obstack
to symbol_obstack.
* gdbtypes.h (TYPE_ALLOC): New macro to allocate space for data
associated with a type, using the same mechanism as was used to
allocate space for the type structure itself.
* coffread.c (patch_type, coff_read_struct_type,
coff_read_enum_type): Use TYPE_ALLOC.
* dwarfread.c (struct_type): Use TYPE_ALLOC.
* gdbtypes.c (create_array_type, check_stub_method,
allocate_cplus_struct_type): Use TYPE_ALLOC.
* mipsread.c (parse_symbol, parse_type): Use TYPE_ALLOC.
* stabsread.c (read_struct_type, read_array_type, read_enum_type,
read_range_type): Use TYPE_ALLOC.
at_import, at_frame_base.
* dwarfread.c (completedieinfo): Remove cases for obsolete
AT_visibility, AT_import, and AT_frame_base attributes.
* breakpoint.h (BREAKPOINT_MAX): Increase from 10 to 16 for
i860, which can keep up to 4 shadow breakpoints.
* tm-stratus.h (USG): Define.
* tm-stratus.h (TARGET_BYTE_ORDER): Define to BIG_ENDIAN.
* xm-stratus.h (HOST_BYTE_ORDER): Define to BIG_ENDIAN.
* xm-stratus.h (xm-sysv4.h): Include, name changed from xm-svr4.h.
* xm-stratus.h (NO_JOB_CONTROL): Remove define.
* config/stratus.mt (TDEPFILES): Include files available from
stratus that are not yet in release pending receipt of paperwork
at FSF.
* buildsym.c (finish_block): Minor code format change.
* gdbtypes.h (struct field): Add to comments.
* gdbtypes.h (virtual_field_bits): Add to comments.
* objfiles.c (allocate_objfile): Change obstack interface to
match FSF merging efforts.
with alloca, to avoid using random values from stack later on.
* defs.h (fprintf_filtered): Add prototype.
* gdbtypes.c (check_stub_method): Fix misleading comments.
* gdbtypes.c (print_arg_types, dump_fn_fieldlists): New maint
support functions.
* gdbtypes.c (print_cplus_stuff, recursive_dump_type): Many
small changes to maint support functions.
* gdbtypes.h (cplus_struct_type): Reorganize member ordering
for some fields and expand comments.
* objfiles.c (allocate_objfile): Use new obstack_alloc_arg
macro and track change to obstack_full_begin macro.
* utils.c (fprintfi_filtered): New function.
* valprint.c (type_print_base): Fixup field printing to not
print extraneous lines and not print bogus "no data fields"
messages for C++ classes with no data members. Also use new
fprintfi_filtered function.
* c-exp.y: Call lookup_signed_typename() after seeing
"signed". This handles "signed char" correctly.
* c-exp.y: Recognize (but ignore) 'const' and 'volatile'
keywords before a type specifier.
to create new blank types, instead of handcrafting them.
* defs.h (printfi_filtered): Add prototype.
* utils.c (printfi_filtered): New function.
* gdbtypes.c (recursive_dump_type): Use printfi_filtered to
to simplify the code. Other cleanups.
* gdbtypes.c (check_stub_method): Demangle using DMGL_ANSI.
* gdbtypes.h (struct cplus_struct_type): Add comments describing
use of various fields.
* gdbtypes.c (print_bit_vector, print_cplus_stuff): New functions.
* c-exp.y (%token): Add CLASS as a token for C++, add grammar
production that currently treats it exactly the same as STRUCT.
* c-exp.y (yylex): Recognize "class" as token CLASS.
* symtab.c (gdb_mangle_name): Rewrite to match current g++ stabs.
* symtab.c (decode_line_1): Fix to pass quoted args on down to
general symbol handling code. Call cplus_mangle_opname with
DMGL_ANSI.
* symtab.c (decode_line_2): Print demangled function names in
breakpoint menus, instead of just file and line number.
* symtab.c (name_match): Call cplus_demangle with DMGL_ANSI.
* valprint.c (type_print_base): Print "class" for C++ classes,
rather than "struct". Print section labels for public, protected
and private members of C++ classes.
* values.c: Include demangle.h.
* values.c (value_headof): Call cplus_demangle with DMGL_ANSI.
* Makefile.in (SFILES_MAINDIR): Add new file maint.c.
* Makefile.in (OBS): Add new file maint.o.
* defs.h (command_class): Add class_maintenance.
* defs.h (MAINTENANCE_CMDS): Default to including maintenance
commands. Allow for them (and other nonessential parts of gdb)
to be selectively left out under special circumstances.
* gdbtypes.c (recursive_dump_type): New function; supports
maintenance print-type command.
* gdbtypes.h (recursive_dump_type, maintenance_print_type):
Add prototypes.
* main.c (maintenancelist, maintenanceinfolist): Add maintenance
command lists.
* main.c (initialize_cmd_lists): Eliminate unnecessary casts on
initializers. Add initializations for setprintlist, showprintlist,
setchecklist, showchecklist, maintenancelist, and maintenanceinfolist.
* printcmd.c (maintenance_print_type): New maintenance cmd.
* valprint.c (setprintlist, showprintlist): Move to main.c, as
implied by comment that all cmd lists are owned by main.c.
* infcmd.c (unsetlist): Move to main.c, as implied by comment
that all cmd lists are owned by main.c.
* language.c (setchecklist, showchecklist): Move to main.c, as
implied by comment that all cmd lists are owned by main.c
* breakpoint.c (enablelist, enablebreaklist, disablelist, cmdlist,
deletelist): Remove redundant declarations (also in gdbcmd.h).
* symmisc.c (printsyms_command): Now maintenance_print_symbols.
* symmisc.c (printmsyms_command): Now maintenance_print_msymbols.
* symmisc.c (printpsyms_command): Now maintenance_print_psymbols.
* symmisc.c (printobjfiles_command): Now maintenance_print_objfiles.
* symtab.h (maintenance_print_symbols, maintenance_print_psymbols,
maintenance_print_msymbols, maintenance_print_objfiles):
Add prototypes.
* symmisc.c (printsyms_command, printpsyms_command,
printmsyms_command, printobjfiles_command): Removed from
_initialize_symmisc.
* main.c (dump_me_command): Moved to maint.c and renamed to
maintenance_dump_me.
* breakpoint.c (all_breakpoints_info): Rename to
maintenance_info_breakpoints.
* breakpoint.c (_initialize_breakpoint): Convert add_info of
all_breakpoints_info to add maintenance_info_breakpoints to the
maintenanceinfolist instead.
main.c (initialize_main): Set up maintenance class commands.
* Makefile.in (DEMANGLE_OPTS): Add, default to -Dnounderscore.
* configure.in: Simplify ncr3000 gdb_host logic, add gdb_target.
* dwarfread.c (struct_type): Apply fix from Peggy Fieland for
proper handling of bit fields.
* gdbtypes.h (struct type): Clarify use of field.bitpos.
* symtab.h: Fix couple of misspellings in comments.
* value.h (struct value): Clarify use of bitpos.
* value.h (unpack_field_as_long): Change prototype, returns
LONGEST.
* values.c (unpack_field_as_long): Change return type to LONGEST,
sign extend unpacked fields that are signed, other rewriting.
* config/ncr3000.mt: New target config file.
functions which handle overwriting of forward-referenced types
for stabs file reading.
(lookup_{reference,pointer,function}_type): These just call
the make_*_type functions with a null storage alloc parameter.
* gdbtypes.h (make_{reference,pointer,function}_type): Declare.
* xcoffread.c (smash_to_pointer_type): Remove, no longer used.
* buildsym.c (dbx_lookup_type): Zero result for (-1,-1) arg.
(dbx_alloc_type): Make it easier to understand. No funct change.
(define_symbol: 't'): Don't put the typedef name into the name of
the struct, union, or enum. Bugfix.
(read_type: '*', '&', 'f'): Add comments. Use make_XXX_type
routines to properly handle overwriting preallocated types so that
forward references will work.
(read_enum_type): Force enum values to file scope, due to bug in
Sun compiler output. FIXME, fix later.
Remove unused header_file_prev_index mechanism. It was already
obsolete in gdb-3.5. These comments appeared in 3.5:
/* This code was used before I knew about the instance codes.
My first hypothesis is that it is not necessary now
that instance codes are handled. */
* dbxread.c (add_new_header_file): Remove header_file_prev_index.
* buildsym.h: Remove it and prev_index that saves it.
* buildsym.c (push_subfile, pop_subfile, start_symtab): Remove it.
* solib.c (special_symbol_handling): When called from core files,
must set up debug_addr. Don't print error messages, just return.
* symmisc.c (print_symbol): Less ascii diarrhea for enums, please.
and use it to decide when to print the actual type name rather than
trying to invent the name of a fundamental type. This clears up the
confusion between int/long when they are the same sizes, removes one
obstacle to multi-language support (previously valprint.c thought
everything was a C type), and allows gdb to support distinctions between
explicitly and implicitly signed types when the compiler supports such
distinction in the debug output (as does every ANSI compiler I tested
except for gcc).
