Remove a comment in compile/compile-cplus-types.c

The include sorter can't handle multi-line comments on the same line
as a #include.  This patch removes the only such comment.

In general I think these sorts of comments do not provide much value:
more often than not, I find that the comment is obsolete in one way or
another, and so the include sorter removes them in most cases.

gdb/ChangeLog
2019-01-22  Tom Tromey  <tom@tromey.com>

* compile/compile-cplus-types.c: Remove a comment by #include.

Include compile-internal.h in gcc-c-plugin.h

gcc-c-plugin.h refers to some types defined in compile-internal.h.
This patch changes the former to include the latter.

gdb/ChangeLog
2019-01-22  Tom Tromey  <tom@tromey.com>

* compile/gcc-c-plugin.h: Include compile-internal.h.

Automatic date update in version.in

Remove the EXTERN define from stabsread.h

This removes the EXTERN define from stabsread.h.  This is the only
spot that still uses this approach, and it interfered with sorting the
includes in stabsread.c.

Tested by rebuilding.

gdb/ChangeLog
2019-01-22  Tom Tromey  <tom@tromey.com>

* stabsread.c (EXTERN): Do not define.
(symnum, next_symbol_text_func, processing_gcc_compilation)
(within_function, global_sym_chain, global_stabs)
(previous_stab_code, this_object_header_files)
(n_this_object_header_files)
(n_allocated_this_object_header_files): Define.
* stabsread.h (EXTERN): Never define.  Use "extern".

Automatic date update in version.in

Fix leak in event-top.c history expansion

E.g. in gdb.base/default.exp, valgrind detects leaks such as
==17663== 1,438 bytes in 101 blocks are definitely lost in loss record 2,804 of 2,884
==17663==    at 0x4C2BE6D: malloc (vg_replace_malloc.c:309)
==17663==    by 0x418A17: xmalloc (common-utils.c:44)
==17663==    by 0x4E6F19C: history_expand (histexpand.c:1061)
==17663==    by 0x4B4490: handle_line_of_input(buffer*, char const*, int, char const*) (event-top.c:685)
==17663==    by 0x4B4562: command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) (event-top.c:753)
...

Fix the leak by using an unique_xmalloc_ptr for history_value.

gdb/ChangeLog
2019-01-22  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* event-top.c (handle_line_of_input): use unique_xmalloc_ptr for
history_value.

Remove duplicate or commented-out #includes

I wrote a little script to detect duplicate or commented-out #includes
and ran it on gdb.  This patch is the result.  Tested by rebuilding.

gdb/ChangeLog
2019-01-21  Tom Tromey  <tom@tromey.com>

* ui-out.c: Fix includes.
* tui/tui-source.c: Fix includes.
* target.c: Fix includes.
* remote.c: Fix includes.
* regcache.c: Fix includes.
* python/py-block.c: Fix includes.
* printcmd.c: Fix includes.
* or1k-tdep.c: Fix includes.
* mi/mi-main.c: Fix includes.
* m32r-tdep.c: Fix includes.
* csky-tdep.c: Fix includes.
* compile/compile-cplus-types.c: Fix includes.
* cli/cli-interp.c: Fix includes.

gdb/gdbserver/ChangeLog
2019-01-21  Tom Tromey  <tom@tromey.com>

* tracepoint.c: Fix includes.
* remote-utils.c: Fix includes.
* linux-x86-low.c: Fix includes.

gdb/stubs/ChangeLog
2019-01-21  Tom Tromey  <tom@tromey.com>

* ia64vms-stub.c: Fix includes.

Fix leak in splay-tree

Philippe Waroquiers noticed a memory leak in gdb, which he tracked
down to a bug in splay-tree.  splay_tree_remove does not call the
`delete_key' function when it removes the old node; but it should.

I looked at every splay tree in GCC and there is only one that passes
a non-NULL delete function -- the one in lto.c.  That file does not
call splay_tree_remove.  So, I think this is safe to check in.

I re-ran the LTO tests to double check.

libiberty/
* splay-tree.c (splay_tree_remove): Delete the key if necessary.

AArch64 AAPCS: Empty structs have non zero size in C++

When gdb.base/infcall-nested-structs.c is complied as C++, the compiler
will not pass structs containing empty structs via float arguments.
This is because structs in C++ have a minimum size of 1, causing padding
in the struct once compiled.  The AAPCS does not allow structs with
padding to be passed in float arguments.

Add padding checks to AArch64 and add C++ compile variant to the test.

Some of the tests fail on X86_64. This has been raised as bug gdb/24104.

gdb/ChangeLog:

* aarch64-tdep.c (aapcs_is_vfp_call_or_return_candidate_1): Check
for padding.

gdb/testsuite/ChangeLog:

* gdb.base/infcall-nested-structs.exp: Test C++ in addition to C.

Updated Brazilian Portuguese translation for the ld/ subdirectory.

Updated translations for various binutils subdirectories.

Fix spelling mistakes in BFD library.

PR 24108
bfd * elf32-nds32.c (nds32_relocate_section): Add space between words
in error message.
* elfnn-riscv.c (riscv_version_mismatch): Fix spelling mistake in
error message.
(riscv_i_or_e_p): Likewise.
(riscv_merge_arch_attr_info): Likewise.

ld * testsuite/ld-riscv-elf/attr-merge-arch-failed-01.d: Update
expected error message.

Testsuite: Ensure stack protection is off for GCC

Using -fstack-protector-strong will cause GDB to break on the wrong line
when placing a breakpoint on a function.  This is due to inadequate dwarf
line numbering, and is being tracked by the GCC bug
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88432

GCC (and Clang) provided by Debian/Ubuntu default to stack protector
being enabled.

Ensure that when running the GDB testsuite, stack protector is always
turned off for GCC 4.1.0 (when stack protector was added) and above.

Ensure that this does not cause infinite recursion due to
test_compiler_info having to compile a file itself.

Add a test to explicitly test breakpoints with various levels of stack
protection on both GCC and Clang, with xfail for the known errors.

Restore change in ovldbreak.exp which worked around the issue.

gdb/testsuite/ChangeLog:

2019-01-18  Alan Hayward  <alan.hayward@arm.com>

* gdb.base/stack-protector.c: New test.
* gdb.base/stack-protector.exp: New file.
* gdb.cp/ovldbreak.exp: Only allow a single break line.
* lib/gdb.exp (get_compiler_info): Use getting_compiler_info
option.
(gdb_compile): Remove stack protector for GCC and prevent
recursion.

Automatic date update in version.in

[MIPS] fix typo in mips_arch_choices.

opcodes/
* mips-dis.c (mips_arch_choices): Fix typo in
gs464, gs464e and gs264e descriptors.

Automatic date update in version.in

Change version to 2.32.51 and regenerate configure and pot files.

Add markers for 2.32 branch to NEWS and ChangeLog files.

Automatic date update in version.in

Automatic date update in version.in

Make minimal symbol range adapter a method on objfile

This removes class objfile_msymbols in favor of a method on the
objfile.

2019-01-16  Tom Tromey  <tom@tromey.com>

* objfiles.h (struct minimal_symbol_iterator): Rename.  Move
earlier.
(struct objfile) <msymbols_range>: Move from top level.
<msymbols>: New method.
(class objfile_msymbols): Remove.
* symtab.c (default_collect_symbol_completion_matches_break_on):
Update.
* symmisc.c (dump_msymbols): Update.
* stabsread.c (scan_file_globals): Update.
* objc-lang.c (info_selectors_command, info_classes_command)
(find_methods): Update.
* minsyms.c (find_solib_trampoline_target): Update.
* hppa-tdep.c (hppa_lookup_stub_minimal_symbol): Update.
* coffread.c (coff_symfile_read): Update.
* ada-lang.c (ada_lookup_simple_minsym)
(ada_collect_symbol_completion_matches): Update.

Simplify minsym iteration

This simplifies the minimal symbol iterator, by using
minimal_symbol_count and just doing a somewhat ordinary array-like
iteration.  array_view is nearly usable, except that it is more
convenient for this iterator to return pointers rather than
references.

gdb/ChangeLog
2019-01-16  Tom Tromey  <tom@tromey.com>

* objfiles.h (class objfile_msymbols) <iterator>: Change argument
type.  Remove no-argument constructor.
<iterator::operator++>: Simplify.
<begin>: Update.
<end>: Use minimal_symbol_count.

Make psymtab range adapter a method on objfile

This removes the objfile_psymtabs class in favor of a method on
objfile and on psymtab_storage.

2019-01-16  Tom Tromey  <tom@tromey.com>

* objfiles.h (struct objfile) <psymtabs>: New method.
(class objfile_psymtabs): Remove.
* psymtab.h (class psymtab_storage) <partial_symtab_range>: New
typedef.
<range>: New method.
(require_partial_symbols): Change return type.
* psymtab.c (require_partial_symbols)
(psym_expand_symtabs_matching): Update.
* mdebugread.c (parse_partial_symbols): Update.
* dbxread.c (dbx_end_psymtab): Update.

Add compunits range adapter to objfile

This removes the objfile_compunits range adapter in favor of using a
method on objfile.

2019-01-15  Tom Tromey  <tom@tromey.com>

* symtab.c (lookup_objfile_from_block)
(lookup_symbol_in_objfile_symtabs)
(basic_lookup_transparent_type_1, find_pc_sect_compunit_symtab)
(find_line_symtab, info_sources_command)
(default_collect_symbol_completion_matches_break_on)
(make_source_files_completion_list): Update.
* symmisc.c (print_objfile_statistics, dump_objfile)
(maintenance_print_symbols, maintenance_info_symtabs)
(maintenance_check_symtabs, maintenance_info_line_tables):
Update.
* source.c (select_source_symtab)
(forget_cached_source_info_for_objfile): Update.
* objfiles.h (class objfile_compunits): Remove.
(struct objfile) <compunits_range>: New typedef.
(compunits): New method.
* objfiles.c (objfile_relocate1): Update.
* mi/mi-cmd-file.c (mi_cmd_file_list_exec_source_files): Update.
* maint.c (count_symtabs_and_blocks): Update.
* linespec.c (iterate_over_all_matching_symtabs): Update.
* cp-support.c (add_symbol_overload_list_qualified): Update.
* coffread.c (coff_symtab_read): Update.
* ada-lang.c (add_nonlocal_symbols)
(ada_collect_symbol_completion_matches)
(ada_add_global_exceptions): Update.

Change all_objfiles_safe adapter to be a method on program_space

This changes the all_objfiles_safe range adapter to be a method on the
program space, and fixes up all the users.

gdb/ChangeLog
2019-01-15  Tom Tromey  <tom@tromey.com>

* progspace.h (program_space) <objfiles_safe_range>: New
typedef.
<objfiles_safe>: New method.
* objfiles.h (class all_objfiles_safe): Remove.
* objfiles.c (free_all_objfiles, objfile_purge_solibs): Update.
* jit.c (jit_inferior_exit_hook): Update.

Change all_objfiles adapter to be a method on program_space

This changes the all_objfiles range adapter to be a method on the
program space, and fixes up all the users.

gdb/ChangeLog
2019-01-17  Tom Tromey  <tom@tromey.com>

* progspace.h (program_space) <objfiles_range>: New typedef.
<objfiles>: New method.
<objfiles_head>: Rename from objfiles.
(object_files): Update.
* guile/scm-progspace.c (gdbscm_progspace_objfiles): Update.
* guile/scm-pretty-print.c
(ppscm_find_pretty_printer_from_objfiles): Update.
* guile/scm-objfile.c (gdbscm_objfiles): Update.
* python/py-xmethods.c (gdbpy_get_matching_xmethod_workers):
Update.
* python/py-progspace.c (pspy_get_objfiles): Update.
* python/py-prettyprint.c (find_pretty_printer_from_objfiles):
Update.
* python/py-objfile.c (objfpy_lookup_objfile_by_name)
(objfpy_lookup_objfile_by_build_id): Update.
* mi/mi-cmd-file.c (mi_cmd_file_list_exec_source_files): Update.
* windows-tdep.c (windows_iterate_over_objfiles_in_search_order):
Update.
* symtab.c (iterate_over_symtabs, matching_obj_sections)
(expand_symtab_containing_pc, lookup_objfile_from_block)
(lookup_static_symbol, basic_lookup_transparent_type)
(find_pc_sect_compunit_symtab, find_symbol_at_address)
(find_line_symtab, info_sources_command)
(default_collect_symbol_completion_matches_break_on)
(make_source_files_completion_list, find_main_name): Update.
* symmisc.c (print_symbol_bcache_statistics)
(print_objfile_statistics, maintenance_print_symbols)
(maintenance_print_msymbols, maintenance_print_objfiles)
(maintenance_info_symtabs, maintenance_check_symtabs)
(maintenance_expand_symtabs, maintenance_info_line_tables):
Update.
* symfile.c (remove_symbol_file_command, overlay_invalidate_all)
(find_pc_overlay, find_pc_mapped_section, list_overlays_command)
(map_overlay_command, unmap_overlay_command)
(simple_overlay_update, expand_symtabs_matching)
(map_symbol_filenames): Update.
* symfile-debug.c (set_debug_symfile): Update.
* spu-tdep.c (spu_overlay_update, spu_objfile_from_frame):
Update.
* source.c (select_source_symtab, forget_cached_source_info):
Update.
* solib.c (solib_read_symbols): Update.
* solib-spu.c (append_ocl_sos): Update.
* psymtab.c (maintenance_print_psymbols)
(maintenance_info_psymtabs, maintenance_check_psymtabs): Update.
* probe.c (parse_probes_in_pspace, find_probe_by_pc): Update.
* printcmd.c (info_symbol_command): Update.
* ppc-linux-tdep.c (ppc_linux_spe_context_inferior_created):
Update.
* objfiles.h (class all_objfiles): Remove.
* objfiles.c (have_partial_symbols, have_full_symbols)
(have_minimal_symbols, qsort_cmp, update_section_map)
(shared_objfile_contains_address_p)
(default_iterate_over_objfiles_in_search_order): Update.
* objc-lang.c (info_selectors_command, info_classes_command)
(find_methods): Update.
* minsyms.c (find_solib_trampoline_target): Update.
* maint.c (maintenance_info_sections)
(maintenance_translate_address, count_symtabs_and_blocks):
Update.
* main.c (captured_main_1): Update.
* linux-thread-db.c (try_thread_db_load_from_pdir)
(has_libpthread): Update.
* linespec.c (iterate_over_all_matching_symtabs)
(search_minsyms_for_name): Update.
* jit.c (jit_find_objf_with_entry_addr): Update.
* hppa-tdep.c (find_unwind_entry)
(hppa_lookup_stub_minimal_symbol): Update.
* gcore.c (gcore_create_callback, objfile_find_memory_regions):
Update.
* elfread.c (elf_gnu_ifunc_resolve_by_cache)
(elf_gnu_ifunc_resolve_by_got): Update.
* dwarf2-frame.c (dwarf2_frame_find_fde): Update.
* dwarf-index-write.c (save_gdb_index_command): Update.
* cp-support.c (add_symbol_overload_list_qualified): Update.
* breakpoint.c (create_overlay_event_breakpoint)
(create_longjmp_master_breakpoint)
(create_std_terminate_master_breakpoint)
(create_exception_master_breakpoint): Update.
* blockframe.c (find_pc_partial_function): Update.
* ada-lang.c (ada_lookup_simple_minsym, add_nonlocal_symbols)
(ada_collect_symbol_completion_matches)
(ada_add_global_exceptions): Update.

Remove cleanups from solib-target.c

This removes the remaining cleanups from solib-target.c, plus a VEC,
by using std::vector and std::unique_ptr.

Tested by the buildbot.

gdb/ChangeLog
2019-01-17  Tom Tromey  <tom@tromey.com>

* solib-target.c (lm_info_target_p): Remove typedef.  Don't
declare VEC.
(solib_target_parse_libraries): Change return type.
(library_list_start_segment, library_list_start_section)
(library_list_end_library, library_list_start_library); Update.
(solib_target_free_library_list): Remove.
(solib_target_parse_libraries): Remove cleanup.  Change return
type.
(solib_target_current_sos): Update.

Replace "the the" with "the"

This replaces "the the" with "the" in various comments.

Tested by rebuilding.  This didn't test the solib-dsbt.c change, but
it looks harmless.

gdb/ChangeLog
2019-01-17  Tom Tromey  <tromey@bapiya>

* valprint.c: Replace "the the" with "the".
* symtab.c: Replace "the the" with "the".
* solib.c: Replace "the the" with "the".
* solib-dsbt.c: Replace "the the" with "the".
* linespec.c: Replace "the the" with "the".
* dwarf2loc.h: Replace "the the" with "the".
* amd64-windows-tdep.c: Replace "the the" with "the".
* aarch64-tdep.c: Replace "the the" with "the".

Arm: Clean up PE GAS testsuite correct THUMB tests.

The PE targets don't support mapping symbols and so the disassembler is unable
to correctly output thumb instructions when the input was thumb.

So for testcases that only have thumb output, I have copied them and skipped the
ones for which auto-detection is supposed to work on PE, and added a new one
that will force thumb output.  This so that the tests still check the mapping
symbols.

For the tests that switch between thumb and arm in one file I just skip them
entirely on PE targets.

This cleans up the PE GAS testsuite.

gas/

* testsuite/gas/arm/archv6t2-1-pe.d: New test.
* testsuite/gas/arm/archv6t2-1.d: Skip pe.
* testsuite/gas/arm/csdb.d: Skip pe.
* testsuite/gas/arm/sb-thumb1-pe.d: New test.
* testsuite/gas/arm/sb-thumb1.d: Skip pe.
* testsuite/gas/arm/sb-thumb2-pe.d: New test.
* testsuite/gas/arm/sb-thumb2.d: Skip pe.
* testsuite/gas/arm/udf.d: Skip pe.

Update objdump's --disassemble=<symbol> feature so that if <symbol> is a function, the entire function will be disassembled, regardless of the presence of interveening symbols.

* objdump.c (disassemble_section): When disassembling from a
symbol only stop at the next symbol if the original symbol was not
a function symbol.  Otherwise continue disassembling until a new
function is reached.
* testsuite/binutils-all/objdump.exp: Add tests of extended
functionality.
* testsuite/binutils-all/disasm.s: New test source file.

Automatic date update in version.in

Actually remove definitions of DEFINE_NON_INLINE_P and DEFINE_INLINE_P

I applied the patch "Do not expand macros to 'defined'" by hand because
I couldn't apply it with git-am, and of course forgot to remove the
macro definitions.  This patch fixes it, and also makes the ChangeLog
entry a bit cleaner.

Don't emit vendor attribute section if there is no attribute to emit.

2019-01-16  Kito Cheng  <kito@andestech.com>
bfd/
* elf-attrs.c (vendor_obj_attr_size): Return 0 if size is 0 even
for OBJ_ATTR_PROC.
gas/
* testsuite/gas/riscv/attribute-empty.d: New.

RISC-V: Merge ELF attribute for ld.

2019-01-16  Kito Cheng  <kito@andestech.com>
    Nelson Chu  <nelson@andestech.com>

bfd/
* elfnn-riscv.c (in_subsets): New.
(out_subsets): Likewise.
(merged_subsets): Likewise.
(riscv_std_ext_p): Likewise.
(riscv_non_std_ext_p): Likewise.
(riscv_std_sv_ext_p): Likewise.
(riscv_non_std_sv_ext_p): Likewise.
(riscv_version_mismatch): Likewise.
(riscv_i_or_e_p): Likewise.
(riscv_merge_std_ext): Likewise.
(riscv_merge_non_std_and_sv_ext): Likewise.
(riscv_merge_arch_attr_info): Likewise.
(riscv_merge_attributes): Likewise.
(_bfd_riscv_elf_merge_private_bfd_data): Merge attribute.
ld/
* testsuite/ld-elf/orphan-region.d: XFAIL for RISC-V, because add new
section.
* testsuite/ld-riscv-elf/ld-riscv-elf.exp: Add new tests.
* testsuite/ld-riscv-elf/attr-merge-arch-01.d: New test.
* testsuite/ld-riscv-elf/attr-merge-arch-01a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-01b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-02.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-02a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-02b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-03.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-03a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-03b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-failed-01.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-failed-01a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-arch-failed-01b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-priv-spec-b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-priv-spec.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align-a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align-b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align-failed-a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align-failed-b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align-failed.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-stack-align.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-01.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-01a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-01b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-02.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-02a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-02b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-03.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-03a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-03b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-04.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-04a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-04b.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-05.d: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-05a.s: Likewise.
* testsuite/ld-riscv-elf/attr-merge-strict-align-05b.s: Likewise.

RISC-V: Support ELF attribute for gas and readelf.

2019-01-16  Kito Cheng  <kito@andestech.com>
    Nelson Chu  <nelson@andestech.com>

bfd/
* elfnn-riscv.c (riscv_elf_obj_attrs_arg_type): New.
(elf_backend_obj_attrs_vendor): Define.
(elf_backend_obj_attrs_section_type): Likewise.
(elf_backend_obj_attrs_section): Likewise.
(elf_backend_obj_attrs_arg_type): Define as
riscv_elf_obj_attrs_arg_type.
* elfxx-riscv.c (riscv_estimate_digit): New.
(riscv_estimate_arch_strlen1): Likewise.
(riscv_estimate_arch_strlen): Likewise.
(riscv_arch_str1): Likewise.
(riscv_arch_str): Likewise.
* elfxx-riscv.h (riscv_arch_str): Declare.
binutils/
* readelf.c (get_riscv_section_type_name): New function.
(get_section_type_name): Add handler for RISC-V.
(riscv_attr_tag_t): Declare.
(riscv_attr_tag): New.
(display_riscv_attribute): New function.
(process_attributes): Add handler for RISC-V.
* testsuite/binutils-all/strip-3.d: Remove .riscv.attribute
section.
gas/
* config/tc-riscv.c (DEFAULT_RISCV_ATTR): Define to 0 if not defined.
(riscv_set_options): Add `arch_attr` field.
(riscv_opts): Set default value for arch_attr.
(riscv_write_out_arch_attr): New.
(riscv_set_public_attributes): Likewise.
(riscv_md_end): Likewise.
(riscv_convert_symbolic_attribute): Likewise.
(s_riscv_attribute): Likewise.
(explicit_arch_attr): Likewise.
(riscv_pseudo_table): Add .attribute to the table.
(options): Add OPTION_ARCH_ATTR and OPTION_NO_ARCH_ATTR
enumeration constants.
(md_longopts): Add `march-attr' and `mno-arch-attr' options.
(md_parse_option): Handle the new options.
(md_show_usage): Document the `march-attr' option.
* config/tc-riscv.h (md_end): Define as riscv_md_end
(riscv_md_end): Declare.
(CONVERT_SYMBOLIC_ATTRIBUTE): Define as
riscv_convert_symbolic_attribute.
(riscv_convert_symbolic_attribute): Declare.
(start_assemble): Declare.
* testsuite/gas/elf/elf.exp: Adjust test case for section2.e.
* testsuite/gas/elf/section2.e-riscv: New.
* testsuite/gas/riscv/attribute-01.d: New test
* testsuite/gas/riscv/attribute-02.d: Likewise.
* testsuite/gas/riscv/attribute-03.d: Likewise.
* testsuite/gas/riscv/attribute-04.d: Likewise.
* testsuite/gas/riscv/attribute-04.s: Likewise.
* testsuite/gas/riscv/attribute-05.d: Likewise.
* testsuite/gas/riscv/attribute-05.s: Likewise.
* testsuite/gas/riscv/attribute-06.d: Likewise.
* testsuite/gas/riscv/attribute-06.s: Likewise.
* testsuite/gas/riscv/attribute-07.d: Likewise.
* testsuite/gas/riscv/attribute-07.s: Likewise.
* testsuite/gas/riscv/attribute-08.d: Likewise.
* testsuite/gas/riscv/attribute-08.s: Likewise.
* testsuite/gas/riscv/attribute-unknown.d: Likewise.
* testsuite/gas/riscv/attribute-unknown.s: Likewise.
* testsuite/gas/riscv/empty.l: Likewise.
* doc/c-riscv.texi (.attribute): Add documentation.
* configure.ac (--enable-default-riscv-attribute): New options.
* configure: Re-generate.
* config.in: Re-generate.
include/
* elf/riscv.h (SHT_RISCV_ATTRIBUTES): Define.
(Tag_RISCV_arch): Likewise.
(Tag_RISCV_priv_spec): Likewise.
(Tag_RISCV_priv_spec_minor): Likewise.
(Tag_RISCV_priv_spec_revision): Likewise.
(Tag_RISCV_unaligned_access): Likewise.
(Tag_RISCV_stack_align): Likewise.

Do not expand macros to 'defined'

Expanding a macro which contains 'defined' PP keyword is UB.  It causes
a compilation failure when -Wexpansion-to-defined is used.

sim/common/Changelog:
2019-01-16  Pavel I. Kryukov  <kryukov@frtk.ru>

* sim-arange.c: eliminate DEFINE_NON_INLINE_P

Introduce dwarf2_cu::get_builder

This patch is an attempt to deal with a variety of bugs reported where
GDB segfaults attempting to access a dwarf2_cu's builder.  In certain
circumstances, this builder can be NULL.  This is especially common
when inheriting DIEs via inlined subroutines in other CUs.  The test
case demonstrates one such situation reported by users.  See gdb/23773,
rhbz1638798, and dups for other concrete examples.

The approach taken here is to save the ancestor CU into the dwarf2_cu of
all CUs with DIEs that are "imported."  This can happen whenever
follow_die_offset and friends are called.  This essentially introduces a
chain of CUs that caused the importation of a DIE from a CU.  Whenever
a builder is requested of a CU that has none, the ancestors are searched
for the first one with a builder.

A design side effect of this is that the builder can now only be
accessed by getter and setter methods because the builder itself
is private.

The bulk of the patch is relatively mindless text conversion from
"cu->builder" to "cu->get_builder ()".  I've included one test which
was derived from one (of the many) bugs reported on the issue in both
sourceware and Fedora bugzillas.

gdb/ChangeLog:

PR gdb/23773
* dwarf2read.c (dwarf2_cu) <ancestor>: New field.
<builder>: Rename to ..
<m_builder>: ... this and make private.
(dwarf2_cu::get_builder): New method.  Change all users of
`builder' to use this method.
(dwarf2_start_symtab): Move to ...
(dwarf2_cu::start_symtab): ... here.  Update all callers
(setup_type_unit_groups): Move to ...
(dwarf2_cu::setup_type_unit_groups): ... here.  Update all
callers.
(dwarf2_cu::reset_builder): New method.
(process_full_compunit, process_full_type_unit): Use
dwarf2_cu::reset_builder.
(follow_die_offset): Record the ancestor CU if it is different
from the followed DIE's CU.
(follow_die_sig_1): Likewise.

gdb/testsuite/ChangeLog:

PR gdb/23773
* gdb.dwarf2/inlined_subroutine-inheritance.exp: New file.

doc: Add table of MI versions

This patch adds a table summarizing the history or MI versions:

- The version number
- Which GDB version introduced it
- Breaking changes compared to the previous version

The goal of the table is to help writers of front ends know which
version of MI they can use with a given GDB version.  It will also help
them update their code to work against a newer MI version.

Right now, we just have 1 and 2, but we expect to add an entry for 3
soon.  I did a bit of archelogy and reverse engineering of the code to
come up with the breaking changes for MI 2.

I did some changes to the text around it, some things that I thought
needed to be clarified, seemed a bit dated or seemed just wrong
(especially "Apart from mi0, new versions of @value{GDBN} will not
support old versions of MI").

gdb/doc/ChangeLog:

* gdb.texinfo (GDB/MI Development and Front Ends): Add table of
MI versions.  Update text around it.

S12Z: gas: Fix bug when a symbol name was the single letter 'c'.

The assembler incorrectly recognised "c" as a register name, and
refused to allow it where it expected a symbol/label.

gas/
* config/tc-s12z.c (lex_reg_name): Compare the length of the strings
before the contents.
* testsuite/gas/s12z/labels.d: New file.
* testsuite/gas/s12z/labels.s: New file.
* testsuite/gas/s12z/s12z.exp: Add them.

S12Z: gas: Permit "extend" instructions which don't actually extend.

Other assemblers permit "extending" a register into a register of a
smaller size or the same size.  It doesn't make much sense to do this
but would appear to be a valid instruction.  So change the error to a
warning.

gas/
* config/tc-s12z.c (tfr): Change as_bad to as_warn.
Also fix message typo and semantics.

S12Z: Emit RELOC_S12Z_OPR instead of RELOC_EXT24 where appropriate.

When assembling instructions which involve OPR references, emit
RELOC_S12Z_OPR instead of RELOC_EXT24.

bfd/
* bfd-in2.h [BFD_RELOC_S12Z_OPR]: New reloc.
* libbfd.h: regen.
* elf32-s12z.c (eld_s12z_howto_table): R_S12Z_OPR takes non zero
source field.  (md_apply_fix): Apply final fix
to BFD_RELOC_S12Z_OPR.
* reloc.c[BFD_RELOC_S12Z_OPR]: New reloc.

gas/
* config/tc-s12z.c (emit_opr): Emit BFD_RELOC_S12Z_OPR instead of
BFD_RELOC_24.
* testsuite/gas/s12z/opr-indirect-expr.d: Expect R_S12Z_OPR instead
of R_S12Z_EXT24.

Automatic date update in version.in

sim: Fix definition of SIM_ARANGE_INLINE

If HAVE_INLINE is false, SIM_ARANGE_INLINE is currently defined as

    #define SIM_ARANGE_INLINE EXTERN

However, EXTERN is not defined anywhere, leading to errors such as:

    In file included from
    /mipt-mips/simulator/../../sim/common/sim-basics.h:147:0,
                     from /mipt-mips/simulator/export/gdb/sim-main.h:13,
                     from /mipt-mips/simulator/export/gdb/gdb_interface.cpp:7:
    /mipt-mips/simulator/../../sim/common/sim-arange.h:71:27: error: ‘EXTERN’
    does not name a type; did you mean ‘EUSERS’?
     #define SIM_ARANGE_INLINE EXTERN
                               ^
    /mipt-mips/simulator/../../sim/common/sim-arange.h:76:1: note: in expansion
    of macro ‘SIM_ARANGE_INLINE’
     SIM_ARANGE_INLINE int
     ^~~~~~~~~~~~~~~~~

I (Simon) have reproduced the problem by simply removing the HAVE_INLINE
definition.  This was originally reported by Pavel I. Kryukov
<kryukov@frtk.ru>.

sim/common/ChangeLog:

* sim-arange.h (SIM_ARANGE_INLINE): Change EXTERN to extern.

Remove cleanup from remote.c

This removes the remaining cleanup from remote.c, by changing
remote_state::buf to be a gdb::char_vector, and then fixing up all the
fallout.  The resulting patch is somewhat ugly, but on the other hand,
it eliminates some possibilities for mistakes.

Regression tested using the
Fedora-x86_64-native-extended-gdbserver-m64 builder on the buildbot.

gdb/ChangeLog
2019-01-15  Tom Tromey  <tom@tromey.com>

* remote.c (class remote_state) <buf>: Now a char_vector.
<buf_size>: Remove.
(remote_target::getpkt): Change type of buf.  Remove sizeof_buf
parameter.
(remote_target::getpkt_or_notif_sane_1)
(remote_target::getpkt_sane)
(remote_target::getpkt_or_notif_sane): Likewise.
(class remote_target) <putpkt>: New overload.
(remote_target::read_frame): Change type of "buf_p".  Remove
sizeof_p parameter.
(packet_ok): New overload.
(packet_check_result): New overload.
Update all uses.

Constify some remote-notif functions

This constifies the "buf" arguments to various remote-notif functions
and updates the users.

gdb/ChangeLog
2019-01-14  Tom Tromey  <tom@tromey.com>

* remote-notif.c (handle_notification, remote_notif_ack)
(remote_notif_parse): Make "buf" const.
* remote-notif.h (struct notif_client) <parse, ack>: Make "buf"
const.
(remote_notif_parse, remote_notif_ack, handle_notification):
Likewise.
* remote.c (remote_notif_stop_parse): Make "buf" const.
(remote_target::remote_parse_stop_reply): Make "buf" const.
(remote_notif_stop_ack): Make "buf" const.

Constify remote_console_output

This constifies the parameter to remote_console_output.

gdb/ChangeLog
2019-01-14  Tom Tromey  <tom@tromey.com>

* remote.c (remote_console_output): Make parameter const.

Constify target_pass_signals and target_program_signals

This constifies the final parameter to target_pass_signals and
target_program_signals and updates the rest of gdb.

Note that I have no way to test the nto-procfs.c change.

gdb/ChangeLog
2019-01-14  Tom Tromey  <tom@tromey.com>

* target-debug.h (target_debug_print_signals): Constify.
* nto-procfs.c (nto_procfs_target::pass_signals): Update.
* procfs.c (procfs_target::pass_signals): Update.
* linux-nat.c (linux_nat_target::pass_signals): Update.
* linux-nat.h (class linux_nat_target) <pass_signals>: Update.
* target-delegates.c: Rebuild.
* remote.c (remote_target::program_signals): Update.
(remote_target::pass_signals): Update.
* target.c (target_pass_signals): Constify argument.
(target_program_signals): Likewise.
* target.h (struct target_ops) <pass_signals, program_signals>:
Constify argument.
(target_pass_signals, target_program_signals): Constify argument.

Automatic date update in version.in

Fix placement of output in TUI mode

The fix for PR tui/28819 regressed gdb command output a bit.  In
"nonl" mode, pressing the Enter key will result in a newline not being
echoed properly, so that gdb output for the command will begin on the
same line as the input.

This patch changes gdb_wgetch to echo the newline.  I have only tested
this interactively, as the TUI doesn't have automated tests in
general.

gdb/ChangeLog
2019-01-14  Tom Tromey  <tom@tromey.com>

PR tui/28819:
* tui/tui-io.c (gdb_wgetch): Print \r when needed.

Include <string.h> to dis-asm.h to get strchr declaration

Include <string.h> to dis-asm.h to get strchr declaration

include/Changelog:
2019-01-11  Pavel I. Kryukov  <kryukov@frtk.ru>

        * dis-asm.h: include <string.h>

[PowerPC] Aliases for vector registers

This patch defines pseudo-registers "v0" through "v31" as aliases that
map to the corresponding raw "vr0" through "vr31" vector registers for
Power.

The motivation behind this is that although GDB defines these
registers as "vrX", the disassembler prints them as "vX", e.g. as the
operands in instructions such as "vaddubm v2,v1,v1".  This can be
confusing to users trying to print out the values of the operands
while inspecting the disassembled code.

The new aliases are made not to belong to any register group, to avoid
duplicated values in "info register vector" and "info register all".
The arch-specific rs6000_pseudo_register_reggroup_p function had
previously been removed since the other pseudo-registers could have
their groups inferred by their type.  It restored with this patch to
handle the aliases.  Membership for the other pseudo-registers is
still determined using the default function.

A new tests checks that GDB prints the expected values of vector
registers after they are filled by the inferior, by using both the raw
names and the aliases.  Two other existing tests are modified to also
test the aliases.

gdb/ChangeLog:
2019-01-14  Pedro Franco de Carvalho  <pedromfc@linux.ibm.com>

* ppc-tdep.h (struct gdbarch_tdep) <ppc_v0_alias_regnum>: New
field.
* rs6000-tdep.c: Include reggroups.h.
(IS_V_ALIAS_PSEUDOREG): Define.
(rs6000_register_name): Return names for the "vX" aliases.
(rs6000_pseudo_register_type): Return type for the "vX" aliases.
(rs6000_pseudo_register_reggroup_p): Restore.  Handle "vX"
aliases.  Call default_register_reggroup_p for all other
pseudo-registers.
(v_alias_pseudo_register_read, v_alias_pseudo_register_write):
New functions.
(rs6000_pseudo_register_read, rs6000_pseudo_register_write):
Handle "vX" aliases.
(v_alias_pseudo_register_collect): New function.
(rs6000_ax_pseudo_register_collect): Handle "vX" aliases.
(rs6000_gdbarch_init): Initialize "vX" aliases as
pseudo-registers.  Restore registration of
rs6000_pseudo_register_reggroup_p with
set_tdesc_pseudo_register_reggroup_p.

gdb/testsuite/ChangeLog:
2019-01-14  Pedro Franco de Carvalho  <pedromfc@linux.ibm.com>

* gdb.arch/vsx-regs.exp: Add tests that use the vector register
aliases.
* gdb.arch/altivec-regs.exp: Likewise.  Fix indentation of two
tests.
* gdb.arch/powerpc-vector-regs.c: New file.
* gdb.arch/powerpc-vector-regs.exp: New file.

gdb/doc/ChangeLog:
2019-01-14  Pedro Franco de Carvalho  <pedromfc@linux.ibm.com>

* gdb.texinfo (PowerPC Features): Document the alias
pseudo-registers for the org.gnu.gdb.power.altivec feature.

[PowerPC] Fix "info vector" test in gdb.arch/altivec-regs.exp

This patch fixes one of the tests in gdb.arch/altivec-regs.exp that
was passing an incorrect list to gdb_expect_list, which always
matched.

gdb/testsuite/ChangeLog:
2019-01-14  Pedro Franco de Carvalho  <pedromfc@linux.ibm.com>

* gdb.arch/altivec-regs.exp: Fix the list passed to
gdb_expect_list when testing "info vector".

Correct PR number in prevvious delta

Add support to GNU ld to separate got related plt entries from normal ones in order to be able to switch the non-plt got entries to read-only after startup, conforming to revised Linux for zSeries ABI.

PR 20133
* emulparams/elf64_s390.sh (SEPARATE_GOTPLT): Define.
* emulparams/elf_s390.sh (SEPARATE_GOTPLT): Define.
* testsuite/ld-s390/gotreloc_31-1.dd: Update expected output.
* testsuite/ld-s390/tlsbin.dd: Likewise.
* testsuite/ld-s390/tlsbin.rd: Likewise.
* testsuite/ld-s390/tlsbin.sd: Likewise.
* testsuite/ld-s390/tlsbin_64.dd: Likewise.
* testsuite/ld-s390/tlsbin_64.rd: Likewise.
* testsuite/ld-s390/tlsbin_64.sd: Likewise.
* testsuite/ld-s390/tlspic.dd: Likewise.
* testsuite/ld-s390/tlspic.rd: Likewise.
* testsuite/ld-s390/tlspic.sd: Likewise.
* testsuite/ld-s390/tlspic_64.dd: Likewise.
* testsuite/ld-s390/tlspic_64.rd: Likewise.
* testsuite/ld-s390/tlspic_64.sd: Likewise.
* testsuite/ld-s390/s390.exp: Skip s390 tests for tpf targets.

Sync config.guess, config.sub from GCC

Merge from GCC:
PR target/88535
* config.guess: Import upstream version 2019-01-03.
* config.sub: Import upstream version 2019-01-01.

Implement the assembly instructions yield, wfe, wfi and sev for ARMv6T2 in both ARM mode and Thumb mode.

* config/tc-arm.c (arm_ext_v6k_v6t2): Define.
(insns) [ARM_VARIANT]: Modified.
(insns) [THUMB_VARIANT]: To implement few ARMv6K instructions
in ARMv6T2 as well.
* testsuite/gas/arm/archv6t2-1.d: New test.
* testsuite/gas/arm/archv6t2-1.s: Likewise.
* testsuite/gas/arm/archv6t2-2.d: Likewise.

ld: Clarify --wrap documentation

ld/

* ld.texi (--wrap): Add example to emphasise that only undefined
references are replaced by the linker.

Automatic date update in version.in

gdb: xtensa: fix register counters for xtensa-linux

Commit 37d9e0623102 ("gdb: xtensa: handle privileged registers") changed
how the tdep->num_regs and tdep->num_pseudo_regs are calculated, but
didn't update these numbers in the gdbarch for the xtensa-linux target.
As a result xtensa-linux-gdb behaves as xtensa-elf-gdb and cannot
communicate with the linux gdbserver.
Fix tdep->num_pseudo_regs calculation and call set_gdbarch_num_regs and
set_gdbarch_num_pseudo_regs in xtensa_linux_init_abi.

gdb/
2019-01-13  Max Filippov  <jcmvbkbc@gmail.com>

* xtensa-linux-tdep.c (xtensa_linux_init_abi): Update
tdep->num_pseudo_regs. Add calls to set_gdbarch_num_regs and
set_gdbarch_num_pseudo_regs.

Automatic date update in version.in

Implement help/show values for 'set|show style'.

Currently, the behaviour is:
  (gdb) show style
  (gdb) set style
  (gdb) show style address
  (gdb) set style address
  (gdb)

With this patch, the behaviour is:
  (gdb) show style
  style address background:  The "address" background color is: none
  style address foreground:  The "address" foreground color is: blue
  style address intensity:  The "address" display intensity is: normal
  enabled:  CLI output styling is enabled.
  style filename background:  The "filename" background color is: none
  style filename foreground:  The "filename" foreground color is: green
  style filename intensity:  The "filename" display intensity is: normal
  style function background:  The "function" background color is: none
  style function foreground:  The "function" foreground color is: yellow
  style function intensity:  The "function" display intensity is: normal
  style variable background:  The "variable" background color is: none
  style variable foreground:  The "variable" foreground color is: cyan
  style variable intensity:  The "variable" display intensity is: normal
  (gdb) set style
  "set style" must be followed by an appropriate subcommand.
  List of set style subcommands:

  set style address -- Address display styling
  set style enabled -- Set whether CLI styling is enabled
  set style filename -- Filename display styling
  set style function -- Function name display styling
  set style variable -- Variable name display styling

  Type "help set style" followed by set style subcommand name for full documentation.
  Type "apropos word" to search for commands related to "word".
  Command name abbreviations are allowed if unambiguous.
  (gdb) show style address
  background:  The "filename" background color is: none
  foreground:  The "filename" foreground color is: green
  intensity:  The "filename" display intensity is: normal
  (gdb) set style address
  List of set style address subcommands:

  set style address background -- Set the background color for this property
  set style address foreground -- Set the foreground color for this property
  set style address intensity -- Set the display intensity color for this property

  Type "help set style address" followed by set style address subcommand name for full documentation.
  Type "apropos word" to search for commands related to "word".
  Command name abbreviations are allowed if unambiguous.
  (gdb)

gdb/ChangeLog
* cli/cli-style.h (class cli_style_option): <add_setshow_commands>
Remove arg prefixname, add do_set and do_show.
Add member functions set_list and show_list.
* cli/cli-style.c (class cli_style_option): Update accordingly.
(style_set_list): Move to file scope.
(style_show_list): Likewise.
(set_style): Call help_list.
(show_style): Call cmd_show_list.
(_initialize_cli_style): New macro STYLE_ADD_SETSHOW_COMMANDS.
Update to use the new macro.

Add RXv3 instructions.

        * rx-decode.opc (DSIZE): New. double size.
        (_ld): New. dmov size attribute.
        (PSCALE): Add double size.
        (DCR, DDR, DDRH, DDRL, DCND): New. Double FPU registers.
        (SCR, SDR, SDRH, SDRL): Likewise.
        (S2DR, S2CR): Likewise.
        (SDD): New. double displacement.
        (DL): New. Set dmov size attribute.
        (rx_decode_opcode): Add RXv3 instructions.
        * rx-decode.c: Regenerate.
        * rx-dis.c (size_names): Add double entry.
        (opsize_names): Likewise.
        (double_register_names): New. Double FPU registers.
        (double_register_high_names): Likewise.
        (double_register_low_names): Likewise.
        (double_register_control_names): Likewise.
        (double_condition_names): dcmp condition.
        (print_insn_rx): Add  bfmov / bfmovz output.
        Add double FPU output.

expand a bit the in-GDB help for the "catch exception" help text

One of our users remarked that the help doesn't mention the fact that
the "catch exception" supports the special argument "unhandled" to catch
exceptions which do not have a handler. This patch changes the output
of...

 | (gdb) help catch exception
 | Catch Ada exceptions, when raised.
 | With an argument, catch only exceptions with the given name.

... to ...

 | (gdb) help catch exception
 | Catch Ada exceptions, when raised.
 | Usage: catch exception [ ARG ]
 |
 | Without any argument, stop when any Ada exception is raised.
 | If ARG is "unhandled" (without the quotes), only stop when the exception
 | being raised does not have a handler (and will therefore lead to the task's
 | termination).
 | Otherwise, the catchpoint only stops when the name of the exception being
 | raised is the same as ARG.

gdb/ChangeLog:

        * ada-lang.c (_initialize_ada_language): Expand the help text
        for the "catch exception" command.

Tested on x86_64-linux, no regression.

gdb/testsuite: Don't allow paths to appear in test name

Having paths in the test names makes it harder to compare results
between two runs in different directories.  Give the test a name so
that the path doesn't appear.

gdb/ChangeLog:

* gdb.base/style.exp: Don't include path in testname.

Fix 'obj' may be used uninitialized warning in symtab.c:matching_obj_sections.

Fix warning:

gdb/symtab.c: In function ‘int matching_obj_sections(obj_section*, obj_section*)’:
gdb/symtab.c:1024:12: warning: ‘obj’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   if (obj->separate_debug_objfile_backlink != NULL

2019-01-12  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* symtab.c (matching_obj_sections): Initialize obj,
declare it closer to its usage.

Automatic date update in version.in

More fallout from PR 23963 change

PR 23963
* testsuite/gas/m68hc11/lbranch-dwarf2.d: Adjust for PR23963 change.
* testsuite/gas/m68hc11/opers12-dwarf2.d: Likewise.

Automatic date update in version.in

Replace inf_threads_iterator with next_iterator

This changes inf_threads_iterator and some range adapters in
thread-iter.h to use next_iterator and next_adapter instead.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* thread-iter.h (inf_threads_iterator): Use next_iterator.
(basic_inf_threads_range): Remove.
(inf_threads_range, inf_non_exited_threads_range)
(safe_inf_threads_range): Use next_adapter.

gdb/23712: Test case for multidictionary

This is a test derived from one of the reproducers in symtab/23010.
The DIE tree used here is typical of compilations with LTO, where an
artificial parent DIE of language C99 imports DIEs of other languages.

gdb/testsuite/ChangeLog:

PR gdb/23712
PR symtab/23010
* gdb.dwarf2/multidictionary.exp: New file.

gdb/23712: Remove dw2_add_symbol_to_list

Finally, we can remove dw2_add_symbol_to_list since the wrapper function
originally introduced to catch this multi-language scenario is no longer
needed.  With multi-language dictionaries, we can now support adding
symbols of multiple languages, negating the need for the assertion
entirely.

This patch should now fix gdb/23712 (and symtab/23010).  At least it will
if the NULL buildsym_compunit problem doesn't strike first (see gdb/23773).

gdb/ChangeLog:

PR gdb/23712
PR symtab/23010
* dwarf2read.c (dw2_add_symbol_to_list): Remove.
(fixup_go_packaging, new_symbol): Use add_symbol_to_list.

gdb/23712: Cleanup/Remove temporary dictionary functions

Now that multidictionary's are being used, there is no longer any need
to retain the four temporary functions introduced in the beginning of
this series.

This patch removes them.

As an additional cleanup, since the single-language dictionaries are
no longer used outside dictionary.c, make all of those functions
static.

gdb/ChangeLog:

PR gdb/23712
PR symtab/23010
* dictionary.c (pending_to_vector): Remove.
(dict_create_hashed_1, dict_create_linear_1, dict_add_pending_1):
Remove _1 suffix, replacing functions of the same name.  Update
all callers.
(dict_create_hashed, dict_create_hashed_expandable)
(dict_create_linear, dict_create_linear_expandable, dict_free)
(dict_add_symbol, dict_add_pending, dict_size, dict_empty):
Make functions static.

gdb/23712: Use new multidictionary API

This patch builds on the previous by enabling the `new' multidictionary
API.  A lot of the hunks are simply textual replacements of "dict_"
with "mdict_" and similar transformations.

A word of warning, even with the use of multidictionaries, the code
still does not satisfactorily fix the reported problems with gdb/23712
(or gdb/23010). We still have additional changes to make before that
happens.

gdb/ChangeLog:

PR gdb/23712
PR symtab/23010
* dictionary.h (struct dictionary): Replace declaration with
multidictionary.
(dict_create_hashed, dict_create_hashed_expandable)
(dict_create_linear, dict_create_linear_expandable)
(dict_free, dict_add_symbol, dict_add_pending, dict_empty)
(dict_iterator_first, dict_iterator_next, dict_iter_match_first)
(dict_iter_match_next, dict_size): Rename to "mdict_" versions
taking multidictionary argument.
[ALL_DICT_SYMBOLS]: Update for multidictionary.
* block.h (struct block) <dict>: Change to multidictionary
and rename `multidict'.
* block.c, buildsym.c, jit.c, mdebugread.c, objfiles.c,
symmisc.c: Update all dictionary references to multidictionary.

gdb/23712: Introduce multidictionary's

gdb/23712 is a new manifestation of the now-infamous (at least to me)
symtab/23010 assertion failure (DICT_LANGUAGE == SYMBOL_LANGAUGE).

An example of the problem (using test case from symtab/23010):

Reading symbols from /home/rdiez/rdiez/arduino/JtagDue/BuildOutput/JtagDue-obj-release/firmware.elf...done.
(gdb) p SysTick_Handler
dwarf2read.c:9715: internal-error: void dw2_add_symbol_to_list(symbol*, pending**): Assertion `(*listhead) == NULL || (SYMBOL_LANGUAGE ((*listhead)->symbol[0]) == SYMBOL_LANGUAGE (symbol))' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n)

This assertion was added specifically to catch this condition (of adding
symbols of different languages to a single pending list).

The problems we're now seeing on systems utilizing DWARF debugging seem to
be caused by the use of LTO, which adds a CU with an artificial DIE of
language C99 which references DIEs in other CUs of language C++.

Thus, we create a dictionary containing symbols of C99 but end up
stuffing C++ symbols into it, and the dw2_add_symbol_to_list triggers.

The approach taken here to fix this is to introduce multi-language
dictionaries to "replace" the standard, single-language dictionaries
used today.

Note to reviewers: This patch introduces some temporary functions to
aide with review.  This and other artifacts (such as "See dictionary.h"
which appear incorrect) will all be valid at the end of the series.

This first patch introduces the new multidictionary and its API (which
is, by design, identical to the old dictionary interface).  It also
mutates dict_create_hashed and dict_create_linear so that they take
a std::vector instead of the usual struct pending linked list.  This will
be needed later on.

This patch does /not/ actually enable multidictionary's.  That is left
for a subsequent patch in the series.

I've done exhaustive performance testing with this approach, and I've
attempted to minimize the overhead for the (overwhelmingly) most common
one-language scenario.

On average, a -g3 -O0 GDB (the one we developers use) will see
approximately a 4% slowdown when initially reading symbols. [I've
tested only GDB and firefox with -readnow.]  When using -O2, this
difference shrinks to ~0.5%.  Since a number of runs with these
patches actually run /faster/ than unpatched GDB, I conclude that
these tests have at least a 0.5% error margin.

On our own gdb.perf test suite, again, results appear to be pretty
negligible.  Differences to unpatched GDB range from -7.8% (yes,
patched version is again faster than unpatched) to 27%.  All tests
lying outside "negligible," such as the 27% slowdown, involve a total
run time of 0.0007 (or less) with smaller numbers of CUs/DSOs (usually 10
or 100).  In all cases, the follow-up tests with more CUs/DSOs is never
more than 3% difference to the baseline, unpatched GDB.

In my opinion, these results are satisfactory.

gdb/ChangeLog:

PR gdb/23712
PR symtab/23010
* dictionary.c: Include unordered_map.
(pending_to_vector): New function.
(dict_create_hashed_1, dict_create_linear_1, dict_add_pending_1):
Rewrite the non-"_1" functions to take vector instead
of linked list.
(dict_create_hashed, dict_create_linear, dict_add_pending): Use the
"new" _1 versions of the same name.
(multidictionary): Define.
(std::hash<enum language): New definition.
(collate_pending_symbols_by_language, mdict_create_hashed)
(mdict_create_hashed_expandable, mdict_create_linear)
(mdict_create_linear_expandable, mdict_free)
(find_language_dictionary, create_new_language_dictionary)
(mdict_add_symbol, mdict_add_pending, mdict_iterator_first)
(mdict_iterator_next, mdict_iter_match_first, mdict_iter_match_next)
(mdict_size, mdict_empty): New functions.
* dictionary.h (mdict_iterator): Define.

Fix tracepoint.c:parse_tracepoint_definition leak (and one more)

Coverity points out that gdb/tracepoint.c:parse_tracepoint_definition
can leak 'cond' in this line:

      cond = (char *) xmalloc (2 * xlen + 1);

That can leak because we're in a loop and 'cond' may have already been
xmalloc'ed into in a previous iteration.  That won't normally happen,
because we don't expect to see a tracepoint definition with multiple
conditions listed, but, it doesn't hurt to be pedantically correct,
in case some stub manages to send something odd back to GDB.

At first I thought I'd just replace the xmalloc call with:

      cond = (char *) xrealloc (cond, 2 * xlen + 1);

and be done with it.  However, my pedantic self realizes that
warning() can throw as well (due to pagination + Ctrl-C), so I fixed
it using gdb::unique_xmalloc_ptr instead.

While doing this, I noticed that these vectors in struct uploaded_tp:

  std::vector<char *> actions;
  std::vector<char *> step_actions;

hold heap-allocated strings, but nothing is freeing the strings,
AFAICS.

So I ended up switching all the heap-allocated strings in uploaded_tp
to unique pointers.  This patch is the result of that.

I also wrote an alternative, but similar patch that uses std::string
throughout instead of gdb::unique_xmalloc_ptr, but in the end reverted
it because the code didn't look that much better, and I kind of
dislike replacing pointers with fat std::string's (3 or 4 times the
size of a pointer) in structures.

gdb/ChangeLog:
2019-01-10  Pedro Alves  <palves@redhat.com>

* breakpoint.c (read_uploaded_action)
(create_tracepoint_from_upload): Adjust to use
gdb::unique_xmalloc_ptr.
* ctf.c (ctf_write_uploaded_tp):
(SET_ARRAY_FIELD): Use emplace_back.
(SET_STRING_FIELD): Adjust to use gdb::unique_xmalloc_ptr.
* tracefile-tfile.c (tfile_write_uploaded_tp):
* tracepoint.c (parse_tracepoint_definition): Adjust to use
gdb::unique_xmalloc_ptr.
* tracepoint.h (struct uploaded_tp) <cond, actions, step_actions,
at_string, cond_string, cmd_strings>: Replace char pointers
with gdb::unique_xmalloc_ptr.

Fix leak in solib-target.c:library_list_start_library

lm_info_target::name is nowadays std::string, so we're leaking the
result of xstrdup.

gdb/ChangeLog:
2019-01-10  Pedro Alves  <palves@redhat.com>

* solib-target.c (library_list_start_library): Don't xstrdup name.

Fix leak in mdebugread.c

Coverity points out that all the "continue;" statements in the switch
case in parse_partial_symbols leak STABSTRING.  This is because we
only release STABSTRING at the end of the scope, with:

              if (stabstring
    && stabstring != debug_info->ss + fh->issBase + sh.iss)
  xfree (stabstring);

but that bit of code is skipped if a case in the switch statement ends
with "continue".

Fix this by using gdb::unique_xmalloc_ptr to manage the heap-allocated
version of 'stabsstring'.

I don't know how to test this.

gdb/ChangeLog:
2019-01-10  Pedro Alves  <palves@redhat.com>

* mdebugread.c (parse_partial_symbols): Use
gdb::unique_xmalloc_ptr to manage heap-allocated 'stabsstring'.

gdb: Fix incorrect variable name in scoped_switch_fork_info

The previous commit:

  commit 1ef8573cc77c91feeef3edab74d383d20809eb33
  Date:   Wed Jan 9 14:02:39 2019 +0000

      gdb: Improve scoped_switch_fork_info class

contained a bug, an incorrect variable name was used. Fixed in this
commit.

gdb/ChangeLog:

* linux-fork.c (scoped_switch_fork_info)
<~scoped_switch_fork_info>: Fix incorrect variable name.

gdb: Improve scoped_switch_fork_info class

After committing this patch I got this feedback:

   https://sourceware.org/ml/gdb-patches/2019-01/msg00181.html

This patch makes the constructor of scoped_switch_fork_info explicit,
and wraps the core of the destructor in a TRY/CATCH block.

I've run this through the testsuite on X86-64/GNU Linux, however, this
code is not exercised, so this patch is untested.

gdb/ChangeLog:

* linux-fork.c (scoped_switch_fork_info)
<scoped_switch_fork_info>: Make explicit.
<~scoped_switch_fork_info>: Wrap core in TRY/CATCH.

Correct changelog entry.

Stop objdump from displaying control codes embedded in symbol names.

PR 23963
binutils* objdump.c (sanitize_string): New function.  Removes control
characters from symbol names.
(dump_section_header): Use new function.
(objdump_print_symname): Likewise.
(objdump_print_addr_with_sym): Likewise.
(show_line): Likewise.
(disassemble_bytes): Likewise.
(disassemble_section): Likewise.
(load_specific_debug_section): Likewise.
(read_section_stabs): Likewise.
(print_section_stabs): Likewise.
(dump_section): Likewise.
(dump_reloc_set): Likewise.
(dump_relocs_in_section): Likewise.
(dump_bfd): Likewise.
(display_any_bfd): Likewise.

gas * testsuite/gas/mips/mips16-branch-absolute-1.d: Adjust for the fact that
control characters are now displayed as escape sequences.
* testsuite/gas/mips/mips16-e.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-0.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-1.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-delay-0.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-delay-1.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-n32-0.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-n32-1.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-n64-sym32-0.d: Likewise.
* testsuite/gas/mips/mips16-pcrel-n64-sym32-1.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-0.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-1.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-delay-0.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-delay-1.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-n32-0.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-n32-1.d: Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-n64-sym32-0.d:
Likewise.
* testsuite/gas/mips/mips16e2@mips16-pcrel-n64-sym32-1.d:
Likewise.
* testsuite/gas/mips/mipsel16-e.d: Likewise.
* testsuite/gas/mips/mipsr6@msa.d: Likewise.
* testsuite/gas/mips/mipsr6@relax-swap3.d: Likewise.
* testsuite/gas/mips/r6-64-n32.d: Likewise.
* testsuite/gas/mips/r6-64-n64.d: Likewise.
* testsuite/gas/mips/r6-n32.d: Likewise.
* testsuite/gas/mips/r6-n64.d: Likewise.
* testsuite/gas/mips/r6.d: Likewise.
* testsuite/gas/mips/tmips16-e.d: Likewise.
* testsuite/gas/mips/tmipsel16-e.d: Likewise.
* testsuite/gas/mn10300/relax.d: Likewise.

Move psymtabs to their own obstack

Previously, the psymtab obstack was just a pointer to the objfile
obstack.  This patch changes psymtabs to use their own obstack,
instead.  A gdb::optional is used to avoid unnecessary allocation when
the obstack is not needed.

After this patch, the psymtab code lifetime model is that, in the core
psymtab code, objects allocated on the psymtab obstack may point to
other such objects, or to objects on the per-BFD obstack -- but never
to the objfile obstack.

Note however that this invariant is only obeyed the core psymtab code,
and even there not quite fully: there is still a link from the psymtab
to the full symtab.

Symbol readers are free to work however they like; and in particular,
even after this patch, in practice all symbol readers violate this
invariant via the read_symtab_private field.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* objfiles.h (objfile::reset_psymtabs): Update.
* objfiles.c (objfile::objfile): Update.
* psymtab.h (psymtab_storage::obstack): Update.
(psymtab_storage::m_obstack): Use gdb::optional.
(class psymtab_storage): Update comment.  Remove objfile
parameter.
* psymtab.c (psymtab_storage::psymtab_storage): Update.

Make psymtab_storage::free_psymtabs private

This adds a new psymtab allocation method to psymtab_storage and
changes the free_psymtabs member to be private.  While not strictly
necessary, this seems like a decent cleanup, and also makes it simpler
to move psymtabs off of obstacks entirely, should that prove
desirable.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* psymtab.h (psymtab_storage::allocate_psymtab): New method.
<free_psymtabs>: Now private.
* psymtab.c (psymtab_storage::allocate_psymtab): Implement.
(allocate_psymtab): Use new method.

Add psymtab_storage::allocate_dependencies

This adds a new method to psymtab_storage to allocate storage for
psymtab dependencies, then changes the symbol readers to use it.  This
has the effect of moving the storage to the psymtab storage obstack.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* xcoffread.c (xcoff_end_psymtab): Use allocate_dependencies.
* psymtab.h (psymtab_storage::allocate_dependencies): New method.
* mdebugread.c (parse_partial_symbols): Use
allocate_dependencies.
* dwarf2read.c (dwarf2_create_include_psymtab): Use
allocate_dependencies.
(process_psymtab_comp_unit_reader)
(build_type_psymtab_dependencies): Likewise.
* dbxread.c (dbx_end_psymtab): Use allocate_dependencies.

Move more allocations to psymtab obstack

This moves a couple more psymtab-related allocations to the psymtab
obstack.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* psymtab.c (add_psymbol_to_bcache): Pass psymtab obstack to
PSYMBOL_SET_LANGUAGE.
(allocate_psymtab): Allocate psymtab on the psymtab obstack.

Allocate the address map on the psymtab obstack

After this patch, the psymtab address map will now be allocated on the
psymtab obstack rather than the objfile obstack.  This also changes
the psymtab storage object to make the obstack private; this will be
used later.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* psymtab.h (psymtab_storage::obstack): New method.
<m_obstack>: Rename from obstack; now private.
* psymtab.c (psymtab_storage): Update.
* dwarf2read.c (create_addrmap_from_index)
(create_addrmap_from_aranges, dwarf2_build_psymtabs_hard):
Update.

Introduce objfile::reset_psymtabs

This introduces a new method, objfile::reset_psymtabs, and changes
reread_symbols to use it.  This method simply destroys the existing
partial symbols and recreates the psymtab_storage object.

This patch fixes a latent bug -- namely, that reread_symbols should
clear objfile::psymbol_map, but does not.  I can submit that
separately if you'd prefer.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* symfile.c (reread_symbols): Call objfile->reset_psymtabs.
* objfiles.h (objfile::reset_psymtabs): New method.

Introduce class psymtab_storage

This introduces a new psymtab_storage class, which holds all
psymbol-related objects that are independent of the objfile.  (This
latter contraint explains why psymbol_map was not moved; though this
could still be done with some work.)

This patch does not yet change where psymtab allocation is done --
that comes later.  This just wraps everything in a single object to
make further transformations simpler.

Note that a shared_ptr is used to link from the objfile to the
psymtab_storage object.  The end goal here is to allow a given symbol
reader to simply attach to the psymtab_storage object to the BFD, then
reuse it in later invocations; shared_ptr makes this simple to reason
about.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* symmisc.c (print_symbol_bcache_statistics): Update.
(print_objfile_statistics): Update.
* symfile.c (reread_symbols): Update.
* psymtab.h (class psymtab_storage): New.
* psymtab.c (psymtab_storage): New constructor.
(~psymtab_storage): New destructor.
(require_partial_symbols): Update.
(ALL_OBJFILE_PSYMTABS_REQUIRED): Rewrite.
(find_pc_sect_psymtab, find_pc_sect_psymbol)
(match_partial_symbol, lookup_partial_symbol, dump_psymtab)
(psym_dump, recursively_search_psymtabs, psym_has_symbols)
(psym_find_compunit_symtab_by_address, sort_pst_symbols)
(start_psymtab_common, end_psymtab_common)
(add_psymbol_to_bcache, add_psymbol_to_list, init_psymbol_list)
(allocate_psymtab): Update.
(psymtab_storage::discard_psymtab): Rename from discard_psymtab.
Update.
(dump_psymtab_addrmap, maintenance_print_psymbols)
(maintenance_check_psymtabs): Update.
(class objfile_psymtabs): Move to objfiles.h.
* psympriv.h (discard_psymtab): Now inline.
(psymtab_discarder::psymtab_discarder): Update.
(psymtab_discarder::~psymtab_discarder): Update.
(ALL_OBJFILE_PSYMTABS): Rewrite.
* objfiles.h (struct objfile) <psymtabs, psymtabs_addrmap,
free_psymtabs, psymbol_cache, global_psymbols, static_psymbols>:
Remove fields.
<partial_symtabs>: New field.
(class objfile_psymtabs): Move from psymtab.h.  Update.
* objfiles.c (objfile::objfile): Initialize partial_symtabs, not
psymbol_cache.
(objfile::~objfile): Don't destroy psymbol_cache.
* mdebugread.c (parse_partial_symbols): Update.
* dwarf2read.c (create_addrmap_from_index)
(create_addrmap_from_aranges, dw2_find_pc_sect_compunit_symtab)
(process_psymtab_comp_unit_reader, dwarf2_build_psymtabs_hard)
(add_partial_subprogram, dwarf2_ranges_read): Update.
* dwarf-index-write.c (write_address_map)
(write_one_signatured_type, recursively_write_psymbols)
(class debug_names, class debug_names, write_psymtabs_to_index):
Update.

Change symbol_set_names to take an objfile_per_bfd_storage

This changes symbol_set_names to take an objfile_per_bfd_storage
argument, and updates the users.  It also changes PSYMBOL_SET_NAMES to
take this argument directly; I feel this clarifies the storage
location of objects created in psymtab.c.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* symtab.h (SYMBOL_SET_NAMES): Update.
(symbol_set_names): Update.
(MSYMBOL_SET_NAMES): Update.
* symtab.c (symbol_set_names): Change argument to be an
objfile_per_bfd_storage.
* psymtab.c (add_psymbol_to_bcache): Update.
* psympriv.h (PSYMBOL_SET_NAMES): Take per_bfd argument.

Change create_demangled_names_hash to take an objfile_per_bfd_storage

This changes create_demangled_names_hash to take an
objfile_per_bfd_storage parameter.  This makes it clearer where it is
storing the objects it allocates.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* symtab.c (create_demangled_names_hash): Change argument to be an
objfile_per_bfd_storage.
(symbol_set_names): Update.

Simplify calls to init_psymbol_list

Existing callers to init_psymbol_list were checking to see if psymbols
had already been initialized.  It seemed better to me to do this check
directly in init_psymbol_list, simplifying the callers.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* xcoffread.c (xcoff_initial_scan): Unconditionally call
init_psymbol_list.
* psymtab.c (init_psymbol_list): Do nothing if already called.
* psympriv.h (init_psymbol_list): Add comment.
* dwarf2read.c (dwarf2_build_psymtabs): Unconditionally call
init_psymbol_list.
* dbxread.c (dbx_symfile_read): Unconditionally call
init_psymbol_list.

Change add_psymbol_to_list to use an enum

This changes add_psymbol_to_list to use an enum, rather than a pointer
to a vector, to decide where to put the new symbol.  This reduces the
number of direct references to the static_psymbols and global_psymbols
members of the objfile, which is handy in a later patch.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* xcoffread.c (scan_xcoff_symtab): Update.
* psymtab.c (add_psymbol_to_list): Replace "list" parameter with
"where".
* mdebugread.c (parse_partial_symbols)
(handle_psymbol_enumerators): Update.
* dwarf2read.c (add_partial_symbol, load_partial_dies): Update.
* dbxread.c (read_dbx_symtab): Update.
* psympriv.h (psymbol_placement): New enum.
(add_psymbol_to_list): Update.

Remove parameters from start_psymtab_common

start_psymtab_common takes references to the global_psymbols and
static_psymbols vectors, but it also has an objfile parameter.  This
is redundant, so this patch simplifies the function by removing those
reference parameters.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* xcoffread.c (xcoff_start_psymtab): Remove global_psymbols and
static_psymbols parameters.
(scan_xcoff_symtab): Update.
* psymtab.c (start_psymtab_common): Remove global_psymbols and
static_psymbols parameters.
* psympriv.h (start_psymtab_common): Update.
* mdebugread.c (parse_partial_symbols): Update.
* dwarf2read.c (create_partial_symtab): Update.
* dbxread.c (read_dbx_symtab): Update.
(start_psymtab): Remove global_psymbols and static_psymbols
parameters.

Remove some unneeded psymtab initializations

allocate_psymtab has long cleared the new psymtab that is returned.
This patch documents this behavior and then removes some redundant
initializations.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* xcoffread.c (xcoff_end_psymtab): Remove some initializations.
* psymtab.c (allocate_psymtab): Add comment.
* psympriv.h (allocate_psymtab): Add comment.
* dwarf2read.c (dwarf2_create_include_psymtab): Remove some
initializations.
* dbxread.c (dbx_end_psymtab): Remove some initializations.

Move some declarations to mdebugread.h

This moves a couple of mdebugread-related declarations from symfile.h
to mdebugread.h, which seemed more appropriate.

gdb/ChangeLog
2019-01-10  Tom Tromey  <tom@tromey.com>

* symfile.h (mdebug_build_psymtabs, elfmdebug_build_psymtabs):
Don't declare.
* mipsread.c: Include mdebugread.h.
* mdebugread.h (mdebug_build_psymtabs, elfmdebug_build_psymtabs):
Declare.
* elfread.c: Include mdebugread.h.

Sync libiberty sources with gcc master versions.

. * libiberty: Sync with gcc.  Bring in:
2019-01-09  Sandra Loosemore  <sandra@codesourcery.com>

PR other/16615

* cp-demangle.c: Mechanically replace "can not" with "cannot".
* floatformat.c: Likewise.
* strerror.c: Likewise.

2018-12-22  Jason Merrill  <jason@redhat.com>

Remove support for demangling GCC 2.x era mangling schemes.
* cplus-dem.c: Remove cplus_mangle_opname, cplus_demangle_opname,
internal_cplus_demangle, and all subroutines.
(libiberty_demanglers): Remove entries for ancient GNU (pre-3.0),
Lucid, ARM, HP, and EDG demangling styles.
(cplus_demangle): Remove 'work' variable.  Don't call
internal_cplus_demangle.

include * Merge from GCC:
2018-12-22  Jason Merrill  <jason@redhat.com>

* demangle.h: Remove support for ancient GNU (pre-3.0), Lucid,
ARM, HP, and EDG demangling styles.

Remove ALL_OBJFILE_PSYMTABS

This removes the ALL_OBJFILE_PSYMTABS macro, replacing its uses with
ranged for loops.

gdb/ChangeLog
2019-01-09  Tom Tromey  <tom@tromey.com>

* dbxread.c (dbx_end_psymtab): Use objfile_psymtabs.
* mdebugread.c (parse_partial_symbols): Use objfile_psymtabs.
* psymtab.c (ALL_OBJFILE_PSYMTABS_REQUIRED): Remove.
(psym_map_symtabs_matching_filename, find_pc_sect_psymtab)
(psym_lookup_symbol, psym_find_last_source_symtab)
(psym_forget_cached_source_info, psym_print_stats)
(psym_expand_symtabs_for_function, psym_expand_all_symtabs)
(psym_expand_symtabs_with_fullname, psym_map_symbol_filenames)
(psym_map_matching_symbols, psym_expand_symtabs_matching)
(psym_find_compunit_symtab_by_address)
(maintenance_print_psymbols, maintenance_info_psymtabs)
(maintenance_check_psymtabs): Use ranged for.
* psymtab.h (class objfile_psymtabs): New.
(require_partial_symbols): Return objfile_psymtabs.
* psympriv.h (ALL_OBJFILE_PSYMTABS): Remove.

Remove ALL_OBJSECTIONS

This removes the ALL_OBJSECTIONS macro, replacing its uses with ranged
for loops.

The special code in this macro for noticing a "break" from the inner
loop was only needed in a single place; so rather than try to
replicate this, I've simply replaced that use with a "goto".

gdb/ChangeLog
2019-01-09  Tom Tromey  <tom@tromey.com>

* symfile.c (overlay_invalidate_all, find_pc_overlay)
(find_pc_mapped_section, list_overlays_command)
(map_overlay_command, unmap_overlay_command)
(simple_overlay_update): Use all_objfiles.
* spu-tdep.c (spu_overlay_update): Use all_objfiles.
* printcmd.c (info_symbol_command): Use all_objfiles.
* objfiles.h (ALL_OBJSECTIONS): Remove.
* maint.c (maintenance_translate_address): Use all_objfiles.
* gcore.c (gcore_create_callback): Use all_objfiles.
(objfile_find_memory_regions): Likewise.

Remove ALL_OBJFILES and ALL_FILETABS

This removes the ALL_OBJFILES and ALL_FILETABS macros, replacing them
with ranged for loops.

gdb/ChangeLog
2019-01-09  Tom Tromey  <tom@tromey.com>

* symtab.c (find_line_symtab, info_sources_command)
(make_source_files_completion_list): Use objfile_compunits.
* source.c (select_source_symtab): Use objfile_compunits.
* objfiles.h (struct objfile): Update comment.
(ALL_OBJFILES): Remove.
(ALL_FILETABS): Remove.
* mi/mi-cmd-file.c (mi_cmd_file_list_exec_source_files): Use
objfile_compunits.

Remove ALL_OBJFILE_FILETABS

This removes ALL_OBJFILE_FILETABS, replacing its uses with ranged for
loops.

gdb/ChangeLog
2019-01-09  Tom Tromey  <tom@tromey.com>

* symmisc.c (print_objfile_statistics, dump_objfile)
(maintenance_print_symbols): Use compunit_filetabs.
* source.c (forget_cached_source_info_for_objfile): Use
compunit_filetabs.
* objfiles.h (ALL_OBJFILE_FILETABS): Remove.
(ALL_FILETABS): Use compunit_filetabs.
* objfiles.c (objfile_relocate1): Use compunit_filetabs.
* coffread.c (coff_symtab_read): Use compunit_filetabs.