Best practice is to not mix lseek/read with fseek/fread on the same
underlying file descriptor, as not all stdio implementations will cope.
Since the plugin uses lseek/read while bfd uses fseek/fread this patch
reopens the file for exclusive use by the plugin rather than trying to
restore the file descriptor. That allows the plugin to read the file
after plugin_call_claim_file too.
bfd/
PR 23254
* plugin.c (bfd_plugin_open_input): Allow for possibility of
nested archives. Open file again for plugin.
(try_claim): Don't save and restore file position. Close file
if not claimed.
* sysdep.h (O_BINARY): Define.
ld/
PR 23254
* plugin.c (plugin_call_claim_file): Revert 2016-07-19 patch.
(plugin_object_p): Don't dup file descriptor.
This removes the last cleanups from the Ada code by changing
ada_lookup_symbol_list's out parameter to be a std::vector, and then
fixing up the fallout.
This is a relatively shallow change. Deeper changes are possible, for
example (1) changing various other functions to accept a vector rather
than a pointer, or (2) changing ada_lookup_symbol_list to return the
vector and omitting the length entirely.
Tested by the buildbot, but I'll wait for Joel to test these as well.
gdb/ChangeLog
2018-06-04 Tom Tromey <tom@tromey.com>
* ada-lang.h (ada_lookup_symbol_list): Update.
* ada-lang.c (resolve_subexp): Update.
(symbols_are_identical_enums): Change type of syms. Remove nsyms
parameter.
(remove_extra_symbols, remove_irrelevant_renamings): Likewise.
(ada_lookup_symbol_list_worker, ada_lookup_symbol_list): Change
results parameter to std::vector.
(ada_iterate_over_symbols, ada_lookup_symbol, get_var_value):
Update.
* ada-exp.y (block_lookup): Update.
(select_possible_type_sym): Change type of syms. Remove nsyms
parameter.
(write_var_or_type, write_name_assoc): Update.
On Windows, using the "-list-thread-groups --available" GDB/MI command
before an inferior is being debugged:
% gdb -q -i=mi
=thread-group-added,id="i1"
=cmd-param-changed,param="auto-load safe-path",value="/"
(gdb)
-list-thread-groups --available
Segmentation fault
Ooops!
The SEGV happens because the -list-thread-groups --available command
triggers a windows_nat_target::xfer_partial call for a TARGET_OBJECT_OSDATA
object. Until a program is being debugged, the target_ops layer that
gets the call is the Windows "native" layer. Except for a couple of
specific objects (TARGET_OBJECT_MEMORY and TARGET_OBJECT_LIBRARIES),
this layer's xfer_partial method delegates the xfer of other objects
to the target beneath:
default:
return beneath->xfer_partial (object, annex,
readbuf, writebuf, offset, len,
xfered_len);
Unfortunately, there is no "beneath layer" in this case, so
beneath is NULL and dereferencing it leads to the SEGV.
This patch fixes the issue by checking beneath before trying
to delegate the request.
gdb/ChangeLog:
* windows-nat.c (windows_nat_target::xfer_partial): Return
TARGET_XFER_E_IO if we need to delegate to the target beneath
but BENEATH is NULL.
gdb/testsuite/ChangeLog:
* gdb.mi/list-thread-groups-no-inferior.exp: New testcase.
xtensa disassembler does not use information from the .xt.prop sections
to switch between code/data disassembly in text sections. This may
result in incorrect disassembly when data is interpreted as code and
disassembler loses synchronization with instruction stream. Use .xt.prop
section information to correctly interpret code and data and synchronize
with instruction stream.
2018-06-04 Max Filippov <jcmvbkbc@gmail.com>
bfd/
* elf32-xtensa.c (xtensa_read_table_entries): Make global.
(compute_fill_extra_space): Drop declaration. Rename function to
xtensa_compute_fill_extra_space.
(compute_ebb_actions, remove_dead_literal): Update references to
compute_fill_extra_space.
include/
* elf/xtensa.h (xtensa_read_table_entries)
(xtensa_compute_fill_extra_space): New declarations.
opcodes/
* xtensa-dis.c (bfd.h, elf/xtensa.h): New includes.
(dis_private): Add new fields for property section tracking.
(xtensa_coalesce_insn_tables, xtensa_find_table_entry)
(xtensa_instruction_fits): New functions.
(fetch_data): Bump minimal fetch size to 4.
(print_insn_xtensa): Make struct dis_private static.
Load and prepare property table on section change.
Don't disassemble literals. Don't disassemble instructions that
cross property table boundaries.
It is currently not possible to correctly match .xt.prop information
for sections with identical VMA. Allow creation of separate property
sections in the BFD. Add assembler option --separate-prop-tables to
allow creation of separate property sections.
2018-06-04 Volodymyr Arbatov <arbatov@cadence.com>
bfd/
* elf32-xtensa.c (elf32xtensa_separate_props): New global
variable.
(xtensa_add_names): New function.
(xtensa_property_section_name): Add new parameter
separate_sections, use it to choose property section name.
(xtensa_get_separate_property_section): New function.
(xtensa_get_property_section): Invoke
xtensa_get_separate_property_section to get individual property
section if it exists, common property section otherwise.
(xtensa_make_property_section): Pass elf32xtensa_separate_props
to xtensa_property_section_name.
gas/
* config/tc-xtensa.c (elf32xtensa_separate_props): New
declaration.
(option_separate_props, option_no_separate_props): New
enumeration constants.
(md_longopts): Add separate-prop-tables option.
(md_parse_option): Add cases for option_separate_props and
option_no_separate_props.
(md_show_usage): Add help for [no-]separate-prop-tables options.
The macro ELF_SECTION_IN_SEGMENT should be used when calculating if
a section maps to a segment.
gdb/
* elfread.c (elf_symfile_segments): Use ELF_SECTION_IN_SEGMENT.
After pulling Alan's change that added aarch64-sve-linux-ptrace.o to
configure.nat, I got an undefined reference to aarch64_sve_get_vq when
doing a "make clean && make". It turns out that re-running configure
(./config.status --recheck) was needed to re-generate the Makefile with
aarch64-sve-linux-ptrace.o included in the object list. Putting
configure.nat in the dependencies of config.status would make sure that
when we modify configure.nat, the configure script is re-ran. I think
it also makes sense because configure.tgt and configure.host are also
there.
gdb/ChangeLog:
* Makefile.in (config.status): Add configure.nat as a
dependency.
Collect per client specific global data items into struct client_state,
which is similar in purpose to remote.c::remote_state.
gdbserver/ChangeLog
* server.h (struct client_state): New.
* server.c (cont_thread, general_thread, multi_process)
(report_fork_events, report_vfork_events, report_exec_events)
(report_thread_events, swbreak_feature, hwbreak_feature)
(vCont_supported, disable_randomization, pass_signals)
(program_signals, program_signals_p, last_status, last_ptid, own_buf):
Moved to client_state.
* remote-utils.c (remote_debug, noack_mode)
(transport_is_reliable): Moved to client_state.
* tracepoint.c (current_traceframe): Moved to client_state.
Update all callers.
* server.c, remote-utils.c, tracepoint.c, fork-child.c,
linux-low.c, remote-utils.h, target.c: Use client_state.
Pedro pointed out in an earlier patch that it would be possible to
make some helper functions in cp-name-parser.y into methods on
cpname_state, cleaning up the code a bit. This patch implements this
idea.
Doing this required moving the %union earlier in the .y file, so the
patch is somewhat bigger than you might expect.
Tested by building with both bison and byacc, and then running the
gdb.cp tests.
gdb/ChangeLog
2018-06-04 Tom Tromey <tom@tromey.com>
* cp-name-parser.y (cpname_state): Add method declarations.
(HANDLE_QUAL): Update.
(cpname_state::d_grab, cpname_state::fill_comp)
(cpname_state::make_operator, cpname_state::make_dtor)
(cpname_state::make_builtin_type, cpname_state::make_name)
(cpname_state::d_qualify, cpname_state::d_int_type)
(cpname_state::d_unary, cpname_state::d_binary): Now methods.
(%union): Move earlier.
Reduce code copy/paste by adding two helper functions for
aarch64_pseudo_read_value and aarch64_pseudo_write
Does not change any functionality.
gdb/
* aarch64-tdep.c (aarch64_pseudo_read_value_1): New helper func.
(aarch64_pseudo_write_1): Likewise.
(aarch64_pseudo_read_value): Use helper.
(aarch64_pseudo_write): Likewise.
<https://sourceware.org/bugzilla/show_bug.cgi?id=22960#c4> reports
that macOS gdb crashes with a null pointer dereference in
push_target(struct target_ops *). This commit fixes that.
The problem is that commit f6ac5f3d63 ("Convert struct target_ops to
C++") left the darwin_ops global uninitialized.
We don't need that global anymore, we can use the (new)
get_native_target function instead for the same effect.
gdb/ChangeLog:
2018-06-04 Pedro Alves <palves@redhat.com>
* darwin-nat.c (darwin_ops): Delete.
(darwin_attach_pid): Use get_native_target.
Enable SVE support for GDB by reading the VQ when creating a
target description.
Also ensurse that SVE is taken into account when creating
the tdep structure, and store the current VQ value directly in tdep.
gdb/
* aarch64-linux-nat.c (aarch64_linux_read_description): Support SVE.
* aarch64-tdep.c (aarch64_get_tdesc_vq): New function.
(aarch64_gdbarch_init): Check for SVE.
* aarch64-tdep.h (gdbarch_tdep::has_sve): New function.
Previously VQ was of type long. Using uint64_t ensures it always matches the
same type as the VG register.
Note that in the Linux kernel, VQ is 16bits. We cast it up to 64bits
immediately after reading to ensure we always use the same type throughout
the code.
gdb/
* aarch64-tdep.c (aarch64_read_description): Use uint64_t for VQ.
* aarch64-tdep.h (aarch64_read_description): Likewise.
* arch/aarch64.c (aarch64_create_target_description): Likewise.
* arch/aarch64.h (aarch64_create_target_description): Likewise.
* features/aarch64-sve.c (create_feature_aarch64_sve): Likewise.
* nat/aarch64-sve-linux-ptrace.c(aarch64_sve_get_vq): Likewise.
* nat/aarch64-sve-linux-ptrace.h (aarch64_sve_get_vq): Likewise.
While reading value_fetch_lazy, I thought it would be good to split it
in small functions (especially the part that handles lval_register).
gdb/ChangeLog:
* value.c (value_fetch_lazy_bitfield): New.
(value_fetch_lazy_memory): New.
(value_fetch_lazy_register): New.
(value_fetch_lazy): Factor out to smaller functions.
This changes "backslashable" and "represented" in cp-name-parser.y to
be const. This lets the compiler make them read-only (though in my
build it seems that GCC inlines them, which seems even better).
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* cp-name-parser.y (backslashable, represented): Now const.
This changes cp-name-parser.y to include parser-defs.h, removing the
copy-pasted declaration of parser_fprintf. This can be done now that
cp-name-parser.y does not define any global variables.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* cp-name-parser.y: Include parser-defs.h.
(parser_fprintf): Remove declaration.
This changes cp-name-parser.y to be a pure parser.
Originally I had thought that doing this would mean that gdb would
always require Bison. However, I've learned that Byacc supports some
of the Bison extensions in this area. So, the new code ought to work
reasonably well with both.
Note that the Byacc documentations says:
%pure-parser
Most variables (other than yydebug and yynerrs) are allocated
on the stack within yyparse, making the parser reasonably
reentrant.
In our case this is ok, first because gdb does not yet actualy require
reentrancy, and second because gdb does not use yynerrs.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* cp-name-parser.y: Use %pure-parser, %lex-param, and
%parse-param.
(lexptr, prev_lexptr, error_lexptr, global_errmsg, demangle_info)
(global_result): Remove globals.
(struct cpname_state): New.
(yyparse): Don't declare.
(yylex, yyerror): Move declarations after %union.
(d_grab, fill_comp, make_operator, make_dtor, make_builtin_type)
(make_name): Add state parameter.
Update all callers.
(d_qualify, d_int_type, d_unary, d_binary, parse_number) Add state
parameter.
(HANDLE_QUAL, HANDLE_SPECIAL, HANDLE_TOKEN2, HANDLE_TOKEN3):
Update.
(yylex): Add lvalp, state parameters.
(yyerror): Add state parameter.
(cp_demangled_name_to_comp): Update.
This changes cp-name-parser.y to use yy-remap.h, rather than its old
manual approach.
This required declaring parser_fprintf in cp-name-parser.y.
parser-defs.h can't be included here because parser-defs.h declares a
global "lexptr", which conflicts with the local one in
cp-name-parser.y. This is only temporary, and will be cleaned up
later in the series.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* cp-name-parser.y (parser_fprintf): Declare.
(GDB_YY_REMAP_PREFIX): Define.
Include yy-remap.h. Don't redefine yy* identifiers.
This removes a static buffer from cp-name-parser.y by replacing the
fixed-sized buffer with a std::string out parameter.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* python/py-type.c (typy_legacy_template_argument): Update.
* cp-support.h (cp_demangled_name_to_comp): Update.
* cp-name-parser.y (cp_demangled_name_to_comp): Change errmsg
parameter to be a "std::string *".
(main): Update.
Move gdb/common/diagnostics.h to include/diagnostics.h so that it can
be used in binutils.
gdb/
* ada-lex.l: Include "diagnostics.h" instead of
"common/diagnostics.h".
* unittests/environ-selftests.c: Likewise.
* common/diagnostics.h: Moved to ../include.
include/
* diagnostics.h: Moved from ../gdb/common/diagnostics.h.
Trying to insert a breakpoint using *FUNC'address with an Ada program
and then running the program until reaching that breakpoint currently
yields the following behavior:
(gdb) break *a'address
Breakpoint 1 at 0x40240c: file a.adb, line 1.
(gdb) run
[1] + 27222 suspended (tty output) /[...]/gdb -q simple_main
Unsuspending GDB then shows it was suspended trying to report
the following error:
Starting program: /home/takamaka.a/brobecke/ex/simple/a
Error in re-setting breakpoint 1: Unmatched single quote.
Error in re-setting breakpoint 1: Unmatched single quote.
Error in re-setting breakpoint 1: Unmatched single quote.
[Inferior 1 (process 32470) exited normally]
The "a'address" is Ada speak for function A's address ("A" by
itself means the result of calling A with no arguments). The
transcript above shows that we're having problems trying to
parse the breakpoint location while re-setting it. As a result,
we also fail to stop at the breakpoint.
Normally, breakpoint locations are evaluated with the current_language
being set to the language of the breakpoint. But, unfortunately for us,
what happened in this case is that parse_exp_in_context_1 calls
get_selected_block which eventually leads to a call to select_frame
because the current_frame hadn't been set yet. select_frame then
finds that our language_mode is auto, and therefore changes the
current_language to match the language of the frame we just selected.
In our case, the language chosen was 'c', which of course is not
able to parse an Ada expression, hence the error.
This patch prevents this by forcing the language_mode to manual
before we call breakpoint_re_set_one.
gdb/ChangeLog:
* breakpoint.c (breakpoint_re_set): Temporarily force language_mode
to language_mode_manual while calling breakpoint_re_set_one.
gdb/testsuite/ChangeLog:
* gdb.ada/bp_fun_addr: New testcase.
Tested on x86_64-linux.
Bump version number to 2.30.52 since _bfd_link_hide_symbol has been added
to bfd_target.
bfd/
* version.m4: Bump version to 2.30.52
* configure: Regenerated.
binutils/
* configure: Regenerated.
gas/
* configure: Regenerated.
gprof/
* configure: Regenerated.
ld/
* configure: Regenerated.
opcodes/
* configure: Regenerated.
TYPE_TAG_NAME has been an occasional source of confusion and bugs. It
seems to me that it is only useful for C and C++ -- but even there,
not so much, because at least with DWARF there doesn't seem to be any
way to wind up with a type where the name and the tag name are both
non-NULL and different.
So, this patch removes TYPE_TAG_NAME entirely. This should save a
little memory, but more importantly, it simplifies this part of gdb.
A few minor test suite adjustments were needed. In some situations
the new code does not yield identical output to the old code.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* valops.c (enum_constant_from_type, value_namespace_elt)
(value_maybe_namespace_elt): Update.
* valarith.c (find_size_for_pointer_math): Update.
* target-descriptions.c (make_gdb_type): Update.
* symmisc.c (print_symbol): Update.
* stabsread.c (define_symbol, read_type)
(complain_about_struct_wipeout, add_undefined_type)
(cleanup_undefined_types_1): Update.
* rust-lang.c (rust_tuple_type_p, rust_slice_type_p)
(rust_range_type_p, val_print_struct, rust_print_struct_def)
(rust_internal_print_type, rust_composite_type)
(rust_evaluate_funcall, rust_evaluate_subexp)
(rust_inclusive_range_type_p): Update.
* python/py-type.c (typy_get_tag): Update.
* p-typeprint.c (pascal_type_print_base): Update.
* mdebugread.c (parse_symbol, parse_type): Update.
* m2-typeprint.c (m2_long_set, m2_record_fields, m2_enum):
Update.
* guile/scm-type.c (gdbscm_type_tag): Update.
* go-lang.c (sixg_string_p): Update.
* gnu-v3-abi.c (build_gdb_vtable_type, build_std_type_info_type):
Update.
* gdbtypes.h (struct main_type) <tag_name>: Remove.
(TYPE_TAG_NAME): Remove.
* gdbtypes.c (type_name_no_tag): Simplify.
(check_typedef, check_types_equal, recursive_dump_type)
(copy_type_recursive, arch_composite_type): Update.
* f-typeprint.c (f_type_print_base): Update. Print "Type" prefix
in summary mode when needed.
* eval.c (evaluate_funcall): Update.
* dwarf2read.c (fixup_go_packaging, read_structure_type)
(process_structure_scope, read_enumeration_type)
(read_namespace_type, read_module_type, determine_prefix): Update.
* cp-support.c (inspect_type): Update.
* coffread.c (process_coff_symbol, decode_base_type): Update.
* c-varobj.c (c_is_path_expr_parent): Update.
* c-typeprint.c (c_type_print_base_struct_union): Update.
(c_type_print_base_1): Update. Print struct/class/union/enum in
summary when using C language.
* ax-gdb.c (gen_struct_ref, gen_namespace_elt)
(gen_maybe_namespace_elt): Update.
* ada-lang.c (ada_type_name): Simplify.
(empty_record, ada_template_to_fixed_record_type_1)
(template_to_static_fixed_type)
(to_record_with_fixed_variant_part, ada_check_typedef): Update.
gdb/testsuite/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* gdb.xml/tdesc-regs.exp (load_description): Update expected
results.
* gdb.dwarf2/method-ptr.exp: Set language to C++.
* gdb.dwarf2/member-ptr-forwardref.exp: Set language to C++.
* gdb.cp/typeid.exp (do_typeid_tests): Update type_re.
* gdb.base/maint.exp (maint_pass_if): Update.
Currently dwarf2read.c will pass the CU's language to
c_type_print_args -- but this doesn't affect all aspects of type
printing, because some code in c-typeprint.c refers to
current_language.
This patch threads the language through more of the type printing
code, adding an overload to c_type_print. Some uses of
current_language remain, but now they are only in top-level functions.
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* dwarf2read.c (dwarf2_compute_name): Pass CU's language to
c_print_type.
* c-typeprint.c (c_print_type_1): Add "language" parameter.
(c_print_type): Update.
(c_print_type): New overload.
(c_type_print_varspec_prefix, c_type_print_args)
(c_type_print_varspec_suffix, c_print_type_no_offsets)
(c_type_print_base_struct_union, c_type_print_base_1)
(cp_type_print_method_args): Add "language" parameter.
(c_type_print_base): Update.
* c-lang.h (c_print_type): Add new overload.
I noticed that c_type_print_varspec_suffix is only called from
c-typeprint.c, so this patch makes it "static".
gdb/ChangeLog
2018-06-01 Tom Tromey <tom@tromey.com>
* typeprint.h (c_type_print_varspec_suffix): Don't declare.
* c-typeprint.c (c_type_print_varspec_suffix): Now static.
In order to prevent gaps in the register numbering, the Z registers
reuse the V register numbers (which become pseudos on SVE).
2018-06-01 Alan Hayward <alan.hayward@arm.com>
* aarch64-tdep.c (aarch64_sve_register_names): New const
var.
* arch/aarch64.h (enum aarch64_regnum): Add SVE entries.
(AARCH64_SVE_Z_REGS_NUM): New define.
(AARCH64_SVE_P_REGS_NUM): Likewise.
(AARCH64_SVE_NUM_REGS): Likewise.
Fix locviews on ia64. If we do not drop the view after copying line
info to slots in the bundle, we may attempt to compute the view more
than once and get very confused.
for gas/ChangeLog
* dwarf2dbg.c (dwarf2_consume_line_info): Drop view.
First of all there's no point in having separate Cpu386 templates - the
respective SReg3 registers can't be specified for pre-386 anyway; see
parse_real_register().
And then we can also make use of D here for the memory forms of the
insn. This cannot be done for the non-64bit GPR forms because of the
IgnoreSize that cannot be dropped from the to-SREG variant.
All REX bits can be specified via individual prefixes. Redundancy should
only be reported on a per-bit basis.
Note that I originally had further checks added to the test case,
checking the effect also on PDEP. I had to strip those, because my patch
to correctly handle those
(https://sourceware.org/ml/binutils/2017-02/msg00280.html) was rejected.
I continue to think that there should not be any new prefix introduced
to handle the VEX case - whether the encoding of an insn requires VEX et
al should not be of immediate interest to the programmer.
... as is already done elsewhere: There's no need for the use of
operand_type_equal() here - the bit identifying control registers isn't
used for any other purposes.
The control register is never expressed by REX.B; this bit only affects
the involved GPR. Also only one of the operands can have its "control"
flag set, so only check the correct operand.
It's long bothered me that setting a Python parameter from the CLI
will print the "set" help text by default. I think usually "set"
commands should be silent. And, while you can modify this behavior a
bit by providing a "get_set_string" method, if this method returns an
empty string, a blank line will be printed.
This patch removes the "help" behavior and changes the get_set_string
behavior to avoid printing a blank line. The code has a comment about
preserving API behavior, but I don't think this is truly important;
and in any case the workaround -- implementing get_set_string -- is
trivial.
Regression tested on x86-64 Fedora 26.
2018-04-26 Tom Tromey <tom@tromey.com>
* NEWS: Mention new "set" behavior.
* python/py-param.c (get_set_value): Don't print an empty string.
Don't call get_doc_string.
gdb/doc/ChangeLog
2018-04-26 Tom Tromey <tom@tromey.com>
* python.texi (Parameters In Python): Update get_set_string
documentation.
This adds a basic Python API for accessing convenience variables.
With this, convenience variables can be read and set from Python.
Although gdb supports convenience variables whose value changes at
each call, this is not exposed to Python; it could be, but I think
it's just as good to write a convenience function in this situation.
This is PR python/23080.
Tested on x86-64 Fedora 26.
2018-04-22 Tom Tromey <tom@tromey.com>
PR python/23080:
* NEWS: Update for new functions.
* python/py-value.c (gdbpy_set_convenience_variable)
(gdbpy_convenience_variable): New functions.
* python/python-internal.h (gdbpy_convenience_variable)
(gdbpy_set_convenience_variable): Declare.
* python/python.c (python_GdbMethods): Add convenience_variable,
set_convenience_variable.
doc/ChangeLog
2018-04-22 Tom Tromey <tom@tromey.com>
PR python/23080:
* python.texi (Basic Python): Document gdb.convenience_variable,
gdb.set_convenience_variable.
testsuite/ChangeLog
2018-04-22 Tom Tromey <tom@tromey.com>
PR python/23080:
* gdb.python/python.exp: Add convenience variable tests.
Alan Modra
Flavio Ceolin
Nick Clifton
GDB Administrator
Tom Tromey
Joel Brobecker
Max Filippov
H.J. Lu
Alan Hayward
Simon Marchi
Stan Cox
Pedro Alves
Jim Wilson
Alexandre Oliva
Jan Beulich