This patch tidies some COFF and PE code that unnecessarily used static
variables to communicate between functions.
* coffcode.h (pelength, peheader): Delete static variables.
(coff_apply_checksum): Instead, define them as auto vars, and pass..
(coff_read_word, coff_compute_checksum): ..to here. Delete
unnecessary forward declarations.
* pei-x86_64.c (pdata_count): Delete static variable.
(struct pex64_paps): New.
(pex64_print_all_pdata_sections, pex64_bfd_print_pdata): Pass
a pex64_paps for counting.
* peicode.h (jtab): Constify.
"document" command executed in python, gdb.execute("document
<comname>\n...\nend\n"), will wait for user input. Python extension stops
working from that point.
multi-line suport was introduced in commit 56bcdbea2. But "document" support
seem to be implemented.
gdb/ChangeLog:
2020-12-02 Rae Kim <rae.kim@gmail.com>
* cli/cli-script.c (do_document_command): Rename from
document_command. Handle multi-line input.
(multi_line_command_p): Handle document_control.
(build_command_line): Likewise.
(execute_control_command_1): Likewise.
(process_next_line): Likewise.
(document_command): Call do_document_command.
* cli/cli-script.h (enum command_control_type): Add
document_control.
gdb/testsuite/ChangeLog:
2020-12-02 Rae Kim <rae.kim@gmail.com>
* gdb.base/document.exp: New test.
Change-Id: Ice262b980c05051de4c106af9f3fde5b2a6df6fe
While working on the expression rewrite, I found a few spots that
called the internal functions of the expression evaluator, just to
pass in an expected type. This patch adds a parameter to
evaluate_expression so that these functions can avoid this dependency.
Regression tested on x86-64 Fedora 28.
gdb/ChangeLog
2020-12-15 Tom Tromey <tom@tromey.com>
* stap-probe.c (stap_probe::evaluate_argument): Use
evaluate_expression.
* dtrace-probe.c (dtrace_probe::evaluate_argument): Use
evaluate_expression.
* value.h (evaluate_expression): Add expect_type parameter.
* objc-lang.c (print_object_command): Call evaluate_expression.
* eval.c (evaluate_expression): Add expect_type parameter.
This adds a new helper method, expression::first_opcode, that extracts
the outermost opcode of an expression. This simplifies some patches
in the expression rewrite series.
Note that this patch requires the earlier patch to avoid manual
dissection of OP_TYPE operations.
2020-12-15 Tom Tromey <tom@tromey.com>
* varobj.c (varobj_create): Use first_opcode.
* value.c (init_if_undefined_command): Use first_opcode.
* typeprint.c (whatis_exp): Use first_opcode.
* tracepoint.c (validate_actionline): Use first_opcode.
(encode_actions_1): Use first_opcode.
* stack.c (return_command): Use first_opcode.
* expression.h (struct expression) <first_opcode>: New method.
* eval.c (parse_and_eval_type): Use first_opcode.
* dtrace-probe.c (dtrace_process_dof_probe): Use first_opcode.
I noticed hat evaluate_subexp_do_call takes an array of arguments and
a count -- but, unlike the usual convention, the count does not
include the first element.
This patch changes this function to match call_function_by_hand --
passing the callee separately, and using an array_view for the
arguments. This makes it simpler to understand.
Regression tested on x86-64 Fedora 28.
gdb/ChangeLog
2020-12-15 Tom Tromey <tom@tromey.com>
* f-lang.c (evaluate_subexp_f): Update.
* expression.h (evaluate_subexp_do_call): Update.
* eval.c (evaluate_subexp_do_call): Add callee parameter. Replace
nargs, argvec with array_view.
(evaluate_funcall): Update.
The Ada component interval handling code, used for aggregate
assignments, does a pre-pass over the sub-expressions so that it can
size an array. For my expression rewrite, it was handy to C++-ify
this.
gdb/ChangeLog
2020-12-15 Tom Tromey <tom@tromey.com>
* ada-lang.c (num_component_specs): Remove.
(assign_aggregate): Update.
(aggregate_assign_positional, aggregate_assign_from_choices)
(aggregate_assign_others, add_component_interval): Change
arguments.
Add accessor methods to elfcpp::Ehdr class for EI_OSABI and EI_ABIVERSION;
use those to simplify initialization of Osabi class and eliminate the need
to template the class.
elfcpp/
* elfcpp.h (class Ehdr): Add get_ei_osabi and get_ei_abiversion methods.
gold/
* dwp.cc (class Dwo_file): Use new Ehdr::get_ei_osabi and
get_ei_abiversion methods.
* incremental.cc (make_sized_incremental_binary): Likewise.
* object.cc (Sized_relobj_file::Sized_relobj_file): Likewise.
(make_elf_sized_object): Likewise.
* object.h (class Osabi): Make the class untemplated.
After Andrew's latest patch, I noticed that the deprecation warnings
could use the (so-called) title style when printing command names.
This patch implements this idea.
gdb/ChangeLog
2020-12-15 Tom Tromey <tromey@adacore.com>
* cli/cli-decode.c (deprecated_cmd_warning): Use title style for
command names.
gdb/testsuite/ChangeLog
2020-12-15 Tom Tromey <tromey@adacore.com>
* gdb.base/style.exp: Add deprecation tests.
On SLE-11, I run into:
...
(gdb) if 1^M
>shell HOME=/dev/null PS1="gdb-subshell$ " /bin/sh^M
>end^M
hostname:/dir> FAIL: gdb.base/multi-line-starts-subshell.exp: \
spawn subshell from multi-line (timeout)
...
The problem is that the PS1 setting has no effect, due to a bug on older
openSUSE/SLE version. The mechanism there is:
- /etc/profile sets ENV=/etc/bash.bashrc
- /bin/sh is started
- /bin/sh executes ENV, in other words /etc/bash.bashrc
- during the execution of /etc/bash.bashrc, PS1 is set unconditionally
Fix this by setting PS1 after spawning the subshell.
Tested on x86_64-linux.
2020-12-15 Tom de Vries <tdevries@suse.de>
PR testsuite/26952
* gdb.base/multi-line-starts-subshell.exp: Set PS1 after spawning
shell.
Add (or suppress) a DT_GNU_FLAGS_1 dynamic section
with a bit flag value of DF_GNU_1_UNIQUE.
bfd/
* elflink.c (bfd_elf_size_dynamic_sections): Call
_bfd_elf_add_dynamic_entry to add a DT_GNU_FLAGS_1 section.
include/
* bfdlink.h (struct bfd_link_info): New field gnu_flags_1.
ld/
* emultempl/elf.em (gld${EMULATION_NAME}_handle_option):
Parse -z unique / -z nounique options.
DT_GNU_FLAGS_1 added to the DT_VALRNGLO-DT_VALRNGHI range.
DT_GNU_FLAGS_1 value DF_GNU_1_UNIQUE added.
* elf/common.h (DT_GNU_FLAGS_1, DF_GNU_1_UNIQUE): Define.
Replace the "SPECIAL_expr" comment with either "DW_FORM_block" or
"DW_FORM_exprloc" in the abbrev.
gdb/testsuite/ChangeLog:
* lib/dwarf.exp (Dwarf::_handle_attribute): Handle SPECIAL_expr
specially, set attr_form_comment to the actual FORM string used.
Since DWARF version 4 expressions are represented by DW_FORM_exprloc
instead of a block form. Support this in the testsuite Dwarf Assembler
by setting the SPECIAL_expr form once we know the CU version.
This doesn't change any testsuite results, it just makes the produced
DWARF valid. gdb also accepts expressions in block form for DWARF
version 4 and above, but this is technically incorrect.
gdb/testsuite/ChangeLog:
* lib/dwarf.exp (Dwarf::_read_constants): Don't set
_constants(SPECIAL_expr) here, but set it...
(Dwarf::cu): ...here based on _cu_version.
When running test-case gdb.base/info-shared.exp, I see in gdb.log:
...
Executing on host: \
gcc ... -fPIC -fpic -c -o info-shared-solib1.c.o info-shared-solib1.c
...
The -fPIC comes from the test-case:
...
if { [gdb_compile_shlib $srcfile_lib1 $binfile_lib1 \
[list additional_flags=-fPIC]] != "" } {
...
but the -fpic, which overrides the -fPIC comes from gdb_compile_shlib.
The proc gdb_compile_shlib adds the -fpic or similar dependent on platform
and compiler. However, in some cases it doesn't add anything, which is
probably why all those test-case pass -fPIC.
Fix this by removing -fPIC from all the calls to gdb_compile_shlib, and
ensuring that gdb_compile_shlib takes care of adding it, if required.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-14 Tom de Vries <tdevries@suse.de>
* lib/gdb.exp (gdb_compile_shlib): Make sure it's not necessary to
pass -fPIC.
* gdb.ada/catch_ex_std.exp: Don't pass -fPIC to gdb_compile_shlib.
* gdb.base/break-probes.exp: Same.
* gdb.base/ctxobj.exp: Same.
* gdb.base/dso2dso.exp: Same.
* gdb.base/global-var-nested-by-dso.exp: Same.
* gdb.base/info-shared.exp: Same.
* gdb.base/jit-reader-simple.exp: Same.
* gdb.base/print-file-var.exp: Same.
* gdb.base/skip-solib.exp: Same.
* gdb.btrace/dlopen.exp: Same.
Some code in GDB will examine the structure of an expression to see if
it starts with OP_TYPE, and then proceed to extract the type by hand.
There is no need to do this dissection manually. evaluate_type does
the same thing via an "allowed" API.
This patch changes such code to use evaluate_type. In two cases this
simplifies the code.
Regression tested on x86-64 Fedora 28.
gdb/ChangeLog
2020-12-14 Tom Tromey <tom@tromey.com>
* dtrace-probe.c (dtrace_process_dof_probe): Use value_type.
* typeprint.c (whatis_exp): Always use evaluate_type.
(maintenance_print_type): Likewise. Simplify.
When running test-case gdb.base/gnu-debugdata.exp on SLE-11, I run into:
...
FAIL: gdb.base/gnu-debugdata.exp: xz
...
The fact that xz is not installed does not mean there's a fail, merely that
the test is unsupported.
Fix this by detecting the "spawn failed" reply in run_on_host and issuing
UNSUPPORTED instead.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-14 Tom de Vries <tdevries@suse.de>
PR testsuite/26963
* lib/gdb.exp (run_on_host): Declare test unsupported if spawn fails.
When running test-case gdb.base/solib-corrupted.exp on SLE-11, I get:
...
(gdb) PASS: gdb.base/solib-corrupted.exp: normal list
p/x _r_debug->r_map->l_next = _r_debug->r_map^M
'_r_debug' has unknown type; cast it to its declared type^M
(gdb) FAIL: gdb.base/solib-corrupted.exp: make solibs looping
...
The reason that _r_debug has unknown type is that glibc debuginfo is not
installed. The test-case attempts to detect this but doesn't handle this
particular error string.
Fix this by adding the "unknown type" line to the regexp detecting missing
glibc debuginfo.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-14 Tom de Vries <tdevries@suse.de>
PR testsuite/26962
* gdb.base/solib-corrupted.exp: Handle "'_r_debug' has unknown type;
cast it to its declared type".
On SLE-11, I run into:
...
FAIL: gdb.base/batch-preserve-term-settings.exp: batch run: spawn shell \
(timeout)
...
The problem is that the shell prompt has PS1="\h:\w> ", but the test expects
a shell prompt ending in a space preceded by either '$' or '#':
...
set shell_prompt_re "\[$#\] "
...
We could easily fix this by adding '>' to shell_prompt_re, but this wouldn't
work for other PS1 setting.
Fix this instead by setting the shell prompt to "gdb-subshell$ " (as in
gdb.base/multi-line-starts-subshell.exp).
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-14 Tom de Vries <tdevries@suse.de>
PR testsuite/26951
* gdb.base/batch-preserve-term-settings.exp: Use "gdb-subshell$ " as
shell prompt.
GNAT can generate a mangled name with "B_N" (where N is a number) in
the middle, like "hello__B_1__fourth.0". This is used for names local
to a block. Multiple levels of block-local name can also occur, a
possibility that was neglected by v1 of this patch. This patch
changes gdb to handle these names.
The wild name matcher is updated a straightforward way. The full
matcher is rewritten. The hash function is updated to ensure that
this works.
This version does not seem to have the performance problems that
affected v1. In particular, the previously-slow "bt" problem has been
fixed.
gdb/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* dictionary.c (language_defn::search_name_hash): Ignore "B".
* ada-lang.c (advance_wild_match): Ignore "B".
(full_match): Remove.
(do_full_match): Rewrite.
gdb/testsuite/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* gdb.ada/nested.exp: Add new tests.
* gdb.ada/nested/hello.adb (Fourth, Fifth): New procedures.
get_var_value is only used when an exact match is needed. This
changes this function to ensure this sort of matching is done.
gdb/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* ada-lang.c (get_var_value): Only consider exact matches.
To handle thick pointers with -fgnat-encodings=minimal, gdb will
rewrite the underlying array type to remove the bounds. However, if
the same DWARF type is used both for a thick pointer and for an
ordinary array, this will have the side effect of removing the bounds
from the array. This breaks the printing of objects of this type.
This patch fixes the problem by copying the array type, its range, and
its bounds.
gdb/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* dwarf2/read.c (rewrite_array_type): New function.
(quirk_ada_thick_pointer_struct): Use rewrite_array_type.
gdb/testsuite/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* gdb.dwarf2/ada-thick-pointer.exp: New file.
fixed_point_binop did not account for division by zero. This would
lead to gdb getting SIGFPE and subsequently cause some test cases to
hang.
gdb/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* valarith.c (fixed_point_binop): Call error on division by zero.
gdb/testsuite/ChangeLog
2020-12-14 Tom Tromey <tromey@adacore.com>
* gdb.dwarf2/dw2-fixed-point.exp: Add test for division by zero.
* libdep_plugin.c: New file: Processes archives that contain a
special library dependencies element.
* Makefile.am: Add build rules for libdep_plugin.
* Makefile.in: Regenerate.
* NEWS: Mention the new plugin.
* ld.texi: Document the new plugin.
The static variables used by bfd_section_from_shdr to detect loops
in ELF sections have a problem: Comparing a BFD pointer doesn't
guarantee that the current bfd is the same as the one previously used
to allocate the sections_being_created array. For example, doing
size bad_elf_1 bad_elf_2
with two corrupted ELF files containing section loops will leave the
section_being_created array allocated for the first file and since
bfd_close is called for bad_elf_1 before bfd_elf_2 is opened, it is
possible that the BFD for the second file is allocated in the same
memory as the first file. If bad_elf_2 has more sections than
bad_elf_1 then we might write beyond the end of the array.
So this patch implements the FIXME Nick put in a comment about
attaching the array to the BFD.
* elf-bfd.h (struct elf_obj_tdata): Add being_created.
* elf.c (bfd_section_from_shdr): Delete static vars for loop
detection. Use new tdata variable instead.
* elfcode.h (elf_object_p): Allocate being_created.
/usr/include/sys/mman.h:81:0: note: this is the location of the previous definition
#define SHARED 0x10
PR 27064
* deffilep.y (SHARED_K): Rename from SHARED. Update uses.
rx-elf is an odd target with non-standard names for default text, data
and bss sections. This patch tweaks a new test to make it pass.
* testsuite/gas/elf/section27.s: Reorder .text, .data and .bss
so that output section order does not depend on those sections
being already created. Use ".section .text" rather than ".text".
During an incremental link, we should simply ignore the Gnu properties
sections. We were not handling them properly -- failing to process the
properties from the base file. While that could be fixed, the property
sections are meant for deployed binaries, and incremental linking is
for development, so keeping the properties sections just adds to the
likelihood of forcing a full link sooner.
gold/
PR gold/23539
* object.cc (Sized_relobj_file::layout_gnu_property_section): Ignore
Gnu properties during incremental links.
When processing the incremental update, incoming .rodata merge sections
do not match the corresponding section in the base file, because the
SHF_MERGE flag had not been masked out of the latter.
gold/
PR gold/24123
* layout.cc (Layout::init_fixed_output_section): Mask out flags that
should be ignored when matching sections.
I noticed that the argumen to parse_and_eval_type could be "const".
This patch implements this change.
I wonder if this could be removed. It's only called via
check_stub_method_group, which seems questionable to me. However, I
didn't look into doing this.
gdb/ChangeLog
2020-12-13 Tom Tromey <tom@tromey.com>
* gdbtypes.c (safe_parse_type): Make argument const.
* value.h (parse_and_eval_type): Make argument const.
* eval.c (parse_and_eval_type): Make argument const.
When running test-case gdb.base/endianity.exp using gcc-4.8, we get:
...
(gdb) x/x &o.v^M
0x7fffffffd120: 0x00000004^M
(gdb) XFAIL: gdb.base/endianity.exp: x/x &o.v
x/xh &o.w^M
0x7fffffffd124: 0x0003^M
(gdb) FAIL: gdb.base/endianity.exp: x/xh &o.w
...
The gcc 4.8 compiler does not support the scalar_storage_order attribute, so
the testcase is compiled without that attribute, and the expected results are
different.
Fix this by rather than xfailing, skipping the tests if the compiler does not
support the scalar_storage_order attribute.
Tested on x86_64-linux, with gcc-4.8, gcc-7, and clang-10.
gdb/testsuite/ChangeLog:
2020-12-13 Tom de Vries <tdevries@suse.de>
PR testsuite/26953
* gdb.base/endianity.exp: Skip tests requiring scalar_storage_order
attribute support if compiler doesn't support it.
The single test-case in the testsuite that creates an ada shared library is
gdb.ada/catch_ex_std.exp.
The test-case does use gdb_compile_shlib, but with a few tweaks that make sure
things are properly handled for ada.
Move the ada-specific code to gdb_compile_shlib, such that gdb_compile_sh can
be used for ada shared libs without tweaks.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-13 Tom de Vries <tdevries@suse.de>
* lib/gdb.exp (gdb_compile_shlib): Handle ada.
* gdb.ada/catch_ex_std.exp: Use gdb_compile_shlib to compile from
source to shared lib. Add ada to options.
There's a single test-case in the testsuite that explicitly calls gnatbind and
gnatlink: gdb.ada/catch_ex_std.exp.
Instead, use gnatmake and pass specific gnatbind and gnatlink options using
gnatmake passthrough options -bargs and -largs.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-12-13 Tom de Vries <tdevries@suse.de>
* gdb.ada/catch_ex_std.exp: Use gnatmake -bargs and -largs instead of
calling gnatbind and gnatlink.
