Document define-prefix command and the use of . in command names.

gdb/ChangeLog
2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>
* NEWS:  Mention define-prefix.  Tell that command names can now
contain a . character.

gdb/doc/ChangeLog
2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.texinfo (Define): Indicate that user-defined prefix can
be used in 'define' command.  Document 'define-prefix' command.

Allow . character as part of command names.

This patch adds . as an allowed character for user defined commands.
Combined with 'define-prefix', this allows to e.g. define a set of Valgrind
specific user command corresponding to the Valgrind monitor commands
(such as check_memory, v.info, v.set, ...).

gdb/ChangeLog
2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* command.h (valid_cmd_char_p): Declare.
* cli/cli-decode.c (valid_cmd_char_p): New function factorizing
the check of valid command char.
(find_command_name_length, valid_user_defined_cmd_name_p): Use
valid_cmd_char_p.
* cli/cli-script.c (validate_comname): Likewise.
* completer.c (gdb_completer_command_word_break_characters):
Do not remove . from the word break char, update comments.
(complete_line_internal_1): Use valid_cmd_char_p.
* guile/scm-cmd.c (gdbscm_parse_command_name): Likewise.
* python/py-cmd.c (gdbpy_parse_command_name): Likewise.

gdb/testsuite/ChangeLog
2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.base/define.exp: Test . in command names.
* gdb.base/setshow.exp: Update test, as . is now part of
command name.

Test define-prefix.

Adds a test testing the new define-prefix command.

2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* gdb.base/define-prefix.exp: New file.

Implement user defined prefix.

This patch adds the new 'define-prefix' command that creates (or mark an
existing user defined command) as a prefix command.
This approach was preferred compared to add a -prefix option to
'define' command : with define-prefix, a command can be defined and
afterwards marked as a prefix.  Also, it is easier to define a
'prefix' only command in one operation.

This patch also adds completers for the 'define' and 'document' commands.
This makes it easier for the user to type the prefixes for 'define'
and type the documented command name for 'document'.

gdb/ChangeLog
2019-11-30  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* cli/cli-script.c (do_define_command): Ensure a redefined
prefix command is kept as a prefix command.
(define_prefix_command): New function.
(show_user_1): Report user defined prefixes.
(_initialize_cli_script):  Create the new 'define-prefix' command.
Add completers for 'define' and 'document'.
* top.c (execute_command):  If command is a user-defined prefix only
command, report the list of commands for this prefix command.

Automatic date update in version.in

gdb: improve debug output of function overload resolution

Function overload resolution prints debug output if turned on via the
'set debug overload' command.  The output includes the badness vector
(BV).  For each function, this vector contains a badness value of the
length of parameters as its first element.  So, BV[0] does not
correspond to a parameter.  The badness values of parameters start
with BV[1].

A badness value is a pair; it contains a rank and a subrank.  Printing
both fields provides useful information.

Improve printing the badness vector along these lines.

gdb/ChangeLog:
2019-11-29  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* valops.c (find_oload_champ): Improve debug output.

Change-Id: I771017e7afbbaf4809e2238a9b23274f55c61f55

gdb: fix segfault in overload resolution debug output

A segfault occurs if overload resolution debug mode is turned on via
the 'set debug overload' command.  E.g.:

~~~
$ gdb ./a.out
...
(gdb) start
...
(gdb) set debug overload 1
(gdb) print foo(5)
-- Arg is int [8], parm is double [9]
Overloaded function instance (null) # of parms 1
Segmentation fault
$
~~~

The problem is, GDB tries to print the badness vector after it has
been std::move'd.  Fix the problem by printing the vector before it is
moved.

gdb/ChangeLog:
2019-11-29  Tankut Baris Aktemur  <tankut.baris.aktemur@intel.com>

* valops.c (find_oload_champ): Print part of debug messages
before the badness vector is std::move'd.

Change-Id: Ia623f9637e82ec332bfeac23eb6b0f2ffdcdde27

Automatic date update in version.in

Fix creal_internal_fn comment

I noticed that the comment before creal_internal_fn refers to $_cimag,
but should refer to $_creal.

gdb/ChangeLog
2019-11-28  Tom Tromey  <tom@tromey.com>

* value.c (creal_internal_fn): Fix comment.

Change-Id: I5665aceb4be5aae7014e914cfb39db184c65d5ea

Make two range_bounds bitfields unsigned

While debugging gdb, I noticed that the bitfields in a range_bounds
were signed, causing the values of these fields to be -1.

I think this is odd; and while we haven't yet committed to boolean
bitfields, I think it is a small improvement to change these types to
unsigned.

gdb/ChangeLog
2019-11-28  Tom Tromey  <tom@tromey.com>

* gdbtypes.h (struct range_bounds) <flag_upper_bound_is_count,
flag_bound_evaluated>: Now unsigned.

Change-Id: Ia377fd931594bbf8653180d4dcb4e60354d90139

Remove unused declaratoin from guile

guile-internal.h declares a function that is never defined.  This
removes the declaration.

gdb/ChangeLog
2019-11-28  Tom Tromey  <tom@tromey.com>

* guile/guile-internal.h (vlscm_scm_from_value_unsafe): Don't
declare.

Change-Id: I2dca228534bc1325d2d4bb319c31328121edecc4

jit: minor improvement to debug logging

gdb/ChangeLog:
2019-11-28  Mihails Strasuns  <mihails.strasuns@intel.com>

* jit.c (jit_bfd_try_read_symtab): Fix printed function name in the
debug output.
* jit.c (jit_unregister_code): Add debug print to match
`jit_register_code`.

Change-Id: Ie66064f3aaa1c74facfc025c8d87f3a057869779

gdb/testsuite: Fix minor bug in skip_btrace*tests procs

The two guard functions skip_btrace_tests and skip_btrace_pt_tests
have a minor bug, if the check function fails to compile then surely
we should skip the btrace tests - currently we return 0 to indicate
don't skip.

gdb/testsuite/ChangeLog:

* lib/gdb.exp (skip_btrace_tests): Return 1 if the test fails to
compile.
(skip_btrace_pt_tests): Likewise.

Change-Id: I6dfc04b4adcf5b9424fb542ece7ddbe751bee301

gas/riscv: Produce version 3 DWARF CIE by default

The flag controlling the default DWARF CIE version to produce now
starts with the value -1.  This can be modified with the command line
flag as before, but after command line flag processing, in
md_after_parse_args targets can, if the global still has the value -1,
override this value.  This gives a target specific default.

If a CIE version is not select either by command line flag, or a
target specific default, then some new code in dwarf2_init now select
a global default.  This remains as version 1 to match previous
behaviour.

This RISC-V has a target specific default of version provided, this
make the return column uleb128, which means we can use all DWARF
registers include CSRs.

I chose to switch to version 3 rather than version 4 as this is most
similar to the global default (version 1).  Switching to version 4
adds additional columns to the CIE header.

gas/ChangeLog:

* as.c (flag_dwarf_cie_version): Change initial value to -1, and
update comment.
* config/tc-riscv.c (riscv_after_parse_args): Set
flag_dwarf_cie_version if it has not already been set.
* dwarf2dbg.c (dwarf2_init): Initialise flag_dwarf_cie_version if
needed.
* testsuite/gas/riscv/default-cie-version.d: New file.
* testsuite/gas/riscv/default-cie-version.s: New file.

ld/ChangeLog:

* testsuite/ld-elf/eh5.d: Accept version 3 DWARF CIE.

Change-Id: Ibbfe8f0979fba480bf0a359978b09d2b3055555e

gas: Check for overflow on return column in version 1 CIE DWARF

In version 1 of DWARF CIE format, the return register column is just a
single byte.  For targets with large numbers of DWARF registers, any
use of a register with a high number for the return column
will (currently) silently overflow giving incorrect DWARF.

This commit adds an error when the overflow occurs.

gas/ChangeLog:

* dw2gencfi.c (output_cie): Error on return column overflow.
* testsuite/gas/riscv/cie-rtn-col-1.d: New file.
* testsuite/gas/riscv/cie-rtn-col-3.d: New file.
* testsuite/gas/riscv/cie-rtn-col.s: New file.

Change-Id: I1809f739ba7771737ec012807f0260e1a3ed5e64

binutils/gas/riscv: Add DWARF register numbers for CSRs

This commit gives DWARF register numbers to the RISC-V CSRs inline
with the RISC-V ELF specification here:

  https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md

The CSRs are defined being numbered from 4096 to 8191.

This adds support to the assembler, required in order to reference
CSRs in, for example .cfi directives.

I have then extended dwarf.c in order to support printing CSR names in
the dumped DWARF output.  As the CSR name space is quite large and
only sparsely populated, I have provided a new function to perform
RISC-V DWARF register name lookup which uses a switch statement rather
than the table base approach that other architectures use.

Any CSR that does not have a known name will return a name based on
'csr%d' with the %d being replaced by the offset of the CSR from 4096.

gas/ChangeLog:

* config/tc-riscv.c (tc_riscv_regname_to_dw2regnum): Lookup CSR
names too.
* testsuite/gas/riscv/csr-dw-regnums.d: New file.
* testsuite/gas/riscv/csr-dw-regnums.s: New file.

binutils/ChangeLog:

* dwarf.c (regname_internal_riscv): New function.
(init_dwarf_regnames_riscv): Use new function.

Change-Id: I3f70bc24fa8b3c75744e6775eeeb87db70c7ecfb

binutils: Make DWARF register name lookup be via a function pointer

Add a layer of indirection for DWARF register name lookup by
introducing a function pointer.  Right now all targets use the same
table based implementation that they always have, however, this will
change in a later commit.

binutils/ChangeLog:

* dwarf.c (typedef dwarf_regname_lookup_ftype): New typedef.
(dwarf_regnames_lookup_func): New static global.
(init_dwarf_regnames_i386): Set dwarf_regnames_lookup_func.
(init_dwarf_regnames_iamcu): Likewise.
(init_dwarf_regnames_x86_64): Likewise.
(init_dwarf_regnames_aarch64): Likewise.
(init_dwarf_regnames_s390): Likewise.
(init_dwarf_regnames_riscv): Likewise.
(init_dwarf_regnames_by_elf_machine_code): Reset
dwarf_regnames_lookup_func to NULL.
(init_dwarf_regnames_by_bfd_arch_and_mach): Likewise.
(regname_internal_by_table_only): New function.
(regname): Make use of dwarf_regnames_lookup_func.

Change-Id: Ibbff61c0abea32927f35c9e478793039ab8bb57d

binutils: Make some functions static in dwarf.c

The architecture specific register name initialisation routines no
longer need to be externally visible, so make them static.

binutils/ChangeLog:

* dwarf.c (init_dwarf_regnames_i386): Make static.
(init_dwarf_regnames_iamcu): Make static.
(init_dwarf_regnames_x86_64): Make static.
(init_dwarf_regnames_aarch64): Make static.
(init_dwarf_regnames_s390): Make static.
(init_dwarf_regnames_riscv): Make static.
* dwarf.h (init_dwarf_regnames_i386): Delete declaration.
(init_dwarf_regnames_iamcu): Delete declaration.
(init_dwarf_regnames_x86_64): Delete declaration.
(init_dwarf_regnames_aarch64): Delete declaration.
(init_dwarf_regnames_s390): Delete declaration.
(init_dwarf_regnames_riscv): Delete declaration.

Change-Id: I9e350f76f98f46e9e3dd88d502f2a2a83e44cb36

binutils: Add a new function to initialise DWARF register name state

Adds a new API function init_dwarf_regnames_by_bfd_arch_and_mach to
initialise the register name state from a BFD architecture and machine
type.

There should be no user visible changes after this commit.

binutils/ChangeLog:

* dwarf.c (init_dwarf_regnames_by_bfd_arch_and_mach): New
function.
* dwarf.h (init_dwarf_regnames_by_bfd_arch_and_mach): Declare.
* objdump.c (dump_dwarf): Call new function instead of calling
specific initialization routines.  Restrucure so that eh_addr_size
is still calculated correctly.

Change-Id: I346d665d2079a18ec4d04bd41893d0e9dc05e4b3

binutils: Rename init_dwarf_regnames

As part of a process to change how dwarf.c figures out the correct
name for a register I wanted to clean up how we initialise the
register name tracking state.

As part of this I rename init_dwarf_regnames to
init_dwarf_regnames_by_elf_machine_code, later commits will add a
different entry point to initialise the register name state.

There should be no user visible changes after this commit.

binutils/ChangeLog:

* dwarf.c (init_dwarf_regnames): Renamed to...
(init_dwarf_regnames_by_elf_machine_code): ...this.
* dwarf.h (init_dwarf_regnames): Renamed to...
(init_dwarf_regnames_by_elf_machine_code): ...this.
* readelf.c (process_file_header): Update call to use new name.

Change-Id: Ic8d2ef5fb62a8590ecd8cbb7e6258e11c6263594

binutils: Be more forgiving of targets with large numbers of registers

Currently if a target has a large ( > 1024 ) number of registers then
we get a warning when dumping the DWARF whenever a register over the
1024 limit is referenced, this occurs in dwarf.c:frame_need_space.

This check was initially introduced to guard against corrupted DWARF
referencing stupidly large numbers of registers.

The frame_need_space function already has a check in place so that, if
a target specifies a set of known DWARF register names then we must
only reference a register within this set, it is only after this check
that we check for the 1024 limit.

What this means is that if a target DOES NOT define a set of known
register names and if we reference more than 1024 registers
frame_need_space will give a warning.

If a target DOES define a set of known registers and there are more
than 1024 defined registers, and we try to reference a register beyond
1024 we will again get an error.

This second case feels wrong to me.  My thinking is that if a target
defines a set of registers then it is not unreasonable to assume the
tools can cope with that number of registers.  And so, if the target
defines 2000 named DWARF registers, frame_need_space should allow
access to all of these registers.

If a target does not define a set of named registers then the 1024
limit should remain.  This is pretty arbitrary, but we do need to have
some limit in place I think, so for now that seems as good as any.

This is an entirely theoretical fix - there are no targets that define
such large numbers of registers, but while experimenting with adding
support for RISC-V CSRs I ran into this issue and felt like it was a
good improvement.

binutils/ChangeLog:

* dwarf.c (frame_need_space): Compare dwarf_regnames_count against
0, and only warn about large numbers of registers if the number is
more than the dwarf_regnames_count.

Change-Id: Ifac1a999ff0677676e81ee373c4c044b6a700827

gas/riscv: Remove unneeded structure

We build a hash table of all register classes and numbers.  The hash
key is the register name and the hash value is the class and number
encoded into a single value, which is of type 'void *'.

When we pull the values out of the hash we cast them to be a pointer
to a structure, however, we never access the fields of that structure,
we just decode the register class and number from the pointer value
itself.

This commit removes the structure and treats the encoded class and
number as a 'void *' during hash lookup.

gas/ChangeLog:

* config/tc-riscv.c (struct regname): Delete.
(hash_reg_names): Handle value as 'void *'.

Change-Id: Ie7d8f46ca3798f56f4af94395279de684f87f9cc

Automatic date update in version.in

Add missing ChangeLog entry for the previous commit

Change-Id: Ibc5788e1879ece9cac637d5c99f92ff4084c8ba1

Add a NEWS entry for multithreaded symbol loading

Just to let people know that this is available and how to use it.

Also updates the description of the setting to say the default is 0.

gdb/ChangeLog:

2019-11-26  Christian Biesinger  <cbiesinger@google.com>

* NEWS: Mention the new multithreaded symbol loading.

Change-Id: I263add6aae03b523f0870ad4d1e972eada4b382a

Turn off threaded minsym demangling by default

Per discussion on gdb-patches with Joel, this patch turns off multihreaded
symbol loading by default. It can be turned on using:
  maint set worker-threads unlimited

To keep the behavior as close as possible to the old code, it still
calls symbol_set_names in the old place if n_worker_threads is 0.

gdb/ChangeLog:

2019-11-27  Christian Biesinger  <cbiesinger@google.com>

* maint.c (n_worker_threads): Default to 0.
(worker_threads_disabled): New function.
* maint.h (worker_threads_disabled): New function.
* minsyms.c (minimal_symbol_reader::record_full): Call symbol_set_names
here if worker_threads_disabled () is true.
(minimal_symbol_reader::install): Skip all threading if
worker_threads_disabled () is true.

Change-Id: I92ba4f6bbf07363189666327cad452d6b9c8e01d

Compute msymbol hash codes in parallel

This is for the msymbol_hash and msymbol_demangled_hash hashtables
in objfile_per_bfd_storage. This basically computes those hash
codes together with the demangled symbol name in the background,
before it inserts the symbols in the hash table.

gdb/ChangeLog:

2019-11-27  Christian Biesinger  <cbiesinger@google.com>

* minsyms.c (add_minsym_to_hash_table): Use a previously computed
hash code if possible.
(add_minsym_to_demangled_hash_table): Likewise.
(minimal_symbol_reader::install): Compute the hash codes for msymbol
on the background thread.
* symtab.h (struct minimal_symbol) <hash_value, demangled_hash_value>:
Add these fields.

Change-Id: Ifaa3346e9998f05743bff9e2eaad3f83b954d071

Precompute hash value for symbol_set_names

We can also compute the hash for the mangled name on a background
thread so make this function even faster (about a 7% speedup).

gdb/ChangeLog:

2019-11-27  Christian Biesinger  <cbiesinger@google.com>

* minsyms.c (minimal_symbol_reader::install): Also compute the hash
of the mangled name on the background thread.
* symtab.c (symbol_set_names): Allow passing in the hash of the
linkage_name.
* symtab.h (symbol_set_names): Likewise.

Change-Id: I044449e7eb60cffc1c43efd3412f2b485bd9faac

gdb/testsuite: Fix race condition compiling fortran test

The Fortran test gdb.fortran/info-modules compiles the files
info-types.f90 and info-types-2.f90 in that order.  Unfortunately
info-types.f90 makes use of a module defined in info-types-2.f90.

This currently doesn't cause a problem if you run all of the Fortran
tests as the info-types.exp test already compiles info-types-2.f90 and
so the module description file 'mod2.mod' will be created, and can
then be found by info-modules.exp during its compile.

If however you try to run just info-modules.exp in a clean build
directory, the test will fail to compile.

Fix this by compiling the source files in the reverse order so that
the module is compiled first, then the test program that uses the
module.

gdb/testsuite/ChangeLog:

* gdb.fortran/info-modules.exp: Compile source files in correct
order.

Change-Id: Ic3a1eded0486f6264ebe3066cf1beafbd2534a91

Test case for BZ 25065

Running a GDB with the fix for BZ 25065 should cause these new tests
to all pass.

When run against a GDB without the fix, there will be 2 unresolved
testcases.  This is what I see in the gdb.sum file when I try it using
a GDB without the fix:

ERROR: GDB process no longer exists
UNRESOLVED: gdb.dwarf2/imported-unit.exp: ptype main::Foo
ERROR: Couldn't send ptype main::foo to GDB.
UNRESOLVED: gdb.dwarf2/imported-unit.exp: ptype main::foo

These are "unresolved" versus outright failures due to the fact that
GDB dies (segfaults) during the running of the test.

gdb/testsuite/ChangeLog:

* gdb.dwarf2/imported-unit.exp: New file.
* gdb.dwarf2/imported-unit.c: New file.

Change-Id: I073fe69b81bd258951615f752df8e95b6e33a271

Fix BZ 25065 - Ensure that physnames are computed for inherited DIEs

This is a fix for BZ 25065.

GDB segfaults when running either gdb.cp/subtypes.exp or
gdb.cp/local.exp in conjunction with using the -flto compiler/linker
flag.

A much simpler program, which was used to help create the test for
this fix, is:

-- doit.cc --
int main()
{
  class Foo {
  public:
    int doit ()
    {
      return 0;
    }
  };

  Foo foo;

  return foo.doit ();
}
-- end doit.cc --

gcc -o doit -flto -g doit.cc
gdb -q doit
Reading symbols from doit...
(gdb) ptype main::Foo
type = class Foo {
Segmentation fault (core dumped)

The segfault occurs due to a NULL physname in
c_type_print_base_struct_union in c-typeprint.c.  Specifically,
calling is_constructor_name() eventually causes the SIGSEGV is this
code in c-typeprint.c:

      const char *physname = TYPE_FN_FIELD_PHYSNAME (f, j);
      int is_full_physname_constructor =
TYPE_FN_FIELD_CONSTRUCTOR (f, j)
|| is_constructor_name (physname)
|| is_destructor_name (physname)
|| method_name[0] == '~';

However, looking at compute_delayed_physnames(), we see that
the TYPE_FN_FIELD_PHYSNAME field should never be NULL.  This
field will be set to "" for NULL physnames:

      physname = dwarf2_physname (mi.name, mi.die, cu);
      TYPE_FN_FIELD_PHYSNAME (fn_flp->fn_fields, mi.index)
= physname ? physname : "";

For this particular case, it turns out that compute_delayed_physnames
wasn't being called, which left TYPE_FN_FIELD_PHYSNAME set to the NULL
value that it started with when that data structure was allocated.

The place to fix it, I think, is towards the end of
inherit_abstract_dies().

My first attempt at fix caused the origin CU's method_list (which is
simply the list of methods whose physnames still need to be computed)
to be added to the CU which is doing the inheriting.  One drawback
with this approach is that compute_delayed_physnames is (eventually)
called with a CU that's different than the CU in which the methods
were found.  It's not clear whether this will cause problems or not.

A safer approach, which is what I ultimately settled on, is to call
compute_delayed_physnames() from inherit_abstract_dies().  One
potential drawback is that all needed types might not be known at that
point.  However, in my testing, I haven't seen a problem along these
lines.

gdb/ChangeLog:

* dwarf2read.c (inherit_abstract_dies): Ensure that delayed
physnames are computed for inherited DIEs.

Change-Id: I6c6ffe96b301a9daab9f653956b89e3a33fa9445

Remove some unnecessary backslashes

I found a couple of unnecessary backslashes in gdb.  This removes
them.

Offhand, I wonder whether this abstract_to_concrete thing could be
done some other way?  This seems possibly expensive.

Anyway, tested by rebuilding.  I'm going to check this in as obvious.

gdb/ChangeLog
2019-11-27  Tom Tromey  <tromey@adacore.com>

* dwarf2read.h (struct dwarf2_per_objfile): Remove unnecessary
backslashes.
* cp-support.c: Remove unnecessary backslashes.

Change-Id: I956c91ae24407eeafec8a731545b45f5222e6a9d

Replace SYMBOL_SET_LINKAGE_NAME with a member function

Easier to read, shorter, and will later make it possible to make the
name field private.

gdb/ChangeLog:

2019-11-27  Christian Biesinger  <cbiesinger@google.com>

* ada-exp.y (write_ambiguous_var): Replace SYMBOL_SET_LINKAGE_NAME
with sym->set_linkage_name.
* coffread.c (coff_read_enum_type): Likewise.
* mdebugread.c (parse_symbol): Likewise.
* stabsread.c (patch_block_stabs): Likewise.
(define_symbol): Likewise.
(read_enum_type): Likewise.
(common_block_end): Likewise.
* symtab.h (struct general_symbol_info) <set_linkage_name>: New
function.
(SYMBOL_SET_LINKAGE_NAME): Remove.
* xcoffread.c (process_xcoff_symbol): Replace SYMBOL_SET_LINKAGE_NAME
with sym->set_linkage_name.

Change-Id: I174a0542c014f1b035070068076308bb8ae79abb

gdb/mi: Add -symbol-info-modules command

Add '-symbol-info-modules', an MI version of the CLI 'info modules'
command.

gdb/ChangeLog:

* mi/mi-cmds.c (mi_cmds): Add 'symbol-info-modules' entry.
* mi/mi-cmds.h (mi_cmd_symbol_info_modules): Declare.
* mi/mi-symbol-cmds.c (mi_cmd_symbol_info_modules): New function.
* NEWS: Mention new MI command.

gdb/testsuite/ChangeLog:

* gdb.mi/mi-fortran-modules-2.f90: New file.
* gdb.mi/mi-fortran-modules.exp: New file.
* gdb.mi/mi-fortran-modules.f90: New file.

gdb/doc/ChangeLog:

* doc/gdb.texinfo (GDB/MI Symbol Query): Document new MI command
-symbol-info-modules.

Change-Id: Ibc618010d1d5f36ae8a8baba4fb9d9d724e62b0f

gdb/mi: Add new commands -symbol-info-{functions,variables,types}

Add new MI commands -symbol-info-functions, -symbol-info-variables,
and -symbol-info-types which correspond to the CLI commands 'info
functions', 'info variables', and 'info types' respectively.

gdb/ChangeLog:

* mi/mi-cmds.c (mi_cmds): Add '-symbol-info-functions',
'-symbol-info-types', and '-symbol-info-variables'.
* mi/mi-cmds.h (mi_cmd_symbol_info_functions): Declare.
(mi_cmd_symbol_info_types): Declare.
(mi_cmd_symbol_info_variables): Declare.
* mi/mi-symbol-cmds.c: Add 'source.h' and 'mi-getopt.h' includes.
(output_debug_symbol): New function.
(output_nondebug_symbol): New function.
(mi_symbol_info): New function.
(mi_info_functions_or_variables): New function.
(mi_cmd_symbol_info_functions): New function.
(mi_cmd_symbol_info_types): New function.
(mi_cmd_symbol_info_variables): New function.
* NEWS: Mention new commands.

gdb/testsuite/ChangeLog:

* gdb.mi/mi-sym-info-1.c: New file.
* gdb.mi/mi-sym-info-2.c: New file.
* gdb.mi/mi-sym-info.exp: New file.

gdb/doc/ChangeLog:

* doc/gdb.texinfo (GDB/MI Symbol Query): Document new MI command
-symbol-info-functions, -symbol-info-types, and
-symbol-info-variables.

Change-Id: Ic2fc6a6750bbce91cdde2344791014e5ef45642d

gdb: Split print_symbol_info into two parts

Split the function print_symbol_info into two parts, the new worker
core returns a string, which print_symbol_info then prints.  This will
be useful in a later commit when some new MI commands will be added
which will use the worker core to fill some MI output fields.

There should be no user visible changes after this commit.

gdb/ChangeLog:

* symtab.c (symbol_to_info_string): New function, most content
moved from print_symbol_info, but updated to return a std::string.
(print_symbol_info): Update to use symbol_to_info_string and print
returned string.
* symtab.h (symbol_to_info_string): Declare new function.

Change-Id: I6454ce43cacb61d32fbadb9e3655e70823085777

gdb: Introduce global_symbol_searcher

Introduce a new class to wrap up the parameters needed for the
function search_symbols, which has now become a member function of
this new class.

The motivation is that search_symbols already takes a lot of
parameters, and a future commit is going to add even more.  This
commit hopefully makes collecting the state required for a search
easier.

As part of this conversion the list of filenames in which to search
has been converted to a std::vector.

There should be no user visible changes after this commit.

gdb/ChangeLog:

* python/python.c (gdbpy_rbreak): Convert to using
global_symbol_searcher.
* symtab.c (file_matches): Convert return type to bool, change
file list to std::vector, update header comment.
(search_symbols): Rename to...
(global_symbol_searcher::search): ...this and update now its
a member function of global_symbol_searcher.  Take account of the
changes to file_matches.
(symtab_symbol_info): Convert to using global_symbol_searcher.
(rbreak_command): Likewise.
(search_module_symbols): Likewise.
* symtab.h (enum symbol_search): Update comment.
(search_symbols): Remove declaration.
(class global_symbol_searcher): New class.

Change-Id: I488ab292a892d9e9e84775c632c5f198b6ad3710

PR23652, Use symbols from debug bfd for _bfd_elf_find_function properly

There were a number of problems with the previous patch.  Firstly,
_bfd_dwarf2_stash_syms didn't do anything when the original file had
dynamic symbols, and secondly, info found by the symbol search didn't
make it out of _bfd_elf_find_nearest_line except in the case of DWARF
functions without external linkage.

PR 23652
* dwarf2.c (_bfd_dwarf2_stash_syms): Break out of loop on finding
matching section.
(_bfd_dwarf2_find_nearest_line): Return an int, with value 2 when
returning info from the symbol table.  Do the _bfd_elf_find_function
search also when !found.  Call _bfd_dwarf2_stash_syms regardless of
symbols.
* elf64-alpha.c (elf64_alpha_find_nearest_line): Accept dwarf2
result of 1 only.
* elfxx-mips.c (_bfd_mips_elf_find_nearest_line): Likewise.
* libbfd-in.h (_bfd_dwarf2_find_nearest_line): Update prototype.
* libbfd.h: Regenerate.

PR23652, Use symbols from debug bfd for _bfd_elf_find_function

Sometimes DWARF info for a function is incomplete, and the function
can be retrieved by examining symbols.  However, when separate debug
files are used it may be that the original file is completely
stripped of symbols.  This patch teaches BFD to look at symbols from
the debug file in that case.

The patch also removes arm_elf_find_function, instead implementing
elf_backend_maybe_function_sym.  arm_elf_find_function was written
before the generic _bfd_elf_find_function called maybe_function_sym.
aarch64 copied arm, so that file gets the same treatment.  There is
some chance this will speed up arm and aarch64 lookup of function/line.

PR 23652
* dwarf2.c (_bfd_dwarf2_stash_syms): New function.
(_bfd_dwarf2_find_nearest_line): Use it here, passing syms to
_bfd_elf_find_function.  Call _bfd_elf_find_function in cases
where _bfd_elf_find_nearest_line would do so.
* elf.c (_bfd_elf_find_nearest_line): Omit _bfd_elf_find_function
for dwarf2.
* elfxx-mips.c (_bfd_mips_elf_find_nearest_line): Similarly.  Tidy.
* elf32-arm.c (elf32_arm_maybe_function_sym): New function.
(elf_backend_maybe_function_sym): Define.
(arm_elf_find_function, elf32_arm_find_nearest_line): Delete.
(bfd_elf32_find_nearest_line): Don't define.
* elfnn-aarch64.c (elfNN_aarch64_maybe_function_sym): New function.
(elf_backend_maybe_function_sym): Define.
(aarch64_elf_find_function, elfNN_aarch64_find_nearest_line): Delete.
(bfd_elfNN_find_nearest_line): Don't define.

Correct R_SH_IND12W handling

Using bfd_vma for insn is to avoid having to worry about sign
propagation in expressions involving insn and sym_value when bfd_vma
is not the same as unsigned long.

* elf32-sh.c (sh_reloc): Use a bfd_vma insn.
(sh_reloc <R_SH_IND12W>): Divide calculated relocation value
by two before applying to insn.  Correct overflow test.
* coff-sh.c (sh_reloc): Likewise.

Automatic date update in version.in

Fix mingw build of gdb

Christian pointed out on irc that the threading series broke the build
on mingw.  This patch fixes the problem, by moving the initialization
of gdb_demangle_attempt_core_dump into the appropriate #if.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tromey@adacore.com>

* cp-support.c (_initialize_cp_support): Conditionally initialize
gdb_demangle_attempt_core_dump.

Change-Id: I9ace0bea75a51f317ea933b607f6b5a94d651eea

Let commands free "name"

This adds a "name_allocated" field to cmd_list_element, so that
commands can own their "name" when necessary.  Then, this changes a
few spots in gdb that currently free the name by hand to instead use
this facility.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* python/py-function.c (fnpy_init): Update.
* value.h (add_internal_function): Adjust declaration.
* value.c (function_destroyer): Remove.
(do_add_internal_function): Don't set destroyer or copy name.
(add_internal_function): Take unique_xmalloc_ptr<char> for name.
Set name_allocated.
* python/py-cmd.c (cmdpy_destroyer): Don't free "name".
(cmdpy_init): Set name_allocated.
* cli/cli-decode.h (struct cmd_list_element) <name_allocated>: New
member.
(~cmd_list_element): Free "name" if needed.

Change-Id: Ie1435cea5bbf4bd92056125f112917c607cbb761

Add add_internal_function overload

add_internal_function sets a command destroyer that frees the doc
string.  However, many callers do not pass in an allocated doc string.

This adds a new overload to clearly differentiate the two cases,
fixing the latent bug.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* value.h (add_internal_function): Add new overload.  Move
documentation from value.h.
* value.c (do_add_internal_function): New function.
(add_internal_function): Use it.  Add new overload.
(function_destroyer): Don't free doc.
* python/py-function.c (fnpy_init): Update.

Change-Id: I3f6df925bc6b3e1bccbad9eeebc487b908bb5a2a

Use cmd_list_element::doc_allocated for Python commands

Python commands manage their "doc" string manually, but
cmd_list_element already has doc_allocated to handle this case.  This
changes the Python code to use the existing facility.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* python/py-cmd.c (cmdpy_destroyer): Don't free "doc".
(cmdpy_init): Set "doc_allocated".

Change-Id: I0014edc117b051bba1f4db267687d231e7fe9b56

Set names of worker threads

This adds some configury so that gdb can set the names of worker
threads.  This makes them show up more nicely when debugging gdb
itself.

2019-11-26  Tom Tromey  <tom@tromey.com>

* gdbsupport/thread-pool.c (thread_pool::set_thread_count): Set
name of worker thread.
* gdbsupport/common.m4 (GDB_AC_COMMON): Check for
pthread_setname_np.
* configure, config.in: Rebuild.

gdb/gdbserver/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* configure, config.in: Rebuild.

Change-Id: I60473d65ae9ae14d8c56ddde39684240c16aaf35

Use run_on_main_thread in gdb.post_event

This changes gdb.post_event to use the new run_on_main_thread
function.  This is somewhat tricky because the Python GIL must be held
while manipulating reference counts.

2019-11-26  Tom Tromey  <tom@tromey.com>

* python/python.c (class gdbpy_gil): New.
(struct gdbpy_event): Add constructor, destructor, operator().
(gdbpy_post_event): Use run_on_main_thread.
(gdbpy_initialize_events): Remove.
(do_start_initialization): Update.

Change-Id: Ie4431e60f328dae48bd96b6c6a8e875e70bda1de

Add maint set/show worker-threads

This adds maint commands to control the number of worker threads that
gdb can use.

2019-11-26  Tom Tromey  <tom@tromey.com>

* NEWS: Add entry.
* maint.c (_initialize_maint_cmds): Add "worker-threads" maint
commands.  Call update_thread_pool_size.
(update_thread_pool_size, maintenance_set_worker_threads): New
functions.
(n_worker_threads): New global.

gdb/doc/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* gdb.texinfo (Maintenance Commands): Document new maint
commands.

Change-Id: I4fb514faa05879d8afe62c77036a4469d57dca2a

Demangle minsyms in parallel

This patch introduces a simple parallel for_each and changes the
minimal symbol reader to use it when computing the demangled name for
a minimal symbol.  This yields a speedup when reading minimal symbols.

2019-11-26  Christian Biesinger  <cbiesinger@google.com>
    Tom Tromey  <tom@tromey.com>

* minsyms.c (minimal_symbol_reader::install): Use
parallel_for_each.
* gdbsupport/parallel-for.h: New file.
* Makefile.in (HFILES_NO_SRCDIR): Add gdbsupport/parallel-for.h.

Change-Id: I220341f70e94dd02df5dd424272c50a5afb64978

Implement a thread pool

This adds a simple thread pool to gdb.  In the end, this seemed
preferable to the approach taken in an earlier version of this series;
namely, starting threads in the parallel-foreach implementation.  This
approach reduces the overhead of starting new threads, and also lets
the user control (in a subsequent patch) exactly how many worker
threads are running.

gdb/ChangeLog
2019-11-26  Christian Biesinger  <cbiesinger@google.com>
    Tom Tromey  <tom@tromey.com>

* gdbsupport/thread-pool.h: New file.
* gdbsupport/thread-pool.c: New file.
* Makefile.in (COMMON_SFILES): Add thread-pool.c.
(HFILES_NO_SRCDIR): Add thread-pool.h.

Change-Id: I597bb642780cb9d578ca92373d2a638efb44fe52

Introduce thread-safe way to handle SIGSEGV

The gdb demangler installs a SIGSEGV handler in order to protect gdb
from demangler bugs.  However, this is not thread-safe, as signal
handlers are global to the process.

This patch changes gdb to always install a global SIGSEGV handler, and
then lets threads indicate their interest in handling the signal by
setting a thread-local variable.

This patch then arranges for the demangler code to use this; being
sure to arrange for calls to warning and the like to be done on the
main thread.

One thing I wondered while writing this patch is if there are any
systems that do not have "sigaction".  If gdb could assume this, it
would simplify this code.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* event-top.h (thread_local_segv_handler): Declare.
* event-top.c (thread_local_segv_handler): New global.
(install_handle_sigsegv, handle_sigsegv): New functions.
(async_init_signals): Install SIGSEGV handler.
* cp-support.c (gdb_demangle_jmp_buf): Change type.  Now
thread-local.
(report_failed_demangle): New function.
(gdb_demangle): Make core_dump_allowed atomic.  Remove signal
handler-setting code, instead use segv_handler.  Run warning code
on main thread.

Change-Id: Ic832bbb033b64744e4b44f14b41db7e4168ce427

Introduce run_on_main_thread

This introduces a way for a callback to be run on the main thread.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* run-on-main-thread.c: New file.
* run-on-main-thread.h: New file.
* unittests/main-thread-selftests.c: New file.
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
main-thread-selftests.c.
(HFILES_NO_SRCDIR): Add run-on-main-thread.h.
(COMMON_SFILES): Add run-on-main-thread.c.

Change-Id: I16ef82f0564e9f8a524bdc64cb31df79a988ad9f

Introduce alternate_signal_stack RAII class

This introduces a new RAII class that temporarily installs an
alternate signal stack (on systems that have sigaltstack); then
changes the one gdb use of sigaltstack to use this class instead.

This will be used in a later patch, when creating new threads that may
want to handle SIGSEGV.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* main.c (setup_alternate_signal_stack): Remove.
(captured_main_1): Use gdb::alternate_signal_stack.
* gdbsupport/alt-stack.h: New file.

Change-Id: I721c047ae9d51a35fd274a6dbc00a58c6440dae6

Add RAII class for blocking gdb signals

This adds configury support and an RAII class that can be used to
temporarily block signals that are used by gdb.  (This class is not
used in this patch, but it split out for easier review.)

The idea of this patch is that these signals should only be delivered
to the main thread.  So, when creating a background thread, they are
temporarily blocked; the blocked state is inherited by the new thread.

The sigprocmask man page says:

    The use of sigprocmask() is unspecified in a multithreaded
    process; see pthread_sigmask(3).

This patch changes gdb to use pthread_sigmask when appropriate, by
introducing a convenience define.

I've updated gdbserver as well, because I had to touch gdbsupport, and
because the threading patches will make it link against the thread
library.

I chose not to touch the NTO code, because I don't know anything about
that platform and because I cannot test it.

Finally, this modifies an existing spot in the Guile layer to use the
new facility.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* gdbsupport/signals-state-save-restore.c (original_signal_mask):
Remove comment.
(save_original_signals_state, restore_original_signals_state): Use
gdb_sigmask.
* linux-nat.c (block_child_signals, restore_child_signals_mask)
(_initialize_linux_nat): Use gdb_sigmask.
* guile/guile.c (_initialize_guile): Use block_signals.
* Makefile.in (HFILES_NO_SRCDIR): Add gdb-sigmask.h.
* gdbsupport/gdb-sigmask.h: New file.
* event-top.c (async_sigtstp_handler): Use gdb_sigmask.
* cp-support.c (gdb_demangle): Use gdb_sigmask.
* gdbsupport/common.m4 (GDB_AC_COMMON): Check for
pthread_sigmask.
* configure, config.in: Rebuild.
* gdbsupport/block-signals.h: New file.

gdb/gdbserver/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* remote-utils.c (block_unblock_async_io): Use gdb_sigmask.
* linux-low.c (linux_wait_for_event_filtered, linux_async): Use
gdb_sigmask.
* configure, config.in: Rebuild.

Change-Id: If3f37dc57dd859c226e9e4d79458a0514746e8c6

Add configure check for std::thread

This adds a configure check for std::thread.  This is needed because
std::thread is not available on some systems, like some versions of
mingw and DJGPP.

This also adds configury to make sure that a threaded gdb links
against the correct threading library (-lpthread or the like), and
passes the right flags (e.g., -pthread) to the compilations.

Note that this also links gdbserver against the thread library.  This
is not strictly necessary at this point in the series, but a later
patch will change gdbsupport to use pthread_sigmask, at which point
this will be needed.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* acinclude.m4: Include ax_pthread.m4.
* Makefile.in (PTHREAD_CFLAGS, PTHREAD_LIBS): New variables.
(INTERNAL_CFLAGS_BASE): Use PTHREAD_CFLAGS.
(CLIBS): Use PTHREAD_LIBS.
(aclocal_m4_deps): Add ax_pthread.m4.
* config.in, configure: Rebuild.
* gdbsupport/common.m4 (GDB_AC_COMMON): Check for std::thread.

gdb/gdbserver/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* Makefile.in (PTHREAD_CFLAGS, PTHREAD_LIBS): New variables.
(INTERNAL_CFLAGS_BASE): Use PTHREAD_CFLAGS.
(GDBSERVER_LIBS): Use PTHREAD_LIBS.
* acinclude.m4: Include ax_pthread.m4.
* config.in, configure: Rebuild.

Change-Id: I00ec55db6077f2615421a93461fc3be57e916aa0

Defer minimal symbol name-setting

Currently the demangled name of a minimal symbol is set when creating
the symbol.  However, there is no intrinsic need to do this.  This
patch instead arranges for the demangling to be done just before the
minsym hash tables are filled.  This will be useful in a later patch.

gdb/ChangeLog
2019-11-26  Tom Tromey  <tom@tromey.com>

* symtab.h (struct minimal_symbol) <name_set>: New member.
* minsyms.c (minimal_symbol_reader::record_full): Copy name.
Don't call symbol_set_names.
(minimal_symbol_reader::install): Call symbol_set_names.

Change-Id: I4fe3993b99fb3a43968067806e294d48e377fd76

Fix crashes due to python GIL released too early

When running GDB tests under Valgrind, various tests are failing due
to invalid memory access.
Here is the stack trace reported by Valgrind, for gdb.base/freebpcmd.exp :
  ==18658== Invalid read of size 8
  ==18658==    at 0x7F9107: is_main (signalmodule.c:195)
  ==18658==    by 0x7F9107: PyOS_InterruptOccurred (signalmodule.c:1730)
  ==18658==    by 0x3696E2: check_quit_flag() (extension.c:829)
  ==18658==    by 0x36980B: restore_active_ext_lang(active_ext_lang_state*) (extension.c:782)
  ==18658==    by 0x48F617: gdbpy_enter::~gdbpy_enter() (python.c:235)
  ==18658==    by 0x47BB71: add_thread_object(thread_info*) (object.h:470)
  ==18658==    by 0x53A84D: operator() (std_function.h:687)
  ==18658==    by 0x53A84D: notify (observable.h:106)
  ==18658==    by 0x53A84D: add_thread_silent(ptid_t) (thread.c:311)
  ==18658==    by 0x3CD954: inf_ptrace_target::create_inferior(char const*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&
  , char**, int) (inf-ptrace.c:139)
  ==18658==    by 0x3FE644: linux_nat_target::create_inferior(char const*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&,
   char**, int) (linux-nat.c:1094)
  ==18658==    by 0x3D5727: run_command_1(char const*, int, run_how) (infcmd.c:633)
  ==18658==    by 0x2C05D1: cmd_func(cmd_list_element*, char const*, int) (cli-decode.c:1948)
  ==18658==    by 0x53F29F: execute_command(char const*, int) (top.c:639)
  ==18658==    by 0x3638EB: command_handler(char const*) (event-top.c:586)
  ==18658==    by 0x36468C: command_line_handler(std::unique_ptr<char, gdb::xfree_deleter<char> >&&) (event-top.c:771)
  ==18658==    by 0x36407C: gdb_rl_callback_handler(char*) (event-top.c:217)
  ==18658==    by 0x5B2A1F: rl_callback_read_char (callback.c:281)
  ==18658==    by 0x36346D: gdb_rl_callback_read_char_wrapper_noexcept() (event-top.c:175)
  ==18658==    by 0x363F70: gdb_rl_callback_read_char_wrapper(void*) (event-top.c:192)
  ==18658==    by 0x3633AF: stdin_event_handler(int, void*) (event-top.c:514)
  ==18658==    by 0x362504: gdb_wait_for_event (event-loop.c:857)
  ==18658==    by 0x362504: gdb_wait_for_event(int) (event-loop.c:744)
  ==18658==    by 0x362676: gdb_do_one_event() [clone .part.11] (event-loop.c:321)
  ==18658==    by 0x3627AD: gdb_do_one_event (event-loop.c:303)
  ==18658==    by 0x3627AD: start_event_loop() (event-loop.c:370)
  ==18658==    by 0x41D35A: captured_command_loop() (main.c:381)
  ==18658==    by 0x41F2A4: captured_main (main.c:1224)
  ==18658==    by 0x41F2A4: gdb_main(captured_main_args*) (main.c:1239)
  ==18658==    by 0x227D0A: main (gdb.c:32)
  ==18658==  Address 0x10 is not stack'd, malloc'd or (recently) free'd

The problem seems to be created by gdbpy_enter::~gdbpy_enter () releasing the GIL lock
too early:
~gdbpy_enter () does:
      ...
      PyGILState_Release (m_state);
      python_gdbarch = m_gdbarch;
      python_language = m_language;

      restore_active_ext_lang (m_previous_active);
    }

So, it releases the GIL lock, does 2 assignments and then leads to the following
call sequence:
  restore_active_ext_lang => check_quit_flag => python.c gdbpy_check_quit_flag
     => PyOS_InterruptOccurred => is_main.
is_main code is:
    static int
    is_main(_PyRuntimeState *runtime)
    {
        unsigned long thread = PyThread_get_thread_ident();
        PyInterpreterState *interp = _PyRuntimeState_GetThreadState(runtime)->interp;
        return (thread == runtime->main_thread
                && interp == runtime->interpreters.main);
    }

The macros and functions to access the thread state are documented as:
    /* Variable and macro for in-line access to current thread
       and interpreter state */

    #define _PyRuntimeState_GetThreadState(runtime) \
        ((PyThreadState*)_Py_atomic_load_relaxed(&(runtime)->gilstate.tstate_current))

    /* Get the current Python thread state.

       Efficient macro reading directly the 'gilstate.tstate_current' atomic
       variable. The macro is unsafe: it does not check for error and it can
       return NULL.

       The caller must hold the GIL.

       See also PyThreadState_Get() and PyThreadState_GET(). */
    #define _PyThreadState_GET() _PyRuntimeState_GetThreadState(&_PyRuntime)

So, we see that GDB releases the GIL and then potentially calls
_PyRuntimeState_GetThreadState that needs the GIL.

It is not very clear why the problem is only observed when running under
Valgrind.  Probably caused by the slowdown due to Valgrind and/or to the 'single
thread' scheduling by Valgrind.

This patch fixes the crashes by releasing the GIT lock later.

2019-11-26  Philippe Waroquiers  <philippe.waroquiers@skynet.be>

* python/python.c (gdbpy_enter::~gdbpy_enter): Release GIL after
restore_active_ext_lang, as GIL is needed for (indirectly)
called PyOS_InterruptOccurred.

Fix declaration of sparc_xfer_wcookie

When building sparc-nat.c with -Wmissing-declarations, we get:

      CXX    sparc-nat.o
    /home/smarchi/src/binutils-gdb/gdb/sparc-nat.c: In function ‘target_xfer_status sparc_xfer_wcookie(target_ops*, target_object, const char*, gdb_byte*, const gdb_byte*, ULONGEST, ULONGEST, ULONGEST*)’:
    /home/smarchi/src/binutils-gdb/gdb/sparc-nat.c:255:1: error: no previous declaration for ‘target_xfer_status sparc_xfer_wcookie(target_ops*, target_object, const char*, gdb_byte*, const gdb_byte*, ULONGEST, ULONGEST, ULONGEST*)’ [-Werror=missing-declarations]
     sparc_xfer_wcookie (struct target_ops *ops, enum target_object object,
     ^~~~~~~~~~~~~~~~~~

Indeed, the declaration is not in sync with the definition, fix that.

sparc_xfer_wcookie is used in sparc_target::xfer_partial.  sparc_target
is only used in the BSD sparc native files.  The error above was
obtained by running "make sparc-nat.o" on Linux with a cross-compiler
for sparc64-linux-gnu.  But I presume that if we were to build for real
with a BSD/sparc compiler, we would end up with an undefined symbol.

gdb/ChangeLog:

* sparc-nat.c (sparc_xfer_wcookie): Sync declaration with
definition.

Change-Id: Id41e706e5516968ff6a49469ddc48eceb29dd3ea

Remove simulator_command declaration, make static

The simulator_command function is not used outside its file, so make it
static.  Remove the declaration, which is not needed and not even in
sync with the definition.

gdb/ChangeLog:

* remote-sim.c (simulator_command): Make static, remove
declaration.

Change-Id: I40bd1e3662f849c4c9970443931ab9ee0ccccea1

Make functions static in unittests

Enabling -Wmissing-declarations points out that a bunch of function in
the unittests can be made static, do that.

gdb/ChangeLog:

* unittests/array-view-selftests.c (check_ptr_size_ctor2): Make
static.
* unittests/basic_string_view/capacity/1.cc (test01): Likewise.
* unittests/basic_string_view/cons/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/cons/char/2.cc (test03): Likewise.
(main): Likewise.
* unittests/basic_string_view/cons/char/3.cc (test05): Likewise.
(main): Likewise.
* unittests/basic_string_view/element_access/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/element_access/char/empty.cc (main): Likewise.
* unittests/basic_string_view/element_access/char/front_back.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/inserters/char/2.cc (test05): Likewise.
(main): Likewise.
* unittests/basic_string_view/modifiers/remove_prefix/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/modifiers/remove_suffix/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/modifiers/swap/char/1.cc (test01): Likewise.
* unittests/basic_string_view/operations/compare/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/compare/char/13650.cc (test01): Likewise.
* unittests/basic_string_view/operations/copy/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/data/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/find/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/find/char/2.cc (test02): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/find/char/3.cc (test03): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/find/char/4.cc (main): Likewise.
* unittests/basic_string_view/operations/rfind/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/rfind/char/2.cc (test02): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/rfind/char/3.cc (test03): Likewise.
(main): Likewise.
* unittests/basic_string_view/operations/substr/char/1.cc (test01): Likewise.
(main): Likewise.
* unittests/basic_string_view/operators/char/2.cc (main): Likewise.
* unittests/optional/assignment/1.cc (test): Likewise.
* unittests/optional/assignment/2.cc (test): Likewise.
* unittests/optional/assignment/3.cc (test): Likewise.
* unittests/optional/assignment/4.cc (test): Likewise.
* unittests/optional/assignment/5.cc (test): Likewise.
* unittests/optional/assignment/6.cc (test): Likewise.
* unittests/optional/assignment/7.cc (test): Likewise.
* unittests/optional/cons/copy.cc (test): Likewise.
* unittests/optional/cons/default.cc (test): Likewise.
* unittests/optional/cons/move.cc (test): Likewise.
* unittests/optional/cons/value.cc (test): Likewise.
* unittests/optional/in_place.cc (test): Likewise.
* unittests/optional/observers/1.cc (test): Likewise.
* unittests/optional/observers/2.cc (test): Likewise.

Change-Id: I66626db864cb877cacc570d4660df633530554f5

Remove declaration of tui_set_var_cmd, make definition static

The declaration of tui_set_var_cmd is not in sync with the definition.
Since tui_set_var_cmd is only used in the file where it's defined,
remove the declaration and make the definition static.

gdb/ChangeLog:

* tui-win.h (tui_set_var_cmd): Remove.
* tui-win.c (tui_set_var_cmd): Make static.

Change-Id: If4bddbfb573347fb7254fb6f1a940052a72f464f

Remove unused rbreak_command_wrapper and other declarations

rbreak_command_wrapper is unused, so remove it.  And while at it, remove
other declarations around it.

gdb/ChangeLog:

* breakpoint.h (hbreak_command_wrapper, thbreak_command_wrapper,
rbreak_command_wrapper): Remove.
* symtab.c (rbreak_command_wrapper): Remove.

Change-Id: If9782f205e4913f8dfc5beeaa526544f25e099c6

Remove info_terminal_command declaration, make definition static

The info_terminal_command declaration in inflow.h does not match the
current definition.  It is not needed anyway, as info_terminal_command
is only used locally, so remove it and make the definition static.

gdb/ChangeLog:

* inferior.h (info_terminal_command): Remove declaration.
* inflow.c (info_terminal_command): Make static.

Change-Id: I22c3fcc44244e3cf877b5e27eff189af11c39503

Remove unused overload of exit_inferior_silent

This function is not used in the code base.

gdb/ChangeLog:

* inferior.c (exit_inferior_silent): Remove.

Change-Id: Ib2b7662744da079185ceac2a165b47590bd3113c

Remove dict_empty/mdict_empty

These functions are not used in the code base, remove them.

gdb/ChangeLog:

* dictionary.c (dict_empty, mdict_empty): Remove.
* dictionary.c (mdict_empty): Remove.

Change-Id: I4c1b08c730f6790b2f3d28b680607618e3c08e48

Make a bunch of functions static

All these functions are only used in their respective files, they are
missing the static keyword, add them.

gdb/ChangeLog:

 * arc-tdep.c (arc_insn_get_memory_base_reg): Make static.
 (arc_insn_get_memory_offset): Likewise.
 (arc_insn_dump): Likewise.
 * cp-support.c (test_cp_symbol_name_matches): Likewise.
 * csky-linux-tdep.c (csky_supply_fregset): Likewise.
 * dictionary.c (dict_iterator_next): Likewise.
 (dict_iter_match_first): Likewise.
 (dict_iter_match_next): Likewise.
 * f-lang.c (evaluate_subexp_f): Likewise.
 * hppa-tdep.c (hppa_read_pc): Likewise.
 * i386-tdep.c (i386_floatformat_for_type): Likewise.
 * parse.c (write_exp_elt_msym): Likewise.
 * ppc-linux-tdep.c (ppc_floatformat_for_type): Likewise.
 * remote.c (remote_packet_size): Likewise.
 (remote_notif_stop_parse): Likewise.
 * rs6000-aix-tdep.c (aix_sighandle_frame_sniffer): Likewise.
 * s12z-tdep.c (s12z_disassemble_info): Likewise.
 * source.c (prepare_path_for_appending): Likewise.
 * sparc64-linux-tdep.c
 (sparc64_linux_handle_segmentation_fault); Likewise.
 * stack.c (frame_selection_by_function_completer): Likewise.

Change-Id: I18e187ad279075b961e3e22e5b034f5c0f6188f0

Remove unused function set_gdb_completion_word_break_characters

gdb/ChangeLog:

* completer.c (set_gdb_completion_word_break_characters):
Remove.

Change-Id: If39b8d01f215a42ea3d01fb8290014613ec0bb8b

Add missing includes in dwarf-index-write.c and mi/mi-interp.c

The following errors show that these files are missing the include of
their matching header, add them.

  CXX    dwarf-index-write.o
/home/smarchi/src/binutils-gdb/gdb/dwarf-index-write.c: In function ‘void write_psymtabs_to_index(dwarf2_per_objfile*, const char*, const char*, const char*, dw_index_kind)’:
/home/smarchi/src/binutils-gdb/gdb/dwarf-index-write.c:1670:1: error: no previous declaration for ‘void write_psymtabs_to_index(dwarf2_per_objfile*, const char*, const char*, const char*, dw_index_kind)’ [-Werror=missing-declarations]
 write_psymtabs_to_index (struct dwarf2_per_objfile *dwarf2_per_objfile,
 ^~~~~~~~~~~~~~~~~~~~~~~

  CXX    mi/mi-interp.o
/home/smarchi/src/binutils-gdb/gdb/mi/mi-interp.c: In function ‘void mi_output_solib_attribs(ui_out*, so_list*)’:
/home/smarchi/src/binutils-gdb/gdb/mi/mi-interp.c:1030:1: error: no previous declaration for ‘void mi_output_solib_attribs(ui_out*, so_list*)’ [-Werror=missing-declarations]
 mi_output_solib_attribs (ui_out *uiout, struct so_list *solib)
 ^~~~~~~~~~~~~~~~~~~~~~~

gdb/ChangeLog:

* dwarf-index-write.c: Include dwarf-index-write.h.
* mi/mi-interp.c: Include mi/mi-interp.h.

Change-Id: I0103b8669e16e0fcaa476f8c5e96f49608157745

Include aarch32-tdep.h in aarch32-tdep.c

The error below shows that aarch32-tdep.c is missing an include for
aarch32-tdep.h, add it.

      CXX    aarch32-tdep.o
    /home/smarchi/src/binutils-gdb/gdb/aarch32-tdep.c: In function ‘const target_desc* aarch32_read_description()’:
    /home/smarchi/src/binutils-gdb/gdb/aarch32-tdep.c:27:1: error: no previous declaration for ‘const target_desc* aarch32_read_description()’ [-Werror=missing-declarations]
     aarch32_read_description ()
     ^~~~~~~~~~~~~~~~~~~~~~~~

Putting the include of aarch32-tdep.h early in aarch32-tdep.c gives us
an error about target_desc not being defined.  Indeed, aarch32-tdep.h
uses target_desc without forward-declaring it or including the proper
header.  Add a forward-declaration for it.

gdb/ChangeLog:

* aarch32-tdep.c: Include aarch32-tdep.h.
* aarch32-tdep.h: Forward-declare struct target_desc.

Change-Id: Ica4be4de0fbd7f22d56a29a40fbf0a31b5abdb16

Use safe_strerror instead of strerror where possible

This provides threadsafety. Unfortunately, since libinproctrace.so
does not link to gnulib, we can't use it there, especially since it
still includes the gnulib headers (so it is difficult to directly
call the system strerror_r).

gdb/ChangeLog:

2019-11-26  Christian Biesinger  <cbiesinger@google.com>

* linux-nat.c (detach_one_lwp): Call safe_strerror instead of
strerror.
* nto-procfs.c (nto_procfs_target::create_inferior): Likewise.
* windows-nat.c (windows_nat_target::create_inferior): Likewise.

gdb/gdbserver/ChangeLog:

2019-11-26  Christian Biesinger  <cbiesinger@google.com>

* debug.c (debug_set_output): Call safe_strerror instead of
strerror.
* linux-low.c (attach_proc_task_lwp_callback): Likewise.
(linux_kill_one_lwp): Likewise.
(linux_detach_one_lwp): Likewise.
(linux_wait_for_event_filtered): Likewise.
(store_register): Likewise.
* lynx-low.c (lynx_attach): Likewise.
* mem-break.c (insert_memory_breakpoint): Likewise.
(remove_memory_breakpoint): Likewise.
(delete_fast_tracepoint_jump): Likewise.
(set_fast_tracepoint_jump): Likewise.
(uninsert_fast_tracepoint_jumps_at): Likewise.
(reinsert_fast_tracepoint_jumps_at): Likewise.
* nto-low.c (nto_xfer_memory): Likewise.
(nto_resume): Likewise.

Change-Id: I9e259cdcaa6e11bbcc4ee6bdc5b7127d73e11abe

Fix gdbserver ChangeLog entry

Christian pointed out that I had accidentally put a ChangeLog entry
into gdbserver that was meant for testsuite.

I'm checking in this patch to fix it.

Change-Id: Iba6124cea6f63539ad66494d3355fb657b78a66d

Introduce new .text.sorted.* sections.

gold/ChangeLog:

2019-11-26  Martin Liska  <mliska@suse.cz>

* layout.cc (Layout::special_ordering_of_input_section):
Add ".text.sorted".
* output.cc: Special case ".text.sorted".
* testsuite/section_sorting_name.cc: Cover also .text.sorted
subsections.
* testsuite/section_sorting_name.sh: Likewise.

ld/ChangeLog:

2019-11-26  Martin Liska  <mliska@suse.cz>

* scripttempl/arclinux.sc: Add .text.sorted.* which is sorted
by default.
* scripttempl/elf.sc: Likewise.
* scripttempl/elf64bpf.sc: Likewise.
* scripttempl/nds32elf.sc: Likewise.
* testsuite/ld-arm/arm-no-rel-plt.ld: Expect .text.sorted.*
in the default linker script.
* testsuite/ld-arm/fdpic-main.ld: Likewise.
* testsuite/ld-arm/fdpic-shared.ld: Likewise.

Fix comparison operations in SH code that trigger warning in clang.

* elf32-sh.c (sh_elf_reloc): Use a signed_vma when checking for a
negative relocated value.
* coff-sh.c (sh_reloc): Likewise.

Automatic date update in version.in

[gdb/contrib] Add -c option to words.sh script

The words.sh script in its current form extracts c comments from files, which
it then transforms into a list of words.

To use the script on the documentation (as I did for commit 6b92c0d3533
"[gdb/doc] Fix typos"), I needed to disable the "extract c comments" part.

Add an option -c that enables extracting c comments, and is off by default.

gdb/ChangeLog:

2019-11-25  Tom de Vries  <tdevries@suse.de>

* contrib/words.sh: Add -c option.

Change-Id: Ifa34d435b3c41b3ff845dc07ae4b0d9f02d92a2d

Replace int with bool in solib.c

This does not touch "int from_tty" and a couple of other instances
that require a bigger change.

gdb/ChangeLog:

2019-11-25  Christian Biesinger  <cbiesinger@google.com>

* solib.c (solib_find_1): Change int to bool.
(exec_file_find): Change int to bool.
(solib_find): Change int to bool.
(solib_read_symbols): Change int to bool.
(solib_used): Change int to bool.
(solib_add): Change int to bool.
(info_sharedlibrary_command): Change int to bool.
(solib_contains_address_p): Change int to bool.
(solib_keep_data_in_core): Change int to bool.
(in_solib_dynsym_resolve_code): Change int to bool.
(reload_shared_libraries_1): Change int to bool.
(gdb_sysroot_changed): Change int to bool.
* solib.h (solib_read_symbols): Change int to bool.
(solib_contains_address_p): Change int to bool.
(solib_keep_data_in_core): Change int to bool.
(in_solib_dynsym_resolve_code): Change int to bool.
(libpthread_name_p): Change int to bool.

Change-Id: Id695ed4ed0c3526af477d4d2bf585a7193c36cab

Fix "psb CSYNC" and "bti C".

psb CYSNC was not finding that CSYNC was a correct spelling.
The problem was upper case version was being put in the
wrong hashtable.  This fixes the problem by using the
correct hashtable.
Also adds testcases for the upper case versions.

        * config/tc-aarch64.c (md_begin): Use correct
        hash table for uppercase version of hint.
        * testsuite/gas/aarch64/system-2.s: Extend psb case to uppercase.
        * testsuite/gas/aarch64/system-2.d: Update.

Change-Id: If43f8b85cacd24840d596c3092b0345e5f212766

[Debugging output] Make remote packet truncation length  adjustable

While debugging, i felt the need to adjust the truncation length of remote
packets so i could see more or less data as needed. The default is currently
set to 512 bytes.

This patch makes this option adjustable through the new "set debug
remote-packet-max-chars" command. It can be set to unlimited if we want to
completely disable truncation.

Update on v5:

- Adjusted function and variable documentation, NEWS entry and GDB manual.

gdb/ChangeLog:

2019-11-25  Luis Machado  <luis.machado@linaro.org>

* NEWS (New Commands): Mention "set debug remote-packet-max-chars".
* remote.c (REMOTE_DEBUG_MAX_CHAR): Remove.
(remote_packet_max_chars): New static global.
(show_remote_packet_max_chars): New function.
(remote_target::putpkt_binary): Adjust to use new
remote_packet_max_chars option.
(remote_target::getpkt_or_notif_sane_1): Likewise.
(_initialize_remote): Register new remote-packet-max-chars option.

gdb/doc/ChangeLog:

2019-11-25  Luis Machado  <luis.machado@linaro.org>

* gdb.texinfo (Debugging Output): Document set debug
remote-packet-max-chars.

Change-Id: I2e871b37bfcaa6376537c3fe3db8f016dd806a7c

Pass section when available to bfd_octets_per_byte

and other tidies.  I think it's better to default to passing the
section to bfd_octets_per_byte, even in cases where we know it won't
make a difference.

A number of the coff reloc functions used bfd_octets_per_byte wrongly,
not factoring it into the offset into the data buffer.  As it happens,
the targets using those files always had bfd_octets_per_byte equal to
one, so there wasn't any detectable wrong behaviour.  However, it is
wrong in the source and might cause trouble for anyone creating a new
target.  Besides fixing that, the patch also defines OCTETS_PER_BYTE
as one in target files where that is appropriate.

bfd/
* archures.c (bfd_octets_per_byte): Tail call
bfd_arch_mach_octets_per_byte.
* coff-arm.c (OCTETS_PER_BYTE): Define.
(coff_arm_reloc): Introduce new "octets" temp.  Use OCTETS_PER_BYTE
with section.  Correct "addr".  Remove ATTRIBUTE_UNUSED.
* coff-i386.c (coff_i386_reloc): Similarly.
* coff-mips.c (mips_reflo_reloc): Similarly.
* coff-x86_64.c (coff_amd64_reloc): Similarly.
* elf32-msp430.c (OCTETS_PER_BYTE): Define.
(rl78_sym_diff_handler): Use OCTETS_PER_BYTE, with section.
* elf32-nds32.c (nds32_elf_get_relocated_section_contents): Similarly.
* elf32-ppc.c (ppc_elf_addr16_ha_reloc): Similarly.
* elf32-pru.c (pru_elf32_do_ldi32_relocate): Similarly.
* elf32-s12z.c (opru18_reloc): Similarly.
* elf32-sh.c (sh_elf_reloc): Similarly.
* elf32-spu.c (spu_elf_rel9): Similarly.
* elf32-xtensa.c (bfd_elf_xtensa_reloc): Similarly.
* elf64-ppc.c (ppc64_elf_ha_reloc, ppc64_elf_brtaken_reloc),
(ppc64_elf_toc64_reloc): Similarly.
* bfd.c (bfd_get_section_limit): Pass section to bfd_octets_per_byte.
* cofflink.c (_bfd_coff_link_input_bfd),
(_bfd_coff_reloc_link_order): Likewise.
* elf.c (_bfd_elf_section_offset): Likewise.
* elflink.c (resolve_section, bfd_elf_perform_complex_relocation),
(elf_link_input_bfd, elf_reloc_link_order, elf_fixup_link_order),
(bfd_elf_final_link): Likewise.
* elf.c (_bfd_elf_make_section_from_shdr): Don't strncmp twice
to set SEC_ELF_OCTETS.
* reloc.c (bfd_perform_relocation): Tidy SEC_ELF_OCTETS special case.
(bfd_install_relocation): Likewise.
(_bfd_final_link_relocate): Don't recalculate octets.
* syms.c (_bfd_stab_section_find_nearest_line): Introduc new
"octets" temp.
* bfd-in2.h: Regenerate.
ld/
* ldexp.c (fold_name): Pass section to bfd_octets_per_byte.
* ldlang.c (init_opb): Don't call bfd_arch_mach_octets_per_byte
unnecessarily.

Introduce new section flag: SEC_ELF_OCTETS

All symbols, sizes and relocations in this section are octets instead of
bytes.  Required for DWARF debug sections as DWARF information is
organized in octets, not bytes.

bfd/
* section.c (struct bfd_section): New flag SEC_ELF_OCTETS.
* archures.c (bfd_octets_per_byte): New parameter sec.
If section is not NULL and SEC_ELF_OCTETS is set, one octet es
returned [ELF targets only].
* bfd.c (bfd_get_section_limit): Provide section parameter to
bfd_octets_per_byte.
* bfd-in2.h: regenerate.
* binary.c (binary_set_section_contents): Move call to
bfd_octets_per_byte into section loop. Provide section parameter
to bfd_octets_per_byte.
* coff-arm.c (coff_arm_reloc): Provide section parameter
to bfd_octets_per_byte.
* coff-i386.c (coff_i386_reloc): likewise.
* coff-mips.c (mips_reflo_reloc): likewise.
* coff-x86_64.c (coff_amd64_reloc): likewise.
* cofflink.c (_bfd_coff_link_input_bfd): likewise.
(_bfd_coff_reloc_link_order): likewise.
* elf.c (_bfd_elf_section_offset): likewise.
(_bfd_elf_make_section_from_shdr): likewise.
Set SEC_ELF_OCTETS for sections with names .gnu.build.attributes,
.debug*, .zdebug* and .note.gnu*.
* elf32-msp430.c (rl78_sym_diff_handler): Provide section parameter
to bfd_octets_per_byte.
* elf32-nds.c (nds32_elf_get_relocated_section_contents): likewise.
* elf32-ppc.c (ppc_elf_addr16_ha_reloc): likewise.
* elf32-pru.c (pru_elf32_do_ldi32_relocate): likewise.
* elf32-s12z.c (opru18_reloc): likewise.
* elf32-sh.c (sh_elf_reloc): likewise.
* elf32-spu.c (spu_elf_rel9): likewise.
* elf32-xtensa.c (bfd_elf_xtensa_reloc): likewise
* elf64-ppc.c (ppc64_elf_brtaken_reloc): likewise.
(ppc64_elf_addr16_ha_reloc): likewise.
(ppc64_elf_toc64_reloc): likewise.
* elflink.c (bfd_elf_final_link): likewise.
(bfd_elf_perform_complex_relocation): likewise.
(elf_fixup_link_order): likewise.
(elf_link_input_bfd): likewise.
(elf_link_sort_relocs): likewise.
(elf_reloc_link_order): likewise.
(resolve_section): likewise.
* linker.c (_bfd_generic_reloc_link_order): likewise.
(bfd_generic_define_common_symbol): likewise.
(default_data_link_order): likewise.
(default_indirect_link_order): likewise.
* srec.c (srec_set_section_contents): likewise.
(srec_write_section): likewise.
* syms.c (_bfd_stab_section_find_nearest_line): likewise.
* reloc.c (_bfd_final_link_relocate): likewise.
(bfd_generic_get_relocated_section_contents): likewise.
(bfd_install_relocation): likewise.
For section which have SEC_ELF_OCTETS set, multiply output_base
and output_offset with bfd_octets_per_byte.
(bfd_perform_relocation): likewise.
include/
* coff/ti.h (GET_SCNHDR_SIZE, PUT_SCNHDR_SIZE, GET_SCN_SCNLEN),
(PUT_SCN_SCNLEN): Adjust bfd_octets_per_byte calls.
binutils/
* objdump.c (disassemble_data): Provide section parameter to
bfd_octets_per_byte.
(dump_section): likewise
(dump_section_header): likewise. Show SEC_ELF_OCTETS flag if set.
gas/
* as.h: Define SEC_OCTETS as SEC_ELF_OCTETS if OBJ_ELF.
* dwarf2dbg.c: (dwarf2_finish): Set section flag SEC_OCTETS for
.debug_line, .debug_info, .debug_abbrev, .debug_aranges, .debug_str
and .debug_ranges sections.
* write.c (maybe_generate_build_notes): Set section flag
SEC_OCTETS for .gnu.build.attributes section.
* frags.c (frag_now_fix): Don't divide by OCTETS_PER_BYTE if
SEC_OCTETS is set.
* symbols.c (resolve_symbol_value): Likewise.
ld/
* ldexp.c (fold_name): Provide section parameter to
bfd_octets_per_byte.
* ldlang (init_opb): New argument s. Set opb_shift to 0 if
SEC_ELF_OCTETS for the current section is set.
(print_input_section): Pass current section to init_opb.
(print_data_statement,print_reloc_statement,
print_padding_statement): Likewise.
(lang_check_section_addresses): Call init_opb for each
section.
(lang_size_sections_1,lang_size_sections_1,
lang_do_assignments_1): Likewise.
(lang_process): Pass NULL to init_opb.

Reverts patches providing octet support in dwarf

Reverts "dwarf2: Align relocation within .debug_line section"
commit 204f543cae7a5dc908264b49d558191d0ceb989c,
Reverts "dwarf2: Pad size of .debug_line section."
commit 145c4477d239fef4e31a457ff8a1ba7153e9a448,
Reverts "dwarf2: Use octets for .debug_line prologue"
commit 38c24f42c97af59ad83505ed735e689c63d3ca45,
Mostly reverts "dwarf2: Use octets for dwarf2 headers"
commit 7235427998571b6d8267e7ac72a52d7b082f4c2b,
Mostly reverts "Symbols with octets value"
commit d18d199917337537713f9fc4b7ae4d6568f740cf.

* dwarf2dbg.c (out_set_addr): Revert 2019-03-13 change.
(out_debug_line, out_debug_aranges, out_debug_info): Likewise.
* symbols.h (symbol_set_value_now_octets, symbol_octets_p): Remove.
* symbols.c (struct symbol_flags): Remove member sy_octets.
(symbol_temp_new_now_octets): Don't set symbol_flags::sy_octets.
(resolve_symbol_value): Revert: Return octets instead of bytes if
sy_octets is set.
(symbol_set_value_now_octets): Remove.
(symbol_octets_p): Remove.

Automatic date update in version.in

Include gdbarch.h in m68k-linux-nat.c

Fix this compilation error, and a bunch of similar ones:

      CXX    m68k-linux-nat.o
    /home/smarchi/src/binutils-gdb/gdb/m68k-linux-nat.c: In function ‘void fetch_register(regcache*, int)’:
    /home/smarchi/src/binutils-gdb/gdb/m68k-linux-nat.c:133:9: error: ‘gdbarch_register_name’ was not declared in this scope
             gdbarch_register_name (gdbarch, regno),
             ^~~~~~~~~~~~~~~~~~~~~

gdb/ChangeLog:

* m68k-linux-nat.c: Include gdbarch.h.

Change-Id: I7cd47bc5d094241b2596e29c244eb55ed11f7a02

Use bool in require_partial_symbols

This changes require_partial_symbols to use bool as its parameter
type.

gdb/ChangeLog
2019-11-24  Tom Tromey  <tom@tromey.com>

* symfile.c (read_symbols): Update.
* psymtab.c (require_partial_symbols): Change type of "verbose" to
bool.
(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): Update.
* psymtab.h (require_partial_symbols): Change type of "verbose" to
bool.

Change-Id: Iae87aa5e4590706bb9e90a33adb86f1fe0fbf3c7

Automatic date update in version.in

Automatic date update in version.in

Restore parameter names in observable.h

Ages ago, when we switched observables to be templates, Joel asked me
to restore the parameter names that were used in the old
observer.texi.

I've finally done this, putting the names into comments.  I also
updated the comments in this file to use the GNU metasyntactic
variable convention as well.

gdb/ChangeLog
2019-11-22  Tom Tromey  <tom@tromey.com>

* observable.h: Update comments.

Change-Id: Id71bea7a7fcaa8f5d4491f33aa8861c56ba9c3f0

Avoid crash in print_ada_task_info

In MI mode, print_ada_task_info can crash in find_thread_ptid when
trying to print an Ada task that is no longer alive.  This patch
avoids the problem by checking for this case.

Because this is Ada-specific, and because Joel approved it internally,
I am checking it in.

gdb/ChangeLog
2019-11-22  Tom Tromey  <tromey@adacore.com>

* ada-tasks.c (ada_task_is_alive): Make parameter const.
(print_ada_task_info): Don't try to fetch thread id if task is not
alive.

gdb/gdbserver/ChangeLog
2019-11-22  Tom Tromey  <tromey@adacore.com>

* gdb.ada/tasks.exp: Add -ada-task-info regression test.
* gdb.ada/tasks/foo.adb: Add another stopping location.

Change-Id: If25eae6507eebb7537eb8adbcbaa1fc1eec88f5c

Replace SYMBOL_*_NAME accessors with member functions

Similar to the MSYMBOL version of this patch, improves readability
and will eventually allow making name private.

gdb/ChangeLog:

2019-11-22  Christian Biesinger  <cbiesinger@google.com>

* ada-exp.y: Update.
* ada-lang.c (sort_choices): Update.
(ada_print_symbol_signature): Update.
(resolve_subexp): Update.
(ada_parse_renaming): Update.
(ada_read_renaming_var_value): Update.
(lesseq_defined_than): Update.
(remove_extra_symbols): Update.
(remove_irrelevant_renamings): Update.
(ada_add_block_symbols): Update.
(ada_collect_symbol_completion_matches): Update.
(ada_is_renaming_symbol): Update.
(aggregate_assign_from_choices): Update.
(ada_evaluate_subexp): Update.
(ada_has_this_exception_support): Update.
(ada_is_non_standard_exception_sym): Update.
(ada_add_exceptions_from_frame): Update.
(ada_add_global_exceptions): Update.
(ada_print_subexp): Update.
* ax-gdb.c (gen_var_ref): Update.
(gen_maybe_namespace_elt): Update.
(gen_expr_for_cast): Update.
(gen_expr): Update.
* block.h: Update.
* blockframe.c (find_pc_partial_function): Update.
* breakpoint.c (print_breakpoint_location): Update.
(update_static_tracepoint): Update.
* btrace.c (ftrace_print_function_name): Update.
(ftrace_function_switched): Update.
* buildsym.c (find_symbol_in_list): Update.
* c-exp.y: Update.
* c-typeprint.c (c_print_typedef): Update.
(c_type_print_template_args): Update.
* cli/cli-cmds.c (edit_command): Update.
(list_command): Update.
(print_sal_location): Update.
* coffread.c (patch_opaque_types): Update.
(process_coff_symbol): Update.
(coff_read_enum_type): Update.
* compile/compile-c-symbols.c (c_symbol_substitution_name): Update.
(convert_one_symbol): Update.
(hash_symname): Update.
(eq_symname): Update.
* compile/compile-cplus-symbols.c (convert_one_symbol): Update.
* compile/compile-cplus-types.c (debug_print_scope): Update.
* compile/compile-loc2c.c (do_compile_dwarf_expr_to_c): Update.
* compile/compile-object-load.c (get_out_value_type): Update.
* cp-namespace.c (cp_scan_for_anonymous_namespaces): Update.
(search_symbol_list): Update.
(cp_lookup_symbol_imports_or_template): Update.
* cp-support.c (overload_list_add_symbol): Update.
* ctfread.c (psymtab_to_symtab): Update.
* dbxread.c (cp_set_block_scope): Update.
* dictionary.c (iter_match_first_hashed): Update.
(iter_match_next_hashed): Update.
(insert_symbol_hashed): Update.
(iter_match_next_linear): Update.
* dictionary.h: Update.
* dwarf2loc.c (func_get_frame_base_dwarf_block): Update.
(locexpr_describe_location_piece): Update.
(locexpr_describe_location_1): Update.
(locexpr_generate_c_location): Update.
(loclist_describe_location): Update.
(loclist_generate_c_location): Update.
* dwarf2read.c (dw2_debug_names_lookup_symbol): Update.
(read_func_scope): Update.
(process_enumeration_scope): Update.
(new_symbol): Update.
(dwarf2_const_value): Update.
(dwarf2_symbol_mark_computed): Update.
* eval.c (evaluate_funcall): Update.
(evaluate_subexp_standard): Update.
* expprint.c (print_subexp_standard): Update.
(dump_subexp_body_standard): Update.
* f-valprint.c (info_common_command_for_block): Update.
* findvar.c (get_hosting_frame): Update.
(default_read_var_value): Update.
* go-lang.c (go_symbol_package_name): Update.
* guile/scm-block.c (bkscm_print_block_smob): Update.
* guile/scm-symbol.c (syscm_print_symbol_smob): Update.
(gdbscm_symbol_name): Update.
(gdbscm_symbol_linkage_name): Update.
(gdbscm_symbol_print_name): Update.
* infcall.c (get_function_name): Update.
* infcmd.c (jump_command): Update.
(finish_command): Update.
* infrun.c (insert_exception_resume_breakpoint): Update.
* linespec.c (canonicalize_linespec): Update.
(create_sals_line_offset): Update.
(convert_linespec_to_sals): Update.
(complete_label): Update.
(find_label_symbols_in_block): Update.
* m2-typeprint.c (m2_print_typedef): Update.
* mdebugread.c (mdebug_reg_to_regnum): Update.
(parse_symbol): Update.
(mylookup_symbol): Update.
* mi/mi-cmd-stack.c (list_arg_or_local): Update.
(list_args_or_locals): Update.
* objc-lang.c (compare_selectors): Update.
(info_selectors_command): Update.
(compare_classes): Update.
(info_classes_command): Update.
(find_imps): Update.
* p-typeprint.c (pascal_print_typedef): Update.
* printcmd.c (build_address_symbolic): Update.
(info_address_command): Update.
(print_variable_and_value): Update.
* python/py-framefilter.c (extract_sym): Update.
(py_print_single_arg): Update.
* python/py-symbol.c (sympy_str): Update.
(sympy_get_name): Update.
(sympy_get_linkage_name): Update.
* python/python.c (gdbpy_rbreak): Update.
* record-btrace.c (btrace_get_bfun_name): Update.
(btrace_call_history): Update.
* rust-lang.c (rust_print_typedef): Update.
* solib-frv.c (frv_fdpic_find_canonical_descriptor): Update.
* stabsread.c (stab_reg_to_regnum): Update.
(define_symbol): Update.
(read_enum_type): Update.
(common_block_end): Update.
(cleanup_undefined_types_1): Update.
(scan_file_globals): Update.
* stack.c (print_frame_arg): Update.
(print_frame_args): Update.
(find_frame_funname): Update.
(info_frame_command_core): Update.
(iterate_over_block_locals): Update.
(print_block_frame_labels): Update.
(do_print_variable_and_value): Update.
(iterate_over_block_arg_vars): Update.
(return_command): Update.
* symmisc.c (dump_symtab_1): Update.
(print_symbol): Update.
* symtab.c (eq_symbol_entry): Update.
(symbol_cache_dump): Update.
(lookup_language_this): Update.
(find_pc_sect_line): Update.
(skip_prologue_sal): Update.
(symbol_search::compare_search_syms): Update.
(treg_matches_sym_type_name): Update.
(search_symbols): Update.
(print_symbol_info): Update.
(rbreak_command): Update.
(completion_list_add_symbol): Update.
(find_gnu_ifunc): Update.
(get_symbol_address): Update.
(search_module_symbols): Update.
(info_module_subcommand): Update.
* symtab.h (SYMBOL_NATURAL_NAME): Remove.
(SYMBOL_LINKAGE_NAME): Remove.
(SYMBOL_DEMANGLED_NAME): Remove.
(SYMBOL_PRINT_NAME): Remove.
(SYMBOL_SEARCH_NAME): Remove.
* tracepoint.c (set_traceframe_context): Update.
(validate_actionline): Update.
(collection_list::collect_symbol): Update.
(encode_actions_1): Update.
(info_scope_command): Update.
(print_one_static_tracepoint_marker): Update.
* typeprint.c (typedef_hash_table::add_template_parameters): Update.
* valops.c (address_of_variable): Update.
(find_overload_match): Update.
(find_oload_champ): Update.

Change-Id: I76bdc8b44eea6876bf03af9d351f8e90cc0154b2

Replace the MSYMBOL_*_NAME macros with member functions

Improves readability. In the future, it will also allow making the name
private, once the name setter functions become member functions.

gdb/ChangeLog:

2019-11-22  Christian Biesinger  <cbiesinger@google.com>

* ada-lang.c (ada_lookup_simple_minsym): Update.
(ada_collect_symbol_completion_matches): Update.
* ada-tasks.c (read_atcb): Update.
* amd64-windows-tdep.c (amd64_skip_main_prologue): Update.
(amd64_windows_skip_trampoline_code): Update.
* arm-tdep.c (skip_prologue_function): Update.
(arm_skip_stack_protector): Update.
* arm-wince-tdep.c (arm_pe_skip_trampoline_code): Update.
(arm_wince_skip_main_prologue): Update.
* ax-gdb.c (gen_expr): Update.
* block.c (call_site_for_pc): Update.
* blockframe.c (find_pc_partial_function): Update.
* breakpoint.c (set_breakpoint_location_function): Update.
* btrace.c (ftrace_print_function_name): Update.
(ftrace_function_switched): Update.
* c-valprint.c (print_unpacked_pointer): Update.
* coffread.c (coff_symfile_read): Update.
* compile/compile-c-symbols.c (convert_symbol_bmsym): Update.
* compile/compile-cplus-symbols.c (convert_symbol_bmsym): Update.
* dwarf-index-write.c (write_psymbols): Update.
* dwarf2loc.c (call_site_to_target_addr): Update.
(func_verify_no_selftailcall): Update.
(tailcall_dump): Update.
(call_site_find_chain_1): Update.
(dwarf_expr_reg_to_entry_parameter): Update.
* elfread.c (elf_gnu_ifunc_record_cache): Update.
* eval.c (evaluate_funcall): Update.
(evaluate_subexp_standard): Update.
(evaluate_subexp_for_sizeof): Update.
* expprint.c (print_subexp_standard): Update.
(dump_subexp_body_standard): Update.
* frame.c (get_prev_frame_always_1): Update.
* frv-tdep.c (frv_skip_main_prologue): Update.
* gnu-v2-abi.c (gnuv2_value_rtti_type): Update.
* gnu-v3-abi.c (gnuv3_rtti_type): Update.
(gnuv3_get_typename_from_type_info): Update.
(gnuv3_skip_trampoline): Update.
* hppa-tdep.c (hppa_lookup_stub_minimal_symbol): Update.
* i386-tdep.c (i386_skip_main_prologue): Update.
(i386_pe_skip_trampoline_code): Update.
* ia64-tdep.c (ia64_convert_from_func_ptr_addr): Update.
* infcall.c (get_function_name): Update.
* linespec.c (minsym_found): Update.
* linux-fork.c (info_checkpoints_command): Update.
* m32c-tdep.c (m32c_m16c_address_to_pointer): Update.
(m32c_m16c_pointer_to_address): Update.
* maint.c (maintenance_translate_address): Update.
* minsyms.c (add_minsym_to_hash_table): Update.
(add_minsym_to_demangled_hash_table): Update.
(lookup_minimal_symbol_mangled): Update.
(lookup_minimal_symbol_demangled): Update.
(lookup_minimal_symbol_linkage): Update.
(lookup_minimal_symbol_text): Update.
(lookup_minimal_symbol_by_pc_name): Update.
(minimal_symbol_is_less_than): Update.
(compact_minimal_symbols): Update.
(build_minimal_symbol_hash_tables): Update.
(find_solib_trampoline_target): Update.
* mips-tdep.c (mips_stub_frame_sniffer): Update.
(mips_skip_pic_trampoline_code): Update.
* msp430-tdep.c (msp430_skip_trampoline_code): Update.
* objc-lang.c (info_selectors_command): Update.
(info_classes_command): Update.
(find_methods): Update.
(find_imps): Update.
* p-valprint.c (pascal_val_print): Update.
* ppc-linux-tdep.c (powerpc_linux_in_dynsym_resolve_code): Update.
* ppc-sysv-tdep.c (convert_code_addr_to_desc_addr): Update.
* printcmd.c (build_address_symbolic): Update.
(info_symbol_command): Update.
* psymtab.c (psymbol_name_matches): Update.
(match_partial_symbol): Update.
(lookup_partial_symbol): Update.
(print_partial_symbols): Update.
(sort_pst_symbols): Update.
(maintenance_check_psymtabs): Update.
* python/py-framefilter.c (py_print_frame): Update.
* python/python.c (gdbpy_rbreak): Update.
* record-btrace.c (btrace_get_bfun_name): Update.
(btrace_call_history): Update.
* rs6000-tdep.c (rs6000_skip_main_prologue): Update.
(rs6000_skip_trampoline_code): Update.
* sol-thread.c (info_cb): Update.
* stabsread.c (scan_file_globals): Update.
* stack.c (find_frame_funname): Update.
(info_frame_command_core): Update.
* symmisc.c (dump_msymbols): Update.
* symtab.c (symbol_natural_name): Rename to..,
(general_symbol_info::natural_name): ...this.
(symbol_demangled_name): Rename to...
(general_symbol_info::demangled_name): ...this.
(symbol_search_name): Rename to...
(general_symbol_info::search_name): ...this.
(symbol_matches_search_name): Update.
(find_pc_sect_line): Update.
(skip_prologue_sal): Update.
(search_symbols): Update.
(print_msymbol_info): Update.
(rbreak_command): Update.
(completion_list_add_msymbol): Update.
(completion_list_objc_symbol): Update.
(get_msymbol_address): Update.
* symtab.h (struct general_symbol_info): Add member functions
natural_name (), linkage_name (), print_name (), demangled_name (),
and search_name ().
(SYMBOL_NATURAL_NAME): Update.
(symbol_natural_name): Move to a member function on general_symbol_info.
(SYMBOL_DEMANGLED_NAME): Update.
(symbol_demangled_name): Move to a member function on
general_symbol_info.
(SYMBOL_SEARCH_NAME): Update.
(symbol_search_name): Move to a member function on general_symbol_info.
(MSYMBOL_NATURAL_NAME): Remove.
(MSYMBOL_LINKAGE_NAME): Remove.
(MSYMBOL_PRINT_NAME): Remove.
(MSYMBOL_DEMANGLED_NAME): Remove.
(MSYMBOL_SEARCH_NAME): Remove.
* x86-tdep.c (x86_in_indirect_branch_thunk): Update.

Change-Id: I65aa529843a9903e174ce799037e41f954a9fcee

Create a correctly-sized demangled names hashtable

If we have a minsym count, we know the demangled names hashtable will
be at least that big.  So use that count to size it, so we don't
have to resize/rehash it as much.

This is a 6% improvement in minsym loading time.

2019-11-22  Christian Biesinger  <cbiesinger@google.com>

* symtab.c (create_demangled_names_hash): Use per_bfd->
minimal_symbol_count for computing the initial size, if greater
than our default size.

Change-Id: I1f074d38e1d90af58705ec852f90c84cc034cd2e

[gdb/contrib] Improve words extraction in words.sh script

Remove more punctuation and quoting in words.sh script.

gdb/ChangeLog:

2019-11-22  Tom de Vries  <tdevries@suse.de>

* contrib/words.sh: Improve words extraction.

Change-Id: I1d9eea165731af4e6c4e1c7e09aed9b07af6395c

[gdb/contrib] Combine sed invocations in words.sh script

Currently running words.sh on all the c source and header files in the repo
takes ~16s in user time:
...
$ time ./gdb/contrib/words.sh \
    $(find -type f -name "*.c" -o -name "*.h") \
    >/dev/null

real    0m7,787s
user    0m16,349s
sys     0m0,367s
...

Rewrite the sed invocations using the -e option from this:
...
   | sed <sedprog1>
   | sed <sedprog2>
...
into this:
...
   | sed \
       -e <sedprog1>
       -e <sedprog2>
...
and reduce user time to ~11s:
...
$ time ./gdb/contrib/words.sh \
    $(find -type f -name "*.c" -o -name "*.h") \
    >/dev/null

real    0m7,243s
user    0m11,220s
sys     0m0,205s
...

gdb/ChangeLog:

2019-11-22  Tom de Vries  <tdevries@suse.de>

* contrib/words.sh: Combine sed invocations.

Change-Id: Ib08453f3712f32ed02d9f503ee960711ebb9421b

Arm: Change CRC from fpu feature to archititectural extension

This patch changes the CRC extension to use the core feature bits instead
of the coproc/fpu feature bits.
CRC is not an fpu feature and it causes issues with the new fpu reset
patch (f439988037a589de3798f44e7268301adaec21a9). CRC can be set using
the '.arch_extension' directive, which sets bits in the coproc bitfield. When
a '.fpu' directive is encountered, the CRC feature bit gets removed and
there is no way to set it back using '.fpu'.
With this patch, CRC will be marked in the feature core bits, which prevents
it from getting removed when setting/changing the fpu options.

gas/ChangeLog:

* config/tc-arm.c (arm_ext_crc): New.
(crc_ext_armv8): Remove.
(insns): Rename crc_ext_armv8 to arm_ext_crc.
(arm_cpus): Replace CRC_EXT_ARMV8 with ARM_EXT2_CRC.
(armv8a_ext_table, armv8r_ext_table,
arm_option_extension_value_table): Redefine the crc
extension in terms of ARM_EXT2_CRC.
* gas/testsuite/gas/arm/crc-ext.s: New.
* gas/testsuite/gas/arm/crc-ext.d: New.

include/ChangeLog:

* opcode/arm.h (ARM_EXT2_CRC): New extension feature
to replace CRC_EXT_ARMV8.
(CRC_EXT_ARMV8): Remove and mark bit as unused.
(ARM_ARCH_V8A_CRC, ARM_ARCH_V8_1A, ARM_ARCH_V8_2A,
ARM_ARCH_V8_3A, ARM_ARCH_V8_4A, ARM_ARCH_V8_5A,
ARM_ARCH_V8_6A): Redefine using ARM_EXT2_CRC instead of
CRC_EXT_ARMV8.

opcodes/ChangeLog:

* opcodes/arm-dis.c (arm_opcodes, thumb32_opcodes):
Change the coproc CRC conditions to use the extension
feature set, second word, base on ARM_EXT2_CRC.

Update the linker documentation regarding the /DISCARD/ linker script section and its interaction with ELF section groups.

* ld.texi (Output Section Discarding): Add note indicating that
/DISCARD/ed sections ignore ELF section grouping.

Replace unnecessary null check with a cast.

* ldlang.h (LANG_FOR_EACH_INPUT_STATEMENT): Use cast instead of
extra check.

Automatic date update in version.in

Rename demangle.c to gdb-demangle.c, and some cleanup

In addition to renaming demangle.c to match the header file naming,
this also makes is_cplus_marker return a bool and removes a duplicate
declaration of "bool demangle" from symtab.h.

gdb/ChangeLog:

2019-11-21  Christian Biesinger  <cbiesinger@google.com>

* Makefile.in: Update.
* demangle.c: Rename to...
* gdb-demangle.c: ..this.
(is_cplus_marker): Change return type to bool.
(_initialize_demangler): Rename to...
(_initialize_gdb_demangle): ...this.
* gdb-demangle.h (is_cplus_marker): Change return type to bool.
* symtab.h (demangle): Remove declaration; instead include
gdb-demangle.h.

Change-Id: I83c3b3f7ee71b2bf6f5b5d0f9eb1d4b5208f2a97

Handle %I64d in format_pieces

We found a bug internally where gdb would crash while disassembling a
certain instruction.  This was tracked down to the handling of %I64d
in format_pieces.

format_pieces will convert %ll to %I64d on mingw -- so format_pieces
should also handle parsing this format.  In this patch, I've made the
parsing unconditional, since I think it is harmless to accept extra
formats.  I've also taken the opportunity to convert the length
modifier test to a "switch".

Tested internally using our failing test case.

gdb/ChangeLog
2019-11-21  Tom Tromey  <tromey@adacore.com>

* gdbsupport/format.c (format_pieces): Parse %I64d.
* unittests/format_pieces-selftests.c (test_windows_formats): New
function.
(run_tests): Call it.

Change-Id: If335c7c2fc8d01e629cd55182394a483334d79c7

Adjust byte order variable display/change if DW_AT_endianity is present.

- Rationale:
It is possible for compilers to indicate the desired byte order
interpretation of scalar variables using the DWARF attribute:
   DW_AT_endianity

A type flagged with this variable would typically use one of:
   DW_END_big
   DW_END_little
which instructs the debugger what the desired byte order interpretation
of the variable should be.

The GCC compiler (as of V6) has a mechanism for setting the desired byte
ordering of the fields within a structure or union.  For, example, on a
little endian target, a structure declared as:
   struct big {
       int v;
       short a[4];
   } __attribute__( ( scalar_storage_order( "big-endian" ) ) );
could be used to ensure all the structure members have a big-endian
interpretation (the compiler would automatically insert byte swap
instructions before and after respective store and load instructions).

- To reproduce
GCC V8 is required to correctly emit DW_AT_endianity DWARF attributes
in all situations when the scalar_storage_order attribute is used.

A fix for (dwarf endianity instrumentation) for GCC V6-V7 can be found
in the URL field of the following PR:
   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82509

- Test-case:
A new test case (testsuite/gdb.base/endianity.*) is included with this
patch.

Manual testing for mixed endianity code has also been done with GCC V8.
See:
   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82509#c4

- Observed vs. expected:

Without this change, using scalar_storage_order that doesn't match the
target, such as

struct otherendian
{
  int v;
} __attribute__( ( scalar_storage_order( "big-endian" ) ) );

would behave like the following on a little endian target:

   Breakpoint 1 at 0x401135: file endianity.c, line 41.
   (gdb) run
   Starting program: /home/pjoot/freeware/t/a.out
   Missing separate debuginfos, use: debuginfo-install glibc-2.17-292.el7.x86_64

   Breakpoint 1, main () at endianity.c:41
   41        struct otherendian o = {3};
   (gdb) n
   43        do_nothing (&o); /* START */
   (gdb) p o
   $1 = {v = 50331648}
   (gdb) p /x
   $2 = {v = 0x3000000}

whereas with this gdb enhancement we can access the variable with the user
specified endianity:

   Breakpoint 1, main () at endianity.c:41
   41        struct otherendian o = {3};
   (gdb) p o
   $1 = {v = 0}
   (gdb) n
   43        do_nothing (&o); /* START */
   (gdb) p o
   $2 = {v = 3}
   (gdb) p o.v = 4
   $3 = 4
   (gdb) p o.v
   $4 = 4
   (gdb) x/4xb &o.v
   0x7fffffffd90c: 0x00    0x00    0x00    0x04

(observe that the 4 byte int variable has a big endian representation in the
 hex dump.)

gdb/ChangeLog
2019-11-21  Peeter Joot  <peeter.joot@lzlabs.com>

Byte reverse display of variables with DW_END_big, DW_END_little
(DW_AT_endianity) dwarf attributes if different than the native
byte order.
* ada-lang.c (ada_value_binop):
Use type_byte_order instead of gdbarch_byte_order.
* ada-valprint.c (printstr):
(ada_val_print_string):
* ada-lang.c (value_pointer):
(ada_value_binop):
Use type_byte_order instead of gdbarch_byte_order.
* c-lang.c (c_get_string):
Use type_byte_order instead of gdbarch_byte_order.
* c-valprint.c (c_val_print_array):
Use type_byte_order instead of gdbarch_byte_order.
* cp-valprint.c (cp_print_class_member):
Use type_byte_order instead of gdbarch_byte_order.
* dwarf2loc.c (rw_pieced_value):
Use type_byte_order instead of gdbarch_byte_order.
* dwarf2read.c (read_base_type): Handle DW_END_big,
DW_END_little
* f-lang.c (f_get_encoding):
Use type_byte_order instead of gdbarch_byte_order.
* findvar.c (default_read_var_value):
Use type_byte_order instead of gdbarch_byte_order.
* gdbtypes.c (check_types_equal):
Require matching TYPE_ENDIANITY_NOT_DEFAULT if set.
(recursive_dump_type): Print TYPE_ENDIANITY_BIG,
and TYPE_ENDIANITY_LITTLE if set.
(type_byte_order): new function.
* gdbtypes.h (TYPE_ENDIANITY_NOT_DEFAULT): New macro.
(struct main_type) <flag_endianity_not_default>:
New field.
(type_byte_order): New function.
* infcmd.c (default_print_one_register_info):
Use type_byte_order instead of gdbarch_byte_order.
* p-lang.c (pascal_printstr):
Use type_byte_order instead of gdbarch_byte_order.
* p-valprint.c (pascal_val_print):
Use type_byte_order instead of gdbarch_byte_order.
* printcmd.c (print_scalar_formatted):
Use type_byte_order instead of gdbarch_byte_order.
* solib-darwin.c (darwin_current_sos):
Use type_byte_order instead of gdbarch_byte_order.
* solib-svr4.c (solib_svr4_r_ldsomap):
Use type_byte_order instead of gdbarch_byte_order.
* stap-probe.c (stap_modify_semaphore):
Use type_byte_order instead of gdbarch_byte_order.
* target-float.c (target_float_same_format_p):
Use type_byte_order instead of gdbarch_byte_order.
* valarith.c (scalar_binop):
(value_bit_index):
Use type_byte_order instead of gdbarch_byte_order.
* valops.c (value_cast):
Use type_byte_order instead of gdbarch_byte_order.
* valprint.c (generic_emit_char):
(generic_printstr):
(val_print_string):
Use type_byte_order instead of gdbarch_byte_order.
* value.c (unpack_long):
(unpack_bits_as_long):
(unpack_value_bitfield):
(modify_field):
(pack_long):
(pack_unsigned_long):
Use type_byte_order instead of gdbarch_byte_order.
* findvar.c (unsigned_pointer_to_address):
(signed_pointer_to_address):
(unsigned_address_to_pointer):
(address_to_signed_pointer):
(default_read_var_value):
(default_value_from_register):
Use type_byte_order instead of gdbarch_byte_order.
* gnu-v3-abi.c (gnuv3_make_method_ptr):
Use type_byte_order instead of gdbarch_byte_order.
* riscv-tdep.c (riscv_print_one_register_info):
Use type_byte_order instead of gdbarch_byte_order.

gdb/testsuite/ChangeLog
2019-11-21  Peeter Joot  <peeter.joot@lzlabs.com>

* gdb.base/endianity.c: New test.
* gdb.base/endianity.exp: New file.

Change-Id: I4bd98c1b4508c2d7c5a5dbb15d7b7b1cb4e667e2