This patch catches invalid initialization of non-POD types with
memset, at compile time.
This is what I used to catch the problems fixed by the previous
patches in the series:
$ make -k 2>&1 | grep "deleted function"
src/gdb/breakpoint.c:951:53: error: use of deleted function ‘void* memset(T*, int, size_t) [with T = bp_location; <template-parameter-1-2> = void; size_t = long unsigned int]’
src/gdb/breakpoint.c:7325:32: error: use of deleted function ‘void* memset(T*, int, size_t) [with T = bp_location; <template-parameter-1-2> = void; size_t = long unsigned int]’
src/gdb/btrace.c:1153:42: error: use of deleted function ‘void* memset(T*, int, size_t) [with T = btrace_insn; <template-parameter-1-2> = void; size_t = long unsigned int]’
...
gdb/ChangeLog:
2017-04-25 Pedro Alves <palves@redhat.com>
* common/common-defs.h: Include "common/poison.h".
* common/function-view.h: (Not, Or, Requires): Move to traits.h
and adjust.
* common/poison.h: New file.
* common/traits.h: Include <type_traits>.
(Not, Or, Requires): New, moved from common/function-view.h.
The delete-memcpy-with-non-trivial-types patch exposed many instances
of this problem:
src/gdb/btrace.h: In function ‘btrace_insn_s* VEC_btrace_insn_s_quick_insert(VEC_btrace_insn_s*, unsigned int, const btrace_insn_s*, const char*, unsigned int)’:
src/gdb/common/vec.h:948:62: error: use of deleted function ‘void* memmove(T*, const U*, size_t) [with T = btrace_insn; U = btrace_insn; <template-parameter-1-3> = void; size_t = long unsigned int]’
memmove (slot_ + 1, slot_, (vec_->num++ - ix_) * sizeof (T)); \
^
src/gdb/common/vec.h:436:1: note: in expansion of macro ‘DEF_VEC_FUNC_O’
DEF_VEC_FUNC_O(T) \
^
src/gdb/btrace.h:84:1: note: in expansion of macro ‘DEF_VEC_O’
DEF_VEC_O (btrace_insn_s);
^
[...]
src/gdb/common/vec.h:1060:31: error: use of deleted function ‘void* memcpy(T*, const U*, size_t) [with T = btrace_insn; U = btrace_insn; <template-parameter-1-3> = void; size_t = long unsigned int]’
sizeof (T) * vec2_->num); \
^
src/gdb/common/vec.h:437:1: note: in expansion of macro ‘DEF_VEC_ALLOC_FUNC_O’
DEF_VEC_ALLOC_FUNC_O(T) \
^
src/gdb/btrace.h:84:1: note: in expansion of macro ‘DEF_VEC_O’
DEF_VEC_O (btrace_insn_s);
^
So, VECs (given it's C roots) rely on memcpy/memcpy of VEC elements to
be well defined, in order to grow/reallocate its internal elements
array. This means that we can only put trivially copyable types in
VECs. E.g., if a type requires using a custom copy/move ctor to
relocate, then we can't put it in a VEC (so we use std::vector
instead). But, as shown above, we're violating that requirement.
btrace_insn is currently not trivially copyable, because it contains
an enum_flags field, and that is itself not trivially copyable. This
patch corrects that, by simply removing the user-provided copy
constructor and assignment operator. The compiler-generated versions
work just fine.
Note that std::vector relies on std::is_trivially_copyable too to know
whether it can reallocate its elements with memcpy/memmove instead of
having to call copy/move ctors and dtors, so if we have types in
std::vectors that weren't trivially copyable because of enum_flags,
this will make such vectors more efficient.
gdb/ChangeLog:
2017-04-25 Pedro Alves <palves@redhat.com>
* common/enum-flags.h (enum_flags): Don't implement copy ctor and
assignment operator.
This patch fixes an internal error exposed by a test that does
something like:
define kill-and-remove
kill inferiors 2
remove-inferiors 2
end
# Start one inferior.
start
# Start another inferior.
add-inferior 2
inferior 2
start
# Kill and remove inferior 1 while inferior 2 is selected.
thread apply 1.1 kill-and-remove
The internal error looks like this:
Thread 1.1 (Thread 0x7ffff7fc2700 (LWP 20677)):
[Switching to inferior 1 [process 20677] (gdb/testsuite/outputs/gdb.threads/threadapply/threadapply)]
[Switching to thread 1.1 (Thread 0x7ffff7fc2700 (LWP 20677))]
#0 main () at src/gdb/testsuite/gdb.threads/threadapply.c:38
38 for (i = 0; i < NUM; i++)
src/gdb/inferior.c:66: internal-error: void set_current_inferior(inferior*): Assertion `inf != NULL' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n) FAIL: gdb.threads/threadapply.exp: kill_and_remove_inferior: try kill-and-remove: thread apply 1.1 kill-and-remove (GDB internal error)
There are several problems around this area of the code. One is that
in do_restore_current_thread_cleanup, we do a look up of inferior by
ptid, which can find the wrong inferior if the previously selected
inferior exited and some other inferior was started with a reused pid
(rare, but still...).
The other problem is that the "remove-inferiors" command rejects
attempts to remove the current inferior, but when we get to
"remove-inferiors" in a "thread apply THR remove-inferiors 2" command,
the current inferior is the inferior of thread THR, not the previously
selected inferior, so if the previously selected inferior was inferior
2, that command still manages to wipe it, and then gdb restores the
old selected inferior, which is now a dangling pointer...
So the fix here is:
- Make make_cleanup_restore_current_thread store a pointer to the
previously selected inferior directly, and use it directly instead
of doing ptid look ups.
- Add a refcount to inferiors, very similar to thread_info's refcount,
that is incremented/decremented by
make_cleanup_restore_current_thread, and checked before deleting an
inferior. To avoid duplication, a new refcounted_object type is
added, that both thread_info and inferior inherit from.
gdb/ChangeLog:
2017-04-19 Pedro Alves <palves@redhat.com>
* common/refcounted-object.h: New file.
* gdbthread.h: Include "common/refcounted-object.h".
(thread_info): Inherit from refcounted_object and add comments.
(thread_info::incref, thread_info::decref)
(thread_info::m_refcount): Delete.
(thread_info::deletable): Use the refcounted_object::refcount()
method.
* inferior.c (current_inferior_): Add comment.
(set_current_inferior): Increment/decrement refcounts.
(prune_inferiors, remove_inferior_command): Skip inferiors marked
not-deletable instead of comparing with the current inferior.
(initialize_inferiors): Increment the initial inferior's refcount.
* inferior.h (struct inferior): Forward declare.
Include "common/refcounted-object.h".
(current_inferior, set_current_inferior): Move declaration to
before struct inferior's definition, and fix comment.
(inferior): Inherit from refcounted_object. Add comments.
* thread.c (switch_to_thread_no_regs): Reference the thread's
inferior pointer directly instead of doing a ptid lookup.
(switch_to_no_thread): New function.
(switch_to_thread(thread_info *)): New function, factored out
from ...
(switch_to_thread(ptid_t)): ... this.
(restore_current_thread): Delete.
(current_thread_cleanup): Remove 'inf_id' and 'was_removable'
fields, and add 'inf' field.
(do_restore_current_thread_cleanup): Check whether old->inf is
alive instead of looking up an inferior by ptid. Use
switch_to_thread and switch_to_no_thread.
(restore_current_thread_cleanup_dtor): Use old->inf directly
instead of lookup up an inferior by id. Decref the inferior.
Don't restore 'removable'.
(make_cleanup_restore_current_thread): Same the inferior pointer
in old, instead of the inferior number. Incref the inferior.
Don't save/clear 'removable'.
gdb/testsuite/ChangeLog:
2017-04-19 Pedro Alves <palves@redhat.com>
* gdb.threads/threadapply.exp (kill_and_remove_inferior): New
procedure.
(top level): Call it.
* lib/gdb.exp (gdb_define_cmd): New procedure.
I left making inferior::detaching a bool to a separate patch, because
doing that makes a make_cleanup_restore_integer call in
infrun.c:prepare_for_detach no longer compile (passing a 'bool *' when
an 'int *' is expected). Since we want to get rid of cleanups anyway,
I looked at converting that to a scoped_restore. However,
prepare_for_detach wants to discard the cleanup on success, and
scoped_restore doesn't have an equivalent for that. So I added one --
I called it "release()" because it seems like a natural fit in the way
standard components call similarly-spirited methods, and, it's also
what the proposal for a generic scope guard calls it too, AFAICS:
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n4189.pdf
I've added some scoped_guard unit tests, while at it.
gdb/ChangeLog:
2017-04-19 Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
unittests/scoped_restore-selftests.c.
(SUBDIR_UNITTESTS_OBS): Add scoped_restore-selftests.o.
* common/scoped_restore.h (scoped_restore_base): Make "class".
(scoped_restore_base::release): New public method.
(scoped_restore_base::scoped_restore_base): New protected ctor.
(scoped_restore_base::m_saved_var): New protected field.
(scoped_restore_tmpl::scoped_restore_tmpl(T*)): Initialize the
scoped_restore_base base class instead of m_saved_var directly.
(scoped_restore_tmpl::scoped_restore_tmpl(T*, T2)): Likewise.
(scoped_restore_tmpl::scoped_restore_tmpl(const
scoped_restore_tmpl<T>&)): Likewise.
(scoped_restore_tmpl::~scoped_restore_tmpl): Use the saved_var
method.
(scoped_restore_tmpl::saved_var): New method.
(scoped_restore_tmpl::m_saved_var): Delete.
* inferior.h (inferior::detaching): Now a bool.
* infrun.c (prepare_for_detach): Use a scoped_restore instead of a
cleanup.
* unittests/scoped_restore-selftests.c: New file.
Currently we can't use gdb::optional<T> as function return type,
because gdb::optional's copy ctor is deleted. For example, with:
gdb::optional<int> function ()
{
gdb::optional<int> opt;
....
return opt;
we get:
src/gdb/foo.c: In function ‘gdb::optional<int> foo()’:
src/gdb/foo.c:75:10: error: use of deleted function ‘gdb::optional<T>::optional(const gdb::optional<T>&) [with T = int]’
return opt;
^
In file included from src/gdb/foo.c:68:0:
src/gdb/common/gdb_optional.h:53:3: note: declared here
optional (const optional &other) = delete;
^
I started by fixing that, and then ran into another missing feature,
also fixed by this patch.
The next feature I'm missing most from gdb::optional<T> compared to
std::optional<T> is construction/move/assignment from a T, instead of
having to default construct an gdb::optional and then use
optional::emplace(....).
For example:
gdb::optional<std::string> function ()
{
gdb::optional<std::string> opt;
std::string str;
...
opt.emplace (std::move (str));
return opt;
vs
gdb::optional<std::string> function ()
{
std::string str;
...
return str;
The copy/move ctor/assign methods weren't initialy implemented because
std::optional supports construction from a type U if U is convertible
to T too, and has rules to decide whether the ctors are
explicit/implicit based on that, and rules for whether the ctor should
be trivial or not, etc., which leads to a much more complicated
implementation.
If we stick to supporting copy/move construction/assignment of/to an
optional<T> from exactly only optional<T> and T, then all that
conversion-related complication disappears, and we still gain
convenience in most use cases.
The patch also makes emplace return a reference to the constructor
object, per C++17 std::optional, and adds a reset method, againt
because std::optional has one and it's trivial to support it. These
two changes are a requirement of the gdb::optional unit testing patch
that will follow.
gdb/ChangeLog:
2017-04-18 Pedro Alves <palves@redhat.com>
* common/gdb_optional.h: Include common/traits.h.
(in_place_t): New type.
(in_place): New constexpr variable.
(optional::optional): Remove member initialization of
m_instantiated.
(optional::optional(in_place_t...)): New constructor.
(optional::~optional): Use reset.
(optional::optional(const optional&)): New.
(optional::optional(const optional&&)): New.
(optional::optional(T &)): New.
(optional::optional(T &&)): New.
(operator::operator=(const optional &)): New.
(operator::operator=(optional &&)): New.
(operator::operator= (const T &))
(operator::operator= (T &&))
(operator::emplace (Args &&... args)): Return a T&. Use reset.
(operator::reset): New.
(operator::m_instantiated):: Add in-class initializer.
* common/traits.h: Include <type_traits>.
(struct And): New types.
As requested, I'm sending this as a separate patch because it is ready
to be included as-is.
The idea here is that both gdb/terminal.h and gdb/gdbserver/terminal.h
share the same code, which is responsible for setting a bunch of
defines on based on the presence of termios.h and a few other headers.
This simple patch just moves this common code to common/gdb_termios.h
and makes the necessary adjustments on both GDB and gdbserver so that
they can use this new header. It also implements the some header
checks on common/common.m4.
As a bonus, gdb/gdbserver/terminal.h can be removed because it's now
empty.
Built on x86_64, no regressions found.
gdb/ChangeLog:
2017-04-12 Sergio Durigan Junior <sergiodj@redhat.com>
* Makefile.in (HFILES_NO_SRCDIR): Add "common/gdb_termios.h".
* common/common.m4: Check headers 'termios.h', 'termio.h' and
'sgtty.h'.
* common/gdb_termios.h: New file, with parts of "terminal.h".
* inflow.c: Include "gdb_termios.h".
* ser-unix.c: Include "gdb_termios.h".
* terminal.h: Move terminal-related defines to
"common/gdb_termios.h".
gdb/gdbserver/ChangeLog:
2017-04-12 Sergio Durigan Junior <sergiodj@redhat.com>
* remote-utils.c: Include "gdb_termios.h" instead of
"terminal.h".
* terminal.h: Delete file.
As a preparation for the next patch, which will move fork_inferior
from GDB to common/ (and therefore share it with gdbserver), it is
interesting to convert a few functions to C++.
This patch touches functions related to parsing command-line arguments
to the inferior (see gdb/fork-child.c:breakup_args), the way the
arguments are stored on fork_inferior (using std::vector instead of
char **), and the code responsible for dealing with argv also on
gdbserver.
I've taken this opportunity and decided to constify a few arguments to
fork_inferior/create_inferior as well, in order to make the code
cleaner. And now, on gdbserver, we're using xstrdup everywhere and
aren't checking for memory allocation failures anymore, as requested
by Pedro:
<https://sourceware.org/ml/gdb-patches/2017-03/msg00191.html>
Message-Id: <025ebdb9-90d9-d54a-c055-57ed2406b812@redhat.com>
Pedro Alves wrote:
> On the "== NULL" check: IIUC, the old NULL check was there to
> handle strdup returning NULL due to out-of-memory.
> See NULL checks and comments further above in this function.
> Now that you're using a std::vector, that doesn't work or make
> sense any longer, since if push_back fails to allocate space for
> its internal buffer (with operator new), our operator new replacement
> (common/new-op.c) calls malloc_failure, which aborts gdbserver.
>
> Not sure it makes sense to handle out-of-memory specially in
> the gdb/rsp-facing functions nowadays (maybe git blame/log/patch
> submission for that code shows some guidelines). Maybe (or, probably)
> it's OK to stop caring about it, but then we should consistently remove
> left over code, by using xstrdup instead and remove the NULL checks.
IMO this refactoring was very good to increase the readability of the
code as well, because some parts of the argument handling were
unnecessarily confusing before.
gdb/ChangeLog:
2017-04-12 Sergio Durigan Junior <sergiodj@redhat.com>
* common/common-utils.c (free_vector_argv): New function.
* common/common-utils.h: Include <vector>.
(free_vector_argv): New prototype.
* darwin-nat.c (darwin_create_inferior): Rewrite function
prototype in order to constify "exec_file" and accept a
"std::string" for "allargs".
* fork-child.c: Include <vector>.
(breakup_args): Rewrite function, using C++.
(fork_inferior): Rewrite function header, constify "exec_file_arg"
and accept "std::string" for "allargs". Update the code to
calculate "argv" based on "allargs". Update calls to "exec_fun"
and "execvp".
* gnu-nat.c (gnu_create_inferior): Rewrite function prototype in
order to constify "exec_file" and accept a "std::string" for
"allargs".
* go32-nat.c (go32_create_inferior): Likewise.
* inf-ptrace.c (inf_ptrace_create_inferior): Likewise.
* infcmd.c (run_command_1): Constify "exec_file". Use
"std::string" for inferior arguments.
* inferior.h (fork_inferior): Update prototype.
* linux-nat.c (linux_nat_create_inferior): Rewrite function
prototype in order to constify "exec_file" and accept a
"std::string" for "allargs".
* nto-procfs.c (procfs_create_inferior): Likewise.
* procfs.c (procfs_create_inferior): Likewise.
* remote-sim.c (gdbsim_create_inferior): Likewise.
* remote.c (extended_remote_run): Update code to accept
"std::string" as argument.
(extended_remote_create_inferior): Rewrite function prototype in
order to constify "exec_file" and accept a "std::string" for
"allargs".
* rs6000-nat.c (super_create_inferior): Likewise.
(rs6000_create_inferior): Likewise.
* target.h (struct target_ops) <to_create_inferior>: Likewise.
* windows-nat.c (windows_create_inferior): Likewise.
gdb/gdbserver/ChangeLog:
2017-04-12 Sergio Durigan Junior <sergiodj@redhat.com>
* server.c: Include <vector>.
<program_argv, wrapper_argv>: Convert to std::vector.
(start_inferior): Rewrite function to use C++.
(handle_v_run): Likewise. Update code that calculates the argv
based on the vRun packet; use C++.
(captured_main): Likewise.
I grew a bit tired of using ptid_get_{lwp,pid,tid} and friends, so I decided to
make it a bit easier to use by making it a proper class. The fields are now
private, so it's not possible to change a ptid_t field by mistake.
The new methods of ptid_t map to existing functions/practice like this:
ptid_t (pid, lwp, tid) -> ptid_build (pid, lwp, tid)
ptid_t (pid) -> pid_to_ptid (pid)
ptid.is_pid () -> ptid_is_pid (ptid)
ptid == other -> ptid_equal (ptid, other)
ptid != other -> !ptid_equal (ptid, other)
ptid.pid () -> ptid_get_pid (ptid)
ptid.lwp_p () -> ptid_lwp_p (ptid)
ptid.lwp () -> ptid_get_lwp (ptid)
ptid.tid_p () -> ptid_tid_p (ptid)
ptid.tid () -> ptid_get_tid (ptid)
ptid.matches (filter) -> ptid_match (ptid, filter)
I've replaced the implementation of the existing functions with calls to
the new methods. People are encouraged to gradually switch to using the
ptid_t methods instead of the functions (or we can change them all in
one pass eventually).
Also, I'm not sure if it's worth it (because of ptid_t's relatively
small size), but I have made the functions and methods take ptid_t
arguments by const reference instead of by value.
gdb/ChangeLog:
* common/ptid.h (struct ptid): Change to...
(class ptid_t): ... this.
<ptid_t>: New constructors.
<pid, lwp_p, lwp, tid_p, tid, is_pid, operator==, operator!=,
matches>: New methods.
<make_null, make_minus_one>: New static methods.
<pid>: Rename to...
<m_pid>: ...this.
<lwp>: Rename to...
<m_lwp>: ...this.
<tid>: Rename to...
<m_tid>: ...this.
(ptid_build, ptid_get_pid, ptid_get_lwp, ptid_get_tid, ptid_equal,
ptid_is_pid, ptid_lwp_p, ptid_tid_p, ptid_match): Take ptid arguments
as references, move comment to class ptid_t.
* common/ptid.c (null_ptid, minus_one_ptid): Initialize with
ptid_t static methods.
(ptid_build, pid_to_ptid, ptid_get_pid, ptid_get_tid,
ptid_equal, ptid_is_pid, ptid_lwp_p, ptid_tid_p, ptid_match):
Take ptid arguments as references, implement using ptid_t methods.
* unittests/ptid-selftests.c: New file.
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
unittests/ptid-selftests.c.
(SUBDIR_UNITTESTS_OBS): Add unittests/ptid-selftests.o.
gdb/gdbserver/ChangeLog:
* server.c (handle_v_cont): Initialize thread_resume::thread
with null_ptid.
The calls to ptid_equal in ptid_lwp_p and ptid_tid_p that compare the
argument to minus_one_ptid and null_ptid are not necessary. The calls
in question are:
if (ptid_equal (minus_one_ptid, ptid)
|| ptid_equal (null_ptid, ptid))
return 0;
minus_one_ptid is { .pid = -1, .lwp = 0, .tid = 0 }
null_ptid is { .pid = 0, .lwp = 0, .tid = 0 }
If the ptid argument is either of them, the statements
return (ptid_get_lwp (ptid) != 0);
and
return (ptid_get_tid (ptid) != 0);
will yield the same result (0/false).
gdb/ChangeLog:
* common/ptid.c (ptid_lwp_p, ptid_tid_p): Remove comparison with
minus_one_ptid and null_ptid.
Commit ecfb656c37 ("dwarf2read.c: Make dir_index and
file_name_index strong typedefs") added a use of gdb::optional that
triggers bogus -Wmaybe-uninitialized warnings:
GCC trunk is complaining like this:
../../binutils-gdb/gdb/dwarf2read.c: In function void read_formatted_entries(bfd*, const gdb_byte**, line_header*, const comp_unit_head*, void (*)(line_header*, const char*, dir_index, unsigned int, unsigned int)):
../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::length may be used uninitialized in this function [-Werror=maybe-uninitialized]
callback (lh, fe.name, fe.d_index, fe.mod_time, fe.length);
^
../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: *((void*)& fe +8) may be used uninitialized in this function [-Werror=maybe-uninitialized]
../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::mod_time may be used uninitialized in this function [-Werror=maybe-uninitialized]
../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: fe.file_entry::name may be used uninitialized in this function [-Werror=maybe-uninitialized]
While some older GCCs are complaining like this:
../../binutils-gdb/gdb/dwarf2read.c: In function void read_formatted_entries(bfd*, const gdb_byte**, line_header*, const comp_unit_head*, void (*)(line_header*, const char*, dir_index, unsigned int, unsigned int)):
../../binutils-gdb/gdb/dwarf2read.c:17779:65: error: uint may be used uninitialized in this function [-Werror=maybe-uninitialized]
callback (lh, fe.name, fe.d_index, fe.mod_time, fe.length);
Looking around the web, I see that boost's optional implementation
triggers this kind of issue often too. See:
http://www.boost.org/doc/libs/1_63_0/libs/optional/doc/html/boost_optional/tutorial/gotchas/false_positive_with__wmaybe_uninitialized.html
I noticed that replacing the gdb::optional uses with real C++17
std::optional uses against GCC 7/trunk makes the warnings go away.
Looking at the implementation, AFAICS, libstdc++ always initializes
its "empty" union payload member (_M_empty, which is defined as an
empty class, like ours). I.e., all payload types have this ctor:
struct _Optional_payload.....
{
constexpr _Optional_payload()
: _M_empty() {}
The constexpr makes a diference too. Without it, GCC7 still warns.
So I'm applying the same treatment to our gdb::optional.
gdb/ChangeLog:
2017-04-05 Pedro Alves <palves@redhat.com>
* common/gdb_optional.h (optional::optional): Make constexpr and
initialize m_dummy.
A while ago, back when GDB was a C program, the sect_offset and
cu_offset types were made structs in order to prevent incorrect mixing
of those offsets. Now that we require C++11, we can make them
integers again, while keeping the safety, by exploiting "enum class".
We can add a bit more safety, even, by defining operators that the
types _should_ support, helping making the suspicious uses stand out
more.
Getting at the underlying type is done with the new to_underlying
function added by the previous patch, which also helps better spot
where do we need to step out of the safety net. Mostly, that's around
parsing the DWARF, and when we print the offset for complaint/debug
purposes. But there are other occasional uses.
Since we have to define the sect_offset/cu_offset types in a header
anyway, I went ahead and generalized/library-fied the idea of "offset"
types, making it trivial to add more such types if we find a use. See
common/offset-type.h and the DEFINE_OFFSET_TYPE macro.
I needed a couple generaly-useful preprocessor bits (e.g., yet another
CONCAT implementation), so I started a new common/preprocessor.h file.
I included units tests covering the "offset" types API. These are
mostly compile-time tests, using SFINAE to check that expressions that
shouldn't compile (e.g., comparing unrelated offset types) really are
invalid and would fail to compile. This same idea appeared in my
pending enum-flags revamp from a few months ago (though this version
is a bit further modernized compared to what I had posted), and I plan
on reusing the "check valid expression" bits added here in that
series, so I went ahead and defined the CHECK_VALID_EXPR macro in its
own header -- common/valid-expr.h. I think that's nicer regardless.
I was borderline between calling the new types "offset" types, or
"index" types, BTW. I stuck with "offset" simply because that's what
we're already calling them, mostly.
gdb/ChangeLog:
2017-04-04 Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
unittests/offset-type-selftests.c.
(SUBDIR_UNITTESTS_OBS): Add offset-type-selftests.o.
* common/offset-type.h: New file.
* common/preprocessor.h: New file.
* common/traits.h: New file.
* common/valid-expr.h: New file.
* dwarf2expr.c: Include "common/underlying.h". Adjust to use
sect_offset and cu_offset strong typedefs throughout.
* dwarf2expr.h: Adjust to use sect_offset and cu_offset strong
typedefs throughout.
* dwarf2loc.c: Include "common/underlying.h". Adjust to use
sect_offset and cu_offset strong typedefs throughout.
* dwarf2read.c: Adjust to use sect_offset and cu_offset strong
typedefs throughout.
* gdbtypes.h: Include "common/offset-type.h".
(cu_offset): Now an offset type (strong typedef) instead of a
struct.
(sect_offset): Likewise.
(union call_site_parameter_u): Rename "param_offset" field to
"param_cu_off".
* unittests/offset-type-selftests.c: New file.
This should help catch mistakes related to mixing the 1-based DWARF
indexes with 0-based std::vector indexes, since the new types do not
implicitly convert to anything.
The change in read_formatted_entries relates to the fact that doing
the seemingly simpler:
- uintp = &fe.dir_index;
+ uintp = (unsigned int *) &fe.dir_index;
would be undefined C/C++. So to address that, I made the function
extract the form before assigning to the file_entry. It felt natural
to use gdb::optional for "do I have this value", and this is what
motivated the previous patch that added the missing observer methods
to gdb::optional.
gdb/ChangeLog:
2017-04-04 Pedro Alves <palves@redhat.com>
* common/underlying.h: New file.
* dwarf2read.c: Include "common/gdb_optional.h" and
"common/underlying.h".
(dir_index, file_name_index): New types.
(file_entry): Use them.
(file_entry::include): Use to_underlying.
(line_header::add_file_name): Use dir_index.
(read_formatted_entries): Use gdb::optional. Read form before
writting to file_entry.
(dwarf_decode_line_header): Use dir_index.
(lnp_state_machine::current_file): Use to_underlying.
(lnp_state_machine::file): Change type to file_name_index.
(dwarf_record_line): Use to_underlying.
(init_lnp_state_machine): Use file_name_index.
(dwarf_decode_lines_1): Use dir_index and file_name_index.
Currently, gdb::optional is really minimal and can only be used for
lazy initialization. There's no way to get at the value contained
inside the optinal. This commit corrects that, by adding observer
methods, mostly copied from libstdc++'s implementation of C++17
std::optional.
This will be used in the following patch.
gdb/ChangeLog:
2017-04-04 Pedro Alves <palves@redhat.com>
* common/gdb_optional.h (gdb::optiona): Add operator->, operator*,
operator bool, has_value and get methods.
Should fix the build failure with Clang mentioned at
<https://sourceware.org/bugzilla/show_bug.cgi?id=21206#c2>:
In file included from ../../binutils-gdb/gdb/dwarf2read.c:72:
../../binutils-gdb/gdb/common/gdb_unlinker.h:35:35: error: '__nonnull__' attribute is invalid for the implicit this argument
unlinker (const char *filename) ATTRIBUTE_NONNULL (1)
^ ~
../../binutils-gdb/gdb/../include/ansidecl.h:169:48: note: expanded from macro 'ATTRIBUTE_NONNULL'
# define ATTRIBUTE_NONNULL(m) __attribute__ ((__nonnull__ (m)))
gdb/ChangeLog:
2017-03-08 Pedro Alves <palves@redhat.com>
PR 21206
* common/gdb_unlinker.h (unlinker::unlinker): Attribute nonnull
goes to argument 2, not 1.
We will need access to the environment functions when we share
fork_inferior between GDB and gdbserver, therefore we simply make the
API on gdb/environ.[ch] available on common/. No extra adjustments
are needed to make it compile on gdbserver.
gdb/ChangeLog:
2017-03-07 Sergio Durigan Junior <sergiodj@redhat.com>
* Makefile.in (SFILES): Replace "environ.c" with
"common/environ.c".
(HFILES_NO_SRCDIR): Likewise, for "environ.h".
* environ.c: Include "common-defs.h" instead of "defs.h. Moved
to...
* common/environ.c: ... here.
* environ.h: Moved to...
* common/environ.h: ... here.
gdb/gdbserver/ChangeLog:
2017-03-07 Sergio Durigan Junior <sergiodj@redhat.com>
* Makefile.in (SFILES): Add "common/environ.c".
(OBJS): Add "common/environ.h".
This commit adds a new function_view type. This type holds a
non-owning reference to a callable. It is meant to be used as
callback type of functions, instead of using the C-style pair of
function pointer and 'void *data' arguments. function_view allows
passing references to stateful function objects / lambdas with
captures as callbacks efficiently, while function pointer + 'void *'
does not.
See the intro in the new function-view.h header for more.
Unit tests included, put into a new gdb/unittests/ subdir.
gdb/ChangeLog:
2017-02-23 Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_UNITTESTS_SRCS, SUBDIR_UNITTESTS_OBS): New.
(%.o) <unittests/%.c>: New pattern.
* configure.ac ($development): Add $(SUBDIR_UNITTESTS_OBS) to
CONFIG_OBS, and $(SUBDIR_UNITTESTS_SRCS) to CONFIG_SRCS.
* common/function-view.h: New file.
* unittests/function-view-selftests.c: New file.
* configure: Regenerate.
... so that we don't need to do it manually, and potentially forget.
For example, this allows to do:
my_flags flags;
...
flags |= some_flag;
gdb/ChangeLog:
* common/enum-flags.h (enum_flags::enum_flags): Initialize
m_enum_value to 0 in default constructor.
This patch introduces a bit of infrastructure -- namely, a minimal
std::optional analogue called gdb::optional, and an RAII template
class that works like make_cleanup_ui_out_tuple_begin_end or
make_cleanup_ui_out_list_begin_end -- and then uses these in the
Python code. This removes a number of cleanups and generally
simplifies this code.
std::optional is only available in C++17. Normally I would have had
this code check __cplusplus, but my gcc apparently isn't new enough to
find <optional>, even with -std=c++1z; so, because I could not test
it, the patch does not do this.
gdb/ChangeLog
2017-02-10 Tom Tromey <tom@tromey.com>
* ui-out.h (ui_out_emit_type): New class.
(ui_out_emit_tuple, ui_out_emit_list): New typedefs.
* python/py-framefilter.c (py_print_single_arg): Use gdb::optional
and ui_out_emit_tuple.
(enumerate_locals): Likewise.
(py_mi_print_variables, py_print_locals, py_print_args): Use
ui_out_emit_list.
(py_print_frame): Use gdb::optional, ui_out_emit_tuple,
ui_out_emit_list.
* common/gdb_optional.h: New file.
A following patch will want to do
string_file str_file;
scoped_restore save_stdout
= make_scoped_restore (&gdb_stdout, &str_file);
where gdb_stdout is a ui_file *, and string_file is a type that
inherits from ui_file, but that doesn't compile today:
src/gdb/top.c: In function ‘std::__cxx11::string execute_command_to_string(char*, int)’:
src/gdb/top.c:710:50: error: no matching function for call to ‘make_scoped_restore(ui_file**, string_file*)’
= make_scoped_restore (&gdb_stdout, &str_file);
^
[...]
In file included from src/gdb/utils.h:25:0,
from src/gdb/defs.h:732,
from src/gdb/top.c:20:
src/gdb/common/scoped_restore.h:94:24: note: candidate: template<class T> scoped_restore_tmpl<T> make_scoped_restore(T*, T)
scoped_restore_tmpl<T> make_scoped_restore (T *var, T value)
^
src/gdb/common/scoped_restore.h:94:24: note: template argument deduction/substitution failed:
src/gdb/top.c:710:50: note: deduced conflicting types for parameter ‘T’ (‘ui_file*’ and ‘string_file*’)
= make_scoped_restore (&gdb_stdout, &str_file);
^
This commit makes code such as the above possible.
gdb/ChangeLog:
2017-01-31 Pedro Alves <palves@redhat.com>
* common/scoped_restore.h
(scoped_restore_tmpl::scoped_restore_tmpl): Template on T2, and
change the value's parameter type to T2.
(make_scoped_restore): Likewise.
This introduces a new class, gdb::unlinker, that unlinks a file in the
destructor. The user of this class has the option to preserve the
file instead, by calling the "keep" method.
This patch then changes the spots in gdb that use unlink in a cleanup
to use this class instead. In one spot I went ahead and removed all
the cleanups from the function.
This fixes one latent bug -- do_bfd_delete_cleanup could refer to
freed memory, by decref'ing the BFD before using its filename.
2017-01-10 Tom Tromey <tom@tromey.com>
* record-full.c (record_full_save_cleanups): Remove.
(record_full_save): Use gdb::unlinker.
* gcore.c (do_bfd_delete_cleanup): Remove.
(gcore_command): Use gdb::unlinker, unique_xmalloc_ptr. Remove
cleanups.
* dwarf2read.c (unlink_if_set): Remove.
(write_psymtabs_to_index): Use gdb::unlinker.
* common/gdb_unlinker.h: New file.
This introduces a new specialization of gdb::ref_ptr that can be used
to manage BFD reference counts. Then it changes most places in gdb to
use this new class, rather than explicit reference-counting or
cleanups. This patch removes make_cleanup_bfd_unref.
If you look you will see a couple of spots using "release" where a use
of gdb_bfd_ref_ptr would be cleaner. These will be fixed in the next
patch.
I think this patch fixes some latent bugs. For example, it seems to
me that previously objfpy_add_separate_debug_file leaked a BFD.
I'm not 100% certain that the macho_symfile_read_all_oso change is
correct. The existing code here is hard for me to follow. One goal
of this sort of automated reference counting, though, is to make it
more difficult to make logic errors; so hopefully the code is clear
now.
2017-01-10 Tom Tromey <tom@tromey.com>
* windows-tdep.c (windows_xfer_shared_library): Update.
* windows-nat.c (windows_make_so): Update.
* utils.h (make_cleanup_bfd_unref): Remove.
* utils.c (do_bfd_close_cleanup, make_cleanup_bfd_unref): Remove.
* symfile.h (symfile_bfd_open)
(find_separate_debug_file_in_section): Return gdb_bfd_ref_ptr.
* symfile.c (read_symbols, symbol_file_add)
(separate_debug_file_exists): Update.
(symfile_bfd_open): Return gdb_bfd_ref_ptr.
(generic_load, reread_symbols): Update.
* symfile-mem.c (symbol_file_add_from_memory): Update.
* spu-linux-nat.c (spu_bfd_open): Return gdb_bfd_ref_ptr.
(spu_symbol_file_add_from_memory): Update.
* solist.h (struct target_so_ops) <bfd_open>: Return
gdb_bfd_ref_ptr.
(solib_bfd_fopen, solib_bfd_open): Return gdb_bfd_ref_ptr.
* solib.c (solib_bfd_fopen, solib_bfd_open): Return
gdb_bfd_ref_ptr.
(solib_map_sections, reload_shared_libraries_1): Update.
* solib-svr4.c (enable_break): Update.
* solib-spu.c (spu_bfd_fopen): Return gdb_bfd_ref_ptr.
* solib-frv.c (enable_break2): Update.
* solib-dsbt.c (enable_break): Update.
* solib-darwin.c (gdb_bfd_mach_o_fat_extract): Return
gdb_bfd_ref_ptr.
(darwin_solib_get_all_image_info_addr_at_init): Update.
(darwin_bfd_open): Return gdb_bfd_ref_ptr.
* solib-aix.c (solib_aix_bfd_open): Return gdb_bfd_ref_ptr.
* record-full.c (record_full_save): Update.
* python/py-objfile.c (objfpy_add_separate_debug_file): Update.
* procfs.c (insert_dbx_link_bpt_in_file): Update.
* minidebug.c (find_separate_debug_file_in_section): Return
gdb_bfd_ref_ptr.
* machoread.c (macho_add_oso_symfile): Change abfd to
gdb_bfd_ref_ptr.
(macho_symfile_read_all_oso): Update.
(macho_check_dsym): Return gdb_bfd_ref_ptr.
(macho_symfile_read): Update.
* jit.c (bfd_open_from_target_memory): Return gdb_bfd_ref_ptr.
(jit_bfd_try_read_symtab): Update.
* gdb_bfd.h (gdb_bfd_open, gdb_bfd_fopen, gdb_bfd_openr)
(gdb_bfd_openw, gdb_bfd_openr_iovec)
(gdb_bfd_openr_next_archived_file, gdb_bfd_fdopenr): Return
gdb_bfd_ref_ptr.
(gdb_bfd_ref_policy): New struct.
(gdb_bfd_ref_ptr): New typedef.
* gdb_bfd.c (gdb_bfd_open, gdb_bfd_fopen, gdb_bfd_openr)
(gdb_bfd_openw, gdb_bfd_openr_iovec)
(gdb_bfd_openr_next_archived_file, gdb_bfd_fdopenr): Return
gdb_bfd_ref_ptr.
* gcore.h (create_gcore_bfd): Return gdb_bfd_ref_ptr.
* gcore.c (create_gcore_bfd): Return gdb_bfd_ref_ptr.
(gcore_command): Update.
* exec.c (exec_file_attach): Update.
* elfread.c (elf_symfile_read): Update.
* dwarf2read.c (dwarf2_get_dwz_file): Update.
(try_open_dwop_file, open_dwo_file): Return gdb_bfd_ref_ptr.
(open_and_init_dwo_file): Update.
(open_dwp_file): Return gdb_bfd_ref_ptr.
(open_and_init_dwp_file): Update.
* corelow.c (core_open): Update.
* compile/compile-object-load.c (compile_object_load): Update.
* common/gdb_ref_ptr.h (ref_ptr::operator->): New operator.
* coffread.c (coff_symfile_read): Update.
* cli/cli-dump.c (bfd_openr_or_error, bfd_openw_or_error): Return
gdb_bfd_ref_ptr. Rename.
(dump_bfd_file, restore_command): Update.
* build-id.h (build_id_to_debug_bfd): Return gdb_bfd_ref_ptr.
* build-id.c (build_id_to_debug_bfd): Return gdb_bfd_ref_ptr.
(find_separate_debug_file_by_buildid): Update.
This adds a new gdb_ref_ptr.h, that implements a reference-counting
smart pointer class, where the user of the class supplies a
reference-counting policy object.
This class will be used in the next patch, which changes most explicit
BFD reference counts to use this new type. Meanwhile, this patch
changes gdbpy_ref to be a specialization of this new class.
This change required adding new nullptr_t overloads some operators in
gdb_ref_ptr.h. I suspect this was needed because some Python header
redefines NULL, but I'm not certain.
2017-01-10 Tom Tromey <tom@tromey.com>
* common/gdb_ref_ptr.h: New file.
* python/py-ref.h (struct gdbpy_ref_policy): New.
(gdbpy_ref): Now a typedef.
This applies the second part of GDB's End of Year Procedure, which
updates the copyright year range in all of GDB's files.
gdb/ChangeLog:
Update copyright year range in all GDB files.
Both libc++ and libstdc++ declare non-throwing new operators as
noexcept and overloads must also be noexcept. This fixes a
-Wmissing-exception-spec warning with clang.
gdb/ChangeLog:
* common/new-op.c (operator new): Mark 'noexcept'.
(operator new[]): Likewise.
This patch fixes a few problems with GDB's time handling.
#1 - It avoids problems with gnulib's C++ namespace support
On MinGW, the struct timeval that should be passed to gnulib's
gettimeofday replacement is incompatible with libiberty's
timeval_sub/timeval_add. That's because gnulib also replaces "struct
timeval" with its own definition, while libiberty expects the
system's.
E.g., in code like this:
gettimeofday (&prompt_ended, NULL);
timeval_sub (&prompt_delta, &prompt_ended, &prompt_started);
timeval_add (&prompt_for_continue_wait_time,
&prompt_for_continue_wait_time, &prompt_delta);
That's currently handled in gdb by not using gnulib's gettimeofday at
all (see common/gdb_sys_time.h), but that #undef hack won't work with
if/when we enable gnulib's C++ namespace support, because that mode
adds compile time warnings for uses of ::gettimeofday, which are hard
errors with -Werror.
#2 - But there's an elephant in the room: gettimeofday is not monotonic...
We're using it to:
a) check how long functions take, for performance analysis
b) compute when in the future to fire events in the event-loop
c) print debug timestamps
But that's exactly what gettimeofday is NOT meant for. Straight from
the man page:
~~~
The time returned by gettimeofday() is affected by
discontinuous jumps in the system time (e.g., if the system
administrator manually changes the system time). If you need a
monotonically increasing clock, see clock_gettime(2).
~~~
std::chrono (part of the C++11 standard library) has a monotonic clock
exactly for such purposes (std::chrono::steady_clock). This commit
switches to use that instead of gettimeofday, fixing all the issues
mentioned above.
gdb/ChangeLog:
2016-11-23 Pedro Alves <palves@redhat.com>
* Makefile.in (SFILES): Add common/run-time-clock.c.
(HFILES_NO_SRCDIR): Add common/run-time-clock.h.
(COMMON_OBS): Add run-time-clock.o.
* common/run-time-clock.c, common/run-time-clock.h: New files.
* defs.h (struct timeval, print_transfer_performance): Delete
declarations.
* event-loop.c (struct gdb_timer) <when>: Now a
std::chrono::steady_clock::time_point.
(create_timer): use std::chrono::steady_clock instead of
gettimeofday. Use new instead of malloc.
(delete_timer): Use delete instead of xfree.
(duration_cast_timeval): New.
(update_wait_timeout): Use std::chrono::steady_clock instead of
gettimeofday.
* maint.c: Include <chrono> instead of "gdb_sys_time.h", <time.h>
and "timeval-utils.h".
(scoped_command_stats::~scoped_command_stats)
(scoped_command_stats::scoped_command_stats): Use
std::chrono::steady_clock instead of gettimeofday. Use
user_cpu_time_clock instead of get_run_time.
* maint.h: Include "run-time-clock.h" and <chrono>.
(scoped_command_stats): <m_start_cpu_time>: Now a
user_cpu_time_clock::time_point.
<m_start_wall_time>: Now a std::chrono::steady_clock::time_point.
* mi/mi-main.c: Include "run-time-clock.h" and <chrono> instead of
"gdb_sys_time.h" and <sys/resource.h>.
(rusage): Delete.
(mi_execute_command): Use new instead of XNEW.
(mi_load_progress): Use std::chrono::steady_clock instead of
gettimeofday.
(timestamp): Rewrite in terms of std::chrono::steady_clock,
user_cpu_time_clock and system_cpu_time_clock.
(timeval_diff): Delete.
(print_diff): Adjust to use std::chrono::steady_clock,
user_cpu_time_clock and system_cpu_time_clock.
* mi/mi-parse.h: Include "run-time-clock.h" and <chrono> instead
of "gdb_sys_time.h".
(struct mi_timestamp): Change fields types to
std::chrono::steady_clock::time_point, user_cpu_time_clock::time
and system_cpu_time_clock::time_point, instead of struct timeval.
* symfile.c: Include <chrono> instead of <time.h> and
"gdb_sys_time.h".
(struct time_range): New.
(generic_load): Use std::chrono::steady_clock instead of
gettimeofday.
(print_transfer_performance): Replace timeval parameters with a
std::chrono::steady_clock::duration parameter. Adjust.
* utils.c: Include <chrono> instead of "timeval-utils.h",
"gdb_sys_time.h", and <time.h>.
(prompt_for_continue_wait_time): Now a
std::chrono::steady_clock::duration.
(defaulted_query, prompt_for_continue): Use
std::chrono::steady_clock instead of
gettimeofday/timeval_sub/timeval_add.
(reset_prompt_for_continue_wait_time): Use
std::chrono::steady_clock::duration instead of struct timeval.
(get_prompt_for_continue_wait_time): Return a
std::chrono::steady_clock::duration instead of struct timeval.
(vfprintf_unfiltered): Use std::chrono::steady_clock instead of
gettimeofday. Use std::string. Use '.' instead of ':'.
* utils.h: Include <chrono>.
(get_prompt_for_continue_wait_time): Return a
std::chrono::steady_clock::duration instead of struct timeval.
gdb/gdbserver/ChangeLog:
2016-11-23 Pedro Alves <palves@redhat.com>
* debug.c: Include <chrono> instead of "gdb_sys_time.h".
(debug_vprintf): Use std::chrono::steady_clock instead of
gettimeofday. Use '.' instead of ':'.
* tracepoint.c: Include <chrono> instead of "gdb_sys_time.h".
(get_timestamp): Use std::chrono::steady_clock instead of
gettimeofday.
Now that we require C++11 and all uses of gdb::unique_ptr and
gdb::move are gone, let's remove their definitions...
With my lazy hat on, I repurposed the header for "generally useful
unique_ptr specializations", and left gdb::unique_xmalloc_ptr in
there. Not sure whether we it'd be better move it out of the gdb
namespace or leave it be. I left it because it's less work and avoids
disrupting yet-unmerged patches that use it.
gdb/ChangeLog:
2016-11-15 Pedro Alves <palves@redhat.com>
* common/common-defs.h: Update comment.
* common/gdb_unique_ptr.h: Update header comment and copyright
year.
(gdb::unique_ptr, gdb::move): Delete.
This introduces the string_printf function. Like asprintf, but
returns a std::string.
gdb/ChangeLog:
2016-11-08 Pedro Alves <palves@redhat.com>
* Makefile.in (COMMON_OBS): Add utils-selftests.o.
* common/common-utils.c (string_printf): New function.
* common/common-utils.h: Include <string>.
(string_printf): Declare.
* utils-selftests.c: New file.
Revert commit f6abaf7a40 (gdb: no longer define
__STDC_CONSTANT_MACROS/__STDC_LIMIT_MACROS), with the tweak suggested
in that commit's log: the macros are now defined before any system
header is included.
This should fix AIX:
https://sourceware.org/ml/gdb-patches/2016-10/msg00682.html
gdb/ChangeLog:
2016-10-25 Pedro Alves <palves@redhat.com>
* common/common-defs.h (__STDC_CONSTANT_MACROS)
(__STDC_LIMIT_MACROS): Define.
Nowadays, if we build GDB with -fsanitize=address, we can get the asan
error below,
(gdb) quit
=================================================================
==9723==ERROR: AddressSanitizer: alloc-dealloc-mismatch (malloc vs operator delete) on 0x60200003bf70
#0 0x7f88f3837527 in operator delete(void*) (/usr/lib/x86_64-linux-gnu/libasan.so.1+0x55527)
#1 0xac8e13 in __gnu_cxx::new_allocator<void (*)()>::deallocate(void (**)(), unsigned long) /usr/include/c++/4.9/ext/new_allocator.h:110
#2 0xac8cc2 in __gnu_cxx::__alloc_traits<std::allocator<void (*)()> >::deallocate(std::allocator<void (*)()>&, void (**)(), unsigned long) /usr/include/c++/4.9/ext/alloc_traits.h:185
....
0x60200003bf70 is located 0 bytes inside of 8-byte region [0x60200003bf70,0x60200003bf78)
allocated by thread T0 here:
#0 0x7f88f38367ef in __interceptor_malloc (/usr/lib/x86_64-linux-gnu/libasan.so.1+0x547ef)
#1 0xbd2762 in operator new(unsigned long) /home/yao/SourceCode/gnu/gdb/git/gdb/common/new-op.c:42
#2 0xac8edc in __gnu_cxx::new_allocator<void (*)()>::allocate(unsigned long, void const*) /usr/include/c++/4.9/ext/new_allocator.h:104
#3 0xac8d81 in __gnu_cxx::__alloc_traits<std::allocator<void (*)()> >::allocate(std::allocator<void (*)()>&, unsigned long) /usr/include/c++/4.9/ext/alloc_traits.h:182
The reason for this is that we override operator new but don't override
operator delete. This patch does the override if the code is NOT
compiled with asan.
gdb:
2016-10-25 Yao Qi <yao.qi@linaro.org>
PR gdb/20716
* common/new-op.c (__has_feature): New macro.
Don't override operator new if asan is used.
This patch replaces many (but not all) uses of
make_cleanup_restore_integer with a simple RAII-based template class.
It also removes the similar restore_execution_direction cleanup in
favor of this new class. Subsequent patches will replace other
similar cleanups with this class.
The class is typically instantiated using make_scoped_restore. This
allows for template argument deduction.
2016-10-21 Tom Tromey <tom@tromey.com>
* common/scoped_restore.h: New file.
* utils.h: Include scoped_restore.h.
* top.c (execute_command_to_string): Use scoped_restore.
* python/python.c (python_interactive_command): Use
scoped_restore.
(python_command, execute_gdb_command): Likewise.
* printcmd.c (do_one_display): Use scoped_restore.
* mi/mi-main.c (exec_continue): Use scoped_restore.
* mi/mi-cmd-var.c (mi_cmd_var_assign): Use scoped_restore.
* linux-fork.c (checkpoint_command): Use scoped_restore.
* infrun.c (restore_execution_direction): Remove.
(fetch_inferior_event): Use scoped_restore.
* compile/compile.c (compile_file_command): Use
scoped_restore.
(compile_code_command, compile_print_command): Likewise.
* cli/cli-script.c (execute_user_command): Use
scoped_restore.
(while_command, if_command, script_from_file): Likewise.
* arm-tdep.c (arm_insert_single_step_breakpoint): Use
scoped_restore.
My gnulib fix at:
https://lists.gnu.org/archive/html/bug-gnulib/2015-11/msg00010.html
was merged upstream meanwhile and our gnulib copy now includes it.
As a concidence, Kevin was telling me today that these macros are
causing a build problem on FreeBSD:
common/common-defs.h:47:0: error: "__STDC_CONSTANT_MACROS" redefined [-Werror]
#define __STDC_CONSTANT_MACROS 1
/usr/include/sys/cdefs.h:408:0: note: this is the location of the previous definition
#define __STDC_CONSTANT_MACROS
(and a similar error for __STDC_LIMIT_MACROS)
The problem seems to be that we should be defining these input macros
before including any system header, but, we're not.
So let's just revert e063da6790 ([C++] Define __STDC_CONSTANT_MACROS
/ __STDC_LIMIT_MACROS for stdint.h). If this causes a problem
somewhere, we can re-define the macros higher up in the file, before
system headers are included.
gdb/ChangeLog:
2016-10-18 Pedro Alves <palves@redhat.com>
* common/common-defs.h (__STDC_CONSTANT_MACROS)
(__STDC_LIMIT_MACROS): Delete.
Many make_cleanup uses in the code base are best eliminated by using a
"owning" smart pointer to manage ownership of the resource
automatically.
The question is _which_ smart pointer.
GDB currently supports building with a C++03 compiler. We have
std::auto_ptr in C++03, but, as is collective wisdom by now, that's
too easy to misuse, and has therefore been deprecated in C++11 and
finally removed in C++17.
It'd be nice to be able to use std::unique_ptr instead, which is the
modern, safe std::auto_ptr replacement in C++11.
In addition to extra safety -- moving (i.e., transfer of ownership of
the managed pointer between smart pointers) must be explicit --
std::unique_ptr has (among others) one nice feature that std::auto_ptr
doesn't --- ability to specify a custom deleter as template parameter.
In gdb's context, that allows easily creating a smart pointer for
memory allocated with xmalloc -- the smart pointer then knows to
release with xfree instead of delete. This is particularly
interesting when managing objects allocated in C libraries, and also,
for C++-fying parts of GDB that interact with other parts that still
return objects allocated with xmalloc.
Since std::unique_ptr's API is quite nice, and eventually we'd like to
move to C++11, this patch adds a C++03-compatible smart pointer that
exposes the subset of the std::unique_ptr API that we're interested
in. An advantage is that whenever we start requiring C++11, we won't
have to learn a new API. Meanwhile, this allows continuing to support
building with a C++03 compiler.
Since C++03 doesn't support rvalue references (boost gets close to
emulating them, but it's not fully transparent to user code), the
C++03 std::unique_ptr emulation here doesn't try hard to prevent
accidentally moving, which is where most of complication of a more
thorough emulation would be. Instead, we rely on the fact that GDB
will be usually compiled with a C++11 compiler, and use the real
std::unique_ptr in that case to catch such accidental moves. IOW, the
goal here is to allow code that would be correct using std::unique_ptr
to be equally correct in C++03 mode, and, just as efficient.
The C++03 version was originally based on GCC 7.0's std::auto_ptr and
then heavily customized to behave more like C++11's std::unique_ptr:
- Support for custom (stateless) deleters. (Support for stateful
deleters could be added, if necessary.)
- unique_ptr<T[]> partial specialization (auto_ptr<T> does not know
to use delete[]).
- Support for all of 'ptr != NULL', 'ptr == NULL' and 'if (ptr)'
using the safe bool idiom to emulate C++11's explicit bool
operator.
- There's no nullptr in C++03, so this allows initialization and
assignment from NULL instead (std::auto_ptr allows neither).
- Variable names un-uglified (ie., no leading __ prefix everywhere).
- Formatting made to follow GDB's coding conventions, including
comment style.
- Converting "move" constructors done differently in order to truly
support:
unique_ptr<Derived> func_returning_unique_ptr (.....);
...
unique_ptr<Base> ptr = func_returning_unique_ptr (.....);
At this point, it no longer shares much at all with the original file,
but, that's the history.
See comments in the code to find out more.
I thought of putting the "emulation" / shim in the "std" namespace, so
that when we start requiring C++11 at some point, no actual changes to
users of the smart pointer throughout would be necessary. Putting
things in the std namespace is technically undefined, however in
practice it doesn't cause any issue with any compiler. However,
thinking that people might be confused with seeing std::unique_ptr and
thinking that we're actually requiring C++11 already, I put the new
types in the "gdb" namespace instead.
For managing xmalloc pointers, this adds a gdb::unique_xmalloc_ptr<T>
"specialization" with a custom xfree deleter.
No actual use of any smart pointer is introduced in this patch.
That'll be done in following patches.
Tested (along with the rest of the series) on:
- NetBSD 5.1 (gcc70 on the compile farm), w/ gcc 4.1.3
- x86-64 Fedora 23, gcc 5.3.1 (gnu++03)
- x86-64 Fedora 23, and gcc 7.0 (gnu++14)
gdb/ChangeLog:
2016-10-18 Pedro Alves <palves@redhat.com>
* common/common-defs.h: Include "gdb_unique_ptr.h".
* common/gdb_unique_ptr.h: New.
gdb/ChangeLog
2016-10-14 Eli Zaretskii <eliz@gnu.org>
* common/common-defs.h [HAVE_STRINGS_H]: Include strings.h if
available, to get prototypes of 'strcasecmp' and 'strncasecmp'.
If xmalloc fails allocating memory, usually because something tried a
huge allocation, like xmalloc(-1) or some such, GDB asks the user what
to do:
.../src/gdb/utils.c:1079: internal-error: virtual memory exhausted.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
Quit this debugging session? (y or n)
If the user says "n", that throws a QUIT exception, which is caught by
one of the multiple CATCH(RETURN_MASK_ALL) blocks somewhere up the
stack.
The default implementations of operator new / operator new[] call
malloc directly, and on memory allocation failure throw
std::bad_alloc. Currently, if that happens, since nothing catches it,
the exception escapes out of main, and GDB aborts from unhandled
exception.
This patch replaces the default operator new variants with versions
that, just like xmalloc:
#1 - Raise an internal-error on memory allocation failure.
#2 - Throw a QUIT gdb_exception, so that the exact same CATCH blocks
continue handling memory allocation problems.
A minor complication of #2 is that operator new can _only_ throw
std::bad_alloc, or something that extends it:
void* operator new (std::size_t size) throw (std::bad_alloc);
That means that if we let a gdb QUIT exception escape from within
operator new, the C++ runtime aborts due to unexpected exception
thrown.
So to bridge the gap, this patch adds a new gdb_quit_bad_alloc
exception type that inherits both std::bad_alloc and gdb_exception,
and throws _that_.
If we decide that we should be catching memory allocation errors in
fewer places than all the places we currently catch them (everywhere
we use RETURN_MASK_ALL currently), then we could change operator new
to throw plain std::bad_alloc then. But I'm considering such a change
as separate matter from this one -- it'd make sense to do the same to
xmalloc at the same time, for instance.
Meanwhile, this allows using new/new[] instead of xmalloc/XNEW/etc.
without losing the "virtual memory exhausted" internal-error
safeguard.
Tested on x86_64 Fedora 23.
gdb/ChangeLog:
2016-09-23 Pedro Alves <palves@redhat.com>
* Makefile.in (SFILES): Add common/new-op.c.
(COMMON_OBS): Add common/new-op.o.
(new-op.o): New rule.
* common/common-exceptions.h: Include <new>.
(struct gdb_quit_bad_alloc): New type.
* common/new-op.c: New file.
gdb/gdbserver/ChangeLog:
2016-09-23 Pedro Alves <palves@redhat.com>
* Makefile.in (SFILES): Add common/new-op.c.
(OBS): Add common/new-op.o.
(new-op.o): New rule.
Ref: https://sourceware.org/ml/gdb-patches/2016-09/msg00203.html
The std::{min,max} patch caused build failures when configuring GDB
with with --disable-nls and using GCC 4.1.
The reason is this bit in common/gdb_locale.h:
#ifdef ENABLE_NLS
...
#else
# define gettext(Msgid) (Msgid)
...
#endif
This causes problems if the <libintl.h> header is first included at
any point after "gdb_locale.h".
Specifically, the gettext&co declarations in libintl.h:
extern char *gettext (__const char *__msgid)
__THROW __attribute_format_arg__ (1);
end up broken after preprocessing:
extern char *(__const char *__msgid)
throw () __attribute__ ((__format_arg__ (1)));
After the std::min/std::max change to include <algorithm>, this now
happens with at least the GCC 4.1 copy of <algorithm>, which includes
<libintl.h> via <bits/stl_algobase.h>, <iosfwd>, and
<bits/c++locale.h>.
The fix is to simply remove the troublesome *gettext and *textdomain
macros, leaving only the _ and N_ ones.
gdb/ChangeLog:
2016-09-19 Pedro Alves <palves@redhat.com>
* common/gdb_locale.h [!ENABLE_NLS] (gettext, dgettext, dcgettext,
textdomain, bindtextdomain): Delete macros.
* main.c (captured_main) [!ENABLE_NLS]: Skip bintextdomain and
textdomain calls.
The ARI complains about this new file:
common/signals-state-save-restore.c:46: warning: gettext: All messages should be marked up with _.
common/signals-state-save-restore.c:59: warning: gettext: All messages should be marked up with _.
common/signals-state-save-restore.c:87: warning: gettext: All messages should be marked up with _.
common/signals-state-save-restore.c:92: warning: gettext: All messages should be marked up with _.
Since these are untranslatable strings, use () instead of _().
gdb/ChangeLog:
2016-08-10 Pedro Alves <palves@redhat.com>
* common/signals-state-save-restore.c
(save_original_signals_state, restore_original_signals_state):
Wrap perror_with_name arguments with '()'.
gdb's (or gdbserver's) own signal handling should not interfere with
the signal dispositions their spawned children inherit. However, it
currently does. For example, some paths in gdb cause SIGPIPE to be
set to SIG_IGN, and as consequence, the child starts with SIGPIPE to
set to SIG_IGN too, even though gdb was started with SIGPIPE set to
SIG_DFL.
This is because the exec family of functions does not reset the signal
disposition of signals that are set to SIG_IGN:
http://pubs.opengroup.org/onlinepubs/7908799/xsh/execve.html
Signals set to the default action (SIG_DFL) in the calling process
image are set to the default action in the new process
image. Signals set to be ignored (SIG_IGN) by the calling process
image are set to be ignored by the new process image. Signals set to
be caught by the calling process image are set to the default action
in the new process image (see <signal.h>).
And neither does it reset signal masks or flags.
In order to be transparent, when spawning new child processes to debug
(with "run", etc.), reset signal actions and mask back to what was
originally inherited from gdb/gdbserver's parent, just before execing
the target program to debug.
gdb/ChangeLog:
2016-08-09 Pedro Alves <palves@redhat.com>
PR gdb/18653
* Makefile.in (SFILES): Add
common/signals-state-save-restore.c.
(HFILES_NO_SRCDIR): Add common/signals-state-save-restore.h.
(COMMON_OBS): Add signals-state-save-restore.o.
(signals-state-save-restore.o): New rule.
* configure: Regenerate.
* fork-child.c: Include "signals-state-save-restore.h".
(fork_inferior): Call restore_original_signals_state.
* main.c: Include "signals-state-save-restore.h".
(captured_main): Call save_original_signals_state.
* common/common.m4: Add sigaction to AC_CHECK_FUNCS checks.
* common/signals-state-save-restore.c: New file.
* common/signals-state-save-restore.h: New file.
gdb/gdbserver/ChangeLog:
2016-08-09 Pedro Alves <palves@redhat.com>
PR gdb/18653
* Makefile.in (OBS): Add signals-state-save-restore.o.
(signals-state-save-restore.o): New rule.
* config.in: Regenerate.
* configure: Regenerate.
* linux-low.c: Include "signals-state-save-restore.h".
(linux_create_inferior): Call
restore_original_signals_state.
* server.c: Include "dispositions-save-restore.h".
(captured_main): Call save_original_signals_state.
gdb/testsuite/ChangeLog:
2016-08-09 Pedro Alves <palves@redhat.com>
PR gdb/18653
* gdb.base/signals-state-child.c: New file.
* gdb.base/signals-state-child.exp: New file.
* gdb.gdb/selftest.exp (do_steps_and_nexts): Add new pattern.
FreeBSD's librt uses SIGLIBRT as an internal signal to implement
SIGEV_THREAD sigevent notifications. Similar to SIGLWP or SIGCANCEL
this signal should be passed through to child processes by default.
include/ChangeLog:
* signals.def: Add GDB_SIGNAL_LIBRT.
gdb/ChangeLog:
* common/signals.c (gdb_signal_from_host): Handle SIGLIBRT.
(do_gdb_signal_to_host): Likewise.
* infrun.c (_initialize_infrun): Pass GDB_SIGNAL_LIBRT through to
programs.
* proc-events.c (signal_table): Add entry for SIGLIBRT.
Eclipse CDT now supports enabling execution recording using two methods
(full and btrace) and both formats for btrace (bts and pt). In the
event that recording is enabled behind the back of the GUI (by the user
on the command line, or a script), we need to know which method/format
are being used, so it can be correctly reflected in the interface. This
patch adds this information to the =record-started async record.
Before:
=record-started,thread-group="i1"
After:
=record-started,thread-group="i1",method="btrace",format="bts"
=record-started,thread-group="i1",method="btrace",format="pt"
=record-started,thread-group="i1",method="full"
The "format" field is only present when the current method supports
multiple formats (only the btrace method as of now).
gdb/ChangeLog:
* NEWS: Mention the new fields in =record-started.
* common/btrace-common.h (btrace_format_short_string): New function
declaration.
* common/btrace-common.c (btrace_format_short_string): New
function.
* mi/mi-interp.c (mi_record_changed): Output method and format
fields in the =record-started record.
* record-btrace.c (record_btrace_open): Adapt record_changed
notification.
* record-full.c (record_full_open): Likewise.
* record.c (cmd_record_stop): Likewise.
gdb/doc/ChangeLog:
* gdb.texinfo (GDB/MI Async Records): Document method and
format fields in =record-started.
* observer.texi (record_changed): Add method and format
parameters.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-record-changed.exp: Adjust =record-started output
matching.
The exceptions-across-readline issue was fixed by the previous commit.
Let's try this again.
gdb/ChangeLog:
2016-04-22 Pedro Alves <palves@redhat.com>
* common/common-exceptions.h (GDB_XCPT_TRY): Remove mention of
the foreign frames issue.
[__cplusplus] (GDB_XCPT): Define as GDB_XCPT_TRY.
If we map GDB'S TRY/CATCH macros to C++ try/catch, GDB breaks on
systems where readline isn't built with exceptions support. The
problem is that readline calls into GDB through the callback
interface, and if GDB's callback throws a C++ exception/error, the
system unwinder won't manage to unwind past the readline frame, and
ends up calling std::terminate(), which aborts the process:
(gdb) whatever-command-that-causes-an-error
terminate called after throwing an instance of 'gdb_exception_RETURN_MASK_ERROR'
Aborted
$
This went unnoticed for so long because:
- the x86-64 ABI requires -fasynchronous-unwind-tables, making it
possible for exceptions to cross readline with no special handling.
But e.g., on ARM or AIX, unless you build readline with
-fexceptions, you trip on the problem.
- TRY/CATCH was mapped to setjmp/longjmp, even in C++ mode, until
quite recently.
The fix is to catch and save any GDB exception that is thrown inside
the GDB readline callback, and then once the callback returns back to
the GDB code that called into readline in the first place, rethrow the
saved GDB exception.
This is similar in spirit to how we catch/map GDB exceptions at the
GDB/Python and GDB/Guile API boundaries.
The next question is then: if we intercept all exceptions within GDB's
readline callback, should we simply return normally to readline? The
callback prototype has no way to signal an error back to readline (*).
The answer is no -- if we return normally, we'll be returning to a
loop inside rl_callback_read_char that continues processing pending
input, calling into GDB again, redisplaying the prompt, etc. Thus if
we want to error out of rl_callback_read_char, we need to long jump
across it, just like we always did before TRY/CATCH were ever mapped
to C++ exceptions.
My first approach built a specialized API to handle this, with a
couple macros to hide the setjmp/longjmp and the struct gdb_exception
saving/rethrowing.
However, I realized that we need to:
- Handle multiple active rl_callback_read_char invocations. If,
while processing input something triggers a secondary prompt, we
end up in a nested rl_callback_read_char call, through
gdb_readline_wrapper.
- Propagate a struct gdb_exception along with the longjmp.
... and that this is exactly what the setjmp/longjmp-based TRY/CATCH
does.
So the fix makes the setjmp/longjmp TRY/CATCH always available under
new TRY_SJLJ/CATCH_SJLJ aliases, even when TRY/CATCH is mapped to C++
try/catch, and then uses TRY_SJLJ/CATCH_SJLJ to propagate GDB
exceptions across the readline callback.
This turns out to be a much better looking fix than my bespoke API
attempt, even. We'll probably be able to simplify TRY_SJLJ/CATCH_SJLJ
when we finally get rid of TRY/CATCH all over the tree, but until
then, this reuse seems quite nice for avoiding a second parallel
setjmp/longjmp mechanism.
(*) - maybe we could propose a readline API change, but we still need
to handle current readline, anyway.
gdb/ChangeLog:
2016-04-22 Pedro Alves <palves@redhat.com>
* common/common-exceptions.c (enum catcher_state, struct catcher)
(current_catcher): Define in C++ mode too.
(exceptions_state_mc_catch): Call throw_exception_sjlj instead of
throw_exception.
(throw_exception_sjlj, throw_exception_cxx): New functions,
factored out from throw_exception.
(throw_exception): Reimplement.
* common/common-exceptions.h (exceptions_state_mc_init)
(exceptions_state_mc_action_iter)
(exceptions_state_mc_action_iter_1, exceptions_state_mc_catch):
Declare in C++ mode too.
(TRY): Rename to ...
(TRY_SJLJ): ... this.
(CATCH): Rename to ...
(CATCH_SJLJ): ... this.
(END_CATCH): Rename to ...
(END_CATCH_SJLJ): ... this.
[GDB_XCPT == GDB_XCPT_SJMP] (TRY, CATCH, END_CATCH): Map to SJLJ
equivalents.
(throw_exception): Update comments.
(throw_exception_sjlj): Declare.
* event-top.c (gdb_rl_callback_read_char_wrapper): Extend intro
comment. Wrap body in TRY_SJLJ/CATCH_SJLJ and rethrow any
intercepted exception.
(gdb_rl_callback_handler): New function.
(gdb_rl_callback_handler_install): Always install
gdb_rl_callback_handler as readline callback.
Pedro Alves
Sergio Durigan Junior
Simon Marchi
Jan Kratochvil
Michael Sturm
Tom Tromey
Joel Brobecker
John Baldwin
Yao Qi
Eli Zaretskii
Nick Clifton