[v1.1,1/3] Introduce gdb::function_view
Commit Message
On 02/22/2017 05:40 PM, Pedro Alves wrote:
> I'll send the updated (squashed) patch as a reply.
Here is is. Thanks for the quick review!
From 997f0e94cbd2d1cdb4bf7612df1de72ce1a6eb56 Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Wed, 22 Feb 2017 17:43:09 +0000
Subject: [PATCH] Introduce gdb::function_view
This commit adds new function_view type. This type holds a a
non-owning reference to a callable. It is meant to be used as
callback type of functions, instead of using C-style pair of function
pointer and 'void *data' arguments. function_view allows passing
references to stateful function objects / lambdas w/ 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. I added a new gdb/unittests/ subdir this time,
instead of putting the tests under gdb/. If this is agreed to be a
good idea, some of the current selftests that exercise gdb/common/
things but live in gdb/ could move here (e.g., gdb/utils-selftests.c).
gdb/ChangeLog:
yyyy-mm-dd Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_UNITTESTS_SRCS, SUBDIR_UNITTESTS_OBS): New.
(SFILES): Add $(SUBDIR_UNITTEST_SRCS).
(COMMON_OBS): Add $(SUBDIR_UNITTEST_OBS).
(%.o) <unittests/%.c>: New pattern.
(INIT_FILES): Add $(SUBDIR_UNITTESTS_SRCS).
* common/function-view.h: New file.
* unittests/function-view-selftests.c: New file.
---
gdb/Makefile.in | 24 ++-
gdb/common/function-view.h | 355 ++++++++++++++++++++++++++++++++
gdb/unittests/function-view-selftests.c | 184 +++++++++++++++++
3 files changed, 560 insertions(+), 3 deletions(-)
create mode 100644 gdb/common/function-view.h
create mode 100644 gdb/unittests/function-view-selftests.c
Comments
On 17-02-22 17:49:22, Pedro Alves wrote:
> On 02/22/2017 05:40 PM, Pedro Alves wrote:
>
> > I'll send the updated (squashed) patch as a reply.
>
> Here is is. Thanks for the quick review!
>
> From 997f0e94cbd2d1cdb4bf7612df1de72ce1a6eb56 Mon Sep 17 00:00:00 2001
> From: Pedro Alves <palves@redhat.com>
> Date: Wed, 22 Feb 2017 17:43:09 +0000
> Subject: [PATCH] Introduce gdb::function_view
>
> This commit adds new function_view type. This type holds a a
Double "a" at the end.
> non-owning reference to a callable. It is meant to be used as
> callback type of functions, instead of using C-style pair of function
> pointer and 'void *data' arguments. function_view allows passing
> references to stateful function objects / lambdas w/ 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. I added a new gdb/unittests/ subdir this time,
> instead of putting the tests under gdb/. If this is agreed to be a
> good idea, some of the current selftests that exercise gdb/common/
> things but live in gdb/ could move here (e.g., gdb/utils-selftests.c).
I wanted to add gdb/unittests for a while, but didn't have a chance to
do so. Yes, it is a good idea to me. How is GDB unit tests, like
disasm-selftests.c? Do you want to move it to gdb/unittests/?
> + C++11 gave us std::function as type-erased wrapper around arbitrary
> + callables, however, std::function is not an ideal fit for transient
> + callbacks such as the use case above. For this use case, which is
> + quite pervasive, a function_view is a better choice, because while
> + while function_view is light and does not require any heap
Double "while"s.
> +namespace gdb {
> +
This is a new namespace in gdb source. What is the rule of using this
namespace?
On 17-02-22 17:49:22, Pedro Alves wrote:
> diff --git a/gdb/Makefile.in b/gdb/Makefile.in
> index 43253d3..a4cac36 100644
> --- a/gdb/Makefile.in
> +++ b/gdb/Makefile.in
> @@ -523,6 +523,12 @@ SUBDIR_PYTHON_DEPS =
> SUBDIR_PYTHON_LDFLAGS =
> SUBDIR_PYTHON_CFLAGS =
>
> +SUBDIR_UNITTESTS_SRCS = \
> + unittests/function-view-selftests.c
> +
> +SUBDIR_UNITTESTS_OBS = \
> + function-view-selftests.o
> +
> # Opcodes currently live in one of two places. Either they are in the
> # opcode library, typically ../opcodes, or they are in a header file
> # in INCLUDE_DIR.
> @@ -1216,7 +1222,8 @@ SFILES = \
> common/xml-utils.c \
> mi/mi-common.c \
> target/waitstatus.c \
> - $(SUBDIR_GCC_COMPILE_SRCS)
> + $(SUBDIR_GCC_COMPILE_SRCS) \
> + $(SUBDIR_UNITTEST_SRCS)
>
> LINTFILES = $(SFILES) $(YYFILES) $(CONFIG_SRCS) init.c
>
> @@ -1800,7 +1807,8 @@ COMMON_OBS = $(DEPFILES) $(CONFIG_OBS) $(YYOBJ) \
> xml-syscall.o \
> xml-tdesc.o \
> xml-utils.o \
> - $(SUBDIR_GCC_COMPILE_OBS)
> + $(SUBDIR_GCC_COMPILE_OBS) \
> + $(SUBDIR_UNITTESTS_OBS)
>
Can we add SUBDIR_UNITESTS_OBS to CONFIG_OBS (in configure.ac) if
GDB_SELF_TEST is defined?
> TSOBS = inflow.o
>
> @@ -1909,6 +1917,10 @@ all: gdb$(EXEEXT) $(CONFIG_ALL)
> $(COMPILE) $<
> $(POSTCOMPILE)
>
> +%.o: ${srcdir}/unittests/%.c
> + $(COMPILE) $<
> + $(POSTCOMPILE)
> +
> # Specify an explicit rule for gdb/common/agent.c, to avoid a clash with the
> # object file generate by gdb/agent.c.
> common-agent.o: $(srcdir)/common/agent.c
> @@ -2124,7 +2136,13 @@ test-cp-name-parser$(EXEEXT): test-cp-name-parser.o $(LIBIBERTY)
> # duplicates. Files in the gdb/ directory can end up appearing in
> # COMMON_OBS (as a .o file) and CONFIG_SRCS (as a .c file).
>
> -INIT_FILES = $(COMMON_OBS) $(TSOBS) $(CONFIG_SRCS) $(SUBDIR_GCC_COMPILE_SRCS)
> +INIT_FILES = \
> + $(COMMON_OBS) \
> + $(TSOBS) \
> + $(CONFIG_SRCS) \
> + $(SUBDIR_GCC_COMPILE_SRCS) \
> + $(SUBDIR_UNITTESTS_SRCS)
> +
If so, SUBDIR_UNITESTS_SRCS is added to CONFIG_SRCS, we don't need this
change.
> diff --git a/gdb/unittests/function-view-selftests.c b/gdb/unittests/function-view-selftests.c
> new file mode 100644
> index 0000000..3e5369b
> --- /dev/null
> +++ b/gdb/unittests/function-view-selftests.c
> @@ -0,0 +1,184 @@
> +/* Self tests for function_view for GDB, the GNU debugger.
> +
> + Copyright (C) 2017 Free Software Foundation, Inc.
> +
> + This file is part of GDB.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +#include "defs.h"
> +#include "selftest.h"
> +#include "common/function-view.h"
> +
> +#if GDB_SELF_TEST
If gdb/unittests/*.c are compiled when GDB_SELF_TEST is defined, we
don't need to check GDB_SELF_TEST here.
> +
> +namespace selftests {
> +namespace function_view {
> +
> +
> +} /* namespace function_view */
> +} /* namespace selftests */
> +
> +#endif
> +
> +void
> +_initialize_function_view_selftests ()
> +{
> +#if GDB_SELF_TEST
This is not needed as well.
> + register_self_test (selftests::function_view::run_tests);
> +#endif
> +}
> --
> 2.5.5
>
>
On 02/22/2017 10:11 PM, Yao Qi wrote:
> Double "a" at the end.
>
> Double "while"s.
Thanks, fixed.
>
>> +namespace gdb {
>> +
>
> This is a new namespace in gdb source. What is the rule of using this
> namespace?
I don't have a hard rule predetermined. I originally put the C++03
unique_ptr shim under the gdb namespace to make it clear at the
call sites that that was not the std type, but gdb's replacement.
I.e., gdb::unique_ptr vs std::unique_ptr. Since then, gdb::unique_ptr is
gone, but gdb::unique_xmalloc_ptr remains, and we've been putting
these generic library-like utilities under the namespace:
common/function-view.h:154:namespace gdb {
common/function-view.h:353:} /* namespace gdb */
common/gdb_unlinker.h:23:namespace gdb
common/gdb_unique_ptr.h:25:namespace gdb
common/gdb_unique_ptr.h:43:} /* namespace gdb */
common/gdb_optional.h:23:namespace gdb
common/gdb_ref_ptr.h:25:namespace gdb
I think putting these new things in _some_ namespace is the
right thing to do. gdb is just the no-brainer namespace name.
IMO, _all_ of GDB should be under "namespace gdb". Then these utilities
would either be put in "namespace gdb" too, or in a "namespace gtl",
for "gdb template library" or something like that. Or we could
put them under "gtl" already.
I have a series that put the whole of gdb under namespace gdb:
https://github.com/palves/gdb/commits/palves/cxx-gdb-namespace
I wrote that while experimenting with ways to use
gnulib's C++ namespace mode, in order to avoid the
"#define open rpl_open" defines gnulib does... I nowadays
believe the put-gdb-in-a-namespace approach is the best one.
However, when we put gdb under a namespace, debugging GDB itself
becomes a bit harder, since you then have to do "b gdb::some_function"
instead of "b some_function". This is just dogfooding pain,
of course, users run into this too...
That's what led to me to work on this branch:
https://github.com/palves/gdb/commits/palves/cp-linespec-2
with that, "b some_function" sets breakpoints on all
functions/methods named some_function in all namespaces.
And it was testing that branch with --target_board=dwarf4-gdb-index
that led to the bug fixed by this series... It's a never
ending story... :-)
Thanks,
Pedro Alves
Pedro Alves <palves@redhat.com> writes:
> I think putting these new things in _some_ namespace is the
> right thing to do. gdb is just the no-brainer namespace name.
Yes, that right.
>
> IMO, _all_ of GDB should be under "namespace gdb". Then these utilities
> would either be put in "namespace gdb" too, or in a "namespace gtl",
> for "gdb template library" or something like that. Or we could
> put them under "gtl" already.
IMO, "gdb template library" or "gtl" is project name, but it doesn't fit
well as a namespace name. I like "gdb::utils" :).
On 02/23/2017 03:11 PM, Yao Qi wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> I think putting these new things in _some_ namespace is the
>> right thing to do. gdb is just the no-brainer namespace name.
>
> Yes, that right.
>
>>
>> IMO, _all_ of GDB should be under "namespace gdb". Then these utilities
>> would either be put in "namespace gdb" too, or in a "namespace gtl",
>> for "gdb template library" or something like that. Or we could
>> put them under "gtl" already.
>
> IMO, "gdb template library" or "gtl" is project name, but it doesn't fit
> well as a namespace name. I like "gdb::utils" :).
"gdb::utils::function_view" is maybe a bit too long for code
that is going to end up used a lot, though.
"gdb::function_view" or "gtl::function_view" also has the "advantage"
that replacing "gdb" with "std" at some point does not require
reindenting the code. (e.g., gdb::optional -> std::optional when
we get to C++17). 1/2 :-)
I think the easiest is to keep using "gdb" for now, and revisit
if/when we move everything under gdb.
Thanks,
Pedro Alves
On Thu, Feb 23, 2017 at 3:20 PM, Pedro Alves <palves@redhat.com> wrote:
>
> I think the easiest is to keep using "gdb" for now, and revisit
> if/when we move everything under gdb.
>
That is fine by me.
@@ -523,6 +523,12 @@ SUBDIR_PYTHON_DEPS =
SUBDIR_PYTHON_LDFLAGS =
SUBDIR_PYTHON_CFLAGS =
+SUBDIR_UNITTESTS_SRCS = \
+ unittests/function-view-selftests.c
+
+SUBDIR_UNITTESTS_OBS = \
+ function-view-selftests.o
+
# Opcodes currently live in one of two places. Either they are in the
# opcode library, typically ../opcodes, or they are in a header file
# in INCLUDE_DIR.
@@ -1216,7 +1222,8 @@ SFILES = \
common/xml-utils.c \
mi/mi-common.c \
target/waitstatus.c \
- $(SUBDIR_GCC_COMPILE_SRCS)
+ $(SUBDIR_GCC_COMPILE_SRCS) \
+ $(SUBDIR_UNITTEST_SRCS)
LINTFILES = $(SFILES) $(YYFILES) $(CONFIG_SRCS) init.c
@@ -1800,7 +1807,8 @@ COMMON_OBS = $(DEPFILES) $(CONFIG_OBS) $(YYOBJ) \
xml-syscall.o \
xml-tdesc.o \
xml-utils.o \
- $(SUBDIR_GCC_COMPILE_OBS)
+ $(SUBDIR_GCC_COMPILE_OBS) \
+ $(SUBDIR_UNITTESTS_OBS)
TSOBS = inflow.o
@@ -1909,6 +1917,10 @@ all: gdb$(EXEEXT) $(CONFIG_ALL)
$(COMPILE) $<
$(POSTCOMPILE)
+%.o: ${srcdir}/unittests/%.c
+ $(COMPILE) $<
+ $(POSTCOMPILE)
+
# Specify an explicit rule for gdb/common/agent.c, to avoid a clash with the
# object file generate by gdb/agent.c.
common-agent.o: $(srcdir)/common/agent.c
@@ -2124,7 +2136,13 @@ test-cp-name-parser$(EXEEXT): test-cp-name-parser.o $(LIBIBERTY)
# duplicates. Files in the gdb/ directory can end up appearing in
# COMMON_OBS (as a .o file) and CONFIG_SRCS (as a .c file).
-INIT_FILES = $(COMMON_OBS) $(TSOBS) $(CONFIG_SRCS) $(SUBDIR_GCC_COMPILE_SRCS)
+INIT_FILES = \
+ $(COMMON_OBS) \
+ $(TSOBS) \
+ $(CONFIG_SRCS) \
+ $(SUBDIR_GCC_COMPILE_SRCS) \
+ $(SUBDIR_UNITTESTS_SRCS)
+
init.c: $(INIT_FILES)
@echo Making init.c
@rm -f init.c-tmp init.l-tmp
new file mode 100644
@@ -0,0 +1,355 @@
+/* Copyright (C) 2017 Free Software Foundation, Inc.
+
+ This file is part of GDB.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+#ifndef COMMON_FUNCTION_VIEW_H
+#define COMMON_FUNCTION_VIEW_H
+
+/* function_view is a polymorphic type-erasing wrapper class that
+ encapsulates a non-owning reference to arbitrary callable objects.
+
+ A way to put it is that function_view is to std::function like
+ std::string_view is to std::string. While std::function stores a
+ type-erased callable object internally, function_view holds a
+ type-erased reference to an external callable object.
+
+ This is meant to be used as callback type of a function that:
+
+ #1 - Takes a callback as parameter.
+
+ #2 - Wants to support arbitrary callable objects as callback type
+ (e.g., stateful function objects, lambda closures, free
+ functions).
+
+ #3 - Does not store the callback anywhere; instead the function
+ just calls the callback directly or forwards it to some
+ other function that calls it.
+
+ #4 - Can't be, or we don't want it to be, a template function
+ with the callable type as template parameter. For example,
+ when the callback is a parameter of a virtual member
+ function, or when putting the function template in a header
+ would expose too much implementation detail.
+
+ Note that the C-style "function pointer" + "void *data" callback
+ parameter idiom fails requirement #2 above. Please don't add new
+ uses of that idiom. I.e., something like this wouldn't work;
+
+ typedef bool (iterate_over_foos_cb) (foo *f, void *user_data),
+ void iterate_over_foos (iterate_over_foos_cb *callback, void *user_data);
+
+ foo *find_foo_by_type (int type)
+ {
+ foo *found = nullptr;
+
+ iterate_over_foos ([&] (foo *f, void *data)
+ {
+ if (foo->type == type)
+ {
+ found = foo;
+ return true; // stop iterating
+ }
+ return false; // continue iterating
+ }, NULL);
+
+ return found;
+ }
+
+ The above wouldn't compile, because lambdas with captures can't be
+ implicitly converted to a function pointer (because a capture means
+ some context data must be passed to the lambda somehow).
+
+ C++11 gave us std::function as type-erased wrapper around arbitrary
+ callables, however, std::function is not an ideal fit for transient
+ callbacks such as the use case above. For this use case, which is
+ quite pervasive, a function_view is a better choice, because while
+ while function_view is light and does not require any heap
+ allocation, std::function is a heavy-weight object with value
+ semantics that generally requires a heap allocation on
+ construction/assignment of the target callable. In addition, while
+ it is possible to use std::function in such a way that avoids most
+ of the overhead by making sure to only construct it with callables
+ of types that fit std::function's small object optimization, such
+ as function pointers and std::reference_wrapper callables, that is
+ quite inconvenient in practice, because restricting to
+ free-function callables would imply no state/capture/closure, which
+ we need in most cases, and std::reference_wrapper implies
+ remembering to use std::ref/std::cref where the callable is
+ constructed, with the added inconvenience that std::ref/std::cref
+ have deleted rvalue-ref overloads, meaning you can't use
+ unnamed/temporary lambdas with them.
+
+ Note that because function_view is a non-owning view of a callable,
+ care must be taken to ensure that the callable outlives the
+ function_view that calls it. This is not really a problem for the
+ use case function_view is intended for, such as passing a temporary
+ function object / lambda to a function that accepts a callback,
+ because in those cases, the temporary is guaranteed to be live
+ until the called function returns.
+
+ Calling a function_view with no associated target is undefined,
+ unlike with std::function, which throws bad_function_call. This is
+ by design, to avoid the otherwise necessary NULL check in
+ function_view::operator().
+
+ Since function_view objects are small (a pair of pointers), they
+ should generally be passed around by value.
+
+ Usage:
+
+ Given this function that accepts a callback:
+
+ void
+ iterate_over_foos (gdb::function_view<void (foo *)> callback)
+ {
+ for (auto &foo : foos)
+ callback (&foo);
+ }
+
+ you can call it like this, passing a lambda as callback:
+
+ iterate_over_foos ([&] (foo *f)
+ {
+ process_one_foo (f);
+ });
+
+ or like this, passing a function object as callback:
+
+ struct function_object
+ {
+ void operator() (foo *f)
+ {
+ if (s->check ())
+ process_one_foo (f);
+ }
+
+ // some state
+ state *s;
+ };
+
+ state mystate;
+ function_object matcher {&mystate};
+ iterate_over_foos (matcher);
+
+ or like this, passing a function pointer as callback:
+
+ iterate_over_foos (process_one_foo);
+
+ You can find unit tests covering the whole API in
+ unittests/function-view-selftests.c. */
+
+namespace gdb {
+
+namespace traits {
+ /* A few trait helpers. */
+ template<typename Predicate>
+ struct Not : public std::integral_constant<bool, !Predicate::value>
+ {};
+
+ template<typename...>
+ struct Or;
+
+ template<>
+ struct Or<> : public std::false_type
+ {};
+
+ template<typename B1>
+ struct Or<B1> : public B1
+ {};
+
+ template<typename B1, typename B2>
+ struct Or<B1, B2>
+ : public std::conditional<B1::value, B1, B2>::type
+ {};
+
+ template<typename B1,typename B2,typename B3, typename... Bn>
+ struct Or<B1, B2, B3, Bn...>
+ : public std::conditional<B1::value, B1, Or<B2, B3, Bn...>>::type
+ {};
+} /* namespace traits */
+
+namespace fv_detail {
+/* Bits shared by all function_view instantiations that do not depend
+ on the template parameters. */
+
+/* Storage for the erased callable. This is a union in order to be
+ able to save both a function object (data) pointer or a function
+ pointer without triggering undefined behavior. */
+union erased_callable
+{
+ /* For function objects. */
+ void *data;
+
+ /* For function pointers. */
+ void (*fn) ();
+};
+
+} /* namespace fv_detail */
+
+/* Use partial specialization to get access to the callable's
+ signature. */
+template<class Signature>
+struct function_view;
+
+template<typename Res, typename... Args>
+class function_view<Res (Args...)>
+{
+ template<typename From, typename To>
+ using CompatibleReturnType
+ = traits::Or<std::is_void<To>,
+ std::is_same<From, To>,
+ std::is_convertible<From, To>>;
+
+ /* True if Func can be called with Args, and the result, and the
+ result is convertible to Res, unless Res is void. */
+ template<typename Callable,
+ typename Res2 = typename std::result_of<Callable &(Args...)>::type>
+ struct IsCompatibleCallable : CompatibleReturnType<Res2, Res>
+ {};
+
+ /* True if Callable is a function_view. Used to avoid hijacking the
+ copy ctor. */
+ template <typename Callable>
+ struct IsFunctionView
+ : std::is_same<function_view, typename std::decay<Callable>::type>
+ {};
+
+ /* Helper to make SFINAE logic easier to read. */
+ template<typename Condition>
+ using Requires = typename std::enable_if<Condition::value, void>::type;
+
+ public:
+
+ /* NULL by default. */
+ constexpr function_view () noexcept
+ : m_erased_callable {},
+ m_invoker {}
+ {}
+
+ /* Default copy/assignment is fine. */
+ function_view (const function_view &) = default;
+ function_view &operator= (const function_view &) = default;
+
+ /* This is the main entry point. Use SFINAE to avoid hijacking the
+ copy constructor and to ensure that the target type is
+ compatible. */
+ template
+ <typename Callable,
+ typename = Requires<traits::Not<IsFunctionView<Callable>>>,
+ typename = Requires<IsCompatibleCallable<Callable>>>
+ function_view (Callable &&callable) noexcept
+ {
+ bind (callable);
+ }
+
+ /* Construct a NULL function_view. */
+ constexpr function_view (std::nullptr_t) noexcept
+ : m_erased_callable {},
+ m_invoker {}
+ {}
+
+ /* Clear a function_view. */
+ function_view &operator= (std::nullptr_t) noexcept
+ {
+ m_invoker = nullptr;
+ return *this;
+ }
+
+ /* Return true if the wrapper has a target, false otherwise. Note
+ we check M_INVOKER instead of M_ERASED_CALLABLE because we don't
+ know which member of the union is active right now. */
+ constexpr explicit operator bool () const noexcept
+ { return m_invoker != nullptr; }
+
+ /* Call the callable. */
+ Res operator () (Args... args) const
+ { return m_invoker (m_erased_callable, std::forward<Args> (args)...); }
+
+ private:
+
+ /* Bind this function_view to a compatible function object
+ reference. */
+ template <typename Callable>
+ void bind (Callable &callable) noexcept
+ {
+ m_erased_callable.data = (void *) std::addressof (callable);
+ m_invoker = [] (fv_detail::erased_callable ecall, Args... args)
+ noexcept (noexcept (callable (std::forward<Args> (args)...))) -> Res
+ {
+ auto &restored_callable = *static_cast<Callable *> (ecall.data);
+ /* The explicit cast to Res avoids a compile error when Res is
+ void and the callable returns non-void. */
+ return (Res) restored_callable (std::forward<Args> (args)...);
+ };
+ }
+
+ /* Bind this function_view to a compatible function pointer.
+
+ Making this a separate function allows avoiding one indirection,
+ by storing the function pointer directly in the storage, instead
+ of a pointer to pointer. erased_callable is then a union in
+ order to avoid storing a function pointer as a data pointer here,
+ which would be undefined. */
+ template<class Res2, typename... Args2>
+ void bind (Res2 (*fn) (Args2...)) noexcept
+ {
+ m_erased_callable.fn = reinterpret_cast<void (*) ()> (fn);
+ m_invoker = [] (fv_detail::erased_callable ecall, Args... args)
+ noexcept (noexcept (fn (std::forward<Args> (args)...))) -> Res
+ {
+ auto restored_fn = reinterpret_cast<Res2 (*) (Args2...)> (ecall.fn);
+ /* The explicit cast to Res avoids a compile error when Res is
+ void and the callable returns non-void. */
+ return (Res) restored_fn (std::forward<Args> (args)...);
+ };
+ }
+
+ /* Storage for the erased callable. */
+ fv_detail::erased_callable m_erased_callable;
+
+ /* The invoker. This is set to a capture-less lambda by one of the
+ 'bind' overloads. The lambda restores the right type of the
+ callable (which is passed as first argument), and forwards the
+ args. */
+ Res (*m_invoker) (fv_detail::erased_callable, Args...);
+};
+
+/* Allow comparison with NULL. Defer the work to the in-class
+ operator bool implementation. */
+
+template<typename Res, typename... Args>
+constexpr inline bool
+operator== (const function_view<Res (Args...)> &f, std::nullptr_t) noexcept
+{ return !static_cast<bool> (f); }
+
+template<typename Res, typename... Args>
+constexpr inline bool
+operator== (std::nullptr_t, const function_view<Res (Args...)> &f) noexcept
+{ return !static_cast<bool> (f); }
+
+template<typename Res, typename... Args>
+constexpr inline bool
+operator!= (const function_view<Res (Args...)> &f, std::nullptr_t) noexcept
+{ return static_cast<bool> (f); }
+
+template<typename Res, typename... Args>
+constexpr inline bool
+operator!= (std::nullptr_t, const function_view<Res (Args...)> &f) noexcept
+{ return static_cast<bool> (f); }
+
+} /* namespace gdb */
+
+#endif
new file mode 100644
@@ -0,0 +1,184 @@
+/* Self tests for function_view for GDB, the GNU debugger.
+
+ Copyright (C) 2017 Free Software Foundation, Inc.
+
+ This file is part of GDB.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+#include "defs.h"
+#include "selftest.h"
+#include "common/function-view.h"
+
+#if GDB_SELF_TEST
+
+namespace selftests {
+namespace function_view {
+
+static int
+plus_one_fn_int (int val)
+{
+ return ++val;
+}
+
+static short
+plus_one_fn_short (short val)
+{
+ return ++val;
+}
+
+static int
+call_callback_int (int val, gdb::function_view <int (int)> callback)
+{
+ return callback (val);
+}
+
+static void
+call_callback_void (int val, gdb::function_view <void (int)> callback)
+{
+ callback (val);
+}
+
+struct plus_one_int_func_obj
+{
+ int operator () (int val)
+ {
+ ++call_count;
+ return ++val;
+ }
+
+ /* Number of times called. */
+ int call_count = 0;
+};
+
+static void
+run_tests ()
+{
+ /* A simple lambda. */
+ auto plus_one_lambda = [] (int val) { return ++val; };
+
+ /* A function_view that references the lambda. */
+ gdb::function_view<int (int)> plus_one_func_view (plus_one_lambda);
+
+ /* Check calling the lambda directly. */
+ SELF_CHECK (plus_one_lambda (0) == 1);
+ SELF_CHECK (plus_one_lambda (1) == 2);
+
+ /* Check calling lambda via the view. */
+ SELF_CHECK (plus_one_func_view (2) == 3);
+ SELF_CHECK (plus_one_func_view (3) == 4);
+
+ /* Check calling a function that takes a function_view as argument,
+ by value. Pass a lambda, making sure a function_view is properly
+ constructed implicitly. */
+ SELF_CHECK (call_callback_int (1, [] (int val)
+ {
+ return val + 2;
+ }) == 3);
+
+ /* Same, passing a named/lvalue lambda. */
+ SELF_CHECK (call_callback_int (1, plus_one_lambda) == 2);
+ /* Same, passing named/lvalue function_view (should copy). */
+ SELF_CHECK (call_callback_int (1, plus_one_func_view) == 2);
+
+ /* Check constructing a function view over a function-object
+ callable, and calling it. */
+ plus_one_int_func_obj func_obj;
+ SELF_CHECK (func_obj (0) == 1);
+ SELF_CHECK (call_callback_int (1, func_obj) == 2);
+ /* Check that the callable was referenced, not copied. */
+ SELF_CHECK (func_obj.call_count == 2);
+
+ /* Check constructing a function_view over a free-function callable,
+ and calling it. */
+ SELF_CHECK (call_callback_int (1, plus_one_fn_int) == 2);
+
+ /* Check calling a function with a
+ compatible-but-not-exactly-the-same prototype. */
+ SELF_CHECK (call_callback_int (1, [] (short val) -> short
+ {
+ return val + 2;
+ }) == 3);
+ /* Same, but passing a function pointer. */
+ SELF_CHECK (call_callback_int (1, plus_one_fn_short) == 2);
+
+ /* Like std::function, a function_view that expects a void return
+ can reference callables with non-void return type. The result is
+ simply discarded. Check a lambda, function object and a function
+ pointer. */
+ call_callback_void (1, [] (int val) -> int
+ {
+ return val + 2;
+ });
+ call_callback_void (1, func_obj);
+ call_callback_void (1, plus_one_fn_int);
+
+ /* Check that the main ctor doesn't hijack the copy ctor. */
+ auto plus_one_func_view2 (plus_one_func_view);
+ auto plus_one_func_view3 (plus_one_func_view2);
+ static_assert (std::is_same<decltype (plus_one_func_view),
+ decltype (plus_one_func_view2)>::value, "");
+ static_assert (std::is_same<decltype (plus_one_func_view),
+ decltype (plus_one_func_view3)>::value, "");
+
+ SELF_CHECK (plus_one_func_view3 (1) == 2);
+
+ /* Likewise, but propagate a NULL callable. If this calls the main
+ function_view ctor instead of the copy ctor by mistake, then
+ null_func_2 ends up non-NULL (because it'd instead reference
+ null_func_1 as just another callable). */
+ constexpr gdb::function_view<int (int)> null_func_view_1 = nullptr;
+ constexpr auto null_func_view_2 (null_func_view_1);
+
+ /* While at it, check whether the function_view is bound using
+ various forms, op==, op!= and op bool. */
+
+ /* op== */
+ static_assert (null_func_view_2 == nullptr, "");
+ static_assert (nullptr == null_func_view_2, "");
+ static_assert (null_func_view_2 == NULL, "");
+ static_assert (NULL == null_func_view_2, "");
+
+ /* op!= */
+ static_assert (!(null_func_view_2 != nullptr), "");
+ static_assert (!(nullptr != null_func_view_2), "");
+ static_assert (!(null_func_view_2 != NULL), "");
+ static_assert (!(NULL != null_func_view_2), "");
+
+ /* op bool */
+ static_assert (!null_func_view_2, "");
+
+ /* Check the nullptr_t ctor. */
+ constexpr gdb::function_view<int (int)> check_ctor_nullptr (nullptr);
+ static_assert (!check_ctor_nullptr, "");
+
+ /* Check the nullptr_t op= */
+ gdb::function_view<int (int)> check_op_eq_null (plus_one_fn_int);
+ SELF_CHECK (check_op_eq_null);
+ check_op_eq_null = nullptr;
+ SELF_CHECK (!check_op_eq_null);
+}
+
+} /* namespace function_view */
+} /* namespace selftests */
+
+#endif
+
+void
+_initialize_function_view_selftests ()
+{
+#if GDB_SELF_TEST
+ register_self_test (selftests::function_view::run_tests);
+#endif
+}