TOC will be added in the final post, since github doesn't recognize [TOC] option...
This file is published as bfilipek.com/C++17 features blog post.
##Intro
The list is mostly done! Still some descriptions could be improved or more example could be provided.
If you have code examples, better explanations or any ideas, let me know! I am happy to update the current post so that it has some real value for others.
The plan is to have a list of features with some basic explanation, little example (if possible) and some additional resources, plus a note about availability in compilers. Probably, most of the features might require separate articles or even whole chapters in books, so the list here will be only a jump start.
The list comes from the following resources:
- SO: What are the new features in C++17?
- cppreference.com/C++ compiler support.
- AnthonyCalandra/modern-cpp-features cheat sheet - unfortunately it doesn't include all the features of C++17, so this is also why I've started a separate try on that.
- plus other findings and mentions
And one of the most important resource: Working Draft, Standard for Programming Language C++
##Language Features
###New auto rules for direct-list-initialization N3922
GCC: 5.0 | Clang: 3.8 | MSVC: 14.0 |
---|
Fixes some cases with auto type deduction. The full background can be found in Auto and braced-init-lists, by Ville Voutilainen.
It fixes the problem of deducing std::initializer_list
like:
auto x = foo(); // copy-initialization
auto x{foo}; // direct-initialization, initializes an initializer_list
int x = foo(); // copy-initialization
int x{foo}; // direct-initialization
And for the direct initialization, new rules are:
- For a braced-init-list with only a single element, auto deduction will deduce from that entry;
- For a braced-init-list with more than one element, auto deduction will be ill-formed.
Basically, auto x { 1 };
will be now deduced as int
, but before it was an initializer list.
###static_assert with no message N3928
GCC: 6.0 | Clang: 2.5 | MSVC: 15.0 preview 5 |
---|
Self-explanatory. It allows just to have the condition without passing the message, version with the message will also be available. It will be compatible with other asserts like BOOST_STATIC_ASSERT
(that didn't take any message from the start).
###typename in a template template parameter
GCC: 5.0 | Clang: 3.5 | MSVC: 14.0 |
---|
Allows you to use typename
instead of class
when declaring a template template parameter. Normal type parameters can use them interchangeably, but template template parameters were restricted to class
, so this change unifies these forms somewhat.
template <template <typename...> typename Container>
// used to be invalid ^^^^^^^^
struct foo;
foo<std::vector> my_foo;
###Removing trigraphs N4086
GCC: 5.1 | Clang: 3.5 | MSVC: Yes |
---|
Removes ??=
, ??(
, ??>
, ...
Makes the implementation a bit simpler, see MSDN Trigraphs
###Nested namespace definition N4230
GCC: 6.0 | Clang: 3.6 | MSVC: 14.3 |
---|
Allows to write:
namespace A::B::C {
//…
}
Rather than:
namespace A {
namespace B {
namespace C {
//…
}
}
}
###Attributes for namespaces and enumerators N4266
GCC: 4.9 (namespaces)/ 6 (enums) | Clang: 3.4 | MSVC: 14.0 |
---|
Permits attributes on enumerators and namespaces. More details in N4196.
enum E {
foobar = 0,
foobat [[deprecated]] = foobar
};
E e = foobat; // Emits warning
namespace [[deprecated]] old_stuff{
void legacy();
}
old_stuff::legacy(); // Emits warning
###u8 character literals N4267
GCC: 6.0 | Clang: 3.6 | MSVC: 14.0 |
---|
UTF-8 character literal, e.g.
u8'a'
. Such literal has typechar
and the value equal to ISO 10646 code point value of c-char, provided that the code point value is representable with a single UTF-8 code unit. If c-char is not in Basic Latin or C0 Controls Unicode block, the program is ill-formed.
The compiler will report errors if character cannot fit inside u8
ASCII range.
Reference:
- cppreference.com/character literal
- SO: What is the point of the UTF-8 character literals proposed for C++17?
###Allow constant evaluation for all non-type template arguments N4268
GCC: 6.0 | Clang: 3.6 | MSVC: not yet |
---|
Remove syntactic restrictions for pointers, references, and pointers to members that appears as non-type template parameters:
For instance:
template<int *p> struct A {};
int n;
A<&n> a; // ok
constexpr int *p() { return &n; }
A<p()> b; // error before C++17
GCC: 6.0 | Clang: 3.6 | MSVC: not yet |
---|
More background here in P0036
Allows to write compact code with variadic templates without using explicit recursion.
Example:
template<typename... Args>
auto SumWithOne(Args... args){
return (1 + ... + args);
}
Articles:
- C++ Truths: Folding Monadic Functions
- Simon Brand: Exploding tuples with fold expressions
- Baptiste Wicht: C++17 Fold Expressions
- Fold Expressions - ModernesCpp.com
###Unary fold expressions and empty parameter packs P0036R0
GCC: 6.0 | Clang: 3.9 | MSVC: not yet |
---|
If the parameter pack is empty then the value of the fold is:
Operator | Value |
---|---|
&& | true |
|| | false |
, | void() |
For any operator not listed above, an unary fold expression with an empty parameter pack is ill-formed.
###Remove Deprecated Use of the register Keyword P0001R1
GCC: 7.0 | Clang: 3.8 | MSVC: not yet |
---|
The register
keyword was deprecated in the 2011 C++ standard. C++17 tries to clear the standard, so the keyword is now removed. This keyword is reserved now and might be repurposed in the future revisions.
###Remove Deprecated operator++(bool) P0002R1
GCC: 7.0 | Clang: 3.8 | MSVC: not yet |
---|
The ++ operator for bool
was deprecated in the original 1998 C++ standard, and it is past time to remove it formally.
###Removing Deprecated Exception Specifications from C++17 P0003R5
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Dynamic exception specifications were deprecated in C++11. This paper formally proposes removing the feature from C++17, while retaining the (still) deprecated throw()
specification strictly as an alias for noexcept(true)
.
###Make exception specifications part of the type system
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Previously exception specifications for a function didn't belong to the type of the function, but it will be part of it.
We'll get an error in the case:
void (*p)();
void (**pp)() noexcept = &p; // error: cannot convert to pointer to noexcept function
struct S { typedef void (*p)(); operator p(); };
void (*q)() noexcept = S(); // error: cannot convert to pointer to noexcept function
###Aggregate initialization of classes with base classes P0017R1
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
If a class was derived from some other type you couldn't use aggregate initialization. But now the restriction is removed.
struct base { int a1, a2; };
struct derived : base { int b1; };
derived d1{{1, 2}, 3}; // full explicit initialization
derived d1{{}, 1}; // the base is value initialized
To sum up: from the standard:
An aggregate is an array or a class with:
- no user-provided constructors (including those inherited from a base class),
- no private or protected non-static data members (Clause 11),
- no base classes (Clause 10) and // removed now!
- no virtual functions (10.3), and
- no virtual, private or protected base classes (10.1).
###Lambda capture of *this P0018R3
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
this
pointer is implicitly captured by lambdas inside member functions (if you use a default capture, like [&]
or [=]
). Member variables are always accessed by this pointer.
Example:
struct S {
int x ;
void f() {
// The following lambda captures are currently identical
auto a = [&]() { x = 42 ; } // OK: transformed to (*this).x
auto b = [=]() { x = 43 ; } // OK: transformed to (*this).x
a();
assert( x == 42 );
b();
assert( x == 43 );
}
};
Now you can use *this
when declaring a lambda, for example auto b = [=, *this]() { x = 43 ; }
. That way this
is captured by value. Note that the form [&,this] is redundant but accepted for compatibility with ISO C++14.
Capturing by value might be especially important for async invocation, paraller processing.
###Using attribute namespaces without repetition P0028R4
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
Other name for this feature was "Using non-standard attributes" in P0028R3 and PDF: P0028R2 (rationale, examples).
Simplifies the case where you want to use multiple attributes, like:
void f() {
[[rpr::kernel, rpr::target(cpu,gpu)]] // repetition
do-task();
}
Proposed change:
void f() {
[[using rpr: kernel, target(cpu,gpu)]]
do-task();
}
That simplification might help when building tools that automatically translate annotated such code into a different programming models.
###Dynamic memory allocation for over-aligned data P0035R4
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
In the following example:
class alignas(16) float4 {
float f[4];
};
float4 *p = new float4[1000];
C++11/14 did not specify any mechanism by which over-aligned data can be dynamically allocated correctly (i.e. respecting the alignment of the data). In the example above, not only is an implementation of C++ not required to allocate properly-aligned memory for the array, for practical purposes it is very nearly required to do the allocation incorrectly.
C++17 fixes that hole by introducing additional memory allocation functions that use align parameter:
void* operator new(std::size_t, std::align_val_t);
void* operator new[](std::size_t, std::align_val_t);
void operator delete(void*, std::align_val_t);
void operator delete[](void*, std::align_val_t);
void operator delete(void*, std::size_t, std::align_val_t);
void operator delete[](void*, std::size_t, std::align_val_t);
###__has_include in preprocessor conditionals P0061R1
GCC: 5.0 | Clang: yes | MSVC: not yet |
---|
This feature allows a C++ program to directly, reliably and portably determine whether or not a library header is available for inclusion.
Example: This demonstrates a way to use a library optional facility only if it is available.
#if __has_include(<optional>)
# include <optional>
# define have_optional 1
#elif __has_include(<experimental/optional>)
# include <experimental/optional>
# define have_optional 1
# define experimental_optional 1
#else
# define have_optional 0
#endif
###Template argument deduction for class templates P0091R3
GCC: 7.0 | Clang: not yet | MSVC: not yet |
---|
Before C++17, template deduction worked for functions but not for classes. For instance, the following code was legal:
void f(std::pair<int, char>);
f(std::make_pair(42, 'z'));
because std::make_pair
is a template function (so we can perform teplate deduction).
But the following wasn't:
void f(std::pair<int, char>);
f(std::pair(42, 'z'));
Although it is semantically equivalent. This was not legal because std::pair
is a template class, and template classes could not apply type deduction in their initialization.
So before C++17 one has to write out the types explicitly, even though this does not add any new information:
void f(std::pair<int, char>);
f(std::pair<int, char>(42, 'z'));
This is fixed in C++17 where template class constructors can deduce type parameters. The syntax for constructing such template classes is therefore consistent with the syntax for constructing non-template classes.
todo: deduction guides.
###Non-type template parameters with auto type P0127R2
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Automatically deduce type on non-type template parameters.
template <auto value> void f() { }
f<10>(); // deduces int
Trip report: Summer ISO C++ standards meeting (Oulu) | Sutter’s Mill
###Guaranteed copy elision P0135R1
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Articles:
###New specification for inheriting constructors (DR1941 et al) P0136R1
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
More description and reasoning in P0136R0. Some excerpts below:
An inheriting constructor does not act like any other form of using-declaration. All other using-declarations make some set of declarations visible to name lookup in another context, but an inheriting constructor declaration declares a new constructor that merely delegates to the original.
This feature changes inheriting constructor declaration from declaring a set of new constructors, to making a set of base class constructors visible in a derived class as if they were derived class constructors. (When such a constructor is used, the additional derived class subobjects will also be implicitly constructed as if by a defaulted default constructor). Put another way: make inheriting a constructor act just like inheriting any other base class member, to the extent possible.
This change does affect the meaning and validity of some programs, but these changes improve the consistency and comprehensibility of C++.
// Hiding works the same as for other member
// using-declarations in the presence of default arguments
struct A {
A(int a, int b = 0);
void f(int a, int b = 0);
};
struct B : A {
B(int a); using A::A;
void f(int a); using A::f;
};
struct C : A {
C(int a, int b = 0); using A::A;
void f(int a, int b = 0); using A::f;
};
B b(0); // was ok, now ambiguous
b.f(0); // ambiguous (unchanged)
C c(0); // was ambiguous, now ok
c.f(0); // ok (unchanged)
// Inheriting constructor parameters are no longer copied
struct A { A(const A&) = delete; A(int); };
struct B { B(A); void f(A); };
struct C : B { using B::B; using B::f; };
C c({0}); // was ill-formed, now ok (no copy made)
c.f({0}); // ok (unchanged)
###Direct-list-initialization of enumerations P0138R2
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
Allows to initialize enum class with a fixed underlying type:
enum class Handle : uint32_t { Invalid = 0 };
Handle h { 42 }; // OK
Allows to create 'strong types' that are easy to use...
###Stricter expression evaluation order P0145R3
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
In a nutshell, given an expression such as f(a, b, c)
, the order in which the sub-expressions f, a, b, c (which are of arbitrary shapes) are evaluated is left unspecified by the standard.
// unspecified behaviour below!
f(i++, i);
v[i] = i++;
std::map<int, int> m;
m[0] = m.size(); // {{0, 0}} or {{0, 1}} ?
Summary of changes:
- Postfix expressions are evaluated from left to right. This includes functions calls and member selection expressions.
- Assignment expressions are evaluated from right to left. This includes compound assignments.
- Operands to shift operators are evaluated from left to right.
Reference:
- C++ Order of evaluation, cppreference
- SO: What are the evaluation order guarantees introduced by C++17?
###constexpr lambda expressions P0170R1
GCC: 7.0 | Clang: not yet | MSVC: not yet |
---|
consexpr can be used in the context of lambdas.
constexpr auto ID = [] (int n) { return n; };
constexpr int I = ID(3);
static_assert(I == 3);
constexpr int AddEleven(int n) {
// Initialization of the 'data member' for n can
// occur within a constant expression since 'n' is
// of literal type.
return [n] { return n + 11; }();
}
static_assert(AddEleven(5) == 16);
Articles
- A 5 min episode of Jason Turner's C++ Weekly about constexpr lambdas
- Lambda expression comparison between C++11, C++14 and C++17
###Different begin and end types in range-based for P0184R0
GCC: 6.0 | Clang: 3.6 | MSVC: 15.0 Preview 5 |
---|
Changing the definition of range based for from:
{
auto && __range = for-range-initializer;
for ( auto __begin = begin-expr,
__end = end-expr;
__begin != __end;
++__begin ) {
for-range-declaration = *__begin;
statement
}
}
Into:
{
auto && __range = for-range-initializer;
auto __begin = begin-expr;
auto __end = end-expr;
for ( ; __begin != __end; ++__begin ) {
for-range-declaration = *__begin;
statement
}
}
Types of __begin
and __end
might be different; only the comparison operator is required. This little change allows Range TS users a better experience.
###[[fallthrough]] attribute
GCC: 7.0 | Clang: 3.9 | MSVC: 15.0 Preview 4 |
---|
Indicates that a fallthrough in a switch statement is intentional and a warning should not be issued for it. More details in P0068R0.
switch (c) {
case 'a':
f(); // Warning emitted, fallthrough is perhaps a programmer error
case 'b':
g();
[[fallthrough]]; // Warning suppressed, fallthrough is intentional
case 'c':
h();
}
###[[nodiscard]] attribute
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
[[nodiscard]]
is used to stress that the return value of a function is not to be discarded, on pain of a compiler warning. More details in P0068R0.
[[nodiscard]] int foo();
void bar() {
foo(); // Warning emitted, return value of a nodiscard function is discarded
}
This attribute can also be applied to types in order to mark all functions which return that type as [[nodiscard]]
:
[[nodiscard]] struct DoNotThrowMeAway{};
DoNotThrowMeAway i_promise();
void oops() {
i_promise(); // Warning emitted, return value of a nodiscard function is discarded
}
A 4 min video about [[nodiscard]] in Jason Turner's C++ Weekly
###[[maybe_unused]] attribute
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
Suppresses compiler warnings about unused entities when they are declared with [[maybe_unused]]
. More details in P0068R0.
static void impl1() { ... } // Compilers may warn about this
[[maybe_unused]] static void impl2() { ... } // Warning suppressed
void foo() {
int x = 42; // Compilers may warn about this
[[maybe_unused]] int y = 42; // Warning suppressed
}
A 3 min video about [[maybe_unused]] in Jason Turner's C++ Weekly
###Ignore unknown attributes
GCC: Yes | Clang: 3.9 | MSVC: not yet |
---|
Clarifies that implementations should ignore any attribute namespaces which they do not support, as this used to be unspecified. More details in P0283R1.
//compilers which don't support MyCompilerSpecificNamespace will ignore this attribute
[[MyCompilerSpecificNamespace::do_special_thing]]
void foo();
###Pack expansions in using-declarations
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Allows you to inject names with using-declarations from all types in a parameter pack.
In order to expose operator()
from all base classes in a variadic template, we used to have to resort to recursion:
template <typename T, typename... Ts>
struct Overloader : T, Overloader<Ts...> {
using T::operator();
using Overloader<Ts...>::operator();
// […]
};
template <typename T> struct Overloader<T> : T {
using T::operator();
};
Now we can simply expand the parameter pack in the using-declaration:
template <typename... Ts>
struct Overloader : Ts... {
using Ts::operator()...;
// […]
};
Remarks
- Implemented in GCC 7.0, see this change.
###Decomposition declarations
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
Helps when using tuples as a return type. It will automatically create variables and tie
them. More details in P0144R0. Was originally called "structured bindings".
For example:
std::tie(a, b, c) = tuple; // a, b, c need to be declared first
Now we can write:
auto [ a, b, c ] = tuple;
Such expressions also work on structs, pairs, and arrays.
Articles:
- Steve Lorimer, C++17 Structured Bindings
- jrb-programming, Emulating C++17 Structured Bindings in C++14
- Simon Brand, Adding C++17 decomposition declaration support to your classes
###Hexadecimal floating-point literals
GCC: 3.0 | Clang: Yes | MSVC: not yet |
---|
Allows to express some special floating point values, for example, the smallest normal IEEE-754 single precision value is readily written as 0x1.0p-126
.
###init-statements for if and switch
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
New versions of the if and switch statements for C++: if (init; condition)
and switch (init; condition)
.
This should simplify the code. For example, previously you had to write:
{
auto val = GetValue();
if (condition(val))
// on success
else
// on false...
}
Look, that val
has a separate scope, without it it will 'leak'.
Now you can write:
if (auto val = GetValue(); condition(val))
// on success
else
// on false...
val
is visible only inside the if
and else
statements, so it doesn't 'leak'.
condition
might be any condition, not only if val
is true/false.
Examples:
###Inline variables
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
Previously only methods/functions could be specified as inline
, now you can do the same with variables, inside a header file.
A variable declared inline has the same semantics as a function declared inline: it can be defined, identically, in multiple translation units, must be defined in every translation unit in which it is used, and the behavior of the program is as if there is exactly one variable.
struct MyClass
{
static const int sValue;
};
inline int const MyClass::sValue = 777;
Or even:
struct MyClass
{
inline static const int sValue = 777;
};
Articles
###DR: Matching of template template-arguments excludes compatible templates
GCC: 7.0 | Clang: 4.0 | MSVC: not yet |
---|
This feature resolves Core issue CWG 150.
From the paper:
This paper allows a template template-parameter to bind to a template argument whenever the template parameter is at least as specialized as the template argument. This implies that any template argument list that can legitimately be applied to the template template-parameter is also applicable to the argument template.
Example:
template <template <int> class> void FI();
template <template <auto> class> void FA();
template <auto> struct SA { /* ... */ };
template <int> struct SI { /* ... */ };
FI<SA>(); // OK; error before this paper
FA<SI>(); // error
template <template <typename> class> void FD();
template <typename, typename = int> struct SD { /* ... */ };
FD<SD>(); // OK; error before this paper (CWG 150)
(Adapted from the comment by IncongruentModulo1) For a useful example, consider something like this:
template <template <typename> typename Container>
struct A
{
Container<int> m_ints;
Container<double> m_doubles;
};
In C++14 and earlier, A<std::vector>
wouldn't be valid (ignoring the typename and not class before container) since std::vector
is declared as:
template <typename T, typename Allocator = std::allocator<T>> class vector;
This change resolves that issue. Before, you would need to declare template <template <typename...> typename Container>
, which is more permissive and moves the error to a less explicit line (namely the declaration of m_ints
wherever the struct A
is implemented / declared, instead of where the struct is instantiated with the wrong template type.
GCC: 6.0 | Clang: 3.7 | MSVC: 14.0 |
---|
More background in the original paper: PDF: N4152 and GOTW issue 47: Uncaught Exceptions.
The function returns the number of uncaught exception objects in the current thread.
This might be useful when implementing proper Scope Guards that works also during stack unwinding.
A type that wants to know whether its destructor is being run to unwind this object can query uncaught_exceptions in its constructor and store the result, then query uncaught_exceptions again in its destructor; if the result is different, then this destructor is being invoked as part of stack unwinding due to a new exception that was thrown later than the object’s construction
The above quote comes from PDF: N4152.
###constexpr
if-statements
GCC: 7.0 | Clang: 3.9 | MSVC: not yet |
---|
The static-if for C++! This allows you to discard branches of an if statement at compile-time based on a constant expression condition.
if constexpr(cond)
statement1; // Discarded if cond is false
else
statement2; // Discarded if cond is true
This removes a lot of the necessity for tag dispatching and SFINAE:
template <typename T, std::enable_if_t<std::is_arithmetic<T>{}>* = nullptr>
auto get_value(T t) {/*...*/}
template <typename T, std::enable_if_t<!std::is_arithmetic<T>{}>* = nullptr>
auto get_value(T t) {/*...*/}
template <typename T>
auto get_value(T t, std::true_type) {/*...*/}
template <typename T>
auto get_value(T t, std::false_type) {/*...*/}
template <typename T>
auto get_value(T t) {
return get_value(t, std::is_arithmetic<T>{});
}
template <typename T>
auto get_value(T t) {
if constexpr (std::is_arithmetic_v<T>) {
//...
}
else {
//...
}
}
Articles:
- LoopPerfect Blog, C++17 vs C++14 - Round 1 - if-constexpr
- SO: constexpr if and static_assert
- Simon Brand: Simplifying templates and #ifdefs with if constexpr
##Library Features
To get more details about library implementation I suggest those links:
- VS 2015 Update 2’s STL is C++17-so-far Feature Complete - Jan 2016
- libstdc++, C++ 201z status
- libc++ C++1z Status
This section only mentions some of the most important parts of library changes, it would be too impractical to go into details of every little change.
###Merged: The Library Fundamentals 1 TS (most parts)
We get the following items:
- Tuples - Calling a function with a tuple of arguments
- Functional Objects - Searchers
- Optional objects
- Class any
- string_view
- Memory:
- Shared-ownership pointers
- Class memory_resource
- Class memory_resource
- Access to program-wide memory_resource objects
- Pool resource classes
- Class monotonic_buffer_resource
- Alias templates using polymorphic memory resources
- Algorithms:
- Search
- Sampling
The wording from those components comes from Library Fundamentals V2 to ensure the wording includes the latest corrections.
Resources:
###Removal of some deprecated types and functions, including std::auto_ptr, std::random_shuffle, and old function adaptors
- Function objects - unary_function/binary_function, ptr_fun(), and mem_fun()/mem_fun_ref()
- Binders - bind1st()/bind2nd()
- auto_ptr
- Random shuffle - random_shuffle(first, last) and random_shuffle(first, last, rng)
###Merged: The Parallelism TS, a.k.a. “Parallel STL.”,
Articles:
###Merged: File System TS,
###Merged: The Mathematical Special Functions IS,
###Improving std::pair and std::tuple
###std::shared_mutex (untimed)
###Variant
Variant is a typesafe union that will report errors when you want to access something that's not currently inside the object.
###Splicing Maps and Sets
From Herb Sutter, Oulu trip report:
You will now be able to directly move internal nodes from one node-based container directly into another container of the same type. Why is that important? Because it guarantees no memory allocation overhead, no copying of keys or values, and even no exceptions if the container’s comparison function doesn’t throw.