Suppose I have some object with an initial name such as quantities_of_widgets.
std::vector<int> quantities_of_widgets = GetQuantities();
I then perform an operation on quantities_of_widgets in-place. The vector quantities_of_widgets no longer represents "a vector of some quantities of things called widgets".
PerformOperationInPlace(quantities_of_widgets);
After this operation, a more appropriate name for the variable would be weights_of_widgets. That would certainly make the code more readable.
I can "change" the name by moving quantities_of_widgets to a new vector weights_of_widgets. This is a terrible idea. The move isn't free, it messes with memory continuity, etc.
std::vector<int> weights_of_widgets = std::move(quantities_of_widgets);
I could also document the meaning of the variable in comments. I might write
std::vector<int> widgets = GetQuantities(); // widgets represents quantities
PerformOperationInPlace(widgets); // widgets represents weights
UseWeights(widgets);
However, comments are less permanent and less readable. It becomes more difficult to keep track of the meaning of variables over multiple lines of code. If widgets is used 20 lines down in the middle of a large expression, I'd have trouble documenting the meaning of the variable at that point.
SomeBigFunction(Func1(widgets, some_other_object_1), some_other_object_2);
I could try aliasing, but I want to make the old name invalid to help with maintainability. Is there a way to change the name of a variable with no run-time cost?
Worrying about such small overhead is almost certainly not worth the effort in all but the most extreme scenarios. However, for the sake of completeness:
The only way to get truly overhead-free renaming of types with non-trivial destructors is via copy-elision.
In this context, you could make use of it via an immediately evaluated lambda and NRVO.
#include <vector>
std::vector<int> GetQuantities(int v);
void PerformOperationInPlace(std::vector<int>&);
void UseWeights(const std::vector<int>&);
template<typename T>
T rename(T&& rhs) {
return rhs;
}
void foo(int x) {
std::vector<int> quantities_of_widgets = GetQuantities(x);
PerformOperationInPlace(quantities_of_widgets);
UseWeights(quantities_of_widgets);
}
void bar(int x) {
std::vector<int> weights_of_widgets = [&]{
std::vector<int> quantities_of_widgets = GetQuantities(x);
PerformOperationInPlace(quantities_of_widgets);
return quantities_of_widgets;
}();
UseWeights(weights_of_widgets);
}
foo() and bar() compile down to effectively the exact same final assembly: (see on godbolt)
The only limitation is that the variable that will be renamed has to be declared within the lambda. Beyond that, implicit capture by reference makes the lambda effectively the same thing as a scope.
It is best to document code by itself. Comments have tendency do degrade.
So I would do this this way:
// name of this function terrible, it should be more like
// ExtractWeightsOfWidgetsFrom or something similar
std::vector<int> OperationInPlace(std::vector<int> data) {
// note argument is a copy!
....
return data;
}
void someCode() {
auto quantities_of_widgets = GetQuantities();
...
auto weights_of_widgets = PerformOperationInPlace(std::move(quantities_of_widgets));
// here `quantities_of_widgets` is empty since contents has been moved
// many tools will report an error if you use this value after that point
....
}
or even better I would just extract more functions:
void someCode() {
auto quantities_of_widgets = GetQuantities();
...
ProcesWeightsOfWidgets(
PerformOperationInPlace(std::move(quantities_of_widgets));
// in this version someCode ends here
}
Related
Suppose the following:
struct C {
... // lots of other stuff
int get(int key) const { return m.at(key); } // This will never throw
private:
std::unordered_map<int, int> m;
};
Due to how the application works, I know that get never throws. I want to make get as fast as possible. So, I would like to make the access unchecked, i.e. I would like to write something like return m[key]. Of course, I cannot write exactly that while keeping get const. However, I want to keep get const, since it is logically const.
Here is the only (ugly) solution I came up with:
struct C {
... // lots of other stuff
int get(int key) const { return const_cast<C *>(this)->m[key]; }
private:
std::unordered_map<int, int> m;
};
Is there a better way?
One approach would be to use std::unordered_map::find:
struct C {
... // lots of other stuff
int get(int key) const { return m.find(key)->second; }
private:
std::unordered_map<int, int> m;
};
I object to the very reasoning behind this question. The overhead (of map.at() vs map[]) associated with catching an error due to unknown key is presumably tiny compared to the cost of finding the key in the first place.
Yet, you willingly take the serious risk of a run-time error just for such a marginal efficiency advantage that you presumably have not even validated/measured. You may think that you know that key is always contained in the map, but perhaps future code changes (including bugs introduced by others) may change that?
If you really know, then you should use
map.find(key)->second;
which makes the bug explicit if the iterator returned is invalid (i.e. equal to map.end()). You may use assert in pre-production code, i.e.
auto it = map.find(key);
assert(it!=map.end());
return it->second;
which in production code (when assert is an empty macro) is removed.
It's my first year of using C++ and learning on the way. I'm currently reading up on Return Value Optimizations (I use C++11 btw). E.g. here https://en.wikipedia.org/wiki/Return_value_optimization, and immediately these beginner examples with primitive types spring to mind:
int& func1()
{
int i = 1;
return i;
}
//error, 'i' was declared with automatic storage (in practice on the stack(?))
//and is undefined by the time function returns
...and this one:
int func1()
{
int i = 1;
return i;
}
//perfectly fine, 'i' is copied... (to previous stack frame... right?)
Now, I get to this and try to understand it in the light of the other two:
Simpleclass func1()
{
return Simpleclass();
}
What actually happens here? I know most compilers will optimise this, what I am asking is not 'if' but:
how the optimisation works (the accepted response)
does it interfere with storage duration: stack/heap (Old: Is it basically random whether I've copied from stack or created on heap and moved (passed the reference)? Does it depend on created object size?)
is it not better to use, say, explicit std::move?
You won't see any effect of RVO when returning ints.
However, when returning large objects like this:
struct Huge { ... };
Huge makeHuge() {
Huge h { x, y, x };
h.doSomething();
return h;
}
The following code...
auto h = makeHuge();
... after RVO would be implemented something like this (pseudo code) ...
h_storage = allocate_from_stack(sizeof(Huge));
makeHuge(addressof(h_storage));
auto& h = *properly_aligned(h_storage);
... and makeHuge would compile to something like this...
void makeHuge(Huge* h_storage) // in fact this address can be
// inferred from the stack pointer
// (or just 'known' when inlining).
{
phuge = operator (h_storage) new Huge(x, y, z);
phuge->doSomething();
}
Given the following:
class ReadWrite {
public:
int Read(size_t address);
void Write(size_t address, int val);
private:
std::map<size_t, int> db;
}
In read function when accessing an address which no previous write was made to I want to either throw exception designating such error or allow that and return 0, in other words I would like to either use std::map<size_t, int>::operator[]() or std::map<size_t, int>::at(), depending on some bool value which user can set. So I add the following:
class ReadWrite {
public:
int Read(size_t add) { if (allow) return db[add]; return db.at(add);}
void Write(size_t add, int val) { db[add] = val; }
void Allow() { allow = true; }
private:
bool allow = false;
std::map<size_t, int> db;
}
The problem with that is:
Usually, the program will have one call of allow or none at the beginning of the program and then afterwards many accesses. So, performance wise, this code is bad because it every-time performs the check if (allow) where usually it's either always true or always false.
So how would you solve such problem?
Edit:
While the described use case (one or none Allow() at first) of this class is very likely it's not definite and so I must allow user call Allow() dynamically.
Another Edit:
Solutions which use function pointer: What about the performance overhead incurred by using function pointer which is not able to make inline by the compiler? If we use std::function instead will that solve the issue?
Usually, the program will have one call of allow or none at the
beginning of the program and then afterwards many accesses. So,
performance wise, this code is bad because it every-time performs the
check if (allow) where usually it's either always true or always
false. So how would you solve such problem?
I won't, The CPU will.
the Branch Prediction will figure out that the answer is most likely to be same for some long time so it will able to optimize the branch in the hardware level very much. it will still incur some overhead, but very negligible.
If you really need to optimize your program, I think your better use std::unordered_map instead of std::map, or move to some faster map implementation, like google::dense_hash_map. the branch is insignificant compared to map-lookup.
If you want to decrease the time-cost, you have to increase the memory-cost. Accepting that, you can do this with a function pointer. Below is my answer:
class ReadWrite {
public:
void Write(size_t add, int val) { db[add] = val; }
// when allowed, make the function pointer point to read2
void Allow() { Read = &ReadWrite::read2;}
//function pointer that points to read1 by default
int (ReadWrite::*Read)(size_t) = &ReadWrite::read1;
private:
int read1(size_t add){return db.at(add);}
int read2(size_t add) {return db[add];}
std::map<size_t, int> db;
};
The function pointer can be called as the other member functions. As an example:
ReadWrite rwObject;
//some code here
//...
rwObject.Read(5); //use of function pointer
//
Note that non-static data member initialization is available with c++11, so the int (ReadWrite::*Read)(size_t) = &ReadWrite::read1; may not compile with older versions. In that case, you have to explicitly declare one constructor, where the initialization of the function pointer can be done.
You can use a pointer to function.
class ReadWrite {
public:
void Write(size_t add, int val) { db[add] = val; }
int Read(size_t add) { (this->*Rfunc)(add); }
void Allow() { Rfunc = &ReadWrite::Read2; }
private:
std::map<size_t, int> db;
int Read1(size_t add) { return db.at(add); }
int Read2(size_t add) { return db[add]; }
int (ReadWrite::*Rfunc)(size_t) = &ReadWrite::Read1;
}
If you want runtime dynamic behaviour you'll have to pay for it at runtime (at the point you want your logic to behave dynamically).
You want different behaviour at the point where you call Read depending on a runtime condition and you'll have to check that condition.
No matter whether your overhad is a function pointer call or a branch, you'll find a jump or call to different places in your program depending on allow at the point Read is called by the client code.
Note: Profile and fix real bottlenecks - not suspected ones. (You'll learn more if you profile by either having your suspicion confirmed or by finding out why your assumption about the performance was wrong.)
(I know) In c++ I can declare variable out of scope and I can't run any code/statement, except for initializing global/static variables.
IDEA
Is it a good idea to use below tricky code in order to (for example) do some std::map manipulation ?
Here I use void *fakeVar and initialize it through Fake::initializer() and do whatever I want in it !
std::map<std::string, int> myMap;
class Fake
{
public:
static void* initializer()
{
myMap["test"]=222;
// Do whatever with your global Variables
return NULL;
}
};
// myMap["Error"] = 111; => Error
// Fake::initializer(); => Error
void *fakeVar = Fake::initializer(); //=> OK
void main()
{
std::cout<<"Map size: " << myMap.size() << std::endl; // Show myMap has initialized correctly :)
}
One way of solving it is to have a class with a constructor that does things, then declare a dummy variable of that class. Like
struct Initializer
{
Initializer()
{
// Do pre-main initialization here
}
};
Initializer initializer;
You can of course have multiple such classes doing miscellaneous initialization. The order in each translation unit is specified to be top-down, but the order between translation units is not specified.
You don't need a fake class... you can initialize using a lambda
auto myMap = []{
std::map<int, string> m;
m["test"] = 222;
return m;
}();
Or, if it's just plain data, initialize the map:
std::map<std::string, int> myMap { { "test", 222 } };
Is it a good idea to use below tricky code in order to (for example)
do some std::map manipulation ?
No.
Any solution entailing mutable non-local variables is a terrible idea.
Is it a good idea...?
Not really. What if someone decides that in their "tricky initialisation" they want to use your map, but on some system or other, or for not obvious reason after a particular relink, your map ends up being initialised after their attempted use? If you instead have them call a static function that returns a reference to the map, then it can initialise it on first call. Make the map a static local variable inside that function and you stop any accidental use without this protection.
§ 8.5.2 states
Except for objects declared with the constexpr specifier, for which
see 7.1.5, an initializer in the definition of a variable can consist
of arbitrary expressions involving literals and previously declared
variables and functions, regardless of the variable’s storage duration
therefore what you're doing is perfectly allowed by the C++ standard. That said, if you need to perform "initialization operations" it might be better to just use a class constructor (e.g. a wrapper).
What you've done is perfectly legal C++. So, if it works for you and is maintainable and understandable by anybody else who works with the code, it's fine. Joachim Pileborg's sample is clearer to me though.
One problem with initializing global variables like this can occur if they use each other during initialization. In that case it can be tricky to ensure that variables are initialized in the correct order. For that reason, I prefer to create InitializeX, InitializeY, etc functions, and explicitly call them in the correct order from the Main function.
Wrong ordering can also cause problems during program exit where globals still try to use each other when some of them may have been destroyed. Again, some explicit destruction calls in the correct order before Main returns can make it clearer.
So, go for it if it works for you, but be aware of the pitfalls. The same advice applies to pretty much every feature in C++!
You said in your question that you yourself think the code is 'tricky'. There is no need to overcomplicate things for the sake of it. So, if you have an alternative that appears less 'tricky' to you... that might be better.
When I hear "tricky code", I immediately think of code smells and maintenance nightmares. To answer your question, no, it isn't a good idea. While it is valid C++ code, it is bad practice. There are other, much more explicit and meaningful alternatives to this problem. To elaborate, the fact that your initializer() method returns void* NULL is meaningless as far as the intention of your program goes (i.e. each line of your code should have meaningful purpose), and you now have yet another unnecessary global variable fakeVar, which needlessly points to NULL.
Let's consider some less "tricky" alternatives:
If it's extremely important that you only ever have one global instance of myMap, perhaps using the Singleton Pattern would be more fitting, and you would be able to lazily initialize the contents of myMap when they are needed. Keep in mind that the Singleton Pattern has issues of its own.
Have a static method create and return the map or use a global namespace. For example, something along the lines of this:
// global.h
namespace Global
{
extern std::map<std::string, int> myMap;
};
// global.cpp
namespace Global
{
std::map<std::string, int> initMap()
{
std::map<std::string, int> map;
map["test"] = 222;
return map;
}
std::map<std::string, int> myMap = initMap();
};
// main.cpp
#include "global.h"
int main()
{
std::cout << Global::myMap.size() << std::endl;
return 0;
}
If this is a map with specialized functionality, create your own class (best option)! While this isn't a complete example, you get the idea:
class MyMap
{
private:
std::map<std::string, int> map;
public:
MyMap()
{
map["test"] = 222;
}
void put(std::string key, int value)
{
map[key] = value;
}
unsigned int size() const
{
return map.size();
}
// Overload operator[] and create any other methods you need
// ...
};
MyMap myMap;
int main()
{
std::cout << myMap.size() << std::endl;
return 0;
}
In C++, you cannot have statements outside any function. However, you have global objects declared, and constructor (initializer) call for these global objects are automatic before main starts. In your example, fakeVar is a global pointer that gets initialized through a function of class static scope, this is absolutely fine.
Even a global object would do provide that global object constructor does the desired initializaton.
For example,
class Fake
{
public:
Fake() {
myMap["test"]=222;
// Do whatever with your global Variables
}
};
Fake fake;
This is a case where unity builds (single translation unit builds) can be very powerful. The __COUNTER__ macro is a de facto standard among C and C++ compilers, and with it you can write arbitrary imperative code at global scope:
// At the beginning of the file...
template <uint64_t N> void global_function() { global_function<N - 1>(); } // This default-case skips "gaps" in the specializations, in case __COUNTER__ is used for some other purpose.
template <> void global_function<__COUNTER__>() {} // This is the base case.
void run_global_functions();
#define global_n(N, ...) \
template <> void global_function<N>() { \
global_function<N - 1>(); /* Recurse and call the previous specialization */ \
__VA_ARGS__; /* Run the user code. */ \
}
#define global(...) global_n(__COUNTER__, __VA_ARGS__)
// ...
std::map<std::string, int> myMap;
global({
myMap["test"]=222;
// Do whatever with your global variables
})
global(myMap["Error"] = 111);
int main() {
run_global_functions();
std::cout << "Map size: " << myMap.size() << std::endl; // Show myMap has initialized correctly :)
}
global(std::cout << "This will be the last global code run before main!");
// ...At the end of the file
void run_global_functions() {
global_function<__COUNTER__ - 1>();
}
This is especially powerful once you realize that you can use it to initialize static variables without a dependency on the C runtime. This means you can generate very small executables without having to eschew non-zero global variables:
// At the beginning of the file...
extern bool has_static_init;
#define default_construct(x) x{}; global(if (!has_static_init()) new (&x) decltype(x){})
// Or if you don't want placement new:
// #define default_construct(x) x{}; global(if (!has_static_init()) x = decltype(x){})
class Complicated {
int x = 42;
Complicated() { std::cout << "Constructor!"; }
}
Complicated default_construct(my_complicated_instance); // Will be zero-initialized if the CRT is not linked into the program.
int main() {
run_global_functions();
}
// ...At the end of the file
static bool get_static_init() {
volatile bool result = true; // This function can't be inlined, so the CRT *must* run it.
return result;
}
has_static_init = get_static_init(); // Will stay zero without CRT
This answer is similar to Some programmer dude's answer, but may be considered a bit cleaner. As of C++17 (that's when std::invoke() was added), you could do something like this:
#include <functional>
auto initializer = std::invoke([]() {
// Do initialization here...
// The following return statement is arbitrary. Without something like it,
// the auto will resolve to void, which will not compile:
return true;
});
With the code below, the question is:
If you use the "returnIntVector()" function, is the vector copied from the local to the "outer" (global) scope? In other words is it a more time and memory consuming variation compared to the "getIntVector()"-function? (However providing the same functionality.)
#include <iostream>
#include <vector>
using namespace std;
vector<int> returnIntVector()
{
vector<int> vecInts(10);
for(unsigned int ui = 0; ui < vecInts.size(); ui++)
vecInts[ui] = ui;
return vecInts;
}
void getIntVector(vector<int> &vecInts)
{
for(unsigned int ui = 0; ui < vecInts.size(); ui++)
vecInts[ui] = ui;
}
int main()
{
vector<int> vecInts = returnIntVector();
for(unsigned int ui = 0; ui < vecInts.size(); ui++)
cout << vecInts[ui] << endl;
cout << endl;
vector<int> vecInts2(10);
getIntVector(vecInts2);
for(unsigned int ui = 0; ui < vecInts2.size(); ui++)
cout << vecInts2[ui] << endl;
return 0;
}
In theory, yes it's copied. In reality, no, most modern compilers take advantage of return value optimization.
So you can write code that acts semantically correct. If you want a function that modifies or inspects a value, you take it in by reference. Your code does not do that, it creates a new value not dependent upon anything else, so return by value.
Use the first form: the one which returns vector. And a good compiler will most likely optimize it. The optimization is popularly known as Return value optimization, or RVO in short.
Others have already pointed out that with a decent (not great, merely decent) compiler, the two will normally end up producing identical code, so the two give equivalent performance.
I think it's worth mentioning one or two other points though. First, returning the object does officially copy the object; even if the compiler optimizes the code so that copy never takes place, it still won't (or at least shouldn't) work if the copy ctor for that class isn't accessible. std::vector certainly supports copying, but it's entirely possible to create a class that you'd be able to modify like in getIntVector, but not return like in returnIntVector.
Second, and substantially more importantly, I'd generally advise against using either of these. Instead of passing or returning a (reference to) a vector, you should normally work with an iterator (or two). In this case, you have a couple of perfectly reasonable choices -- you could use either a special iterator, or create a small algorithm. The iterator version would look something like this:
#ifndef GEN_SEQ_INCLUDED_
#define GEN_SEQ_INCLUDED_
#include <iterator>
template <class T>
class sequence : public std::iterator<std::forward_iterator_tag, T>
{
T val;
public:
sequence(T init) : val(init) {}
T operator *() { return val; }
sequence &operator++() { ++val; return *this; }
bool operator!=(sequence const &other) { return val != other.val; }
};
template <class T>
sequence<T> gen_seq(T const &val) {
return sequence<T>(val);
}
#endif
You'd use this something like this:
#include "gen_seq"
std::vector<int> vecInts(gen_seq(0), gen_seq(10));
Although it's open to argument that this (sort of) abuses the concept of iterators a bit, I still find it preferable on practical grounds -- it lets you create an initialized vector instead of creating an empty vector and then filling it later.
The algorithm alternative would look something like this:
template <class T, class OutIt>
class fill_seq_n(OutIt result, T num, T start = 0) {
for (T i = start; i != num-start; ++i) {
*result = i;
++result;
}
}
...and you'd use it something like this:
std::vector<int> vecInts;
fill_seq_n(std::back_inserter(vecInts), 10);
You can also use a function object with std::generate_n, but at least IMO, this generally ends up more trouble than it's worth.
As long as we're talking about things like that, I'd also replace this:
for(unsigned int ui = 0; ui < vecInts2.size(); ui++)
cout << vecInts2[ui] << endl;
...with something like this:
std::copy(vecInts2.begin(), vecInts2.end(),
std::ostream_iterator<int>(std::cout, "\n"));
In C++03 days, getIntVector() is recommended for most cases. In case of returnIntVector(), it might create some unncessary temporaries.
But by using return value optimization and swaptimization, most of them can be avoided. In era of C++11, the latter can be meaningful due to the move semantics.
In theory, the returnIntVector function returns the vector by value, so a copy will be made and it will be more time-consuming than the function which just populates an existing vector. More memory will also be used to store the copy, but only temporarily; since vecInts is locally scoped it will be stack-allocated and will be freed as soon as the returnIntVector returns. However, as others have pointed out, a modern compiler will optimize away these inefficiencies.
returnIntVector is more time consuming because it returns a copy of the vector, unless the vector implementation is realized with a single pointer in which case the performance is the same.
in general you should not rely on the implementation and use getIntVector instead.