C++17: The Language Features - Nicolai Josuttis

yeah my name is Nikolay Azotus I'm [Music] couple of city suspects years now doing super specialization and I learn and voluntario about c++ sleeper cell 17 so we have established c plus or 17 and i want to tell you what it's all about [Music] yes c purposes just one the next step of a couple of seasons of learnings we had in the past we started with C versus nineteen eight I was already involved there and then we had just the back six in 2003 and then where was the big step of modern C++ how we call it with C++ 11 and with some fixes and some enhancement but only small enhancement in C++ 14 and now C / 17 is around the corner which to some extent was announced as the next major version but it turned out it's not a next major version it's probably what I would call a medium version now so something between the innovation of C plus 14 and C++ 11 the status is that we count V so it's not done yet we are not done yet the status is that we look with a draft for national body feedback so that formerly some countries can say this should change and then otherwise I have a problem with this new standard and we want to fix all these commands we started to fix them already in in autumn and want to fix them again in February in Kona Hawaii so there we should finish the semester seventies done it if ssing cents goes fine unfortunately I can't be there so the SEO history it turned out that we now have this three-year schedule of C++ and that means that we are also think about trying out new things before we standardize them so one thing that happened in the past we have so-called TSS technical specifications for a couple of topics and as you can see here in dark green only a few of them went into C++ 17 most of them went not which to some extent surprised some people including the father of fitness of 17 who said we are too slow but on the other hand we are doing now standardization with about a hundred people in the world and we as a hundred people have to have some consensus about the next step and you see well yes it's like in politics sometimes you have to find ya it's sometimes tricky to have condenses even about the direction so there might be although although there are a couple of new things there might be some disappointment in the community about what's coming well I think it's good enough to update some books and that's good for me at least so let's see is there support already yes while the standard is not is not out yet we already have support for a couple of features most of the features I can even say the best support you have when you try a GCC or D plus plus there is the base is a version set o at work and this seven dot over them will have support for C plus 17 most if not all of it where they account be working on that it's still in development but the goal is to have support for C++ 70 in there and if you want to try it out just in case you you don't know there are a couple of online compilers available so if you want to try out C++ 11 feature a C++ 17 series also other features try for example rent box which is a compiler that overnight please gives you then the newest version of the GCC and also of the clang compiler there here for example it's a head of some 7.0 and I can try out here what compiles here of a couple of new features you might be able to read here or not that become too all these examples later so as I said most of the things I can test I could test already with that so I have some application programming experience with superstar 17 although it's not finalized and that's good we still find some issues especially when we bring together different features because in combination of some surprising behavior sometimes comes as we had in the past so what was not voted in C++ 17 the biggest things that are not there and a lot of people as I said are waiting for them that's modules concepts ranges co-routines reflection overloading operator dot default comparison operators uniform course in text or a library such as boost SEO which we for example now have in 80s and in the detail standard what but it will not make it into C++ 17 most of it for what the reason was we are not we are not convinced that this is mature enough for both for language features or library features if they didn't come in this might surprise to some people for example for concepts a lot of discussion is already there about the past 10 years but yes it's not done yet so I was that that was probably the most thing I was surprised about yeah for anything else there might be good reasons or not some of them I'm happy that they are not already there because I myself think they are not well done or surprising behavior so yeah that's the way it is unfortunately every opinion is different in the Senate committee why isn't it working now some interesting Windows program took my focus you so one small part before we jump into details and generally change that C++ a lesson should be on top of c11 instead of c99 so we still have to go to support C functionality from the C library but it will switch to c11 and with some minor fixes but in general we should support all the data from there so let's talk in this first talk about the language features and then after the break in the next talk I talked about the library features knowing that there sometimes it's a little bit difficult to know is it a library or language feature because we have language support in the library let's see how it works out so the first thing here is structured bindings structured bindings is something you see here which means you can declare an object that in fact declares multiple objects so it initializes a variable U and V by what is returned by function foo and you can use to initialize U and V by a structure by some class that has a gap support like couple or by written by arrays or return arrays here you see we have a structure internally in the stack life and integer and a double so you will be deduced to be an integer we will be deduced to be a double and they will get the corresponding values what is returned here from foo same here if the function returns a couple of character float and string a B and C will have the corresponding types character float and string and will have the corresponding values depending on what the value was which was returned from G and also I can direct the initializes with an array here an array of 47 and 11 so that means x and y integers initialized by these various let's just makes it some things more convenient you can still have the usual qualifiers you can say I want to have the reference I want to have the Const value of this but you have to use Auto if you have a reference of course that means that you have access to what was returned from the function is it was returned by reference and yeah so you can change it if there still is it's still valid to modify it some interesting consequences this asks for example this is a nice example a matter of string and double what we can write now is this we iterate over this map by saying well I I want to iterate over the element the elements are key value pairs so I want to have them by reference and that means instead of having the strange pair where we use first and second and nobody can understand that then unless they know about C++ I can read this perfectly if key is greater than 20 I double the value and I can use this loop now to iterate over the elements and for the same reason I can now use the auto f4 key and value and print key and baryons is definitely more readable code than using our first and second in my opinion he is doing and greater than 20 yeah yes there was a hidden secret here yeah no you're right I changed the slides but not completely yeah thank you so yes of course the first thing should be a so yeah with the usual route you have for autograph extending temporaries account autograph extending 10% usual woods apply next thing um we were so successful with the extension of the for loop to becoming a range bass follow that we thought oh that's a cool set to go so let's do something else and one thing we extend it now was if and switch so that we can now besides the condition in front we can have an initialization that means if you have something like this this probably the typical example you have an initialization of something which you want to check then and which you still need them because if you don't need it then you can do it in one expression but if you still need it then you have to do something like this you can now write it as this you can say initialization is status s is the result of check and then I check whether s is not equal success and I use this as with the scope that the assets only valid inside the scope of this expression that also helps for lock guards for example if lock guards you read lock guards to do something like I want to have a vector and if it's not empty I print front I now can write this as follows to saying I initialize the lock gut and then I do a check and then I do something with thee with a vector that's also possible and it's also possible for a switch statement so if I have an initialization and then switch over the result I can now do the initialization inside the switch statement and I'm in the scope of this switch statement okay yeah so the question is if I initialize it just a temporary will it survive yeah actually I'm not sure about the route that I didn't try that out there must be some correlation and I will think about it so I have no answer yet Thanks yeah but here's something also bringing two of these features together just to show you what happens if I knew if I use structured bindings and then the the initialization is so if I have something like I have on a insert into a collection of value of 42 the return value is the pair where the first element refers to the element if it was exerted and the second element tells you whether this was successful or not so you have to write this strange code saying well if it was not successful I can tell you we i we already have there another element and that's this who can read that so we can do it now better we can say if initialize with auto with sector finding a position and done from insert why we are not done or if we are not done then we initialize key and value from what position refers to and then we can use the key and that's probably better readable code than before good something different because I could one valid name of these variables are just an underscore yeah but some people love that I am getting a little bit nervous when I see where I ever file it like that but it's possible yes you can say Const Auto rev key comma underscore if you don't care you could but you could also use the underscore if you care but that's your problem then oh yeah yeah and any new features is as a new ability to obscure behavior by I don't know encryption mobile server argumented action for class templates we have now the ability that we don't necessarily have to specify template arguments they can be reduced by the constructor like template arguments can be deduced by parameters when I have function templates now we have that in classes so do you see here usually if I have a class B with two types and I say I want to have an object of this type I can say I have to say please instantiate this for two strings with or without the initialization now I can skip the types if it if the types can be deduced from the parameters like and function templates so I can write here D 2 is high work the types by the way are the first type is and three characters and the second is six characters and here we have it's an int and it's a double and by the way I have I think no I'm this is also not correct because this is using the argument by value this will decay so these two types will be charged as if they would the argument would be defined by reference then this would be the two types of T 1 and T 2 it's again it's the same rules as for function templates but now for class templates you don't have to specify the types if they can be deduced it's an of course an error if they can't be deduced so you can't write that and you can use it for simple things like this you can say complex now it does not need an int here as a complex number is initialized by 5 and 3 both are integers so this will deduce a complex of int it will not work according to class complex if the types differ because complex requires that both arguments are the same and as for function templates if the deduction gives the contradiction and there's no automatic type conversion then this is an error the rule is either you specify all or none you can't partially specify this class template deduction argument so here I specify all of them well there's a special behavior if I as a default type then you don't have specialists like that one but if I say something like I don't specialize anything but here I special I hire a past one and here I passed two and that's fine but if I here say I pass two arguments but I specialize only one this is an error so we can only specify all that have no default where you'll or none so anything else is an error that's by the way something to help you because here consider this if we could sparsely specify things a couple partially specified for in taking two ends then we would not take this as an error we would say yeah we would say this is a table by to end but yeah so this seems seemed not to be useful and therefore we said no we either all or nothing - that we can't do the following now we can't say I want to have a set of ends and my sorting criteria wrong is passed as argument because it's this has a specific type which is the second parameter I still have to specify the second parameter we we are thinking about placing something here like comma question mark or something like that so that we can say it's our intention that we have two to ten parameters but that we still want to have only the second one to be deduced but that's not in C + 417 that's something that might come later another interesting thing is that sometimes deduction doesn't work or gives you surprising behavior so one thing here is pair I no longer need for a pair to specify what is the type of the first and the second element of the pair I can do it that way however this will not compile and the reason is that it's not allowed to copy or initialize array parameters and this is an array so I still need make pair because this decays and make payable decay to a contra star here and so sometimes the convenient functions functions are still there another good example shared pointer you can say I want to skip to initialize a shared point of end by a new in I want to skip this in because I can deduce this from the constructor but this is an error and it will probably remain to be an error why is it an error because in class shared pointer we have a template parameter T that's this one but the constructor takes a different template parameter because the construction happens with another template parameter to have to support automatic type conversion and there is no rule to convert this Y into this T so therefore this will not work we can deduce this Y but we can't abuse this T now there is something interesting something interesting you can specify roots for that which is called so-called deduction guides you can specify how I want to use this mechanism to use types and I can say for example the following if I initialize a shared pointer by a y star and Y is a template parameter then please did use that chat pointer of Y is used so the Y is used st in shared pointer I can do that so that's a deduction guide and if that if you define that then this is fine and will work we are currently discussing whether we want to have that and it will probably not be the case in C++ 17 because then we can initialize such a shared pointer by a new integer array because internal areas are also in stuff and that will though may be causing more interesting problems in your programs so there will be some surprises in the future about this in general oh yeah here's the example in general we will have that these deduction guys are at work and we are still working on them and for example one if he is one example if you have a vector you initialize it by two iterators you want to deduce the element type from the iterators but you have to say that i want to then have a vector which uses the iterator tripes traits Vario type as element type so that's that will be that will probably be in the standard for all the containers and some of them also yeah and you can also do it yourself you can have something like if I have a struct F and it it doesn't have to be a template the deduction guy you can say if I have a s for a string literal or chakhan star please did use s to use as a as a template parameter to be string and then this will automatically happen if you then initialize s by a spring little it will be an S string that's possible you have to do it in the same scope where you define the type so if you want to extend behavior of the standard library you have to define these deduction guides in this namespace SCD which is officially not allowed but possible so you have to do it in the same context where the original data type is different good very attic templates we introduced in C++ 11 very attic template which means we have here parameter take a list of parameters and we can use them to do something with them one thing we can do now which we couldn't do before to say we wanna we want to have this recursion and not the point is we take arguments the first argument is handled differently it's printed out and the other remaining arguments are printed are passed to print again so that way one argument off of the other we print all the elements but we need an end criterion because somewhere the recursion has to end and the problem is we could write the end caterham here because we had no if at compile time saying we want to hear only use this call if the number of elements is 0 is not 0 of course so now we have that we have a compile time if statement the name of under which it was proposed both expert if it was changed into its context but the term is still the name is still context raised but the syntax is is cons expert and that means here there must be a compile time expression if this a compile time expression is true and only then this will be are used in the template to generate code otherwise this is ignored and you can have an adze part the same way we call that a discarded statement then here you okay let's cool by the way that will change a lot of template code and we have even more for these variadic templates we have now so called fold expressions instead of using all these recursions to process one argument after the other we can say let's combine all these arguments with one Operator for example by saying dot dot dot plus s and s is a list of arguments of different types that comes here any length we can say we want to sum up all the elements here so this will expand to s1 plus s2 plus s3 as one two three are the different elements of this parameter which is a very a deck template parameter so representing different values so this will expand - this is left binding and with the dot dot at the end it's right binding and we can have an initialization so you can do something like this I want a initializing have my see out and I want to print all the elements I pass here to see out I can write it that way unfortunately not easily with some white space in between so that's that's a problem here that we have no syntax for that yet you have to do some tricks with some I don't know factors also to do that okay another thing template auto of course Auto more and more Auto we have now we can say we have a template a non type tender so a template for an integer or longer forever where we are not defining inner hand which type it is so we can say we instantiate this s by 42 or by a character and then the type will depend on that I didn't have a slide just at the moment I present this of course now with class template argument deduction we could again now say let's deduce this from the constructor I have to try it out at least so but that's something yeah I could try out later it this is no magic to to allow something that was not allowed before so we can't still in instantiate this for double as we can't declare a template for a double n that's still not possible you can partly specialize it so you can say this is a general guideline but for integers I have special implementation or for pointers I have a special implementation and you can do some interesting street tricks you can for example say Auto dot or dot here then you have a value list a template parameter are lists of values of different types and if you want to have this list homogenious you say my first template parameter is auto so it's deduced from the argument and all the other arguments have to have the same type so it's deco type of this type you will find out what this will be used for in practice I'm pretty sure something else this is really something nice who of you doesn't have the problem saying I only need this C++ file this this compilation unit just to clear one global object because a global object according to the one definition rule or the hour rule has to be in one translation unit but not in everything so you can't declare it in your header file you need a CPP file for that that's gone you can say by placing inline here you have an inline static object and that means the standard not guarantees that this object is exists once in your program is visible in all the files that include this header file and doesn't need outside a definition so you can include this and include this here and both can access energy and they have it you have it once in your program and you can do it for a couple of places you can for example do something like this you have a class money tour sorry a single thing you have a class and you want to have this monitor once in your program just in your head as I write in line monitor I create my global money tour and that's enough all your translation units want to use this monitor includes the code the peasant matter how much translation units you have you will have this once and it will be shared by all the different translation units you are just cooking looking very straight and the other thing is that every context breh implies in line which was by the way the fact already before so that's nothing new but the funny thing is that we can now in C++ say this context / implies in life so this still makes sense but that's nothing new you could wrote write that before in C++ 11 yeah yeah that's nothing new but but now we we see this expression under a different rule saying this is an inline definition of the back string okay I'm pretty sure it has been considered in the context of modules but I have no clue about the outcome because I'm happy I understood all the features that made it into C++ 17 I still not take an expert in modules sorry so I can't answer that question another rule we have now aggregates that can derive from other aggregates I mean this has been possible before you could define a structure deriving form a structure for example a C structure that's nothing new but what is new is that this is still an aggregate so aggregate rules apply and that might be some special language rules apply and there's a specific new way to initialize this completely so you can now write something like this I declare a variable d1 of this type and I have 42 which shows the Shadley pass through the base class to initialize the base class and then comma 1 to initialize the derived class and if you don't want to initialize the base class you use empty armed curly braces this is now an error just initializing v3 with the empty curly braces it's now an error I expect this will still compile for some why because this might break existing programs but according to the language this is now a change and should make it easier to deal with yeah as you say some extensions to existing types to initialize the object because without that we always had the need to provide a corresponding constructor here that then passes the arguments to the base class and that's no longer necessary yeah then something else Sean and me were talking yesterday already privately on the on a trip with about we now have a change in the routes about optimisation of return values you know we have two things in C++ we have return values that have a name like this X and we return values that don't have a name and there's a rule that some optimization happens so that not necessarily a copy constructor is called here the return value optimization of the names return value optimization and in C++ 17 this optimization is no longer optional it's mandatory it's required and that has an interesting effect that there is no longer the need to have a copy or move constructor to support that so even if you are not allowed to copy or move your objects this is no valid code which will change from what we have right now so you can say something like this I have a class X I delete my copy constructor I delete my move constructor but I can still return a temporary object here of this type and it will compile and you can also assign it to another object yeah so this will change the some things I'm also sure you attribute C++ defines a couple of you attributes of formal annotations and these are all things I think that people who know lint and C know already so something like no discard that means we want to have no we want to have a warning if we throw away the return value of something and one good example is acing those people of you who know async async has a problem it starts an asynchronous task and as in Cronus yeah functionality but it blocks in the destructor for the end of that so if you don't use the return value this means this is a synchronous call of this function and that that's a contradiction in itself so we now say a thing should be s be marked at the compilers warned explicitly about if you if you don't use the return value of async and there might be other examples where we think about changing things but existing practice is an interesting issue here - whether we really want to enable it or not so some people thought about operator new saying no discard or something like that so but let's see whether this will work out this is a language feature where we are still not you we still don't know where to apply it in the library that's a general problem we have we are making car language changes until the end and the library people have to follow that so it's not like an ordinary project management yet to say let's let's stop developing the framework and then work on the application of the framework it's unfortunately a little bit different here so I have no clue what happens in the last week of C++ analyzation with the question where should be the apply notice card in the language in the library another thing is that you can intentionally say yes I assigned the return value to Y but I didn't use it so that's also an attribute you have now and fall through is for switch cases where you're intentionally fall through in another case and if you have even if you have between the two case some statements some small things needs that namespaces you don't have to spell them out that way you can now spell them out that way if you want to use them you can now say in static assertions that the message here is no longer mandatory you don't have to write a message you can just say this is the assertion that should be checked by the compiler and if it's a little of the compiler too has a good message about that the funny thing is that the compiler then has to care about internationalization so here we have to care and yeah so that's one thing in lambdas we can now capture star this this means if you have a lambda lambda can exist inside member functions so if you are if you have an object of Class C and inside your member function you calling a lambda then you have this and usually what happens if you pass this either by reference or by value in general or by passing the name you will pass a pointer and that means you refer to the original object now guess what your start is a new thread and in the new thread you are working on the original object without copying this object and to be able to be safe there you can say now star this I'm honest I want to have that when I call the lambda that it uses a copy of the object where I call this lambda for new on delete now support arm alignment in memory we had alignment introduced for four stack objects so we had a line of a line as keywords introduced but they were not used in allocated memory and memory allocation on the heat and that's that has some interesting consequences because it means that if you call if you define your operator new operator you might take into account the alignment of the objects I don't know the details already but there will be new operators new and deletes etc which have additional arguments for alignment of your data so that you can use this argument to fulfill when you when you allocate the memory on the heap that you fully fulfill the requirement that it should be aligned according to some value on on the heap then so this will also be there yeah the question is what happens we'll see with the existing implementations yeah so the default is that an operator should choose one of these two which means in backward compatibility you only have this one and then this should be used with the existing behavior so of course this feature has to be backward countable compatible otherwise we put have would be in big trouble yeah you can now you have no new preprocessor statement which asked for whether a header file exists or can be found according to the part of edify so you can write something like this if I find the header file optional then I do something if I find the experimental optional line then I do something else and otherwise I do something different this way I have in my program some way to find out whether a feature which is in some header files is supported or not and I don't have to use some other techniques we have and then we have a couple of minor things I think I don't have time to to go through a lot of them one interesting thing is expression evaluation order we are starting to define now for operators the order of evaluation we had it in the past only for boolean operators so that we said end on our operation shall be evaluates from left to right but anything else we said it's up to the compiler to decide whether they evaluate first the right operand or the left operand this is going to change intentionally I don't have a slide about that because I don't want to have that you write code that depends on that so and we still fight a little bit about it some compiler writers are not very happy about this thing so yeah so be in ramadi remember C++ 17 is not done yet so there might happen something in the last meeting and it's getting very word that this might not be the last meeting the next meeting for about six or seventeen okay maybe maybe I tell you two or three things a based on the time I have left one interesting thing is this one this is uncaught exceptions uncaught exceptions with an S at the end we have already under standard uncaught exception without an S at the end which returns a boolean value and it tells you whether you are in a face handling an exception and then you can react accordingly the problem is and you could use that code to say well oh when I when I'm handling an exception I should not commit my transaction I should roll it back something like that unfortunately it could happen that while you are handling an exception you start some code where it's perfectly valid to fulfill the transaction and that was not possible with the old interface which is now possible with the new interface where we can say I can ask for the number of uncaught exceptions I have here and can be a number of zero one two or more and I when I keep it in my constructor I can check it against that in the destructor and again and then I can decide whether I want to commit or rollback I give you an example here this is a little bit tricky to code so I have here I have this class C C's has to decide whether to commit or rollback the code I have here a class D and E Class E so what happens is the following I create a C objects so the C object nothing with D and E is involved remember now when I declared this object I was not in the context of handling an exception good then I create a D now when I create Adi Adi is yep that's the an ordinary constructor and then I now I throw an exception so what happens now is because we throw an exception we have to call the destructors in the other order so the destructor for D and here you see the interesting thing destructive because of funky called clean up the cleanup funky creates a C now what we have we are in the context of handing districts so and except but it's it's not we are not rolling back we are using why we handle this with this exception we have just created this object see so we have now in the constructor we see that this C was created why we are in one uncaught exception and we create now an object e and inside a try-catch class we create another C and another e and now we say throw an exception so why we are handling an exception we are get another exception here which is okay as long as we catch it and handle it so now now the interesting things happens now we say we call the destructor for e the destructor for e uses a C again so they create a C and now the C knows that when I get created I was in the context of two unhandled exceptions now the destructor finds out nothing happens between construction and destruction so I can commit whatever was at a store C so this was fine code now when I then call the destructor of the outer C so this was here now with this C I find I am a context of to uncork instructions but exceptions but when I close in the construction I had one uncaught exception so I'm now in a rollback mode for this C and that means I call rollback so that that's the way you can you can now program and do the right thing or to the better thing to decide whether this is due to a rollback over this it's just why we are handling exceptions just the destruction code destructor call okay we can skip the rest some other minor things here for example some relaxations that we say if I have an enumeration and I want to initialize this enumeration by a by a ordinary value whether it's in on class on ordinary enum usually that's not allowed but now we allow it if we use curly braces but only then so that means that it becomes easier to say something like I won't use enum to define my own integral type which has the same values like an unsigned char and I can initialize it by 42 well it's it's just a support to make this more convenient because in the past you had to use a static cast here we have four new utf-8 character literals we don't have utf-8 support we only have the literals and the literals are only valid when they are valid in the the normal as a normal character and then we also are still thinking about no except no except if discussed to become part of the type so that you say for example something like this I have here a P at a pointer to P to a pointer to a function that gets no argument and returns nothing but it's not signaled to be no except so it's a function pointer that might refer to a function that might throw an exception and in the past currently I can assign this pointer to a reference referring to a function not allowing to throw an exception and this should be fixed by by this type that by by making no except part of the type which does of course not mean you can overload no functions for no accept and not know except like you can't do it for different return types this is still under discussion because it turned out over the last two or three months a lot of people found a lot of interesting problems that fall out from this so this is also not clear whether it will come so what else we have a huge number of bug fixes I will show you one as an example we for example one one small well not bug fixes but making it consistent for example the deprecated attribute can now apply to enumeration in namespaces we removed some stuff so we we no longer have operator plus plus for bool and - - of course also not try grave paragraphs are not supported the keyboard register is no longer supported also it's still a reserved keyword so we also clean some things up in the language here's one example that is for example fixed as a back author let's see as an interesting behavior we have already if you write this look here well let's start here Auto X is inertia lies by the integer 42 everybody of you would read accessing in the good news is it is correct access it used to be an end but the interesting thing happens if we use the modern form of initialization the curly braces I put that that most of you would say oh axis and initialized by 42 but according to the standard X is an initializer list having one element named 42 the funny thing is compilers also already make it correct and if you have this Auto X 1 comma 2 what is this in your opinion and we clarified for example things like this it's interesting how we clarify that we say now a direct initialization with curly braces it's not an initializer this when you use auto here so this initialize an integer if you have a comma here that's an error now but if you use the copy initialization so the assignment operator here then this on the Left will be initialize honest initialize a list of one element of multiple elements so that I'm not sure whether this is not more surprising than before but because I'm personally I'm a friend of of reducing the difference between initialization without and with the equal sign but I'm not a car guy I'm just an application programmer but now you know the rules and by the way these rules are already implemented in the compilers because too many people were too confused by what is happening here things like this we fix which Mike to some extent even break existing programs but yeah so yeah let's that's progress so that's it for the moment that's only the language features that only the language features I had time for or I have understood enough to present them there's more definitely as I said after a break I will come to the library features and I think we have time next for three minutes or so for some questions I don't guarantee that I'm able to answer them yeah structured binding just about circle - yeah so the question is can I use cine to restrict which structure bindings are allowed or I see no reason that the swinish should not work here and the way I usually handle this is I don't guarantee I want to try it out at least in two compilers or find the corresponding wording but I see no reason not to allow it here because to some extent we don't we don't want to introduce you a really new feature so it should be something like Auto uke and Auto V are initialized here and all the other rules should apply as before so I would imagine that you can spin out things here are the template constructor destruction okay okay yeah so here's again the question could be used in a to disable this deduction I would also assume that yeah because the trick of a sign is make it not compile and whenever template code does not compile it's not the an error it's just ignored and I see no reason that this would not apply here you so say it again so we yeah we have yeah we that's right so we could use Finny instead of the deduction guides I mean to some extent deduction guides a convenient way to to 3d out everything here but it's not only that I mean yeah where are they I can't speak and move at the same time so here they are yeah so and I I would expect to some extent that both is possible spinning out and use deduction guide to do the right thing certainly the deduction guys as a more intuitive interface I mean I I hate funny I have to say I would really prefer to have Sweeney as a language feature instead of a template act using using template trick nobody can understand what happens there unless people I don't know have smart minds also so if we just would say something like requires yeah yeah I mean that's more or less that's more or less saying I want to have concepts yeah and now with concepts we could we could skip a lot of fini unfortunately we can't skip it now yeah yeah okay one last question so the question is can be used equals delete yeah I see no reason that it should not go up mine you can you can apply equal to delete everywhere yeah but I will try it out and later so just using even a deduction guide with equal to it oh no I'm not sure about that I don't think that specified yeah but yeah mmm probably not memory no I I have no clue you see so let's find out yeah so you are saying a call Bangkok's temporaries initialized in in the in the essence which destroyed I be careful with that so no because because I think all these which are in a very early stage so I don't think this is a proof for anything and even this might still be under consideration as a consequence of what we standardized here so I would not say that this is a proof that this is the case I personally would be a little bit surprised if the chain if the woods would this was there because in the S state mental I consider to be very dangerous to say that it doesn't extend the lifetime until the end of the if statement personally and it would be at least inconsistent to the existing rules we will we have to find it out okay that's it thank you very much if you want in 20 minutes or 15 minutes we talk about the library thank you

Channel: NDC Conferences

Views: 5,779

Keywords: c++, c++17, Nicolai Josuttis, ndc, ndc london

Id: pEzV32yRu4U

Length: 61min 50sec (3710 seconds)

Published: Tue Apr 18 2017