Modern C++ auto
-
Rick York wrote:
With auto the compiler knows exactly the appropriate type to use and gives you an error if it can't figure it out.
If it "knows exactly", how can it not "figure it out"? Sounds like a chick/egg scenario to me.
".45 ACP - because shooting twice is just silly" - JSOP, 2010
-----
You can never have too much ammo - unless you're swimming, or on fire. - JSOP, 2010
-----
When you pry the gun from my cold dead hands, be careful - the barrel will be very hot. - JSOP, 2013If it knows exactly what the type should be, it'll substitute the appropriate type. If it doesn't know exactly what the type should be, it knows exactly that it doesn't know what the type should be. How's that a chicken/egg-scenario? Either the compiler knows the type or it doesn't. That's the beginning of the causality chain. If the compiler doesn't know the type, it throws you an error. That's the end of the causality chain.
-
At first glance
vardoes seem lazy, I use it regularly while working on a large codebase with a lot of 'technical debt'. I use it quite a lot in my professional code development having been encouraged to do so. There again I work in an environment where comments are frowned upon, the thinking being that well written code should not have to be documented - a philosophy which I don't agree with. I think the use ofvarfits in with this 'no comments' philosophy as it is not explicitly stated what the type of the variable is and you have to figure it out with intellisense or by inspecting the method's return type.“That which can be asserted without evidence, can be dismissed without evidence.”
― Christopher Hitchens
If your code has a lot of 'technical debt' it is probably not well written code, and thus arguing if well written code should be documented commented or not is irrelevant for that particular case. --- I agree with a lot of the philosophy you don't like... I have been anti-comment and pro
varfor a long time. I believe comments should not say what (names are for that) or how (instructions are for that)... yet, I think comments that explain why and for what are good. At the end the motivation for having less comments is that comments are not checked, and could be forgotten in refactoring, and thus there is a risk that they will be outdated... sure, we can argue dicipline, yet we use strictly typed languages for a reason. Thus, instead we want to express what we would have said in comments in code. With that siad, I can tell you that usingvaras an extension of a no-comments philosophy is retrograde. The idea is to make the code express as much as it can (so it is explicit, that is what they mean by well written code, please do not confuse with verbose), so that we do not have things to communicate in comments... from that point of view,varis counterproductive. Let us be clear,varis not dynamic typing. Yes, names can help with knowing the type※... yet, no, I am not advocating for hungarian notation either. So how can I be anti-comment and provarif they are at odds? I belive in the use ofvaras a way to protect the code from reasons to change. Same goes forauto. And yeah, I use it virtually everywhere. It eases refactoring (If I change the return type, usingautoavoids a maintenance ripple of updating types everywhere the code is used), thus increases maintainability. Addendum: You know what, I do realize it goes both ways, because if I did a poor job and returned something bad, auto will not complain. Although, I would expect it to break where we actually try to use the value. If the code follows the robustness principle ("Be conservative in what you send, be liberal in what you accept"), we will not be usingautofor the return type, instead the return type should be as specific as posible (without breaking encapsulation, if any). On the other hand, we want to assign the return value to a variable, and the return type is probably much more specific than you actually need in client code. In that situation we pro -
If it knows exactly what the type should be, it'll substitute the appropriate type. If it doesn't know exactly what the type should be, it knows exactly that it doesn't know what the type should be. How's that a chicken/egg-scenario? Either the compiler knows the type or it doesn't. That's the beginning of the causality chain. If the compiler doesn't know the type, it throws you an error. That's the end of the causality chain.
The compiler knows. The developer doesn't (necessarily).
-
The compiler knows. The developer doesn't (necessarily).
That's pretty much the point. In most cases, namely when the type doesn't matter as long as it works, the developer doesn't need to know. In edge cases, such as auto i=1 where i is required to be, let's say, an unsigned value somewhere later down the line, the developer can still forego the auto and make it a unsigned int i=1 or at least an auto i=static_cast1. An example from my own work: I've been using API functions like GetTickCount quite a lot and instead of looking up the exact return type, a simple auto s=GetTickCount does the job. API functions returning some value are documented as "Returns 0 if the operation succeeded", in that case, a if 0==s is still enough, I don't need to know the type.
-
Is this just me? I am starting to see the "auto" keyword abuse growing to an extraordinary proportions. Reminds me of the "var" type in JavaScript or "void*" in C/C++. The program, where all variables are declared as void*, would be considered atrocious, yet auto seems to be littered like an empty beer cans nowadays.
The use of auto is not laziness, nor is it abusive. It is correct and idiomatic modern C++. Bjarne Stroustrup and most members of the ISO C++ standardization committee actively advocate for its use, to the point where AAA - Almost Always Auto - has become a common mantra. The simple fact is that, most of the time, the compiler is smarter than you, and understands your code on a level that you never could. Allowing the compiler to determine the type automatically, as often as possible, allows for optimizations that may not be possible if you coerce an explicit type. People who reject evolutionary features of C++ are the same sort of people who would reject fuel injection on cars, because they learned how to drive a car with a carburetor, so everyone else should be fine with it. Technology advances. Try to keep up, or be left behind.
-
Is this just me? I am starting to see the "auto" keyword abuse growing to an extraordinary proportions. Reminds me of the "var" type in JavaScript or "void*" in C/C++. The program, where all variables are declared as void*, would be considered atrocious, yet auto seems to be littered like an empty beer cans nowadays.
I agree it can be easily abused. It is especially true if you are maintaining someone else's code. Yes, it's great for templated iterators and such, but it can also throw more work onto someone else down the road, which I consider rude or lazy. Say I need to add some functionality. I see
auto foo = SomeFunctionReturningObjOrRef(bar);
foo.SomeMethod(blah);Great. That's really easy for the original dev, and really easy for the compiler. Wonderful. Now I need to add some code. What the heck is a "foo"? What members does it have? What methods are available? Is it a base class, or a derived class that has the functions I need? The dev who wrote the code knew, but didn't bother to declare it. The compiler knows, and I suppose I can just try
foo.SomeMethodIHopeTheObjectHas();
And see if I get a compilation error. And hope that it really is the right type and not a base class or derived class of the one I expect... But practically speaking I have to do the work the original dev didn't do and look up
SomeFunctionReturningObjOrRef()
and see what it does, and what it returns so I can be sure I'm getting the right type or that it supports the methods I need, and I don't actually want to be using "bar" or something else or upcast or downcast etc.
-
That's pretty much the point. In most cases, namely when the type doesn't matter as long as it works, the developer doesn't need to know. In edge cases, such as auto i=1 where i is required to be, let's say, an unsigned value somewhere later down the line, the developer can still forego the auto and make it a unsigned int i=1 or at least an auto i=static_cast1. An example from my own work: I've been using API functions like GetTickCount quite a lot and instead of looking up the exact return type, a simple auto s=GetTickCount does the job. API functions returning some value are documented as "Returns 0 if the operation succeeded", in that case, a if 0==s is still enough, I don't need to know the type.
Even "auto i=1" can be made explicit with "auto i=1U".
-
I agree it can be easily abused. It is especially true if you are maintaining someone else's code. Yes, it's great for templated iterators and such, but it can also throw more work onto someone else down the road, which I consider rude or lazy. Say I need to add some functionality. I see
auto foo = SomeFunctionReturningObjOrRef(bar);
foo.SomeMethod(blah);Great. That's really easy for the original dev, and really easy for the compiler. Wonderful. Now I need to add some code. What the heck is a "foo"? What members does it have? What methods are available? Is it a base class, or a derived class that has the functions I need? The dev who wrote the code knew, but didn't bother to declare it. The compiler knows, and I suppose I can just try
foo.SomeMethodIHopeTheObjectHas();
And see if I get a compilation error. And hope that it really is the right type and not a base class or derived class of the one I expect... But practically speaking I have to do the work the original dev didn't do and look up
SomeFunctionReturningObjOrRef()
and see what it does, and what it returns so I can be sure I'm getting the right type or that it supports the methods I need, and I don't actually want to be using "bar" or something else or upcast or downcast etc.
I usually do "all that work" by hovering the mouse over the function call and seeing what the IDE shows as the prototype of the function. There is no guesswork involved.
"They have a consciousness, they have a life, they have a soul! Damn you! Let the rabbits wear glasses! Save our brothers! Can I get an amen?"
-
The use of auto is not laziness, nor is it abusive. It is correct and idiomatic modern C++. Bjarne Stroustrup and most members of the ISO C++ standardization committee actively advocate for its use, to the point where AAA - Almost Always Auto - has become a common mantra. The simple fact is that, most of the time, the compiler is smarter than you, and understands your code on a level that you never could. Allowing the compiler to determine the type automatically, as often as possible, allows for optimizations that may not be possible if you coerce an explicit type. People who reject evolutionary features of C++ are the same sort of people who would reject fuel injection on cars, because they learned how to drive a car with a carburetor, so everyone else should be fine with it. Technology advances. Try to keep up, or be left behind.
Andy Hoffmeyer wrote:
The simple fact is that, most of the time, the compiler is smarter than you, and understands your code on a level that you never could. Allowing the compiler to determine the type automatically, as often as possible, allows for optimizations that may not be possible if you coerce an explicit type.
So, by that logic if I was using an IDE with good intellisense and hovered over a variable declared initially with auto, which showed me what the omniscient compiler will decide the type should be, and then explicitly declared the variable to be that exact type it would somehow break the multi-dimensional optimization the compiler would perform. Are you seriously saying that or did I misread your comment??
-
The use of auto is not laziness, nor is it abusive. It is correct and idiomatic modern C++. Bjarne Stroustrup and most members of the ISO C++ standardization committee actively advocate for its use, to the point where AAA - Almost Always Auto - has become a common mantra. The simple fact is that, most of the time, the compiler is smarter than you, and understands your code on a level that you never could. Allowing the compiler to determine the type automatically, as often as possible, allows for optimizations that may not be possible if you coerce an explicit type. People who reject evolutionary features of C++ are the same sort of people who would reject fuel injection on cars, because they learned how to drive a car with a carburetor, so everyone else should be fine with it. Technology advances. Try to keep up, or be left behind.
The community is very much divided over AAA, so it's hardly a no brainer and being against it hardly makes one a Luddite. Of course the people who design the language and compilers aren't the people who are maintaining my code, so it's pretty easy for them to make such an assertion. And the compiler doesn't understand my code anywhere NEAR as well I do. If it did, then it would know that if I changed the right side of a statement that had a auto type to something wrong, that it was actually wrong, even if the new type was syntactically compatible with the code to follow (which might be really simple and only use some operators that many, many types would support.) But if course it wouldn't know that error if it sat down on it, because it has absolutely no idea what the INTENT of that code is or what the DESIGN of my code is. And so that would create a silent failure that isn't caught because I was too lazy to make my intent clear to the compiler, using what tools are available, and explicit types are a key indication of intent.
Explorans limites defectum
-
Andy Hoffmeyer wrote:
The simple fact is that, most of the time, the compiler is smarter than you, and understands your code on a level that you never could. Allowing the compiler to determine the type automatically, as often as possible, allows for optimizations that may not be possible if you coerce an explicit type.
So, by that logic if I was using an IDE with good intellisense and hovered over a variable declared initially with auto, which showed me what the omniscient compiler will decide the type should be, and then explicitly declared the variable to be that exact type it would somehow break the multi-dimensional optimization the compiler would perform. Are you seriously saying that or did I misread your comment??
You're performing a good bit of mental gymnastics to arrive at that interpretation of what I said. Clearly, if you know the exact type that would be deduced, there would be no penalty for explicitly using it.
-
You're performing a good bit of mental gymnastics to arrive at that interpretation of what I said. Clearly, if you know the exact type that would be deduced, there would be no penalty for explicitly using it.
But you always DO know the type, well 99.9999% of the time. So clearly there's no penalty. So how exactly does the compiler know more than us, particularly enough to risk the potential silent bugs that auto could introduce?
Explorans limites defectum
-
I usually do "all that work" by hovering the mouse over the function call and seeing what the IDE shows as the prototype of the function. There is no guesswork involved.
"They have a consciousness, they have a life, they have a soul! Damn you! Let the rabbits wear glasses! Save our brothers! Can I get an amen?"
Really the above isn't the issue. The issue is this:
auto whatever = GetSomething()
while (somecondition)
whatever++;In a large code base, there are probably a hundred or more things that would be syntactically compatible with that, so that accidentally changing GetSomething(), either manually or via search and replace, such that it returned one of those types, would create a completely silent error. Is it going to happen every day? No, obviously not. But that's not the point. The point is that, this:
FailureCounter& failCnt = GetFailCounter();
while (somecondition)
failCnt++;is just far less likely to be subject to such a silent error because you have to make two parallel errors for that to happen. You are expressing your intent to the compiler by using an explicit type, which is the only way the compiler can know if your intent is not being followed. And how much extra work did that take to get the extra safety? Almost nothing.
Explorans limites defectum
-
But you always DO know the type, well 99.9999% of the time. So clearly there's no penalty. So how exactly does the compiler know more than us, particularly enough to risk the potential silent bugs that auto could introduce?
Explorans limites defectum
Can you give an example of such silent bugs? In fact, I think the ISO C++ committee would probably be interested in hearing about these bugs so they could address them in the next release.
-
Can you give an example of such silent bugs? In fact, I think the ISO C++ committee would probably be interested in hearing about these bugs so they could address them in the next release.
I gave one below and I'm sure that they know about them and they cannot address this, because it's fundamental to why auto is dangerous.
auto whatever = GetSomething()
while (somecondition)
whatever++;If you accidentally change the right side to anything that provides a ++ operator (anything that is syntactically valid for the loop), the compiler will never know that's wrong, because you are not providing the compiler with information about your intent. The compiler is only being given SYNTACTICAL guidance when you use auto, not SEMANTIC guidance, which is what it needs to help you in this situation. If you provide the actual type, then you are telling the compiler what your intent is, i.e. semantic information, and so you have to make two parallel errors in most cases for this to silently cause a bug. Otherwise, you won't know until you somehow realize that something isn't getting incremented as it should be which could have most likely been caught at compile time with explicit typing.
Explorans limites defectum
-
I gave one below and I'm sure that they know about them and they cannot address this, because it's fundamental to why auto is dangerous.
auto whatever = GetSomething()
while (somecondition)
whatever++;If you accidentally change the right side to anything that provides a ++ operator (anything that is syntactically valid for the loop), the compiler will never know that's wrong, because you are not providing the compiler with information about your intent. The compiler is only being given SYNTACTICAL guidance when you use auto, not SEMANTIC guidance, which is what it needs to help you in this situation. If you provide the actual type, then you are telling the compiler what your intent is, i.e. semantic information, and so you have to make two parallel errors in most cases for this to silently cause a bug. Otherwise, you won't know until you somehow realize that something isn't getting incremented as it should be which could have most likely been caught at compile time with explicit typing.
Explorans limites defectum
That sounds like your API has issues. Functions are interface contracts and if you don't know that function's return type changed, then the interface has been broken. That's outside the scope of language keywords, in my opinion.
-
That sounds like your API has issues. Functions are interface contracts and if you don't know that function's return type changed, then the interface has been broken. That's outside the scope of language keywords, in my opinion.
So you are saying only one class in the entire code base can have a ++ operator? Or a += operator? or an add() method or a push() method?
Explorans limites defectum
-
So you are saying only one class in the entire code base can have a ++ operator? Or a += operator? or an add() method or a push() method?
Explorans limites defectum
I never said anything even close to that. I am saying if the maintainer of the GetSomething() function changes the return type without informing the consumer, that's a major problem that has nothing to do with language features.
-
I never said anything even close to that. I am saying if the maintainer of the GetSomething() function changes the return type without informing the consumer, that's a major problem that has nothing to do with language features.
So you are saying that mistakes shouldn't happen? Of course GetSomething()'s return could be changed by accident, and that would get caught also. But the more likely scenario is that someone accidentally changes the call, either by editing the wrong thing, or by search and replace, so that something besides GetSomething() is being called. Either of those things would become silent failures that could be taken care of by using an explicit type. Are they going to happen every day? No, they won't. But it's those type of silent errors that are the killers. Those are the ones that suddenly six months later the code stops working in the field and no one understands why.
Explorans limites defectum
-
Really the above isn't the issue. The issue is this:
auto whatever = GetSomething()
while (somecondition)
whatever++;In a large code base, there are probably a hundred or more things that would be syntactically compatible with that, so that accidentally changing GetSomething(), either manually or via search and replace, such that it returned one of those types, would create a completely silent error. Is it going to happen every day? No, obviously not. But that's not the point. The point is that, this:
FailureCounter& failCnt = GetFailCounter();
while (somecondition)
failCnt++;is just far less likely to be subject to such a silent error because you have to make two parallel errors for that to happen. You are expressing your intent to the compiler by using an explicit type, which is the only way the compiler can know if your intent is not being followed. And how much extra work did that take to get the extra safety? Almost nothing.
Explorans limites defectum
I always use the highest warning level available and set warnings to be failures and I have done this for many years. Since the auto keyboard has been available I have never seen a compiler allow anything through that could cause a failure.
"They have a consciousness, they have a life, they have a soul! Damn you! Let the rabbits wear glasses! Save our brothers! Can I get an amen?"