Whose idea was this in C\?

[obermd]

var should be banished to the depths of time! If you can't explicitly type the variable then you obviously don't understand the problem.

[Greg Utas]

var is like auto in C++, right? I use it wherever possible. I know what the type is. If the reader doesn't, they need to learn so that they'll truly understand the code. :-D

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.

[Slacker007]

We use var all the time at our shop, where applicable.

[Lost User]

So,

Greg Utas wrote:

auto in C++, right? I use it wherever possible. I know what the type is. If the reader doesn't, they need to learn so that they'll truly understand the code.

You hide your variable types behind the auto keyword and have an elevated THREAD_PRIORITY_HIGHEST thread in all your multi-threaded processes. Any other tips you want to share? It just keeps getting better.

[TNCaver]

I used to explicitly type all my variables until recently when it seems to have become fashionable in my department to use var. I don't normally follow coding fashions unless they make sense, and this sure makes the code look cleaner. And as long as IntelliSense knows what type it is and shows me the correct properties and methods, I'm happy.

If you think 'goto' is evil, try writing an Assembly program without JMP.

[Greg Utas]

The type is on the RHS.

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.

[Lost User]

Yep, I'd like to get your opinion on something. Would you be willing to do that? How does the auto keyword impact the reliability, debuggability and robustness of C++ code?

[Greg Utas]

I'm always willing to play the straight man! The main problem with auto is that the type ends up being something that you need to consider more carefully:

for(auto i = container.size() - 1; i >= 0; --i)...

and you've got an infinite loop because i is unsigned. I've been burned by this a few times, so have learned to write int instead of auto here. But that overrides the type; "correctly" saying size_t would cause the same problem.

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.

[Lost User]

Thanks, I'm still learning the ropes. But I do enjoy hanging out with you guys in the Lounge. :rose:

[Keefer S]

Awesome to see such productive back and forth comments. Seemed to border on getting offtrack, but was rescued at the last moment. Great work guys!

[Dave B 2022]

Quote:

var is like auto in C++, right?

I used to think that. But I am not so sure anymore in terms of the net effect in each language. In our C# code, I demand that the type be determinable by reading the line the type is defined on. I.e. if a variable is initialized by a value returned from a method, you best not use var. The reasoning is not that a developer can't determine the type with enough effort. It is the time it takes someone reading the code to KNOW what the code is doing and meant to do. At the end of the day it is about developer productivity. Experienced developers aren't questioning if they can make something work or focusing on the time it takes to key in their code. It is how effective they are at solving problems and maintaining volumes of code. Using var is optimizing the writing of code at the expense of the total cost of the code over the life of the code. But, when moving over to C++ and understanding the design best practice of (Always Use Auto), I had to step back and evaluate why someone would recommend a coding pattern that made code take longer to understand and would drop developer productivity. In the end, I came to the conclusion that the C# patterns that we tend to use don't use structs, almost always use classes, and as a result, put the memory on the heap. In these scenarios there is no runtime benefit the compiler can help you with that makes your code more effective since all non primitive variables are just references. var only gives an opportunity to save time typing code at the expense of maintaining it. C++ on the other hand appears to prefer patterns where local variables do not use the heap and the compiler is both optimizing the implicit type conversion and if it can use a reference instead of a copy with every variable use. In the patterns I am aware of, it is minimizing the copying of data and the running of class move/copy/ctor/dtor methods for you. Without using auto here, the compiler's hands have been tied. All of that being said, I would like the input of those who have written a lot of modern C++ to see if I am missing anything. I also believe that if our C# code used structs that had implicit type conversions more, the reasoning of not allowing var may be misguided. Although I have not pulled on this thread and do not know much about the C# compiler optimization in these cases.

[Greg Utas]

I'm unaware of any way that auto helps to produce more efficient code. If you don't want a deep copy, you have to write auto& id =.... And if it's const, you have to write const auto id =.... Yes, Herb Sutter wrote an article "Almost Always Auto (AAA)." As far as someone reading the code having to look up the type returned by a function goes, my take is that knowing the type is only the first step to understanding. The reader also needs to know the function's purpose, which means reading its interface documentation. And given that we sometimes fall short on that front, it can also mean reading its implementation. Providing the type can therefore be detrimental by giving a false sense of security. Although I haven't used C#, I'd be surprised if best practices for when to use the heap versus the stack weren't the same in both languages.

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.

[Kirk 10389821]

Did I ever mention why I hated MSFT? and most of their products? THIS. Exactly This! The last straw was when they stopped adding changes to the 16 Bit C/C++ compiler that they were putting into the 32 bit version. Our lead dev made a 32 bit library that we were forced to write a Thunking layer to use. He used almost every new feature he could. In the end, I forcibly recompiled the code using a Borland 16 bit compiler. The ONE thing I LIKED about Oracle was that for DECADES we would simply DUMP our DB and Code. Import it into a newer version, and it worked. Hundreds of upgrades, and we barely ever ran into something that no longer compiled. Something I can honestly say NEVER happened with MSFT stuff. From VB breaking every version, to the above, to MSSQL T-SQL changes. (Heck, SqlCmd has a :Connect command. Try to use it in Azure hosting! Because it does NOT support choosing the Database. So it fails. It's as if ONE HAND has no idea what the other is doing). Good ideas are great... But going in to make a small change to a system, and finding out you cannot even begin to recompile it because of the new compiler. Imagine if Linux was built on those precepts! I feel your pain!

[Kate X257]

I'm probably missing a joke or 2 here, but I'm having a hard time seeing the benefit of auto-typing. Since you 100%, always, with 0 possible exceptions know the complete type when you're writing the code.. Why would you want to obscure it? It's like having both a pig, and some lipstick, and feeling compelled to apply the latter to the former just because you can. :rose:

[Kate X257]

It has come to my attention that you did not read the memo. I do not know what to tell you. We all did a zoom call and agreed on this months ago. ;P

[Kate X257]

At the end of the day, it's still sacrificing ease of maintenance for better CPU performance. No matter how you slice it, that's not a trade-off I'm comfortable with, given the average amount of defects per KLoC in C++ code. Then again, I avoid writing C++ for specifically that reason. (that and my hate for having to maintain and patch a couple of C++ projects that have long been abandoned by their creators)

[Lost User]

Nah, I was twisting the dagger down to the last letter[^]. I just wrote it using double entendre so that only a poet could know it. I like Greg, this is just my unique way of showing it. I think he can handle it. :)

[Dave B 2022]

Greg, thank you for the reply. It looks like the reason auto makes this more efficient that I was recalling was wrong. Although my conclusion is somewhat the same. Rereading the AAA article, I picked out the point:

Quote:

It is efficient by default and guarantees that no implicit conversions (including narrowing conversions), temporary objects, or wrapper indirections will occur. In particular, prefer using auto instead of function<> to name lambdas unless you need the type erasure and indirection.

The same is true of C# if one uses implicit conversions as can be seen in the following code:

public class T1 {
int m_count;

public T1(int cnt) 
{
    m\_count = cnt;
}

public static implicit operator T2(T1 t1) {
    return new T2(t1.m\_count.ToString());
}

}

public class T2 {
string m_count;
public T2(string count) {
m_count = count;
}
}

public class Class1
{
T1 GenerateValue() {
return new T1(22);
}

void temp() {
    
    // The following creates an instance of T1, converts it to T2 and 
    // leaves T1 for the GC to clean up.
    // Is it clear to a developer that they just created this overhead?
    T2 t2\_implicit = GenerateValue();

    // The following creates and holds an instance of T1
    T1 t1\_explicit = GenerateValue();

    // The following also creates and holds an instance of T1
    var t1\_implicit = GenerateValue();
}

}

This means my original statement was in error. The reason we "benefit" from not using var has little to do with us not using structs. It is because we don't use implicit conversions on the vast majority of our classes/structs. So the potential "mistake" of accidentally forcing an unnecessary type conversion is minimal and is outweighed by putting type information at the developers fingertips.

Quote:

Although I haven't used C#, I'd be surprised if best practices for when to use the heap versus the stack weren't the same in both languages.

The thing that makes C# different from C++ in this case is that all memory created to hold instances of a class cannot be placed on the stack. Class instance memory is always placed on the heap and managed by the garbage collector (GC). On the other hand, a struct is always considered a value type and storage follows the same as it would for

[Greg Utas]

That's interesting how C# treats classes and structs differently. As far as implicit type conversion in C++ goes, a constructor that takes a single argument can be tagged explicit to the avoid unintended creation of an object. I've rarely used implicit construction because it can make the code opaque.

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.

[Greg Utas]

When using templates in C++, the type can be a mess, so it looks like a pig. Using auto also lets me keep lines to 80 characters while rarely having to spill them. So the pig disappears, and all that's left is the lipstick. Call it the Cheshire Pig!

Robust Services Core | Software Techniques for Lemmings | Articles
The fox knows many things, but the hedgehog knows one big thing.