Cleanup: remove using namespace std;

[arrufat]

Fixes #3015

[ChristopheDumeunier]

Thanks for the quick reaction. This fixes my problem.
Note that you may encounter the same problem with some 'using namespace std::chrono' in some headers too. ;-)

[arrufat]

Yes, I saw a couple of those, too. Might fix it later.

[arrufat]

Not too sure about this one ab82c4f

I am ready to revert it.

[davisking]

Eh, what is the specific error we are talking about @ChristopheDumeunier ? Because doing using namespace inside a function scope is totally fine. It doesn't matter if that function is in a header or not. What's bad is doing it at a global scope in a header, as that brings symbols into client code.

Was there any of that in dlib? There shouldn't be.

I don't really want to go mass changing code that isn't breaking. That's invariably how we find out later on some platform something is now broken.

[ChristopheDumeunier]

@davisking Doing using namespace inside a function is not fine if the inline/template function is defined in a header because you don't know the compilation context of your code.

Here is a code that reproduce the problem: https://godbolt.org/z/r7T5nqMr5
error: reference to 'numeric_limits' is ambiguous

[arrufat]

So, in that example, you declare your namespace in a CPP file and include dlib after that?

If you just move the code that simulates including dlib to the top of the C++ (where the #include directives usually are), the error is gone.

I am just curious, why would you do that?

[ChristopheDumeunier]

It's a reduced example. In our application, some parts of the unit tests of the code are generated by scripts and macro, the the order of includes can be strange and vary depending on (de)activated dependencies. We use dlib from a long time, without this error. The error appeared when we added a struct named "numeric_limits" somewhere else in our namespace.

[arrufat]

The using namespace thing in dlib isn't the issue, if you name anything like something that's already inside dlib and then use using namespace <your_namespace> before including dlib, you'll have the same problem.

That's why they say you shouldn't use them globally in header files: so that you don't clash with other people's code.

But what you're doing amounts to that: using namespace before importing a header is akin to using it globally in a header file.

I am closing this PR.

[davisking]

Well, it's not exactly the same @arrufat

Since name lookup is preferred for names in your own namespace. Like if you did class matrix{}; before including any dlib headers it's still fine and nothing bad would happen, since dlib is all in the one monolithic dlib namespace any code in it will still prefer the dlib::matrix. But for symbols outside the dlib namespace, if someone goes and defines them at global scope like in the example yeah it's going to become ambiguous.

Which is generally a thing you shouldn't be doing. But that's fine, dlib is all about being convenient for users, so making these changes is fine with me. I get that people get pushed into doing stuff that is sub-optimal in software and we should make those cases not a pain when we can :)

That is to say, I just wanted to know if this was a real problem from real code and work or a "if I break it on purpose it's broken" example. It's with real code, so let's make that real code work since it's probably not going to mess up anything in dlib anyway :)

[arrufat]

I see. Actually, that's why I hurried to make this PR so that I could solve @ChristopheDumeunier's issue.

However, you scared me that I might have just broken some quirky compilers/setups by doing that stuff (I saw workarounds for GCC 4.9 while doing this PR).

And I wouldn't want to take more of your time with more people opening issues because of me...

[davisking]

I see. Actually, that's why I hurried to make this PR so that I could solve @ChristopheDumeunier's issue.

However, you scared me that I might have just broken some quirky compilers/setups by doing that stuff (I saw workarounds for GCC 4.9 while doing this PR).

Yeah, maybe just leave that one. IDK. It's an old compiler so I doubt anyone cares.

And in general, the easy thing to do is often to replace any using namespace std; with a using std::whatever_the_thing_is_its_using; and then leave the rest of the code alone. Kinda depends on how many lines need to change and how readable/unreadable it makes them. Like the chrono stuff gets kinda unreadable if it's all fully scoped on each use.

[arrufat]

I see. Actually, that's why I hurried to make this PR so that I could solve @ChristopheDumeunier's issue.
However, you scared me that I might have just broken some quirky compilers/setups by doing that stuff (I saw workarounds for GCC 4.9 while doing this PR).

Yeah, maybe just leave that one. IDK. It's an old compiler so I doubt anyone cares.

I think it's safe to remove, as dlib now uses C++14, it doesn't look like GCC 4.9 fully supports C++14: https://en.cppreference.com/w/cpp/compiler_support/14
GCC 5 was the first version to fully support C++14.

[davisking]

Thanks, looks good :)