c++ templates: header files still broken? - c++

It used to be that to use a template class in C++, the implementations had to be in the header file or #included into the header file at the bottom.
I've not used C++ templates for a few years; I just started using them again and observed that this behavior seems to persist. Is this still the case? Or are the compilers smart enough now to have the implementation separate from the interface?

Technically they do not need to be in the header file.
An example of this usage is when you have a template class with a fixed set of versions (lets say for arguments sake char and wchar_t). Then you can put all the method delcarations into a source file and explicitly instanciate these two versions. This has the safety that others can not istanciate the template for types it was not meant to be used for.
// X.h
template<typename T>
class X
{
// DECLARATION ONLY OF STUFF
public:
X(T const& t);
private:
T m_t;
};
// X.cpp
#include "X.h"
// DEFINTION OF STUFF
template<typename T>
X<T>::X(T const& t)
:m_t(t)
{}
// INSTANCIATE The versions you want.
template class X<char>;
template class X<wchar_t>;
// Main.cpp
#include "X.h"
int main()
{
X<chat> x1('a');
X<wchar_t> x2(L'A');
// X<int> x3(5); // Uncomment for a linker failure.
}
Assuming people can't just directly include X.cpp (because it is not provided by the distribution) then others can not use X<int> or X<float> etc. But the abovr classes are fully defined.
I have also seen this technique used to reduce compilatio time. Because each compilation unit is not re-generating the same version of X we only get the defintion in one place (thus one compilation cost). The downsize to this is that you must manually instanciate each seprate version of X that you use.

To separate the implementation from the declaration the standard forces you to use the export keyword. As far as I know there's only one compiler that knows how to handle it: Comeau.
However, C++0x will include a mechanism that tells the compiler not to instantiate certain specializations automatically (extern templates). So, if you want to cut compilation time you will be able to do so by explicitly instantiating some specializations in one compilation unit and declaring them in the header as extern.

You are referring to exported templates (using the export keyword), which seem to be supported only by Comeau C++ (according to this section of the C++ FAQ Lite).
A common technique to keep the interface devoid of implementation code is to put the inline function definitions into a separate "implementation" header that can be included at the end of the declaration header.

Export is only support by the EDG frontend, comercially only available in the Comeau compiler as far as I know.
Export doesn't eliminate the need for source disclosure, nor does it reduce compile dependencies, while it requires a massive effort from compiler builders.
So Herb Sutter himself asked compiler builders to 'forget about' export. As the time investment needed would be better spend elsewhere... so I don't think export will ever be implemented by other compilers after they saw how long it took, and how little was gained.
The paper is called "Why we can't afford export", it's listed on Sutters blog but no pdf there (a quick google should turn it up though), it's six years old now, I suppose they all listened and never bothered :)
Many people use two header files (e.g. .hpp and .ipp), one with only the declaration, and one with the definitions, then it's simply a matter of including one in the other.
foo.hpp
#ifndef MY_TEMPLATES_HPP
#define MY_TEMPLATES_HPP
template< class T >
void foo(T & t);
#include "foo.ipp"
#endif
foo.ipp
#ifdef MY_TEMPLATES_IPP
nonsense here, that will generate compiler error
#else
#define MY_TEMPLATES_IPP
template< class T >
void foo(T & t) {
... // long function
}
#endif
This only gains some clarity of course, nothing really changes compared to simply inlining everything in one header file.

GCC goes through a lengthy collect stage unless you explicitly instantiate all templates. VC++ seems to cope, but I prefer to avoid this step anyway and in cases where I know how template is going to be used, which is usually the case for applications, not so much for libraries, I put template definitions into a separate file. This also makes code more readable by making declarations less cluttered with implementation details.

Related

Can I put `extern template` into a header file?

Is putting an extern template in a header file and then do the explicit template instantiation in a unit compilation file valid ?
For example in the compiling example for g++, is this working to avoid the instancation of nothing<int> twice ? Why doesn't anybody write it like this and prefer to copy pase the extern template line in each .cpp file ?
A.hpp:
#ifndef HEADERC_A
#define HEADERC_A
template< typename T > struct nothing {};
extern template struct nothing<int>;
#endif
A.cpp:
#include "A.hpp"
template struct nothing<int>;
main.cpp:
#include "A.hpp"
#include <iostream>
int main()
{
nothing<int> n;
return 0;
}
Well, this is certainly "valid" insofar as gcc will compile this, and do pretty much what you expect to happen.
As far as why doesn't everyone to do this, well, once you go beyond a trivial situation like this, and you start managing a large collection of widely used templates, it will quickly reach the point where it simply becomes not practical to keep track of every parameter that each one of your templates gets used with, so that it can be instantiated explicitly, in this manner.
It's going to be much easier for the compiler to keep track of it, for you.
As previously stated this is a perfectly valid use case and if it fits your programming model then you should use it. But buyer beware:
There are several reasons why extern templates are not commonly declared in header files and then explicitly instantiated in the cpp files.
A very common model for implementing template classes/functions is to place the definitions in the header file and the implementation in an "inl" or other named file. But then include that file at the bottom of the header file. There are reams of code that use this approach to resolving the template/header/implementation separation problem. Putting an "extern" at the top of the implementation makes the code much easier to read and maintain, especially when multiple classes get involved. Heres an example:
A.hpp
#pragma once
template< typename T > struct nothing {
void donothing(T input); // fastest func around
};
#include "A.inl"
A.inl
// does NOT include A.hpp
extern template struct nothing<int>; // save time and space
template<typename T> nothing<T>::donothing { return; }
Instance.h
#include "A.hpp"
template struct nothing<int>; // compiler generates code now
But there is a hidden caveat in all this...
If this gets implemented as you suggest then what happens when another person comes along and wants:
nothing<float> mynothing;
The compiler will see the header file but never find an implementation for float. So it may compile just fine, but at link time there will be unresolvable symbols.
So they try this:
template struct nothing<float>;
nothing<float> mynothing;
WRONG! Now the compiler can't find the implementation and all you get is MORE errors.
Now you could go back to your A.cpp file and add another instance for float...can you say maintenance, headache, carpal tunnel nightmare? With the commonly used solution you get to have your cake and eat it to. (mostly)
Now you may be thinking why even bother with the extern? Because as your post implies there is a typical use case where most of the time "nothing" will be used with an int template type. Having this occur in potentially hundreds of files can lead to serious compile time and code size ramifications.
Why doesn't the standards committee do something about this mess? They did! They added extern templates! In all fairness it is a difficult issue to resolve after the fact.

Should I always define my entire template class in the header file?

For some reason, it seems that C++ doesn't like it when you split a template class (i.e. a class declared as template <typename T> class Thing) into a .h and a .cpp file, like you would do with any other class.
Does that mean that when writing a template class, I should just write it all in the header file? What do C++ programmers do in these cases?
EDIT: I understand that there are alternatives to writing it all in the .h file. But what's the best option? Or what the most common option?
It is popular to split the definition of a template out into another file. It is so popular it may even be the most common practice, but I am not positive about that.
I think this is poor practice (as opposed to putting the definitions in the same header) for the following reasons.
My priorities when writing code are (most important first):
Correctness.
Run time performance.
Compile time performance.
Maintainability / Easy to read.
Easy to write.
As long as the template definition is in the same header, or in another file that the header includes, the number one priority Correctness is not impacted one way or the other.
The second priority Run time performance is not impacted in the least.
The third priority Compile time performance is negatively impacted by putting the definition in a separate file. By how much is debatable. But there is no way that a compiler can compile X amount of code and open/close two files, as fast as it can compile the exact same code while opening/closing only one file. A friend of mine who is an excellent compiler writer once told me that the slowest thing I could ask his compiler to do was to open a file. This was before we did as much compile-time computation as we do today, but still...
The fourth priority Maintainability / Easy to read is quite subjective. But in my opinion a separate file is a negative impact on this priority. When I'm reading code, it is typically easier for me to understand it if it is all in one file. And I really get annoyed when I have to go hunting for the file when its name or location is not obvious. And I get even more annoyed when the definition is split into many files. For my money, one file optimizes this priority.
The fifth priority Easy to write is also quite subjective. I don't see much advantage either way on this one. It is very slightly easier to put everything in one file. But it is certainly not difficult to create a new file. I give the "one file approach" a very slight advantage on this priority.
So in summary, out of my five priorities, the two most important, this decision makes absolutely no difference on, and separate files have small negative consequences for the 3 lower priorities. Of the 3 lower priorities, the most important of these, compile-time performance is objectively negatively impacted. The two lowest priorities are admittedly subjective as to whether or not they are positively or negatively impacted.
I see no benefit and a small cost to putting your template definitions in a separate header.
"Does that mean that when writing a template class, I should just write it all in the header file? What do C++ programmers do in these cases?"
What you can do (and is wide spread/popular practice), is separate implementation code out to special template implementation files, that will in turn be in íncluded by the template headers, containing the declarations.
The gain of this technique is considered little for most of the cases, though it has it's points, if you want to hide the implementation details and not spill header files to become large.
The point is not to have the template definition code, or specializations in separate translation units, such these can be seen directly by other translation units, including the template header file.
The common pattern is
MyTemplate.h
#if !defined(MYTEMPLATE_H_)
#define MYTEMPLATE_H_
namespace myspace {
template <typename T>
class MyTemplate {
public:
void foo(T value);
};
#include "MyTemplate.tcc"
}
#endif // MYTEMPLATE_H_
MyTemplate.tcc
// NOTE: There aren't header guards for this file intentionally!
template<typename T>
void MyTemplate<T>::foo(T value) {
// ...
}
Other popular extensions for template implementation files are .icc, .ipp, .impl. Just important, it shouldn't be .cpp, since most IDE's or build system frameworks will track this as translation unit, unless it's explicitly excluded (here's a sample why).
"So simply instead of the .cpp #includeing the header, the header #includes the .tpp (which contains the implementations)?"
Template classes work a bit differently regarding the ODR (one definition rule). Regular header files, that provide class declarations shouldn't contain implementations because the ODR would be violated when these are included from different translation units:
MyClass.h
class MyClass {
public:
void foo(int);
};
// Violates ODR if MyClass.h is included from different translation units (.cpp files)
void MyClass::foo(int param) {
}
The templated version
template<typename T>
class MyClass {
public:
void foo(T);
};
// Doesn't violate ODR if MyClass.h is included from different translation units
// (.cpp files), since the template parameter isn't instatiated here yet.
template<typename T>
void MyClass<T>::foo(T param) {
}
As soon one translation unit instantiated something like MyClass<int>, other translation units instatiating the same template signature, will use the 1st one seen.
The latter implementation part could be replaced with an #include "MyClass.impl" that contains that code, if you think it messes up readability or maintainability of your header file too much.
As a minor drawback of the #include "MyClass.tcc" technique, you should notice, that most of the popular IDE's handle syntax highlighting and intellisense poorly for these kind of template implementation files.

Separating C++ Class Code into Multiple Files, what are the rules?

Thinking Time - Why do you want to split your file anyway?
As the title suggests, the end problem I have is multiple definition linker errors. I have actually fixed the problem, but I haven't fixed the problem in the correct way. Before starting I want to discuss the reasons for splitting a class file into multiple files. I have tried to put all the possible scenarios here - if I missed any, please remind me and I can make changes. Hopefully the following are correct:
Reason 1 To save space:
You have a file containing the declaration of a class with all class members. You place #include guards around this file (or #pragma once) to ensure no conflicts arise if you #include the file in two different header files which are then included in a source file. You compile a separate source file with the implementation of any methods declared in this class, as it offloads many lines of code from your source file, which cleans things up a bit and introduces some order to your program.
Example: As you can see, the below example could be improved by splitting the implementation of the class methods into a different file. (A .cpp file)
// my_class.hpp
#pragma once
class my_class
{
public:
void my_function()
{
// LOTS OF CODE
// CONFUSING TO DEBUG
// LOTS OF CODE
// DISORGANIZED AND DISTRACTING
// LOTS OF CODE
// LOOKS HORRIBLE
// LOTS OF CODE
// VERY MESSY
// LOTS OF CODE
}
// MANY OTHER METHODS
// MEANS VERY LARGE FILE WITH LOTS OF LINES OF CODE
}
Reason 2 To prevent multiple definition linker errors:
Perhaps this is the main reason why you would split implementation from declaration. In the above example, you could move the method body to outside the class. This would make it look much cleaner and structured. However, according to this question, the above example has implicit inline specifiers. Moving the implementation from within the class to outside the class, as in the example below, will cause you linker errors, and so you would either inline everything, or move the function definitions to a .cpp file.
Example: _The example below will cause "multiple definition linker errors" if you do not move the function definition to a .cpp file or specify the function as inline.
// my_class.hpp
void my_class::my_function()
{
// ERROR! MULTIPLE DEFINITION OF my_class::my_function
// This error only occurs if you #include the file containing this code
// in two or more separate source (compiled, .cpp) files.
}
To fix the problem:
//my_class.cpp
void my_class::my_function()
{
// Now in a .cpp file, so no multiple definition error
}
Or:
// my_class.hpp
inline void my_class::my_function()
{
// Specified function as inline, so okay - note: back in header file!
// The very first example has an implicit `inline` specifier
}
Reason 3 You want to save space, again, but this time you are working with a template class:
If we are working with template classes, then we cannot move the implementation to a source file (.cpp file). That's not currently allowed by (I assume) either the standard or by current compilers. Unlike the first example of Reason 2, above, we are allowed to place the implementation in the header file. According to this question the reason is that template class methods also have implied inline specifiers. Is that correct? (It seems to make sense.) But nobody seemed to know on the question I have just referenced!
So, are the two examples below identical?
// some_header_file.hpp
#pragma once
// template class declaration goes here
class some_class
{
// Some code
};
// Example 1: NO INLINE SPECIFIER
template<typename T>
void some_class::class_method()
{
// Some code
}
// Example 2: INLINE specifier used
template<typename T>
inline void some_class::class_method()
{
// Some code
}
If you have a template class header file, which is becoming huge due to all the functions you have, then I believe you are allowed to move the function definitions to another header file (usually a .tpp file?) and then #include file.tpp at the end of your header file containing the class declaration. You must NOT include this file anywhere else, however, hence the .tpp rather than .hpp.
I assume you could also do this with the inline methods of a regular class? Is that allowed also?
Question Time
So I have made some statements above, most of which relate to the structuring of source files. I think everything I said was correct, because I did some basic research and "found out some stuff", but this is a question and so I don't know for sure.
What this boils down to, is how you would organize code within files. I think I have figured out a structure which will always work.
Here is what I have come up with. (This is my class code file organization/structure standard, if you like. Don't know if it will be very useful yet, that's the point of asking.)
1: Declare the class (template or otherwise) in a .hpp file, including all methods, friend functions and data.
2: At the bottom of the .hpp file, #include a .tpp file containing the implementation of any inline methods. Create the .tpp file and ensure all methods are specified to be inline.
3: All other members (non-inline functions, friend functions and static data) should be defined in a .cpp file, which #includes the .hpp file at the top to prevent errors like "class ABC has not been declared". Since everything in this file will have external linkage, the program will link correctly.
Do standards like this exist in industry? Will the standard I came up with work in all cases?
Your three points sound about right. That's the standard way to do things (although I've not seen .tpp extension before, usually it's .inl), although personally I just put inline functions at the bottom of header files rather than in a separate file.
Here is how I arrange my files. I omit the forward declare file for simple classes.
myclass-fwd.h
#pragma once
namespace NS
{
class MyClass;
}
myclass.h
#pragma once
#include "headers-needed-by-header"
#include "myclass-fwd.h"
namespace NS
{
class MyClass
{
..
};
}
myclass.cpp
#include "headers-needed-by-source"
#include "myclass.h"
namespace
{
void LocalFunc();
}
NS::MyClass::...
Replace pragma with header guards according to preference..
The reason for this approach is to reduce header dependencies, which slow down compile times in large projects. If you didn't know, you can forward declare a class to use as a pointer or reference. The full declaration is only needed when you construct, create or use members of the class.
This means another class which uses the class (takes parameters by pointer/reference) only has to include the fwd header in its own header. The full header is then included in the second class's source file. This greatly reduces the amount of unneeded rubbish you get when pulling in a big header, which pulls in another big header, which pulls in another...
The next tip is the unnamed namespace (sometimes called anonymous namespace). This can only appear in a source file and it is like a hidden namespace only visible to that file. You can place local functions, classes etc here which are only used by the the source file. This prevents name clashes if you create something with the same name in two different files. (Two local function F for example, may give linker errors).
The main reason to separate interface from implementation is so that you don't have to recompile all of your code when something in the implementation changes; you only have to recompile the source files that changed.
As for "Declare the class (template or otherwise)", a template is not a class. A template is a pattern for creating classes. More important, though, you define a class or a template in a header. The class definition includes declarations of its member functions, and non-inine member functions are defined in one or more source files. Inline member functions and all template functions should be defined in the header, by whatever combination of direct definitions and #include directives you prefer.
Do standards like this exist in industry?
Yes. Then again, coding standards that are rather different from the ones you expressed can also be found in industry. You are talking about coding standards, after all, and coding standards range from good to bad to ugly.
Will the standard I came up with work in all cases?
Absolutely not. For example,
template <typename T> class Foo {
public:
void some_method (T& arg);
...
};
Here, the definition of class template Foo doesn't know a thing about that template parameter T. What if, for some class template, the definitions of the methods vary depending on the template parameters? Your rule #2 just doesn't work here.
Another example: What if the corresponding source file is huge, a thousand lines long or longer? At times it makes sense to provide the implementation in multiple source files. Some standards go to the extreme of dictating one function per file (personal opinion: Yech!).
At the other extreme of a thousand-plus line long source file is a class that has no source files. The entire implementation is in the header. There's a lot to be said for header-only implementations. If nothing else, it simplifies, sometimes significantly, the linking problem.

Code organization across files that has to deal with template functions and inlining

I'm maintaining a large library of template classes that perform algebraic computations based on either float or double type. Many of the classes have accessor methods (getters and setters) and other functions that run small amounts of code, therefore such functions need to be qualified as inline when the compiler locates their definitions. Other member functions, in contrast, contain sophisticated code and thus would better be called rather than inlined.
A substantial part of the function definitions are located in headers, actually in .inl files included by headers. But there are also many classes whose function definitions happily live in .cpp files by means of explicit instantiation for float and double, which is rather a good thing to do in case of a library (here explained why). And finally, there is a considerable number of classes whose function definitions are broken across .inl files (accessor methods) and .cpp files (constructors, destructors, and heavy computations), which makes them all pretty difficult to maintain.
I would have all my class implementations in .inl files only if I knew a reliable way to prevent some functions from being inlined, or in .cpp files if inline keyword could strongly suggest compiler to inline some of the functions, which, of course, it does not. I would really prefer all the function definitions in the library to reside in .cpp files, but since accessor methods are used extensively throughout the library, I have to make sure they are inlined whenever referenced, not called.
So, in this connection, my questions are:
Does it make any sense to mark the definition of a template function with inline in view of the fact that, as I've recently learnt here, it is going to be automatically qualified as inline by the compiler regardless of whether it's marked with inline or not?
And most importantly, since I would like to have the definitions of all the member functions of a template class gathered together in a single file, either it's .inl or .cpp (using explicit instantiation in case of .cpp), preferably still being able to hint the compiler (MSVC and GCC) which of the functions should be inlined and which shouldn't, sure if such thing is possible with template functions, how can I achieve this or, if there is really no way (I hope there is), what would be the most optimal compromise?
----------
EDIT1: I knew that inline keyword is just a suggestion to the compiler to inline a function.
EDIT2: I really do know. I like making suggestions to the compiler.
EDIT3: I still know. It's not what the question is about.
----------
In view of some new information, there is also third question that goes hand in hand with the second one.
3. If compilers are so smart these days that they can make better choices about which function should be inlined and which should be called and are capable of link-time code generation and link-time optimization, which effectively allows them looking into a .cpp-located function definition at link time to decide its fate about being inlined or called, then maybe a good solution would be simply moving all the definitions into respective .cpp files?
----------
So what's the conclusion?
First of all, I'm grateful to Daniel Trebbien and Jonathan Wakely for their structured and well-founded answers. Upvoted both but had to choose just one. None of the given answers, however, presented an acceptable solution to me, so the chosen answer happened to be the one that helped me slightly more than others in making the final decision, the details of which are explained next for anyone who's interested.
Well, since I've always been valuing the performance of code more than how much convenient it is to maintain and develop, it appears to me that the most acceptable compromise would be to move all the accessor methods and other lightweight member functions of each of the template classes into the .inl file included by the respective header, marking these functions with inline keyword in an attempt to provide the compiler with a good hint (or with a keyword for inline forcing), and move the rest of the functions into the respective .cpp file.
Having all member function definitions located in .cpp files would hinder inlining of lightweight functions while unleashing some problems with link-time optimization, as has been ascertained by Daniel Trebbien for MSVC (in an older stage of development) and by Jonathan Wakely for GCC (in its current stage of development). And having all function definitions located in headers (or .inl files) doesn't outweigh the summary benefit of having the implementation of each class sorted into .inl and .cpp files combined with a bonus side effect of this decision: it would ensure that only the code of primitive accessor methods is visible to a client of the library, while more juicy stuff is hidden in the binaries (ensuring this wasn't a major reason, however, but this plus was obvious for anyone who is familiar with software libraries). And any lightweight member function that doesn't need to be exposed by the include files of the library and is used privately by its class can have its definition in the .cpp file of the class, while its declaration/definition is spiced with inline to encourage the inline status of the function (don't know yet whether the keyword should be in both places or just one in this particular case).
In short: Put the template code in a header file. Use compiler-specific forceinline or noinline keywords if the optimizer fails to make good decisions about inlining.
You can and should put definitions of template members into header files. This ensures that the compiler has access to the definition at the point of use when it finds out what the actual template parameters are, and is able to perform implicit instantiaion.
The inline keyword has very little impact on templates, since template functions are already exempted from the single definition requirement (The One Definition Rule still requires that all definitions be the same). It is a hint to the compiler that the function should be inlined. And you can omit it as a hint to the compiler to not inline the function. So use it that way. But the optimizer will still look at other factors (function size) and make its own choice on inlining.
Some compilers have special keywords, like __attribute__(always_inline) or __declspec(noinline) to override the optimizer's choice.
Mostly, though, the compiler is smart enough not to inline "complex code that makes more sense as a function call". You shouldn't have to worry about it, just let the optimizer do its thing.
Portable inlining control isn't beneficial, because the trade-offs of inlining are very platform-specific. The optimizers should already be aware of those platform-specific tradeoffs, and if you do feel the need to override the compiler's choice, do so on a per-platform basis.
1. Does it make any sense to mark the definition of a template function with inline in view of the fact that, as I've recently learnt, it is going to be automatically qualified as inline by the compiler regardless of whether it's marked with inline or not? Is the behavior compiler-specific?
I think you are referring to the fact that a member function defined in its class definition is always an inline function. This is per the C++ Standard, and has been since the first publication:
9.3 Member functions
...
A member function may be defined (8.4) in its class definition, in which case it is an inline member function (7.1.2)
So, in the following example, template <typename FloatT> my_class<FloatT>::my_function() is always an inline function:
template <typename FloatT>
class my_class
{
public:
void my_function() // `inline` member function
{
//...
}
};
template <>
class my_class<double> // specialization for doubles
{
public:
void my_function() // `inline` member function
{
//...
}
};
However, by moving the definition of my_function() outside of the definition of template <typename FloatT> my_class<FloatT>, it is not automatically an inline function:
template <typename FloatT>
class my_class
{
public:
void my_function();
};
template <typename FloatT>
void my_class<FloatT>::my_function() // non-`inline` member function
{
//...
}
template <>
void my_class<double>::my_function() // non-`inline` member function
{
//...
}
In the latter example, it does make sense (as in, it's not redundant) to use the inline specifier with the definitions:
template <typename FloatT>
inline void my_class<FloatT>::my_function() // `inline` member function
{
//...
}
template <>
inline void my_class<double>::my_function() // `inline` member function
{
//...
}
2. And most importantly, since I would like to have the definitions of all the member functions of a template class gathered together in a single file, either it's .inl or .cpp (using explicit instantiation in case of .cpp), preferably still being able to hint the compiler (MSVC and GCC) which of the functions should be inlined and which shouldn't, sure if such thing is possible with template functions, how can I achieve this or, if there is really no way (I hope there is), what would be the most optimal compromise?
As you know, the compiler may elect to inline a function, whether or not it has the inline specifier; the inline specifier is just a hint.
There is no standard way to force inlining or prevent inlining; however, most C++ compilers support syntactic extensions for accomplishing just that. MSVC supports a __forceinline keyword to force inlining and #pragma auto_inline(off) to prevent it. G++ supports always_inline and noinline attributes for forcing and preventing inlining, respectively. You should refer to your compiler's documentation for details, including how to enable diagnostics when the compiler is unable to inline a function as requested.
If you use those compiler extensions, then you should be able to hint to the compiler whether a function is inlined or not.
In general, I recommend to have all "simple" member function definitions gathered together in a single file (usually the header), by which I mean, if the member function does not require very many more #includes above the set of #includes required to define the classes/templates. Sometimes, for example, a member function definition will require #include <algorithm>, but it is unlikely that the class definition requires <algorithm> to be included in order to be defined. Your compiler is able to skip over function definitions that it does not use, but the larger number of #includes can noticeably lengthen compile times, and it is unlikely that you will want to inline these non-"simple" functions anyway.
3. If compilers are so smart these days that they can make better choices about which function should be inlined and which should be called and are capable of link-time code generation and link-time optimization, which effectively allows them looking into a .cpp-located function definition at link time to decide its fate about being inlined or called, then maybe a good solution would be simply moving all the definitions into respective .cpp files?
If you place all of your function definitions into CPP files, then you will be relying on LTO for mostly all function inlining. This may not be what you want for the following reasons:
At least with MSVC's LTCG, you give up the ability to force inlining (See inline, __inline, __forceinline.)
If the CPP files are linked to a shared library, then programs linking with the shared libraries will not benefit from LTO inlining of library functions. This is because the compiler intermediate language (IL)—the input to LTO—has been discarded and is not available in the DLL or SO.
If Under The Hood: Link-time Code Generation is still correct, "calls to functions in static libraries can't be optimized".
The linker would be performing all inlining, which might be a lot slower than having the compiler perform some inlining at compile time.
The compiler's LTO implementation might have bugs that cause it to not inline certain functions.
Use of LTO might impose certain limitations on projects using your library. For example, according to Under The Hood: Link-time Code Generation, "precompiled headers and LTCG are incompatible". The /LTCG (Link-time Code Generation) MSDN page has other notes, such as "/LTCG is not valid for use with /INCREMENTAL".
If you keep the likely-to-be-inlined function definitions in the header files, then you could use both compiler inlining and LTO. On the other hand, moving all function definitions into CPP files will restrict compiler inlining to only within the translation units.
I don't know where you learnt that, but templates are not "automatically qualified as inline by the compiler regardless of whether it's marked with inline or not". Templates and inline functions both have what is sometimes called "vague linkage" meaning their definitions can be present in multiple objects without error and the linker will use one of the definitions and discard the others. But the fact templates and inline functions both have vague linkage doesn't mean templates are automatically inline. Lions and tigers are both big cats but that doesn't mean lions are tigers.
Unless you know all the instantiations you are using in advance you can't always use explicit instantiation e.g. if you're writing a template library for others to use then you can't provide all the explicit instantiations, so you must define the template in .h (or .inl) files that the user of the code can #include. If you do know all the instantiations in advance then using explicit instantiations in .cpp files has the advantage of improving compilation time, because the compiler only instantiates the templates once in the file containing the explicit instantiations, not in every file that uses them. But that has nothing to do with inlining. For a function to be inlined its definition must be visible to the code calling it, so if you only define function templates (or member functions of class templates) in a .cpp file then they can't be inlined anywhere except in that file. If you define them in a .cpp file and do qualify them as inline then you might cause problems trying to call them from other files, which can't see the inline keyword (if a function is declared inline in one translation unit it must be declared inline in all translation units in which it appears, [dcl.fct.spec]/4.)
For what it's worth, I don't generally bother using .inl files, I just define templates directly in .h files, which gives one less file to deal with. Everything's in one place, and it just works, all files that use the templates can see the definitions and choose to inline them if desired. You can still use explicit instantiations in that case too, to improve compilation time and reduce object file size, without sacrificing inlining opportunites.
Why would that be better than just defining your template code in headers, where it belongs? What exactly are you trying to achieve? If it's fewer files, put the template code in headers, that will always work, the compiler can choose to inline everything without needing LTO, and you only have one file per class template (and you can still use explicit instantiation to improve compilation times). If you're trying to move all your code into .cpp files (which I think you're focusing on too much) then go ahead and do it. I think it's a bad idea, and will probably cause problems (link-time optimisation still has issues with the only compiler I've tried using it with, and certainly won't make compilation any faster) but if that's what you want, do whatever floats your boat.
It seems like your questions revolve around a misunderstanding here:
I would have all my class implementations in .inl files only if I knew a reliable way to prevent some functions from being inlined,
If all your template definitions are in header files you don't need "a reliable way to prevent some functions from being inlined" ... as I said above, templates are not automatially inline just because they're in headers, and if they're too large to inline the compiler won't inline them. First problem solved. Secondly:
or in .cpp files if inline keyword could strongly suggest compiler to inline some of the functions, which, of course, it does not, especially if a function marked with inline is located in a .cpp file.
As I said above, a function marked inline in a .cpp file is ill-formed unless it's also marked inline in the header, and never used in any other .cpp file. So doing this is just making life difficult and possibly causing linker errors. Why bother.
Again, all signs point to just put your template definitions in headers. You can still use explicit instantiation (as GCC does for std::string, as mentioned in the post you link to) so you get the best of both worlds. The only thing it doesn't achieve is hiding the implementations from users of the templates, but it doesn't sound like that's your aim anyway, if it is then provide non-template function API, which can be implemented in terms of templates in a single .cpp file.
This is not a complete answer.
I read that clang and llvm are able to do very comprehensive link time optimization. This includes link time inlining! To enable this, compile with optimization level -O4 when using clang++. The object files will be llvm bytecode instead of machine code. This is what makes this possible. This feature should therefore allow you to put all of your definitions in the cpp files, knowing that they will still be inlined where necessary.
Btw, the length of a function body is not the only thing that determines whether it will be inlined. A lengthy function that is only called from one location can easily be inlined at that location.

Implications of template declaration & definition

From what I understand template classes and template functions (for the most part) must be declared and defined in the same header file. With that said:
Are there any other ways to achieve separate compilation of template files other than using particular compilers? If yes, what are those?
What, if any, are the drawbacks of having the declaration and definition in the same file?
What is considered best-practice when it comes to template declaration & definition?
How To Organize Template Source Code
Basically, you have the following options:
Make the template definition visible to compiler in the point of instantiation.
Instantiate the types you need explicitly in a separate compile unit, so that linker can find it.
Use keyword export (if available)
One of the drawbacks encountered by implementing templates in the .h files is that any time you make a small change to the implementation, all the code that uses the template must be recompiled. There's really no way around this aside from not using templates, or declaring & defining them in the CPP file where you use them.
You can implement templates in a seperate file, and then include that file from the .h file. Such as:
templ.h
template<class V> V foo(const V& rhs);
#include "templ.inc"
templ.inc
template<class V> V foo*const V& rhs)
{
// do something...
return val;
}
My personal preference is to implement templates right in the h file unless they become large, and then I'll break it up in to h and inc files.
Not really. The definition of the template must be available at compile time, since templates are instantiated depending on the template arguments you give them. This is why they must be placed in headers, so the compiler can have the code to write a new instantiation. You pretty much need a compiler with support for the export keyword.
People can see your code, if that's a drawback to you. It might also be less "neat" to some people, but I don't think that's an issue.
A further problem is the compile times every time you change the .h (especially if it is included in a lot of places)
Not really. There is the export keyword, but most compilers don't support this. The only mainstream one that I know of that does support this is the Comeau compiler.
If your template is part of a public API then you're exposing your code to the world. (most people don't consider this a problem, but some do. It depends on your business).
Put them both in the same header file.