Handling stdafx.h in cross-platform code

Question

I have a Visual Studio C++ based program that uses pre-compiled headers (stdafx.h). Now we are porting the application to Linux using gcc 4.x.

The question is how to handle pre-compiled header in both environments. I've googled but can not come to a conclusion.

Obviously I want leave stdafx.h in Visual Studio since the code base is pretty big and pre-compiled headers boost compilation time.

But the question is what to do in Linux. This is what I found:

1. Leave the stdafx.h as is. gcc compiles code considerable faster than VC++ (or it is just my Linux machine is stronger ... :) ), so I maybe happy with this option.

2. Use approach from here - make stdafx.h look like (set USE_PRECOMPILED_HEADER for VS only):

   #ifdef USE_PRECOMPILED_HEADER
   ... my stuff
   #endif 
   

3. Use the approach from here - compile VC++ with /FI to implicitly include stdafx.h in each cpp file. Therefore in VS your code can be switched easily to be compiled without pre-compiled headers and no code will have to be changed.
   I personally dislike dependencies and the mess stdafx.h is pushing a big code base towards. Therefore the option is appealing to me - on Linux you don't have stdafx.h, while still being able to turn on pre-compiled headers on VS by /FI only.

4. On Linux compile stdafx.h only as a precompiled header (mimic Visual Studio)

Your opinion? Are there other approaches to treat the issue?

Cricoid

Answer 1

You're best off using precompiled headers still for fastest compilation.

You can use precompiled headers in gcc as well. See here.

The compiled precompiled header will have an extension appended as .gch instead of .pch.

So for example if you precompile stdafx.h you will have a precompiled header that will be automatically searched for called stdafx.h.gch anytime you include stdafx.h

Example:

stdafx.h:

#include <string>
#include <stdio.h>

a.cpp:

#include "stdafx.h"
int main(int argc, char**argv)
{
  std::string s = "Hi";
  return 0;
}

Then compile as:

> g++ -c stdafx.h -o stdafx.h.gch
> g++ a.cpp
> ./a.out

Your compilation will work even if you remove stdafx.h after step 1.

Oxcart

I meant just that after you run g++ -c stdafx.h -o stdafx.h.gch, you can remove stdafx.h and your compilation would still work (g++ a.cpp). You wouldn't actually do this, but I mentioned it as proof that it is using the precompiled header. – Oxcart

@Brian Maybe I'm wrong but this looks like option 4 I listed :) I tried to remove stdafx.h from a.cpp and it didn't compiled - precompiled header will be searched only if respective header file is included. I confirmed it with strace :)). – Cricoid

@idimba: Yes it is option 4, but you modified your question and added #4 three minutes after I posted my answer. The respective header file should be included anyway in source. – Oxcart

Answer 2

I used option 3 last time I needed to do this same thing. My project was pretty small but this worked wonderfully.

Outage

Answer 3

I'd either go for option 4 or option 2. I've experimented with precompiled headers on both various VS versions and GCC on Linux (blog posts about this here and here). In my experience, VS is a lot more sensitive to the length of the include paths, number of directories in the include path and the number of include files than G++ is. When I measured build times properly arranged precompiled headers would make a massive difference to the compile time under VS whereas G++ was pretty much unimpressed by this.

Actually, based on the above what I did the last time I worked on a project where this was necessary to rein in the compile time was to precompile the equivalent of stdafx.h under Windows where it made sense and simply used it as a regular file under Linux.

Chastitychasuble

Answer 4

Very simple solution. Add a dummy file entry for "stdafx.h" in Linux environment.

Hutton

Answer 5

I would only use option 1 in a big team of developers. Options 2, 3, and 4 will often halt the productivity of other members of your team, so you can save a few minutes a day in compile time.

Here's why:

Let's assume that half of your developers use VS and half use gcc. Every now and then some VS developer will forget to include a header in a .cpp file. He won't notice, because the stdafx.h implicitly includes it. So, he pushes his changes in the version control, and then a few other members of the gcc team will get compiler errors. So, for every 5 minutes-a-day you gain by using precompiled headers, 5 other people waste by fixing your missing headers.

If you don't share the same code across all of your compilers, you will run into problems like that every day. If you force your VS developers to check for compilation on gcc before pushing changes, then you will throw away all your productivity gains from using precompiled headers.

Option 4 sounds appealing, but what if you want to use another compiler at some point in time ? Option 4 only works if you only use VS and gcc.

Notice that option 1 may make gcc compilation suffer a few seconds. Although it may not be noticeable.

Tumescent

Answer 6

It's simple, really:

Project->Project Settings (Alt + F7)

Project-Settings-Dialog:
C++ -> Category: Precompiled Headers -> Precompiled Headers radio buttons --> disable

Jitter

Answer 7

Since stdafx.h is by default all the Windows-specific stuff, I've put an empty stdafx.h on my other platform. That way your source code stays identical, while effectively disabling stdafx on Linux without having to remove all the #include "stdafx.h" lines from your code.

Counterfactual

Answer 8

If you are using CMake in your project, then there are modules which automate it for you, very convenient, for example see cmake-precompiled-header here. To use it just include the module and call:

include( cmake-precompiled-header/PrecompiledHeader.cmake )
add_precompiled_header( ${target} ${header} FORCEINCLUDE SOURCE_CXX ${source} )

Another module called Cotire creates the header file to be precompiled (no need to manually write StdAfx.h) and speeds up builds in other ways - see here.

Cyrillus

Answer 9

I've done both option 2 (#ifdef) and option 4 (PCH for gcc) for cross platform code with no issues.

I find gcc compiles much faster than VS so the precompiled headers are generally not that important, unless you are referencing some huge header file.

Quotient

Answer 10

I have a situation where #2 in particular didn't work for me (There are numerous VS build configs where a #ifdef around #include "stdafx.h" does not work). Other solutions were suboptimal because the files themselves were cross-project as well as being cross-platform. I did not want to force preprocessor macros to be set or force linux or even windows builds to use (or not use) pch, so...

What I did, given a file named notificationEngine.cpp, for example, was removed the #include stdafx.h line entirely, created a new file in the same directory called pchNotificationEngine.cpp with the following contents:

#include "stdafx.h"
#include "notificationEngine.cpp"

Any given project can just include the correct version of the file. This admittedly is probably not the best option for cpp files that are only used by a single project.

Trabeated

Let's not worry about my recent comment on a different question: The compiler for main just needs the function/class declarations (.hpp), not the implementations/definitions. The linker finds those. #includeing .cpp files can be done but is incredibly bizarre. – Trabeated