Cpp Dependency Analysis

Header files, used in the wrong way, can make CeePlusPlus compilations really slow. The only reason is that they include more source code to compile. And the headers they include have some more source code, and so on.

There are things you can do to prevent including more than you really need to, but the first step is to figure out where all your compilation time is going. I whipped up a small tool in python that helps you find the culprits, you can get it here: http://csteinbach.com/public/cppdep-0.2.tar.gz

It contains three python scripts. explore.py lets you see the include tree together with the cumulative number of lines included for each file. It should look something like this:

explore.py takes preprocessor output as its input. There is also a script tree.py which is simply a text version of explore.py.

Another script report.py helps pinpoint header file bloat for a whole project. The output is like this,

 . . .
 . . .
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/bits/stl_algobase.h lines in file(842), all included lines(82076), places included from(22)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/bits/stl_algo.h lines in file(5148), all included lines(91564), places included from(22)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/algorithm lines in file(71), all included lines(91674), places included from(22)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/ios lines in file(53), all included lines(94203), places included from(18)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/bits/char_traits.h lines in file(376), all included lines(94817), places included from(22)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/istream lines in file(774), all included lines(154096), places included from(18)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/sstream lines in file(643), all included lines(161358), places included from(18)
 /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/string lines in file(60), all included lines(278947), places included from(22)
 ../include/log4cpp/OstringStream?.hh lines in file(52), all included lines(294740), places included from(22)

In this case OstringStream?.hh that has produced 294740 lines of code for compilation and it has been included in 22 different places.

All these scripts take preprocessor output as their input. Here's how you get hold of preprocessor output:

Preprocessor output for Windows developers

If you are programming with Visual Studio change the project C++ settings to produce preprocessor output (this works for VS 7, not sure for VS 6 or earlier). After a build you will see a bunch of files with a .i extension. These can be passed directly to the scripts. If you are using the CL (Windows) compiler program from the command line or a script, use the compiler option /P to generate preprocessor output.

Preprocessor output from GnuCpp

If you are using makefiles, then GnuCpp has a flag --save-temps that will leave the preprocessor output in a file with a .ii extension. The -E flag will also produce preprocessor output, but will stop the compilation process at that point.

Script usage syntax

 $ python explore.py <module.i> # or module.ii for g++
 $ python tree.py <module.i>
 $ python report.py module1.i module2.i ...

--ChrisSteinbach

Note: On Linux the explore.py requires the package tkinter to be installed to get the graphics to work. Once this is in it all works nicely.

-- JohnFletcher

A small function was added so that it displays the "include chain" from a source to a header. Would've emailed you a patch, but was unable to find a contact. My email is my first-name at last-name dot net. It's posted here http://progzer.spaces.live.com/blog/cns!733D363B1B9A8BA9!1013.entry -- Pranav Tekchand

Some techniques that help reduce header bloat: PimplIdiom, RedundantIncludeGuards, SelfContainedHeaders, ForwardDeclaration

CategoryCpp