The include cost of a header is its size plus the size of all transitively included headers. The goal is to measure the impact on the parsing time in clang. That's why

• the cost is measured in bytes and not in lines of code
• the size of files without header guards is added each time the file is included
• code excluded by the pre-processor is not counted towards the total because clang is able to skip such blocks of code efficiently
• comments over certain length are not counted towards the total because clang is able to skip such blocks of code efficiently

To obtain the data we

1. check out fresh sources of libc++
2. build libc++
3. for each file required by C++ standard and provided by libc++
• for each supported language version
• measure the cost of including such file
4. TODO: should we exclude the sizes of built-in and libc headers?