podusowski
commented
9 years ago Ok, I did some quick research. There are 50 compilation units (cpp files), each has 50 levels of includes (1.cpp includes 1_1.hpp, 1_1.hpp includes 1_2.hpp etc), the deepest header includes some system headers (listed in the table). Scanning means including /usr/include and /usr/lib in dependency graph (we use compiler as a scanner so we need to manually exclude them - this is why "clean build" takes longer in case when we don't check /usr/*). Some side observation: note how clean build gets prolonged just by including something.
| without scanning |
with scanning |
|||
|---|---|---|---|---|
| system headers | clean build | no rebuild | clean build | no rebuild |
| string | 0:05.47 0:05.86 0:05.68 |
0:00.06 0:00.06 0:00.07 |
0:05.25 0:05.38 0:05.15 |
0:00.10 0:00.10 0:00.10 |
| string, iostream, algorithm, memory | 0:10.58 0:10.73 0:10.33 |
0:00.08 0:00.09 0:00.06 |
0:10.20 0:10.18 0:10.59 |
0:00.13 0:00.12 0:00.14 |
| string, iostream, algorithm, memory boost/type_traits.hpp, boost/foreach.hpp, boost/any.hpp boost/shared_ptr.hpp, boost/signals2/signal.hpp |
1:39.78 | 0:00.06 | 1:38.42 | 0:00.83 |
Cmake make/ninja seems to ignore system headers (like the ones in /usr/include). Performance gain should be evaluated in pake.