optimise includes in math

[pseyfert]

cleaning up a few includes in math/mathmore and math/physics.
NB:

1. root builds for me with these changes but i didn't check for platform independence (when it comes to system headers)
2. removing includes from headers may break 3rd party code including those headers i manipulated.

[sbogacz]

This is my first foray into this project, but there are a few instances of commenting out the includes instead of fully removing them. If they can in fact be safely removed regardless of the target platform, then the lines should be removed to avoid cluttering up the code base.

[vgvassilev]

@pseyfert, good work! Could you rebase and remove the redundant lines instead of commenting them out?

[phsft-bot]

Can one of the admins verify this patch?

[pseyfert]

• Rebased and deleted includes instead of commenting out.
• I didn't rerun IWYU so I'm curious if the build still succeeds.
• Also the rebase failed on math/physics/inc/TVector3.h where I used the master for simplicity.

[vgvassilev]

@phsft-bot: build.

[vgvassilev]

@phsft-bot: build!

[vgvassilev]

@phsft-bot: build!

[vgvassilev]

@phsft-bot build!

[vgvassilev]

@pseyfert, do you think you can optimise the rest of the includes in ROOT in a similar manner in a new PR?

[pseyfert]

need to check. i didn't get iwyu running with cmake and haven't used ./configure since quite a while.
i had tried it back then for roofit and realised it caused problems in my own "third party" code if a class A has a method to return a pointer to an instance of class B, but A.h doesn't include B.h.

[vgvassilev]

It seems that there are two ways to use it within cmake: http://stackoverflow.com/questions/30951492/how-to-use-the-tool-include-what-you-use-together-with-cmake-to-detect-unused-he Vassil Msg sent from my phone. Please excuse my brevity.

…

On Mar 10, 2017, at 17:21, pseyfert ***@***.***> wrote: need to check. i didn't get iwyu running with cmake and haven't used ./configure since quite a while. i had tried it back then for roofit and realised it caused problems in my own "third party" code if a class A has a method to return a pointer to an instance of class B, but A.h doesn't include B.h. — You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or mute the thread.

[vgvassilev]

Merged, thanks! I'd be happy to review a PR, running IWYU on the rest of the ROOT codebase!

[pseyfert]

cool, will notify you once i came around running it.

[pseyfert]

@vgvassilev I put the output of a first run /afs/cern.ch/work/p/pseyfert/public/iwyu.log It's quite big (and i didn't even build all components) and also involves parts I don't dare to touch (llvm). I'll clean up my changes to the cmake files and factorise the suggestions a bit.

[vgvassilev]

Good progress. I do not understand why it says we should remove the .o files. Do you have an in-source build somehow?

We should ignore the fixes in llvm at first. Then we should ask the iwyu to order the suggested includes by generality (i.e. stl, libc go last).

[pseyfert]

changes to cmake now available here: https://github.com/pseyfert/root/tree/IWYU .
It's not an in-source build, but I iterated a few times with

cd ../build
rm -rf *
cmake ../root -G Ninja .......
cmake --build .

and once i saw the first iwyu output (realising i put the parts together correctly), I didn't do an rm -rf (don't touch a running system) and ran:

cmake --build . --clean-first -- -v

[vgvassilev]

Ok, looks very good. Can we implement this as a travis check. I don't expect people to run it locally.