split-off input_context.h from input.h; run iwyu to always include input.h directly

[Brambor]

Summary

None

Purpose of change

Split-off input_context.h from src/input.h to reduce compilation time when changing something in header files.
Test how well did I do.
I want to split src/input.h in another PR.

I realize that, since there are unneeded includes, it still can happen, that I recompile more than I need to (when working with input.h). But the ratio (of files needing recompilation after the future input.h split vs before) will say what the ideal (lowest) recompilation file count would be if all indirect includes were sorted out.

Sorting out these unneeded includes would take too much time. I do only small bites.

Describe the solution

run iwyu

• First commit: accept (not all) changes that only remove includes / forward declarations (don't add any). I might as well do this since I run iwyu anyway. (almost pure, had to add lightmap.h include)
• Second commit: includes in input.h, input.cpp according to iwyu.
• Third commit: expand includes so that #include "input.h" is directly exposed everywhere. On affected files, remove whatever iwyu reports to remove, but add only the minimum needed for compilation (take advantage of indirect includes).
• check iwyu (tests below)
• Fourth commit: split-off input_context.h and input_enums.h from input.h and input_context.cpp from input.cpp
• • input_enums.h contains things that both of the other header files need. It is included in both of the header files.
• Fifth commit: like third, but expand input.h and input_context.h, if neither is needed, but input_enums.h, include it.
• Test how effective this is by counting the number of files needing recompilation before and after this change.

Conclusion

Test results bellow.

I set out to reduce re-compilation times for class input_manager in src/input.h. It is now (indirectly) included in 36 rather than 240 files! So this was a resounding success.

I believe, however, that the success comes from splitting the input.h file, not from running iwyu alone. As can be seen from not reducing includes for split-off input_context.h. Therefore I don't think running iwyu on the whole codebase is worth my time right now. However, it seems that splitting monolithic header files would have a significant impact.

Describe alternatives you've considered

Running over the whole codebase. But I want something merge-able. And small bites.

Testing

compiling locally
compiling in this pr

(running) iwyu

• suggests no changes for input.h and input.cpp
• suggests no changes for input_context.h and input_context.cpp
• doesn't suggest to add or remove input.h or input_context.h anywhere
• suggests to add input_enums.h only if input.h or input_context.h are already included

Additional context

Not a draft, so that all tests run. Ready to merge now!

[github-actions]

Auto-requesting reviews from non-collaborators: @andrei8l

[Brambor]

If the tests pass, this is ready to be merged!

[akrieger]

Clang tidy says

Error: /home/runner/work/Cataclysm-DDA/Cataclysm-DDA/src/lightmap.h:45:36: error: no member named 'log' in namespace 'std' [clang-diagnostic-error]
    return static_cast<int>( -std::log( LIGHT_AMBIENT_LOW / b ) * ( 1.0 /

[Brambor]

Clang tidy says

Error: /home/runner/work/Cataclysm-DDA/Cataclysm-DDA/src/lightmap.h:45:36: error: no member named 'log' in namespace 'std' [clang-diagnostic-error]
    return static_cast<int>( -std::log( LIGHT_AMBIENT_LOW / b ) * ( 1.0 /

It should be resolved now.

[ehughsbaird]

Test how effective this is by counting the number of files needing recompilation before and after this change.

Before

% cat obj/*.inc | grep input.h | wc -l
243

After

% cat obj/*.inc | grep input_context.h | wc -l
240
% cat obj/*.inc | grep input_enums.h | wc -l
244
% cat obj/*.inc | grep input.h | wc -l
36

[Brambor]

I wondered if the includes in that file could repeat, so I did this, and they don't seem to be repeating.

% grep -m 1 input_context.h obj/*.inc | wc -l
240

[Brambor]

Nooooooooooooooooo
A merge conflict :(
Why do you do this to me?

Looks so simple.
But I will have to rerun this.
Will take about an hour....

[Maleclypse]

Sorry! Ping me in discord when it's passed tests after merge conflict resolution.

[Brambor]

Let's just try including both, what could go wrong?

[Brambor]

test failed, restarting

[Brambor]

trying to restart again

Clang 12 Asan cancelled during compilation for some reason, so all tests after that also didn't run. Someone should give this a kick probably.

[akrieger]

widget_test segfaulted on both windows and mac, which is slightly suspicious. If it were just one, I wouldn't care, but those are two wildly different platforms.

[andrei8l]

widget_test segfaulted on both windows and mac, which is slightly suspicious. If it were just one, I wouldn't care, but those are two wildly different platforms.

That's my fault. See #71213

[Brambor]

As promised, I now need to recompile only 38 files, which is over 6x improvement. I love it!