CMake 2022 C++ Modules and More - Bill Hoffman - CppNow 2022

[@fnizzelwhoop]

Great presentation -- he was clearly hyped about giving this talk and showing off the progress being made in CMake. With that said, this presentation made me decide to finally give up on C++. What a wonderful mess it is.

[@piotrarturklos]

The topic of modules starts at 6:13

[@TheLavaBlock]

nice to see progress is being made

[@DrUlrih]

Yep, great😃 I have to type so much attribute words (internal fileset etc) to produce build graph, oh no :(

[@puyadaravi3109]

If CMake is asking the compiler for the dependencies of each source file, doesn't it mean the compiler will potentially have to scan parts of the source tree multiple times? (making it inefficient) If so is there a caching mechanism to help with this?

[@mapron1]

a bit inefficient. 1. parsing source code is much much faster then code generations. like 10x and more (for heavy templates). 2. OS usually do good caching job itself, bottleneck is usually disk reads, not like 'parsing'. 3. if you working with IDE, you end up loading each file multiple times anyway. clang, highlight/editor, doxygen/test/todo parsers, build system modules, actual compilation, you can complete the list. so, after project load and code model parsed, you get disk cache hit. Another concern is CI build times (cache probably won't be as good as a lot of project building same time). Anyway (1) is still applied. Like, today, if you have qt project and enable automoc it will scan you entire tree for 'Q_OBJECT' macro. and it can be done in a second or two for 1 MSLOC codebase or something. So, yeah, overhead, but it's negligible.

[@mapron1]

Well, on windows, creating a new process is slow, so I can imagine process spawn time can be a concern here (having 5ms spawn and 1ms actual work)

[@lukasklause2239]

Nice Talk, thank you for uploading ^^

[@BoostCon]

Thank you too!

[@neuzen]

Why there is no unified project management tool (like cargo) and packages repo (like crates)?