Header inclusion guidance for C++ in 1.0.0+

[jeongseok-meta]

I've noticed that since version 1.0.0, individual C++ headers seem to have changed in terms of their self-containment. Specifically, including a single header now results in compilation errors. For instance, I've found that including drjit/fwd.h fails to build due to a missing definition for JitBackend.

Could someone please clarify if this change was intentional? If so, I'd greatly appreciate guidance on the recommended way to include headers for C++-only use cases. If not, I'd be happy to submit a patch to fix this issue. Thank you in advance for your time and assistance.

[dvicini]

I just want to add a +1 here: it would be great to again get to a state where individual headers compile if possible.

[wjakob]

Is this mainly about fwd.h or also other header files? (e.g. internal header files in the src directory). If the latter, for this kind of header file wrangling, do you know of good tooling to solve this?

[dvicini]

We internally use https://github.com/include-what-you-use/include-what-you-use, but I have no idea how easy that is to use in open source.

Fixing this is not urgent, I just wanted to flag that I also ran into these problems :)

[jeongseok-meta]

Thanks for confirming, @dvicini. Yeah, it's not urgent, but it would be good to have it fixed :)

I would say a simple approach would be to create a unit test that includes only fwd.h and resolve any issues with it as a starting point.

[merlinND]

IWYU seems fairly easy to use!
You can build it locally for a Clang version of your choice, and then it integrates nicely with CMake. When building Mitsuba:

cmake -DCMAKE_CXX_INCLUDE_WHAT_YOU_USE=/path/to/bin/include-what-you-use ..

It will write out a ton of missing, as well as unneeded, includes (for each file).
I haven't tried yet, but apparently you can then apply the fixes automatically from that log with:

python3 fix_includes.py < /tmp/iwyu.out