C++ Insights is a Clang-based tool that does a source-to-source transformation. The goal of C++ Insights is to make things visible that normally and intentionally happen behind the scenes. It's about the magic the compiler does for us to make things work.
Take this piece of code for example:
Nothing special, and of course, it compiles. This is the compiler's view on it:
You can see all the compiler-provided special member functions and the upcast from Derived to Base.
C++ Insights is a Clang-based tool that does a source-to-source transformation. The goal of C++ Insights is to make things visible that normally and intentionally happen behind the scenes. It's about the magic the compiler does for us to make things work. Or looking through the classes of a compiler.
In 2017, I started looking into some new things we got with C++11, C++14, and C++17. Amazing things like lambdas, range-based for-loops, and structured bindings. I put it together in a talk. You can find the slides and a video online.
However, all that research and some of my training and teaching got me to start thinking about how it would be if we could see with the eyes of the compiler. Sure, there is an AST dump, at least for Clang. We can see what code the compiler generates from a C++ source snippet with tools like Compiler Explorer. However, what we see is assembler. Neither the AST nor the Compiler Explorer output is in the language I write code. Hence, I'm not very familiar with this output. Plus, when teaching students C++, showing an AST and explaining that it is all there was not quite satisfying for me.
I started to write a Clang-based tool that can transform a range-based for-loop into the compiler-internal version. Then, I did the same for structured bindings and lambdas. In the end, I did much more than initially planned. It shows where operators are invoked and places in which the compiler does some casting. C++ Insights can deduce the type behind auto or decltype. The goal is to produce compilable code. However, this is not possible in all places.
You can see, for example, the transformation of a lambda, range-based for-loop, or auto. Of course, you can transform any other C++ snippet.
See yourself. C++ Insights is available online: cppinsights.io.
Still, there is work to do.
I do not claim to get all the things right. I'm also working on supporting features from new standards, like C++20, at the moment. Please remember that C++ Insights is based on Clang and its understanding of the AST.
I did a couple of talks about C++ Insights since I released C++ Insights. For example, at C++ now. Here are the slides and the video.
C++ Insights can be built inside the Clang source tree or outside.
To build with extra/clang use the following extra flags: -DINSIGHTS_USE_SYSTEM_INCLUDES=off -DCLANG_LINK_CLANG_DYLIB=on -DLLVM_LINK_LLVM_DYLIB=on
See https://github.com/andreasfertig/cppinsights/issues/186 for an explanation of why INSIGHTS_USE_SYSTEM_INCLUDES needs to be turned off.
extra/clang and extra/llvm provide /usr/lib/{libclangAST.so,libLLVM*.a,libLLVM.so}. libclangAST.so needs libLLVM.so and there would be a conflict if libLLVM*.a (instead of libLLVM.so) are linked. See https://bugs.archlinux.org/task/60512
You need to have a Clang installation in the search path.
The resulting binary (insights) can be found in the build folder.
The easiest way to build C++ Insights inside the Clang source tree is using the LLVM_EXTERNAL_PROJECTS option.
There are a couple of options that can be enabled with cmake:
| Option | Description | Default |
|---|---|---|
| INSIGHTS_STRIP | Strip insight after build | ON |
| INSIGHTS_STATIC | Use static linking | OFF |
| INSIGHTS_COVERAGE | Enable code coverage | OFF |
| INSIGHTS_USE_LIBCPP | Use libc++ for tests | OFF |
| DEBUG | Enable debug | OFF |
It seems best to supply the architecture during configuration:
Then, in Cevelop Import -> General -> Existing Project into Workspace. Select build_eclipse. Enjoy editing with Cevelop.
Using C++ Insights is fairly simple:
Things get complicated when it comes to the system-include paths. These paths are hard-coded in the binary, which seems to come from the compiler C++ Insights was built with. To help with that, check out scripts/getinclude.py. The script tries to collect the system-include paths from the compiler. Without an option, getinclude.py uses g++. You can also pass another compiler as a first argument.
Here is an example:
The script can be used together with C++ Insights:
In case you have a custom build of the GCC compiler, for example, gcc-11.2.0, and NOT installed in the compiler in the default system path, then after building, Clang fails to find the correct libstdc++ path (GCC's STL). If you run into this situation, you can use "`--gcc-toolchain=/path/GCC-1x.x.x/installed/path`" to tell Clang/C++ Insights the location of the STL:
Here "`${GCC_11_2_0_INSTALL_PATH}`" is the installation directory of your customized-built GCC. The option for Clang is described here.
There is also another GitHub project that sets up a docker container with the latest C++ Insights version in it: C++ Insights - Docker
A plugin for Vim is available at here.
An extension for Visual Studio Code is available at the VS Code marketplace: C++ Insights - VSCode Extension.
At least for macOS, you can install C++ Insights via Homebrew thanks to this formulae:
I aim for the repository to compile with the latest version of Clang and at least the one before. The website tries to stay close to the latest release of Clang. However, due to certain issues (building Clang for Windows), the website's version is often delayed by a few months.
I created a YouTube channel where I release a new video each month. In these videos, I use C++ Insights to show and explain certain C++ constructs, and sometimes I explain C++ Insights as well.
See TODO.
If you like to support the project, consider submitting a patch. Another alternative is to become a GitHub Sponsor or a Patreon supporter.
