How to run clang analyzer with emscripten?

[ravisumit33]

Recently I am trying to run clang analyzer with emscripten. It provides scan-build which wraps the command and analyze. But that doesn't work with emcc. I tried to explore what it does internally and found that it invokes clang with --analyze. I tried passing it manually with emcc only to get below error:

clang++: warning: argument unused during compilation: '-c' [-Wunused-command-line-argument]
error: action RunAnalysis not compiled in
make[2]: *** [CMakeFiles/hello_world.dir/src/main.cpp.o] Error 1
make[1]: *** [CMakeFiles/hello_world.dir/all] Error 2
make: *** [all] Error 2

Is there a way to run static analyzer on cpp code written for Emscripten?

[sbc100]

I don't think anyone has tried to add support for this.

I think the idea is that scan-build will replace emcc in the build system, but I don't think we need emcc itself to support --analyze (Since IIUC when scan-build runs it doesn't actually run emcc at all, but replaces it with clang --analyze.

If you have time to work adding support that would be most welcome.

[sbc100]

Oh wow.. it looks like scan-build is a perl script :) Looks like you could maybe just modify that list of known compiler in RunBuildCommand here: https://github.com/llvm/llvm-project/blob/b24c33a9d745bd2a3009f1d52f31247772e954e5/clang/tools/scan-build/bin/scan-build#L1105-L1109

[ravisumit33]

@sbc100 Issue is scan-build uses its own clang which uses its own flags where things go wrong. I use boost which do checks on platform and compiler to include certain header files. I tried providing the exact command (as found in compile-commands.json) to scan-build. But then boost started to give compilation error that it can't find some mac os specific header (my current system). Ideally scan-build should have used the same complier for which the code was written for, isn't it?

[sbc100]

I'm afraid I don't know anything about scan-build myself.

[ravisumit33]

@sbc100 I can try building emscripten's clang with the RunAnalysis action in it. Can you guide me how emscripten does it?

[ravisumit33]

Here is what all I have to do to make clang-static-analyzer run (throught clang-tidy) with emscripten on my code base:

1. Run export LLVM_CMAKE_ARGS=-DCLANG_ENABLE_STATIC_ANALYZER=ON,-DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;lld" to add required environment variables
2. Build llvm from source using emsdk. Run ./emsdk install llvm-git-main-64bit
3. Activate custom llvm: ./emsdk activate llvm-git-main-64bit
4. Run clang-tidy by running clang-tidy -p=<folder-containing-compilation-database> <file-path> --extra-arg=--target=wasm32 --extra-arg=-D__EMSCRIPTEN__

@sbc100 Right now this is not possible with the latest emscripten and cpp linters such as clang-tidy doesn't work. Given the steps what are the chances of this getting into emscripten ?

[sbc100]

There is nothing special about emscripten's version of clang. Its just an completely normal/unmodified version of clang.

It sounds like the issue is that clang-tidy is not part of emsdk? But if you have have a working version of clang-tidy on your system it should be possible to just run clang-tidy --extra-arg=--target=wasm32 --extra-arg=-D__EMSCRIPTEN__. Is that right? The clang-tidy binary doesn't have to come from emsdk itself right?

Even it we add clang-tidy (which seems reasonable to do) it still won't work out of the box because emsdk does not put clang binaries in your PATH by default. It only puts upstream/emscripten and not upstream/bin in the user's PATH. So a user would have to know where to find clang-tidy and which arguments to pass.

But if you think its still worth adding the clang-tidy binary I would be fine with that.

The code that we use to build the emsdk binaries is here: https://chromium.googlesource.com/emscripten-releases/+/main/src/build.py#704

[ravisumit33]

Two things will be required:

1. Emscripten's LLVM need to be built with -DCLANG_ENABLE_STATIC_ANALYZER=ON,-DLLVM_ENABLE_PROJECTS="clang;clang-tools-extra;lld" so that it produces clang that supports static analyzing cpp code and it produces tools such as clang-tidy.
2. On the user side, he/she will need to pass the clang-tidy and clang path into its editor integration or if invoking manually will have to use the absolute path. Also the extra args. I think this is pretty doable.

Does this sound good?

[sbc100]

That sounds reasonable yes.

Of course anyone with clang-tidy installed locally can already do this today, right?

[ravisumit33]

Actually clang-tidy invokes the clang it was built with. I checked this by trying three different llvm versions: my own build, emsdk custom llvm build & homebrew llvm package. Each one invoked its own clang binary. So, if the user has clang-tidy installed and the corresponding clang binary supports static analysis and webassembly target (for example, homebrew package supports both but mac os default one doesn't support webassembly target), it should work.

I think it will be better if emscripten provides both the clang-tidy tool and the clang binary so that users can do seamless integration.

[sbc100]

Sure. Most clang builds support the default set of architectures and webassembly is in the default set, so I think recommending folks use their own clang is often fine.

If you would like to open a PR to add clang-tidy to the emsdk distribution please feel free: https://chromium.googlesource.com/emscripten-releases/+/main/src/build.py#704

[ravisumit33]

I tried with multiple editor integrations and each of them worked without needing me building llvm by my own. As you rightly said, webassembly target seems to be built in all of them (except the default clang on mac). I just needed to add --target=wasm32 -D__EMSCRIPTEN__ extra complier flags, thats all. Now, I believe there is no need for Emscripten to provide clang-tidy binary. May be just a documentation of how to use it.

[sbc100]

The __EMSCRIPTEN__ macro should be automatically defined when --target=wasm32-emscripten: https://github.com/llvm/llvm-project/blob/cc5237c7af6417139fb32f3e312b9489a6e0bf62/clang/lib/Basic/Targets/OSTargets.h#L960

[ravisumit33]

Yes it indeed does. Thanks for this.