Skip to content

useclang.md

Latest commit

 

History

History
107 lines (80 loc) · 3.97 KB

useclang.md

File metadata and controls

107 lines (80 loc) · 3.97 KB

How to use clang

From a developer point of view, we highly recommend clang over other compilers. The toolset provided is very convenient and the errors are easier to track.

Advanced debugging options

Here is a non-exhaustive list of the debugging tips

Use -fsanitize=address

Use the address sanitizer from clang by adding the -fsanitize=address to the compilation and linker flags.

To get a decent stack trace you should compile with -O1 and -fno-omit-frame-pointer. To further improve the trace, use -fno-optimize-sibling-calls.

For more information, refer to the clang documentation

Use -fsanitize=undefined

Use the undefined behavior sanitizer from clang by adding the -fsanitize=undefined to the compilation and linker flags.

To get a decent stack trace you should compile with -O1 and -fno-omit-frame-pointer. Also, run with UBSAN_OPTIONS=print_stacktrace=1.

For more information, refer to the clang documentation

Use valgrind

Please, use the fsanitize instead! It's much more powerfull and robust. The classical vagrind is useful but sometimes not enough (see the sanitize). To run it:

mpirun -n X valgrind ./murphy...

You can also add some options like --show-leak-kinds=all --leak-check=full --track-origins=yes --error-limit=no to help detect the leaks.

Clang Optimization viewer

To run (and view the optimization) analysis, add the flags -fsave-optimization-record to the compilation.

Then, simply use the tool from llvm to create html webpages (run the command directly in the Docker):

python3 /usr/lib/llvm-11/share/opt-viewer/opt-viewer.py --output-dir opt_reports build/*.opt.yaml

VSCode integration

configure your C/C++ vscode extension

  • install the Intelisens extension
  • create a new configuration profile (either open c_cpp_properties.json or use the UI: cmd+P + type ">C/C++: Edit configurations (UI)")
  • my configuration is something like:
{
    "name": "murphy-container-clang",
    "includePath": [
        "src",
        "src/clients",
        "src/core",
        "src/grid",
        "src/operator",
        "src/poisson",
        "src/time",
        "src/tools",
        "src/wavelet",
        "test/src",
        "/soft/flups/include",
        "/soft/p4est-github/include",
        "/usr/local/hdf5/include",
        "/usr/local/openmpi/include",
        "/usr/local/fftw/include",
        "/soft/googletest/include"
    ],
    "defines": [],
    "compilerPath": "/usr/bin/clang",
    "cStandard": "c17",
    "cppStandard": "c++17",
    "intelliSenseMode": "linux-clang-x64",
    "compileCommands": "${workspaceFolder}/compile_commands.json",
    "browse": {
        "path": [
            "${workspaceFolder}/src/**",
            "${workspaceFolder}/test/src/**"
        ],
        "limitSymbolsToIncludedHeaders": true,
        // "databaseFilename": ""
    }
}
  • choose the configuration: cmd+P + type ">C/C++: Select a configuration")

Now the IntelliSense of VSCode is using clang to guide you and generate error/warnings

configure the extension Clang-Tidy

Clang-Tidy is helpfull to track and detect all the c++ errors (like objects declaration etc).

VSCode has an extension that allows you to visualize the output directly in the editor.

  • install the extension if not already done (should be done automatically as part of the .devcontainer.json file)
  • you need to generate a compilation database for clang (a file compile_commands.json). As we now have clang in the docker the easiest is by far to use clang directly
  • we implemented a rule in the makefile to do so: make compdb (or make compdb_full to include the tests)

Clang resources:

Other useful resources