sammccall added a comment. Details below, but TL;DR is running only on files in `compile_commands.json` is definitely the established way to analyze a project. If you're proposing a *change* then that might be worth considering, but this probably isn't the place: mailing lists are more public and offline is higher-bandwidth :-)
In D113902#3137311 <https://reviews.llvm.org/D113902#3137311>, @kuhnel wrote: > In D113902#3131707 <https://reviews.llvm.org/D113902#3131707>, @sammccall > wrote: > >> Hmm, clang-tidy should only be running on entries in compile_commands.json. >> are these files listed somehow? > > No, they are not. > > The question is: > Do the tools our contributors use care about this? This is not the default > behavior of clang-tidy, it's rather implemented in a LLVM-specific wrapper > script. So do we require our contributors to use that script to check their > code? --- A few things to unpack here: - contributor status quo: our contributors mostly don't currently run clang-tidy, and we don't require them to. - clang tools in general: when running tools on a codebase, we use the CDB to define the scope of the sources as well as to ensure we're only running on sources with reliable flags. - clang-tidy proper. Most clang tools support `--executor=all-TUs` which iterates over the CDB to run on a project. It looks like clang-tidy does not have this, only explicitly running on one file at a time. - wrapper script: I assume you're talking about `run-clang-tidy.py`? It provides the `--executor=all-TUs` behavior (and parallelism) as an external script. This is not LLVM-specific, it's part of the clang-tidy apt package etc along with the clang-tidy binary. > I was playing with the VS Code extension > <https://marketplace.visualstudio.com/items?itemName=notskm.clang-tidy> to > see the findings right in the IDE and that calls clang-tidy directly. So it > ignores compile_commands.json (In fact it doesn't quite, it'll use it to pick a best approximate set of flags via InterpolatingCompilationDatabase. clangd behaves the same way.) But yes, there's a difference in behavior between what happens when you run a tool on a file (you get best effort, even if the file is unknown) vs when you run a tool on a project (you get only the files that are considered part of the project). This is pretty consistent: e.g. clangd will try to assist in editing any file, but will only index files in compile_commands.json. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D113902/new/ https://reviews.llvm.org/D113902 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
