Automating clang-format and clang-tidy targets

[junglie85]

I've liked the look of please for a while but have been working on Windows so it's not been possible to try it out. Now I'm back in the *nix world, I'm looking forward to getting my hands dirty with it.

I'm working with a very simple hello world in c++ and would like to do the following:

• Use a system installed version of clang-format and configure please to manually invoke it via a target.
• Use a system installed version of clang-tidy and configure please to manually invoke it via a target.
• Configure please to automatically invoke these tools as part of the build process for a cc_binary target.

For clang-format, I need to find all relevant files, which gives me a task a bit like this:

sh_cmd(
    name = "clangformat",
    cmd = "find . -iname *.h -o -iname *.cpp | xargs clang-format -i",
)

I'd rather something like:

sh_cmd(
    name = "clangformat",
    cmd = "clang-format -i **/*.cpp" **/*.h,
)

But that fails if there are no header files. How can I use please to recursively find all c,cpp,cxx,h,hpp,hxx,inl etc files and pass them as arguments to clang-format?

Similarly, how can I do this for clang-tidy?

sh_cmd(
    name = "clangtidy",
    cmd = "clang-tidy **/*.cpp -- **/*.cpp",
)

clang-tidy works much better with a compilation database. CMake can generate a compilation database; is this something please is able to do or in the roadmap if not currently possible?

How do I link these sh_cmds to the cc_binary target as they're not tools in the tools folder?

[Tatskaari]

The short answer is no, we don't have that tight integration with the C/C++ ecosystem. As you've discovered, you can use Please as a task runner to run commands via sh_cmd(). You can also use commands like plz query inputs, plz query print --field srcs, or plz query graph to get information about the build graph including the inputs to each target, but they won't directly give you what you're after. You may be able to build a compilation database off the back of these tools though.

Another approach that might work would involve extending the C/C++ rules. Similar to how the Golang Please package driver works, we could output the compilation database info for each cc_library() rule as a sub-rule. For context, Golang has this concept of a package json file that describes something very similar i.e. all the source files in a particular translation unit which can be used by other tools to understand foreign build systems like Please. We have implemented this in the go rules by adding an internal #pkg_info target to the go_library() rules that outputs a json file describing things about the package, including the source files. A similar approach for C/C++ rules might make sense.

By the way, there was a bit of discussion a while back about linters here, which you may find interesting:
#1294

Nothing really happened off the back of this, but the idea was that rules would define lint actions so you'd just be able to plz lint //src/... and it would run clang-tidy because cc_library() etc. define that as a lint action. Unfortunately a lot of linters require access to the whole source graph, and we never really came up with an elegant design that would work for this use case.