I guess my point is that make doesn't scale, and unavoidably gets messy over time as each project solves the same problems over and over in different, buggy ways.

make solves problems you get early in a project ("let's get this handful of files automatically built") and doesn't even try to address issues of modern software development, like build parallelism and dependency management. Its facilities for debugging problems are laughable.

Go look up plan9's mk. It's a remake of the program written by the designers a few decades later with everything they learned from it. Not only are the variables idiomatic, and the spaces problem fixed, but you can ask for output at each stage and (IIRC) view the graph that it builds.

Of course you quickly get to million-line Makefiles in bigger projects. In C++, if you depend on stuff like Boost, a single file can easily depend on thousands of header files, and every single one must be tracked by Make.

But of course, nobody in his right mind would write these dependencies by hand. At my dayjob, the core Makefile is 700 lines. These 700 lines do cross-compilation with gcc/clang for various platforms, unit testing, valgrind, cppcheck, clang-tidy, coverage, and several other things. It is completely non-recursive, runs in parallel without problems, tracks all dependencies, and also works on Windows.

Of course this stuff is complicated. Make is very low-level compared to a build-generator like CMake. But if you want a bespoke build system for your project which lets you control every last detail of your build, it is still a good choice.