On 2023-06-09 10:12, Kenton Wilson wrote:
> Hello All,
>
> I came across a paper by Nathan Sidwell entitled, 'Make Me a Module':
>
>
> https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2019/p1602r0.pdf
Yikes, that whole thing is just grotesque. What is he thinking?
The fundamental problem is that a module system is supposed to
make a tool like Make unnecessary!
As it incrementally compiles, the module system determines what
else needs to be compiled. The transitive closure of the dependency
graph is interleaved with the compiling.
Interfacing with make brings nothing to the table, other than leveraging
some parallelism?
If you know what to build, and what the dependencies are, you can
check them yourself rather than sending a message to make over
some protocol.
Most of the complexity in GNU make revolves around being a tool which can
be programmed to build anything, using almost any imaginable file system
structure. It has expansions and patterns, and re-evaluation, inclusion,
interpolation of outputs of external commands, control via environment
and command line variables, ad nauseum ...
The problem of building C++ modules with a particular toolchain is very
specific; it doesn't require any of that. It is a focused and specific
problem. It *has* to be solved by the C++ toolchain itself in order
for that module system to demonstrate its worth.
A comprehensive module build can be plugged into a Makefile; it can
be invoked as some unconditional step. The program to be built can
be treated as a phony target so that the module builder is always
invoked.
.PHONY: $(PROGRAM)
$(PROGRAM):
c++-module-build-system $(PROGRAM) # does everything
install: $(PROGRAM)
... steps to install program
tests: $(PROGRAM)
... steps to run tests
