On 01-04-18 17:13, Ggh via Boost wrote:

wondering...if I'm reinventing the wheel ( again as many do)....I guess there's loads of similar stuff

Similar to what? Shell scripts that write a makefile? Also, your link was broken. The fixed link leads to incomplete code ( You  don't state the target implementations (just GNU make? what versions?), features required, features already supported, etc. I think what I've seen is trivial. I'm not at all convinced that cramming the options all on some command line to be parsed by a ... questionable script is better than writing the equivalent makefile (or Makefile.am, m4 template or CMakeLists.txt, ... if you intend to generate the makefile). The code is... what I'd call bad code. You should probably have posted it for review instead (https://codereview.stackexchange.com/ e.g.). It fails to use all the conventions in place to make Makefiles maintainable. You subvert default variable naming (CC for CXX, really?). Here's the same program converted to actual C++ - in 306 lines of code (instead of the incomplete 560 lines): https://gist.github.com/sehe/57693dedda3de679c33848074d61c8e6/e48398e97f4207... Now note that *LOTS* of things have been fixed with respect to quoting of path names, but lots of things still don't work (no way to pass compiler/linker flags, static libraries aren't built (because the target rule was wrong), static libraries cannot link (because it needs position indepent code, but, again, compiler flags cannot be passed), the linker is always using `g++` instead of `gcc` even if all source files were compiled with the C compiler etc. To fix things, I cleaned things up **A LOT** MORE: https://gist.github.com/sehe/57693dedda3de679c33848074d61c8e6/3f22091fc4f393... Now the code is 351 lines of code, but actually organized (you could split headers and implementations here), it has actually working          - STATICLIBTARGET and LIBTARGET      - much simplified rules      - dependency detection / creation      - using OBJS instead of hardcoding names and even `*.o` (ugh)      - separating discover_sourcefiles, generate_makefile, run_make      - deduceLinker (instead of using g++ to link C programs...)      - allow for compiler flags      - don't repeat includes and libraries with each (intermediate) target A screenshot of interactively bootstrapping the program using itself: https://imgur.com/a/AlYRz, using the command line: ./bootstrap/mymakemaker -hd -t mymakemaker -cxxflags -std=c++1z -tt exe -wd bootstrap -mk -cc 'clang++' -i ~/custom/boost -L ~/custom/boost/stage/lib/ -l boost_program_options -l stdc++fs Reading directory:    "bootstrap" Creating makefile:    makefile.mk Making... ------------------------------------------------------ "clang++"  -I"/home/sehe/custom/boost" -MMD    -std=c++1z -c test.cpp -o "./test.o" "clang++"  -I"/home/sehe/custom/boost" -MMD    -std=c++1z "./test.o"   -L"/home/sehe/custom/boost/stage/lib/" -l"boost_program_options" -l"stdc++fs" -o "mymakemaker" ------------------------------------------------------ The resulting generated makefile looks like https://paste.ubuntu.com/p/rMhGFxnYHy/

On 05-04-18 06:58, Ggh via Boost wrote:

Seth!! THANK YOU!!!THANK YOU!!!

:)

Greg Cheers. I don't mean to rain on your parade, but perhaps I forgot to stress how... futile this still is.

For one thing, the complete same thing can be achieved with a relatively straightforward standard makefile, just replacing `OBJS` with a pattern substitution over a wildcard() call, and replacing all those "fixed" rules with some pattern rules like: %.o:%.c     $(CC) $(CFLAGS) -c $< -o $@ %o:%.cpp     $(CXX) $(CXXFLAGS) -c $< -o $@ -include $(OBJS:%.o=%.d) In fact as written, the rules for %.o are redundant as they match the builtin defaults (at least for GNU Make), as you can find out by using make -p. As an example, here's a simple FIXED `Makefile` that basically reproduces all the things you had in your C++ program, but without any external code: https://gist.github.com/sehe/2720161cec8452c38a5eb5808a7dd919 It can build all the targets from one file, regardless of how many files there are. In fact it's better in some respects (this makefile correctly deals with e.g. having `a.cpp`, `a.cxx` and `a.c` all in the same directory). Example runs to bootstrap the same program again, this time using **just** the one Makefile: https://i.imgur.com/boBdfID.png Now, look at how CMake already works with ninja, Make and MSBuild. I don't really think you are ready going to improve things yet. Of course, you can generate your own makefiles for your own convenience, though I'd probably write it in Python, Perl or something like that. Regards, Seth