A make build tool that sucks less
GNU Make could be replaced by... redo
Published: Wednesday, Jan 13, 2021 Last modified: Monday, Dec 9, 2024
djb introduced redo’s elegant design in a couple of pages. @apenwarr then created a Python version with good documentation and fleshed out examples.
The design of redo unlike GNU make is elegant and can be shown to implemented in 1000LOC of C or even in shell!
redo in Golang introduced by Sergey Matveev, is where I began to learn and be persuaded:
I understand you very good there! I spent several months at home and work converting Makefiles without assurance that it is worth of it and not just making another kind of bicycle without any visible profits. But now my scepticism is lost completely and I literally have no Makefile in nearly all of my (and work) projects (of course except various projects forks). redo is life-changing :-).
- no pain and struggle with various Make implementations (I use BSD, GNU/Linux systems use GNU Make, that also have slightly different behaviour in its versions)
- cool autodependency on .do files and deterministic behaviour with work directories (some Make implementation change to directory when doing “$(MAKE) -C dir”, some are not)
- I can create dynamic targets in Makefiles now, like autodetecting what “#include"s exist in .c files to depend them on
- guaranteed working of parallel builds, that are often just does not work properly on many many existing Makefile
- no complete rebuilding of the project “just to be sure”
All of that just saves much time and greatly helps to make the job done.
Why hasn’t redo eaten Make’ lunch?
Some combination of multiple fragmented implementations and lack of critical mass?
Tests collected by Sergey are becoming more comprehensive. Uses the same test harness as git.
I want a single file, like a Makefile
instead of bunch of *.do files?
Yes you can have one file, but making a modification to this build file has a huge disadvantage… all targets will expire!