text
_
_ _ ___ _____ ___| |_ ___
| | | | | .'| '_| -_|
|___|_|_|_|_|_|__,|_,_|___|
Work in progress
unmake is a makefile linter that promotes extreme portability.
Too many makefiles online are restricted to building only as Works On My Machine^TM.
unmake bucks this trend, encouraging the makefile author to think more critically about what level of platform support they want for their software builds.
Do you want to be able to build your software on macOS? On Linux? On FreeBSD? On Windows (Command Prompt, PowerShell, and/or WSL)? On fish terminals?
All of the above?
unmake can help to catch vendor-lock issues earlier in the SDLC process. So that your apps can build more reliably, for more contributors to enjoy.
```console $ cd examples
$ unmake Makefile; echo "$?" 0
$ unmake bsd/makefile; echo "$?" error at 1:16: expected one of " ", "$(", "${", ":", "t", [^ (' ' | 't' | ':' | ';' | '#' | 'r' | 'n')] 1 ```
See unmake -h for more options.
https://crates.io/crates/unmake
https://docs.rs/unmake/latest/unmake/
https://github.com/mcandre/unmake/releases
console
$ cargo install --force --path .
The unmake linter serves several purposes.
unmake provides a strict replacement for make -n, in case the local make implementation has BSD, GNU, etc. extensions. unmake encourages validating makefiles for syntactic wholeness as part of their projects' linter suites, before any tasks are actually run.
unmake encourages long or subtle shell snippets to be moved to dedicated shell script files, where they are more amenable to scanning with shell script linters. Most linters for common shell snippet wrapping languages (e.g., Ansible, Dockerfile, makefile, Vagrantfile, various CI/CD pipelines) perform very limited scanning for potential flaws in the embedded snippets, compared with linters that specifically scan shell script files.
make is a natural candidate for working around limitations in provisioning scripts. For example, go mod / cargo do not track linters or other dev dependencies, and sh defaults to ignoring errors during provisioning. make's default semantics prepare it well for provisioning and other activities. make can do many things! unmake helps it do them better.
unmake discourages vendor locking in makefile scripts. Numerous makefiles online assume a highly specific development environment. For example, assuming that (GNU) findutils, (GNU) sed, (GNU) awk, (non-PowerShell) curl are installed, with a GNU bash or zsh user interpreter, on a GNU/Linux operating system. So the typical makefile is likely to fail for (non-WSL) Windows users, or macOS users, or FreeBSD users, and so on. Ideally, our makefiles strive for portability, so that our projects can be enjoyed on a wider variety of computers. Portability, naturally restricts makefile contents to a portable subset, of what the various make implementations allow people to write. But a portability linter can curtail some of the worst offenders.
The dream is for every makefile to behave as a polyglot script, cabable of running well on most any computer platform with a make implementation installed.
unmake follows a stiff reading of the POSIX make standard:
https://pubs.opengroup.org/onlinepubs/9699919799/utilities/make.html
Briefly, characters in makefiles that are explicitly rejected by the standard, may be treated as parse errors. Implementation-defined behavior, undefined behavior, and certain ill-advised syntax, may be treated as parse errors.
Common examples of makefile syntax that may trigger parse errors;
\r) and Windows CRLF (\r\n) line endings are out of spec. If you have a need to contribute to projects with makefiles from a Windows machine, configure your text editor to use LF (\n) line endings (and a final LF as well).\t) at the beginning of rule commands, are out of spec.\\ \n) is out of spec.\\\n<eof>), are out of spec.=1) are out of spec.A) are out of spec.") are out of spec.include...\\\n) are out of spec.Certain extensions beyond the POSIX make subset, such as GNU-isms, or BSD-isms, etc., may also trigger parse errors.
Repeat: This is a linter focusing on extreme portability. We break things in testing, so that your software breaks less often in production.
Coming soon.
We do our best to catch POSIX make violations, but some may slip by. For example:
include directivesmake script executionunmake assumes the 2008 edition of the POSIX standard, and that make implementations fully comply with the standard.
Despite the release name, the POSIX edition is as of this writing still being actively collated in 2023 (!)
Neither make, nor linters, can perfectly read the mind of the makefile author. A rule that is syntactically valid, but accidentally neglects to declare a relevant prerequisite, can result in make misbehaving.
A rule with zero prerequisites, zero inline commands, and zero indented commands, may trigger a parse error. But a syntactically valid rule that happens to omit relevant prerequisite tasks or files (e.g. C/C++ source files), can run poorly in make, and may not be safeguarded by any linter messages.
Users are expected to have working knowledge of make, including the basic semantics of how make resolves task trees.
make caches artifact files using timestamps. In fact, the common touch command (UNIX, Windows) provides a useful way to force make to rebuild downstream targets. However, the POSIX standard observes that make can experience problems when run on platforms that lack sufficient, sub-second timestamp precision.
Neither make nor unmake are not expected to resolve problems caused by builds running on environments that lack a realtime clock with sufficient timestamp precision, such as when running builds directly on microcontrollers or other embedded devices.
Fortunately, some workarounds are available for problems arising from timestamp precision:
make on a machine with a realtime clock. For example, use a conventional workstation to cross-compile. Then copy the resulting artifacts onto the embedded device..PHONY, which disables more caching features.make implementation family.(None)
For more details on developing crit itself, see DEVELOPMENT.md.
FreeBSD