Knit is a build tool inspired by Make and Plan9 Mk. You define rules with a Make-like embedded syntax within a Lua program. Rules can be passed around as Lua objects, and generated by Lua code. You can use the Lua module system to make reusable modules for building any kind of source code. Knit combines the readability of a Make-style rules language will the power and expressiveness of Lua. If you are familiar with Make, you can learn Knit very quickly.
Knit tracks more of your build to give you better incremental builds. For example, Knit automatically adds an implicit dependency on a rule's recipe, so if you change a recipe (either directly or through a variable change), Knit will automatically re-run all rules that were affected. Knit will also skip build steps dynamically if it can determine that they will be unchanged from the previous run (even if earlier dependencies were changed).
Knit has support for namespaced sub-builds that execute relative to their
directory, but Knit avoids build fragmentation because sub-builds don't rely on
spawning build sub-processes. No more make -C
to do sub-builds! Everything is
tracked by the root Knitfile, but you can still make directory-specific rules.
Knit's rules language is heavily inspired by Plan9 Mk. In some ways, Knit can be considered a modern version of Mk with a Lua meta-programming system built on top of it (there are some differences compared to Mk).
Why make yet another build system? Because it's fun and useful to me! Maybe it will be useful to you too. Everyone hates something about their build system so if you have feedback or a request, let me know! The project is new enough that your feedback may be seriously taken into account.
Articles: I have written two articles with more details about Knit here and here (this article is specifically about D, but the optimization Knit applies is general and not tied to D specifically).
- Knit uses Lua for customization. This makes it possible to write reusable build libraries, and in general makes it easier to write powerful and expressive builds.
- Knit has built-in syntax for a rules language inspired by Make and Plan9 Mk. This makes it very familiar to anyone who has used Make/Mk.
- Knit has direct support for sub-builds (compared to Make, which usually involves spawning a separate make sub-process to perform a sub-build).
- Knit can hash files to determine if they are out-of-date, rather than just
relying on file modification times.
- Knit additionally uses hashes for "dynamic task elision": if Knit can dynamically determine that a prerequisite that was rebuilt actually changed nothing, it won't re-run the dependent build step, allowing for even better incremental builds compared to timestamp-based approaches (Make, Ninja, etc.).
- Knit tracks recipe changes, so if you update a variable (in the Knitfile or at the command-line), any dependent rules will be automatically rebuilt.
- Knit supports
%
meta-rules and regular expression meta-rules. - Knit uses rule attributes instead of using targets such as
.SECONDARY
to indicate special processing. - Knit supports virtual rules that are independent of the file system.
- Knit uses sane variable names like
$input
,$output
, and$match
instead of Make's$^
,$@
, and$*
. - Knit supports rules with multiple outputs, and treats them like Make's group targets by default.
- Knit supports sub-tools that implement various build utilities including:
- Generating a graph visualization using graphviz (dot).
- Showing build status information (whether targets are out-of-date and why).
- Exporting a compile commands database for use with a language server.
- Automatically cleaning all build outputs.
- Converting your build into a shell script, Makefile, or Ninja file.
- Knit will search up the directory hierarchy for a Knitfile, allowing you to run your build from anywhere in your project.
- Knit supports parallel builds and uses all cores by default.
- Cross-platform support (Windows support is still experimental).
- Knit uses a shell to execute commands. By default, Knit searches for
sh
on your system and uses that. If it cannot findsh
, it uses an internal (cross-platform) shell.
- Knit uses a shell to execute commands. By default, Knit searches for
Here is a very basic Knitfile for building a simple C project.
return b{
$ hello: hello.o
cc -O2 $input -o $output
$ %.o: %.c
cc -O2 -c $input -o $output
}
The syntax for rules is nearly the same as Make, and Knit supports %
meta-rules just like Make. However, rather than using a custom language
to configure the build, Knit uses Lua.
Here is a more complex example Knitfile used for building a simple C project.
This time the Knitfile supports various configurations (changing cc
and
enabling debug flags), and automatically detects the source files.
local knit = require("knit")
local conf = {
cc = cli.cc or "gcc",
debug = tobool(cli.debug) or false,
}
local cflags := -Wall
if conf.debug then
cflags := $cflags -Og -g
else
cflags := $cflags -O2
end
local src = knit.glob("*.c")
local obj = knit.extrepl(src, ".c", ".o")
local prog := hello
return b{
$ build:VB: $prog
$ $prog: $obj
$(conf.cc) $cflags $input -o $output
$ %.o:D[%.d]: %.c
$(conf.cc) $cflags -MMD -c $input -o $output
}
Running knit hello
would build all the necessary .o
files and then link
them together. Running knit hello debug=1
would change the flags and re-run
the affected rules. Running knit build
will build hello
(effectively an
alias for knit hello
). The VB
attributes on the build rule means that it
is virtual (not referring to a file on the system), and should always be built
(out-of-date).
Running knit -t clean
will run a sub-tool that automatically removes all
generated files.
Header dependencies are automatically handled by using the -MMD
compiler flag
with the D[%.d]
attribute. To explicitly name the dependency file (e.g., to
put it in a .dep
folder), you could instead use:
$ %.o:D[.dep/%.dep]: %.c
$(conf.cc) $cflags -MMD -MF $dep -c $input -o $output
Note that Knitfiles are Lua programs with some modified syntax: special syntax
using $
for defining rules, and special syntax using :=
for defining raw
strings (no quotes) with interpolation.
See the docs for more information.
See examples for a few examples, and see this repository's Knitfile and the tests for even more examples.
Prebuilt binaries are available from the release page.
You can install one automatically using eget.
eget zyedidia/knit
Or you can build from source (requires Go 1.19):
go install github.com/zyedidia/knit/cmd/knit@latest
- Ninja to Knit converter (for compatibility with cmake, and for benchmarking). See knitja for the converter tool.
- Performance optimizations.
- Knit can already be used to build large projects, such as CVC5 (using the knitja converter). For massive builds though, like LLVM, Knit suffers from some performance problems that could be improved.
- Better support for dynamic dependencies. Currently it is possible to handle
dynamic dependencies by generating rules, but I would like to explore the
possibility of a more clean and cohesive solution.
- Ptrace enabled automatic dependency discovery (Linux-only feature). See the xkvt project for some experiments on this front.
- Global build file cache (similar to
ccache
, but for every command that is executed). - A restrictive mode for build sandboxing.
It is always useful useful to get feedback from others to improve Knit. If you have feedback, or questions about how to use it, please open a discussion. It would be great to discuss the good and bad parts of the current design, and how it can be improved.
Usage of knit:
knit [TARGETS] [ARGS]
Options:
-B, --always-build unconditionally build all targets
--cache string directory for caching internal build information (default ".")
--cpuprofile string write cpu profile to 'file'
-D, --debug print debug information
-C, --directory string run command from directory
-n, --dry-run print commands without actually executing
-f, --file string knitfile to use (default "knitfile")
--hash hash files to determine if they are out-of-date (default true)
-h, --help show this help message
--keep-going keep going even if recipes fail
-q, --quiet don't print commands
--shell string shell to use when executing commands (default "sh")
-s, --style string printer style to use (basic, steps, progress) (default "basic")
-j, --threads int number of cores to use (default 8)
-t, --tool string subtool to invoke (use '-t list' to list subtools); further flags are passed to the subtool
-u, --updated strings treat files as updated
-v, --version show version information
Available sub-tools (knit -t list
):
list - list all available tools
graph - print build graph in specified format: text, tree, dot, pdf
clean - remove all files produced by the build
targets - list all targets (pass 'virtual' for just virtual targets)
compdb - output a compile commands database
commands - output the build commands (formats: knit, json, make, ninja, shell)
status - output dependency status information
If you find a bug or have a feature request please open an issue for discussion. I am sometimes prone to being unresponsive to pull requests, so I apologize in advance. Please ping me if I forget to respond. If you have a feature you would like to implement, please double check with me about the feature before investing lots of time into implementing it.
If you have a question or feedback about the current design, please open a discussion.