mise
magus installs through mise with no plugin, via its ubi backend, which pulls the GitHub release asset matching your platform:
# The hosted docs substitute the latest release tag below; on GitHub it reads
# literally - get the real value from https://github.com/egladman/magus/releases
mise use -g ubi:egladman/magus@v0.4.2
Pin it per repository by putting it in that repo's mise.toml instead:
[tools]
"ubi:egladman/magus" = "v0.4.2"
Three things to know before choosing this route.
mise owns the version, not magus self update. Upgrade with mise upgrade.
Running magus self update on a mise-managed install would replace a binary mise
believes it controls, and the next mise install would undo it.
Nothing verifies the release signature. ubi fetches the asset over HTTPS and
trusts GitHub; it performs no signature or checksum check of its own. The platform
installers and magus self update both verify the artifact against the
Ed25519-signed manifest. If that guarantee matters,
verify the release by hand afterwards, or use the platform
installer instead.
You may get a different binary than the installer hands you. The installer
defaults to the static build; ubi resolves to the dynamically linked one where a
release publishes both. darwin/amd64 currently ships only a static build, so
that is what it gets there.
Why not aqua, or the go backend?
aqua is mise's preferred
backend where a tool is registered, and it is the better target long-term because
the aqua registry carries checksum and signature metadata that ubi has no
equivalent for. It is not usable yet: magus has no aqua-registry entry, so
aqua:egladman/magus fails with no aqua-registry found. Getting one is an
upstream pull request to
aquaproj/aqua-registry, and it would
close the verification gap above.
The go backend does not currently work at all. It is documented here because
people find it anyway, not because it is supported.
Not the golden path, and not a route to adopt. magus is a build tool, and a build tool installed through the package manager of a toolchain it manages puts a circle in your foundation: you need a working Go toolchain to obtain the thing you use to manage Go builds, and when something breaks the remedy is to upgrade the toolchain you were using magus to pin. Those failures are oblique, hard to guard against, and there is usually nowhere sensible to attach an error message explaining what happened. The same objection applies to installing any build tool with npm, cargo, or pip. Use the install script.
# fails outright, before it ever reaches a compile step
mise use -g go:github.com/egladman/magus/cmd/magus@latest
go.mod requires github.com/egladman/magus/libs/diagnostics and
.../libs/gopherbuzz - nested modules with their own go.mod - at v0.0.0,
resolved only through this repo's own LOCAL replace directives; neither nested
module has ever had a tagged release. go install pkg@version (what the go
backend runs under the hood) refuses outright to build any module whose go.mod
contains a replace directive, local or not, unless that module is the main
module of the build - so the install dies on the replace directives themselves,
before dependency resolution or compilation ever starts. There is no flag that
gets around this from the consuming side; it is fixed only by the nested modules
getting real tags, at which point the require lines above would point at a real
version and the replace directives could be dropped for a downstream install.
Even if that were fixed, go install still cannot pass the -ldflags that stamp
the version, commit, and build date - a go-backend install would present itself
to magus as unknown (unknown) built unknown, the dev-build sentinel magus keys
on internally to fingerprint an unstamped build, rather than the release it came
from.