The language-agnostic, universal package manager
gx is a packaging tool built around the distributed, content addressed filesystem IPFS. It aims to be flexible, powerful and simple.
gx is Alpha Quality. It's not perfect yet, but it's proven dependable enough for managing dependencies in go-ipfs and ready for pioneering developers and early users to try out and explore.
- Background
- Requirements
- Installation
- Usage
- Dependencies
- Updating
- Repos
- Hooks
- The vendor directory
- Using gx as a Go package manager
- Using gx as a package manager for language/environment X
- Why is it called gx?
- License
gx was originally designed to handle dependencies in Go projects in a distributed fashion, and pulls ideas from other beloved package managers (like npm).
gx was designed with the following major goals in mind:
- Be language/ecosystem agnostic by providing git-like hooks for adding new ecosystems.
- Provide completely reproducible packages through content addressing.
- Use a flexible, distributed storage backend.
Users are encouraged to have a running IPFS daemon of at least version 0.4.2 on their machines. If not present, gx will use the public gateway. If you wish to publish a package, a local running daemon is a hard requirement.
$ go get -u github.com/whyrusleeping/gx
This will download the source into $GOPATH/src/github.com/whyrusleeping/gx
and build and install a binary to $GOPATH/bin
. To modify gx, just change
the source in that directory, and run go build
.
Creating and publishing new generic package:
$ gx init
$ gx publish
This will output a 'package-hash' which is unique to the exact content of your package at the time of publishing. If someone were to download your package and republish it, it would produce the exact same hash.
It should be noted that gx is meant to work with existing package.json
files. If you are adding a package to gx that already has a package.json
file in its root, gx will try and work with it. Any shared fields will have the same types, and any fields unique to gx will kept separate.
E.g. A single package.json
file could be used to serve both gx and another packaging tool, such as npm. Since gx is Alpha Quality there may be some exceptions to the above statements, if you notice one, please file an issue.
If you've cloned down a gx package, simply run gx install
or gx i
to
install it (and its dependencies).
To add a dependency of another package to your package, simply import it by its hash:
$ gx import QmaDFJvcHAnxpnMwcEh6VStYN4v4PB4S16j4pAuC2KSHVr
This downloads the package specified by the hash into the vendor
directory in your
workspace. It also adds an entry referencing the package to the local package.json
.
Gx has a few nice tools to view and analyze dependencies. First off, the simple:
$ gx deps
go-log QmSpJByNKFX1sCsHBEp3R73FL4NF6FnQTEGyNAXHm2GS52 1.2.0
go-libp2p-peer QmWXjJo15p4pzT7cayEwZi2sWgJqLnGDof6ZGMh9xBgU1p 2.0.4
go-libp2p-peerstore QmYkwVGkwoPbMVQEbf6LonZg4SsCxGP3H7PBEtdNCNRyxD 1.2.5
go-testutil QmYpVUnnedgGrp6cX2pBii5HRQgcSr778FiKVe7o7nF5Z3 1.0.2
go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
This just lists out the immediate dependencies of this package. To see
dependencies of dependencies, use the -r
option: (and optionally the -s
option to sort them)
$ gx deps -r -s
go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
go-datastore QmbzuUusHqaLLoNTDEVLcSF6vZDHZDLPC7p4bztRvvkXxU 1.0.0
go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
go-keyspace QmUusaX99BZoELh7dmPgirqRQ1FAmMnmnBn3oiqDFGBUSc 1.0.0
go-libp2p-crypto QmVoi5es8D5fNHZDqoW6DgDAEPEV5hQp8GBz161vZXiwpQ 1.0.4
go-libp2p-peer QmWXjJo15p4pzT7cayEwZi2sWgJqLnGDof6ZGMh9xBgU1p 2.0.4
go-libp2p-peerstore QmYkwVGkwoPbMVQEbf6LonZg4SsCxGP3H7PBEtdNCNRyxD 1.2.5
go-log QmSpJByNKFX1sCsHBEp3R73FL4NF6FnQTEGyNAXHm2GS52 1.2.0
go-logging QmQvJiADDe7JR4m968MwXobTCCzUqQkP87aRHe29MEBGHV 0.0.0
go-multiaddr QmYzDkkgAEmrcNzFCiYo6L1dTX4EAG1gZkbtdbd9trL4vd 0.0.0
go-multiaddr-net QmY83KqqnQ286ZWbV2x7ixpeemH3cBpk8R54egS619WYff 1.3.0
go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
go-net QmZy2y8t9zQH2a1b8q2ZSLKp17ATuJoCNxxyMFG5qFExpt 0.0.0
go-testutil QmYpVUnnedgGrp6cX2pBii5HRQgcSr778FiKVe7o7nF5Z3 1.0.2
go-text Qmaau1d1WjnQdTYfRYfFVsCS97cgD8ATyrKuNoEfexL7JZ 0.0.0
go.uuid QmcyaFHbyiZfoX5GTpcqqCPYmbjYNAhRDekXSJPFHdYNSV 1.0.0
gogo-protobuf QmZ4Qi3GaRbjcx28Sme5eMH7RQjGkt8wHxt2a65oLaeFEV 0.0.0
goprocess QmSF8fPo3jgVBAy8fpdjjYqgG87dkJgUprRBHRd2tmfgpP 1.0.0
mafmt QmeLQ13LftT9XhNn22piZc3GP56fGqhijuL5Y8KdUaRn1g 1.1.1
That's pretty useful, I now know the full set of packages my package depends on.
But what's difficult now is being able to tell what is imported where. To
address that, gx has a --tree
option:
$ gx deps --tree
├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
├─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
│ ├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
├─ go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
│ ├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
│ └─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
│ ├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
├─ go-log QmNQynaz7qfriSUJkiEZUrm2Wen1u3Kj9goZzWtrPyu7XR 1.1.2
│ ├─ randbo QmYvsG72GsfLgUeSojXArjnU6L4Wmwk7wuAxtNLuyXcc1T 0.0.0
│ ├─ go-net QmZy2y8t9zQH2a1b8q2ZSLKp17ATuJoCNxxyMFG5qFExpt 0.0.0
│ │ ├─ go-text Qmaau1d1WjnQdTYfRYfFVsCS97cgD8ATyrKuNoEfexL7JZ 0.0.0
│ │ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
│ └─ go-logging QmQvJiADDe7JR4m968MwXobTCCzUqQkP87aRHe29MEBGHV 0.0.0
└─ go-libp2p-crypto QmUEUu1CM8bxBJxc3ZLojAi8evhTr4byQogWstABet79oY 1.0.2
├─ gogo-protobuf QmZ4Qi3GaRbjcx28Sme5eMH7RQjGkt8wHxt2a65oLaeFEV 0.0.0
├─ go-log Qmazh5oNUVsDZTs2g59rq8aYQqwpss8tcUWQzor5sCCEuH 0.0.0
│ ├─ go.uuid QmPC2dW6jyNzzBKYuHLBhxzfWaUSkyC9qaGMz7ciytRSFM 0.0.0
│ ├─ go-logging QmQvJiADDe7JR4m968MwXobTCCzUqQkP87aRHe29MEBGHV 0.0.0
│ ├─ go-net QmZy2y8t9zQH2a1b8q2ZSLKp17ATuJoCNxxyMFG5qFExpt 0.0.0
│ │ ├─ go-text Qmaau1d1WjnQdTYfRYfFVsCS97cgD8ATyrKuNoEfexL7JZ 0.0.0
│ │ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
│ └─ randbo QmYvsG72GsfLgUeSojXArjnU6L4Wmwk7wuAxtNLuyXcc1T 0.0.0
├─ go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
│ ├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
│ └─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
│ ├─ go-base58 QmT8rehPR3F6bmwL6zjUN8XpiDBFFpMP2myPdC6ApsWfJf 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
└─ go-msgio QmRQhVisS8dmPbjBUthVkenn81pBxrx1GxE281csJhm2vL 0.0.0
└─ go-randbuf QmYNGtJHgaGZkpzq8yG6Wxqm6EQTKqgpBfnyyGBKbZeDUi 0.0.0
Now you can see the entire tree of dependencies for this project. Although,
for larger projects, this will get messy. If you're just interested in the
dependency tree of a single package, you can use the --highlight
option
to filter the trees printing:
$ gx deps --tree --highlight=go-crypto
├─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
├─ go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
│ └─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
├─ go-log QmNQynaz7qfriSUJkiEZUrm2Wen1u3Kj9goZzWtrPyu7XR 1.1.2
│ └─ go-net QmZy2y8t9zQH2a1b8q2ZSLKp17ATuJoCNxxyMFG5qFExpt 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
└─ go-libp2p-crypto QmUEUu1CM8bxBJxc3ZLojAi8evhTr4byQogWstABet79oY 1.0.2
├─ go-log Qmazh5oNUVsDZTs2g59rq8aYQqwpss8tcUWQzor5sCCEuH 0.0.0
│ └─ go-net QmZy2y8t9zQH2a1b8q2ZSLKp17ATuJoCNxxyMFG5qFExpt 0.0.0
│ └─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
└─ go-ipfs-util QmZNVWh8LLjAavuQ2JXuFmuYH3C11xo988vSgp7UQrTRj1 1.0.0
└─ go-multihash QmYf7ng2hG5XBtJA3tN34DQ2GUN5HNksEw1rLDkmr6vGku 0.0.0
└─ go-crypto Qme1boxspcQWR8FBzMxeppqug2fYgYc15diNWmqgDVnvn2 0.0.0
This tree is a subset of the previous one, filtered to only show leaves that end in the selected package.
The gx deps command also has two other smaller subcommands, dupes
and
stats
. gx deps dupes
will print out packages that are imported multiple
times with the same name, but different hashes. This can be useful to see if
different versions of the same package have been imported in different places
in the dependency tree. Allowing the user to more easily go and address the
discrepancy. gx deps stats
will output the total number of packages imported
(total and unique) as well as the average depth of imports in the tree. This
can be used to give you a rough idea of the complexity of your package.
I firmly believe that packages are better when:
This means restructuring your code in such a way that flattens (and perhaps widens as a consequence) the tree. For example, in Go, this often times means making an interface its own package, and implementations into their own separate packages. The benefits here are that flatter trees are far easier to update. For every package deep a dependency is, you have to update, test, commit, review and merge another package. That's a lot of work, and also a lot of extra room for problems to sneak in.
This should be fairly common sense, but striving to import packages only where they are actually needed helps to improve code quality. Imagine having a helper function in one package, simply because it's convenient to have it there, that depends on a bunch of other imports from elsewhere in the tree. Sure it's nice, and doesn't actually increase the 'total' number of packages you depend on. But now you've created an extra batch of work for you to do any time any of these are updated, and you also now force anyone who wants to import the package with your helper function to also import all those other dependencies.
Adhering to the above two rules should (I'm very open to discussion on this) improve overall code quality, and make your codebase far easier to navigate and work on.
Updating packages in gx is simple:
$ gx update mypkg QmbH7fpAV1FgMp6J7GZXUV6rj6Lck5tDix9JJGBSjFPgUd
This looks into your package.json
for a dependency named mypkg
and replaces
its hash reference with the one given.
Alternatively, you can just specify the hash you want to update to:
$ gx update QmbH7fpAV1FgMp6J7GZXUV6rj6Lck5tDix9JJGBSjFPgUd
Doing it this way will pull down the package, check its name, and then update that dependency.
Note that by default, this will not touch your code at all, so any references
to that hash you have in your code will need to be updated. If you have a
language tool (e.g. gx-go
) installed, and it has a post-update
hook,
references to the given package should be updated correctly. If not, you may
have to run sed over the package to update everything. The bright side of that
is that you are very unlikely to have those hashes sitting around for any other
reason so a global find-replace should be just fine.
Gx by default will not let you publish a package twice if you haven't updated
its version. To get around this, you can pass the -f
flag. Though this is not
recommended, it's still perfectly possible to do.
To update the version easily, use the gx version
subcommand. You can either set the version manually:
$ gx version 5.11.4
Or just do a 'version bump':
$ gx version patch
updated version to: 5.11.5
$ gx version minor
updated version to: 5.12.0
$ gx version major
updated version to: 6.0.0
Most of the time, your process will look something like:
$ gx version minor
updated version to: 6.1.0
$ gx publish
package whys-awesome-package published with hash: QmaoaEi6uNMuuXKeYcXM3gGUEQLzbDWGcFUdd3y49crtZK
$ git commit -a -m "gx publish 6.1.0"
[master 5c4d36c] gx publish 6.1.0
2 files changed, 3 insertions(+), 2 deletions(-)
To automate this, you can use the release
subcommand. gx release <version>
will automatically do a version update (using the same inputs as the normal
version
command), run a gx publish
, and then execute whatever you have set
in your package.json
as your releaseCmd
. To get the above git commit flow,
you can set it to: git commit -a -m \"gx publish $VERSION\"
and gx will
replace $VERSION
with the newly changed version before executing the git
commit.
You can use a .gxignore
file to make gx ignore certain files during a publish.
This has the same behaviour as a .gitignore
.
Gx also respects a .gitignore
file if present, and will not publish any file
excluded by it.
gx supports named packages via user configured repositories. A repository is simply an ipfs object whose links name package hashes. You can add a repository as either an ipns or ipfs path.
Add a new repo
$ gx repo add myrepo /ipns/QmPupmUqXHBxikXxuptYECKaq8tpGNDSetx1Ed44irmew3
List configured repos
$ gx repo list
myrepo /ipns/QmPupmUqXHBxikXxuptYECKaq8tpGNDSetx1Ed44irmew3
List packages in a given repo
$ gx repo list myrepo
events QmeJjwRaGJfx7j6LkPLjyPfzcD2UHHkKehDPkmizqSpcHT
smalltree QmRgTZA6jGi49ipQxorkmC75d3pLe69N6MZBKfQaN6grGY
stump QmebiJS1saSNEPAfr9AWoExvpfGoEK4QCtdLKCK4z6Qw7U
Import a package from a repo:
$ gx repo import events
gx can support a wide array of use cases by having sane defaults that are extensible based on the scenario you are in. To this end, gx has hooks that get called during certain operations.
These hooks are language specific, and gx will attempt to make calls to a
helper binary matching your language to execute the hooks, for example, when
writing go, gx calls gx-go hook <hookname> <args>
for any given hook.
Currently available hooks are:
post-import
- called after a new package is imported and its info written to package.json.
- takes the hash of the newly imported package as an argument.
post-init
- called after a new package is initialized.
- takes an optional argument of the directory of the newly init'ed package.
pre-publish
- called during
gx publish
before the package is bundled up and added to ipfs. - currently takes no arguments.
- called during
post-publish
- called during
gx publish
after the package has been added to ipfs. - takes the hash of the newly published package as an argument.
- called during
post-update
- called during
gx update
after a dependency has been updated. - takes the old package ref and the new hash as arguments.
- called during
post-install
- called after a new package is downloaded, during install and import.
- takes the path to the new package as an argument.
install-path
- called during package installs and imports.
- sets the location for gx to install packages to.
Gx by default will install packages 'globally' in the global install location
for your given project type. Global gx packages are shared across all packages
that depend on them. The location of this directory is not set in stone, if
for your specific environment you'd like it somewhere else, simply add a hook
to your environments extension tool named install-path
(see above) and gx
will use that path instead. If your language does not set a global install
path, gx will fallback to installing locally as the default. This means that
it will create a folder in the current directory named vendor
and install
things to it.
When running gx install
in the directory of your package, gx will recursively
fetch all of the dependencies specified in the package.json
and save them to
the install path specified.
Gx supports both local and global installation paths. Since the default is
global, to install locally, use --local
or --global=false
. The global flag
is passed to the install-path
hook for your extension code to use in its
logic.
If you want (like me) to use gx as a package manager for go, it's pretty easy. You will need the gx go extensions before starting your project:
$ go get -u github.com/whyrusleeping/gx-go
Once that's installed, use gx like normal to import dependencies. You can import code from the vendor directory using:
import "gx/ipfs/<hash>/packagename"
for example, if i have a package foobar, you can import with gx it like so:
$ gx import QmR5FHS9TpLbL9oYY8ZDR3A7UWcHTBawU1FJ6pu9SvTcPa
And then in your go code, you can use it with:
import "gx/ipfs/QmR5FHS9TpLbL9oYY8ZDR3A7UWcHTBawU1FJ6pu9SvTcPa/foobar"
Then simply set the environment variable GO15VENDOREXPERIMENT
to 1
and run
go build
or go install
like you normally would. Alternatively, install
your dependencies globally (gx install --global
) and you can leave off the
environment variable part.
See the gx-go repo for more details.
If you want to extend gx to work with any other language or environment, you
can implement the relevant hooks in a binary named gx-X
where the 'X' is the
name of your environment. After that, any package whose language is set to 'X'
will call out to that tools hooks during normal gx
operations. For example, a
'go' package would call gx-go hook pre-publish
during a gx publish
invocation before the package is actually published. For more information on
hooks, check out the hooks section above.
No reason. "gx" stands for nothing.
If you're interested in gx, please stop by #gx and #ipfs on freenode irc!
MIT. Jeromy Johnson.