peru

Maybe sometimes better than copy-paste.

Peru is a tool for including other people's code in your projects. It fetches from anywhere -- git, hg, svn, tarballs -- and puts files wherever you like. Peru helps you track exact versions of your dependencies, so that your history is always reproducible. And it fits inside your scripts and Makefiles, so your build stays simple and foolproof.

Why?

If you build with make, you don't have to do anything special when you switch branches or pull new commits. Build tools notice those changes without any help. But if you depend on other people's code, the tools aren't so automatic anymore. You need to remember when to git submodule update or go get -u or pip install -r. If you forget a step you can break your build, or worse, you might build something wrong without noticing.

Peru wants you to automate dependency management just like you automate the rest of your build. It doesn't interfere with your source control or install anything global, so you can just throw it in at the start of a script and forget about it. It'll run every time, and your dependencies will never be out of sync. Simple, and fast as heck.

The name "peru", along with our love for reproducible builds, was inspired by Amazon's Brazil build system. It also happens to be an anagram for "reup".

Installation

Peru supports Linux, Mac, and Windows. It requires python (3.5 or later) and git, and optionally hg and svn if you want fetch from those types of repos. Use pip to install it:

pip install peru

Note that depending on how Python is set up on your machine, you might need to use sudo with that, and Python 3's pip might be called pip3. Also, if you have to use Python 3.3 or 3.4, those were supported up to peru 1.1.4.

On Arch, you can install peru from the AUR.

Getting Started

Here's the peru version of the first git submodules example from the Git Book. We're going to add the Rack library to our project. First, create a peru.yaml file like this:

imports:
    rack_example: rack/  # This is where we want peru to put the module.

git module rack_example:
    url: git://github.com/chneukirchen/rack.git

Now run peru sync.

What the heck just happened?

Peru cloned Rack for you, and imported a copy of it under the rack directory. It also created a magical directory called .peru to hold that clone and some other business. If you're using source control, now would be a good time to put these directories in your ignore list (like .gitignore). You usually don't want to check them in.

Running peru clean will make the imported directory disappear. Running peru sync again will make it come back, and it'll be a lot faster this time, because peru caches everything.

Getting Fancy

For a more involved example, let's use peru to manage some dotfiles. We're big fans of the Solarized colorscheme, and we want to get it working in both ls and vim. For ls all we need peru to do is fetch a Solarized dircolors file. (That'll get loaded somewhere like .bashrc, not included in this example.) For vim we're going to need the Solarized vim plugin, and we also want Pathogen, which makes plugin installation much cleaner. Here's the peru.yaml:

imports:
    # The dircolors file just goes at the root of our project.
    dircolors: ./
    # We're going to merge Pathogen's autoload directory into our own.
    pathogen: .vim/autoload/
    # The Solarized plugin gets its own directory, where Pathogen expects it.
    vim-solarized: .vim/bundle/solarized/

git module dircolors:
    url: https://github.com/seebi/dircolors-solarized
    # Only copy this file. Can be a list of files. Accepts * and ** globs.
    pick: dircolors.ansi-dark

curl module pathogen:
    url: https://codeload.github.com/tpope/vim-pathogen/tar.gz/v2.3
    # Untar the archive after fetching.
    unpack: tar
    # After the unpack, use this subdirectory as the root of the module.
    export: vim-pathogen-2.3/autoload/

git module vim-solarized:
    url: https://github.com/altercation/vim-colors-solarized
    # Fetch this exact commit, instead of master.
    rev: 7a7e5c8818d717084730133ed6b84a3ffc9d0447

The contents of the dircolors module are copied to the root of our repo. The pick field restricts this to just one file, dircolors.ansi-dark.

The pathogen module uses the curl type instead of git, and its URL points to a tarball. (This is for the sake of an example. In real life you'd probably use git here too.) The unpack field means that we get the contents of the tarball rather than the tarball file itself. Because the module specifies an export directory, it's that directory rather than the whole module that gets copied to the import path, .vim/autoload. The result is that Pathogen's autoload directory gets merged with our own, which is the standard way to install Pathogen.

The vim-solarized module gets copied into its own directory under bundle, which is where Pathogen will look for it. Note that it has an explicit rev field, which tells peru to fetch that exact revision, rather than the default branch (master in git). That's a Super Serious Best Practice™, because it means your dependencies will always be consistent, even when you look at commits from a long time ago.

You really want all of your dependencies to have hashes, but editing those by hand is painful. The next section is about making that easier.

Magical Updates

If you run peru reup, peru will talk to each of your upstream repos, get their latest versions, and then edit your peru.yaml file with any updates. If you don't have peru.yaml checked into some kind of source control, you should probably do that first, because the reup will modify it in place. When we reup the example above, the changes look something like this:

diff --git a/peru.yaml b/peru.yaml
index 15c758d..7f0e26b 100644
--- a/peru.yaml
+++ b/peru.yaml
@@ -6,12 +6,14 @@ imports:
 git module dircolors:
     url: https://github.com/seebi/dircolors-solarized
     pick: dircolors.ansi-dark
+    rev: a5e130c642e45323a22226f331cb60fd37ce564f

 curl module pathogen:
     url: https://codeload.github.com/tpope/vim-pathogen/tar.gz/v2.3
     unpack: tar
     export: vim-pathogen-2.3/autoload/
+    sha1: 9c3fd6d9891bfe2cd3ed3ddc9ffe5f3fccb72b6a

 git module vim-solarized:
     url: https://github.com/altercation/vim-colors-solarized
-    rev: 7a7e5c8818d717084730133ed6b84a3ffc9d0447
+    rev: 528a59f26d12278698bb946f8fb82a63711eec21

Peru made three changes:

• The dircolors module, which didn't have a rev before, just got one. By default for git, this is the current master. To change that, you can set the reup field to the name of a different branch.
• The pathogen module got a sha1 field. Unlike git, a curl module is plain old HTTP, so it's stuck downloading whatever file is at the url. But it will check this hash after the download is finished, and it will raise an error if there's a mismatch.
• The vim-solarized module had a hash before, but it's been updated. Again, the new value came from master by default.

At this point, you'll probably want to make a new commit of peru.yaml to record the version bumps. You can do this every so often to keep your plugins up to date, and you'll always be able to reach old versions in your history.

Commands

• sync
  • Pull in your imports. sync yells at you instead of overwriting existing or modified files. Use --force/-f to tell it you're serious.
• clean
  • Remove imported files. Same --force/-f flag as sync.
• reup
  • Update module fields with new revision information. For git, hg, and svn, this updates the rev field. For curl, this sets the sha1 field. You can optionally give specific module names as arguments.
• copy
  • Make a copy of all the files in a module. Either specify a directory to put them in, or peru will create a temp dir for you. You can use this to see modules you don't normally import, or to play with different module/rule combinations (see "Rules" below).
• override
  • Replace the contents of a module with a local directory path, usually a clone you've made of the same repo. This lets you test changes to imported modules without needing to push your changes upstream or edit peru.yaml.

Module Types

git, hg, svn

For cloning repos. These types all provide the same fields:

• url: required, any protocol supported by the underlying VCS
• rev: optional, the specific revision/branch/tag to fetch
• reup: optional, the branch/tag to get the latest rev from when running peru reup

The git type also supports setting submodules: false to skip fetching git submodules. Otherwise they're included by default.

curl

For downloading a file from a URL. This type is powered by Pythons's standard library, rather than an external program.

• url: required, any kind supported by urllib (HTTP, FTP, file://)
• filename: optional, overrides the default filename
• sha1: optional, checks that the downloaded file matches the checksum
• unpack: optional, tar or zip

Peru includes a few other types mostly for testing purposes. See rsync for an example implemented in Bash.

Creating New Module Types

Module type plugins are as-dumb-as-possible scripts that only know how to sync, and optionally reup. Peru shells out to them and then handles most of the caching magic itself, though plugins can also do their own caching as appropriate. For example, the git and hg plugins keep track of repos they clone. Peru itself doesn't need to know how to do that. For all the details, see Architecture: Plugins.

Rules

Some fields (like rev and unpack) are specific to certain module types. There are also fields you can use in any module, which modify the tree of files after it's fetched. Some of these made an appearance in the fancy example above:

• copy: A map or multimap of source and destination paths to copy. Works like cp on the command line, so if the destination is a directory, it'll preserve the source filename and copy into the destination directory.
• move: A map or multimap of source and destination paths to move. Similar to copy above, but removes the source.
• drop: A file or directory, or a list of files and directories, to remove from the module. Paths can contain * or ** globs.
• pick: A file or directory, or a list of files and directories, to include in the module. Everything else is dropped. Paths can contain * or ** globs.
• executable: A file or list of files to make executable, as if calling chmod +x. Also accepts globs.
• export: A subdirectory that peru should treat as the root of the module tree. Everything else is dropped, including parent directories.

Note that these fields always take effect in the order listed above, regardless of the order they're given in peru.yaml. For example, a move is always performed before a pick. Also note that these fields can't be given twice. For example, instead of using two separate move fields (one of which would be ignored), use a single move field containing multiple moves. In practice, things work this way because these fields are parsed as keys in a dictionary, which don't preserve ordering and can't repeat.

Besides using those fields in your modules, you can also use them in "named rules", which let you transform one module in multiple ways. For example, say you want the asyncio subdir from the Tulip project, but you also want the license file somewhere else. Rather than defining the same module twice, you can use one module and two named rules, like this:

imports:
    tulip|asyncio: python/asyncio/
    tulip|license: licenses/

git module tulip:
    url: https://github.com/python/asyncio

rule asyncio:
    export: asyncio/

rule license:
    pick: COPYING

As in this example, named rules are declared a lot like modules and then used in the imports list, with the syntax module|rule. The | operator there works kind of like a shell pipeline, so you can even do twisted things like module|rule1|rule2, with each rule applying to the output tree of the previous.

Recursion

If you import a module that has a peru file of its own, peru can include that module's imports along with it, similar to how git submodules behave with git clone --recursive. To enable this, add recursive: true in a module's definition.

It's also possible to directly import modules that are defined in the peru.yaml file of another module. If your project defines a module foo, and foo has a peru file in it that defines a module bar, you can use foo.bar in your own imports. This works even if you never actually import foo, and it does not require setting recursive: true.

Configuration

There are several flags and environment variables you can set, to control where peru puts things. Flags always take precedence.

• --file=<file>: The path to your peru YAML file. By default peru looks for peru.yaml in the current directory or one of its parents. This setting tells peru to use a specific file. If set, --sync-dir must also be set.
• --sync-dir=<dir>: The path that all imports are interpreted relative to. That is, if you import a module to ./, the contents of that module go directly in the sync dir. By default this is the directory containing your peru.yaml file. If set, --file must also be set.
• --state-dir=<dir>: The directory where peru stashes all of its state metadata, and also the parent of the cache dir. By default this is .peru inside the sync dir. You should not share this directory between two projects, or peru sync will get confused.
• --cache-dir=<dir> or PERU_CACHE_DIR: The directory where peru keeps everything it's fetched. If you have many projects fetching the same dependencies, you can use a shared cache dir to speed things up.
• --file-basename=<name>: Change the default peru file name (normally peru.yaml). As usual, peru will search the current directory and its parents for a file of that name, and it will use that file's parent dir as the sync dir. Incompatible with --file.

Links

• Discussion and announcements (Google Group)
• Architecture doc
• Using peru with make