Crates.io | lorevault |
lib.rs | lorevault |
version | 1.4.0 |
source | src |
created_at | 2024-09-15 08:41:10.753371 |
updated_at | 2024-11-17 16:00:43.858737 |
description | A simple program to create a directory from a recipe. |
homepage | |
repository | https://github.com/JanNeuendorf/lorevault |
max_upload_size | |
id | 1375327 |
size | 187,560 |
Lorevault is a simple program that creates a directory from a declarative configuration file.
When I ran that test ten minutes ago, did I forget to to delete the old log files? Is that why it failed?
-- Me, every five minutes
The main motivation for this project is to define directories in a way that can be made completely reproducible. This, of course, could also be done by copying a reference directory or cloning a git-repo. There are a few problems with this:
To combat those problems, we can use:
While you can be pedantic, you do not have to be, so you can use this for simple templates.
This can also be used to manage your dotfiles. (skip here)
You can install the latest release
cargo install --git https://github.com/JanNeuendorf/lorevault
or the latest version from this repository
cargo install --git https://github.com/JanNeuendorf/lorevault
Then run
lorevault example
to get a basic example.
The command:
lorevault sync config.toml targetdir --tags=tag1,tag2
creates the directory at targetdir
according to the recipe.
The directory is always deleted and recreated. This ensures that there are no subtle changes that can be missed. If the directory existed before, it is used as a reference. If a file has a defined hash and the file in the directory matches it, it can be taken from there.
Other commands are:
Usage: lorevault <COMMAND>
Commands:
sync Sync to a specified directory
clean Remove files controlled by corresponding sync operation
config Shortcut for syncing to ~/.config with -S
example Writes out an example configuration file
hash Prints the SHA3-256 hash of a file
tags Lists all the tags defined in the file
list Lists all the files that would be in the directory
show Shows the contents of a single source (as utf8)
help Print this message or the help of the given subcommand(s)
The configuration file can be read in from a local or remote git-repo with the syntax repo#id:path
.
It does not have to be stored in your project's directory.
The config file is a .toml
file that consists of a list of file descriptions.
We might include individual files in our directory. Here is an example:
[[file]]
path = "my_subdir/my_file.txt"
hash = "741C077E70E4869ADBC29CCC34B7935B58DDAC16A4B8007AC127181E2148F468"
tags = ["tag1","tag2"]
sources=["/some/local/file.txt","repo#id:path/to/file.txt"]
The first variable, path
defines where the file will be located in the target folder.
The directory my_subdir
will be created automatically.
Here, we specified the optional SHA3-256
hash of the file. This has two advantages: we get an error whenever we are trying to load a file with a wrong hash and we might avoid downloading files if the file already matches the hash.
We can specify a list of tags. The file will then only be included if at least one of the tags is activated. It will replace untagged files at the same path.
The last line specifies a list of possible sources for the file. The list is checked in order, so a local copy should be listed first.
There are several kinds of sources:
You can give a path to a local file, but it must be an absolute path. This ensures that it does not matter, where our config file is stored and from where we call the sync
command.
Using the syntax repo#id:path
, we can load files from git repositories. They can be local, in which case the path to the repository must be absolute, or remote.
Remote repos are cloned to a cache directory that persists until the end of the process. This ensures that the same repo is not cloned multiple times.
The remote path can be ssh: user@machine:repo.git#id:path
or http: https://website.com/repo.git#id:path
.
Authentication for cloning repos is handled by auth-git2-rs, so you can clone private repos with the correct ssh-key. If and how the key is unlocked is up to the user's machine.
The id
can be a commit hash, a tag or a branch. When a branch is specified, we get the latest commit to that branch.
Technically, the repos are not cloned but mirrored. This preserves other branches and their tags, but it is slow. To speed things up, one should add a local clone of the repository to the list of sources.
Submodules are not supported!
You can give a URL starting with http
or https
. It must return a file-response and there is no support for authentication or caching.
The syntax user@machine:some/file
loads the file over sftp. The default port is 22.
We can specify the contents of the file as text. For this, we need a slightly different .toml
syntax:
[[file]]
path="my_file.txt"
[[file.source]]
type = "text"
content = """
Hello,
this file was written with Lorevault.
"""
The other sources can be written in this way too.
We might want to include a file with a slight modification. It would be unfortunate if we had to store the edited copy, especially if we have multiple sources for the original. If the file's content is an utf8-encoded string, we can make small edits like this:
[[file]]
path = "my_dotfile.in"
hash = "741C077E70E4869ADBC29CCC34B7935B58DDAC16A4B8007AC127181E2148F468"
sources=["/some/path","repo#id:path"]
[[file.edit]]
type="insert"
content="# The document begins\n\n"
position="prepend" # could be "append" or after a line number.
[[file.edit]]
type="replace"
from="setting=false"
to="setting=true"
tags=["flip"] # Will be skipped if the tag is not active.
[[file.edit]]
type="delete"
start=30 # line numbers (inclusive)
end=100
The hash always refers to the hash before any edits are made. Line numbers are counted from 1. The edits are made in sequence, so the line numbers change.
We can include entire directories
[[directory]]
path="my_included_directory"
count=5 # optional
sources=["/path/to/dir","repo#id:path/to/dir"]
ignore_hidden=false # This is the default
tags = ["tag1","tag2"]
This will try to list the directory and copy all contents to the new directory at path
.
While the directory can be nested, it can not contain any objects that are not files. This includes empty directories.
We have the option to specify the expected number of files as a check. The possible sources are local directories and directories in git repos. They work the same as for single files.
The first working source is used for listing the directory and fetching the files.
In practice, the directory is expanded and the files are added to the list of files individually.
(We can also specify hashes for directories. They consist of the hash of the filepaths followed by the hashes of the individual files. This should not be set manually but only with the lock
subcommand. While identical hashes for the filepaths guarantee reproducible results, sources with identical results can give different hashes.
)
To avoid repetition, variables can be set at the beginning of the file and used in the following way:
var.user = "you"
var.mypath = "subdir/for/{{user}}"
[[file]]
path = "{{mypath}}/file.txt"
[[file.source]]
type = "text"
content = "This file was written by {{user}}."
ignore_variables=false # This is the default. If true, the text is protected.
They can not be used inside hashes, tags, types or editing positions.
We can include other configuration files.
[[include]]
config="/path/to/included.toml" # Can be repo#id:path
subdir="files/go/here" # Defaults to directory root.
required_tags=["tag1"] # If not set, the file will not be included.
with_tags=["tag2"] # Will be passed to the other file.
# We can specify the hash of the included `.toml` file itself.
hash = "741C077E70E4869ADBC29CCC34B7935B58DDAC16A4B8007AC127181E2148F468"
# We can ensure that the loaded config must be locked.
enforce_locked = true # default false
Variables are not shared between files. Tags for included files can only be activated in the way shown above and are not influenced by the tags activated on the CLI.
The behavior should be the same as building the directory with the required tags first and then including it.
There is currently no check for cyclic dependencies.
We can specify tags that are activated by default. For this, we just need to put:
default=["some_tag","some_other_tag"]
in the configuration file.
If we then want to deactivate the tag, we can do it with an exclamation mark:
lorevault sync myconf.toml mydir -t '!some_tag'
Note that some shells require single quotes to prevent !
to be read as a special character.
To avoid confusion, tags can not start with !
or be called default
.
If we include a .toml
file, its default tags are active unless they are deactivated with
with_tags=["!my_tag"]
In general, relative paths are not allowed inside config files.
It might, however, be useful to refer to data stored together with the config. This is especially true if the config is inside a repository.
For this, we can use built-in variables.
If the config file is read from a git-repo, the variables
SELF_REPO
and SELF_ID
are set automatically.
If it is a local file, SELF_PARENT
is set.
SELF_ROOT
gives either repo#id:
or the parent directory.
It is therefore a good convention to put the config file in the root of the project, regardless of whether the project is a git-repo or just a local directory.
Here is an example:
project/
│
├─── config.toml
│
└─── data/
└── file.txt
In config.toml
:
[[file]]
path = "new/filename.txt"
sources=["{{SELF_ROOT}}/data/file.txt"]
If the config file is referred to as repo#commit:config.toml
(from the CLI or by inclusion in another config),
the contents of new/filename.txt
will match the state of data/file.txt
at the time of that commit.
If it is referred to with a path, it is the current version in the directory.
We might want to include files with secret contents in our directory. One way to do that is to use lorevault to fetch the encrypted files and then decrypt them with a script. For convenience, lorevault has build-in support for age (a tool and format for file-encryption). The Rust-implementation of age used here is not yet stable and in general this should not be used in situations where there is the possibility of an advanced attack.
Here is an example of how to include an encrypted file:
[[file]]
path="decrypted.txt"
sources=["/path/to/encrypted.age"]
hash = "741C077E70E4869ADBC29CCC34B7935B58DDAC16A4B8007AC127181E2148F468"
decrypt="agev1"
The hash always refers to the encrypted file not the decrypted one. This means that the file will always be regenerated.
In order for this to work, we need to provide the path to a private key.
lorevault sync config.toml targetdir -i /path/to/key
We can provide multiple key-files (containing multiple keys) and all keys will be tried on all encrypted files. Currently this only supports the original key format of age (no ssh-keys).
Sometimes we do not want to control the entire directory. A good example might be managing dotfiles in ~/.config
.
Resetting the entire directory is probably not what we want. Maybe the configuration files for some programs are managed in some other way.
To only update parts of the directory we can use:
lorevault sync -S config.toml target_dir
where the -S
stands for skip first level.
This will preserve paths that differ from the controlled files at the first level.
Great, what does that mean? It means that if your config.toml
creates a file in a specific subdirectory (directly or by import)
this subdirectory is deleted and recreated according to the config. If no such file exists, the subdirectory (or file) is left as it was.
Let's walk through an example:
The config file defines the following files (as can be seen with the list
subcommand).
subdir1/subsubdir/file1.txt
file2.txt
The directory currently looks like this:
target_directory/
│
├─── file2.txt
│
├─── file3.txt
│
├─── subdir1/
│ └── file
│
└───subdir2/
└── file
What will happen when running sync -S
is the following:
subdir1
is defined in the config file. Therefore, the program assumes that we want to control the entire folder subdir1
. It is completely replaced. It does not matter that we only defined a single file in a subdirectory.file2.txt
is also defined. Therefore the file is replaced.file3.txt
is not defined and there is no file starting with subdir2/
. These paths are not deleted or changed.Unless we use the -Y
option, we will get a list of all controlled paths for confirmation.
On linux you can use the subcommand
lorevault config config.toml
This will find ~/.config
and sync to it with the -S
option.
Of course this can also be used to load someone elses dotfiles if they host a lorevault file on their git.
Especially if we use this in a script, we might want to undo the sync operation.
The subcommand clean
takes the same arguments as sync
and it deletes all paths that were synced.
If we did not pass the -S
option, the operation can be undone by removing the directory, so that is all that it does. If -S
is used however, it finds all the paths that the corresponding sync
command would have altered and deletes them.
There is one potential risk when using this command: the list of paths controlled by the config file might have changed since the sync
command was run. This could have happened for three reasons:
.toml
file itself has changed.Issues can be avoided by not referring to local files or directories and by not using git-IDs like branch-name
or HEAD
, which can change.
A config file is called locked when all of the following requirements are true:
enforce_locked
set to trueThis does not mean that the file actually works, but it guarantees that if it works, it always produces the same output.
We can enforce that the file is locked by passing the --locked
flag when syncing the directory.
We can use:
lorevault lock configfile.toml
This will try to fill in missing hashes.
You can look at the contents of a single file with
lorevault show some/file
This simply prints the contents of the file to standard output.
The advantage is that it also works with repo#id:path/to/file
and user@machine:path/to/file
,
so it can be used in scripts where you don't know what kind of source will be used.
It assumes that the contents are utf-8
encoded.
If you want to write binary contents to a file, use the -o
option instead of a pipe.
It only works on Unix systems. (Only tested on Linux.)
The contents of the directory are created in memory, so very large files are to be avoided.
There is no control over metadata/permissions.
All contributions are very welcome, but most of all this project needs testing.
There are a few tests in the justfile
to get started.
It is, however, very hard to test alone.
I am thankful for every bug report.