**exa** is a modern replacement for the command-line program ls that ships with Unix and Linux operating systems, with more features and better defaults. It uses colours to distinguish file types and metadata. It knows about symlinks, extended attributes, and Git. And it’s **small**, **fast**, and just one **single binary**.
- Valid sort fields are **accessed**, **changed**, **created**, **extension**, **Extension**, **inode**, **modified**, **name**, **Name**, **size**, **type**, and **none**. Fields starting with a capital letter sort uppercase before lowercase. The modified field has the aliases **date**, **time**, and **newest**, while its reverse has the aliases **age** and **oldest**.
- Valid time fields are **modified**, **changed**, **accessed**, and **created**.
exa is written in [Rust](http://www.rust-lang.org). You will need rustc version 1.17.0 or higher. The recommended way to install Rust is from the official download page.
If you intend to compile for musl you will need to use the flag vendored-openssl if you want to get the Git feature working: `cargo build --release --target=x86_64-unknown-linux-musl --features vendored-openssl,git`
`exa` is also installable through [the derivation](https://github.com/NixOS/nixpkgs/blob/master/pkgs/tools/misc/exa/default.nix) using the [nix package manager](https://nixos.org/nix/) by running:
exa uses [Vagrant][] to configure virtual machines for testing.
Programs such as exa that are basically interfaces to the system are [notoriously difficult to test][testing]. Although the internal components have unit tests, it’s impossible to do a complete end-to-end test without mandating the current user’s name, the time zone, the locale, and directory structure to test. (And yes, these tests are worth doing. I have missed an edge case on more than one occasion.)
The initial attempt to solve the problem was just to create a directory of “awkward” test cases, run exa on it, and make sure it produced the correct output. But even this output would change if, say, the user’s locale formats dates in a different way. These can be mocked inside the code, but at the cost of making that code more complicated to read and understand.
An alternative solution is to fake *everything*: create a virtual machine with a known state and run the tests on *that*. This is what Vagrant does. Although it takes a while to download and set up, it gives everyone the same development environment to test for any obvious regressions.
The first command downloads the virtual machine image, and then runs our provisioning script, which installs Rust, exa’s dependencies, configures the environment, and generates some awkward files and folders to use as test cases. This takes some time, but it does write to output occasionally. Once this is done, you can SSH in, and build and test:
host$ vagrant ssh
vm$ cd /vagrant
vm$ cargo build
vm$ ./xtests/run
All the tests passed!
### Running without Vagrant
Of course, the drawback of having a standard development environment is that you stop noticing bugs that occur outside of it. For this reason, Vagrant isn’t a *necessary* development step — it’s there if you’d like to use it, but exa still gets used and tested on other platforms. It can still be built and compiled on any target triple that it supports, VM or no VM, with `cargo build` and `cargo test`.
