Activating environments

Environments are a central concept in Flox, representing the tools you want to use, all of their dependencies, various environment variables necessary to make them function properly, and all of your customizations. Given that environments are such an important part of Flox, it stands to reason that how you use them is also an important part of Flox.

There are three different ways to use an environment, and two different modes that an environment can be activated in. At the end of the day, though, it all boils down to properly configuring a shell. The hook and profile scripts specified in your manifest are run as part of configuring that shell. Understanding when and why they're run will help you take full advantage of Flox, so let's walk through what it means to "activate" an environment and how it works.

Configuring the shell

When you "activate" an environment, Flox configures a shell, making all of the packages and environment variables specified in your manifest available, as mentioned above.

The most basic way to activate an environment is simply by calling flox activate, which puts you into a subshell with everything configured:

$ flox activate
flox [myenv] $ # Now you can use your packages

One of the core features that makes Flox so attractive for development is that the packages in your manifest are available when the environment is active, and they're gone when it's inactive. We do this by carefully setting a collection of environment variables, some of which you may be familiar with, such as PATH, and others which you may not have heard of like ACLOCAL_PATH, RUST_SRC_PATH, and others.

As an example, if you create an environment myenv and install hello, Flox will place <path to myenv>/.flox/run/<your system>.myenv.dev/bin at the beginning of your PATH variable so that hello will be selected from your environment rather than from elsewhere on your system.

Three different ways to activate

We mentioned above that there are three different ways to use an environment.

Subshell

We've already mentioned the first method, which is to put you into a subshell. When you activate this way your existing shell is paused and you're put into a new one configured by Flox. Once this shell exits (via exit or Ctrl-D, for example), your original shell is resumed and your are put back into it.

In-place

This method is similar to the first in that you're still in an interactive shell, but in this case it's the original shell.

To make this happen you could do one of these options in Bash:

# Option 1
source <(flox activate)

# Option 2
eval "$(flox activate)"

In both cases Flox emits a script that configures the shell, and the shell executes that code to configure itself.

In order to configure a default environment that's activated for every new shell, you would put a line like this in your .bashrc, .zshrc, .tcshrc, or config.fish. You could do this manually, but Flox will also prompt you to do it for you the first time you attempt to install a package in a directory without an environment and with no environments currently active.

Command

Sometimes you just want to run a command in the context of your environment, maybe because you have some tools available in your environment that aren't available outside the environment.

You could do this in a subshell: - Enter the subshell via flox activate - Run the command - Exit the subshell via exit

You could do a similar thing with an in-place activation: - Configure your existing shell with source <(flox activate). - Run the command - Your shell is still configured by Flox when you're done

That's a lot of ceremony to run that one command though, and the in-place activation leaves the environment still activated in your shell, which you may not want.

The easy way to do this is:

$ flox activate -- <your command>

This starts a Flox-configured subshell, runs your command, and immediately exits to put you back into your shell.

Activation flow

In order to understand where hook and profile fit into the picture, we need to explore the timeline of what an activation looks like. Much of this is dictated by what gets inherited when you create subshells. Don't feel like you need to understand this entirely in order to use Flox, it's just here to help you if you want a deeper understanding.

What's inherited by a subshell?

When you create a subshell, you create a new process that happens to be a shell. A new process inherits the environment of its parent process by default, meaning that it inherits all of the environment variables set by that process. If the parent process is a shell, the functions and aliases defined in the parent process (shell) are not passed down to the subshell (unless you use special options like export -f). This means that if we want you to be able to define functions and aliases to be used by your shell in your environment, we have to make your shell source their definitions.

Timeline

In order to meet our constraints and user experience goals, we activate an environment in a number of steps. The steps for a subshell activation are shown in the diagram below, and they're very similar for the other types of activation. Let's break it down step by step.

For a variety of reasons it's convenient to have the same process ID (PID) throughout the lifecycle of the activated environment. The way you tell the current process to run a different program is via the exec command (which calls the exec system call).

flox activate execs a Bash subshell with a script that's bundled with Flox, and sets some environment variables to be present in that Bash subshell. We use this Bash subshell to prepare the way for putting you into a configured instance of your shell of choice. In the diagram above your shell of choice is represented with the FLOX_SHELL variable, which is also the variable you can use to override which shell Flox uses when you activate an environment.

As part of this activation script, Flox runs some initial setup, then sets the variables you've provided in the [vars] section of your manifest. Next, the script sources the hook.on-activate script that you've provided in the [hook] section of your manifest. Since this script is run by the Bash shell we're using, you only need to worry about the synax and oddities of one shell when writing this script. This is convenient, but it comes with some tradeoffs.

This Bash subshell is non-interactive. Some programs behave differently when they execute in a shell that's not interactive, and most shells will not (by default) expand aliases when run non-interactively. Also, remember that any functions or aliases you define in this Bash subshell via the hook.on-activate script will not be inherited by your shell later.

Finally, we exec your shell with some overrides that allow us to inject our own configuration, such as sourceing the scripts defined in the [profile] section of your manifest. Once those scripts have been sourced, we hand control back to you. Since the scripts in the [profile] section are sourced by your shell, this is where you can define aliases and functions that you'd like to be available in your activated environment.

hook vs. profile in a nutshell

So, that was a lot of technical detail. To make life easier for you when it comes to writing scripts for your environment, here is some simple guidance:

hook.on-activate:

• Always Bash, so there's only one shell syntax and set of quirks to keep in mind.
• Can't use aliases.
• Can define functions to use within the hook, but they won't be passed down to other shells.
• Can define environment variables that need to be computed.

profile:

• Syntax depends on the shell.
• Can define functions and aliases.
• Can source scripts needed for other programs to work properly e.g. the activate script for a Python virtual environment.
• Can define environment variables that need to be computed.

In short, it's probably best to put as much as you can in hook.on-activate until you have shell-specific needs, you need aliases, or you need to source a third-party script into your shell.

Attaching

Everything we've discussed so far is about what happens when you start a new activation of an environment. However, if you're simply activating a second instance of an environment, all of the setup done in your hook and script will already have been done, so you would be doing the same exact thing. In addition, you would be setting all of the same exact environment variables from your [vars] [hook] sections as before.

In short, this would be doing a lot of pointless work. For that reason, we record those environment variables and apply them to subsequent activations rather than computing them again. We call this "attaching" to an activation, and we do it automatically for you to make activation as fast as possible.

If you edit your manifest and activate in a new shell, the whole activation process is run again and subsequent activations will attach to this new version of the environment.

Development vs. runtime mode

In addition to the different ways to use an activation, there are two different "modes" to activate in: development mode and runtime mode.

The default mode at the moment is development mode. In development mode a package and all of its development dependencies are made available. As the name implies, this is useful at development time. However, this may causes unexpected failures when layering environments or when activating an environment system-wide.

For these reasons we also provide "runtime" mode, which simply puts the requested packages in PATH (and makes their man pages available). This behavior is more in line with what you would expect from a system-wide package manager like apt, yum, or brew.

Conclusion

As you can see, there's a lot going on under the hood, but at the end of the day it's just Unix fundamentals: processes and environment variables.

This is what makes activating a Flox environment so fast. There's no container to build and there's no VM to boot up. It really is a return to basics, with our own special twist on it.

Hopefully after reading this you have a deeper understanding of how a Flox environment is activated, and you feel confident that you can write [hook] and [profile] scripts that prepare your environment just how you like them.