GNU ELPA - buffer-env

When called interactively, buffer-env-update asks for a script file and executes it (in the sense detailed below). The role of the script is to set some environment variables. Then the Emacs variables process-environment and exec-path are made buffer-local and their values are set so as to replicate the environment defined by the script.

When buffer-env-update is called from a hook, a file named buffer-env-script-name is looked up in the current directory or one of its parents. If found, the same procedure as in the interactive case takes place. Otherwise, nothing happens.

It remains to clarify what “executing a script” means in the paragraphs above. Normally, it simply means to execute the script as a shell script and collect all the exported variables. However, certain script names are treated specially. These are:

• .env: These files, used by Docker, Node.js and others, are simple lists of VARIABLE=value pairs. They are still executed as shell scripts (which dictates when and how quotes are to be used, for instance), but no export statements are needed.
• guix.scm and manifest.scm: The development environment of the Guix package is loaded and exported to Emacs. Make sure you have entered guix shell at least once before to install the dependencies, otherwise you may block Emacs for a long time.
• flake.nix and shell.nix: These files are used by the Nix package manager and are handled similarly to Guix.
• pyproject.toml: If you are using Poetry, Hatch or PDM, buffer-env can infer the project environment from them.
• *.ps1: Similar to a regular shell script, but interpreted by PowerShell.

For instructions on how to extend this list, see the documentation of the variable buffer-env-command-alist.

Most Emacs packages are not written with the possibility of a buffer-local process environment in mind. This leads to issues with a few commands; specifically, those which start an external process after switching to a different buffer or remote directory. Examples include:

• compile and project-compile (C-x p c) in Emacs 27 and older,
• async-shell-command (M-&).

Fortunately, the problem has an easy fix provided by the inheritenv package, which see.

Alternatively, if you speak Elisp and want to keep your configuration lean, you can just copy the function below and apply it as an :around advice to any affected commands.

(eval-when-compile (require 'cl-lib))
(defun buffer-env-inherit (fn &rest args)
  "Call FN with ARGS using the buffer-local process environment.
Intended as an advice around commands that start a process after
switching buffers."
  (cl-letf (((default-value 'process-environment) process-environment)
	    ((default-value 'exec-path) exec-path))
    (apply fn args)))

This package is essentially a knockoff of the envrc package by Steve Purcell. The main difference is that envrc depends on and tightly integrates with the direnv program, while buffer-env is minimalist and has no extra dependencies.

For a comparison of the buffer-local approach to environment variables with the global approach used by most of the similar packages, see envrc's design notes.

There is a large number of Emacs packages interfacing with the Python virtualenv system. They all seem to take the global approach and, therefore, the comparisons and caveats in the envrc design notes also apply, mutatis mutandis.

Discussions, suggestions and code contributions are welcome! Since this package is part of GNU ELPA, contributions require a copyright assignment to the FSF.