Work in progress manual: Eshell - the emacs shell
There are several shells for Emacs, but none can match the versatility and integration with Emacs like Eshell. Eshell is a shell written entirely in Emacs-Lisp, and it replicates most of the features and commands from GNU CoreUtils and the Bourne-like shells. So by re-writing common commands like ls and cp in Emacs-Lisp, Eshell will function identically on any environment Emacs itself runs on.
- 1 Overview
- 2 Commands
- 3 Command History
- 4 Commandline Interaction
- 5 Redirection
- 6 Argument Predicates
- 7 Eshell smart display
- 8 Bugs
Unlike the other shells in Emacs, Eshell does not inherit from comint-mode, the default mode for interacting with inferior processes in Emacs. But because Eshell is not an inferior process, it does not have to use comint; but while that may seem like a good thing, it does mean that hooks and routines written for comint-mode won’t work with Eshell.
However, almost all the Emacs commands common to comint-mode are reimplemented natively in Eshell — and most share the same keybinds — but there are a few new advances that haven’t been ported over to Eshell, like the spiffy
comint-history-isearch-backward-regexp in Emacs 23.2, bound to M-r.
Eshell works well on any platform Emacs itself runs on, as Eshell interacts with a common middleware (namely the Emacs-Lisp/C source library) and that middleware will in turn communicate with your OS on how to go about copying files and what have you. That middleware support enables Eshell to take advantage of TRAMP as well.
Given Emacs’ UNIX origin, Eshell emulates traditional UNIX shells like bash and the GNU toolchain. This is good news if you are using Windows and cannot be bothered fidgeting with cygwin, or if you require a completely portable Emacs with few or no external dependencies.
And actually, the Windows support in Eshell is a lot better, in many ways, than cygwin’s bash. You do not have the
/cygdrive/c crud to contend with, as Eshell natively supports Windows/MS-DOS drive paths (so
cd D: and
D: both work equally well.)
Despite all the advantages offered by Eshell, there are some points I want to make that seem to confuse some people:
- Eshell is not a terminal emulator. It does not talk to a shell, for it is the shell. Everything it does — from displaying stuff on the screen, to fetching the contents of a directory — it does through Emacs, and Emacs in turn talks to your operating system.
- Because of the way Eshell talks to other processes (asynchronous ones especially) there may be issues with the way it buffers text and how interrupts work.
- Eshell does not support interactive (or “visual” in Eshell parlance) programs, like top, directly; you must tell Eshell to launch them in a separate
- It is not bash or zsh or even csh; do not treat it as such, even though it is heavily inspired by them. To use Eshell effectively you should treat it as if you are using a completely alien shell.
CommandsEshell is capable of invoking almost any elisp function loaded in Emacs. That sort of flexibility is unmatched; there are no shells out there capable of approximating what Eshell can do. In fact, this functionality is heavily used (and encouraged!) by Eshell. If you want to open the file foobar.txt in Emacs you simply invoke
All commands evaluated by Eshell have an evaluation order, which is an ordered list your command must pass through to determine what part of Eshell handles it. If there is nothing on the list that wants to evaluate your command, you will be told your command is invalid.
Assuming you want to execute the command cp, the evaluation order is:
- A full filepath (e.g. /bin/cp) runs cp in /bin
- Look for the command prefix, (default is *), and if it is found then look for the command in the search path.
- Look for a shell-defined alias (alias command)
- Look for cp in the search path, $PATH (or )
- Look for a Lisp function named cp or the elisp function eshell/cp
Eshell has a handful of commands written in Emacs-Lisp that closely emulate a large subset of what the real GNU Coreutils (or your favorite shell) has. Those commands are called "Alias functions."
EShell only implements a subset of the functionality provided by the real commands, but if you pass an unknown argument to Eshell it will defer to the real commandline tool (if it is installed) automatically.
Here’s what Eshell currently re-implements in elisp:
cat, cp, ls, cd, export, dirs, du, echo, env, kill, ln, mkdir, mv, alias, popd, pushd, pwd, rm, rmdir, time, umask.
There is a big emphasis on adhering to the original GNU functionality, so the fact they are emulated is unlikely to cause you any trouble.
Eshell has a cool mechanism where certain commands are intercepted and passed on to Emacs proper. This enables you to invoke a command like man or ls and have Emacs’s built-in man formatter handle it instead. This functionality is especially important for interactive commands (as they will not work properly in Eshell) as Eshell would not be able to call them otherwise.
But where the feature really shines is with complex commands like grep, or diff as Emacs comes with awesome grep and diff tools built in. This feature alone shows the power of Eshell.
The following commands are redirected to Emacs proper:
agrep, diff, egrep, fgrep, glimpse, grep, info, jobs, locate, man, occur, su, sudo, whoami.
The commands su, sudo and whoami are TRAMP aware commands (in Emacs 23.2), so if you are connected to a remote shell they work as expected.
You can use $() to in-line elisp calls and use their output as arguments, in much the same way as you would in bash. The only caveat here is you cannot use the backquote (backtick) to spawn a subshell, but that syntax was never universally supported anyway. It’s also possible (though I would not recommend it, for there are cases where it does not work) to use a standard elisp form like this: (form ...) — so the same as the subshell syntax I explained before, but without the $.
Useful Elisp Commands
Eshell comes with a selection of helper functions that make your day-to-day life just . That, combined with the power to invoke almost any elisp function, means you have incredible flexibility and control over your shell. Some of the commands I’ve listed in the table below were written for Eshell specifically, and the rest are elisp commands I find useful.
I’ve compiled a table of elisp functions (some are made for Eshell; others are not.); then enter the text in that file's own buffer.
|listify ARGS||Parses an argument string into elisp list notation and prints it to the screen. It’s clever enough to handle both MS-DOS/Windows and POSIX-style argument syntax.|
|addpath PATH||Adds the argument, which must be a path, to the $PATH environment variable. If no argument is specified the existing paths are pretty-printed to the screen.|
|unset ENV-VAR||Unsets an existing environment variable|
|find-file FILE||Finds the file FILE and opens it in Emacs. This function is TRAMP aware and will therefore work remotely.|
|dired DIRECTORY||Opens a dired buffer in DIRECTORY.|
|calc-eval EXPR||Runs EXPR through the Emacs calculator.|
|upcase STR/downcase STR||Converts STR to upper- or lowercase.|
|vc-dir DIRECTORY||Reports the status of a version controlled directory (equivalent to the status command in most VCS)|
|ediff-files FILE1 FILE2||Diffs FILE1 and FILE2 using ediff, Emacs’ diff engine.|
Aliasing in Eshell works in much the same way as it does in other mainstream shells, except you can freely mix elisp and Eshell commands. The command alias takes an alias-name and a definition. The definition must be surrounded by single quotes. You can use the usual argument references known from other shells: $1 for the first argument, $2 for the second, …, or $* to use all arguments, or omit them entirely as Eshell will magically append them on to the end of a command if they weren’t referenced in the definition.
alias ff 'find-file $1'
alias d 'dired $1'
To delete an alias, simply leave out the definition argument and it will be removed automagically. To list all the aliases, leave out both arguments.
Eshell will write the alias definitions to eshell-aliases-file, which in turn is governed by the Eshell-directory-name and that put together means your alias file will be put in
~/.Eshell/alias by default. This is done every time you alter an alias.
Another useful thing to know is the auto-correcting aliasing. If you type an invalid command too many times (governed by eshell-bad-command-tolerance, which is 3 by default) Eshell will offer to alias it to its intended command for you. If you don’t like that, you can bump up the aforementioned variable to a large number.
Some commands are too complex to be displayed by Eshell directly, and require special handling. An example would be top, a program that won’t work with a dumb terminal. To support these commands Eshell will run a term session when you invoke a command Eshell considers visual.
To modify the list of visual commands, you can alter eshell-visual-commands, like:
(add-to-list 'eshell-visual-commands "htop")
Eshell comes with a feature-rich command history facility. Because Eshell does not use comint-mode it does not have all the history features available to it, but most of them are reimplemented.
|M-r / M-s||Search backwards or forwards for a command by regexp|
|M-p / M-n||Goes backwards or forwards in the command history list|
|C-p / C-n||Jump to the previous or next command position in Eshell|
|C-c M-r / C-c M-s||Jumps to the previous or next command that shares the command currently used as input. So it jumps to other instances of the command foo if that is the current input.|
Unfortunately, the new-and-improved comint-history-isearch-backward-regexp (bound to M-r in comint) doesn’t work in Eshell because it doesn't inherit from comint (and therefore misses out on upgrades.).
Like bash and other shells, Eshell has support for history modification and interaction. It’s probably easier to refer you to the bash info manual for detailed information on how the history interaction works. I’ve included a small table below that describes most of the history syntax Eshell supports.
|!!||Repeats the last command|
|!ls||Repeats the last command beginning with ls|
|!?ls||Repeats the last command containing ls|
|!ls:n||Extract the nth argument from the last command beginning with ls|
|!ls||Using pcomplete, show completion results matches ls|
|^old^new||Quick substitution. Using the last command, replaceold with new and run it again. Appears to be buggy.|
|$_||Returns the last parameter in the last executed command.|
Eshell also has some support for bash history modifiers (like
!!:s/old/new/) and the bash reference on history interaction would be a good place to brush up on that.
The Eshell Prompt
You can customize the Eshell prompt by modifying eshell-prompt-function, a variable that takes a function that defines what the prompt should contain. By relegating prompt configuration to elisp you can do just about anything you like with it. The only problem is, of course, that Eshell will need to be told what the prompt “looks” like, so you must also edit the variable eshell-prompt-regexp so Eshell knows what the prompt is.
Eshell supports to escape newlines and supports rudimentary multi-line input that way. Another way of doing multi-line literal strings is with single quotes: begin a single quote and hit enter, and you are free to enter text until the closing quote delimiter is encountered. If you use double quotes Eshell will expand subshell commands and do variable expansion.
Due to the way Eshell works, you can even go back and modify the text you entered, in quotes. This is very handy as you can go back and change stuff you don’t like, and get it right the first time.
Eshell comes equipped with a couple of quality-of-life improvements that make interacting with Emacs and Eshell a lot easier.
|C-c M-b||Inserts the printed buffer name at point|
|C-c M-i||Inserts the printed process name at point|
|C-c M-v||Inserts an environment variable name at point|
|C-c M-d||Toggles between direct input and delayed input (send on RET).|
Useful for some programs that don’t work correctly with buffered input.
Redirection in Eshell works in much the same way as it does in other shells. The key difference is that Eshell has to emulate the pseudo-devices as they may not be present (or may not be present in the same form) on platforms such as Windows where
/dev/null is actually NUL.
Another caveat is that Eshell does not support input redirection, though it does support output redirection. To skirt around the lack of input redirection you should use pipes instead.
Redirection to stdout, stdin and stderr work as you would expect, and you can send things to multiple targets as well, which is very nice.
Because Eshell has to reimplement pseudo-devices internally it is not at the mercy of dealing with just UNIX device files — it is actually capable of implementing its own pseudo-devices.
A good example would be redirection to a buffer of your choosing, and that can be done with the following syntax:
$ cat mylog.log >> #<buffer *scratch*>
The keybind I mentioned before C-c M-b will insert the printed name of a buffer.
You can also output straight to an elisp symbol (but be careful you don’t fry the wrong settings):
$ echo foo bar baz > #'myvar $ echo $(cadr myvar) bar
If you set eshell-buffer-shorthand to t you can use the shorthand #'*scratch* instead, but it means you will not be able to redirect straight to elisp symbols.
Eshell reimplements the following pseudo-devices:
|/dev/eshell||Prints the output interactively to Eshell.|
|/dev/null||Sends the output to the NULL device.|
|/dev/clip||Sends the output to the clipboard.|
|/dev/kill||Sends the output to the kill ring.|
The usual redirection rules like overwrite (>) and append (>>) apply here.
To custom virtual targets
You can design your own virtual targets by modifying eshell-virtual-targets, an alist that takes the name of the pseudo-device you want to create, and a function that takes one parameter, mode, that determines if it’s overwrite, append or insert.
Argument predicates are a cool way of quickly filtering lists of files or even elisp lists. The predicate syntax is based on the one used in zsh, so if you are familiar with argument predication in zsh, you can apply most of your knowledge to Eshells’ version.
Unlike most other areas of Eshell, argument predicates are documented in Eshell itself. You can access the help files by typing eshell-display-predicate-help or eshell-display-modifier-help.
Filtering globbed lists of files is very useful, as it saves you the hassle of using tools like find or abusing ls to do your thing.
The help file is fairly spartan and only serves as a simple reference, so I’ve included a small guide here; but actually, the only real way to learn something as flexible as argument predication is simply by trial and error.
I’ve opted not to reprint the sizeable list of predicates and modifiers, as the Eshell manual (see the commands above) do a good enough job of explaining how they work.
Globbing in Eshell follow the same rules as it does in most other common shells: it is the shell that does the expansion of globs and it passes the expanded list of matches on to commands like ls. That’s why when you use find and xargs together it’s critical that you pass
-print0 to find, and
-0 to xargs. If you don’t, filenames with obscure characters or spaces in them may trip up xargs; by using the NUL character as a separator ensures tokenization takes place correctly as the NUL character is an invalid character (along with /) in files.
Eshell’s “lists” are actually elisp lists in their printed form as well as internally. That makes life a lot simpler if you think about it, as Eshell can paw off list handling to elisp, which is something Lisp does well.
Simplest glob example is
echo *, which echos a list of all the wildcard matches in the current directory. Because — as I just mentioned above — wildcard expansion takes place inline, I can immediately apply a modifier to the * wildcard above.
Let’s uppercase the globbed result set:
$ echo *(:U) ("BAR" "BIN/" "DEV/" "ETC/" "FOO" "HOME/" "LIB/" "TMP/" "USR/" "VAR/")
Notice how I used () immediately following the glob pattern. The brackets are what makes argument modifiers or predicates possible. Modifiers are things that modify (big surprise!) the resulting list. Modifier commands always begin with
:, and predicates do not.
Another example, but this time I filter directories using a predicate:
$ echo *(^/) ("bar" "foo")
The circumflex, ^, in this case, like in regular expressions, is negation. The / means “directories” only.
But I don’t have to use globs to apply modifiers or predicates to lists:
$ echo ("foo" "bar" "baz" "foo")(:gs/foo/blarg/) ("blarg" "bar" "baz" "blarg")
This time I replaced all occurrences of foo with blarg. Observe that the syntax is identical, except instead of using globs to get a list of files, I used a list of my own choosing.
The advantages provided by argument predicates and modifiers will greatly reduce commandline clutter as the predicates cover permissions, ownership, file attributes, and much more.
Adding New Modifiers and Predicates
You can even add your own predicates (eshell-predicate-alist) or modifiers (eshell-modifier-alist):
(add-to-list 'eshell-modifier-alist '(?X . '(lambda (lst) (mapcar 'rot13 lst))))
Here I’ve bound X to rot13, the substitution cipher:
$ echo ("foo" "bar" "baz")(:X) ("sbb" "one" "onm")
Eshell smart display
Eshell comes with a pared-down facsimile of Plan 9′s terminal, called the Eshell smart display. The smart display is meant to improve the write-run-revise cycle all commandline hackers go through. It works by not letting the point follow the output of a command you execute, like a normal terminal would. Instead, the point is kept on the line of the command you executed, letting you revise it easily without having to use M-p and M-n or the history modification commands.
Note: In shell-mode, one would use C-c C-r to go to the beginning of the last output and C-c C-p to go on the precedent one.
If smart display is enabled it will also let you review the output of long-running commands by using SPC to move down a page and BACKSPACE to move up a page. If any other key is pressed it will jump the end of the buffer, essentially acting in the same way as if smart display wasn’t enabled.
Essentially, if Eshell detects that you want to review the last executed command, it will help you do so; if, on the other hand, you do not then Eshell will jump to the end of the buffer instead. It’s pretty clever about it, and there are switches you can toggle to fine-tune the behavior.
Where the smart display really shines is that it lets you modify the command you just executed by using the movement keys — like you normally would — to change the command, say to fix a typo or tweak an argument.
The smart display can also be set not to use this extended “edit mode” if the command returns successfully, and without displaying output, like chown for instance. This is how I prefer it.
To enable it put this in your .emacs file:
(require 'eshell) (require 'em-smart) (setq eshell-where-to-jump 'begin) (setq eshell-review-quick-commands nil) (setq eshell-smart-space-goes-to-end t)
If Eshell has already initialized (that is, you’ve already launched an instance of Eshell in Emacs) then evaluating the changes above will not work. You must switch to the Eshell buffer and type
M-: (shell-smart-initialize) (or restart Emacs.)
The smart display is a pretty useful feature and it won’t get in your way once you’re used to it. Simply typing in new commands will make Eshell jump to the end of buffer as if the point was already there.
Can not quit eshell: text is read-only
In that case, we can't exit eshell nor emacs. The way to quit both is to evaluate that bit of elisp with M-:
(let ((inhibit-read-only t)) (kill-this-buffer))}}