Babblings of an aging geek in love with the Absurd, his family, and his own hubris.... oh, and Lisp.

Eschewing Zshell for Emacs Shell

I’ve done it. I’m now done with Zshell and Fish and Bash and all of those guys…mostly. While they all have nice features, I find that what drives my workflow is my editor. I start in Emacs, and then pop over to shell for file manipulation and whatnot. I don’t start in the shell and move around and then edit files.

Most Emacs users split the Emacs window and start a shell inside Emacs, and pop over and back to that window. However, I’m finding that Emacs’ eshell to be a better fit, for the more I use it, the better I like it.

It’s problem is the lack of documentation…and a bit of misunderstanding. Hence this essay. Before I get started, I would like to put a plug for Mickey Petersen’s new book, Mastering Emacs, as he has an excellent chapter on mastering the eshell (which happens to be free).


A shell is a command-driven REPL. You type in a command and view the results, type in another command…rinse and repeat. Fine until you type something you expect to have a few lines, but if it responds with hundreds of lines, you redo the command and pipe it through less.

But with EShell, you don’t need to bother with a pager, since if you received too much information, hit C-c C-p which jumps you to the top of the last command, and then C-v your way down. Or better yet, just search for what you want. EShell means every command goes through the Emacs pager.

To be even cooler, try the Plan 9-like approach using the Eshell smart display, where you automatically begin at the top of the command until you type a non-cursor-movement key.

The Emacs Shell has a few other advantages:

  • Since it is written in Emacs Lisp, it has the same shell behavior between different operating systems.
  • Along with access to all your scripts and programs, you also have access to Emacs functions … write your shell scripts in Lisp? Sure!
  • Craft and extend your shell experience.

Eshell’s primary disadvantage shows when a program attempts to control the terminal.1

While you might have tried EShell in the past, I doubt you inhaled and noticed its uniqueness. Let’s level up…

Starting the Shell

Since my workflow is driven from Emacs, shells are temporary. I pop out to a shell for a few commands, and then return to my work. When I say pop out to the shell, I use the following function which creates a buffer-specific window in the lower third portion and start Eshell (which picks up that buffer’s directory).

(defun eshell-here ()
  "Opens up a new shell in the directory associated with the
current buffer's file. The eshell is renamed to match that
directory to make multiple eshell windows easier."
  (let* ((parent (if (buffer-file-name)
                     (file-name-directory (buffer-file-name))
         (height (/ (window-total-height) 3))
         (name   (car (last (split-string parent "/" t)))))
    (split-window-vertically (- height))
    (other-window 1)
    (eshell "new")
    (rename-buffer (concat "*eshell: " name "*"))

    (insert (concat "ls"))

(global-set-key (kbd "C-!") 'eshell-here)

My function, x exits that shell and closes that window.

(defun eshell/x ()
  (insert "exit")

Lisp REPL? Almost

EShell is a Lisp REPL. The following works as you’d expect from such a REPL:

$ (message "hello world")
"hello world"

However, in a shell, we care more for simplicity and speed of typing that we do for semantic clearness, so we can, in this case, drop the parens with the same results:

$ message "hello world"
"hello world"

Functions that begin with eshell/ are available in Eshell without the prefix, so calling the eshell/echo function makes the shell experience less surprising:

$ echo "hello world"
"hello world"

If you put it in parens, you need to give it the full name:

$ (eshell/echo "hello world")
"hello world"

What about types? In a normal shell, everything is a string, but EShell has a foot in both worlds:

$ echo hello world
("hello" "world")

A list of two strings. However, you can NOT attempt to pass that echo to car… at least not directly:

$ car echo hello world

Returns an error, as does:

$ car (list hello world)

You see, once you bring in parens, you also bring in syntactic specific-ness, so you would need to do this:

$ car (list "hello" "world")

EShell has a listify that converts its arguments to a list of strings:

$ listify hello world
("hello" "world")

But if you want to pass that data to something like car, you need to surround it in curly braces, which is EShell’s way of saying, call some shell-like-goodness, but return it like Lisp:

$ car { listify hello world }

Not really sure what is the real difference between list and listify, as they seem to behave the same to me:

$ listify hello world
("hello" "world")

$ list hello world
("hello" "world")

$ listify 1 2 3
(1 2 3)

$ list 1 2 3
(1 2 3)

$ list "hello world"
(#("hello world" 0 11
   (escaped t)))

$ listify "hello world"
(#("hello world" 0 11
   (escaped t)))

I guess my point to this little exploration is that you can pretend it is a shell, and you can pretend it is a Lisp REPL, and you have ways to weave between the two, but be a bit careful.


As the documentation says,

Since Eshell is just an Emacs REPL(1), it does not have its own scope, and simply stores variables the same you would in an Elisp program.

Running printenv only displays the environment variables:

$ setenv A "hello world"
$ getenv A
"hello world"

Use the setq to assign normal, Emacs variables:

$ setq B hello world
$ echo $B
$ setq B "hello world"
$ echo $B
hello world

Preface with a $, you can access any Emacs variable:

$ echo $recentf-max-menu-items

Keep in mind that environment variables over-shadow Emacs variables:

$ setenv C hello
$ setq C goodbye
$ echo $C

Finally, you can source Eshell variables from a file:

$ cat blah.eshell
setq FOO 42
setq BLING "bongy"

$ . blah.eshell

$ echo $FOO

$ echo $BLING


Executing a series of commands on each matching file is a standard use case for shells. While you could use a Lisp-like dolist, EShell attempts to give you a similar shell-like syntax:

$ for file in *.org {
  echo "Upcasing: $file"
  mv $file $file(:U)

The (:U) converts the contents before it to upper case format. It is a modifier, and I’ll babble on about this in the next section (since this is one of Eshell’s best feature).

I find it interesting to note that *.org gives the for loop a list to iterate over, but if there is more than one argument, a list is created, as in:

$ for i in 1 2 3 4 { echo $i }

Passing more than one list flattens them into one list, so the following works as you expect:

$ for file in emacs* zsh* { ... }

File Selection

If all you were doing was renaming a single file, or changing access permissions on all files in a directory, you’d hardly need a flexible shell, as dired or even Finder is sufficient for those tasks. A shell comes in handy when selecting a subset of files based on a pattern, and EShell really shines here, because of its filters (that it stole from Zshell’s modifiers):

$ ls -al *.mp3(U)   # Show songs I own

The *.mp3 part is just a normal globbing pattern we all know and love, but the (U) part further filters the selection. In this case limiting the selection to files you own.

The help for this feature is available in the shell, so type the following:

$ eshell-display-predicate-help
$ eshell-display-modifier-help

While you may have seen the predicates before (since they are similar to ZShell’s), the coolest part is that you can write some ELisp code to add your own predicates and modifiers.

File Filter Predicates

Here is a list of the filter predicates. These can be stacked, so typing, ls **/*(IW) will display all the files in the current directory (and sub-directories) that are readable by the group owner or all accounts on the system.

/ Directories (may accept d … gotta verify that)
. Regular files
* Executable files
@ Symlinks
p named pipes
s sockets
U Owned by current UID
u Owned by the given user account or UID, e.g. (u'howard')
g Owned by the given group account or GID, e.g. (g100)
r Readable by owner (A is readable by group)
R Readable by World
w Writable by owner (I is writable by group)
W Writable by World
x Executable by owner (E is executable by group)
X Executable by world
s setuid (for user)
S setgid (for group)
t Sticky bit
% Other file types.

These are fairly straight-forward. For example, list all the directories:

ls -ld *(/)

Some symbols take options, like to list all files owned by the howard account, specify the string with single quotes:

ls -ld *(u'howard')

The % requires a second parameter to specify the file type to filter. These are taken from the ls, so %c will display char devices. Here is the list if stole from the ls man page:

b Block special file
c Character special file
d Directory
l Symbolic link
s Socket link
- Regular file

The options can be combined. For instance, list all symbolic links owned by your account:

ls -l *(@U)

Or, list all symbolic links that you don’t own by prefixing the ^ symbol:

ls -l *(@^U)

Gets more interesting with time and size filters which take parameters. This is the cheat-sheet-like output from eshell-display-predicate-help:

access time +/-/= N months/weeks/hours/mins/secs (days if unspecified) if FILE specified, use as comparison basis; so a+’file.c’ shows files accessed before file.c was last accessed.
modification time…
change time…
file size +/-/= N Kb/Mb/blocks

The following examples should clarify how to use these:

To display all org-mode files in my directory that I’ve modified since yesterday, I would type:

ls *.org(m-1)

Where the m is the modification time, the - means less than and 1 refers to the day, since we didn’t specify any other time period. To display the files we’ve modified over the last 8 hours, we’d enter:

ls *.org(mh-8)

Compress everything which hasn’t been accessed in 30 days:

bzip2 -9v **/*(a+30)

The ** symbol is recursive access to sub-directories.

Shell scripts (that end with a .sh and are executable (we specify the * character first) that are 50k or larger (we use the + symbol):

ls ***/*.sh(*Kl+50)

To specify 50K, we first write k then + (to mean or larger) and finally the size. The three stars, *** is a recursive search into sub-directories, but not to follow symbolic links.


Modifiers are similar to filters mentioned above, except that begin with a colon symbol, and they change the string, file or list that precedes it. For instance, :U upper-cases a string or file name:

for f in *(:U) { echo $f }



The modifiers can also affect a variable. The following example behaves the same as the previous example:

for f in * { echo $f(:U) }

Here is the complete list of modifiers for an individual string or file name:

:L lowercase
:U uppercase
:C capitalize
:h dirname
:t basename
:e file extension
:r strip file extension
:q escape special characters
:S split string at any whitespace character
:S/PAT/ split string at each occurrence of /PAT/
:E evaluate again

Here is the list of modifiers for a list:

:o sort alphabetically
:O reverse sort alphabetically
:u unique list (typically used after :o or :O)
:R reverse the list
:j join list members, separated by a space
:j/PAT/ join list members, separated by PAT
:i/PAT/ exclude all members not matching PAT
:x/PAT/ exclude all members matching PAT
:s/pat/match/ substitute PAT with MATCH
:g/pat/match/ substitute PAT with MATCH for all occurrences

To append the string, -foobar, to all files owned by you, before the extension, you would type:

for F in *(U) { mv $F $F(:r)-foobar.$F(:e) }

Custom Filter Predicates

As you know, the best part of Emacs is its ability to customize everything…including your shell experience.

As Mickey Petersen mentions, we can create our own predicates to filter out files. Wouldn’t it be nice if we could specify files based on their internal org-mode #+TAGS entry. For instance, at the top of my files, I add the following headers:

#+TITLE:  Alfred
#+AUTHOR: Howard Abrams
#+DATE:   [2013-05-15 Wed]
#+TAGS:   mac technical

I would like Eshell to be able to list files that end in org, but with contents that include the mac tag, like:

ls *.org(T'mac')

If the filter used a single symbol, we would append a tuple to eshell-predicate-alist that specifies the symbol and the predicate function (that returns true or nil). Something like

(add-to-list 'eshell-predicate-alist '(?P . eshell-primary-file))

However, in our example, the T symbol takes the tag as a parameter. In this case, we need a two-step process:

  1. A function to parse the Eshell buffer to look for the parameter (and move the point past the parameter)
  2. A predicate function that takes a file as a parameter

For the first step, we have our function called as it helps parse the text at this time. Based on what it sees, it returns the predicate function used to filter the files:

(add-to-list 'eshell-predicate-alist '(?T . (eshell-org-file-tags)))

I combine the two steps into a single function, as after the function is done with Step 1, we will return a lambda expression for Step 2.

The first step is to parse the text following the point looking for the tag (surround in single quotes), and move the point forward over the option for our filter function (notice the goto-char to the end of the match).

(defun eshell-org-file-tags ()
  "Helps the eshell parse the text the point is currently on,
looking for parameters surrounded in single quotes. Returns a
function that takes a FILE and returns nil if the file given to
it doesn't contain the org-mode #+TAGS: entry specified."

  ;; Step 1. Parse the eshell buffer for our tag between quotes
  ;;         Make sure to move point to the end of the match:
  (if (looking-at "'\\([^)']+\\)'")
      (let* ((tag (match-string 1))
             (reg (concat "^#\\+TAGS:.* " tag "\\b")))
        (goto-char (match-end 0))

        ;; Step 2. Return the predicate function:
        ;;         Careful when accessing the `reg' variable.
        `(lambda (file)
             (insert-file-contents file)
             (re-search-forward ,reg nil t 1))))
    (error "The `T' predicate takes an org-mode tag value in single quotes.")))

The returned function loads the given file into a temp-buffer, and search the contents for the regular expression. Returns nil if the match wasn’t found (false), and anything else is interpreted as true.

Now I can search for Homebrew commands without bringing up my beer notes:

$ grep brew *.org(T'mac')

Since the grep is really the Emacs grep function, it brings the results up in a buffer that I can click to load/edit.


Of course the real beauty of EShell is the Emacs integration, for instance, using highlight-regexp to colorize key words from expected output, or piping the results to an Emacs buffer:

$ ls -al > #<buffer some-notes.org>

And then issue a C-c | on that results to create an org-mode table you can continue to manipulate.

While the Eshell is baked into Emacs, and requires no customization, I have made a few improvements that may be helpful to others.



Programs like top wouldn’t work well in Eshell since it these programs are still attempting to manipulate the world with ancient VT100 control codes, and the Eshell mostly is expecting standard textual output.

However, if you type top, eshell notices top on its naughty list (actually, the list is called eshell-visual-commands), and will farm it out to a special comint buffer.

In practice, I don’t notice this limitation, since most applications I would like, I usually just use a similar program re-written in Emacs. However, if you find an app that doesn’t work well in EShell, append it to this list.