(defblog exordium) Emacs and Lisp musings

My fancy Zsh prompt

It’s been several years since I updated this blog. Let’s dust it off a bit.

I updated my Zsh prompt to be more efficient and more useful. Features:

  • Display current path abbreviated to the 3 last subdirectories:

zsh prompt

  • Automatically detect git repos. Display the name of the repo and the current branch:

zsh prompt

  • Display indicators if the git repo contains untracked/unstaged/staged files:

zsh prompt

Here is the code:

setopt prompt_subst
autoload -U colors && colors

# Colors for the Tomorrow Night theme
# xterm-256color codes: https://jonasjacek.github.io/colors/
color_path='117'      # SkyBlue1
color_repo='148'      # Yellow3
color_branch='246'    # Gray58
color_modified='167'  # IndianRed
color_untracked='221' # LightGoldenrod2
color_staged='148'    # Yellow3
color_host='93'       # Purple

# Return a colored string containing the name of the git branch in parentheses, like (master).
git_branch() {
    local BRANCH=$(git rev-parse --abbrev-ref HEAD 2> /dev/null | sed -e 's/\(.*\)/on \1/g')
    echo " %F{$color_branch}$BRANCH%f"
}

# Return a colored string containing bullets of different colors indicating
# staged/untracked/modified files in a git repo.
git_status() {
    # Populate flags with bullets of different colors for staged/untracked/modified files
    local MODIFIED="%F{$color_modified}●%f"
    local UNTRACKED="%F{$color_untracked}●%f"
    local STAGED="%F{$color_staged}●%f"
    local -a FLAGS

    if ! git diff --cached --quiet 2> /dev/null; then
        FLAGS+=( "$STAGED" )
    fi

    if [[ -n $(git ls-files --other --exclude-standard `git rev-parse --show-toplevel` 2> /dev/null) ]]; then
        FLAGS+=( "$UNTRACKED" )
    fi

    if ! git diff --quiet 2> /dev/null; then
        FLAGS+=( "$MODIFIED" )
    fi

    # Format flags to add a space if not empty
    local -a STATUS
    STATUS+=( "" )
    [[ ${#FLAGS[@]} -ne 0 ]] && STATUS+=( "${(j::)FLAGS}" )
    echo "${(j: :)STATUS}"
}

# If in a git repo, return git_branch() and git_status().
# Otherwise return nothing.
prompt_git_info() {
    # Exit if not inside a Git repository
    ! git rev-parse --is-inside-work-tree > /dev/null 2>&1 && return
    echo "$(git_branch)$(git_status)"
}

# Return a colored string containing the path, shortened to the last 3 directories.
# The current directory is highlighted.
# If in a git repo, the string starts at the root of the repo, highlighted in different color.
prompt_path() {
    local PROMPT_PATH=""
    local CURRENT=$(basename $PWD)
    if [[ $CURRENT = / ]]; then
        # At /
        PROMPT_PATH="%F{$color_path}/%f"
    elif [[ $PWD = $HOME ]]; then
        # At ~
        PROMPT_PATH="%F{$color_path}~%f"
    else
        local GIT_REPO_PATH=$(git rev-parse --show-toplevel 2>/dev/null)
        if [[ -d $GIT_REPO_PATH ]]; then
            # Inside a git repo.
            ROOT=$(basename $GIT_REPO_PATH)  # repo name

            if [[ $PWD -ef $GIT_REPO_PATH ]]; then
                # At the root of the repo.
                PROMPT_PATH="%B%F{$color_repo}$ROOT%f%b"
            else
                # Below the root of the repo.
                REAL_PWD=$PWD:A
                PATH_TO_CURRENT="${REAL_PWD#$GIT_REPO_PATH}"  # path from root, without root
                PATH_TO_CURRENT="${PATH_TO_CURRENT%/*}"       # remove last dir

                if [[ -z "$PATH_TO_CURRENT" ]]; then
                    # Just one level below root
                    PROMPT_PATH="%F{$color_repo}$ROOT/%f%B%F{$color_path}$CURRENT%f%b"
                else
                    # More than one level below root
                    local -a DIRS=(${(s|/|)PATH_TO_CURRENT})  # split string into array using /
                    local LENGTH="${#DIRS[@]}"
                    if [[ $LENGTH -gt 2 ]]; then
                        PATH_TO_CURRENT="/…/$DIRS[$LENGTH-1]/$DIRS[$LENGTH]"
                    fi
                    PROMPT_PATH="%F{$color_repo}$ROOT%f%F{$color_path}$PATH_TO_CURRENT/%B$CURRENT%b%f"
                fi
            fi
        else
            # Not in a git repo.
            # Note: this expression checks for 4 elements long: %(4~|true|false)
            PATH_TO_CURRENT=$(print -P "%(4~|…/%3~|%~)")      # shortened pwd
            PATH_TO_CURRENT="${PATH_TO_CURRENT%/*}"           # remove last dir
            PROMPT_PATH="%F{$color_path}$PATH_TO_CURRENT/%B$CURRENT%b%f"
        fi
    fi
    echo "$PROMPT_PATH"
}

# Return a colored string containing the local host name, if ssh.
# Return an empty string otherwise.
remote_hostname() {
    local MACHINE=""
    if [ -n "$SSH_CLIENT" ]; then
        MACHINE="%F{$color_host}[%m]%f "
    fi
    echo "$MACHINE"
}

export PROMPT=$'$(remote_hostname)$(prompt_path)$(prompt_git_info) %F{$color_path}❯ %f'

# Simple/uncolored prompt for troubleshooting
noprompt() {
    export PROMPT='%2~ ❯ '
}

Or course, if you are just starting with zsh you are better off just adopting Oh My Zsh and Powerlevel10k.

My fancy Zsh prompt

I switched from bash to zsh a few years ago and I am never looking back! It has awesome tab completion: for example cd doc/sh/pl becomes cd Documents/shared/planning. It also expands git commands and branches, environment variables and other things. It has extended file globbing which provides a good replacement for find. If you spend a lot of time in the shell, you should really give zsh a try.

Whatever shell you use, it’s worth spending a few minutes configuring your prompt. The prompt is something you will see literally thousands of times a day. So why not make it useful?

Here is mine. First, let’s open a new shell in the home directory:

zsh prompt

Simple and to the point. ~ sweet ~! Let’s go into some directory:

zsh prompt

It displays the path, from the home directory, with the current directory displayed in bold (“planning”). To keep the prompt short, we only display the last 3 subdirectories:

zsh prompt

And now let’s move into a git repo:

zsh prompt

The prompt displays the path starting at the root of the repo (“org”), and also the current branch in green (“master”). Now let’s move into a subdirectory in the same repo:

zsh prompt

The root of the repo is highlighted (“org”), and the current directory is displayed in bold like before. We could also display something indicating if the repo is clean or if it has uncommitted changes, but I prefer to keep it simple.

The colors are from the Tomorrow Night theme, which also happens to be the default theme in Exordium.

Here is the code (in ~/.zshrc):

# Zsh options
setopt prompt_subst
autoload -U colors && colors

# Colors
BLACK=$'\033[0m'
GREEN=$'\033[38;5;148m'
BLUE=$'\033[38;5;117m'
DARK_BLUE='\033[38;5;4m'

current_git_branch() {
    git rev-parse --abbrev-ref HEAD 2> /dev/null | sed -e 's/\(.*\)/(\1)/g'
}

current_directory() {
    PROMPT_PATH=""
    CURRENT=`dirname ${PWD}`
    if [[ $CURRENT = / ]]; then
        PROMPT_PATH=""
    elif [[ $PWD = $HOME ]]; then
        PROMPT_PATH=""
    else
        GIT_REPO_PATH=$(git rev-parse --show-toplevel 2>/dev/null)
        if [[ -d $GIT_REPO_PATH ]]; then
            # We are in a git repo. Display the root in color, then the path
            # starting from the root.
            if [[ $PWD -ef $GIT_REPO_PATH ]]; then
                # We are at the root of the git repo.
                PROMPT_PATH=""
            else
                # We are not at the root of the git repo.
                BASE=$(basename $GIT_REPO_PATH)
                GIT_ROOT="%{$fg_bold[red]%}%{$DARK_BLUE%}${BASE}%{$reset_color%}"
                REAL_PWD=$PWD:A
                PATH_TO_CURRENT="${REAL_PWD#$GIT_REPO_PATH}"
                PATH_TO_CURRENT="%{$BLUE%}${PATH_TO_CURRENT%/*}%{$reset_color%}"
                PROMPT_PATH="${GIT_ROOT}${PATH_TO_CURRENT}/"
            fi
        else
            # We are not in a git repo.
            PATH_TO_CURRENT=$(print -P %3~)
            PATH_TO_CURRENT="%{$BLUE%}${PATH_TO_CURRENT%/*}%{$reset_color%}"
            PROMPT_PATH="${PATH_TO_CURRENT}/"
        fi
    fi
    echo "${PROMPT_PATH}%{$reset_color%}%{$fg_bold[red]%}%{$BLUE%}%1~%{$reset_color%}"
}

export PROMPT=$'$(current_directory) %{$GREEN%}$(current_git_branch)%{$BLACK%}%# '

Notice the function current_git_branch: this is the fastest way I have found to get the name of the branch. The trick is to make the execution of the prompt as fast as possible, since it gets executed every time you hit Enter.

Org mode part 2

This is the second article in a series about Emacs Org mode. It assumes that you read the first one.

Today we’ll talk a bit about literate programming. The idea is to write an Org mode document that includes snippets of code, and possibly the result of their execution as well. It kind of reverses the way you think about documenting code: instead of adding comments inside your program, you add your program’s code inside the documentation file. In the end the result is the same, you still get executable code.

Org mode includes a feature named Babel which allows for embedding code in any programming language you fancy. Let’s get started.

Enabling Babel

First, we need to make sure that Babel is enabled for a few other languages than ELisp (the only one by default). Add the following in your emacs configuration and evaluate it with M-C-x (or restart Emacs):

(org-babel-do-load-languages
 'org-babel-load-languages '((emacs-lisp . t)
                             (ruby . t)
                             (python . t)
                             (sh . t)))

Hello World

First open a new buffer literate.org, with this content:

#+TITLE: Literate Programming

#+begin_src python
def hello(str):
    return "Hello, " + str + "!"

return hello("Dude")
#+end_src

The #+TITLE directive is just to add a title to the Org file; it is not really needed. The other directives we use are #+begin_src and #+end_src which respectively introduce and close a code block for a particular language. Note that Emacs automatically applies the proper syntax highlighting.

Type C-c C-c to evaluate the code block under the point. After confirmation, Babel will insert a new block #+RESULTS with the result of the evaluation, like so:

Org-mode20

What happened here? Emacs forked a process to execute the code using Python, and got back the returned value, and inserted it into the buffer.

By default the result is the returned value i.e. the last expression that was executed, but you can also use the output of the program as the result with a directive like: #+begin_src python :result output. For example, evaluate the shell statement below, again using C-c C-c:

#+begin_src sh :results output
   echo "Hello $USER! Today is `date`"
#+end_src

Org-mode21

Calling code blocks

Babel also allows you to refer to code blocks from elsewhere in your document, by labeling each block with a name, using directive #+name. Let’s say we have some Ruby code to revert a string (yes, I know we could use the native reverse!):

#+name: reverse_str
#+begin_src ruby
def reverse_str(s)
  len = s.length
  reversed = ""
  for i in 1..len do
    reversed += s[-1*i]
  end
  return reversed
end

reverse_str(str)
#+end_src

We can now call this block. Note that we get the result of the block evaluation. So if you want to use the result of a function in the block, you also need to add the call to that function (see the last line).

To call the block, add a #+call directive referring to the block’s name and binding the input variable str, and type C-c C-c with the point onto it:

#+call: reverse_str(str="The Quick Brown Fox")

Org-mode22

More fun

Here is a more practical example to illustrate the power of Babel, using different languages to get the job done. This is actually similar to an iPython notebook.

Suppose I want to produce a report containing statistics about the words that are used in a collection of files. Assuming all these files are in the same directory (let’s say they are Markdown files), this shell script counts the number of unique words in each file:

#+name: words
#+begin_src sh
  for F in /Users/phil/Documents/philippe-grenet.github.io/_posts/*.md
  do
     cat $F | tr -cs A-Za-z '\n' | tr A-Z a-z | sort | uniq -c | wc -l
  done
#+end_src

The result has multiple lines, one per file. Therefore it is stored into an Org table:

Org-mode23

Now let’s compute the standard deviation. We could follow up with another shell script, but it is easier to do this in ELisp. Add this block in the buffer:

#+begin_src elisp :var samples=words
(defun standard-deviation (samples)
  (let* ((samples    (mapcar #'car samples))
         (n          (length samples))
         (average    (/ (reduce #'+ samples) n))
         (deviations (mapcar #'(lambda (x)
                                 (expt (- x average) 2))
                             samples)))
    (sqrt (/ (reduce #'+ deviations) n))))

(standard-deviation samples))
#+end_src

The #+begin_src directive uses the :var keyword to bind the Lisp variable samples to the output of the words code block (which is the table above). In Lisp that samples variable will be set with a list of lists. Basically each row in the table is a sublist, and since the table has only one column the sublists only contain a single number, like this:

;; Value of 'sample':
'((55) (274) (560) (296) (650) (394) (253) (264))

If you evaluate the block with C-c C-c, you get the final result:

Org-mode24

You can of course re-run the report anytime you want. Once we have our report ready, Org mode is able to export it into plenty of different formats, choosing exactly what should be included and excluded. But that’s a discussion for another day.

Calc

If you are a programmer, there probably isn’t a day that goes by when you don’t need a calculator, for example for things like base conversion. After all, like Dijkstra said, programming is math.

Emacs includes a calculator, which is actually quite sophisticated. It is capable of both algebraic and symbolic calculations; it has a ton of functions and it can handle matrices or solve equations, if you need that sort of things. It can even create graphs if you have GnuPlot installed on your machine.

Here we will just review the basics. Check out the manual for more. There a nice Info manual embedded in Emacs which you can bring up using C-h i g (calc).

Start the calculator using M-x calc. It will show up at the bottom of the screen. Type q to quit. Once the calculator is started, you can hide and redisplay it using C-x * *.

RPN logic

Calc uses the RPN (Reverse Polish Notation) logic. Which is only fair, since Lisp uses the Polish notation! Anyway, if you are a fan of the HP calculators, you will be just at home.

Calc displays a stack of numbers. When you type a number, it accumulates in the stack. When you type an operator such as + or *, it operates on the last 2 numbers of the stack, and replaces those numbers with the result.

Give it a try: Type M-x calc, then the following:

  • 2 Enter
  • P (no need to press Enter)
  • 10 Enter

Org-mode1

The P inserts π. Now you have 3 numbers in the stack, and you can calculate the circumference of a circle of radius 10 (C = 2 π r). Press * once, the stack now shows 2 and 31.4159265359. Press * another time and you get a single number which is the result.

Because the dash key (-) is reserved for subtraction, you should use an underscore (_) to enter a negative number. Note that the calculator supports arbitrarily big numbers. If you want to challenge it, try calculating 2^200 (that’s 2 Enter, 200 Enter, then ^). The default precision is 12 but you can change it using p.

Type n to negate the last number, and TAB to flip the last 2 numbers. The delete key deletes the last number on the stack.

If you need to use complex numbers, enter them like (a,b). If you need fractions, enter them like 1:3 (1/3). You can also enter an algebraic expression directly by typing ' (the quote key) then your expression, such as 2 + (10 / 5). The result will be put on top of the stack.

Most functions use a single key. For example, type s for Sin, c for Cos, t for Tan, etc. The more complex commands are modal: they start with a prefix key, which leads to a prompt for the subcommand key. If you get lost, type ? and the mini-buffer will show you what is available.

If you make a mistake, type U to undo and D to redo. Of course this is Emacs, you have an infinite number of undos and redos. You can also reset the calculator using C-x * 0.

Base conversion

You can enter any number in the format base#number. Example: 16#FF is immediately converted to 255.

For the reverse, you need to set the output display mode. In this example, d r 16 followed by Enter sets the display to base 16. d r 2 sets it to binary. Set it to base 10 to get the default behaviour again.

Emacs Tips

In this short post we’ll review a few features of Emacs that are not well known, but really worth knowing.

Repeating commands

The simplest way to repeat a command is to type Control-a number, then the command. For example if I type C-6 C-0 ~ (that’s 6 then 0 with the Control key down, then the tilde character), I get a line with 60 tilde:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Another way is to use the universal argument: we saw how to write an interactive Lisp function that accepts a numeric argument in the previous article. forward-line is such a function; it goes down N lines (by default one line). For example C-u 1 0 DOWN (Control-u then 1 then 0 then the down arrow key) goes down 10 lines. You can do the same by calling the function explicitly with C-u 1 0 M-x forward-line.

Registers

Registers are used for saving temporary data in the current session. They can store positions or bookmarks, text, numbers etc. Registers have a name which is a letter or a number: a to z, A to Z (so case-sensitive) and 1 to 9, which gives you a total of 62 registers. That’s more than enough.

C-x r is the prefix for all register operations. Generally you type this prefix, then a key to specify what operation you want (e.g. save or read), then the register name which is one extra key.

Buffer positions

For example, open any file you have in one of your projects, go to some position in the file, and hit C-x r SPC a (where SPC is the space bar). This command stores the buffer position in register a. Now go to a different buffer such as the scratch buffer, and type C-x r j a: this command will jump back to the buffer and the position you just stored in register a.

Note that if you close the file and then retype the same command, Emacs will ask you if you want to reopen the file, and it will bring you back to the same position again. This can be useful if you want to keep bookmarks in a large project, for example if you keep going to the same files and bits within these files, but you don’t want to keep them open all the time.

You can view the content of register a with command M-x view-register then a at the prompt. It will show a window saying that this register stores a position in your buffer (note that the position is a number, which represents the offset from the beginning of the buffer). Type C-x 1, or q in the other window, to dismiss it.

Bookmarks

A bookmark is similar to a buffer position but with a twist: you give it a name and it is persisted between Emacs sessions. Note that bookmarks are not actually associated with registers, but they use the same C-x r prefix.

The command is C-x r m (mark) which prompts for a name. By default it proposes the current buffer name but you can choose whatever you want. You can view the list of bookmarks with C-x r l (list). Just click on a bookmark to jump to it. You can also jump to a bookmark with C-x r b (bookmark) which prompts for the name using auto-complete. The nice thing about bookmarks is that they are saved on the filesystem when you exit Emacs, and they are available when you restart it. You can also force save with M-x bookmark-save.

Text

Sometimes you want to save a snippet of text somewhere, so you can paste it later. One way is to use the kill ring: M-w to save the selected text (which is what Emacs calls the region), then C-y to paste. The kill ring saves all copy operations you have done so far, so if you want to paste the second previous thing you copied, type C-y followed by M-y (repeat M-y to go back in history).

Another way is to save the region in a register, which you do with something like C-x r s a (save in register a). Now if you want to insert the content of a register at the current point, type C-x r i a (insert the content of a).

Windows

This is less useful than the above, but you can also save a window configuration in a register. For example, split the screen horizontally with C-x 2, then split the current window vertically with C-x 3. You now have 3 windows displayed, which you can resize as you see fit.

Let’s save this layout in register a with C-r w a. Now dismiss all other windows than the current one using C-x 1. If you want to restore the layout, type C-r j a (it is the same key for jumping to a buffer position). Note that this only saves window configurations and not buffers content: if you close one of the buffers Emacs will not reopen it for you.

Summary

The table below summarizes the keys we just learned.

Key binding Description
C-x r SPC a Store the current position to register a.
C-x r j a Jump to the position stored in register a,
or restore the window positions stored in register a.
C-x r s a Save the selected region in register a.
C-x r i a Insert the text saved into register a.
C-x r w a Save window positions in register a.
C-x r m Save a bookmark.
C-x r b Go to a bookmark.
C-x r l List bookmarks.

Macros

Editing macros are a very powerful feature of Emacs. After all, Emacs stands for “Editing Macros” 1.

There are several keys for macros, but really you only need to remember two of them: F3 and F4. The first one records a new macro. The second one terminates the recording, if you were recording a macro; otherwise it executes the macro. If you make an mistake while recording a macro, hit C-g to abort, and start over.

Let’s take an example. Suppose I have this text:

the quick brown fox
jumps over
the lazy dog

Now suppose I want to make each line start with a capital letter and end with a period. I could edit the text manually because it is only 3 lines, but just imagine that it is much longer for argument’s sake, in order to make the use of a macro more compelling.

The way to do this with a macro is simple: fix the first line, while recording a macro. Then execute the macro N times, one time per remaining line. To record the macro do the following:

  • Move the cursor to the beginning (e.g. M-<).
  • F3 to start recording.
  • M-c to capitalize the first word (The).
  • C-e to go to the end of the line.
  • . to insert a period. Now the first line is good.
  • C-a to go to the beginning of the line (where we started), and the down arrow to go to the next line.

Now type F4 to stop recording. Then F4 again to run the macro on the second line. Then F4 again to run the macro on the 3rd line. You’re done!

The quick brown fox.
Jumps over.
The lazy dog.

You could also run the macro N times using Control-a number then F4, as we saw earlier. You can also apply the macro to a whole region by selecting the region and running M-x apply-macro-to-region-lines, which is neat.

If you want to see macros at their best (and incidentally Emacs beat Vim at its own game), check out this quick video:

The entire series of Emacs Rocks is worth watching.


That’s it for today. Lots more to come. Stay tuned!

  1. Note that Emacs editing macros have nothing to do with Lisp macros: one is a trick to save a sequence of keys and repeat it, the other is a Lisp function that executes twice, at compilation time and at run time.