Daudin - a Python command-line shell

daudin is a UNIX command-line shell based on Python.

The aim is to provide an interactive shell that is as convenient to use as the regular shell (in particular providing pipelines) but which has Python as its programming language.

Contents: Installation · Usage · Examples · Pipelines · Changing directory · Command substitution · Readline · Init file · Prompts · More on usage · Exiting · Command interpretation · Pipeline execution · Shell execution · Debugging · Version · Background · TODO.

Installation

$ pip install daudin

Usage

Run daudin and enter commands interactively.

Should run fine on a recent version of Python 3 (I am using 3.7.3).

Examples

The following examples all assume you have already run daudin (which prints the >>> prompt).

Like a regular shell, you have direct access to UNIX tools:

>>> ls -l
total 44
-rw-r--r-- 1 terry terry   635 Oct 12 17:34 Makefile
-rw-rw-r-- 1 terry terry 16619 Oct 12 23:05 README.md
-rwxrwxr-x 1 terry terry  1261 Oct 12 22:42 daudin
drwxrwxr-x 3 terry terry  4096 Oct 12 22:51 daudinlib
-rw-rw-r-- 1 terry terry  2309 Oct 12 23:05 example-functions.py
-rw-r--r-- 1 terry terry  1546 Oct 12 17:43 setup.py
drwxrwxr-x 3 terry terry  4096 Oct 12 22:48 test
>>> ls | wc -l
7
>>> echo hello there > /tmp/xxx
>>> cat /tmp/xxx
hello there

(If regular shell commands are slow to execute, it's probably because your regular shell is slow to start. Change the default underlying shell used by daudin to something faster, as described below.)

But in fact it's Python all the way down:

>>> from math import pi
>>> pi
3.141592653589793
>>> def area(r):
...   return r ** 2 * pi
...
>>> area(2.0)
12.566370614359172

Pipelines

Shell pipelines are super cool. As you've seen above daudin, has pipelines that look just like the shell. But there are a few added extras.

You can mix Python and the shell in a daudin pipeline:

>>> import this | grep 'better than'
Beautiful is better than ugly.
Explicit is better than implicit.
Simple is better than complex.
Complex is better than complicated.
Flat is better than nested.
Sparse is better than dense.
Now is better than never.
Although never is often better than *right* now.

In Python commands in a daudin pipeline, the current pipeline value is kept in a variable named _, which may be of any type:

>>> -6 | abs(_) | _ * 7
42
>>> 'hello' | _.title()
Hello

UNIX commands produce lists of strings in _:

>>> ls | for name in _:
...   prefix = name.split('.')[0]
...   print(len(prefix), prefix.upper())
...
8 MAKEFILE
6 README
6 DAUDIN
9 DAUDINLIB
17 EXAMPLE-FUNCTIONS
5 SETUP
4 TEST

That means you should use _[0] if you want to act on just the first line of UNIX command output:

>>> def triple(x):
...   return int(x) * 3
...
>>> echo a b c | wc -w | triple(_[0])
9

Here's the same thing, but with cat reading from the terminal:

>>> def triple(x):
...   return int(x) * 3
...
>>> cat | wc -w | triple(_[0])
a b c
^D
9

You can really mix things up (UNIX, Python, UNIX, Python):

>>> seq 0 9 | map(lambda x: 2 ** int(x), _) | sum(_)
1023
>>> seq 0 9 | list(map(lambda x: 2 ** int(x), _)) | tee /tmp/powers-of-two | sum(map(int, _))
1023
>>> cat /tmp/powers-of-two
1
2
4
8
16
32
64
128
256
512

You can hit ENTER in the middle of a pipeline without disrupting it. So this:

>>> echo a b c | wc -w
3

is equivalent to this

>>> echo a b c |
>>> wc -w
3

And if you forget to end a pipeline command-line with a | you can just put one at the start of the next line to continue the pipeline:

>>> echo a b c
a b c
>>> | wc -w
3

If you need to write a helper function in the middle of the pipeline before continuing processing, you can do it:

>>> def triple(x):
...   # Doesn't use int() to convert its argument.
...   return x * 3
...
>>> echo a b c | wc -w
3
>>> f = lambda line: int(line[0])
>>> | f(_) | triple(_)
9

Similarly, you can change directories in the middle of a pipeline (see Changing directory below).

You can put comments into the middle of a pipeline

>>> ls -1
Makefile
README.md
daudin
daudinlib
setup.py
test
>>> # The pipeline is still alive!
>>> _
['Makefile', 'README.md', 'daudin', 'daudinlib', 'setup.py', 'test']
>>> | wc -l
6

You can also pipe the output of a multi-line Python command directly into a following command:

>>> ls | for name in _:
...   prefix = name.split('.')[0]
...   print(len(prefix), prefix.upper()) | | sort
17 EXAMPLE-FUNCTIONS
4 TEST
5 SETUP
6 DAUDIN
6 README
7 LICENSE
8 MAKEFILE
9 CHANGELOG
9 DAUDINLIB

There are two | symbols before the sort above because an empty command is necessary to terminate the compound Python command.

The output above should have been numerically sorted. The pipeline can be immediately continued using a leading |:

>>> | sort -n
4 TEST
5 SETUP
6 DAUDIN
6 README
7 LICENSE
8 MAKEFILE
9 CHANGELOG
9 DAUDINLIB
17 EXAMPLE-FUNCTIONS

Here's another example where two | symbols can be used to terminate the Python command in order to continue the pipeline on a single line:

>>> ls | for i in _: print(i[:3]) | | wc -l
11

The above could instead be piped into the sus function I have in my ~/.daudin.py file (see Init file for details on this). The sus Python function does the equivalent of the shell sort | uniq -c | sort -nr trick for finding the most common inputs:

>>> ls | for i in _: print(i[:3]) | | sus()
3 dau
1 CHA
1 LIC
1 Mak
1 REA
1 dis
1 exa
1 set
1 tes

Just to repeat: the || provides an empty command (the zero-length string between the pipe symbols) to terminate the Python for block. The space between the two pipe symbols is optional.

Undo in a pipeline

If you run a command that alters the pipeline content and you want to restore it to its former value, you can undo with %u.

There is only a single undo at the moment. This could obviously be improved, and a redo command could be added.

You can of course always save the current pipeline value into a Python variable:

>>> echo a b c
>>> a = _

Changing directory

Changing directory has to be handled a little specially because although a cd command can be handed off to a shell, the change of directory in that shell will have no effect on your current process. That's why cd has to be a special "built-in" command in regular shells.

In daudin you can change dir using regular Python:

>>> import os
>>> os.chdir('/tmp')
>>> pwd
/tmp

but that's far too laborious for interactive use. So there's a cd function provided for you:

>>> cd('/tmp')
>>> pwd
/tmp

To ease this, at the price of a litle ugliness, there's also a "built-in" special command called %cd:

>>> %cd /tmp
>>> pwd
/tmp

Changing directory does not affect the current pipeline value. So you can change directory in the middle of a pipeline:

>>> mkdir /tmp/a /tmp/b
>>> %cd /tmp/a
>>> touch x y z
>>> ls | %cd /tmp/b | for i in _:
...   with open(i + '.txt', 'w') as fp:
...     print('I am file', i, file=fp)
>>> pwd
/tmp/b
>>> cat x.txt
I am file x

Command substitution

In regular shells there is a way to have part of a command line executed in a sub-shell and the output of that sub-shell replaces that part of the original command.

For example, suppose you want to get the value of date into a variable. In the bash shell you could do this:

$ d=$(date)
$ echo $d
Sat Oct  5 22:36:24 CEST 2019

In daudin there is a sh function that you can use to pass commands to a sub-shell. So, equivalently:

>>> d = sh('date')
>>> d
Sat Oct  5 22:36:24 CEST 2019

If you then wanted to extract the month from the date output, in the shell you could do this:

# Working from the d variable set above:
$ month=$(echo $d | cut -f2 -d' ')
# Or you could call date again:
$ month=$(date | cut -f2 -d' ')
$ echo $month
Oct

Same thing in daudin:

# Working from the d variable set above:
>>> month = d.split()[1]
# Calling date again:
>>> month = sh('date').split()[1]
>>> month
Oct

You could also have the shell do all the work via sh() but to do that you'll need to use \| in the to ensure that daudin doesn't incorrectly split the shell command into two pieces:

>>> month = sh('date \| cut -f2 -d" "')

Readline

daudin uses the GNU readline library to make it easy to edit and re-enter commands. The history is stored in ~/.daudin_history.

daudin provides file and directory name completion, as well as Python completion (the latter using rlcompleter).

Init file

daudin will initially read and execute code in a ~/.daudin.py file, if any. This is a good place to put convenience functions you write that you want readily accessible. The file example-functions.py has some functions that give a flavor of how you can add functionality to daudin. I have all these in my ~/.daudin.py file.

Use the --noInit argument when invoking daudin to disable loading the init file.

Use the special %r (reload) command to re-read your init file.

Changing prompts

There are --ps1 and --ps2 options that can be given on the command line to daudin. You can also set sys.ps1 or sys.ps2 while running:

>>> sys.ps1 = '% '
% 3 + 4
7

You can also set sys.ps1 or sys.ps2 to be a function that returns a string. In the example-functions.py file mentioned above, you can see sys.ps1 assigned to a simple function returning a string containing the basename of the current working directory (or ~ if you are in your home directory) plus the name of the current git branch (if any). The string also contains escape sequences to color the prompt. This is all extremely simplistic and for now is just an example. It would make sense to port a more sophisticated prompt package from a shell to Python and incorporate that.

More on usage

As you've seen, you can simply invoke daudin with no arguments and it will start an interactive read-eval-print loop, reading from standard input.

You can also write scripts and run them by giving the script filename(s) on the command line:

$ cat daudin-script
#!/usr/bin/env daudin

def triple(x):
    return int(x) * 3

echo a b c d | wc -w | triple(_[0])

$ daudin daudin-script
12

You can provide several script filenames on the command line and they will be run in turn. If you give a filename of - it will cause daudin to read from standard input. Note that if you provide filenames, input piped into daudin will be ignored (to read standard input, use a - as just mentioned).

Finally, you can pipe commands into daudin:

$ echo 33 + 34 | daudin
77

Command-line options

daudin understands the following command-line options (run daudin --help to see this):

usage: daudin [-h] [--ps1 PS1] [--ps2 PS2] [--shell SHELL] [--noInit]
              [--noPtys] [--debug] [--tracebacks]
              [FILE [FILE ...]]

A Python shell.

positional arguments:
  FILE           A file of commands to run non-interactively. Use "-" to
                 indicate reading from standard input.

optional arguments:
  -h, --help     show this help message and exit
  --ps1 PS1      The primary shell prompt. Note that this value will be
                 ignored if the user's init file (if any) sets sys.ps1.
  --ps2 PS2      The secondary shell prompt. Note that this value will be
                 ignored if the user's init file (if any) sets sys.ps2.
  --shell SHELL  The shell executable (and its initial argument(s)) that
                 should be used to execute UNIX commands. Default is
                 "$DAUDIN_SHELL" if DAUDIN_SHELL is set in your environment,
                 else "$SHELL -c" if SHELL is set in your environment, else
                 "/bin/sh -c".
  --noInit       Do not load the ~/.daudin.py start-up file.
  --noPtys       Do not run any shell commands in pseudo-ttys.
  --debug        Start in debug mode.
  --tracebacks   Print exception tracebacks (implies --debug).

Exiting daudin

Just use control-d as you would in any other shell. Or you can call a Python builtin function exit() or quit(), or use sys.exit() (all of which can be given an int exit status).

How commands are interpreted

daudin first tries to run a command with eval. If that succeeds, the result becomes _ for the next command. If eval fails, code.compile_command is used to try to compile the command. If a full command is found, it is given to exec to execute. If a partial command (e.g., the beginning of a function definition or a dictionary or list etc.) is found, a secondary prompt (sys.ps2) is printed. If a command cannot be compiled or executed, execution is attempted via the shell (/bin/sh) using subprocess. The current _ is provided to the shell on standard input. The output of the shell command, if any, is converted to a Python list of strings (though daudin initially prints this as a single string for the user). The list of strings becomes the next _. If the shell command produces no output, _ is set to [] for the next command (a value of None could be used instead, but it's more consistent to have all shell commands return a list of strings, even if empty).

If a command returns a value (or if None is returned but the command prints something) that value becomes the new pipeline value:

>>> 4
4
>>> _
4
>>> [3, 6, 9]
[3, 6, 9]
>>> print('hello')
hello
>>> echo hello too
hello too
# This echo command actually returns a list of one string, and that is the
# value that _ is set to.  But, as mentioned above, when daudin first
# prints the output from the shell command the lines are joined with '\n'.
>>> _
['hello too']

Shortcoming

Although the above works well almost all the time, it is not perfect. In particular it is possible that you enter a valid Python expression but that eval and exec cannot run it (e.g., len(None)). In that case the command is fed to the shell, which results in an error similar to

>>> len(None)
/bin/sh: 1: Syntax error: word unexpected (expecting ")")

You can turn on debugging via the special %d command (see below for more detail) to dig into what happened:

>>> %d
>>> len(None)
                    Processing 'len(None)'.
                    Not in pipeline.
                    Trying eval 'len(None)'.
                    Could not eval: object of type 'NoneType' has no len().
                    Trying to compile 'len(None)'.
                    Command compiled OK.
                    Could not exec: object of type 'NoneType' has no len().
                    Trying shell 'len(None)' with stdin None.
                    In _shPty, stdin is None
/bin/sh: 1: Syntax error: word unexpected (expecting ")")
                    Shell returned '/bin/sh: 1: Syntax error: word unexpected (expecting ")")\n'

Pipeline Python execution environment

When a Python command is run, it has access to the following:

• cd - a function for changing directory.
• sh - a function for running a shell command.
• self - the instance of daudinlib.pipeline.Pipeline. This allows full access to the internals of the running daudin shell. So you can do things like self.debug = True, and anything else you can think of.

In addition, the variables or functions you define or import in your ~/.daudin.py are also present.

Shell execution environment

By default, daudin will use your SHELL environment variable (with a -c argument) as the shell to execute non-Python commands with. You can specify another shell on the command line via the --shell argument, e.g.,

$ daudin --shell '/bin/bash -c'

or by setting an environment variable, DAUDIN_SHELL to a value such as /bin/bash -c. If you normally use a shell with slow start-up, and invoke daudin from that shell, you will certainly want to use one of these options to set the shell that daudin uses, otherwise shell commands executed by daudin will all be slow.

You can also change the shell in use during a daudin session:

>>> self.shell = ['/bin/bash', '-c']
>>> echo $BASH_VERSION
5.0.3(1)-release
>>> echo testing-{1,2,3}
testing-1 testing-2 testing-3
>>> for i in testing-{1,2,3}; do echo $i; done
testing-1
testing-2
testing-3

>>> self.shell = ['/bin/zsh', '-c']
>>> echo $ZSH_VERSION
5.5.1
>>> echo {2015..2019}
2015 2016 2017 2018 2019
>>> foreach i (testing-{1,2,3}); do echo $i; done
testing-1
testing-2
testing-3

>>> self.shell = ['/usr/local/bin/fish', '-c']
>>> # Slow!!
>>> echo $FISH_VERSION
3.0.0
>>> echo testing-{1,2,3}
testing-1 testing-2 testing-3

>>> self.shell = ['/bin/sh', '-c']
>>> echo testing-{1,2,3}
testing-{1,2,3}

Note that this shell is only used by daudin when it encounters shell commands on a command line, not when you use the built-in sh command. When using sh, you can provide whatever arguments you like, either as a string or a list of strings, to be passed to subprocess.run (or subprocess.Pipe in the case of a pseudotty - see below).

Pseudottys

When a shell command is the final command on a line, it is run in a pseudotty. So commands that check to see if they're running with standard output connected to a terminal will think they are. In that case, a command like git status or ls --color=auto will produce colored output that will be correctly displayed.

Debugging

You can turn on debugging output using the special %d command, or set self.debug to a true value:

>>> self.debug = 1
>>> 4
                    Processing '4'.
                    Trying eval '4'.
                    Eval returned 4.
4

or run self.toggleDebug().

For more information you can enable printing of tracebacks via the %t special command.

There are --debug and --tracebacks command-line options that can be given on daudin invocation to immediately enable debugging and traceback printing.

Special commands

All special % commands have been described above, but here's list of them in one place for reference:

• %cd - change directory.
• %d - toggle debug output.
• %r - reload init file.
• %t - toggle traceback output (also turns on debugging output).
• %u - undo the last change to the _ pipeline variable.

It's worth pointing out that none of these special commands is actually needed. They're just syntactic sugar to make some actions easier. Their effects can all be achieved using regular Python inside daudin (usually via self), if you know what you're doing.

Version

To get the daudin version:

>>> import daudinlib
>>> daudinlib.__version__
0.0.7

Background & thanks

Daudin is the surname of François Marie Daudin, a prolific French zoologist who in 1826 gave the name "Python", to a genus in the Pythonidae family. Pythonidae in turn is a member of the wonderfully named superfamily, Pythonoidea.

I wrote daudin on the evening of Oct 4, 2019 following a discussion about shells with Derek Smith after he overheard me talking to a student. I was saying how awesome the shell is (really meaning its pipelines and the power you get from it falling back onto external programs in $PATH when it encounters a non-keyword).

Derek said he thinks the shell and having to use one in the first place really sucks. He asked two questions, illustrating his strong objections:

1. First, he asked why we have to program the shell in this archaic painful language when we use a completely different language to get our "real" work done. Why can't it all just be one language?

   I told him how I love pipelines and how natural they are. They are completely unlike what we normally do as programmers and that we somehow got used to. In "normal" programming you need to either a) nest functions inside-out (innermost is executed first) in a typical language, or b) use prefix syntax and nested inside-out functions as in Lisp, or c) perhaps worst of all, push all your arguments onto a stack and then use postix, as in a Reverse Polish Notation (RPN) calculator or other stack-based language.

   Unlike in those other environments, in a shell pipeline you indicate what you want to get done in the "natural" (for many people) left-to-right order. You give arguments in a natural place. The shell takes care of making the data flow through the pipeline behind the scenes, hooking up standard input and output between successive commands.

   I mentioned an experimental new shell (nushell) to Derek, that Nelson Minar had recently pointed me to. I told Derek how nushell lets structured data flow along the pipeline.

2. He gave an outraged snort and asked his second deadly question: but why structured data? Why can't a number, a string, a list, an object, a function (or even "structured data"), flow along the pipeline? Derek is a Lisp programmer, and has been for decades. He's never seen a reason to move to any other programming environment. If you know about Lisp and its typical programming environment (not to mention the Lisp Machine) you can easily appreciate why.

I recently wrote a Python RPN calculator that lets you put anything onto the stack and operate on it, so I've been thinking about very general and syntactically easy use of Python like this, but in the context of a stack, not the shell. I also wrote pystdin to allow you to easily process standard input in Python, without having to do so much typing.

With Derek's two provocative questions burning brightly in my mind, I started thinking about how to write a shell that was all Python, with the elegance of shell pipelines (both conceptually and syntactically), and that would allow anything to flow along the pipeline. And, for bonus points, make it easy to use and seamlessly tie in to all the UNIX commands a regular shell provides access to.

Once I had the basic idea of what to write, the code was pretty straightforward to put together due to Python's strong support for parsing, compiling, evaluating, and execing Python code and the nice subprocess library. The rest was just glue and a REPL loop. An initial working version was about 280 lines of code and could be written in one evening. The code is still quite ugly and brittle (only a few tests and various exceptions are either not handled as well as they could be or may even cause daudin to exit). But it works fine as an initial proof of concept.

I find it interesting to note that it feels like daudin generalizes my RPN calculator and pystdin (both mentioned above).

I'm going to try using daudin for real and see what kinds of additional helper functions I end up adding and how things go in general. It's easy to imagine some things, like a smart cd command (I've written quite a few shell cd commands over the years, including a client-server one :-)). The prompt could be a function. Many things could be done with history. Etc.

Thanks for reading, and thanks Derek & Nelson.

Terry Jones (@terrycojones)
terry@jon.es

TODO

Here are some concrete things I'd like to (possibly) add

• Just have one sub-shell and send commands to it instead of forking a new one for each command. That would allow persistent shell variables. cd commands could be run simultaneously in both shells.
• Add some way to deal with standard error?
• Some of what might also be wanted in a pipeline with _ can be done with tee.
• Make it so code can return IGNORE to explicitly preserve the pipeline.
• Guess at auto-indent level for incomplete commands.
• Add a specially-named function that (if defined) is used to produce the prompt.
• Add a specially-named function that (if defined) run after each command (or command-line).
• Add variable export.