Skip to main content

Python one-liners in the shell in the spirit of Perl and AWK

Project description

Python one-liners in the spirit of Perl and AWK.

pyfil stands for PYthon FILter. One of the tenants of the Unix design is that every program is a filter. It’s especially obvious of programs, like grep, sed, sort, tr, etc.

One notable example is awk – a Turing-complete, interpreted language for parsing text. While AWK scripts are still in use and it’s a fine language, it has been superseded for parsing scripts by more general languages like Perl and later Python and Ruby. However, AWK was designed to be especially useful in the shell as a filter, and it is still in very commonly used for that today (in part because it is on every *nix system, but also because it’s great at what it does). AWK is able to be any arbitrary text filter that doesn’t come as a coreutil. perl -e is also quite good as a filter, and Ruby has made a valiant attempt to do so as well.

While python does have a few good one-line uses (python -m http.server), some elements of its design make it less suited than the afore-mentioned languages. pyfil is one of several attempts to address this issue. In particular, it takes a lot of queues in the design of its CLI from AWK and Perl, and aims fundamentally to be a capable text filter, though it will evaluate any arbitrary Python expression and print its value (with modules being imported implicitly as required).

As a more modern touch, it also has a special emphasis on interoperability with JSON. If the return value of the evaluated expression is a container type, python will attempt to serialize it as JSON before printing, so you can pipe output into other tools that deal with JSON, store it to a file for later use, or send it over http. This, combined with the ability to read JSON from stdin (with –json) make pyfil a good translator between the web, which tends to speak JSON these days, and the POSIX environment, which tends to think about data in terms of lines in a file (frequently with multiple fields per line).

pyfil is in pypi (i.e. you can get it easily with pip, if you want)

pyfil has only been tested with python3, and only has wheels available for python3

similar projects

pyfil ain’t the first project to try something like this. Here are some other cracks at this problem:

Don’t worry. I’ve stolen some of their best ideas already, and I will go on stealing as long as it takes!


pyfil [-h] [-l] [-x] [-q] [-j] [-o] [-b PRE] [-e POST] [-s] [-F PATTERN]
      [-n STRING] [-R] [-S] [-H EXCEPTION_HANDLER]
      expression [expression ...]
positional arguments:
expression expression(s) to be executed. If multiple expression
arguments are given, and –exec is not used, the value of the previous expression is available as ‘x’ in the following expression. if –exec is used, all assignment must be explicit.
optional arguments:
-h, --help show this help message and exit
-l, --loop for n, i in enumerate(stdin): expressions
-x, --exec use exec instead of eval. statements are allowed, but automatic printing is lost. doesn’t affect –post
-q, --quiet suppress automatic printing. doesn’t affect –post
-j, --json load stdin as json into object ‘j’; If used with –loop, treat each line of stdin as a new object
-J, --real-dict-json
 like -j, but creates real dictionaries instead of the wrapper that allows dot syntax.
-o, --force-oneline-json
 outside of loops and iterators, objects serialzed to json print with two-space indent. this forces this forces all json objects to print on a single line.
-b PRE, --pre PRE
 statement to evaluate before expression args. multiple statements may be combined with ‘;’. no automatic printing
-e POST, --post POST
 expression to evaluate after the loop. always handeled by eval, even if –exec, and always prints return value, even if –quiet. implies –loop
-s, --split split lines from stdin on whitespace into list ‘f’. implies –loop
-F PATTERN, --field-sep PATTERN
 regex used to split lines from stdin into list ‘f’. implies –loop
-n STRING, --join STRING
 join items in iterables with STRING
-R, --raise-errors
 raise errors in evaluation and stop execution (default: print message to stderr and continue)
-S, --silence-errors
 suppress error messages
 specify exception handler with the format ‘Exception: alternative expression to eval’

available objects

pyfil automatically imports any modules used in expressions.

If you’d like to create any other objects to use in the execution environment ~/.config/ and put things in it.

default objects:

  • l = []
  • d = {}

These are empty containers you might wish to add items to during iteration, for example.

  • x is the value of the previous expression unless –exec was used.

The execution environment also has a special object for stdin, creatively named stdin. This differs from sys.stdin in that it strips trailing newlines when you iterate over it, and it has a property, stdin.l, which returns a list of the lines (without newlines). If you do want the newlines, access sys.stdin directly.

stdin inherits the rest of its methods from sys.stdin, so you can use to get a string of all lines, if that’s what you need.

Certain other flags may create additional objects in the evaluation context.

  • –loop (or anything that implies –loop) create n and i.
  • –json creates j.
  • –split or –field_sep create f

Check the flag descriptions for further details.


automatic printing

By default, pyfil prints the return value of expressions. Different types of objects use different printing conventions.

  • None does not print (as in the REPL)
  • strings are sent directly to to print()
  • iterators (not other iterables) print each item on a new line.
  • other objects are serialized as json. If an object cannot be serialized as json, it is sent directly to print().
  • all of these are overridden by –join

Iterators will also try to serialize each returned object as json if they are not strings. json objects will be indented if only one object is being printed. If –loop is set or several of objects are being serialzed from an iterator, it will be one object per-line. –force-oneline-json extends this policy to printing single json objects as well.


$ # None gets skipped
$ pyfil None
$ # strings and numbers just print
$ pyfil sys.platfrom
$ pyfil math.pi
$ # objects try to print as json
$ pyfil sys.path
$ pyfil '{i: n for n, i in enumerate(sys.path)}'
  "/usr/lib/python3.5/plat-linux": 3,
  "/usr/lib/": 1,
  "/usr/lib/python3.5": 2,
  "/usr/lib/python3.5/lib-dynload": 4,
  "/usr/lib/python3.5/site-packages": 6,
  "/home/ninjaaron/.local/lib/python3.5/site-packages": 5,
  "/home/ninjaaron/.local/bin": 0
$ # unless they can't
$ pyfil '[list, print, re]'
[<class 'list'>, <built-in function print>, <module 're' from '/usr/lib/python3.5/'>]
$ # iterators print each item on a new line, applying the same conventions
$ pyfil 'iter(sys.path)'
$ pyfil '(i.split('/')[1:] for i in sys.path)'
["home", "ninjaaron", "src", "py", "pyfil", "venv", "bin"]
["home", "ninjaaron", "src", "py", "pyfil"]
["usr", "lib", ""]
["usr", "lib", "python3.5"]
["usr", "lib", "python3.5", "plat-linux"]
["usr", "lib", "python3.5", "lib-dynload"]
["home", "ninjaaron", "src", "py", "pyfil", "venv", "lib", "python3.5", "site-packages"]

Most JSON is also valid Python, but be aware that you may occasionally see null instead of None along with true and false instead of True and False, and your tuples will look like list. I guess that’s a risk I’m willing to take. (The rational for this is that pyfil is more about composability in the shell than printing valid Python literals. JSON is becoming the defacto standard for serialization.)

suppressing output and using statements

Because these defaults use eval() internally to get value of expressions, statements may not be used. exec() supports statements, but it does not return the value of expressions when they are evaluated. When the -x/–exec flag is used, automatic printing is suppressed, and expressions are evaluated with exec, so statements, such as assignments, may be used. Values may still be printed explicitly.

–quite suppresses automatic printing, but eval is still used.

The –post option is immune from –quiet and –exec. It will always be evaluated with eval(), and it will always try to print. The only difference is that if –quiet or –exec was used, json will be printed with indentation unless –force-oneline-json is used.

using files for input and output

pyfil doesn’t have any parameters for input and output files. Instead, use redirection.

pyfil -s 'i.upper()' > output.txt < input.txt

using multiple expression arguments

pyfil can take as many expressions as desired as arguments. When used with –exec, this works pretty much as expected, and assignment must be done manually.

Without –exec, the return value of each expression is assigned to the variable x, which can be used in the next expression. The final value of x is what is ultimately printed, not any intermediate values.

$ pyfil 'reversed("abcd")' 'i.upper() for i in x'

looping over stdin

one can do simple loops with a generator expression. (note that any expression that evaluates to an iterator will print each item on a new line unless the --join option is specified.)

$ ls / | pyfil 'i.upper() for i in stdin'

However, with the -l/--loop flag, pyfil loops over stdin in a context like this:

for n, i in enumerate(stdin):

Therefore, the above loop can also be written thusly:

$ ls / | pyfil -l 'i.upper()'

--pre and --post (-b and -e) options can be used to specify actions to run before or after the loop. Note that the –pre option is run with exec instead of eval, and therefore output is never printed, and statements may be used. This is for things like initializing container types. –post is automatically printed and statements are not allowed (even if –exec is used). –loop is implied if --post is used. --pre can be used without a –loop to do assignments (or whatever else you may want to do with a statement).

Using -s/--split or -F/--field-sep for doing awk things also implies –loop. The resulting list is named f in the execution environment, in quazi-Perl fashion. (oh, and that list is actually a subclass of collections.UserList that returns an empty string if the index doesn’t exist, so it acts more like awk with empty fields, rather than throwing and error).

json input

pyfil can parse json objects from stdin with the -j/--json flag. They are passed into the environment as the j object. combining with the –loop flag will treat stdin as one json object per line. json objects support dot syntax for attribute access, e.g. j.someattr.attr_of_someattr

There are occasionally functions that require “real” dictionaries and won’t work with this special subclass that supports dot access. In such cases, use -J/--real-dict-json to get unadultered Python dictionaries.

formatting output (and ‘awk stuff’)

It’s probably obvious that the most powerful way to format strings is with Python’s str.format method and the -F or -s options.

$ ls -l /|pyfil -s '"{0}\t{2}\t{8}".format(*f)'
IndexError: tuple index out of range
lrwxrwxrwx    root    bin
drwxr-xr-x    root    boot/
drwxr-xr-x    root    dev/
drwxr-xr-x    root    etc/
drwxr-xr-x    root    home/
lrwxrwxrwx    root    lib

However, you will note that using string.format(*f) produces an error and does not print anything to stdout (error message is sent to stderr; see error handling for more options) for lines without enough fields, which may not be the desired behavior when dealing with lines containing arbitrary numbers of fields.

For simpler cases, you may wish to use the -n/--join option, which will join any iterables with the specified string before printing, and, in the case of the f list, will replace any none-existent fields with an empty string.

$ ls -l /|pyfil -sn '\t' 'f[0], f[2], f[8]'
lrwxrwxrwx    root    bin
drwxr-xr-x    root    boot/
drwxr-xr-x    root    dev/
drwxr-xr-x    root    etc/
drwxr-xr-x    root    home/
lrwxrwxrwx    root    lib

In this case, the first line of ls -l / provides values for all available fields.

Technical note:
The separator specified with the --join option is implemented internally as ast.literal_eval("'''"+STRING.replace("'", r"\'")+"'''"). If one works hard at it, it is possible to pass values which will cause pyfil to crash; i.e. patterns ending with a backslash. Keep in mind rules about escape sequences in the shell and in python if you absolutely must have a pattern that terminates with a backslash. (The reason it is implemented this way is to allow the use of escape sequences that are meaningful to the python, but not the shell, such as \n, \t, \x, \u, etc.)


I realize that it’s much better to do most of these things with the original utility. This is just to give some ideas of how to use `pyfil`

replace wc -l:

$ ls / | pyfil 'len(stdin.l)'

replace fgrep:

$ ls / | pyfil '(i for i in stdin if "v" in i)'
$ ls / | pyfil -l 'i if "v" in i else None'

replace grep:

$ ls / | pyfil 'filter(lambda x:"^m", x), stdin)'
$ ls / | pyfil -lS '"^m", i).string)'
$ # using the -S option to suppress a ton of error messages

replace sed 's/...:

$ ls / | pyfil -l 're.sub("^([^aeiou][aeiou][^aeiou]\W)", lambda m:, i)'

This example illustrates that, while you might normally prefer sed for replacement tasks, the ability to define a replacement function with re.sub does offer some interesting possibilities. Indeed, someone familiar with coreutils should never prefer to do something they already comfortable doing the traditional way with pyfil (coreutils are heavily optimized). Python is interesting for this use-case because it offers great logic, anonymous functions and all kinds of other goodies that only full-fledged, modern programming language can offer. Use coreutiles for the jobs they were designed to excel in. Use pyfil to do whatever they can’t… and seriously, how will coreutils do this?:

$ wget -qO- | pyfil -j 'j.urls[0].filename'
$ ls -l | pyfil -qSs \
"d.update({f[8]: {'permissions': f[0], 'user': f[2], 'group': f[3],
                  'size': int(f[4]), 'timestamp': ' '.join(f[5:8])}})" \
--post 'd'
  "README.rst": {
    "group": "users",
    "user": "ninjaaron",
    "permissions": "-rw-r--r--",
    "timestamp": "Sep 6 20:55",
    "size": 18498
  "pyfil/": {
    "group": "users",
    "user": "ninjaaron",
    "permissions": "drwxr-xr-x",
    "timestamp": "Sep 6 20:20",
    "size": 16
  "": {
    "group": "users",
    "user": "ninjaaron",
    "permissions": "-rw-r--r--",
    "timestamp": "Sep 6 20:30",
    "size": 705
  "LICENSE": {
    "group": "users",
    "user": "ninjaaron",
    "permissions": "-rw-r--r--",
    "timestamp": "Sep 3 13:32",
    "size": 1306

Other things which might be difficult with coreutils:

$ ls / | pyfil -n '  ' 'reversed(stdin.l)'
var/  usr/  tmp/  sys/  srv/  sbin@  run/  root/  proc/  opt/  ...
$ # ^^ also, `ls /|pyfil -n '  ' 'stdin.l[::-1]'

error handling

If pyfil encounters an exception while evaluating user input the default is to print the error message to stderr and continue (if looping over stdin), as we saw in the section on formatting output. However, errors can also be silenced entirely with the -S/--silence-errors option. In the below example, the first line produces an error, but we don’t hear about it.

$ ls -l /|pyfil -sS '"{0}\t{2}\t{8}".format(*f)'
lrwxrwxrwx    root    bin
drwxr-xr-x    root    boot/
drwxr-xr-x    root    dev/
drwxr-xr-x    root    etc/
drwxr-xr-x    root    home/
lrwxrwxrwx    root    lib

Alternatively, errors may be raised when encountered, which will stop execution and give a (fairly useless, in this case) traceback. This is done with the -R/--raise-errors flag.

$ ls -l /|pyfil -sR '"{0}\t{2}\t{8}".format(*f)'
Traceback (most recent call last):
  File "/home/ninjaaron/src/py/pyfil/venv/bin/pyfil", line 9, in <module>
    load_entry_point('pyfil', 'console_scripts', 'pyfil')()
  File "/home/ninjaaron/src/py/pyfil/pyfil/", line 242, in main
    run(expressions, a, namespace)
  File "/home/ninjaaron/src/py/pyfil/pyfil/", line 164, in run
    handle_errors(e, args)
  File "/home/ninjaaron/src/py/pyfil/pyfil/", line 134, in handle_errors
    raise exception
  File "/home/ninjaaron/src/py/pyfil/pyfil/", line 162, in run
    value = func(expr, namespace)
  File "<string>", line 1, in <module>
IndexError: tuple index out of range

In addition to these two handlers, it is possible to specify a rudimentary custom handler with the -H/--exception-handler flags. The syntax is -H 'Exception: expression', where Exception can be any builtin exception class (including Exception, to catch all errors), and expression is the alternative expression to evaluate (and print, if not –quiet).

$ ls -l /|pyfil -sH 'IndexError: i' '"{0}\t{2}\t{8}".format(*f)'
total 32
lrwxrwxrwx    root    bin
drwxr-xr-x    root    boot/
drwxr-xr-x    root    dev/
drwxr-xr-x    root    etc/
drwxr-xr-x    root    home/
lrwxrwxrwx    root    lib

In this case, we’ve chosen to print line without any additional formatting. If other errors are encountered, it will fall back to other handlers (-S, -R, or the default). For more sophisticated error handling… write a real Python script, where you can handle to your heart’s content.

Also note that this case is possible to handle with a test instead of an exception handler because f is a special list that will return an empty string instead of throw an index error if the index is out of range:

ls -l / | pyfil -s '"{0}\t{2}\t{8}".format(*f) if f[2] else i'


Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
pyfil-1.9.2-py3-none-any.whl (20.3 kB) Copy SHA256 hash SHA256 Wheel py3
pyfil-1.9.2.tar.gz (18.3 kB) Copy SHA256 hash SHA256 Source None

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page