Easily run Python at the shell! Magical, but never mysterious.
Run pip install pypyp (note the extra "yp"!)
pyp requires Python 3.8 or above.
pyp will statically analyse the input code to detect undefined variables. Based on what it finds,
it will proceed to transform the AST of the input code as needed. We then compile and execute the
result, or if using --explain, unparse the AST back to source code.
This section will walk you through the details of using pyp, and hopefully replace your needs
for many common shell utilities. For a cheatsheet / tldr, run pyp --help.
Just use one of the magic variables x, l or line to refer to the current line.
# pyp like cut
ls | pyp 'x[:3]'
ps x | pyp 'line.split()[4]'Use the magic variable lines for a list of rstripped lines or stdin for sys.stdin.
# pyp like wc -c
cat /usr/share/dict/words | pyp 'len(stdin.read())'
# pyp like awk
seq 1 5 | pyp 'sum(map(int, lines))'# pyp like sh
echo echo echo | pyp 'subprocess.run(lines[0], shell=True); pass'
# pyp like jq
curl -s 'https://api.github.com/repos/hauntsaninja/pyp/commits?per_page=1' | pyp 'json.load(stdin)[0]["commit"]["author"]'
# pyp like egrep
cat /usr/share/dict/words | pyp 'x if re.search("(p|m)yth", x) else None'For collections, math, itertools, pathlib.Path, pprint.pp, pyp will figure it out even
if you don't use the qualified name.
# pyp like bc
pyp 'sqrt(5)'
# pyp like ${x##*.}
ls | pyp 'Path(x).suffix'# pyp like line numbers
cat setup.py | pyp 'f"{idx+1: >3} {x}"'By default, pyp will print the last expression in your code — except if it evaluates to None
(or the last statement is pass).
And you can always explicitly call print yourself, in which case pyp will stay out of your way.
# pyp like grep
cat /usr/share/dict/words | pyp 'x if "python" in x else None'
cat /usr/share/dict/words | pyp 'if "python" in x: print(x); "this will not be printed"'This makes the output of pyp easier to compose with shell tools.
Again, explicit printing will stop this magic, but pyp makes the function pypprint available if
you do want to explicitly opt back in.
# pyp like tail
ls | pyp 'lines[-10:]'
# pyp like sort
ls | pyp 'sorted(lines)'
ls | pyp 'print(f"Sorting {len(lines)} lines"); pypprint(sorted(lines))'
# pyp like sort | uniq
ls | pyp 'sorted(set(lines))'Note if you run into trouble with semicolons and want a new line (without using a multiline string in your shell), you can just pass another string to pyp. You can also always pipe pyp to pyp!
# pyp like anything!
ps aux | pyp -b 'd = defaultdict(list)' 'user, pid, *_ = x.split()' 'd[user].append(pid)' -a 'del d["root"]' -a 'd'Use --explain or --script and pyp will output a script equivalent to what it would run. This can also serve as a
useful starting point for more complex scripts.
pyp --explain -b 'd = defaultdict(list)' 'user, pid, *_ = x.split()' 'd[user].append(pid)' -a 'del d["root"]' -a 'd'#!/usr/bin/env python3
from collections import defaultdict
from pyp import pypprint
import sys
d = defaultdict(list)
for x in sys.stdin:
x = x.rstrip('\n')
(user, pid, *_) = x.split()
d[user].append(pid)
del d['root']
if d is not None:
pypprint(d)And if your command hits an exception, pyp will reconstruct a traceback into the generated code.
Point the environment variable PYP_CONFIG_PATH to a file containing, for example:
import numpy as np
import tensorflow as tf
from pipetools import *
def p95(data):
return np.percentile(data, 95)
class PotentiallyUsefulClass: ...When attempting to define undefined names, pyp will statically* analyse this file as a source of
possible definitions. This means that if you don't use tf, we won't import tensorflow! And of
course, --explain will show you exactly what gets run (and hence what doesn't!):
pyp --explain 'print(p95(list(map(float, stdin))))'#!/usr/bin/env python3
import sys
import numpy as np
def p95(data):
return np.percentile(data, 95)
stdin = sys.stdin
print(p95(list(map(float, stdin))))Note, importing things from libraries like pipetools in your configuration can allow you to achieve high levels of syntax sugar:
seq 1 110 | pyp 'lines > foreach(int) | where(X > 100) | group_by(X % 3) | sort_by(X[0])'*If you use wildcard imports, we will need to import those modules if there remain undefined
names, though we skip this in the happy path. If this matters to you, definitely don't
from tensorflow import * in your config!
If definitions in your config file depend on magic variables, pyp will substitute them in the way that makes sense. For example, put the following in your config...
n = int(x)
f = x.split()
j = json.load(stdin)
import pandas as pd
csv = pd.read_csv(stdin)...to make pyp easier than ever for your custom use cases:
ps | pyp 'f[3]'
cat commits.json | pyp 'j[0]["commit"]["author"]'
< cities.csv pyp 'csv.to_string()'There's additional documentation and examples at FAQ. If that doesn't answer your question, please open an issue!
Pyed Piper aka Python Power at the Prompt
pypyp takes inspiration (and the command name!) from here.
Pyed Piper had been dead and Python 2 only for a decade
when pypyp was written; it appears it's recently been resurrected. Pyed Piper is further away
from normal Python syntax and APIs than pypyp aims to be. In particular, Pyed Piper has an emphasis
on piping within Python similar to the way you can combine pypyp with pipetools in the
configuration example above.
Pyped is very similar; enough so that I probably wouldn't have written pyp had I known about it. But I'm glad I didn't, since Pyped doesn't do the AST introspection and manipulation that we do. This means:
- Pyped relies on you to pass in flags to tell it what to do, when pyp can unambiguously infer intention from the input.
- It doesn't provide easy automatic printing, or smart printing of iterables and dicts.
- It hardcodes a list of imports and installs some libraries on your system. This project's automatic import will work for any library you use.
- It doesn't have anything like
--explain/--script.
However,
- It has some conveniences, like regex splitting of input, that you'd have to do for yourself here.
- It supports Python 2 and early versions of Python 3.
- It's been around for much longer.
Since writing pyp, it turns out there are more alternatives out there than I thought :-) Some quick notes:
- Most of them rely on the user passing in flags, like Pyped.
- Most of them have limitations around automatic printing, like only being able to automatically print single expressions or not handling iterables and dicts well.
- Some of them have custom syntax for in-process command chaining, which can be convenient.
- Some of them have specialised support for things like JSON input or running shell commands.
- Some of them expose the input in interesting ways with custom line / file / stream objects.
- Some of them have more advanced options for error handling (although none of them have pyp's great tracebacks).
- None of them have powerful configuration like pyp.
- None of them have anything like
--explain.
For whatever it's worth, I've listed the projects above in approximate order of my personal preference.
mario is a featureful take on shell processing with Python. It doesn't use undefined name
detection, instead relying on a pluggable subcommand system. While the subcommands can be more
verbose than pyp, mario makes up some ground by automatic application of functions and a custom
command chaining syntax. The result can feel a little DSL-like, while pyp tries to feel very close
to writing Python.
Consider using mario if:
- You find yourself stringing together long sequences of pyp commands and want to be able to command chain within a single process out of the box.
- You find yourself often needing to reuse complex pyp commands or doing a lot of domain specific shell processing that you wish you could reuse with a single command.
- You want to easily be able to use async functions.
Consider pyp if:
- You want to minimise keystrokes for things that should be quick and easy.
- You want something minimal and lightweight that feels very close to Python. You don't want to have to remember commands.
- You're happy using Python libraries to do domain specific heavy lifting, for easy command chaining
or syntax sugar. You don't mind (or want to be able to) fall back to a script via
--scriptto deal with complexity.
xonsh is a shell whose language is a superset of Python; this is more ambitious and pretty
different from pyp. pyp is easier to use for the one-liner piping use case, but if you need
more Python in your shell, check out xonsh.
If awk works for you, how did you end up here?
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent