26 comments

[ 3.1 ms ] story [ 54.7 ms ] thread
> Space: O(n). No code is ever duplicated, so the one-lined code produced is linear in the size of the input code.

Seems like a non sequitur?

I'm not sure why you think that. Do you mean that normally, in this context "space" means the memory usage of a program, which isn't particularly related to the size of the program?

In which case, yeah, it's a little incongruous. The "space" bit is talking about the textual representation of the program produced, while the "time" bit is talking about the runtime characteristics of the program.

Oh, you're right, I'm using "space" incorrectly. Fixed :)
Not really, what I meant was that it's pretty easy to go over O(n) without ever duplicating any code. Like just imagine if you had a string literal that held all the numbers from 1 to n^2.
For a very limited definition of "any python file".

Still brings me back to highschool, where we toyed with that kind of tricks (someone builds something, the others try to figure out how it works). Fun times!

Yeah, like trying to do a C file without semicolons
Nice.

This made me think of this video with Jim Weirich demonstrating functional patterns, transformations and the Y combinator:

https://vimeo.com/45140590

Highly recommened. It's very related to what seems to be going on here, and actually quite funny.

> Your code was not valid: 'try-except' is an open problem.

Well that was disappointing...

Call me when the results are compliant with PEP8
Brrring
haha, I think the resulting one liners are a little over 80 characters :)
Eh, you can insert newlines wherever you want and it'll still work, plus the "80 char line" rule is at the top of the list of PEP8 rules you can break.
True, but then it's not on one line anymore, and the code likely ends up on more lines. If only there was some other version of the code that took up less lines and characters... :)
> However, since while loops and for loops are implemented with recursion, you might encounter maximum recursion depth exceeded errors during runtime if your loops go on for too long.

Okay, this was enough to scare me away from it :)

is it me or is it continuation passing style ?
+1 for the amusing faq. And I think this is a cool project. I'll use it just to learn better ways to make one-liners for use in other shell scripts and makefiles. Python has always been harder than the other languages I know to make one-liners, but on occasion it's quite useful.
On what occasion would a one-liner be useful? I'm genuinely curious, because I can't think of any such case, but I wouldn't doubt there being a few.
Read the previous sentence :)
Command line use. You grep some files, and think, "man, this really needs this ONE LITTLE FILTER". So, you pipe it to `perl -e "..."`, and soon have a very short program that you iterate on a few times to make sure you have right, and then you pipe it through the rest of the toolchain.

I don't write Perl much anymore, but that was far and away my __favorite__ way to use it. I love coding in Python now, more than I ever liked programming in any language (except perhaps lisp), but I still don't feel I can use it on the command line that way.

In the context of a Python program, I can't see the value of a Lovecraftian lambda (unless maybe in code golf comments?), but I found this article very appealing mainly because it's quite a feat. Doing anything nontrivial in a lambda is a pain in the ass, so the fact that someone wrote a tool to do it is awesome.

(comment deleted)
I in the distant past when I was still in Uni attempted to do something similar, though I never finished it. I think I should take up my project again and try to finish it this time. I got stuck last time trying to figure out how to do a try/catch.

The one big thing I notice this is doing "wrong", is that it is not linear in recursion depth: Each statement increases the stack depth by 1. The way around this is to instead sequence the commands using arrays, since python guarantees the execution order is left to right. So his example could have been rewritten as:

  (lambda s: [s.__setitem__('x', 3), s.__setitem__('y', 4), print(s['x'] < s['y'] < 5)])({})
(though for this example you could just modify the result of globals())

Then you can also do trickery to get for and while loops into generator expressions as opposed to lambdas. The hard part is handling continues and breaks. For instance:

  while <cond>:
    <stmts>
can get compiled into something like

  list(<compiled_stmts> for _ in itertools.takewhile(lambda _: <compiled_cond>, itertools.cycle([1])))
(though you'll want to replace list with a function that evaluates the iterator but doesn't take up O(N) space where N is the number of iterations, something like a no-op accumulator)

To handle breaks and continues, you'd have to make your <compiled_stmts> be aware that a break/continue was called (a flag on some meta table was my plan), so they can terminate early. You'd also have to change <compiled_cond> to terminate the loop (on a break) and reset the continue flag.

Why pass up a perfectly good excuse to abuse lambda functions, ternary expressions, list comprehensions, and even the occasional Y combinator? Never pass up an opportunity to use the Y combinator.

^^ Legendary

First try: Your code could not be one-lined. Open problem: try-except

Second try: Your code could not be one-lined. Open problem: yield

Third try: Your code could not be one-lined. Open problem: raise

Fourth try: Your code could not be one-lined. Not yet implemented: classdef

So, any file without class definitions, try/except/raise/yield (and probably some more). Not my definition of 'any' :)