allcombinations - generating combinations of python structures

Alright, on a whim I decided to make another tricky thing in Python. This one is less of a hack, and is more likely to be useful.

So let's say you want to do something with all combinations of... something.

[5, 6, oneOf(7,8,9), oneOf(10, 11, "shazaam")]

So you want to turn this structure into all the possibilities represented within:

[
[5, 6, 7, 10],
[5, 6, 7, 11],
[5, 6, 7, "shazaam"],
[5, 6, 8, 10],
[5, 6, 8, 11],
[5, 6, 8, "shazaam"],
[5, 6, 9, 10],
[5, 6, 9, 11],
[5, 6, 9, "shazaam"],
]

Well with allcombinations, you can do just that:

from allcombinations import allCombinations

allCombinations( [5, 6, oneOf(7,8,9), oneOf(10, 11, 12)] )

Here it is. The gist includes more complicated example.

Features
The oneOf should be able to reside almost anywhere in your expression. It can be in a list (as seen here), in a dict, or even in the attribute of an object. It can also reside in a list within a dict within an object's attribute, etc, as long as it's nowhere within an unsupported container type.

Limitations:
This will only work if oneOf resides in a list, dict, or an object's attributes. It shouldn't work anywhere within a set, or any other structure I can't think of. If you try it in something unsupported, the oneOf object should just stick around in all your combinations.

I'm probably actually going to use this, particularly (again) for testing. Anyone else think they'd find it useful? Should I package it?

Testing (Django) views with pyquery

PyQuery is basically what it sounds like. Using jQuery syntax, you can query and even manipulate XML files. Obviously we don't (yet!) have Python in the browser, so it's not useful in the same domain, but it can help out in dealing with XML in general, in the same way as, say, lxml, but without having to learn about things like ElementTree for simple cases. It's particularly good for XHTML because jQuery (and thus PyQuery) uses CSS syntax for class= and id=. Which brings me to how I'm using it:

from django.test.client import Client
from pyquery import PyQuery
from django.test.testcases import TestCase

...

class TestSomeViews(TestCase):

   def testAView(self):

       client = Client()

       ...

       response = client.get("/someurl/")

       self.assertTrue("expected text" in PyQuery(response.content)("#someid").html() )

(If you're unfamiliar with testing Django views, see this.)

For some basic tests, you can just search the entire response html, and not have to worry about where it shows up. But suppose you're searching for a username in a particular part of your response. You're pretty likely to find that username elsewhere on the page, so you have to select out the part of the file you expect it. I think this is much easier than using a regex.

So what this bit of PyQuery does is find the tag with the id of "someid" (presumably there's only only one, being an id), and returns the html within that tag. (If you search for a class that returns multiple tags, it seems that a simple call to .html() will only return the contents of the first one. This very well may match jQuery's behavior, I'm admittedly not that familiar, but just a head's up.) For more details look at the PyQuery API.

Django Model Validation

I'm really excited about model validation, in Django circa 1.2. It's going to save me from the disastrous hack of ModelForms I've used so far. One wants something to validate data before saving it, since validating it by saving it is apparently a Bad Idea, since you may have already made changes to the database by the time the error is thrown.

I wanted this done automatically, and Django only had form validation until 1.2, so I hacked forms into some sort of all-purpose wrapper for models. I've since learned to program Django like an adult, and model validation came around just in time anyway. But there's a problem I have with it.

First I should point out that there's a few different things that get validated, but the two I'll point out are A) checking that required fields are set and B) whatever I put into the clean() function. Judging by some errors I've gotten while debugging/testing, Django seems to be checking B before A.

Now, when I use a ModelForm, sometimes I want to use the commit=False option when I save. This returns a model that hasn't been committed to the database. Sometimes there's extra data I want to add to the model that the form didn't supply. Sometimes that data is in fact necessary for the model to be valid. So clearly Django shouldn't check A, and it doesn't. Here's the funny thing though: it does check B. Why would it do that? I can understand checking when I call is_valid(), I can control when that's called, making sure I added everything first.

So far the consequences of checking B early have been trying to access members that haven't yet been set, in my clean() function. So in clean() I just check for those items, and just let it pass if they don't exist. I figure, yeah, it'll pass certain tests when it shouldn't. But if it's really running through validation (not just B as in the save (commit=False) case) that means it'll catch the missing members on the A pass anyway.

Maybe I got something wrong, but I thought it was a weird design decision.

Significant Whitespace in Python Data Structures

I recently wrote a program in Python for parsing files. I'm pretty naive still when it comes to functional programming, but I'm still excited about it, so I wanted it to be more functional in style. It had a complicated data structure representing the file structure, instead of a loop with bunch of if-thens. By Python standards I may have gone a bit overboard. Guido probably would not have approved of my code (not to mention what follows in this blog post).

So as a result, most of the program became whitespace irrelevant. Huge dicts of lists of tuples, etc. It made me think that relevant whitespace might become handy for data too. And while talking on IRC about it this morning I realized I could sortof hack it using decorators and generators. So here's what it looks like:

http://gist.github.com/611646

So I have two examples:

• example.py - This shows how you can define a more complicated structure with whitespace instead of a bunch of ){(}[].
  

• inlinefunc.py - This demonstrates a sort of side-effect benefit. You can have multi-line functions inline in a list (or tuple or dict). Usually you're stuck with lambdas, and of course that starts to look confusing too.

I'd like to clean up the syntax. Obviously having to have a decorator before and a yield after isn't great. I'll have to think about how I could do that. Maybe make an even dirtier hack by doing introspection.

Any thoughts?

A useful script: copytoclipboard

Another quick tool I figure I'd share. Made this one a while ago. When you're doing stuff on the command line and you need to copy something for a desktop app, you end up having to select text on your terminal screen. Well if it's a big text file, you have to copy, paste, scroll to the next part, copy (make sure you didn't copy any part twice), paste, etc.

Well, copytoclipboard takes standard input and puts it into the gtk clipboard (haven't tried this on KDE).

#!/usr/bin/env python

import sys

import pygtk
pygtk.require('2.0')
import gtk

c = gtk.Clipboard()
c.set_text(''.join(sys.stdin.read()))
c.store()

For instance, to paste this code, I went into my terminal and typed:

cat copytoclipboard | copytoclipboard

and pasted the results here. For a less confusing example:

uptime | copytoclipboard

lets me paste this:

17:42:14 up 3:23, 2 users, load average: 0.35, 0.39, 0.41

EDIT: Or I could just do a google search and find that xclip already exists. Oh well.

EDIT2: No, xclip seems to be its own thing. I just tried it, doesn't seem to work with gtk. I guess I still win!

"New Music" script

I just thought of a script (can't test it at work so I'll actually write it later, but it'd be only a few lines long). Would work thusly:

Desktop>newmusic britney_spears_leak_trance_remix.ogg ~/music/various/

It would act like mv, just move the song over to the "various" directory, but it would also add a symlink to

~/music/new/

That way, I'll remember to give it a listen. And when it's no longer novel I can just delete the symlink. Otherwise this crap gets accumulated on my Desktop. Maybe happens to you too.

Maybe it could be made into a Nautilus script or something, but that seems a bit more tricky if you're dealing with drag n drops.

Making new twitter followers marginally more convenient

Well, one tiny annoyance that I just randomly decided I felt like itching - I get a handful of emails from twitter saying so-and-so is following me. Usually it's somebody I'm not interested in, but on occasion it's a friend. What annoys me is that I have to open their profile to see. I'd prefer to see the bio and their last few tweets in the email itself. Would help me decide much faster. Well, I've started on a real hack of a solution. It should make it a little less annoying:

http://gist.github.com/304335

Look it over. If you trust it, run it on a command prompt (works on Linux, all I can say). It asks for your gmail password. It opens browser tabs for all the twitter accounts that have followed you that are still in your gmail inbox.

Now, I'm making some wild assumptions about the emails that Twitter is sending us (namely that the first Twitter url is the profile in question), so this may not open up everything correctly. Especially if Twitter changes the email. So, I also open up a tab with a gmail search for all those emails. You can look it over and confirm that it opened up the right tabs, plus this helps you archive them right away.

If it doesn't work, do this first:

https://www.google.com/accounts/DisplayUnlockCaptcha

I guess it tells Google that your computer isn't engaging in any anti-human activities. I had to do it.

I may eventually have it generate an html page with all the useful info right there, so you can look it over quicker. I'll also look for a way to have a GUI password entry thing, so you won't need a cmd prompt. Or, you know, if any of you Open Sourcerers want to do it yourself and send back the update, that would be great too.