Subscribe: Ned Batchelder's blog
Added By: Feedage Forager Feedage Grade B rated
Language: English
batt  def  dict  eighths  fibonacci  find  good  level  line  numbers  pairs  print  python  return  rules  small  text  triangular 
Rate this Feed
Rate this feedRate this feedRate this feedRate this feedRate this feed
Rate this feed 1 starRate this feed 2 starRate this feed 3 starRate this feed 4 starRate this feed 5 star

Comments (0)

Feed Details and Statistics Feed Statistics
Preview: Ned Batchelder's blog

Ned Batchelder's blog

Ned Batchelder's personal blog.


New backups: Arq to Wasabi


This week CrashPlan announced they were ending consumer services, so I had to replace it with something else. Backups are one of those things at the unpleasant intersection of tedious, difficult, and important.

A quick spin around the latest alternatives showed the usual spectrum of possibilities, ranging from perl hackers implementing rsync themselves, to slick consumer tools. I need to have something working well not just on my computer, but others in my family, so I went the consumerish route.

Arq backing up to Wasabi seems like a good choice for polish and price.

One thing I always struggle with: how to ensure my stuff is backed up, without needlessly copying around all the crap that ends up in my home directory that I don't need backed up. On a Mac, the ~/Library directory has all sorts of stuff that I think I don't need to copy around. Do I need these?:

  • Library/Application Support
  • Library/Caches
  • Library/Containers

I add these directories to the exclusions. Should my Dropbox folder get backed up? Isn't that what Dropbox is already doing?

Then as a developer, there's tons more to exclude. Running VirtualBox? You have have a 10Gb disk image somewhere under your home. I have something like 20,000 .pyc files. The .tox directory for is 350Mb.

So I also exclude these:

  • .git
  • .hg
  • .svn
  • .tox
  • node_modules
  • .local
  • .npm
  • .vagrant.d
  • .vmdk
  • .bundle
  • .cache
  • .heroku
  • .rbenv
  • .gem
  • *.pyc
  • *.pyo
  • *$py.class

Of course, as a native Mac app for consumers, Arq doesn't provide a way that I can supply all these once, I have to fiddle with GUI + and - buttons, and enter them one at a time...

Lastly, some files don't seem comfortable with backups. Thunderbird's storage files are large, and while Arq copies only certain byte ranges, they still amount to about 300Mb each time. Should I even back up my email? Should I still be using Thunderbird? Too many uncertainties.... podcast


I was a guest on the Podcast.__init__ podcast this week: with Ned Batchelder. We talk about, how I got started on it, why it's good, why it's not good, how it works, and so on:

And in the unlikely case that you want yet more of my dulcet tones, I was also on the Python Test podcast, mentioned in this blog post: The Value of Unit Tests.

Look around you


I've been trying my Instagram experiment for a year now. I've really liked doing it: it gives me a prompt for looking around me and seeing what there is to see. One of the things that surprised me when I looked back is how many pictures I took in a very small area, one that I would have thought of as uninteresting: the few blocks between where I swim in the morning, and where I work. Of the 197 pictures I took in the last year, 38 of them are from that neighborhood: I'm not saying these are all masterpieces. But I wouldn't have thought to take them at all if I hadn't been explicitly looking for something interesting to shoot.Look around you: not much is truly uninteresting. [...]

Finding fuzzy floats


For a 2D geometry project I needed to find things based on 2D points. Conceptually, I wanted to have a dict that used pairs of floats as the keys. This would work, except that floats are inexact, and so have difficulty with equality checking. The "same" point might be computed in two different ways, giving slightly different values, but I want them to match each other in this dictionary.I found a solution, though I'm surprised I didn't find it described elsewhere.The challenges have nothing to do with the two-dimensional aspect, and everything to do with using floats, so for the purposes of explaining, I'll simplify the problem to one-dimensional points. I'll have a class with a single float in it to represent my points.First let's look at what happens with a simple dict:>>> from collections import namedtuple >>> Pt = namedtuple("Pt", "x") >>> >>> d = {} >>> d[Pt(1.0)] = "hello" >>> d[Pt(1.0)] 'hello' >>> d[Pt(1.000000000001)] Traceback (most recent call last):   File "", line 1, in  KeyError: Pt(x=1.000000000001) As long as our floats are precisely equal, the dict works fine. But as soon as we get two floats that are meant to represent the same value, but are actually slightly unequal, then the dict is useless to us. So a simple dict is no good.To get the program running at all, I used a dead-simple structure that I knew would work: a list of key/value pairs. By defining __eq__ on my Pt class to compare with some fuzz, I could find matches that were slightly unequal:>>> class Pt(namedtuple("Pt", "x")): ...     def __eq__(self, other): ...         return math.isclose(self.x, other.x) ... >>> def get_match(pairs, pt): ...     for pt2, val in pairs: ...         if pt2 == pt: ...             return val ...     return None ... >>> pairs = [ ...     (Pt(1.0), "hello"), ...     (Pt(2.0), "goodbye"), ... ] >>> >>> get_match(pairs, Pt(2.0)) 'goodbye' >>> get_match(pairs, Pt(2.000000000001)) 'goodbye' This works, because now we are using an inexact closeness test to find the match. But we have an O(n) algorithm, which isn't great. Also, there's no way to define __hash__ to match that __eq__ test, so our points are no longer hashable.Trying to make things near each other be equal naturally brings rounding to mind. Maybe that could work? Let's define __eq__ and __hash__ based on rounding the value:>>> class Pt(namedtuple("Pt", "x")): ...     def __eq__(self, other): ...         return round(self.x, 6) == round(other.x, 6) ...     def __hash__(self): ...         return hash(round(self.x, 6)) ... >>> d = {} >>> d[Pt(1.0)] = "apple" >>> d[Pt(1.0)] 'apple' >>> d[Pt(1.00000001)] 'apple' Nice! We have matches based on values being close enough, and we can use a dict to [...]

Triangular Fibonacci numbers


Yesterday in my post about 55, I repeated Wikipedia's claim that 55 is the largest number that is both triangular and in the Fibonacci sequence. Chris Emerson commented to ask for a proof. After a moment's thought, I realized I had no idea how to prove it.The proof is in On Triangular Fibonacci Numbers, a dense 10-page excursion into number theory I don't understand.While I couldn't follow the proof, I can partially test the claim empirically, which leads to fun with Python and itertools, something which is much more in my wheelhouse.I started by defining generators for triangular numbers and Fibonacci numbers:def tri():     """Generate an infinite sequence of triangular numbers."""     n = 0     for i in itertools.count(start=1):         n += i         yield n          print(list(itertools.islice(tri(), 50))) [1, 3, 6, 10, 15, 21, 28, 36, 45, 55, 66, 78, 91, 105, 120, 136, 153, 171, 190, 210, 231, 253, 276, 300, 325, 351, 378, 406, 435, 465, 496, 528, 561, 595, 630, 666, 703, 741, 780, 820, 861, 903, 946, 990, 1035, 1081, 1128, 1176, 1225, 1275]def fib():     """Generate an infinite sequence of Fibonacci numbers."""     a, b = 1, 1     while True:         yield a         b, a = a, a+b          print(list(itertools.islice(fib(), 50))) [1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, 1597, 2584,  4181, 6765, 10946, 17711, 28657, 46368, 75025, 121393, 196418, 317811,  514229, 832040, 1346269, 2178309, 3524578, 5702887, 9227465, 14930352,  24157817, 39088169, 63245986, 102334155, 165580141, 267914296, 433494437,  701408733, 1134903170, 1836311903, 2971215073, 4807526976, 7778742049,  12586269025, 20365011074] The Fibonacci sequence grows much faster!My first impulse was to make two sets of the numbers in the sequences, and intersect them, but building a very large set took too long. So instead I wrote a function that took advantage of the ever-increasing nature of the sequences to look for equal elements in two monotonic sequences:def find_same(s1, s2):     """Find equal elements in two monotonic sequences."""     try:         i1, i2 = iter(s1), iter(s2)         n1, n2 = next(i1), next(i2)         while True:             while n1 < n2:                 n1 = next(i1)             if n1 == n2:                 yield n1                 n1 = next(i1)       [...]

Math factoid of the day: 55


55 is in the Fibonacci sequence:

1 1 2 3 5 8 13 21 34 55 ...

55 is a triangular number:

1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10 = 55

It is the largest number that is both Fibonacci and triangular.

It is also a Kaprekar number:

55² = 3025 and 30+25 = 55

Re-ruling .rst


Sometimes, you need a small job done, and you can write a small Python program, and it does just what you need, and it pleases you. I have some Markdown files to convert to ReStructured Text. Pandoc does a really good job. But it chooses a different order for heading punctuation than our house style, and I didn't see a way to control it.But it was easy to write a small thing to do the small thing:import re import sys # The order we want our heading rules. GOOD_RULES = '#*=-.~' # A rule is any line of all the same non-word character, 3 or more. RULE_RX = r"^([^\w\d])\1\1+$" def rerule_file(f):     rules = {}     for line in f:         line = line.rstrip()         rule_m =, line)         if rule_m:             if line[0] not in rules:                 rules[line[0]] = GOOD_RULES[len(rules)]             line = rules[line[0]] * len(line)         print(line) rerule_file(sys.stdin) If you aren't conversant in .rst: there's no fixed order to which punctuation means which level heading. The first rule encountered is heading 1, the next style found is heading 2, and so on.There might be other ways to do this, but this makes me happy. [...]

Shell = Maybe


A common help Python question: how do I get Python to run this complicated command line program? Often, the answer involves details of how shells work. I tried my hand at explaining it what a shell does, why you want to avoid them, how to avoid them from Python, and why you might want to use one: Shell = Maybe.

Text-mode menu bar indicators


I recently upgraded my Mac operating system, and decided to try out a new feature: automatically hiding the menu bar. This gives me back another sliver of vertical space. But it has a drawback: I no longer have the time, battery life, and speaker volume indicators available at a glance.I went looking for a thing that I figured must exist: a Mac app that would display that information in a dock icon. I already have a dock clock. I found a dock battery indicator, though it tried so hard to be cute and pictorial, I couldn't tell what it was telling me.Asking around, I got a recommendation for GeekTool. It lets you draw a panel on your desktop, and then draw in the panel with the output of a script. Now the ball was back in my court: I could build my own thing.I'd long ago moved the dock to the left side of the screen (again, to use all the vertical space for my own stuff.) This left a small rectangle of desktop visible at the upper left and lower left, even with maximized windows. I drew a panel in the upper left of the desktop, and set it to run this script every five seconds:#!/usr/bin/env python3.6 import datetime import re import subprocess def block_eighths(eighths):     """Return the Unicode string for a block of so many eighths."""     assert 0 <= eighths <= 8     if eighths == 0:         return "\u2003"     else:         return chr(0x2590 - eighths) def gauge(percent):     """Return a two-char string drawing a 16-part gauge."""     slices = round(percent / (100 / 16))     b1 = block_eighths(min(slices, 8))     b2 = block_eighths(max(slices - 8, 0))     return b1 + b2 now = print(f"{now:%-I:%M\n%-m/%-d}") batt = subprocess.check_output(["pmset", "-g", "batt"]).decode('utf8').splitlines() m ="\d+%", batt[1]) if m:     level =     batt_percent = int(level[:-1]) else:     level = "??%" if "discharging" in batt[1]:     arrow = "\u25bc"        # BLACK DOWN-POINTING TRIANGLE elif "charging" in batt[1]:     arrow = "\u25b3"        # WHITE UP-POINTING TRIANGLE else:     arrow = "" print(level + arrow) print(gauge(batt_percent) + "\u2578")   # BOX DRAWINGS HEAVY LEFT vol = subprocess.check_output(["osascript", "-e", "get volume settings"]).decode('utf8') m ="^output volume:(\d+), .* muted:(\w+)", vol) if m:     level, muted = m.groups()     if muted == 'true':         level [...]

Clean-text bookmarklet


I love the text-based web. I love that people can speak their minds, express opinions, encourage each other, and create a lively world of words. This also means they are free to design their text in, shall we say, expressive ways. Those ways are not always ideal for actually reading the words.Today I really liked Tiberius Hefflin's Part of That World, about the need to recognize non-code contributions in open source projects. You should read it, it is good and true.But when I first got to the page, I saw this:To start with the positive, this text has an elegance to it. It gives a peaceful quiet impression. It pairs perfectly with the mermaid illustration on the page. But I find it hard to read. This typeface is too weak to be light-on-dark, and letterspacing is almost always a bad idea for body text. It isn't even white-on-black, it's 70% white on black, so the letters seem to be hiding in the dark.I don't mean to pick on this page. It's a well-designed page. There's clearly a mood being created here, and it's been established well. There are many pages online that veer much farther from the usual than this.My solution for pages like this is a bookmarklet to strip away idiosyncracies in text layout. It changes text to almost-black on white, it removes letterspacing and shadows, and changes full-justified text to left-justified. When I use the bookmarklet on Part of That World, it looks like this:You might prefer the original. That's fine, to each their own. You might feel like the personality has been bleached from this text. To some extent, that's true. But I saw the original, and can choose between them. This helped me to read the words, and not get snagged on the design of the page.This is the bookmarklet: Clean text.This is the JavaScript code in the bookmarklet, formatted and tweaked so you can read it:javascript:(function () {     var newSS = document.createElement('link'),         styles = (             '* { ' +                 'background: #fff; color: #111; ' +                 'letter-spacing: 0; text-shadow: none; hyphens: none; ' +             '}' +             ':link, :link * { color: #0000EE; } ' +             ':visited, :visited * { color: #551A8B; }'         ).replace(/;/g,' !important;');     newSS.rel = 'stylesheet';     newSS.href = 'data:text/css,' + escape(styles);     document.getElementsByTagName('head')[0].appendChild(newSS);     var els = document.getElementsByTagName('*');     for (var i = 0, el; el = els[i]; i++) {         if (getComputedStyle(el).textAlign === 'justify') {    = 'left';  &#[...]