Transition to New Site

This blog is now being retired in favor of the notes on my new site, which has an explanation of the transition. This blog will remain but will no longer be updated.

Steve Jobs, Rest in Peace

Today a great entrepreneur, one of the great American capitalists, Steve Jobs, passed away at the rather young age of 56. As others have said, we are all a bit poorer for it.

I didn't always agree with his style, I wasn't always a fan of his company's products, and I certainly didn't like the personality cult that surrounded him, but I have great respect for the man who believed in what he did, thought big, sought to empower people, and did it with style and aesthetics. He is an inspiration to us all.

Finding Files with PowerShell, Part 2

This version (see Finding Files with PowerShell) gives the full path of the items:

Finding Files with PowerShell

I miss the UNIX find command in Windows. I could get it with Cygwin or GNUWin32, but that requires littering my machine with extra stuff. So, PowerShell to the rescue:

Reddit's Unique Web Server

Reddit has quite a unique web server.

How Likely Are You to Be Dealt a Royal Flush?

In the game of poker, you have a royal flush when you have a 10, J, Q, K, and A all of the same suit. Assuming a uniform random distribution of cards in the deck, what is the probability of being dealt a royal flush?

A royal flush consists of a certain five cards of the same suit. Assume we are dealt one of these cards. Then the suit is determined, and we now need the remaining four cards of that suit. So, let's take the probability of being dealt one of the certain five cards (10, J, Q, K, A) of any suit. Since there are five such cards per suit, and there are four suits (the set of suits is \( \{ \diamondsuit, \heartsuit, \clubsuit, \spadesuit \} \)), then we can get any one of those cards to start building a royal flush. Therefore, the probability of getting a starting card is \[ P(\text{starting card}) = \frac{20}{52}. \] Once we have a starting card of a certain suit, our suit is restricted. That is, we may only get the remaining four cards from the same suit as our starting card. Then the probability of getting one of the four cards from the same suit is \[ P(\text{one of the four from the same suit}) = \frac{4}{51}. \] Likewise, after we get the second card, there are three left to get, so the probabilities are \( 3/50 \), then \( 2/49 \), and then \( 1/48 \). What, then, is the probability that we will get all of the necessary cards? It is the product of all these probabilities: \[ P(\text{all}) = \frac{20}{52} \cdot \frac{4}{51} \cdot \frac{3}{50} \cdot \frac{2}{49} \cdot \frac{1}{48} = \frac{1}{649\,740} = 1.539\ldots \times 10^{-6} = 0.000\,153\,9\ldots \%. \] Not very good odds.

Calculating the Number of Trailing Zeros in a Factorial

The factorial function, denoted \( n! \), where \( n \in \mathbb{Z}^* \), is defined as follows: \[ n! = \begin{cases} 1 & \text{if $n = 0$,} \\ n(n - 1)! & \text{otherwise}. \end{cases} \] In the case of \( n = 10 \), \( 10! = 3\,628\,800 \). Note that that value has two trailing zeros. When \( n = 20 \), \( 20! = 2\,432\,902\,008\,176\,640\,000 \) and has four trailing zeros. Can we devise an algorithm to determine how many trailing zeroes there are in \( n! \) without calculating \( n! \)?

Let us define a function \( z : \mathbb{N} \to \mathbb{N} \) that accepts an argument \( n \) that yields the number of trailing zeros in \( n! \). So, using our two examples above, we know that \( z(10) = 2 \) and \( z(20) = 4 \). Now, what is \( z(100) \)?

An integer \( a \) has \( k \) trailing zeros iff \( a = b \cdot 10^k \) for some integer \( b \). Now, the prime factors of \( 10 \) are \( 2 \) and \( 5 \). So, what if we consider the prime factors of each factor in a factorial product? For example, consider the first six terms (after \( 1 \)) of \( 100! \) and their prime factors. We omit \( 1 \) because of its idempotence: \[ \begin{align} 2 &= 2 \\ 3 &= 3 \\ 4 &= 2^2 \\ 5 &= 5 \\ 6 &= 2 \cdot 3 \\ 7 &= 7 \end{align} \] We see that \( 7! \) contains one \( 5 \) and (more than) one \( 2 \). Therefore, \( 7! \) contains a factor of \( 10 \), so we should expect there to be one trailing zero in \( 7! \). And we see that \( 7! = 5040 \), which does indeed have one trailing zero. This suggests that we could, for any integer \( n \) in \( n! \), cycle through the integers from \( 2 \) to \( n \), finding the prime factorization of each integer, and counting pairs of \( 2 \) and \( 5 \) that we find.

You might have noticed from our example of \( 7! \) that there are four \( 2 \)s and only one \( 5 \). Consider that there are \( n/2 \) even factors in \( n! \)—that is, every other integer factor of \( n! \), starting at \(2\), is even. So there are \( n/2 \) even factors. Each of these has at least one \( 2 \) in its prime factorization, and many of them have more than one. Now consider how many factors of \( 5 \) there are in \( n! \): there are, in fact, \( n / 5 \). So, we can see that, for every \( 5 \) we find in \( n! \), there is a \( 2 \) we can match with it. This suggests that we only need count the \( 5 \)s and not the pairs of \( 2 \) and \( 5 \).

This now suggests an algorithm, presented here in Java:

Using this algorithm, we find that \( z(100) = 24 \).

Adobe Reader's Big Problem

I use PDFs often. Nothing excels them in fidelity of design to documents for which design is important. The Adobe Reader, in its 5.5 days and earlier, was the standard PDF reader of choice. But with version 6, serious bloat occurred, and Adobe Reader is only now, in its latest versions, recovering from that problem. I have a new problem with Adobe Reader, though. Take a look:

This is a document that appears to be made of scanned images placed in text pages, produced by \( \mathrm{\TeX} \). Note how the scanned images appear as completely useless gray boxes. Here is how the document should look, and does look, in other PDF viewers:

I have searched and searched on this matter, but the results are few and old, and whenever I try to employ a solution I do find, it doesn't work.

Generate a Zero-Valued UUID

Say you want a UUID of 00000000-0000-0000-0000-000000000000, but you just want to make one fast, without having to find out the number and grouping of digits. Here's a way to do it:

The Beauty of Scheme

Bertrand Russell once noted that

Mathematics, rightly viewed, possesses not only truth, but supreme beauty—a beauty cold and austere, like that of sculpture, without appeal to any part of our weaker nature, without the gorgeous trappings of painting or music, yet sublimely pure, and capable of a stern perfection such as only the greatest art can show. The true spirit of delight, the exaltation, the sense of being more than Man, which is the touchstone of the highest excellence, is to be found in mathematics as surely as poetry.

I think much the same can be said of code, especially code written in certain languages. There is something so elegantly simple and beautiful about Scheme (and Racket), for example, that shares the same aesthetics of mathematics.

A Human-Usable URL Parser for Python

Yes, there is the urlparse module in Python 2.7 and the urllib.parse function in Python 3, but each of them just creates a 6-tuple of URL parts, and it's up to you to know that the query part, for example, is at position 4.

So here's a neat little wrapper around that functionality which gives named fields to each part:

So, to get the query part, you just do parsed_url.query instead of parsed_url[4].

How I Saved My Friends Lots of Money

Last night I got a phone call from one of my neighbors. She told me that she needed to use some courseware for a class at her university and said that she was unable to use it, and that she was told by someone that she needed to upgrade Windows to use it. They further said that, since a Windows license costs on the order of $200, she should just buy a new laptop. Now, this courseware is a web application—just a web site—yet she was told to get a new laptop. Naturally, my BS-meter went off, and I asked if I could come take a look.

I went over and looked, and saw that the real problem was not that the web site didn’t work—it worked just fine—but that the course documents were Microsoft Word .docx files. Not having Office or any other office apps, she had nothing with which to open them. So, I told her the nature of the problem and installed LibreOffice. Problem solved for $0.

The moral of this story? These points:

• “Computer literacy,” so called, is woefully scarce in our society. I don’t really fault my neighbor so much as the yahoo she spoke with who told her such outrageous nonsense.
• Professors should make their course documents available as PDFs, which is the proper interchange format for important documents. (What’s interesting about this is that Microsoft Word 2007 and 2010, which use the .docx format natively, make producing PDFs from Word easier than ever before.)

Making System.Data.SQLite Work in an NServiceBus Endpoint

Today a colleague and I worked on getting an an NServiceBus endpoint in a .NET 4.0 project that referenced System.Data.SQLite to work properly. It turns out that this is not as straightforward as it may seem.

The System.Data.SQLite binaries are targeted for the .NET 2.0 runtime. When referencing a .NET 2.0 mixed mode assembly in a .NET 4.0 project, additional configuration is required:


OK, that's fine. But when running an NServiceBus endpoint in the NServiceBus.Host.exe executable, it seems that this setting in App.config does not get applied in the context of NServiceBus.Host.exe. So, to get this to work, we had to turn the NServiceBus endpoint into a console application with a Main method that used reflection to invoke the Main method of NServiceBus.Host.exe. Doing this enables the above setting to be applied in the context of NServiceBus.Host.exe.

It took us the better part of a day to figure this out, so I'm memorializing it here for (my own) future reference.

Generate Test Files From the Command Line

Note to self: I had to generate some files quickly to test the behavior of the .NET FileSystemWatcher. This little bash script will create 6 zero-filled files of of size \( 2 \times 10^i \) KiB, where \( i \in \{0, 1, \dots, 5\} \).

Retrying Code Execution

I have a bit of code I'm working on at work that makes use of an HTTP endpoint to notify a legacy system of certain events. Because the network is not reliable, we have to handle the case when the network may be down. In this situation, it means to try again a few times, after which the processing request goes into an error queue, where it can be handled manually. So, I started out with something like this:

This calls the HTTP endpoint up to five times. If one of those times succeeds, it returns the result and stops trying. If it gets to the sixth time, it gives up.

Now, in my project, I have several of these types of calls to make. Much of the above code, namely the exception handling and flow control stuff, is common to all of them. So, using some functional .NET and an extension method, we can do this:

This will attempt to execute and return the value of func up to times times, and give up after that. With this extension method, we can now put our endpoint-calling code in a Func<int> and tell it to try to call the endpoint up to five times, like this: Now, is this the best way to do this? Maybe; maybe not. It's just something I've been toying around with in my attempts to eliminate duplication of the aforementioned ugly flow control code. I'll have to walk around in these shoes for a while and see whether I like them.