Adium IM Stats
Adium keeps its chat logs in a pretty regular XML format. This makes it easy to collect some stats without mucking around too much with XML. Below are some code snippets that will give you a sorted list of who has been IMing you the most:
cd ~/Library/Application\ Support/Adium\ 2.0/Users/Default/Logs/
# for each folder (username) in this directory, run the following:
# (obviously change screenname to whatever your s/n is for each directory. note that screenname comes up twice)
grep -R "<message sender=" ./AIM.billybob123/* | grep -v "<message sender=\"billybob123\"" >> all_logs
cut -f1 -d " " all_logs | cut -f3 -d "/" | sort | uniq -c | sort -r -n -k1 > yes
# yes will contain your sorted list of messages received
macports compilation errors
Recently, my macports have been going crazy: I wasn’t able to install anything because I kept running into ‘"/lib/cpp" fails sanity check’ errors. I figured this was a config error because /lib doesn’t exist on OSX. Turns out, I had installed MinGW64 header files into /opt/local/include which were Win32-only. When I built my ports, they would use those headers and invariably fail. The quick solution was to clean out that include directory and reinstall all Macports.
Installing from a CD in Red Hat
I recently had to install a copy of Red Hat 5.4 for a client at work. Now, I needed to compile something, so I went to install GCC. However, since I used the VMWare easy install option, I was never prompted for my Red Hat Network account or password. Since I had the full DVD, I knew I could just install GCC from the repo on the disc. But how? Adding a source to yum gave me a weird error: ‘URL must be http, ftp, file or https not ""’. Given that my url started with file://, I couldn’t figure out what that meant. It turns out the spaces in the filename were the culprit: neither quoting, escaping, or replacing the spaces with %20 worked. Unfortunately, the Red Hat automounter mounted the disc using its volume name, which had spaces!
The conclusion I had: if you want to install RHEL packages from a DVD post-install, you need to unmount the disc (look in /media), remount it using a common location (/media/cd), and add the following to a .repo (my DVD’s repo was in Server/):
baseurl=file:///media/cd/Server
Implementing Lexical Closures on a Stack Machine
So, I’ve been studying how to implement things on LLVM, and I wanted to write down a key insight I had over the past few days. How the hell would I make continuations? A continuation is a closure of a function bound with the entire global state of the program. This lets you ‘warp back’ to a program state at any time, which is surprisingly useful.
Now, the question arises: how to you provide such an environment as a data structure? If I was implementing a language with dynamic scope, the answer is simple: copy and restore the stack. However, most modern languages use lexical scoping. This means that variables must be managed individually instead of “walking up the stack.”
My first idea was to hash out something like this:
- Set up a global symbol structure
- Whenever these are modified in a block, make a copy of the structure and modify that element
- Pass the new modified environment structure to any called functions
- When the current block returns, flip the environment back
Obviously, this involved a lot of copying, but I couldn’t see any other way of implementing this while making sure state propagated down. After all, if function a calls b, and function a locally modifies variable x, it has to propagate that change down somehow.
Or does it? Did you spot the error that was plaguing my implementation? I half-forgot that we are using lexical scoping — a statement like “a calls b, b must respect the local state of a” is very false. Luckily, this dramatically simplifies things for us! All we have to do now is keep one copy of the global environment around, if anything is locally changed this can be explicitly passed to the new function via an argument: internally, the function will ignore the global symbol and use the argument instead.
This is all very obvious looking back but it was very relieving to figure it out.
llvm2cpp — forget it
Instead of trying to resurrect llvm2cpp, I’ve just decided to write LLVM assembly directly to a text file. You lose the ability to do JIT and some automated stuff, but this is easier for now.
Accessing Array Pointers in LLVM
When using LLVM, it’s not immediately clear how to convert arrays into pointers. This may not be the best way to go about it, but this seems to work: use getelementptr on a pointer to the array, then dereference it twice. Let’s say you have the following array:
@cool = internal constant [2 x i64] [i64 1, i64 2]
and you want to get into the juicy bits of the array. The following will create an i64* for you based on the 0th element:
%intptr = getelementptr [2 x i64]* @cool, i32 0, i32 0
which you can then use with load and store.
Progressbar Marquee Fix
For work I needed to add a small screen with a progressbar using the Windows API. I wanted to use the marquee property to specify that the operation was running, but would take an indefinite amount of time. Now, I coded the progressbar and got it to display, but it wouldn’t marquee! If I set a position it would display fine, implying that most of the code was working properly. It turns out that the marquee property needs to be enabled by activating a newer version of a library. This MSDN page shows several options to enable this, but it was a pretty annoying problem to solve.
Found an LLVM Tip
I found some better LLVM tips online: this tutorial should be a bit more updated than the one on the site. Unfortunately there seems to be no specification of the virtual machine itself; I am currently determing how the LLVM stack works? Does it go up or down, what are the stack/frame pointers, and how is a frame saved?
The Fate of llvm2cpp
Since the LLVM docs are so out-of-whack, the best method to figure out how LLVM works seems to be through the live demo. This lets you type in a (complete) C program and get LLVM bytecode and even C++ API output. The backend of the API output is a program called llvm2cpp. However, it seems this tool has been deprecated for… nothing. The docs say llc may duplicate this functionality, but it currently does not. Unfortunately, the web demo doesn’t do a lot of things llvm2cpp does, mainly function-by-function API generation. This would have made my experimentation a lot easier; I guess I’ll try to piece together things with the web interface or get it to compile on its own.
Installing Whisper
After writing my own web publishing software, I’ve decided to use someone else’s for a while. I’ve settled on using William Morgan’s Whisper because I like doing things slightly different. There have been a few things I’ve been tweaking, but I’m leaving the design for a later date. Right now it’s a very abrupt change from my main site, but deal with it.