And so today, just for giggles, I tried to get Sifflet to work. Along the way, I encountered a number of... "glitches", if you will. First of all, I tried to get it to work on Windows. I fired up a new Windows VM and installed Haskell Platform 2010.1.0.0. It seems that (finally) this includes the cabal-install tool, which is nice. It seems it includes cabal-install 0.8.0, so as soon as I tried "cabal update", it tells me a new version is available. It claims I just need to "cabal install cabal-install" (which is amusingly riddle-like). Unfortunately, although doing this *does work*, the new cabal.exe is installed lower down the search path than the existing one, so you still get the old version (unless you manually fiddle with the search path). In fact, it seems that the HP install folder is higher up the search path than the "global" binary target, which is higher than the "local" binary target. It seems to me like this ought to be the other way around. The next problem is that "cabal install sifflet" gets mighty confused and outright fails. Basically it can't figure out how to resolve all the dependencies. It seems that Sifflet demands GTK 0.11.0 (i.e., gtk-0.11.0, pango-0.11.0, glib-0.11.0, etc.) However, Cabal looks at gtk-0.11.0, sees it depends on (say) "glib-0.11.*", and wants to use glib-0.11.1 (the latest one). But then Sifflet wants glib-0.11.0, not glib-0.11.1, and it seems Cabal just can't figure out what the heck to do. Which is slightly surprising, really. The solution (of sorts) is to painstakingly resolve the dependencies by hand, by asking Cabal to install the correct packages one at a time in the correct order. (I still love the way Gtk2hs *actually compiles* on Windows now. That's pretty sweet!) That reminds me. What the heck is actually *in* file 126? I don't know why, but compiling file 126 (Graphics.UI.Gtk.Gdk.Cursor) takes up 75% of the entire Gtk2hs build time! What's that all about? Anyway, having finally built Gtk2hs version 0.11.0 successfully, I continued trying to get Sifflet working... only to discover it wants the "curl" package. And when I ask Cabal to build it, it just retorts that it has a configure script. *sigh* So that's the end of that. I have absolutely no idea why a tool like Sifflet would need access to the Curl library. Presumably this is just another one of those obscure dependency-chasing artifacts that happen from time to time? (After dealing with Linux, I'm used to this kind of weirdness.) OK, so it's only possible to run Sifflet under Linux. Let's give that a try... So I fire up a new OpenSUSE VM. I quickly discovered that gtk seems to want Alex and Happy, but cabal-install is defaulting to doing a "local" rather than "global" install, and thereafter it can't "find" Alex or Happy, even though they're installed. (Yeah, great, thanks for that...) So I rewind the VM, build the latest version of cabal-install, and edit the configuration to do global installs instead of local. The amusing part is, if you "sudo cabal install" so it has permission to put the installed files into place, it then uses root's configuration file instead. *sigh* Well anyway, I managed to work around that. But... Cabal *still* fails to find Alex or Happy, even though they're now in the search path. Oh, wait. They're in *my* search path. They're not in root's search path. (As per good security practise, root's search path is rather short.) I can pass some CLI switches to tell it where these are, but then gtk2hs-buildtools makes a whole bunch of stuff which Cabal also can't find. Eventually, the easiest thing I could come up with was to do "cabal unpack" to get a source tree, configure and build as me, and sudo for the install. Except that then it tries to reconfigure...? Wuh? So instead of "sudo cabal install", I tried "sudo runhaskell Setup install", which works just fine (although obviously it's rather wordy!) So, I manually hold Cabal's hand through the process of building all the 0.11.0 packages one at a time, in a way that it can find all the stuff at configure-time. Gosh this is a faff! Well anyway, it worked. Trips over when I reach Curl, but that's because I need to ask YaST to install the curl-devel package. And, finally, I can build and install Sifflet itself. After 4 hours or so of typing commands, it was nice to do some stuff with my mouse. ;-) I spent about 20 minutes trying to figure out how to build high-order functions before I discovered that you can't. Still, the stated design goal of Sifflet is "to teach beginners about recursion". Clearly it's design achieves this goal quite well. Equally clearly, Haskell has damaged my mind. I actually cannot think of *any* interesting low-order functions! 5 years of programming in Pascal with low-order *monomorphic* functions, and today I can't think of a single interesting use for such a thing. ;-) Heh, well anyway... I don't know how many of the things I've mentioned are known or likely to be fixed. Just thought I'd share.