I nominated SaVi to the Sourceforge community choice awards in the 'Best project for academia' category, despite the chances of SaVi winning or being nominated by anyone else being roughly zero. Prove me wrong.

I'm ever the optimist.

Microsoft's great file copying

We'll continue to support XP, and XP is a great operating system, but keep in mind, it will be 12 years old next year.

-- Kevin Turner, Chief Operating Officer for Microsoft,
Vista SP2 and Windows 7 More Secure than Linux and Mac OS X Leopard, Microsoft claims, Softpedia.

Great operating system? Windows XP's file handling is a joke. For example, try copying files from one disk to another by dragging in Windows Explorer. If XP runs into a single problem, it throws up a dialog box and stops copying entirely; figuring out what got copied and what didn't is your problem. COPY under DOS (sorry, cmd) is a joke. The included XCOPY and its switches attempts to make up for COPY's shortcomings, but is still no cp -pir.

So you wind up using third-party utilities, such as the free Ycopy, to do something as basic to a "great operating system" as copying files.

Writing a robust file copy function strikes me as something that is a lot easier than implementing good security, and I've yet to see Microsoft do that in its many patches and service packs. (Was XP "great" before all the patches and service packs, or after?)

Update: Ycopy skips copying all Eudora *.mbx files, even though Eudora is not running and the files are not open. And it's entirely unsupported. So I'm still looking for a robust free Windows file copy utility; next up: evaluating TeraCopy (1.22 hung on me a lot) and RichCopy (which seems to be adequate).

Cygwin and X regressions

Previously, I said that Cygwin was so much easier than the Mac (at least, as far as Macintosh Tcl/Tk went).

I updated my Cygwin install, and inadvertently picked up their new X server code announced in November. I tried the new X installation out with SaVi and Geomview. There's no hardware acceleration yet, but Geomview finally builds with just configure and make; no command-line nudges needed.

Geomview runs for about thirty seconds before the X server dies and takes everything with it. I've watched the X server exhaust resources so that I can't even pull down a menu in a Windows program. Having a slightly easier build environment does not compensate for the loss of a robust and reliable working environment (and an X server that had hardware acceleration, to boot).

Previously, Cygwin was so much easier than a Mac.

Tcl/Tk on Mac OS X

...is complex, bizarre, and quite oddly broken.

I maintain SaVi, which uses Tcl/Tk, and is reasonably crossplatform and portable thanks to relying on the unix C libraries, turning on every useful warning flag, and keeping its decades-old code uptodate with compiler changes etc.

Porting SaVi to Mac OS X 10.4 Tiger meant updating SaVi's ten-year-old menubar code to use newer menu commands which know about the Mac menubar. (This also produced more Windows-like menus under Cygwin's Insight Tcl/Tk.) Workarounds for Mac-only Tk crashes - don't make your listbox too wide or Aqua Tk will hang in an infinite drawing loop! - were found.

Then 10.5 Leopard came out and I bought an Intel Macbook, expecting to run SaVi in style. The Tcl/Tk menuing code that worked in Tiger didn't work in Leopard's Xcode with its Tcl/TK 8.4.7 install. At all. (Other Leopard bugs with OpenGL linking and so on took a while to find workarounds for, but not having working menus is a bit of a showstopper.)

So, I immediately retrofitted my old code, with a command-line flag to select menu drawing methods. Popop menus from the window on a Mac aren't elegant, but they're functional. And I started to look into the menubar problem and Tcl/Tk on the Mac more closely.

Over a year later, I'm giving up. I haven't found a fix for the menubar problem with the default Tcl/Tk 8.4.7 install. I've attempted to update the installed version of Tcl/Tk, using ActiveState, Fink, and reputable Aqua installers. These all install something, but don't dare touch the libraries in /usr/lib, to avoid breaking things. So, SaVi always gets compiled with Tcl/Tk 8.4.7, because /usr/lib is where the compiler looks.

Now, is the menubar problem fixed in later versions of Mac Tcl/Tk? The answer to that question doesn't matter, because the chances of SaVi's Mac Leopard userbase building SaVi with anything but Tcl/Tk 8.4.7 is vanishingly small. Saying 'just double-click this Tcl/Tk installer after you install Xcode' is a non-starter, because those installers don't update any libraries in the default locations that I can link to. Those packages seem to be emphasising the Mac-specific Framework stuff, rather than the crossplatform unix libraries.

(Complicating things slightly is that on the Mac there's the 'Aqua Tk' and the 'X Tk'. SaVi is a unix program, but lost its X dependencies years ago to work with Cygwin's Insight Tcl. So SaVi takes full advantage of Aqua Tk and the Mac interface. Apart from the menubar, of course. I haven't needed to experiment with the X Tk.)

SaVi 1.4.2, released last week, is my admission of failure. It is the first version of SaVi to default to popup menus on the Mac; Tiger users will just have to turn this off.

Cygwin was so much easier and simpler than the Mac.

SaVi satellite constellation Visualization

I've released SaVi 1.4.2, fixing a couple of minor bugs in SaVi 1.4.1.

1.4.2 has a new feature that lets you view the source to the simulation scripts that create the satellite systems. I don't know if showing the internal workings will encourage people to go from playing with SaVi to doing more serious simulation work, but one can always hope.

And it's gratifying to know that, in these uncertain times, people are still buying ++ungood; T-shirts. It's a view that sums up the economy.