McCLIM 0.9.3 "All Souls Day"

Thanks to tireless Release Engineer Andreas Fuchs, McCLIM version 0.9.3 has been released! I was about to write a blog entry about how this was a release short on shiny features, and that it mostly had solid bugfixes and changes of that sort. But as I have already used up my quota of unsubstantiated claims for this week, I decided to check the list of changes, just to be sure, which resulted in me being reminded that this release actually has some pretty major additions. The most shiny new feature is probably Gtkairo - a new backend using GTK widgets and Cairo for drawing. It's not yet as functional as the default CLX backend, but it presents the interesting future prospect of McCLIM applications not being as ugly as they are now. Also, using high-level Cairo drawing primitives easily beats raw X drawing for convenience. Personally, the McCLIM change I find most interesting is that Andreas Fuchs removed a nasty memory leak in the incremental redisplay code, that would eventually cause applications using incremental redisplay to gobble up all available memory. This means that Climacs will now run for more than ten minutes before getting unusably slow due to constant large-scale garbage collections! (It's still slow, but at least it no longer degrades even further with use.)

Also, making the freeze period so short (McCLIM was frozen last saturday) seems to work much better than long freeze periods that eventually just fizzle out. I doubt freeze periods cause the McCLIM hackers to start testing obsessively anyway, so the primary motivation and goal for the freeze period should be to make sure that no-one will commit any experimental buggy code just before a release.

McCLIM 0.9.3 probably works best on a threaded SBCL, and doesn't work on CMUCL and CLISP. The reason for the CMUCL breakage is the lack of proper support for forward-referenced classes in CMUCL's CLOS implementation, but that has been fixed in very-recent CMUCLs, so you might be able to find a patch so you can run McCLIM (it's fixed in 19d patch1 I'm told). The reason for the CLISP breakage is somewhat similar - CLISP chokes when compiling certain files that contain function definitions in whose body can be found a form resembling (typep x 'foo), where foo is the name of a class that has not yet been defined. I have not yet been able to create a small test case for demonstrating this, but CLISP hackers are most welcome to look at the issue by compiling McCLIM. I had an idea about how to create a hack that defined stub classes in advance, but for some reason CLISP suddenly compiles McCLIM without problems on my system, so I'm unable to get a list of the affected classes. You might be just as lucky. No matter what, remember that your CLISP will need to use MIT-CLX, NEW-CLX is known not to work.