Squeak People diary for rowledge

15 May 2005

Sun, 15 May 2005 20:18:38 -0700

Cool. We just had a Black Bear swimming in our pond. Not every day you see that, even out here.

The performance issues with the 64b VM seems have been solved now; the use of inline static functions works ok but isn't optimised quite as well as 'real' code. Making macros available for platforms with such problems restored benchmark figures and other fixes done in the process of finding the problem have made small improvements over that.

11 May 2005

Wed, 11 May 2005 22:48:08 -0700

Another six months, another diary entry. Hope you all like the 'squeak people moves to Germany' logo!

My email address and website have had to move following a set of senility related problems on sumeru - it is after all a seven year old pentium - 66 running a version of linux so old that just logging in raises a cloud of dust. So, I have 'www.rowledge.org' all set up and ready to go and the email address should now be 'tim@rowledge.org'.

The VM is now almost clean for 64bit machines. There is work left to do on making it easy to use and a number of changesets need adding to the update stream. Some plugins still need cleaning up so volunteer to fix any you depend upon. Neither I nor Ian have time to hit them all. I'm a bit concerned that my performance results have dropped significantly with these changes but so far nothing stands out as a likely cause. So far noinfo on performance results for other platforms.

In my last entry I referred to the release of the host windows work; since then there has been zero, zilch, none, no, zip, interest in doing anything with it. Makes me wonder why I bother.

6 Nov 2004

Sat, 06 Nov 2004 20:40:57 -0700

Golly doesn't six months go by quickly?

I've been busy though: a job, trips to LA (twice), Portland (for Camp Smalltalk), Seattle (twice, once for Smalltalk Solutions), Victoria (three times) and Vancouver for OOPSLA. Far too much flying and I have to go to SFBA next week for yet another con.

We released Ariethfa Ffenestri a few weeks ago, providing a first version of portable host mulitiple windows. It currently works well on RISC OS and Mac with a Windows port beginning to get on the way. We (which is to say John McIntosh & I) have only made it work with Tweak so far since that is the target UI framework for the TK4 project which is paying for the work. Ned feels it should be a reasonably simple job to make it work for Morphic as well - I'm just about clueless on morphic so I'll take his word for it. For documentation about the VM framework, see the swiki page (http://minnow.cc.gatech.edu/squeak/3862).

22 May 2004

Sat, 22 May 2004 03:47:51 -0700

Damn 'int' ! It was a really bad idea to have variables in the Slang code default to int; anytime ObjectMemory stretches past the 2Gb range in your machine's memory map uses of int comparison are going to try to bite you in the delicate parts.

It took me three days of debugging why my new VM wouldn't load the package loader properly - was it my new file system? the new socket code? something basic in prim dispatching? Why did it work _with the same VM excutalbe_ just a couple of days before? I got to test that the sockets were fetching the same bytes, the files were writing and reading correctly and even that unzipping was working ok. Finally, a check on the actual oops being loaded in primloadsegment revealed that the magical 2Gb boundaryhad been breached! And the reason it worked a few days previously? Such a memory address is caused (or not) by a random number of previous application startup/closedowns claiming memory before Squeak is run.... grrrr.

Still, all I have to do now is fix one offending routine and at least that particular problem will go away. I've managed to get rid of several big irritants in the last couple of weeks. By reimplementing a RISC OS specific version of the basic file access prims I have been able to fake out the 'you cant open a file multiple times if it is open for write' problem. This is a peculiarity of RISC OS that has been pissing me off since the first release of Squeak but it was nicely disguised by a loadable filesystem called !raFS; you probably don't know what RISC OS loadable filesystems are but basically it is possible to trap file system calls and do things like making a big file look just like a DOS disk, or to treat a zip file as a directory, or... whatever. !raFS did this in a way that provided workarounds for several annoying aspects of older RISC OS releases. Sadly it doesn't work for my latest machine and I've been surviving without the ability to use a lot of useful stuff in Squeak for most of a year. Finally I got annoyed enough to get off my bum (or rather, back onto it) and sit at the keyboard to write a loadacode to keep a cache of previously opened files and their name, handle and a reference count. Now when a file is opened I check the list (just the once, I'm not Santa) and if it's not already open I open it and make a new list entry. The Squeak file structure is then given the handle as normal. Any later attempt to open the same file will get the same handle; closes simply decrement the reference count and actually close the file when appropriate. Think of it - Squeak now has a refernce counting garbage collector for file handles! Oh, since I don't need to have to silly variable for 'lastOp' anymore (not using those stupid posix fread things) I use that to store the last set position of the filestream so that there is no overlap between different users of the same handle. It turned out to only take a day to write and I even managed to get truncate into it! So far after a couple of weeks usage there seem to be no bugs. It's a load faster too. Now I get to see my recent chages log! Timestamps in my browser annotations! Monticello works!

I also needed to update the sockets code; it's been left alone apart from emergency fixes for several years. since I understand sockets about as well as Gumby understands hyperspatial trajectory analysis I was really dreading this. Then I had a brainwave - the old code was merely a hackedup copy of Ian P's unix socket code - could it possibly be... Yes! a few trivial hacks to compensate for some (hopefully ) small diferences and whammo, holy bits Batman, it works. So thanks again to Ian's neat code and the unsung good guys that wrote the RISC OS socket libraries to do a pretty good emulation of unix sockets.

So, good improvements in my life as a user of Squeak. I can finally get around to making use of Monticello on a regular basis.

2 May 2004

Sun, 02 May 2004 05:46:46 -0700

Time for more news, eh. We bought a house! Outside Qualicum Beach, up in the local hills, next to Little Qualicum River. Couple of acres, nice views, pond, trees. Not a terrifically exciting house but nice enough for now. Move in is last week of June. Only real problem at themoment is that broadband internet doesn't reach out that far yet so I'm going to be dailup for a while :-(

VMMaker is now pretty much done. Since we're in beta there will be nothing except critical bug fixes until 3.7 final is wrapped up. The latest version is VMMaker 3.7b4 and the only cheng from b2 is that primitive function addresses are cached in the MCache and accessed directly to call the prim. No particular use is made of this yet but it does allow us to cache a specialised version of a prim to suit the class format of the receiver. By avoiding format checks we can save a little time in the prim.

17 Mar 2004

Wed, 17 Mar 2004 01:59:42 -0700

Another month another diary entry.

We're settled in a slightly more comfortable place now with more space and a decent network connection. At last I have the RISC OS machine online again - no more pratting about with stoopid winXP to get my mail. What a relief. All our paperwork is now done, even the healthcare cards are here. It's two whole months now without having to feel any embarrassment over odious george.

SO, Squeak work is proceeding quite well for now. A new VMMaker is taking shape slowly with a lot of cleanups of a small sort and a major change in the way primitives are called. On my RISC OS machine this has a major performance improvement (better than doubling it) but on x86 machines and pMac it seems to be worth 5-8% or so. Some more work is needed on cleanly handling prims that take a long time and potentially disturb the timer interrupts. What I have works but it's ugly.

More ordinary fixups include the object-as-method stuff, display depth allowing the -ve numbers, image startup speed improvements, jpegwriter bugfix, primitiveCopy, primitiveYield and so on.

20 Feb 2004

Fri, 20 Feb 2004 01:30:21 -0700

Yowzah, a month since last entry. Good grief.

Busy, busy. We had to do all sorts of stuff over the last month to get assorted bureacratic stuff dealt with. Sign up for the health coverage, apply for our SIN cards, meet the bank manager and get accounts all sorted out, find a new cellphone provider, all that stuff. Amazingly nobody has been less than very helpful. Even the DMV equivalent people - and for californians that will sound impossible. I even got the BS over the car 'exporting' sorted out eventually, which meant I could get it locally registered and insured. Insurance is actually a bit less than it was in CA. The neat thing is that insurance and registration go together here so there is basically none of that unisured motorist crap that causes so much trouble.

So we're settling in and I'm trying to get back to work again. Things are a bit tricky because of the limited network access and space (I'm at the kitchen table) and no printer and awkward lighting. I have at last been able to review and incorporate various VM related bugfixes and build a 3.7 beta release of VMMaker. On my machine it provides a few percent performance improvement but nothing drastic. So far I've been told it works ok for MacOSX as well.

22 Jan 2004

Thu, 22 Jan 2004 09:40:22 -0700

Well we made it to Vancouver Island at last. Right now we're staying at the White House (really!) whilst we look for a more permanent residence.

The journey was quite fun. We started from Milpitas last week and drove up to Reddding,CA for the first stop. No problems apart from missing the junction for the 505 Sacramento bypass and suddenly finding ourselves almost in the middle of the city. We just gave Ahnold a salute and moved on. Redding is not a terribly exciting place but not bad. We found the La Quinta Inn to be a reasonable place to stay and they are happy with dogs.

Next morning, up for breakfast and push on to Ashland, famous for the Shakespeare festival and definitely one of the last remaining habitats of the lesser striped hippy. Dinner was found at a pretty decent facsimile of a British pub - Post Office phonebox in the corner, shepherds pie on the menu, the whole thing. Another La Quinta Inn provided a comfy little suite and free, decent, coffee. By now it was getting pretty cold for someone with twelve years experience of California weather; we had seen snow and evidence of the serious blizzard conditions of a few weeks previous. Of course, since I had prepared properly - snowchains, shovel, boots, drinking water etc - we had no problems with snow.

Onwards to Eugene next and the problem is finding the dang hotel; the directions don't quite match up to reality. I just love when you find a road that appears to have two names, neither of which matches what you're told it should be. Eventually we find another La Quinta next to a really impressive park. The river, I think the Rogue, is in full flow and I'd hate to fall in. The only hassle is that the room is up six flights of stairs and it takes a while to get the dog up and down them for walkies. In case you didn't know, the dog is blind which makes stairs something of an adventure. Oh, we had to walk a couple of miles into downtown to find something to eat; La Quinta Inns are not generally equipped with kitchnes etc and all the nearby eateries were closed when we needed food. Still, the Steelhead micro-brewery was open and the food was very satisfactory.

After being woken at 7am by some nincumpoop in reception setting off the fire alarms(!) we got moving towards Portland. It was kind of grey and foggy a lot of the time so we really didn't get to see much mountain. Since the I5 gets a bit slow through inner Portland we used the 205 bypass and stayed at yet another La Quinta, this one near the airport. I'm really glad I didn't have to fly into Portland in that sort of cloud. It stayed cold and wet all night, not fun when doing late night walkies for the dog. Since the hotels's breakfast offering was a bit feeble we went over the road and stuffed ourselves at the local Shari's. Nice people and very satisfying comfort food.

Last stop in the US was Seattle - or more precisely another La Quinta, in Kirkland. This was by far the most expensive hotel and the least pleasant. It needs updating Real Soon Now. Still, we had dinner with a friend and a decent nights sleep for the final push to the border. It only took about three hours to get to the Peace Park crossing, where we officially landed in Canada. The immigration folks took a brief look at our paperwork, asked us to sign a couple of places and that was it; no hassles, no interviews, no testing us on why we were there. Very pleasnat, polite and quick. Then we did the customs work for importing all our household goods- no hassles, again pleasant polite and friendly people.

Finally, we were told that if we wanted to ever drive the car back to the US for some reason (like visiting friends) we should probably make sure to officially export the car from the US. Why this should even be a problem I can't imagine. To make it even sillier, you have to deal with it in person at a _different_ border crossing. So, we had to drive several miles to this other place and get back to the border. We had been warned that the US customs people have a nasty habit of impounding vehicles for several days as part of the process, so I left the car on the Canadian side and walked in to the customs office. What a difference! Rude, swaggering officials, guns everywhere, attitude galore. Even the female agents seemed to be suffering from testosterone excess. Apparently I had 'violated federal law, boy' by having the effrontery to take a car to Canada. Yup, they wanted to impound the car though after I pointed out that we needed it to travel on and no, we only have the one car (apparently also a federal offense) they begrudgingly allowed that I could just leave the title doc with them and would have to return three plus days later to be inspected. I dread to think what would have happened if I had simply pointed out that the car was in Canada and no way was it going to be left with them. Probably would have been shot or something. So now I have to drive/ferry all the way back to the border post _again_ next week. To be honest, it made me so happy to have escaped from the new police state mentality that I could almost cry.

After a night at the Tsawwassen Inn we got the ferry across to VanIsle this morning and after a short drive found the White House. It's on the lakefront, has pretty good views and even has a float plane dock - with docked plane right now. This (as in VanIsle) is the frst place we've ever actively chosen to live rather than simply landed up as part of work. It's really quite an odd feeeling.

18 Jan 2004

Sun, 18 Jan 2004 08:30:10 -0700

Yay! It's Radio Ranter back on the air for a few minutes as we stop off at the La Quinta Inn in Eugene, Oregon. Believe it or not they have broadband and wireless access that works on the third floor! So here we are listening to The Archers (see www.bbc.co.uk.radio4/archers - it's an addiction) and making a diary entry.

The last few weeks have been chaos of the worst sort; packing, rushing around organising, more packing, attemptingto get some programming done and more packing. Currently every computer I own except for an iBook is packed in a truck that is taking the long way round (boise, coeur d'alene, several stops in washington, finally Nanaimo) and so I won't have access to my main machines until at least the 28th of Jan. This may be a problem for making any new release of VMMaker for 3.7 ready for the proposed Feb 7 release. Currently I'm not aware of any major changes from the current beta that are needed; let me know if there are any truly crucial ones and we'll see what can be done.

Further to my entry of dec18th, I did get to do a little work on the morhpic performance problem that improved things a bit; it's still terrible but somewhat less so. In the process I reworked an ancient prim to replace the ProcessScheduler>yield method and slightly reduce the thrashing of delays. I'm extremely puzzled by this performance issue; the Acorn windowing system is amazingly fast for native apps and I really can't see where all the time is going. Sadly I don't have much in the way of OS level profiling to help here. It's particularly galling that a 600MHz P3 machine I have (which benchmarks about twice as fast as my 600MHz XScale Acorn, mostly because of the huuuge disparity in cache sizes I suspect) has no serious UI performance problems - even weirder when you consider that the KDE UI on it is appallingly slow. I almost feel as if some important vm plugin is failing on the Acorn but I've never been able to actually spot one. Suggestions welcome.

So, goodness knows when I'll have net access again. This will have to do for now. Meanwhile, spare a few monents to think of a fellow squeaker that is now not only jobless but homeless as well......

18 Dec 2003

Thu, 18 Dec 2003 23:51:14 -0700

Finally, I got around to releasing my ancient extensions to class reference browsing. And the save-image-somewhere-else patch. Amazing - at this rate I'll catch up in only 119.567 years!

I'm convinced there is something serioulsy wrong somewhere in Morphic-as-it-works-on-RISCOS. For the first time in a while I checked out an MVC project - and what a difference in performance! Menus that popup before my finger goes to sleep! Wow! Of course, the chances of me finding time to investigate are minimal right now, but never mind.