Squeak-dev March 2003

squeak-dev@lists.squeakfoundation.org

202 participants
818 discussions

[FUN] Who's Where
by Chris Burkert 03 Apr '03

03 Apr '03

See http://www.chrisburkert.de/index.php?node_id=50 Not present at the map? Send me Longitude and Latitude of your town. have fun! Regards Chris Burkert PS: I didn't want to distribute this on the swiki because every new Image will be ~100KB and AFAIK it isn't possible to delete the older images. -- ------------------------------------------------------------------------ Student of applied Computer Science at Chemnitz University of Technology http://www.chrisburkert.de/ chris(a)chrisburkert.de ------------------------------------------------------------------------ "I invented the term Object-Oriented, and I can tell you I did not have C++ in mind." - Alan Kay --> http://www.whysmalltalk.com/

14 20

TestBrowser freezes - how to debug
by Martin Drautzburg 02 Apr '03

02 Apr '03

I just noticed that the TestBrowser does not come up anymore. This must have something to do with stuff I filed in recently or with things I messed up personally. Sometimes I am able to interrupt with Alt. and sometimes even that does not work. In the latter case I get the emergency debugger with the following backtrace. http://minnow.cc.gatech.edu/squeak/3144 My question is: how would an experienced smalltalker deal with such a situation. I even have an older image around, but it shows the same problem.

2 8

[FIX] senderOfSplitName-asm ([sm][et][er] approved of by tpr)
by Tim Rowledge 02 Apr '03

02 Apr '03

I like this. Strongly approve for inclusion. tim -- Tim Rowledge, tim(a)sumeru.stanford.edu, http://sumeru.stanford.edu/tim Useful random insult:- They must have done a clean boot on him.

2 1

[BUG][TEST] Unwind blocks fail to evaluate when closing debugger
by Andreas Raab 02 Apr '03

02 Apr '03

Hi Guys, Here's a weird one - if you step within a debugger and then close the window, unwind blocks will not be evaluated correctly. The attached test illustrates both, the correct and the incorrect behavior. Any ideas?! Cheers, - Andreas

5 11

[FIX][Comanche] persistent connections
by Ian Piumarta 02 Apr '03

02 Apr '03

Comanche was not sending the correct 'Connection: {Keep-Alive,Close}' in the response header. The attached fix cures the problem allowing browsers that aggressively use persistent connections (e.g., Safari) to work without endless annoying "connection reset" errors. Ian

4 5

YASJITCI -- Yet Another Squek Just In Time Compiler Idea
by Andrew Berg 02 Apr '03

02 Apr '03

I'm writing this for two reasons: (1) To get some expert opinion on the feasibility of my idea, and (2) to distract this otherwise quite enjoyable list from its semiannual not-quite-flamewar-over-licensing. (That's a little joke, but it's probably only funny because it hurts so much.) Anyhow, this idea occurred to me about a month ago, and so far I have not figured out why it would not work, so I'm posting it up here for you all to shoot holes in so I can get back to my day job. Write the JIT compiler in Squeak itself. This is similar to what Bryce has been working on with Exupery, but I propose that simplicity is the order of the day. FORTH, where I spent a reasonable amount of my misspent youth, is pretty darn fast at runtime because all of the subroutine calls are lightweight, and the compiled "methods" are natively executable. I don't think we can make the method calls as lightweight as FORTH, nor do we really want to. But we can write compiled methods that are natively executable. The Exupery way is to write a real compiler, and there is my way (which I make no bones about: It will not perform as well) which is to have a table of bytecode implementations, which just get chained together between some somewhat generic prefix/postfix blocks, and then get the various jump addresses fixed up. For message sends, it seems obvious that we want to inline when the receiver's implementation is short enough, and we have a high enough confidince that it is the right one, to cache the receiver's implementation address when it is not, and to emit fallback-code for the rest. Proper register allocation is an important thing that I am almost completely throwing away. Years ago, when I wrote a FORTH compiler, it worked by emitting little chunks of code which were hand-written in assembly. My initial implementation assumed that at each chunk boundary there would be the top three stack elements in three specific registers, and that the pointer to the rest of the data stack lived in another register. What I quickly found was that it was often convenient to also be able to have one, two or four top values in registers, so I tried again with a slightly more complicated table where I kept track of how many stack elements were in registers and then provided multiple implementations of each basic word (think: bytecode) with various numbers of values in registers as preconditions, and with the most efficient number for the word as the postcondition. Along with a little table of nop instructions for filling holes in my hand-written table, it worked pretty good. My gut feeling is that this approach will give us XX% of what a full blown register allocation scheme will, for xx% the cost, where XX and xx are probably something around 85 and 15 respectively, just like everything else. What will it do for us performancewise? Well, on deeply pipelined superscalar cpu's, the worst thing you can do is to introduce a bubble in the pipe, and the easiest way to introduce a big bubble is to miss a branch prediction. Strictly speaking, a memory access that misses all the way out to the swapfile is probably bigger, but let's assume we can't do much about that case. Right now, we are probably suffering a branch mispredict on every bytecode, since the case statement implementing the bytecode interpreter is branching to all of the bytecodes from a single branch instruction. By naively chaining together sequences of machine code, without a doubt we will be generating more instructions than necessary, but with multiple instructions being dispatched per clock, the cost of these is quite low in comparison. What will it require for implementation? Well, as I see it, code in this implementation has three levels, and maybe more. Source code, compiled to bytecode and compiled to machine code are the obvious ones. In order to produce receiver class metrics for message send optimizing and inlining, it will probably be handy to add a fourth one, between bytecodes and machine code. Whether this layer should be bytecode or machine code, I don't really know. If message recipient's implementation addresses are to be placed directly into the machine code as jump addresses, then they need special handling from the VM to prevent movement during GC. I thought maybe if the jump addresses were stored in the constants table, then the VM would know to update them and everything would be fine. Yeah, there is an extra lookup for the jump indirect, but as long as branch prediction is usually correct it is probably OK. As I see it, the thing to do is to put a second layer of compiled to bytecode to gather receiver classes. Could we dedicate a seperate set of send bytecodes for this? Once the receivers are known, generate new bytecode which uses a new version of send something like 'sendMessage w to x ifItsClassMatches y returningTo z' and leaves a new return address which skips over the code for a regular, unknown send. Then this bytecode is what we use chunk-chaining to create executable machine code for. Ok, there's the idea. Anyone out there understand where I am trying to go with this? Anyone out there can tell me why it is a dumb idea? By the way, the last project I spent about 2 months worth of evenings working on was embedding Mozilla into Squeak. I gave up because: I realized that I really don't want to fiddle with Mozilla. It is a big, ugly, complicated beast, and it is written in C++. I use C++ at work because they pay me to. The last thing I want to do is to use it at home. :) -andrew -- andrew_c_berg(a)yahoo.com

2 2

Running Squeak apps off the net
by jennyw 01 Apr '03

01 Apr '03

One of the things that drew me to Squeak initially was that I was under the impression that one could run Squeak applications over the net ... kind of like going to a Web site but getting a full-blown app. instead of a Web page. Was this a misimpression on my part? Or is this something in the works? I got this impression because I read something comparing Squeak to REBOL saying that Squeak had it first ... Thanks! Jen

3 7

Re: [ANN] Exupery, yet another compiler project
by Daniel Vainsencher 01 Apr '03

01 Apr '03

Bryce Kampjes <bryce(a)kampjes.demon.co.uk> wrote: > Currently I only target x86 on Linux. It shouldn't be hard to extend > both to other operating systems and also to other CPUs. Perfect on both counts - that's what I use :-) > > For a first step - do you have the obligatory 3 + 4 working? > Compiling to assembly it can handle an iterative factorial. Compiling > inside the image is still very limited because of the assembler. That > is my current project. Good idea. > Once the assembler can handle compiling an iterative factorial the > next step will be message sends. They are required for the integer > byte codes anyway, overflows are handled by sending and so is > non-SmallInteger arithmetic. I'll tell you one thing that sounds to me like it might be both useful and relatively easy to improve on - running the simulator. Since it's actually Slang (no dynamic sends), it's already a subset of the full Smalltalk problem (project 1.5?), and since it's currently either compiled externally (clumsy) or interpreted as full Smalltalk (relatively slow), you might make it more easily useful. If the simulator was fast enough to comfortably run regular size images, I guess people would use it more to learn about/debug problem images, and study the VM. Oh, we need to fix it in the image, too, but Craig's doing that... > I don't want to start trying to speed up sends until after > I've got useful byte code speed ups. Just curious - why? to partition it more clearly? > It might be better to think of it as three projects. The first one is > as small as possible to be easily achievable, compiling individual > methods to speed up bytecode performance. The second one is larger and > should be easily achievable on top of the first, a useful general > purpose dynamic compiler for Squeak. The third is a lot of dreams > about what computers could be and involves redrawing the abstraction > boundaries in systems software. I'm currently focusing on the first > which is why I'm interested in easy uses for the compiler now. Sounds like you have very clearly planned. Release early, release often - I'm curious :-) Daniel

2 1

[ANN] SerialTTY-nk registered on SqueakMap
by Ned Konz 01 Apr '03

01 Apr '03

SerialTTY builds on Ian Piumarta's Telnet package to provide a serial terminal. It will load the Telnet package if necessary, and will add it to the World menu if you're using 3.4. For 3.2 and 3.4. -- Ned Konz http://bike-nomad.com GPG key ID: BEEA7EFE

2 2

PluggableMultiColumnListMorph How To
by German Arduino 01 Apr '03

01 Apr '03

Hi Folks: I'm working in a Morphic project and need to put some date in a grid-style morph, something that can show 4 or 5 columns with a row title and several data rows. (All numeric). I've made several searchings in the image and I think that PluggableMultiColumnListMorph can help me, but I can't to make work him as I need. Someone have experience with this morph and can bring me some advices? or know some examples or related sites?. Thanks in Advance and Best Regards. --- German Arduino http://www.dimensionO.com

5 6

← Newer
1
2
3
4
5
6
7
8
...
82
Older →

Jump to page: