Vm-dev June 2010

vm-dev@lists.squeakfoundation.org

42 participants
52 discussions

[commit] r2216 - add platforms dir for new apple codebase
by commits＠squeakvm.org 25 Jun '10

25 Jun '10

Author: root Date: 2010-06-25 13:32:47 -0700 (Fri, 25 Jun 2010) New Revision: 2216 Added: trunk/platforms/iOS/ Log: add platforms dir for new apple codebase

1 0

infrastructure for builds
by stephane ducasse 25 Jun '10

25 Jun '10

Andreas et al. What we can offer to the community is a hudson server (maintained). Right now we are experimenting with pharo and we want to see how we can push the maintenance of hudson to INRIA admin (because well it takes 3 or 4 versions of hudson to find the working one) since they have a service were we could be hosted. Now for the VMs, INRIA has a platform/access with a set of different OSes to help people porting software. I wanted to ask them if they could bridge this platform access to hudson so that we could have automatic build on all the platforms they offer. I have to discuss with them, my impression is that they are willing to support that. First we should get credible with the simple hudson service and build pharo versions and start to burn our/their servers with that. After I imagine that we could nicely asked for suhc a bridge. So if it makes sense to the VM-dev people we would be happy to push. Stef

1 0

Unix VM alias resolving bug
by Henrik Johansen 24 Jun '10

24 Jun '10

There's a bug in the alias resolution sqGetFilenameFromString in sqUnixMain.c In the line filenameLength = readLink(aCharBuffer, aCharBuffer, MAXPATHLEN) It replaces the old link path with the resolved link path, however links may be relative, eg. /bin/sh linking to bash rather than /bin/bash. Thus, file opening, and other operations using this function will fail if encountering a relative link. F.ex. print (on Ubuntu, where sh -> dash): |openedRelative| openedRelative := nil. [FileStream readOnlyFileNamed: '/bin/sh' do: [:fs | openedRelative := true]] on: FileDoesNotExistException do: [openedRelative := false]. openedRelative and you get false. Cheers, Henry

3 3

Re: [squeak-dev] COG crash while recompiling
by Eliot Miranda 24 Jun '10

24 Jun '10

Hi Nicolas, On Wed, Jun 23, 2010 at 12:30 PM, Nicolas Cellier < nicolas.cellier.aka.nice(a)gmail.com> wrote: > I got several crash dump while recompiling some classes. > For example try to load > http://source.squeak.org/trunk/SUnit-dtl.79.mcz in a cog image. > I've tracked this down to the primitive trace log, traceInterpreterPrim, which is causing GC failures. I'm intending to delete this code anyway but before I do so I want to understand the bug, which may take a little time. So as a quick fix you could edit the C source of src/vm/gcc3x-cointerp.c and src/vm/cointerp.c and delete the function traceInterpreterPrim and its two calls. Those of you who want fixes for the simple block crash [ | b | 5 ] value and/or the GeneratorTest crash can load VMMaker-oscog.14.mcz from http://www.squeaksource.com/VMMaker into the VMMaker image, delete traceInterpreterPrim and its two senders and regenerate the sources. HTH Eliot > Nicolas > >

1 0

[squeak-dev-noreply@lists.squeakfoundation.org: VM Maker: VMMaker-oscog.12.mcz]
by David T. Lewis 24 Jun '10

24 Jun '10

The following update message was blocked from the vm-dev list because of the large size of the diff listing. Point your Montecello browser at Sqs/VMMaker to see the details. Note - after refreshing your Montecello browser you may need to scroll down the listing to find the new VMMaker-oscog.12 and VMMaker-oscog.14 entries. Dave ----- Forwarded message from squeak-dev-noreply(a)lists.squeakfoundation.org ----- Date: Wed, 23 Jun 2010 21:27:41 0000 From: squeak-dev-noreply(a)lists.squeakfoundation.org To: vm-dev(a)lists.squeakfoundation.org, lewis(a)mail.msen.com Reply-To: squeak-dev(a)lists.squeakfountation.org Subject: VM Maker: VMMaker-oscog.12.mcz Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://www.squeaksource.com/VMMaker/VMMaker-oscog.12.mcz ==================== Summary ==================== Name: VMMaker-oscog.12 Author: eem Time: 22 June 2010, 9:51:54 am UUID: a34fa6af-c3a0-453e-8450-8525342057ac Ancestors: VMMaker-oscog.11 Fix co-routining and cannotReturn: crashes. Key is to arrange that when baseFrameReturn returns to a frame in the middle of a stack page the frames above it (which may be in use in coroutining) are preserved, i.e. by movign them to another page. Fix simple blocks with temps. JIT compiles simple blocks without a stack frame and so the stack must balance. The JTI must strip any leading pushNils that initialize temps otherwise these will be interpreted as return addresses causing a jump into la la land. Various typos and niceties from mainline Cog. =============== Diff against VMMaker-ar.165 =============== <snip> ----- End forwarded message -----

1 0

VM Maker: VMMaker-oscog.14.mcz
by squeak-dev-noreply＠lists.squeakfoundation.org 23 Jun '10

23 Jun '10

Eliot Miranda uploaded a new version of VMMaker to project VM Maker: http://www.squeaksource.com/VMMaker/VMMaker-oscog.14.mcz ==================== Summary ==================== Name: VMMaker-oscog.14 Author: eem Time: 23 June 2010, 12:11:20 pm UUID: d50a4520-06c6-490c-8d24-900eff2ef35e Ancestors: VMMaker-eem.13 CogVM: Fix saving frames above interior frame for co-routining in non-local return. Remove unnecessary check in B3DAccelleratorPlugin>>stackMatrix: =============== Diff against VMMaker-oscog.12 =============== Item was changed: VMStructType subclass: #CogBytecodeDescriptor + instanceVariableNames: 'generator spanFunction numBytes stackDelta needsFrame needsFrameInBlock isBranch isBackwardBranch isReturn isBlockCreation isMapped' - instanceVariableNames: 'generator spanFunction numBytes needsFrame needsFrameInBlock isBranch isBackwardBranch isReturn isBlockCreation isMapped' classVariableNames: '' poolDictionaries: '' category: 'VMMaker-JIT'! !CogBytecodeDescriptor commentStamp: '<historical>' prior: 0! + I am an entry in the Cogit's dispatch table for bytecodes. I hold the routine to call to generate code for the partcular bytecode I represent and the number of bytes the bytecode has. For eliminating temps in frameless blocks I maintan a stack delta for bytecodes that are valid in a frameless block.! - I am an entry in the Cogit's dispatch table for bytecodes. I hold the routine to call to generate codee for the partcular bytecode I represent and the number of bytes the bytecode has.! Item was changed: ----- Method: B3DAcceleratorPlugin>>stackMatrix: (in category 'primitive support') ----- stackMatrix: index "Load a 4x4 transformation matrix from the interpreter stack. Return a pointer to the matrix data if successful, nil otherwise." | oop | <inline: false> <returnTypeC:'void*'> oop := interpreterProxy stackObjectValue: index. oop = nil ifTrue:[^nil]. - oop = interpreterProxy nilObject ifTrue:[^nil]. ((interpreterProxy isWords: oop) and:[(interpreterProxy slotSizeOf: oop) = 16]) ifTrue:[^interpreterProxy firstIndexableField: oop]. ^nil! Item was changed: (excessive method size, no diff calculated) Item was changed: (excessive method size, no diff calculated) Item was changed: ----- Method: ObjectMemory class>>requiredMethodNames (in category 'translation') ----- requiredMethodNames "return the list of method names that should be retained for export or other support reasons" + ^ #(byteSwapped: checkedLongAt: firstAccessibleObject noteAsRoot:headerLoc: splObj: printReferencesTo:)! - ^ #(checkedLongAt: firstAccessibleObject noteAsRoot:headerLoc: splObj: printReferencesTo:)!

1 0

VM does not always exit
by Chris Muller 23 Jun '10

23 Jun '10

I think I've noticed that, with the 4.0.3-2202 VM, after receiving an error that a particular (Cog) image cannot be loaded, the process is still active in Linux even after pressing CR. I've had to kill -9 them..

1 0

Setting the Cog bit
by David T. Lewis 23 Jun '10

23 Jun '10

Does the Cog VM update to a new image format version number when snapshotting? If not, then I have the following proposal: An image that is snapshotted by a Cog VM is going to expect Cog support when next loaded. It would be good if the image snapshot declares this explicitly so the VM can exit in a controlled manner if unable to provide the required support. I propose that we add a Cog bit to the image format version number, as illustrated in the attached change set. This can be extended in the future such that an image that requires "foo" support from a VM will set the "foo bit". The original Squeak images had image format number 6502. With closure support, this became 6504. Requiring Cog implies a requirement for closure support, so an image that requires Cog support would be 6505. For 64-bit images, the corresponding image format numbers are 68000, 68002, and 68003. Note that bit 0 was previously unused, which means that Cog would become the recipient of the highly-coveted bit zero position ;) There are plenty of free bits available to be allocated for future capabilities, and the high order bit would be reserved for expansion if the need ever arises. Dave

4 4

Cog direction (was: Cog - dumb question)
by David T. Lewis 23 Jun '10

23 Jun '10

Excerpted from thread on pharo list: On Tue, Jun 22, 2010 at 10:22:18AM -0700, Eliot Miranda wrote: > > What I hope will > happen is that the Stack VM will replace the standard VM pretty soon, and > that, because the Stack VM and the Cog VM can read each other's images that > once the VM maintainers have integrated the Stack VM into the VM sources > then images will move forward to the Cog format. This sounds like a very reasonable and achievable goal to me. Dave

1 0

adding iOS to platforms
by John M McIntosh 23 Jun '10

23 Jun '10

Ok if someone has the proper SVN permission I'd like to add an iOS sub directory to the platforms directory on SVN squeakvm.org It will contain the iphone, ipad, and os-x cocoa version of Squeak as I move the source tree from squeakvm.org I can sent the helpful party the iOS.zip file for importing... -- =========================================================================== John M. McIntosh <johnmci(a)smalltalkconsulting.com> Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===========================================================================

1 0

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

Vm-dev June 2010