Vm-dev March 2005

vm-dev@lists.squeakfoundation.org

16 participants
29 discussions

Start a nNew thread

TEST
by Alan Grimes 03 Apr '05

03 Apr '05

TEST.

2 2

changes to platforms/Cross
by Ian Piumarta 31 Mar '05

31 Mar '05

Folks, I'm just about to release the 'mainstream' 64-bit VM support. Part of this requires changes to the plugin-related files in platforms/Cross. I have tried to ensure that the changes remain backward compatible with pre-64-bit source code, but a few things might need tweaking (minimally) in platform-specific files to make old generated sources compile with the new headers. (I can compile a 3.7 Unix VM using my 64-bit support code, with all plugins built correctly, but I cannot guarantee anything about the other platforms which may or may not require intervention from their maintainers to ensure they build old-world generated code correctly after the new-world 64-bit changes are committed.) Unless someone complains before 15:00 PST, I will commit the 64-bit changes to the trunk. I consider the danger of committing these changes to be minimal. The danger of not committing these changes today (including leaving them to linger in the branch they currently occupy) is that they start rotting once more -- and it's not clear when I'll next have another week to waste on fixing them again. Cheers, Ian

3 2

changes to platforms/Cross [addendum]
by Ian Piumarta 31 Mar '05

31 Mar '05

Sorry -- I forgot to include a list of the affected files: Files 3.7-7/platforms/Cross/plugins/FilePlugin/FilePlugin.h and merge64/platforms/Cross/plugins/FilePlugin/FilePlugin.h differ Files 3.7-7/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c and merge64/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c differ Files 3.7-7/platforms/Cross/vm/sq.h and merge64/platforms/Cross/vm/sq.h differ Files 3.7-7/platforms/Cross/vm/sqNamedPrims.c and merge64/platforms/Cross/vm/sqNamedPrims.c differ Files 3.7-7/platforms/Cross/vm/sqVirtualMachine.c and merge64/platforms/Cross/vm/sqVirtualMachine.c differ Files 3.7-7/platforms/Cross/vm/sqVirtualMachine.h and merge64/platforms/Cross/vm/sqVirtualMachine.h differ (SocketPlugin.h will likely follow, as I 64-bit clean the Unix support code this afternoon.) Cheers, Ian

1 0

This week's VM build problem...
by Alan Grimes 29 Mar '05

29 Mar '05

Looking to update my workhorse VM with the current one, I sync'd my SVN sources and did my usual routine. (with the SVN-provided configure script), Here's the result. bash-2.05b$ make gawk -f /home/atg/source/trunk/platforms/unix/config/gnuify /home/atg/source/trunk/src/vm/interp.c > gnu-interp.c.out mv gnu-interp.c.out gnu-interp.c gcc -g -O2 -fomit-frame-pointer -DLSB_FIRST=1 -DHAVE_CONFIG_H -DSQUEAK_BUILTIN_PLUGIN -I/home/atg/source/trunk/bld -I/home/atg/source/trunk/platforms/unix/vm -I/home/atg/source/trunk/platforms/Cross/vm -I/home/ atg/source/trunk/platforms/Cross/vm -I/home/atg/source/trunk/platforms/unix/vm -I/home/atg/source/trunk/src/vm -I/home/atg/source/trunk/platforms/Cross/plugins/FilePlugin -I/home/atg/source/trunk/platforms/unix/p lugins/B3DAcceleratorPlugin -I/home/atg/source/trunk/bld -I/home/atg/source/trunk/platforms/unix/vm -I/home/atg/source/trunk/platforms/Cross/vm -c -o gnu-interp.o gnu-interp.c In file included from /home/atg/source/trunk/platforms/Cross/vm/sq.h:198, from /home/atg/source/trunk/platforms/unix/vm/sqGnu.h:45, from gnu-interp.c:8: /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:52:28: sqMemoryAccess.h: No such file or directory In file included from /home/atg/source/trunk/platforms/Cross/vm/sq.h:198, from /home/atg/source/trunk/platforms/unix/vm/sqGnu.h:45, from gnu-interp.c:8: /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:54: error: parse error before "minHeapSize" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:55: error: parse error before "sqGrowMemoryBy" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:55: error: parse error before "oldLimit" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:55: warning: data definition has no type or storage class /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:56: error: parse error before "sqShrinkMemoryBy" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:56: error: parse error before "oldLimit" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:56: warning: data definition has no type or storage class /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:57: error: parse error before "sqMemoryExtraBytesLeft" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:57: error: parse error before "includingSwap" /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:57: warning: data definition has no type or storage class /home/atg/source/trunk/platforms/unix/vm/sqPlatformSpecific.h:67: error: parse error before "sqInt" make[1]: *** [gnu-interp.o] Error 1 make: *** [vm/vm.a] Error 2 bash-2.05b$ -- Someone forgot to check in sqMemoryAccess.h?

2 1

3.7-7 "final" VM for Unix
by Ian Piumarta 25 Mar '05

25 Mar '05

Unix Hackers, I am pleased (and relieved) to announce the "final" 3.7-7 VM for Unix. This VM is very close to 3.7b-5 (which has been working well for many people for some time) with the addition of several critical fixes. Archives are in the usual place (http://squeak.hpl.hp.com/unix) for sources and a few pre-compiled binaries: Darwin 7.8 PowerPC Mac OS X PowerPC FreeBSD 4.10 i386 NetBSD 1.6.1 i386 OpenBSD 3.4 i386 Solaris 2.9 i386 Solaris 2.9 Sparc GNU/Linux PowerPC GNU/Linux i686 (.rpms and .debs will follow shortly. Due to limited physical access to machines I have only been able to test this release on Darwin-PPC, MacOSX-PPC, and GNU/Linux-i686.) If you prefer to fetch the sources from our Subversion repository: svn export http://squeak.hpl.hp.com/svn/squeak/tags/unix-3.7-7 (The repository source now includes everything needed to build the Unix VM. No more guessing which magical combination of image-side stuff is needed to generate a working Interpreter and plugins. FWIW, the generated sources used in this release are those found in the released "final" 3.7-5898-full image [following my usual rule] plus [exceptionally] Ned's "2GB oop" patches, which are included in the 'unix/misc' directory of both the source archive and repository.) Enjoy! Ian PS: A small reminder. Please send VM-related problems/bugs/fixes to the VM-Dev list (see discuss.squeakfoundation.org). VM-related posts to the general-purpose squeak-dev list are not guaranteed to be seen, let alone acted upon, by the relevant people. If you're sending me Unix-specific VM fixes, please send *entire* modified files (not 'diffs', of any flavour) since I read and manually apply all patches, one line at a time, using ediff in Emacs. Fixes should usually be sent to me directly (ian [point] piumarta [chez] squeakland [point] org) and not copied to any mailing list, *please*. (Partly traffic, but mostly to keep my remaining 'uncompromised' email addresses out of publicly-archived lists.) Thanks! PPS: Here are the release notes (relative to the 3.6 stable release) for anyone interested... VM: Netscape plugin understands "imageName" and "failureURL" attributes in <embed ...> tags. Less noise from dlopen() and friends when probing for plugins. Problems preventing more than one external plugin from loading in Mac OS X fixed. Obscure bug fixed that prevented socket options being set on some platforms. NPSqueak improvements and bug fixes from Bert Freudenberg and Jens Lincke. New environment variable SQUEAK_USERDIR used to find plugin image; defaults to 'My Squeak'. (NPS may not work for users of shells other than bash. Contact Bert if you have problems.) Interpreter changes from Ned Konz that avoid image segment loading errors caused by oops >= 2GB. VM drivers: Drag-and-drop support added to the X11 display driver. vm-sound-NAS driver explicitly depends on libaudio. New option '-glxdebug <n>' sets OpenGL debug message level for X11 display. Plugins: Additional checks for external data in SocketPlugin. New OSProcessPlugin (including XDpyPlg and AioPlg). Build: Optimisation flags default to '-O' if compiler not recognised. sqGnu.h asm() statements fixed for gcc-3.1 compatibility. Use of alloca() made more portable. SVN repository now includes contents of 'src' directory used to build Unix binaries. No more endless searching for the right combination of obscure VMM version and image updates. Build instructions updated for SVN and recent organisational changes. 'gnuify' inserts a #error into gnu-interp.c if it detects a problem (causing cpp to barf cleanly with a comprehensible error message, rather than leaving the compiler to spew hundreds of obscure error messages). Documentation: URIs updated to reflect recent organisational changes.

3 2

VMMaker 3.8b3 released
by Tim Rowledge 25 Mar '05

25 Mar '05

This is a temporary version of VMMaker mostly to provide a workaround for that irritating >2Gb issue. It is also an experiment in releasing as an MC file. It should build a useful VM for people keeping up to date. It is possible that there will be SVN tree related issues as the 64bit enhancements work their way in but everyone is working to prevent problems and begs your indulgence whilst the patient's chest cavity is open. tim -- Tim Rowledge, tim(a)sumeru.stanford.edu, http://sumeru.stanford.edu/tim Useful Latin Phrases:- Sentio aliquos togatos contra me conspirare = I think some people in togas are plotting against me.

1 0

3.8a-1 (Unix) and 32-bit compatibility for 64-bit support code (all platforms)
by Ian Piumarta 23 Mar '05

23 Mar '05

[apologies for multiple reception. if it's possible to subscribe to this list with several posting email addresses for just a single reception address, someone please point me at the relevant stuff] VM Hackers, The trunk of the unix tree is now at (an unstable, monotonically advancing) 3.8a-1. I'm telling you this because there are significant widespread changes since 3.7-7. In particular, the core Unix VM support is the result of merging 3.7-7 with my previous 64-bit support code released a while ago. This 64-bit compatible code is now the official support code for Unix versions 3.8 and onwards. While waiting for Tim to pull the changes Dan and I sent him into the Interpreter and CCodeGenerator, I've pre-emptively checked in Cross/vm/sqMemoryAccess.h, which is a slightly modified copy of the one I released with the original 64-bit code. It allows the support code to use the new type names, but still compile with a 'src' tree generated from a 32-bit Interpreter/CCodeGenerator. There may be a few unexpected warnings about mixing integers and pointers for the moment, but these are harmless. Needless to say, if your support code is still of the old 32-bit flavour, it won't ever include this header file. Note that this does not mean you can compile a 64-bit VM for 32/64-bit hardware, nor does it mean you can run a 32-bit VM on 64-bit hardware. The only thing it kludges into working order is compiling 32-bit 'src' to run on 32-bit hardware, mixing 'int's in the generated stuff with 'sq{Int,Long,Whatever}'s in the support code. For all four combinations of 32/64 VM + 32/64 h/w to work, we have to wait for our 64-bit image changes to percolate into the released image-side stuff. Backwardly-compatible cheers, Ian

2 1

SVN 3.7 platforms?
by Tim Rowledge 23 Mar '05

23 Mar '05

Do we have a 3.7 platform tree in SVN? As in, the code that one would point people to to make a nice shiny 3.7b6 vm for any platform? And assuming we have such a thing, is it in a form that would allow one to update some of the files? It is entirely possible for example that one or more of my platform files would be out of date. tim -- Tim Rowledge, tim(a)sumeru.stanford.edu, http://sumeru.stanford.edu/tim There are never any bugs you haven't found yet.

2 2

%USERNAME% on Windows 95 and 98
by Yoshiki Ohshima 22 Mar '05

22 Mar '05

Hello, Korakurider-san (it is a disguised name; his true identity is kept secret here) wrote a proposed fix for a problem. The problem is that the environment variable %USERNAME% is not usually defined on Windows 95 and 98, so the creation of the untrusted folder wasn't working. His change is to replace the line: lstrcpy(untrustedUserDirectory, TEXT("C:\\My Squeak\\%USERNAME%")); in ioInitSecurity with lines like: /* 2005/3/21 Tooru Nosse: Previous code assume environment variable USERNAME is defined, though it was not supported on Win95/98. That leave %USERNAME% unresolved. */ lstrcpy(untrustedUserDirectory, TEXT("C:\\My Squeak\\")); dirLen = lstrlen(untrustedUserDirectory); dwSize = MAX_PATH-dirLen; GetUserName(untrustedUserDirectory+dirLen, &dwSize); I guess that even GetUserName() fails, or even the user name is null, it doesn't create a bogus folder at least. -- Yoshiki

1 0

VMMaker 3.7b6 released
by Tim Rowledge 22 Mar '05

22 Mar '05

See SM. What it says on the can. tim -- Tim Rowledge, tim(a)sumeru.stanford.edu, http://sumeru.stanford.edu/tim BASIC is to computer programming as QWERTY is to typing. - Seymour Papert

2 3

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

Vm-dev March 2005