On Thu, Sep 22, 2011 at 11:01 AM, Igor Stasenko siguctua@gmail.com wrote:
On 22 September 2011 19:37, Eliot Miranda eliot.miranda@gmail.com wrote:
...in http://www.mirandabanda.org/files/Cog/VM/VM.r2493/. These VMs are
not much changed from the 2487 VMs, BUT the 2487 VMs introduced a bug while providing faster doesNotUnderstand: processing. These 2493 VMs contain code to help identify that bug. If you're already using a 2487 VM or newer /please/ upgrade to these 2493 VMs asap. I want to see C stack backtraces from these VMs whenever they crash. Typically the VMs produce a crash.dmp file somewhere in the file system (may be / on Mac OS, may be in the directory containing the VM or the current directory on other OSs, but it will be called crash.dmp, and on Mac OS info will be printed to the console or to a crash report window). The C backtrace part of things looks like
Just a little nitpick: can you make a crash.dmp to appear in more predictable places? :) I vote for same directory where image located.
Yes, good suggestion.
this on windows: Stack backtrace: [004193EA] ??? + 4297706 in CogCode [00442337] ??? + 4465463 in CogCode [00442685] ??? + 4466309 in CogCode [00583018] ??? + 5779480 in CogCode [0040124B] ??? + 4198987 in CogCode [00401298] ??? + 4199064 in CogCode [7C816FE7] RegisterWaitForInputIdle + 73 in kernel32.dll [7C913BA7] RtlDosApplyFileIsolationRedirection_Ustr + 1824 in ntdll.dll this on Mac C stack backtrace: 0 nsvm 0x00038689 reportStackState + 105 1 nsvm 0x0003893e sigsegv + 110 2 libsystem_c.dylib 0x9b2e259b _sigtramp + 43 3 ??? 0xffffffff 0x0 + 4294967295 4 nsvm 0x0008b2a5 sufficientSpaceAfterGC
- 69
5 nsvm 0x0008d11d primitiveNewWithArg +
221
6 ??? 0x068b9899 0x0 + 109811865 7 nsvm 0x00098d10
initStackPagesAndInterpret + 512
8 nsvm 0x0002c530 EventLoopEventHandler
- 16
Etc. So please feel free to email me the crash.dmp's and/or the C
backtraces of any crashes you have with 2493 VMs. Hopefully I'll track down this regression quickly.
For the curious, what's the bug? I don't know yet. The symptom is that
very rarely the inline cache linking machinery for MNU PICs relinks the call of an MNU PIC to the address 0x00000013, which causes a crash when the call is next executed, long after the bug bit. These VMs contain code that will raise an error when the relinking attempt is made, which should reveal the bug.
Here's the 2493 readme:
CogVM binaries as per VMMaker.oscog-eem.125/r2493. Add callsite
link/relocate
checks to catch the call 0x00000013 MNU callsite relinking bug. Reduce the size of the simStack to something proportional to
LargeContextSize.
In the Newspeak VM, don't cd to the image's directory on win32. Fix
off-by-one
error in Win32OSProcessPlugin>primitiveGetCurrentWorkingDirectory so it
doesn't
include an erroneous trailing null.
Fix the 1Gb allocation bug.
-- best, Eliot
-- Best regards, Igor Stasenko.