[Openmcl-devel] 32-bit Windows lisp didn't runs on 64-bit Windows

[Gary Byers]

The register dump suggests that it's a problem in low-level exception-handling
code: some unknown exception occurs, a handler is called, the handler causes
an exception, and we eventually overflow the stack.  We can't tell from that
information what the original problem was (or what caused the recursive
failure), and it's hard to do that without poking around in GDB.  I can't
easily do that poking around, since the image that Matt committed last week
seemed to work fine for me on 64-bit Windows 7.

Matt fixed some bugs that're related to the changes necessary to get the
Win32 CCL running in Win64 after committing the new image.  The nature of
the bugs was that some function that referenced thread-local storage referenced
an address that was a few kb beyond the correct location.  That's clearly
wrong, but:

  - it seemed that there was generally "something there": the memory references
    referenced the wrong values, but didn't fault (or at least no one reported
    that it did.)  AFAIK, the addresses involved get randomized a bit by
    Windows, so whether or not the bad memory access faults or not may
    not be predictable.
  - the functions in general only run in code that handles exceptions; it's
    not clear that any of that code ordinarily runs at startup, and you're
    apparently crashing at startup.

I checked in an image that contains Matt's fixes about 6 hours ago.  I don't
have any real idea whether or not this'd fix the problem, but if you could
do an 'svn update' and try again that'd either fix the problem (if we're lucky)
or eliminate one possibility.


On Tue, 8 Feb 2011, ????? wrote:

> Dear all,
> I tried to run the trunk version of the Windows/x86 port on 64-bit Windows 7
> .
> But an exception occurred.
> 
> this is log and ticket.
> http://trac.clozure.com/ccl/ticket/816
> 
> Are some settings ?necessary ?
> 
> Rregards,
> lambda_sakura
> 
> 
> 
>