Raymond Toy wrote:
> >>>>> "William" == William Brew <[EMAIL PROTECTED]> writes:
>
> William> Well, it looks like I spoke too soon. Because the
> William> compile time in the 2004_04 snapshot was so long, I had
> William> tried saving part of the application with purification
> William> turned on and it worked. From this I extrapolated that
> William> purification was working in the 2004_04 snapshot. Over
> William> the weekend, I tried letting the application build go to
> William> completion; purification dumps
>
> William> core just like in previous versions of cmucl (18e, 2003_12).
>
> William> So the current state is:
> William> - purify works fine in "small situations" in all
> William> versions of cmucl that I have tried
> William> - purify dumps core when I try the whole application
> William> - non-purify works fine for the whole application in 18e and the
> 2003_12 snapshot
> William> - non-purify fails with a div by 0 in 2004_04
>
> This is not good. Would you be able to try your app with an older
> version of cmucl? I'm just wondering if this ever worked.
>
> Do you use FFI? Are you sure you haven't trashed the heap?
>
> Previously, you said the size of the application seems to matter. How
> big is the application?
>
> If it truly dumps core, you might want to run it with gdb. At least
> you might be able to figure out where it's dying and that might have
> some hints on where to look.
>
> Ray
I have succeed in developing two test programs that are completely independent
of my application that demonstrate the problems described above in the
2004_04 snapshot.
- purify fails on large images
- non purify save fails on large images
These could be related but I suspect that there are really two bugs here.
To demonstrate the purify problem, I just increased the size numbers in
the original test. To demonstrated the non-purify problem, I wrote a
program that writes a bunch of dummy functions to files. You then
compile and load these files, save (without purify) and then the saved core
will not restart as described previously.
This seems like it is getting more hard core cmucl developer'ish. I am new to this and
am unsure of the protocols. Should I just continue the discussion here and send the
above test programs to this list or is there a more appropriate forum.
Thanks.
-bill-
