Subject: Re: Prolem setting up Vortex/Cecil 3.0 on Solaris (problem in gc)
From: James McCarron (jmccarro@maplesoft.com)
Date: Mon Apr 17 2000 - 19:25:00 PDT
According to Craig Chambers:
>
> Hmmm. There is no more recent version of Vortex. How did you fix the GC
> library? I think that when you ran Vortex, you ran the prebuilt version that
> contained the old (and apparently not working for you) GC library, and so you
> got the same error. Where is the newer GC library? What kind of Sun are you
> running? (If it's something really new or really old, then it's possible that
> the stack layouts are different than what we expect.)
>
> If you're just interested in Cecil, then be aware that the Cecil interpreter
> executable might be more convenient for small programs than the Vortex
> optimizing compiler.
>
> -- Craig Chambers
Thanks for replying so quickly!
I used Version 4.14 of the Boehm collector. (I don't recall where I
got it from; probably http://reality.sgi.com/boehm/gc.html.) I simply
replaced the directory
runtime/gc/conservative-gc/src
in your distribution with the directory from the gc-4.14 distribution
and copied in your Makefile. (The `README' file in the
runtime/gc/conservative-gc/src directory containing the collector
indicates that Version 4.11 was used in your distribution.)
The machine I am using is an Ultra 10 (sun4u), with Solaris 7 installed.
(US-IIi processor). (It's one of these new PCI based systems; I just
got it.) So, yes, it is quite a new machine.
I think there must be a bug in the heuristics used by the collector
to detect the direction of stack growth that has been fixed (or made
compatible with this machine) in the version I tried. The message
seems to indicate that it was built with the correct stack direction
defined. In /usr/include/sys/stack.h on my machine appears:
#define STACK_GROWTH_DOWN /* stacks grow from high to low addresses */
which seems to agree with the message. But, the "GC_stackbottom = 0x0"
part of the message seems suspicious to me. (Wait, I'll poke around a bit...)
Yes, the function GC_get_stack_base() differs between the two versions
of the collector. Version 4.14 appears to contain a correction for
this problem exactly:
# ifdef STACK_GROWS_DOWN
if (result == 0) result = (ptr_t)(signed_word)(-sizeof(ptr_t));
# endif
appears just before the return in GC_get_stack_base() in 4.14, but
is absent from Version 4.11 of the collector. (Well, there are alot
of differences, but this looks to be a likely candidate.)
I am interested in both the design of the language Cecil and the compiler
technology you've developed for it. For the latter, I can likely get
by with all the technical reports and papers I found at your site,
and it appears that most of the Cecil source for the compiler is
included, so yes, I can likely do with just the Cecil interpreter
(certainly, it will do at least for awhile).
Unfortunately, I also get:
[pts/11][9196][20:50:38]excalibur% bin/solaris/cecil
STACK_GROWS_DOWN is defd, but stack appears to grow up
sp = 0xffbef04c, GC_stackbottom = 0x0
stack direction 3
Abort (core dumped)
so it does not appear that I can use the interpreter either.
(I get the same thing if I invoke it properly using `run-cecil'.)
I'm not sure what can be done at this point, but if you have any
suggestions, that would be great.
Thanks very much,
James
-- James McCarron E Pluribus Solaris! "memory leaks are quite acceptable in many applications" -- Bjarne Stroustrup, The Design and Evolution of C++, page 220 cat(``,op(zip(cat,[seq(_,_="Js nte al akr")],[seq(_,_="utAohrMpeHce ")])));
This archive was generated by hypermail 2b25 : Tue Oct 03 2000 - 15:21:30 PDT