[Gcl-devel] Re: reporting GC bugs in GCL

[Matt Kaufmann]

Thanks, Camm.

>> Matt, is this windows?  This is the same bug I think we discussed
>> earlier, which I hope has been resolved in

Yes, I believe it is, and in fact I think it's a pretty old GCL (from last
November or earlier).  Sorry about that.

>> BTW, I am still working on two issues you've brought to my attention.... 

FYI, it occurred to me to try the full test (of which the various recent
profiles etc. have been a part), with certain needless compilation omitted.
Here are results.

Allegro CL (development environment) time:
3839.780u 19.680s 1:04:24.64 99.8%      0+0k 0+0io 85891pf+0w

GCL time:
14599.720u 39.610s 4:04:13.28 99.9%     0+0k 0+0io 54777pf+0w

Thanks --
-- Matt
   cc: address@hidden
   From: Camm Maguire <address@hidden>
   Date: 22 Aug 2003 15:29:11 -0400
   User-Agent: Gnus/5.09 (Gnus v5.9.0) Emacs/21.2
   Content-Type: text/plain; charset=us-ascii

   Greetings!

   Matt Kaufmann <address@hidden> writes:

   > Hi, Camm --
   > 
   > An ACL2 user just reported the following, which looks to me like a GCL gc 
bug.
   > 
   >    ACL2 !>>[GC for 100 FIXNUM pages..[GC for 100 FIXNUM pages..
   >    Process acl2 exited abnormally with code 128
   > 

   Matt, is this windows?  This is the same bug I think we discussed
   earlier, which I hope has been resolved in

   revision 1.15
   date: 2002/11/19 03:05:59;  author: mjthomas;  state: Exp;  lines: +34 -11
   make_fixnum() called GBC(t_fixnum) called runtime() called make_fixnum...
   ----------------------------

   Releases after this point should be fine.  If not, please let me
   know.  It would be most helpful if you could give me the commands to
   reproduce it on a Linux system.


   > What is the preferred way to proceed when such bugs occur?  For example, is
   > there a web page where a list appears of various known GCL bugs in various
   > versions?  Or should I just ask this user to send you a bug report with 
the GCL
   > version and platform?

   The most useful for me is the gcl-devel list, address@hidden  (the
   list seems to be down/delayed at the moment).  Second to this is the
   savannah website.  We do try to keep a log of bugs there which aren't
   fixed quickly, but we tend to be short on the manpower needed to
   maintain this database.  Of course, if someone would so volunteer, it
   would be a great service.

   BTW, I am still working on two issues you've brought to my attention
   -- profiling, and the odd contiguous page gbc behavior when sgc is
   enabled and bignums are exercised.  I've verified that the latter is
   behaving as designed -- your allocation of a few relblock pages puts
   the available contiguous just below the amount needed for the
   computation specified -- but of course I feel the design can be
   improved.  What is basically happening is that certain page types are
   allowed to sgc when enabled and certain others are not.  In the bignum
   case, the bignum struct itself is on a page which does not sgc, and
   the body of the bignum is on a contiguous page which does.  The bignum
   body contiguous pages are never swept when sgc is on as the only step
   which would sgc-mark these pages occurs when sgc is turned off.  I
   should have a solution soon, together with a new profiling mode, gprof
   support.  At that point I'd like to release 2.5.4.

   Take care,

   > 
   > -- Matt
   > 
   > 
   > 

   -- 
   Camm Maguire                                         address@hidden
   ==========================================================================
   "The earth is but one country, and mankind its citizens."  --  Baha'u'llah