I have found another crash in ferret; this one just uses a regular
search. It’s similar to an issue reported by Matt S. a while ago,
but unlike his, mine does not go away if I turn off omit_norms. It does
go away if I turn on the garbage collector more often, but I’m not sure
that’s a stable workaround under the circumstances.
This one isn’t a segfault, but more complicated versions have been. This
bug manifests as a error in the garbage collector:
$ ruby rw_ferret_test.rb
…rw_ferret_test.rb:20: [BUG] rb_gc_mark(): unknown data type
0x28(0x402ba3a4) non object
ruby 1.8.2 (2005-04-11) [i386-linux]
Aborted
Here’s the code:
require ‘rubygems’
require ‘ferret’
include Ferret
def phrase(num)
“a_fairly_long_phrase%05d”%num
end
LOOPS=(ARGV[1]||40).to_i
_fields = Index::FieldInfos.new
_fields.add_field :phrase, :term_vector => :no, :store => :no, :index =>
:untokenized
$phrases=I.new(:field_infos=>_fields)
phrase(0).upto(phrase(LOOPS)){|phrase| $phrases << {:phrase=> phrase} }
@phrases=Ferret::Search::Searcher.new $phrases.reader
count=0
loop{
@phrases.search(
Search::TermQuery.new(:phrase,
“a_fairly_long_phrase%05d”%rand(LOOPS+1))
)
if ((count+=1)%1000).zero? #1309
print '.'
STDOUT.flush
GC.start #enable this to workaround
end
}
$phrases.close
On 07.12.2006, at 10:38, Caleb C. wrote:
bug manifests as a error in the garbage collector:
$ ruby rw_ferret_test.rb
…rw_ferret_test.rb:20: [BUG] rb_gc_mark(): unknown data type
0x28(0x402ba3a4) non object
ruby 1.8.2 (2005-04-11) [i386-linux]
Aborted
I tried your code and I get lots of dots but no error 
Maybe you should use a newer Ruby version. I’m using
ruby 1.8.5 (2006-08-25) [powerpc-darwin8.8.0]
Cheers,
Andy
On 07.12.2006, at 14:56, Patrick R. wrote:
[BUG] rb_gc_mark(): unknown data type 0x28(0x402ba3a4) non object
ruby 1.8.2 (2005-04-11) [i386-linux] Aborted
I tried your code and I get lots of dots but no error Maybe you
should use a newer Ruby version. I’m using ruby 1.8.5 (2006-08-25)
[powerpc-darwin8.8.0]
I get the same crash on ruby 1.8.5 (2006-08-25) [i386-cygwin],
perhaps it’s specific to certain architectures?
Not sure if this is of any use, but I tried it on two other machines.
Linux (Ubuntu):
…test.rb:21: [BUG] rb_gc_mark():
unknown data type 0x0(0xb7d9e208) non object
ruby 1.8.4 (2005-12-24) [i486-linux]
Linux (Debian):
ruby 1.8.5 (2006-08-25) [i486-linux]
All test (including the first one on Mac OS X) were run with Ferret
0.10.13
Which version of Ferret are you using? The combination Ruby 1.8.5/
Ferret 0.10.13 seems to work best.
–
Andy
Hi Ben!
On 07.12.2006, at 15:44, Benjamin K. wrote:
Which version of Ferret are you using? The combination Ruby 1.8.5/
Ferret 0.10.13 seems to work best.
1.8.4 and 0.10.13 is fine, too 
Yes, I used that combo successfully on my staging system for quite
some time. However, the test case that Caleb posted fails on that
same machine.
…test.rb:21: [BUG] rb_gc_mark():
unknown data type 0x0(0xb7d9e208) non object
ruby 1.8.4 (2005-12-24) [i486-linux]
Perhaps David should take a look at this. BTW, any news from him? Is
he still missing in the outbacks down under?
Cheers,
Andreas
Perhaps David should take a look at this. BTW, any news from him? Is
he still missing in the outbacks down under?
Hey …
Well, no news… just some very short notices that he’s still lost and
without an internet connection… If I hear anything i’ll send it to this
list.
A bunch of things are left on Daves agenda and he wanted to implement a
lot of great features this year, to finally reach a 1.0 status. But I
guess we will need to wait a couple of more weeks until a new ferret
release will be available.
Ben
I met the same problem before,i thought it caused by ruby
Multithreading.
So i make a new searcher before search every time,it works well,maybe
you should try this.
Which version of Ferret are you using? The combination Ruby 1.8.5/
Ferret 0.10.13 seems to work best.
1.8.4 and 0.10.13 is fine, too 
for other test results take a look at
http://camping.davebalmain.com/smoke_alarm/
Ben
On 12/7/06, Caleb C. [email protected] wrote:
…rw_ferret_test.rb:20: [BUG] rb_gc_mark(): unknown data type
include Ferret
$phrases=I.new(:field_infos=>_fields)
if ((count+=1)%1000).zero? #1309
print '.'
STDOUT.flush
GC.start #enable this to workaround
end
}
$phrases.close
This problem has been fixed.