Xapian discuss - Feb 2010 - Xapian core dumping on Solaris

Hi everyone,

I'm having a problem with xapian (the matchspy branch, with Python
bindings) on Solaris 10 / SPARC. Users can run queries for a few hours
with no problems, then it returns "inflate failed (invalid code
lengths set)"  to Python and dumps core. gdb reports:

#0  0xfe6dc910 in inflate_table () from /usr/lib/libz.so.1
#1  0xfe6d9d6c in inflate () from /usr/lib/libz.so.1
#2  0xfe2ebe20 in FlintTable::read_tag (this=0x75ed70, C_=0x75ede0,
    tag=0xfdcf8568, keep_compressed=false)
    at backends/flint/flint_table.cc:1254
#3  0xfe2eff70 in FlintTable::get_exact_entry (this=0x75ed70, key=@0xfdcf8578,
    tag=@0xfdcf8568) at backends/flint/flint_table.cc:1190
#4  0xfe2ddc04 in FlintSpellingTable::open_termlist (this=0x75ed70,
    word=@0xfdcf8998) at backends/flint/flint_spelling.h:46
#5  0xfe20dcb0 in Xapian::Database::get_spelling_suggestion (this=0x75b5a8,
    word=@0xfdcf8998, max_edit_distance=2) at include/xapian/base.h:476
#6  0xfe39ba90 in Xapian::QueryParser::Internal::parse_query (this=0x75b590,
    qs=@0x798288, flags=128, default_prefix=@0xfdcf8a44)
    at queryparser/queryparser.lemony:948
#7  0xfe38fa0c in Xapian::QueryParser::parse_query (this=0x798218,
    query_string=@0x798288, flags=128, default_prefix=@0xfdcf8be0)
    at include/xapian/base.h:154
#8  0xfe5551f0 in _wrap_QueryParser_parse_query (self=0x0, args=0x712648)
    at
/usr/sfw/lib/gcc/sparc-sun-solaris2.10/3.4.3/../../../../include/c++/3.4.3/ext/new_allocator.h:69
#9  0x00117290 in PyCFunction_Call (func=0x60d9e0, arg=0x712648, kw=0x0)
    at Objects/methodobject.c:116

This doesn't seem to be related to any particular query, and I haven't
worked out a simple way to reproduce the error. There is no database
writer while all this happens.

Any ideas? I might try rebuilding with the latest version of zlib -
does that sound sensible? Could it even be a low-level I/O error?

cheers,
Tom

On 1 February 2010 09:41, Tom <banoffi at gmail.com>
wrote:
> I'm having a problem with xapian (the matchspy branch, with Python
> bindings) on Solaris 10 / SPARC. Users can run queries for a few hours
> with no problems, then it returns "inflate failed (invalid code
> lengths set)" ?to Python and dumps core. gdb reports:
To be clear - when you say "returns "inflate failed ..." to
Python",
do you mean that it prints that message out on the console?  Or does
it return a value to python, and then dump core shortly afterwards?
> This doesn't seem to be related to any particular query, and I
haven't
> worked out a simple way to reproduce the error. There is no database
> writer while all this happens.
Is this in a single threaded environment?  The most likely cause of
this kind of error is accidentally calling a xapian object from two
threads simultaneously.
> Any ideas? I might try rebuilding with the latest version of zlib -
> does that sound sensible?
Sounds worth a try - what zlib version are you using currently?
> Could it even be a low-level I/O error?
Could be, but it doesn't look particularly likely to me, from that
traceback.

-- 
Richard

On 1 February 2010 09:41, Tom <banoffi at gmail.com>
wrote:
>
> I'm having a problem with xapian (the matchspy branch, with Python
> bindings) on Solaris 10 / SPARC. Users can run queries for a few hours
> with no problems, then it returns "inflate failed (invalid code
> lengths set)" ?to Python and dumps core. gdb reports:
If anyone's curious, we finally fixed this by including -D_REENTRANT
and -D_PTHREADS (probably only one is required) in the c++ flags
(we're using python threads at the app level).

Tom