There is special code in dbi_bootinit() in DB.xs for Perl5.8 only that calls fp_dup()
to change the logfp variable to something other than PerlIO_stderr(). Although the
comment seems suspicious, the code seems to work fine for a normal single threaded
perl script. However, I have a C++ server that creates a pool of perl interpreters
using perl_clone(). In that case, any call into any child interpreter (all descended
from a single parent that parses the original perl script with perl_parse()) causes a
core dump. The crash happens by just adding "use DBI;" - no other calls are required.
Eliminating the call to fp_dup() and leaving the log as stderr resolves the core dump.
My server has been purified and executes numerous other perl modules correctly, so
there must be some problem with either the arguments or the call itself
#ifdef DBI_USE_THREADS
#if (PERL_VERSION >= 8) /* fp_dup prototype changed in 5.8.0 */
/* XXX the fp_dup type param is a fudge, but fp_dup doesn't use it anyway */
DBIS->logfp = parent_dbis ? fp_dup(parent_dbis->logfp,'>',0) : PerlIO_stderr();
#endif
#endif
Thanks,
Todd Richmond
