Berkeley DB Reference Guide: Architecture independent FAQs
Berkeley DB Reference Guide:
Building Berkeley DB for UNIX systems
Architecture independent FAQs
When compiling with gcc, I get unreferenced symbols, e.g.,:
symbol __muldi3: referenced symbol not found
symbol __cmpdi2: referenced symbol not found
On systems where they're available (e.g., HP-UX, Solaris), Berkeley DB uses
64-bit integral types. As far as we can tell, some versions of gcc
don't support these types. The simplest workaround is to reconfigure
Berkeley DB using the --disable-bigfile configuration option, and then rebuild.
My C++ program traps during a failure in a DB call on my
gcc-based system.
We believe there are some severe bugs in the implementation of exceptions
for some gcc compilers. Exceptions require some interaction between
compiler, assembler, runtime libraries, and we're not sure exactly what
is at fault, but one failing combination is gcc 2.7.2.3 running on SuSE
Linux 6.0. The problem on this system can be seen with a rather simple
test case of an exception thrown from a shared library and caught in the
main program.
A variation of this problem seems to occur on AIX, although we believe it
does not necessarily involve shared libraries on that platform.
If you see a trap that occurs when an exception might be thrown by the DB
runtime, we suggest that you use static libraries instead of dynamic
(shared) libraries. See the documentation for configuration. If this
doesn't work, and you have a choice of compilers, try using a more recent
gcc or a non-gcc based compiler to build Berkeley DB.
Finally, you can disable the use of exceptions in the C++ runtime for
Berkeley DB by using the DB_CXX_NO_EXCEPTIONS flag with
db_env_create or db_create. When this flag is on, all
C++ methods fail by returning an error code rather than throwing an
exception.
I get unexpected results and database corruption when running
threaded programs.
I get error messages that mutex (e.g., pthread_mutex_XXX or
mutex_XXX) functions are undefined when linking applications with Berkeley DB.
On some architectures, the Berkeley DB library uses the ISO POSIX standard
pthreads and UNIX International (UI) threads interfaces for underlying
mutex support, e.g., Solaris and HP-UX. You can specify compilers,
compiler flags or link with the appropriate thread library when loading
your application, to resolve the undefined references:
cc ... -lpthread ...
cc ... -lthread ...
xlc_r ...
cc ... -mt ...
See the appropriate architecture-specific Reference Guide pages for more
information.
On systems where more than one type of mutex is available, it may be
necessary for applications to use the same threads package from which
Berkeley DB draws its mutexes, e.g., if Berkeley DB was built to use the POSIX
pthreads mutex calls for mutex support, the application may need to be
written to use the POSIX pthreads interfaces for its threading model.
While this is only conjecture at this time and we know of no systems that
actually have this requirement, it's not unlikely that some exist.
In a few cases, Berkeley DB can be configured to use specific underlying mutex
interfaces. You can use the --enable-posixmutexes and
--enable-uimutexes configuration options to specify the POSIX and Unix
International (UI) threads packages. This should not, however, be
necessary in most cases.
In some cases, it is vitally important to make sure that you load
the correct library. For example, on Solaris systems, there are POSIX
pthread interfaces in the C library, and so applications can link Berkeley DB
using only C library and not see any undefined symbols. However, the C
library POSIX pthread mutex support is insufficient for Berkeley DB and Berkeley DB
cannot detect that fact. Similar errors can arise when applications
(e.g., tclsh) use dlopen to dynamically load Berkeley DB as a library.
If you are seeing problems in this area after you've confirmed that you're
linking with the correct libraries, there are two other things you can
try. First, if your platform supports inter-library dependencies, we
recommend that you change the Berkeley DB Makefile to specify the appropriate
threads library when creating the Berkeley DB dynamic library, as an
inter-library dependency. Second, if your application is using dlopen to
dynamically load Berkeley DB, specify the appropriate thread library on the link
line when you load the application itself.
I get core dumps when running programs that fork children.
Berkeley DB handles should not be shared across process forks, each forked
child should acquire its own Berkeley DB handles.
I get reports of uninitialized memory reads and writes when
running software analysis tools (e.g., Rational Software Corp.'s Purify
tool).
For performance reasons, Berkeley DB does not write the unused portions of
database pages or fill in unused structure fields. To turn off these
errors when running software analysis tools, build with the
--enable-umrw configuration option.
Berkeley DB programs or the test suite fail unexpectedly.
The Berkeley DB architecture does not support placing the shared memory regions
on remote filesystems, e.g., the Network File System (NFS) or the Andrew
File System (AFS). For this reason, the shared memory regions (normally
located in the database home directory) must reside on a local filesystem.
See Shared Memory Regions for more
information.
With respect to running the test suite, always check to make sure that
TESTDIR is not on a remote mounted filesystem.
The db_dump185 utility is the utility that supports conversion
of Berkeley DB 1.85 and earlier databases to current database formats. If
the errors look something like:
cc -o db_dump185 db_dump185.o
ld:
Unresolved:
dbopen
it means that the Berkeley DB 1.85 code was not found in the standard
libraries. To build db_dump185, the Berkeley DB version 1.85 code
must have already been built and installed on the system. If the Berkeley DB
1.85 header file is not found in a standard place, or the library is
not part of the standard libraries used for loading, you will need to
edit your Makefile, and change the lines:
DB185INC=
DB185LIB=
So that the system Berkeley DB 1.85 header file and library are found, e.g.,