Building Ice-3.3.0 on Red Hat 9

[pjtait]

I have openssl-0.9.8g installed, but get the following error during make in the cpp directory. Clues welcome:

make[2]: Entering directory `/home/philip/Ice-3.3.0/cpp/src/IceSSL'
/usr/local/gcc/4.3.1/bin/c++ -c -I.. -I../../include -DICE_SSL_API_EXPORTS -ftemplate-depth-128 -Wall -D_REENTRANT -I/usr/include/nptl -I/usr/kerberos/include -I//usr/local/ssl/include -fPIC -g Instance.cpp
Instance.cpp: In function ‘long unsigned int IceSSL_opensslThreadIdCallback()’:
Instance.cpp:76: error: invalid static_cast from type ‘__opaque_pthread*’ to type ‘long unsigned int’
make[2]: *** [Instance.o] Error 1
make[2]: Leaving directory `/home/philip/Ice-3.3.0/cpp/src/IceSSL'
make[1]: *** [all] Error 1
make[1]: Leaving directory `/home/philip/Ice-3.3.0/cpp/src'
make: *** [all] Error 1

[matthew]

Sorry, this isn't a supported platform. For a full list of supported platforms see http://www.zeroc.com/platforms.html.

For this particular problem, assuming you can't upgrade to supported version of Linux, and you need the SSL plug-in, you can probably use the reinterpret_cast from the previous code block.

unsigned long
IceSSL_opensslThreadIdCallback()
{
#if defined(_WIN32)
    return static_cast<unsigned long>(GetCurrentThreadId());
#elif defined(__FreeBSD__) || defined(__APPLE__) || defined(__osf1__)
    //
    // On some platforms, pthread_t is a pointer to a per-thread structure.
    //
    return reinterpret_cast<unsigned long>(pthread_self());
#elif (defined(__linux) || defined(__sun) || defined(__hpux)) || defined(_AIX)
    //
    // On Linux, Solaris, HP-UX and AIX, pthread_t is an integer.
    //
    return static_cast<unsigned long>(pthread_self());
#else
#   error "Unknown platform"
#endif
}

Note that you might well run into further problems down the road, however!

[pjtait]

I don't need the SSL plugin, but I can't see how to unselect it. Is it somewhere in config/Make.rules.Linux ?

[dwayne]

If you don't want to compile SSL, edit cpp/src/Makefile and remove IceSSL from the list of subdirectories to compile.

[pjtait]

Thanks, that was a big help.

Next hurdle:

make[3]: Entering directory `/home/philip/Ice-3.3.0/cpp/test/IceBox/configuration'
make[3]: *** No rule to make target `../../../include/IceBox/IceBox.h', needed by `Service.o'. Stop.

Looks like editing items out of cpp/src/Makefile disrupted dependency resolution. Solved by changing directory to src/IceBox and running make there.

[mes]

Hi,

IceBox.h is generated as part of the build process, but you have to build the cpp/src/IceBox subdirectory.

As you can see, if you are building only certain portions of the Ice source distribution, you will encounter build failures elsewhere in the tests and demos that depend on the components you've skipped.

Take care,
Mark

[pjtait]

I reported that Freeze/BackgroundSaveEvictorI.cpp gave a compilation error. Some further Googling gave me the answer - the '--enable-cxx' option must be provided when building BerkeleyDB.4.6. Hopefully, this will help someone else, too.

[bernard]

Hi Philip,

When building Ice from source, I recommend reading the INSTALL.<OS> for your system, for example INSTALL.LINUX.

Another useful file is the README in the third-party source distribution available from our download page. It describes in particular the requirements for Berkeley DB.

Although INSTALL.LINUX points to this third-party source distribution, I agree that some build failures can be confusing. Maybe a FAQ entry would help?

Cheers,
Bernard

[pjtait]

You're right, I shouldn't have missed that.

For the next Ice release, it might be worth having the tarball set to unpack into a subdirectory, so that important files don't get overlooked in cluttered home directories... :-)

It's very encouraging to get such helpful responses.

[pjtait]

*** running tests 2/61 in /home/philip/Ice-3.3.0/cpp/test/IceUtil/thread
*** configuration: Default
*** test started: 09/04/08 08:11:16
starting client... ok
running mutex test... Cond.cpp:314: IceUtil::ThreadSyscallException:
syscall exception: Invalid argument failed
test mutex failed
('test in /home/philip/Ice-3.3.0/cpp/test/IceUtil/thread failed with exit status', 256)

[bernard]

If you look at the corresponding code (line 314 in IceUtil/Cond.cpp), you'll see the issue:

#if !defined(__hpux) && !defined(__APPLE__)
    rc = pthread_condattr_setclock(&attr, CLOCK_MONOTONIC); 
    if(rc != 0)
    {
        throw ThreadSyscallException(__FILE__, __LINE__, rc);
    }
#endif

It looks like condition variables on Red Hat 9 don't support a monotonic clock, at least not with the standard POSIX API. You should ifdef this out on your system as well.

Cheers,
Bernard

[pjtait]

Skipping test 8/61, I found the failure message for 9/61 indicated the problem was the lack of the epoll() system call. Undefining ICE_USE_EPOLL in src/Ice/Selector.h fixed the problem with 9/61, and 8/61 also. Onward...

[pjtait]

1 thru 60 all pass, 61 gives:

*** running tests 61/61 in /home/philip/Ice-3.3.0/cpp/test/Glacier2/staticFiltering
*** configuration: Default
*** test started: 09/05/08 16:10:50
testing category filter..Network.cpp:346: Ice:NSException:
DNS error: Servname not supported for ai_socktype
host:
. DNS error: Servname not supported for ai_socktype
host:
/home/philip/Ice-3.3.0/cpp/test/Glacier2/staticFiltering/client: Network.cpp:1387: Ice:NSException:
DNS error: Servname not supported for ai_socktype
host:
('test in /home/philip/Ice-3.3.0/cpp/test/Glacier2/staticFiltering failed with exit status', 256)

[benoit]

Hi,

Most likely this failure occurs when using an empty -h option in endpoints. You can try to reproduce it with the hello demo, the hello client should throw a similar exception. It should work if you add "-h localhost" to the endpoints of the Hello.Proxy in the demo/Ice/hello/config.client file. To workaround this problem, you can try replacing:

            rs = getaddrinfo(0, "1", &hints, &info);

with

            rs = getaddrinfo(0, 0, &hints, &info);

in the src/Ice/Network.cpp file, line 335 and 1365.

Cheers,
Benoit.

[pjtait]

Thanks for the guidance, Benoit.

I now recall having problems with getaddrinfo(3) on RH9 a couple of years ago - it doesn't behave quite the same as current implementations.

Your suggestion of setting the 'service' parameter to 0 didn't fix the problem, but setting it to "tcpmux" did.

Now I have all 61 tests passing for 'cpp'. :-)

Next up: 'py'!