IcedTea7 HEAD build error on RHEL 5 x86_64

Dr Andrew John Hughes ahughes at redhat.com
Sun Sep 25 15:58:31 PDT 2011 

On 14:20 Fri 23 Sep     , Deepak Bhole wrote:
> * Dr Andrew John Hughes <ahughes at redhat.com> [2011-09-23 11:57]:
> > On 22:21 Thu 22 Sep     , Deepak Bhole wrote:
> > > * Dr Andrew John Hughes <ahughes at redhat.com> [2011-09-22 22:09]:
> > > > On 11:49 Thu 22 Sep     , Deepak Bhole wrote:
> > > > > * Pavel Tisnovsky <ptisnovs at redhat.com> [2011-09-22 09:53]:
> > > > > > Deepak Bhole wrote:
> > > > > > > * Pavel Tisnovsky <ptisnovs at redhat.com> [2011-09-22 08:37]:
> > > > > > >> Greetings,
> > > > > > >>
> > > > > > >> there's a new IcedTea7 build error thrown when IcedTea7 HEAD is build on RHEL 5 x86_64.
> > > > > > >>
> > > > > > >> Config log & shell script used to build IcedTea7 is stored in an attachment.
> > > > > > >>
> > > > > > > 
> > > > > > > Is libattr-devel installed on the system?
> > > > > > 
> > > > > > Nope (and ./configure did not complaint ;-). I've installed it and trying to build IcedTea7 again.
> > > > > > 
> > > > > > Maybe it's the right moment to update the wiki page http://icedtea.classpath.org/wiki/BuildRequirements
> > > > > > 
> > > > > > Do you think it's a good idea to split this page to IcedTea6 part and IcedTea7(8) part?
> > > > > > 
> > > > > 
> > > > > Hi Pavel,
> > > > > 
> > > > > Are the requirements radically different? From what I've seen, they are
> > > > > very similar if not the same for v6 and 7. If we can maintain the same
> > > > > page for both, I am all the more for it since it reduces changes of
> > > > > pages getting out of sync.
> > > > > 
> > > > 
> > > > You're wrong.  NIO introduces a number of new dependencies including
> > > > GIO and calls to extended attribute functions.
> > > > 
> > > 
> > > Yeah I realized that after the conversation on IRC this morning. I was
> > > going by the # of deps I had to add to the 1.7 spec file for Fedora
> > > (none) vs 1.6. But clearly that was not adequate.
> > > 
> > 
> > You should check out the recent IcedTea changes.  There are a number of latent dependencies
> > that, in upstream OpenJDK, don't get activated until a particular piece of runtime code is
> > activated.  With IcedTea, I've provided the option to build against these dependencies at
> > runtime instead (as part of the general system library update for jpeg, png, etc.).
> > 
> > Some probably also apply to 6, although they'll remain runtime there (do your 6 packages
> > depend on Gtk+, Gnome-VFS and GConf?)
> > 
> 
> They don't. I didn't know until yesterday that OpenJDK has certain
> run-time only dependencies. I guess so far we've just lucked out that
> nobody noticed (at least with Fedora) it because they happened to have
> the necessary deps already installed :/
> 

It's probably more that they are installed as standard anyway on
Fedora, if you use GNOME.  They are designed upstream so that they
build without any of the development headers needing to be in place at
compile-time, so you only get a failure if you hit the right piece of
code and cause the dlopen to be run.  Even then, in most cases, it
just appears like the feature doesn't work, rather than crashing out,
if it fails.

For 6:

* Gtk+ is needed for the Gtk look and feel which is automatically turned on
if GNOME is detected as running.  It dlopens gtk+ and gthread libraries at
runtime.
* GConf is needed if system proxies are turned on (this is changed to GIO
and GSettings in 7 as part of the bug I recently fixed, and I intend to
backport that to 6 too and upstream it - it will still fall back to GConf).
* GNOME-VFS is used to open a URL in java.awt.Desktop.  Again, I've changed
this to use GIO in IcedTea7.

7 also needs GIO and a number of extended attribute functions for NIO, which
again only fail at runtime.  This motivates the GIO change for proxies and
GNOME-VFS, as it actually reduces dependencies in 7.

The epoll system calls 

Here's a quick scan of dlopen calls in OpenJDK6:

../../../../jdk6/jdk/src/solaris/native/sun/nio/ch/FileChannelImpl.c:    if (dlopen("/usr/lib/libsendfile.so.1", RTLD_GLOBAL | RTLD_LAZY) != NULL) {
../../../../jdk6/jdk/src/solaris/native/sun/security/jgss/wrapper/NativeFunc.c:    gssLib = dlopen(libName, RTLD_NOW);
../../../../jdk6/jdk/src/solaris/native/sun/security/smartcardio/pcsc_md.c:    hModule = dlopen(libName, RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/security/pkcs11/j2secmod_md.c:    void *hModule = dlopen(libName, RTLD_NOLOAD);
../../../../jdk6/jdk/src/solaris/native/sun/security/pkcs11/j2secmod_md.c:    hModule = dlopen(libName, RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/security/pkcs11/wrapper/p11_md.c:    hModule = dlopen(libraryNameStr, RTLD_NOW);
../../../../jdk6/jdk/src/solaris/native/sun/security/pkcs11/wrapper/p11_md.c:    hModule = dlopen(libraryNameStr, RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/net/spi/DefaultProxySelector.c:  if (dlopen("libgconf-2.so", RTLD_GLOBAL | RTLD_LAZY) != NULL ||
../../../../jdk6/jdk/src/solaris/native/sun/net/spi/DefaultProxySelector.c:      dlopen("libgconf-2.so.4", RTLD_GLOBAL | RTLD_LAZY) != NULL) {
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_UNIXToolkit.c:    void* hSplashLib = dlopen(0, RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/awt/fontpath.c:    libfontconfig = dlopen("libfontconfig.so.1", RTLD_LOCAL|RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/awt/fontpath.c:        libfontconfig = dlopen("libfontconfig.so", RTLD_LOCAL|RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_GraphicsEnv.c:    libHandle = dlopen(XineramaLibName, RTLD_LAZY | RTLD_GLOBAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_GraphicsEnv.c:    libHandle = dlopen(XineramaLibName, RTLD_LAZY | RTLD_GLOBAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_GraphicsEnv.c:    void *pLibRandR = dlopen("libXrandr.so.2", RTLD_LAZY | RTLD_LOCAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/CUPSfuncs.c:  void *handle = dlopen("libcups.so.2", RTLD_LAZY | RTLD_GLOBAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/CUPSfuncs.c:    handle = dlopen("libcups.so", RTLD_LAZY | RTLD_GLOBAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_Mlib.c:        handle = dlopen("libmlib_image_v.so", RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_Mlib.c:        handle = dlopen("libmlib_image.so", RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/awt/gtk2_interface.c:        lib = dlopen(GTK2_LIB, RTLD_LAZY | RTLD_LOCAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/gtk2_interface.c:    gtk2_libhandle = dlopen(GTK2_LIB, RTLD_LAZY | RTLD_LOCAL);
../../../../jdk6/jdk/src/solaris/native/sun/awt/awt_LoadLibrary.c:    awtHandle = dlopen(buf, RTLD_LAZY | RTLD_GLOBAL);
../../../../jdk6/jdk/src/solaris/native/sun/xawt/awt_Desktop.c:    vfs_handle = dlopen("libgnomevfs-2.so.0", RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/xawt/awt_Desktop.c:    gnome_handle = dlopen("libgnome-2.so.0", RTLD_LAZY);
../../../../jdk6/jdk/src/solaris/native/sun/jdga/dgalock.c:        handle = dlopen("libxinerama.so", RTLD_NOW);
../../../../jdk6/jdk/src/solaris/native/sun/jdga/dgalock.c:        handle = dlopen(libName, RTLD_NOW);
../../../../jdk6/jdk/src/solaris/native/sun/java2d/opengl/OGLFuncs_md.h:        OGL_LIB_HANDLE = dlopen(libGLPath, RTLD_LAZY | RTLD_LOCAL); \
../../../../jdk6/jdk/src/solaris/native/java/io/UnixFileSystem_md.c:    void *handle = dlopen(0, RTLD_LAZY);

There are more for 7 (though the sendfile one seems to have been
removed there).  My recent work on 7, as well as moving the system
library changes to the tree, also involved adding options to compile
against these libraries and their proper headers at runtime.  You can
see this in the recent IcedTea7 changes which turns it on by default.

There's also stuff like this in both 6 & 7:

    epoll_create_func = (epoll_create_t) dlsym(RTLD_DEFAULT, "epoll_create");
    epoll_ctl_func    = (epoll_ctl_t)    dlsym(RTLD_DEFAULT, "epoll_ctl");
    epoll_wait_func   = (epoll_wait_t)   dlsym(RTLD_DEFAULT, "epoll_wait");

    if ((epoll_create_func == NULL) || (epoll_ctl_func == NULL) ||
        (epoll_wait_func == NULL)) {
         JNU_ThrowInternalError(env, "unable to get address of epoll functions, pre-2.6 kernel?");
    }

which is instead now linked properly at compile-time using COMPILE_AGAINST_SYSCALLS.
I don't think we're ever going to hit that one, but there is similiar stuff with
*xattr calls returning an error number in NIO2.

You can see why this is done in the Oracle environment where they want the binary to
only depend on libc (even libstdc++ is statically linked), but we should make sure
as much stuff works at *compile-time* as possible instead of waiting for users to
hit failures at run-time.

This is even more important when we don't yet have a TCK for 7.

> > I'll be blogging about this shortly.
> > 
> 
> Looking forward to reading about it in more detail!

Good.  I think it's interesting stuff.  Now you also know why 7 has been taking a while;
I wanted to get this in a good state before the first release and provide as much compile-time
testing as possible.

> 
> Cheers,
> Deepak

-- 
Andrew :)

Free Java Software Engineer
Red Hat, Inc. (http://www.redhat.com)

Support Free Java!
Contribute to GNU Classpath and IcedTea
http://www.gnu.org/software/classpath
http://icedtea.classpath.org
PGP Key: F5862A37 (https://keys.indymedia.org/)
Fingerprint = EA30 D855 D50F 90CD F54D  0698 0713 C3ED F586 2A37

More information about the distro-pkg-dev mailing list