[RFC]: make IcedTea-web to be compatible with RHEL5 libraries

[Deepak Bhole]

* Jiri Vanek <jvanek at redhat.com> [2011-08-25 05:53]:
> On 08/23/2011 01:33 PM, Jiri Vanek wrote:
> >On 08/22/2011 10:55 PM, Deepak Bhole wrote:
> >>* Jiri Vanek<jvanek at redhat.com> [2011-08-22 09:21]:
> >>>On 07/21/2011 03:21 PM, Dr Andrew John Hughes wrote:
> >>>>On 12:08 Thu 21 Jul , Jiri Vanek wrote:
> >>>>>
> >>>>>Hi!
> >>>>>
> >>>>>This patch is another attempt o make icedtea-web buildable on rhel 5 with older libraries. Before build of plugin itself is started, then testing c++ program is tried to be linked with same options. If this compilation fails, then original IcedTeaNPPlugin.cc is backuped, glibBuildable.patch is applied to IcedTeaNPPlugin.cc and then plugin is linked.
> >>>>>During clean-plugin original IcedTeaNPPlugin.cc is restored from backup (and when test application is compiled, then this binary is deleted, also backup is delete)
> >>>>>
> >>>>>Please, when you will ask me to include this "solution" more to configure (or autotools or anywhere) then I will need a little bit of help.
> >>>>>
> >>>>
> >>>>I suggest taking a look at the autoconf manual and function detection:
> >>>>
> >>>>http://www.gnu.org/software/autoconf/manual/autoconf.html#Library-Functions
> >>>>
> >>>>Then, rather than patching, I'd add a header file with #ifdefs to handle both cases.
> >>>>
> >>>
> >>>2011-08-22 Jiri Vanek<jvanek at redhat.com>
> >>>Added functionality to allow icedtea web to be buildable with
> >>>rhel5 libraries
> >>>* testGlibs.cc: testing file. If not compiled, then Rhel5Compatible
> >>>macro is defined for compiling IcedTeaNPPlugin.cc.
> >>>* plugin/icedteanp/IcedTeaNPPlugin.cc: added replacements for incompatible
> >>>functions, added #define sections for use this function instead of glib ones
> >>>*Makefile.am: (stamps/patch-for-glib): new target, compiling testGlibs.cc
> >>>($(PLUGIN_DIR)/%.o): now depends on stamps/patch-for-glib and define
> >>>Rhel5Compatible if testGlibs.cc was compiled successfully
> >>>($(PLUGIN_DIR)/IcedTeaPlugin.so): same as ^
> >>>(clean-IcedTeaPlugin): removes compiled testGlibs if exists
> >>>
> >>
> >>Is this RHEL 5 specific or glib version specific? It looks to be the
> >>latter. In which case no test program should be needed -- just a check
> >>for glib version and conditional compilation accordingly.
> >
> >Very well then :)
> >I will add this to acinclude.m4
> >
> >
> >AC_DEFUN_ONCE([IT_GET_GLIBCVERSION],
> >[
> >AC_MSG_CHECKING([for glibc version >= 2.12])
> >basicVersion=`ldd --version | grep "@<:@0123456789@:>@\.@<:@0123456789@:>@" | sed "s/ldd (GNU libc) //"`;
> >echo "- $basicVersion -"
> >versions=(`echo $basicVersion | sed "s/\./ /g"`);
> >if test ${versions@<:@0@:>@} -lt 2 ; then
> >GLIBCVERSION="no"
> >else
> >if test ${versions@<:@0@:>@} -gt 2 ; then
> >GLIBCVERSION="yes"
> >else
> >#maybe
> >if test ${versions@<:@1@:>@} -ge 12 ; then
> >GLIBCVERSION="yes"
> >else
> >GLIBCVERSION="no"
> >fi;
> >fi;
> >fi;
> >AC_MSG_RESULT([${GLIBCVERSION}])
> >AC_SUBST(GLIBCVERSION)
> >])
> >
> >
> >
> >this will add make variable GLIBCVERSION for conditional compilation in makefile.
> >The docmentation says "since 2.16" but it can not be true, as rhel5 have 2.5 and have missing those functions, and rhel 6 have 2.12.5 and my F13 have 2.12 and both HAVE those functions .... :-/ But it is not firs time glibc documentation was wrong.
> >
> >I still think compilation is much more bulet-proof, but if you are ok with this then I will post updated patch.
> >
> >Regards J.
> >
> >(maybe useless) ps: @<:@ is replaced by [ and @:>@ by ] during configure generation... no other way here...long live the Autotools :(
> >pps: I'm happy autotools-lame, please feel free to correct me in above script as much as possible.
> >
> >
> >>
> >>Cheers,
> >>Deepak
> >>
> >>>
> >>>Regards J.
> >>
> 
> Following Omair's advice, instead of version checking I have used AC_TRY_LINK:
> 

Hi Jiri,

Just noticed this now.. why is the variable called GLIBC*? You are
checking GLIB, not GLIBC..

Also, I am still not sure why we are checking by functions.. the
iterator functions were added between 2.14 and 2.15 according to the git
history. So we just need to check if GLIB is < 2.15 and then define
RHEL5Compatible.. although I think that should be renamed. It should
instead be called -DGLIB214 or something, as this is a glib thing, not a
RHEL thing.

Comments below:

>  
>  $(PLUGIN_DIR)/%.o: $(PLUGIN_SRCDIR)/%.cc
> +	if [ "$(GLIBCVERSION)" == "no" ] ; then \
> +	  RHEL5COMPATIBLE="-DRhel5Compatible" ;\
> +	else \
> +	  RHEL5COMPATIBLE="" ;\
> +	fi; \
> +	echo "rhel5comaptible="$$RHEL5COMPATIBLE ;\

Rhel5Compatible should be renamed do GLIB214. Also, the if is not
needed, see below.

>  
> +AC_DEFUN_ONCE([IT_GET_GLIBCVERSION],
> +[
> +AC_MSG_CHECKING([for glibc  version >= 2.12])
> +AC_LANG_PUSH(C++)
> +CXXFLAGS_BACKUP=$CXXFLAGS
> +CXXFLAGS=$CXXFLAGS" "$GLIB_CFLAGS" "$GTK_CFLAGS" "$GLIB_LIBS" "$GTK_LIB
> +AC_TRY_LINK([
> +#include <glib.h>
> +],[
> +GHashTable* instance_to_id_map = g_hash_table_new(NULL, NULL);
> +GHashTableIter iter;
> +gpointer id, instance;
> +g_hash_table_iter_init (&iter, instance_to_id_map);
> +g_hash_table_iter_next (&iter, &instance, &id);
> +g_strcmp0 ("hell", "hello");
> +],[
> +AC_MSG_RESULT(yes)
> +GLIBCVERSION="yes"
> +],[
> +GLIBCVERSION="no"

Just a version check will suffice, and it can set a variable
"GLIB214=-DGLIB214" or keep it empty if version is >= 2.15

Cheers,
Deepak