[patch] Re: OpenOffice.org-2.0.4rc1 compile problems on -CURRENT

[Jung-uk Kim]

On Monday 18 September 2006 05:58 am, Kostik Belousov wrote:
> On Fri, Sep 15, 2006 at 02:17:12PM -0400, Jung-uk Kim wrote:
> > On Thursday 14 September 2006 10:19 pm, Nicolas Blais wrote:
> > --- >8 --- SNIP!!! --- >8 ---
> >
> > > =============
> > > Building project udkapi
> > > =============
> > > /usr/ports/editors/openoffice.org-2.0/work/OOD680_m3/udkapi/com
> > >/sun /star/uno mkout -- version: 1.7
> > > idlc @/tmp/mkmJ4vAL
> > > idlc: compile 'Exception.idl' ...
> > > idlc: could not load registry dll.
> > > idlc: detected 1 errors
> > > Sun Microsystems (R) idlc Version 1.1
> > >
> > > dmake:  Error code 1, while
> > > making '../../../../unxfbsdi.pro/misc/urd_cssuno.don'
> > > '---* tg_merge.mk *---'
> > >
> > > ERROR: Error 65280 occurred while
> > > making
> > > /usr/ports/editors/openoffice.org-2.0/work/OOD680_m3/udkapi/com
> > >/sun /star/uno dmake:  Error code 1, while making
> > > 'build_instsetoo_native' '---* *---'
> > > *** Error code 255
> > >
> > > Stop in /usr/ports/editors/openoffice.org-2.0.
> > >
> > > The port tree is up to date and so is -CURRENT. Any
> > > suggestions?
> >
> > src/libexec/rtld-elf/rtld.c 1.117 has serious regression:
> >
> > http://docs.freebsd.org/cgi/mid.cgi?200609081459.k88ExtSR084500
> >
> > You can revert to 1.116 and pass this but there is some other
> > issues. :-(
> >
> > Jung-uk Kim
>
> Ok, below is the patch that shall fix the issue. It was tested by
> Jung-uk Kim' regression test. Also, adopted patch (for STABLE) was
> tested with KDE (opening https:// sites) and OpenOffice 2.0.3.
> The latter was simply run and several documents of different
> types where opended. I have no resources to rebuild the beast
> myself. Konqueror issued a lot of warning for undefined symbols
> (like kdecore (KLibLoader): WARNING: KLibrary: Undefined symbol
> "OPENSSL_add_all_algorithms" ), but it does it with both patched
> and non-patched rtld on 6-STABLE. In both cases it worked normally,
> allowed to open the sites, and I see no excessive VM usage (I
> assume 30-40Mb is normal for it).
>
> Original patch (committed as rev. 1.117) scanned objects from the
> obj->dagmembers list for symbol retrieved by dlsym(). This appeared
> to be wrong by two reasons:
>
> 1. dagmembers are initialized only when dlopen actually load the
> dso. This caused the second test from Jung-uk Kim' regression set
> to fail, because it dlopen() depended dso and tries to lookup
> symbol from there. 2. In the case dso A depends on B and C, usage
> of dagmembers would cause B to expose symbols from C, that is wrong
> too.
>
> Attached patch uses obj->needed, that is filled for each dso, and
> is a real DAG instead of list. Patch (hopefully) implements
> SUSv3-required breadth-first lookup for dependent symbols.
>
> Please, give it a try.
--- >8 --- SNIP!!! --- >8 ---

Similar patch is committed to -CURRENT:

http://docs.freebsd.org/cgi/mid.cgi?200609191247.k8JClDaT048713

If you have time and electrons to burn, please cvsup your system, 
rebuild/reinstall world, and test the attached patch.  You will have 
to remove patch-i66667 from files directory and drop this patch in 
there instead.

Thanks!

Jung-uk Kim
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch-bridge
Type: text/x-diff
Size: 4493 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-openoffice/attachments/20060919/dc3bed21/patch-bridge.bin