Home | Docs | Forums | Lists | Bugs | Planet | Store | GMN | Get Gentoo!

Full Text Bug Listing

GMSoft and i talked about moving the 64bit.eclass:64bit() to return only "true"
on
64bit architectures. Currently, 64bit() returns true on all 64bit archs _and_
on hppa, as they depend on all those -fPIC Makefile patches, too. 

For that purpose I made a list of all occurences of 64bit and fPIC in the whole
tree; will attach it to this bug.

TODO: 

1) make sure there is no: "64bit && <fpic-patch>"; move to unconditional
epatch.
   Gentoo Policy sais to apply -fPIC to all shared libraries, regardless of
   ${ARCH}. In case the ebuild both produces executables and binaries, a patch
   to the ebuild's Makefile(s) is neccessary.
2) move "use [amd64, *64] && <some-64bit-patch" to either "64bit && epatch" in
   case the patch breaks 32bit archs or to an unconditional epatch.
3) remove hppa from 64bit().

Created an attachment (id=34193) [details]
List of 64bit

Created an attachment (id=34194) [details]
List of -fPIC

...got a patch to make all those fPIC ones do (64-bit || use hppa)? also...
doesnt hppa have 64bit?

we can remove this if hppa doesnt mind and all these ebuilds use the above
syntax. i strongly agree that if a function is named 64-bit, it damn well
better be returning true only on 64bit archs...

Created an attachment (id=38481) [details]
Updated list of ebuild which uses 64-bit.eclass

Ok, let's start by this list.

Here is what was done :

##./sys-apps/shadow/shadow-4.0.4.1-r1.ebuild:inherit eutils libtool gnuconfig 64-bit flag-o-matic
#./app-text/chasen/chasen-2.3.3-r2.ebuild:inherit perl-module 64-bit flag-o-matic
##./dev-libs/DirectFB/DirectFB-0.9.20-r1.ebuild:inherit eutils 64-bit
#./dev-libs/gmetadom/gmetadom-0.2.1-r1.ebuild:inherit 64-bit flag-o-matic eutils
##./dev-libs/libaio/libaio-0.3.15.ebuild:inherit eutils 64-bit
##./media-gfx/sodipodi/sodipodi-0.34.ebuild:inherit gnome2 64-bit eutils
#./media-libs/libsamplerate/libsamplerate-0.1.1.ebuild:inherit 64-bit
##./media-libs/tse3/tse3-0.2.7.ebuild:inherit eutils 64-bit
##./media-sound/cheesetracker/cheesetracker-0.9.9.ebuild:inherit 64-bit eutils
##./media-video/dvdauthor/dvdauthor-0.6.10.ebuild:inherit eutils 64-bit
#./sys-libs/ncurses/ncurses-5.4-r4.ebuild:inherit eutils flag-o-matic 64-bit gnuconfig
##./x11-libs/libast/libast-0.5-r2.ebuild:inherit 64-bit eutils
#./x11-wm/waimea/waimea-0.5.0_pre040506.ebuild:inherit eutils 64-bit

# done
## unrelated to -fPIC or newer version does not have it


Now, 64-bit isn't used anymore for -fPIC. hppa will be removed from the 64-bit.eclass when I'll mark stable ncurses-5.4-r5 on hppa.

# These two ebuilds are the only ones showing up for me that 
# are conditionally patched right now with the following commands.

wget -q -O - 'http://bugs.gentoo.org/attachment.cgi?id=4194' | grep -v filter-flags  | grep patch | grep '&&'

media-video/mjpegtools/mjpegtools-1.6.2-r1.ebuild:	[ "$ARCH" != x86 ] && epatch ${FILESDIR}/${P}-fPIC.patch

net-www/gplflash/gplflash-0.4.10-r3.ebuild:	use amd64 && epatch ${FILESDIR}/${P}-fPIC.patch

--------------------------------------------------------------------------------
# To help narrow your list of packages down I did.

wget -q -O - 'http://bugs.gentoo.org/attachment.cgi?id=34194'|grep -v filter-flags|grep patch|grep -v '#'|cut -d : -f 1|cut -d / -f 0-2|sort -u

# And I don't see anything on this bug that is directly related to 
# the toolchain@. Clue me in? Anything we need to change?
--------------------------------------------------------------------------------
# Kugelfang: Narrow down the official list of fPIC down a little please.
--------------------------------------------------------------------------------

What about use arch && append-flags -fPIC ?
This should be unconditional too or better a patch to -fPIC only libs.

for this I used "find . -name \*.ebuild | xargs grep -B 3 -i fpic | less"
It's a bit longer but I found thoses ebuild which either conditionaly use append-flags -fPIC or apply conditionaly a fPIC patch.

sys-apps/filelight
sys-apps/pciutils
app-misc/krusader
app-sci/blas
app-sci/lapack
app-text/gpdf
dev-lang/perl
dev-lang/python
dev-libs/ffcall
dev-libs/nettle
dev-perl/PDL
dev-tcltk/tclx
dev-util/cvs
dev-util/kdevelop
mail-client/evolution
mail-client/mozilla-thunderbird
media-gfx/imagemagick
media-libs/flac
media-libs/gst-plugins
media-libs/clibpdf
media-libs/imlib2
media-libs/jbigkit
media-libs/libcdaudio
media-libs/libddmpeg
media-libs/libmatroska
media-libs/libmpeg2
media-libs/libmpeg3
media-libs/plib
media-plugins/live
media-video/ffmpeg
media-video/transcode
net-dialup/wvdial
net-dns/bind-tools
net-ftp/kbear
net-im/silc-toolkit
net-irc/bitchx
net-irc/ctrlproxy
net-libs/c-client
net-libs/openh323
net-mail/checkpassword
net-misc/hylafax
net-wireless/bluez-libs
net-www/gplflash
net-www/mozilla
net-www/mozilla-firefox
sys-devel/patch
sys-libs/libcap
x11-wm/fvwm


This doesn't contain ebuild which unconditionaly use append-flags -fPIC when a fPIC specific patch would be better.


I'll try to produce fPIC patch for the ebuild listed above or at least use fPIC unconditionally.

I worked a bit on Guy's list:

app-misc/krusader - -fPIC no longer necessary (Potential problem w/
  libtool/gnuconfig on amd64 which is now solved ?)

app-sci/blas - Will be removed from tree soon. Replacement is
  app-sci/blas-{atlas, reference}

app-sci/lapack - see above

dev-lang/perl - Guy: this doesn't seem to be necessary anymore !
  Can you test just w/o -fPIC on hppa ? See ChangeLog Entry for dcccflags.

dev-libs/ffcall - Already OK. Only library here, no executables

dev-util/kdevelop - fixed

media-gfx/imagemagick - No problems in the latest stable version(all arches).
  Old ebuild will be removed.

--

solar: I remove toolchain@g.o now. amd64, alpha and hppa can handle this. Still can't remember why i CC'ed you guys in the first place ! ;-)

And of course, I started to work on thoses at the top of the list :)

I already commited theses :
sys-apps/filelight
sys-apps/pciutils
app-sci/blas

and was working on other in your "not todo" list.

I'll continue my work and notice you if I find a suspect ebuild :)

app-text/gpdf - kloeri tested and removed fPIC not necessary anymore

dev-lang/python - needs no -fPIC on amd64.

dev-libs/nettle - -fPIC no longer necessary (Potential problem w/
  libtool/gnuconfig on amd64 which is now solved ?)

dev-perl/PDL - fixed

dev-tcltk/tclx - -fPIC no longer necessary

mail-client/mozilla-thunderbird - works w/o append-flags -fPIC on amd64.

x11-wm/fvwm needs checking for alpha it compiled without -fPIC on hppa.
sys-lib/libcap doesn't have -fPIC in the latest version
sys-devel/patch has a hardened/amd64 stuff which may need checking
net-wireless/bluez-libs needs checking on amd64. works on my hppa without -fPIC
net-irc/bitchx did a nice fPIC.patch

dev-util/cvs I don't think it's still needed for alpha -> needs test
media-libs/flac newer version is already inconditionnal

Little summary of what still need to be done.

Alpha and amd64, can you test thoses plz ?

Need amd64 checking :
sys-devel/patch
net-wireless/bluez-libs

Need alpha checking :
x11-wm/fvwm
dev-util/cvs

TODO :
mail-client/evolution
media-libs/gst-plugins
media-libs/clibpdf
media-libs/imlib2
media-libs/jbigkit
media-libs/libcdaudio
media-libs/libddmpeg
media-libs/libmatroska
media-libs/libmpeg2
media-libs/libmpeg3
media-libs/plib
media-plugins/live
media-video/ffmpeg
media-video/transcode
net-dialup/wvdial
net-dns/bind-tools
net-ftp/kbear
net-im/silc-toolkit
net-irc/ctrlproxy
net-libs/c-client
net-libs/openh323
net-mail/checkpassword
net-misc/hylafax
net-www/gplflash
net-www/mozilla
net-www/mozilla-firefox

The fPIC thing in sys-devel/patch was done to solve an amd64/hardened bug. I'll
ask lv if it is still necessary in 2.5.9.

As for bluez-libs: I introduced that -fPIC there myself for version 2.5 when
the
shared libs didn't get build with it. 2.10 builds object files for shared and
static libs now in separate directories. The sed-fPIC line has already been
removed.

net-misc/hylafax is done. Btw latest version of if is already applying it
unconditionally.

x11-wm/fvwm and dev-util/cvs is fixed on alpha.

hppa moved out of 64bit(), the list is horribly messy and last activity was
half a year ago.. closing as fixed, i don't think we still need this bug. feel
free to reopen ;)

how about we close the bug when we remove the 64-bit.eclass

only rxvt-unicode (Bug 91511) and dev-lang/R (Bug 91512) use the 64-bit eclass
anymore

ive updated the 64-bit function to call 'die' if $PN != rxvt-unicode and $PV !=
R

vapier: this bug never wanted to remove the 64bit.eclass. all we wanted is that
only 64bit arches get a 64bit() == true. it used to also return true on hppa,
because hppa relies on -fPIC patches too. this misbehaviour was corrected, so i
can't see a reason to close the bug.

well regardless, the 64bit eclass is an ugly hack which is plain wrong

only two packages are left which use it and once those are fixed i'll punt the eclass