IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
It's not reasonable for use/firmware/laptop to depend on
use/firmware/wireless as some laptops come without WiFi
cards and wireless userspace to use those is specified
elsewhere anyways.
This partially reverts commit 30d3838: trying to use/rescue
with e.g. distro/simply results in conflict between SysVinit
and systemd-sysvinit; INIT_TYPE had to relation to RESCUE_LISTS
in the first place. Ugh.
This has long been a TODO item but an elegant solution
just didn't come until the night before starterkits...
some services (mostly those operating on real hardware)
do not fit virtual environments at all, won't even start.
shaba@ asked if it's feasible to extend 50-net-eth
with a generator for systemd-networkd style configs
having provided examples; here it is (depends on
/etc/systemd/network/ being packaged into that one).
(fixed up by shaba@'s removal of superfluous quotes)
gdm2.20 seems rather obsolete by now, let's move on;
and m-p doesn't just lump a huge bunch of stuff in,
vector fonts for installer are requested explicitly.
...by moving reference to a package list that *deducts*
packages from a feature (that should lend itself for reuse)
to a particular distribution's configuration (that can have
some specific polish).
The problem was that basing junior on slinux feature while
adding some KDE/Qt-based packages to it failed miserably
in a hard-to-debug manner: adding every package that's been
requested but not installed by hand suddenly made it build,
see also http://altlinux.org/mkimage/debug [ru]
mixin/desktop-installer became *quite* inobvious
even for me over time, and it's not easy to grep up;
let's introduce explicit targets where one is expected
to expect those.
rootfs scripts should hit installer some day; the problem
is with variables (dumping 'em wholesale looks dirty,
and proxying those sort of defeats the approach)
rather than with scripts.
Until then, transform the data from the single variable
into a file containing one facility per line for
installer-1.8.31+ to consume.
As noted in the comment, these include a few quite strong ones:
- sshd(8) will only allow in "wheel" and "users" members
by keys, no password access is allowed;
- password change even by root is subject to quality checks;
- su(8) is only useful to lower privileges and not gain those
(so root access is available either through local console
or via use of ssh keys).
Don't use if frowned upon.
This is based on distro/regular-jeos but torn into two
and somewhat updated for sisyphus-going-to-bring-p8:
1) libcap-ng is now required by util-linux;
2) bridge-utils might be needed for subsequent images.
Those packages which are *required* should be available
for standalone use; and those which are optional should go
into extras.
Adjust server feature accordingly.
The issue with these "; @:" thinglets is that mkimage-profiles
relies on target tracing (see commit 788cad8 some four years ago);
and this tracing approach relies on non-empty recipes which do call
shell (which gets (ab)used) unlike empty ones which oviously don't.
So this _will_ be traced properly:
a: b
@echo "hello world"
and this will too:
a: b; @:
but this will result in a broken graph with REPORT=1:
a: b
icon-theme-oxygen is required by kf5-oxygen <- kde5 <- kde5-big
by now so it's not needed to specify it explicitly anymore;
and zerg@ has just packaged some translations, let's jump in!
There's /usr/share/qt5/translations/ now too, handle that.
OTOH we've got some CJK support already, time to split up
those "cleanups" into some target locale set dependent form.
This has no users in master but out-of-tree branches might need
a trivial update.
The rationale is that it's actually for *any* stage2 and not related
to specifically "install" at all (otherwise it should have been moved
to install2 feature altogether).
Note that there's no reason to add nfs-utils similarly as make-initrd
requires kinit-utils which includes its own nfsmount.
The "best" "feature" of systemd "init system" has just
emerged once again: it will happily sit there idling
given startup or shutdown loops resulting in what looks
like a hang to an unsuspecting user; let's provide the
suspecting one with at least some tools described at
http://www.freedesktop.org/wiki/Software/systemd/Debugging
The issue with this was that plain use/browser/firefox/esr
didn't actually pull in the feature as such; it MUST NOT
pull in use/browser/firefox since it will change semantics
from "if it's Firefox make it ESR" to outright "use FX ESR"
(starterkits depend on the former and it was intentional).
FX_FLAVOUR variable can be set anywhere to switch
use/browser/firefox to prefer ESR packages, including
the appropriate localization ones.
Note that there's no dependency as it can be set in e.g.
starterkits (still unset in regular builds) wholesale
but shouldn't affect those of them lacking firefox.
The reason behind this silly patch is that the default URL
can be left alone with no rebuilds neccessary but with the
intranet services delivered through a "captive portal" or
a redirecting proxy; we definitely don't want the canonical
wiki URL, http://www.altlinux.org, blocked by a rule made
for redirecting the default homepage, so let it be another
one which is served but not widely known or linked to.
This makes use of IM_PACKAGES variable processed by
newly added im feature so that DE-specific targets
could tell which DE-specific IM packages they'd like
on a system *iff* use/im has been requested.
Might be lacking right now, to be sorted out with
the actual users.
This one has been brewin' for quite a while but has been
completed finally; some tweaks sure can come in later but
it's working.
Please note that it's rather needed for "proper" distros
with specific branding and docs packages prepared for those;
one should use l10n feature most likely too.
The "full" target should care for rescue bits as well
(remember that THE_* won't go there); thus regular-rescue.iso
will receive these couple hundred useful kilobytes as well.
It's the very same problem that must be solved within mkimage:
some package lists get expanded early and some late thus having
no chance to influence apt's choices of alternatives made early
(in fact, too early).
Until that, here's another kludge...
PS: turns out that ^systemd- is not "drop ^systemd" but rather:
systemd-analyze
systemd-coredump
systemd-journal-gateway
systemd-networkd
systemd-sysvinit
-- thus one /really/ wants something else.
This one was an experimental but the server is long
offline and isn't going back up; remove the obsolete
config snippet, if/when it's done again it's the easiest
part to be restored (the implementation should provide
HTTP/FTP/NFS-publishable deliverables without the need
to extract those from ISO images).
This one relies on the controversial polkit-sysvinit package
that subverts policykit using well known groups to make it
"work" for things like NM and shutdown helpers.
See also http://altlinux.org/sysvinit and feel free to improve.
/etc/sudoers is persistent with regard to userdel(8)
so removing a LiveCD user isn't going to drop this kind
of the added privilege and might result in an unintended
grant of those by adding a user with the same name after
permanent LiveCD installation.
This has been spotted by Speccyfighter:
https://bugzilla.altlinux.org/31071
This one is alike to install2's one; it's not a shared rootfs
script/variable though as contexts differ a lot, let's be careful.
The commit has been missing from 1.1.64 somehow, found in patch
series while figuring out why LIVE_CLEANUP_KDRIVERS seems to be
just ignored in live-privacy *after* the massive rebase of that
branch...
There's a convention that syslinux configuration snippets
carrying the names of subprofiles involved are picked up
automatically; there were a few special cases already
when this is actually inconvenient, and there's another
one at hand so let's just step up and do it.
NB: this is a sort of a hacky hook though, wish an elegant
interface would come to mind some day.
The added initscript used to be purged by 98-init-rescue
which has been somewhat overlooked during vain attempts
to build an image that would actually run it!
This one provides cmdline arguments for startup-rescue >= 0.24
which would bring up networking and sshd in its turn thus allowing
remote access to the host booted in this mode.
The feature has been asked for by many people including mithraen@
and valintinr@ (and I'd make use of it another day too).
See the appropriate startup-rescue commit description for notes
on implementation; this default set of variable values should be
both useful and illustrative though.
A recent commit has dropped wireless support from
regular server images; staging modules might still
come handy in some situations, let's keep those in
but not as a part of default installation.
This one is likely to get just a single user right now
but the future potential is clearly higher.
Please do review libzmalloc implementation if concerned.
This is sort of laying the ground for the future dismantling
of 10-stage2 (which was sub.in/stage1/modules just recently);
things look like tagged lists might become due some day, e.g.
"net+usb" or "scsi+raid" -- time will tell.
These are aimed to test the modules.d/ and auto-pickup
implementation as well as to present an example.
At least 50-net might change (or just get renamed to avoid
auto-pickup) some day as the "net" feature's meaning is
to provide networking upon bootup and these modules are
only needed within stage1 if we're going to netboot;
and that's quite different thing.
armh-cubox bits are prone to get renamed/generalized too
since e.g. ArmadaXP based server images are going to need
this as well.
These were produced off the single sub.in/stage1/modules
file using this scriptlet to prefix/annotate the names:
grep '\.ko$' modules \
| grep -v / \
| while read m; do \
echo "$(find /lib/modules/$(uname -r)/kernel/{drivers,fs} \
-name "$m" -printf %P $m $(modinfo -d "${m%.ko}" 2>&1)"; \
done
...with subsequent sorting and manual separation.
This is meant to be the second stage in monolithic modules
file split, so the lists themselves are largely unmolested
otherwise. The plan is to further split those into prefix-
and module-specific ones.
Add a note clarifying 10-stage2's status, by the way.
What was a static sub.in/stage1/modules (and the only one)
is now features.in/stage2/stage1/modules.d/10-stage2
(basically a compatibility file that might go some day).
It will be auto-picked as its name corresponds to the
NN-SUFFIX pattern specified in stage1 subprofile now
with $(FEATURES) going into default STAGE1_MODLISTS.
stage1's got prepare-modules target collecting
modules file snippets all over stage1/modules.d/
subdirectories within individual features.
stage2 now adds names of all the features going into
a particular image as snippet file suffix list so that
individual features don't have to register themselves
twice (as a feature and as a propagator modules.d
snippet carrier).
This is going to allow both "uncommon" modules getting
included with no problem (sin@ has wanted cifs ones
for quite some time, for example, and some want e.g.
infiniband modules) *and* to reduce the actual list
below the common mark as well (which is the case with
live-privacy image, for one).
And stage1 memory consumption does matter in some cases
as it's highly critical with no chance to use swap yet.
...and split off use/live/.base *without* use/deflogin/live.
There's need for live images without predefined logins
(like e.g. live-privacy image).
NB: this commit might break things for someone, please notify.
The unfortunate thing is that we have to take care
for sessions, somehow; still there are only two for now
(LXQt and KDE5 Plasma Desktop) so this doesn't look like
a disaster just yet.
Commit 657c0bf has silently added use/bootloader
to the base use/install2 target thus breaking
experimental distro/netinst; it seems better to
require *a* bootloader in the target that's been
specifically designed to cover the common case
(thus linked to by +installer shortcut) but still
to have our base lightweight and flexible.
This doesn't hurt the actual distros as these use
+installer of course.
The former approach to handling "LiveCD with sessions"
has been to mangle "automatic=method:cdrom" into
"automatic=method:disk,label:ALT*" within gfxboot
so that propagator and make-initrd-propagator would
try and discover/create a filesystem labelled
"alt-live-storage" on a LiveFlash's free space.
Then "live_rw" handling has been unified in
make-initrd-propagator (as of 0.18-alt1) to accept
any of "label" subparameter or "live_rw" argument
to go and create_disk_slice().
Then propagator's cdrom.c has been fixed to actually
try sdX1 before sdX (as of 20150306-alt1).
And now it's all been tested to verify that:
- flash "ro" and "rw" boot is OK
- CD-ROM "ro" boot is OK
- CD-ROM "rw" boot is fine given that there's
a partition labeled "alt-live-storage" elsewhere
This is a can of worms indeed :-/
References
~~~~~~~~~~
* http://altlinux.org/initrd-propagator
* http://altlinux.org/make-initrd-propagator
* http://bugzilla.altlinux.org/28289
It's entirely unclear to an unsuspecting curious user
where the actual results of a proposed example hasher
build end up; that's ~/hasher/repo, just state that.
The former install2-only "bloated binary" purge script
happened to hit stage2 (which is a lot more than just
install2); a kind of safety net has been stuck into it
to guard installable LiveCDs against this particular
cleanup but seems it was not enought for ildar@ who
reported this problem almost three years after it was
introduced.
This change re-places the script back into install2
section; the binaries in question amount for ca. 8 Mb
(except openssl ildar@ asked about); if these are deemed
unneccessary within any other stage2-based subprofiles,
please step up with details.
use/vmguest/vbox/base used to pull in DRM modules
which are required for vboxvideo but useless without
xorg bits; and all of these aren't needed in jeos.
Things might break, doublecheck please.
When installer-feature-systemd-stage3 hits BASE_PACKAGES
it pulls install2-init-functions in which is wrong
(one of the consequences is that alterator-browser-qt
lands into even a very basic server installation).
And install2 doesn't even need that package as init feature
carries a script hook that does the same...
This project has evolved/merged into LXQt which has been
packaged for both p7/t7 and sisyphus by now, no need to
carry on deprecated bits.
NB: 0.6.x still have it as t6/p6 still bear razorqt.
The installer feature added is a trivial wrapper around
apt-cache nodeps to uninstall the ^lib packages that have
no more dependencies upon those when the temporarily
installed packages like alterator-browser-qt get removed.
This has only been useful for plymouth feature,
and +installer shortcut included this target
for all the wrong reasons as it seems today
(thus blocking the DRM-free server installers,
for example).
This authorized_keys file has been downloaded to get incorporated
into a script hook but was looking common enough to be forgotten
during pre-commit feature cleanup unfortunately; fix that.
A few more leftover libraries tend to hang around after
purging extra alterator packages that have fired already
during installation stage3; this change might hurt someone,
please do notify if that is the case (OTOH one isn't forced
to use it or to inherit intermediate targets that do so).
This feature operates LIVE_* variables specifically
(as opposed to the more generic THE_* ones) so +alsa
isn't exactly suitable but reusing the pkglist that's
just been factored out is fine.
Split package lists:
- base alsa packages (also needed in pulseaudio-based installations)
- additional alsa packages (needed only if not using pulseaudio)
- pulseaudio packages
live-webkiosk.iso regressed into trying to boot off the local disk
immediately, the exact cause is not yet discovered but this one
has popped in build trace (distcfg.mk) already.
Desktop images are still likely to depend on it though.
There were quite a few things missing:
- packager pseudo for both rpm and hasher;
- proper target (so that i686 or athlon don't confuse people);
- hasher repo for apt so that mkimage would use it too;
- last but not least, a friendly bootstrap message! ;-)
THE_* and friends are all unneeded for live builder image;
this commit makes dev feature basically incompatible with
install2 feature (which relies on the conventional behaviour
of main subprofile), will have to think it all over if the
combination of these two ever becomes needed.
The problem with initial implementation (commit 62e7e9c)
is that there's no systemd-services package in p7/branch
thus apt complains about an attempt to remove something
that doesn't even exist in the first place.
Aimed at live images at first but should cover installers as well.
This has been brewing for quite some time and while the proper
implementation is considerably more complex (and hard to do)
looks like there's demand for the particular important use case,
namely LiveCDs for Russian users, so this code has been shared
with a few people before merge.
E19 would ask the user if they want to shut down
when facing power button event; it won't get a chance
though as the system will hurl down immediately as per
acpid-events-power package provided configuration.
This should avoid ruining principle of the least surprise
with ROOTPW_EMPTY=0 or ROOTPW_EMPTY=n actually *enabling*
empty root password; overriding an already set "1" with "0"
becomes possible either.
This one has been inspired by these guys:
http://www.informatimago.com/linux/emacs-on-user-mode-linux.htmlhttps://raymii.org/s/blog/Vim_as_PID_1_Boot_to_Vim.html
It's aimed at building images running their main userspace
piece instead of ramdisk's init, that means PID=1, UID=0.
Mostly fun of course but it suddenly became interesting with
kernel IP autoconfiguration and e.g. elinks running this way
(NB: requires patched make-initrd 0.8.8 at the moment to get
resolver configured).
And startup times are way better than sysvinit and systemd combined!
This function's got its argument order chosen for "aesthetical"
reason of $(2) following $(1) in the macros but the logical order
is exactly the opposite: we care for kernel flavour much more than
for module set (which is dependent upon it).
So while silent dropout of kernel-image if KFLAVOURS is set
but KMODULES is empty could be fixed by testing for $(2) only,
it looks like a good time to fix this discrepancy altogether.
stage2 has been thinking it's synonymous with propagator
and used to usurp kernel's belongings either; carefully
tear scripts apart so that kernel feature makes sure
initrd gets generated, and stage2 (which is still all
about propagator) cares for its bits.
xorg-drv-vmware is desirable for guests with X11
but undesirable for text-only ones; let's provide
this knob at least but ideal m-p would figure out
that an image with use/x11 and use/vmguest/vmware
should receive this intersection either.
Maybe firmware feature should be merged into kernel feature
as the firmware binaries added by it are only used by kernel
but let's clean up a bit at a time.
"use/stage2/net-eth use/net-eth" would be common enough
for installers or livecd images to just get these pulled in
together by a convenient shortcut thus reducing confusion
and chances to just forget one of these counterparts.
It looks *ugly* on-screen, at least within regular builds,
even if the screen is 166dpi.
Based on a quick experiment this morning I'd suggest using
fonts-otf-adobe-source-{code,sans}-pro instead -- and it's
available as use/fonts/otf/adobe now, incidentally.
The documentation is still built with it though as a2x/fop look
unhappy otherwise (as in replacing Cyrillic glyphs with "#"s).
Font packages are sprinkled all over the metaprofiles,
let's try and help make their use more systematic.
This is a sort of a feature abuse as it was conceived
for fontconfig setup originally but spawning features
with confusing names looks grim; so let all things fonts
live within a feature named "fonts" for the time being.
There's not much sense in overduplication of documentation
(tends to get stale faster then), still it's not good to
just refer to the code as the PDF/HTML book is less useful
then; maybe drifting towards "recommended" bits with more
"advanced" things being impleentation-defined is better.
That's a part of ALT Linux conveniences: system log
messages at tty12 (helps immensely in case of disk crash
or cable problems as running anything, including utilities
to view logs, becomes painful to impossible in such cases).
systemd lacks this kind of setup out-of-box for sure
so zerg@ hacked a substitute together; just pull that in.
fonts-ttf-droid have been superseded by fonts-ttf-google-droid-sans
in Sisyphus which might be lacking; I've considered replacing it
with Fira anyways, so let's just do that.
This needs further refinement regarding p7/t7 specifically:
NM behaviour regarding defaults differs in sisyphus and this
has led to livecds booting with DHCP networking but installed
systems booting without configured interfaces.
Non-GUI packages moved to base+nm pkglist to enable standalone
installation of those; and GTK bits left in desktop+nm for use
by images lacking their own new and improved(tm) variant.
Note that both GNOME3 and KDE4 aren't lacking anymore.
This is a similar trouble: p7/t7 branches had
plasma-applet-networkmanager while sisyphus has
switched to kde4-plasma-nm* (there's a bunch of
subpackages there, basically all of them desired).
The current branches lack both firefox 29+ and
firefox-classic_theme_restorer, correspondingly;
sisyphus has those; the feature shouldn't pose
any problems in both cases, should it?
Firefox was the very reasonable default for initial livecd
implementation but now that at least initial browser chooser
infrastructure is in place it's time to un-hardwire its use.
It's _the_ default but switchable now so that images providing
a comprehensive browser can avoid feature duplication.
This one has been asking to be implemented for too long already,
and zerg@ was interested in a bit more lean and mean regular-kde4
either (there are two browsers provided with it via metapackage).
There's another reason to do it recently: Firefox Australis UI
is not exactly the best for many of us, and good ol' seamonkey
seems preferable for "vintage"/low-resource images coming with
icewm or windowmaker.
led-ws kernel flavour has gained kernel-modules-vmware
recently, let's add this to the appropriate targets.
It's used in regular-jeos already but THE_ part was missing.
dm service is set up to autostart when installed anyways,
and explicit `chkconfig dm on' results in it being turned on
at runlevels 2, 3 and 4 too which is really not needed.
Thanks led@ for spotting and reporting this.
cfg.in/README should be explicit regarding
"automatic=method:cdrom" being usable for
flash media too (propagator has been fixed
since 20101130-alt10 or so, and gfxboot is
able to tweak the cmdline having figured
out it's running off the flash either).
It's by no means substitution for proper l10n feature
but forcing users into POSIX locale for recovery ops
is no good at all.
This is basically a fork of live feature's 20-locale,
a font has been changed to save some face though.
VMware specific bits went into use/install2/vmware target,
and all of those targets are worth their use/install2/vmguest
collective one instead of just sticking the kitchen sink into
use/install2/full immediately.
This feature intrinsically depends on predictable
ethernet interface names and makes no sense without
those; so it only seems reasonable to bring this
nice package in, huge thanks go to shaba@ of course.
This value is used to authenticate rescue rootfs image
by verifying the squashfs file's sha256sum before use
(propagator-20140419+).
Looks like this check might be useful for other stage2
images as well but let's get started with this one.
Thanks Maxim Suhanov <suhanov/group-ib.ru> for both
http://www.forensicswiki.org/wiki/Forensic_Live_CD_issues
and propagator patches.
Thanks Maxim Suhanov (suhanov <AT> group-ib.ru,
http://www.forensicswiki.org/wiki/User:.FUF)
for taking the time to review regular-rescue image.
Note that there are more than just filesystems:
arrays, logical volumes and swaps aren't activated either;
startup-rescue >= 0.18 should make that clear enough.
syslinux shortcut handling is case-insensitive,
let's find yet another letter...
A variant of rescue that marks the need to be careful
towards block devices and filesystems thus reducing
the amount of auto-activation done by startup-rescue
(0.17 or newer).
This is a refactored result of Zabbix-related experiments;
we can do a rough zabbix server sketch that still requires
its own setup to go.
NB: both the pkglist and the target are describing several
distinct things actually: zabbix server, zabbix agent,
and the underlying SQL/HTTP/SMTP servers which might get
their own smaller targets some day.
It appears that live feature has been buggy regarding user
groups: its 30-users script would create a predefined account
with fixed supplementry groups list, and even if deflogin feature
got used too it would fail to add any groups to already existing
account since its useradd(8) call would fail.
Let's drop this duplication which has been long overdue anyways.
Thanks dd@ for both reporting the problem and carrying out
initial investigation.
There have been several problems with this feature:
- a typo;
- non-existant GROUPS (even a single one) would block setting
all of the supplementary groups but separately-set 'wheel';
- this feature isn't used much actually so sees no battle testing.
The typo has been just fixed; GROUPS are now applied by iteration
which is less effective but more reliable; an additional script
hook to write down login invitation for the first passwordless
account (if any) has been implemented; and several more group
managing targets have been added (based on live feature's script).
This relates to commit f2892ad3e4
as there's an obvious need to be able to set empty root password
for LiveCDs but previous implementation was very fragile (and is
going to stay that way) -- so clear and separate knob for making
an image defenseless looks better.
Whoops, the very first build of a real distro with gfxboot
has shown that the label isn't picked up there... and things
are actually worse: iso.needscheck gfxboot test seems to look
up "check=1" in sectors where it might have been landing back
then but it's just not there by now; some kludgery is due in
branding-altlinux-sisyphus unfortunately.
No use to hunt make or diff file-by-file.
Well this chroot should have been more lean
in the first place (or a few files in initrd)
but life is short so better use/baby/steps.
This one is quite different already and utility-based name
was pretty clumsy; meet the new feature and retire the old
experimental one.
Please note that quite aggressive cleanups are implemented
within this stage2-based subprofile for the simple reason
that it has a single task to do; nothing else is expected
to be configured into it for that matter.
This functionality asks to be further moved into initrd of course;
adding it there will take a few more decisions to be made, mostly
regarding user interaction in failure scenarios, and it looks like
mkimage will have to be patched in case this doesn't just go into
full.cz under some sort of conditional check.
It's not exactly obvious how install2_size, live_size or rescue_size
get defined since the variable names themselves get constructed;
help git grep these down.
acpid is not enough since power button handling configuration
has been split apart; and tracking this in zillion places is
utterly useless in face of a specially trained power feature.
Just use it.
This one has been missing for quite some time (infiniband modules
should have triggered a commit like this back then), finally there
in very crude and draft form for the starters.
By the time these hooks run the font packages' %post scriptlets
should have fired already; no need to carry the utilities on.
Yes these are bit-by-bit savings. No it's too expensive still.
My gut feeling is that we're not going to see glib2's
messages a lot within installer environment anyways.
And there's a forgotten /usr/share/X11/locale/ too.
An installer needs video playback acceleration
when it has some content to show and some means to;
as long as these are not supported just drop this
unconditionally.
These are only needed for alterator-vm when making
LUKS encrypted partitions; ideally the extra libraries
would be omitted automatically when luks isn't included.
Looks like today's xorg won't autoload radeon_drv but
insists on ati_drv falling back to fbdev if it's not there;
FlightGear runs definitely slow on C-60 APU with that.
I didn't specify ati since it pulls r128 and mach64 modules in
which are rather useless in this context (accelerated 3D graphics).
