xdm rc scripts

[Dejan Lesjak]

On Saturday 19 of February 2005 20:24, Jose M Rodriguez wrote:
> El Sábado, 19 de Febrero de 2005 19:55, Dejan Lesjak escribió:
> > On Friday 18 of February 2005 22:25, Jose M Rodriguez wrote:
> > > Hi,
> > > I launch PRs ports/74000 and ports/74003 time ago, to make launch
> > > of xdm from a rcNG script possible (Not mandatory).
> > >
> > > I think that this must be part of X11 clients ports, but if someone
> > > find any issue with this, let me know.
> >
> > Hi,
> > Some of my issues:
> > In my opinion there is no need for a rc script to start xdm, since
> > this is already taken care of by /etc/ttys which also starts gettys
> > on other ttys - so all of gettys and xdm, which "[...] provides
> > services similar to those provided  by [...] getty [...]" are in one
> > place. This putting of configuration of things which are similar
> > seems to be exactly what you intended, or am I completely mistaken?
>
> Not really.
> First, the offered scripts are the only method I know to unifor launch
> xdm/gdm/kdm.  gdm can't be launch from /etc/ttys.

Yes, but as I already explained, having a script that would take care of (at 
least) 4 ports in one port doesn't seem the right way to go, so I was 
explicitly talking only about xdm here. The only way that such script would 
make sense (in my opinion) would be if it was installed in base (which I 
don't think would be good for other reasons) or by separate port (which would 
seem overkill for installing an optional script as opposed to rc_subr which 
installs script that is required by several ports).
However it would perhaps be nice if you could submit this alternative way of 
starting display managers to doc@ and they might consider adding it to 
Handbook as an example.

> > Furthermore the PRs you submitted would require us to patch XFree86
> > and X.Org code which is not necessary, because that code is perfectly
> > fine and has worked, works and probably will work for some time to
> > come. We have in X11 ports quite a few of patches which are needed to
>
> The patches to Xorg/X11 code are not needed for the scripts.  Only
> install the rc script.  The patches try to solve a real race problem
> between init (the gettys) and gdm (which must be launch from
> localpackage).
>
> This is documented in main XFree86/Xorg docs.  If gdm/xdm/kdm doesn't
> have 'hard assigned' an vt, it may stole vt (in the FreeBSD case,
> vt0/vt1) if it get running before init launch gettys.
>
> If this happens (I often get this in fast machines), you may end in
> front of a gdm login script, with mouse, but without any keyboard
> input.

Changing things so that xdm starts from rc script seems like a regression 
then. See as far as I understand things work just fine now and I want to know 
if changing then really brings enough benefits that would merit patching 
vendor sources and adding two files in ports tree. I'm also concerned about 
possible confusion that would result in having these rc files installed - 
users changing both /etc/ttys and rc.conf and weird problems that could 
result in that.

> > split installation of X11 distribution into separate pieces to
> > hopefully ease maintenance for users in case where only one component
> > needs updating. These are the patches that will never be submitted to
> > upstreams, since they are completely ports specific and both of X11
> > build fine without them. Your patch to
> > programs/xdm/config/Imakefile would increase the burden of
> > maintaining local patches for what, at least to me so far, doesn't
> > seem like something that actually needs patching - xdm on FreeBSD
> > will take the first virtual terminal available so hardcoding default
> > doesn't seem to be the right way. Which brings me to another point...
> > In your PR, you mention "race problems". Could you please explain
>
> Allready done.

Thanks.

> > what do you mean by this. If there is a problem that would be
> > introduced with starting xdm through rc script, then that is another
> > reason not to abandon the long time documented way of doing things,
> > which works quite well. Note that this only goes for xdm. If there is
>
> I still have some FreeBSD-2.x machines running, but I prefer install
> FreeBSD-5.  I think there're really good reasons to _permit_, not
> _force_ the use of a rcNG script.

And it is permitted, nothing prevents user to create a rc script and run xdm 
from that. There are many things that can be done in a way that is neither 
recommended nor documented, but I believe that the way that presently is 
documented works well and don't see a reason to complicate it.

> > any script eventually included in either X11 -clients, then that
> > script cannot take responsibility of starting things which come from
> > other ports such as gdm or kdm (BTW, you forgot wdm). Kdm and gdm
>
> Point me to the port and I'll take a look on this.

It's conveniently named wdm and resides in x11/wdm.

> > have their own maintainer teams who know how to start their programs
> > properly and intruding into their territory with this script seems
> > neither appropriate nor wise (consider that the way in which wdm
> > starts changes in one version - how intuitive would it be to expect
> > people to upgrade xorg-clients to get wdm working, not to mention why
> > would people who don't use wdm or any foodm for that matter need to
> > upgrade their ports). In short: stuffing startup of all display
> > managers into one script would seem a bad idea.
>
> rcNG have enough resorces to cope with this without need of futher
> scripts cahnges.

I apologise, I don't understand what you mean by this or how it relates to the 
concern I've written.

> > So I haven't been convinced so far that making rc script for purpose
> > of starting a kind of getty would be either needed or something that
> > would simplify things. I don't believe that rc.conf is the only file
> > users edit after their FreeBSD installation. I also don't believe
> > that it would be good if things were changed so that rc.conf would be
> > the only file users would need to edit. There was mentioned a
> > question of policy vs. features among the thread discussing this... I
> > don't believe a policy is dictated by having an example of starting
> > xdm in /etc/ttys file - users can still make their own rc script if
> > they want and configure it as it fits their purpose, be it "old
> > style" rc script or rcNG one.
> >
> >
> > Dejan
>
> well, I'll close the PRs.
>
> thanks for your time,

Oh, no problem about time and I'm also not necessarily right. I just wanted to 
voice my concerns with introducing this scripts and if I've managed to sound 
to negative I do apologise. But I will stress once again that such a change 
would in my opinion (and again this is my opinion if there are many people 
that think otherwise, I'm sure they will followup) need serious benefits to 
be done.
The script for starting xfs that you also suggested was on the other hand a 
great idea and it also doesn't affect things in the way xdm one would so 
thank you for your time and suggestions.


Dejan