Re: Report on X

[Prabhu Ramachandran <prabhu@aero.iitm.ernet.in>]

hi,

>>>>> "Steve" == Steve Bowman <sbowman@frostwork.net> writes:

<snip>
    Steve> 	login as sbowman ps works su works startx fails su now
    Steve> fails ("Killed") ps now fails ("Killed")

    Steve> and now many other things seems to be failing, like ping.
    Steve> ls is one thing that still works.  Logging out and back in
    Steve> again seems to reset the situation, a reboot isn't
    Steve> required.

    Steve> The output of startx is: _X11TransSocketUNIXConnect: Can't
    Steve> connect: errno = 1073741826 which repeats until
    Steve> interrupted.

    Steve> I just upgraded to gnumach 1.2-6 and this behaviour still
    Steve> occurs.  Is anyone else seeing this kind of instability?

I see _exactly_ the same behaviour.  With the exact same message (the
errno is the same too.)  I am running a rebuilt gnumach1.2-5 kernel
with the fd0 patch.  I tried suiding the XF86_SVGA binary but no go.
I also gave everyone read/write access to the console.  


There are some other problems I'd like to report.

(1) I installed mailx and tried mail.  Here is what approxly happened.

$ mail
N <Some new test message>...
> q
unable to lock mailbox: permission denied.
$

The mailx program has the right perms (owned by root and group mail)
it is also suid'd.  The file /var/spool/mail/prabhu was had owner
prabhu and group mail.  I have no idea what is wrong.

(2) I tried to make a few debs from source.  Here is what happened.

	(a) I wanted to compile numeric python and started to compile
	the python-llnl package (which originally was the source).  I
	had installed the relevant python packages.  I found that,
	the python-dev package doesnt ship with all the header
	files.   My linux install has 69 files in
	/usr/include/python1.5/	 but hurd had 63.  I simply copied
	the missing ones from linux and it worked.  I got stuck up
	later on another problem but realized that python-numeric had
	a separate source and dropped compiling python-llnl.

	(b) While compiling python-numeric.  I found that a lot of
	important modules were not present in hurd.  glob and shutil
	come to mind.

	$ ls -la /gnu/usr/lib/python1.5/ | wc -l  
	190
	$ ls -la /usr/lib/python1.5/ | wc -l
	424
	$

	As you can see the python package has some serious problem
	with the modules.  The python interpreter does work though.  I
	will look into this problem and recompile python for the hurd
	this weekend.  In any case, I copied the relevant modules from
	my linux machine and I compiled python-numeric.  It installs
	fine.  I have to check it.

	(c) I ran X as root with blackbox as the wm and decided to try
	using X for a while.  I tried to compile mesa.  I did this
	under X.  It was crazy.  I had 4 xterms running and a remote
	login.  I did dpkg-buildpackage -B -uc inside the mesa source
	tree.  At first things were pretty ok.  But soon, hurd started
	swapping like crazy.  I let it swap for almost 2 hours.  The
	machine was practically unusable.  There werent too many
	processes (as far as a ps uax was concerned) but it was
	swapping like hell.  It finally got to the point that I was
	scared to burning my hard disk and I tried to kill the
	processes but to no avail. After I killed a lot of the
	packages, hurd froze and I hit the reset.  Mesa still hadnt
	finished building.  I then went back to the console and redid
	the dpkg-buildpackage this time round it got much farther but
	stopped a some error that I wasnt in a state to debug - it was
	too early in the morning.

	Basically, X has some serious problem.  I have no idea what.
	Maybe there are too many threads (I remember that Marcus
	mentioned that X uses up some 2000 of them).  How does one
	check the number of threads that are running??  Under linux
	pthreads registers the thread as a process but hurd is
	different I guess, cos I didnt see more than 200 processes.

	(d) The tcl and tk packages dont ship with the respective
	libraries. Under linux they do.  For instance the package
	tk8.0 under linux contains the file /usr/lib/libtk8.0.so.1 but
	there is no such file in the hurd package.  Similar is the
	case for the tcl8.0 package.  Because of this python-tk doesnt
	work.  I have downloaded the soruces and will see if a rebuild
	fixes this.


That is pretty much what happened for most of last night. :)  I am
sorry that this message is extremely verbose but I thought that it
might be useful.

	On the happy side, exim seems to work fine for me.  I sent a
test message from under hurd to another account and it worked (I used
the same config options as i used under linux).  I will try to send a
message to the list later from the hurd.


thanks,
prabhu