nanogui: Thread: ELKS problems

Al,
	I worked on getting up to date this weekend on the linux-86 devkit (v0.14.8)
and ELKS (v0.77).  After quite a few compile time problems, I got everything
compiling and working.  I'll detail the problems later, and submit a patch.

	I can't get Micro-Windows to work with the serial mouse driver.  It appears
there's still possibly a bug in select().  Basically, the mouse works for about 1 second
and then freezes.  This is only the case if I move the mouse rapidly on startup.  I
modified mou_ser.c to use /dev/ttys0 and mousetype "pc", as I use a logitech mouse,
but this shouldn't affect anything.  If mou_ser.c is compiled -DTEST=1, it will
run as a standalone app, and everything works fine. (it prints the x, y, buttonstatus
to the console.)  If I run Microwindows and then move the mouse, nothing happens
after 1 second, and then when I hit ESC to quite Microwindows, and run the mou_ser
test program, it spits out 50 or so mouse moves that appeared to be queued up in
the kernel??  It appears the serial driver queues stuff inbetween close/open's.
Have you seen any of this activity?


	Also, when trying to set environment variables, what shell do you use?
I can't get sash to set environment variables, and when I run sh, it chops the PATH
and other env vars short, as though the env space is too small...

	Finally, I don't yet understand how you build images.zip.  There are
root, boot and  comb images, but the Makefiles in elkscmd and elks appear only to make
the following:

	elks "make disk" makes boot, which seems to be called root in distribution.
	elkscmd "make floppy" makes root, which seems to be called comb...
Isn't there a standard makefile function to make these images?

	On my system (Caldera 2.2) the dev86 package won't make.  It
requires an old /usr/bin/ar (this core dumps), and a change in elksemu to include
<sys/time.h>.  I will send a patch for this.

	In addition, the elkscmd requires changes in byacc/Makefile
and ash/mknode.c, and uses a redefinition of utmp stuff in utent.c.  I'll send
a patch on this as well.

Greg

On Mon, 28 Jun 1999, Greg Haerr wrote:

> 	I can't get Micro-Windows to work with the serial mouse driver.
> It appears there's still possibly a bug in select().  Basically, the
> mouse works for about 1 second and then freezes.

As I read this, the first thing that pops into my head is that you're not
re-setting the struct timeval.  Try setting up strace in the kernel, and
see what happens when you run it.

> It appears the serial driver queues stuff inbetween close/open's.  Have
> you seen any of this activity?

My guess here (maybe) is a flow-control issue.  You might find the
buffer/queue is actually in the mouse, and as soon as you open the device,
the flow-control lets the data come in.

Davey

On Monday, June 28, 1999 1:37 PM, David Murn ####@####.#### wrote:
: On Mon, 28 Jun 1999, Greg Haerr wrote:
: 
: > 	I can't get Micro-Windows to work with the serial mouse driver.
: > It appears there's still possibly a bug in select().  Basically, the
: > mouse works for about 1 second and then freezes.
: 
: As I read this, the first thing that pops into my head is that you're not
: re-setting the struct timeval.  Try setting up strace in the kernel, and
: see what happens when you run it.
: 
	OK.  Al, does Micro-windows work for you?  How about anyone
else?

Greg

On Monday, June 28, 1999 1:37 PM, David Murn ####@####.#### wrote:
: On Mon, 28 Jun 1999, Greg Haerr wrote:
: 
: > 	I can't get Micro-Windows to work with the serial mouse driver.
: > It appears there's still possibly a bug in select().  Basically, the
: > mouse works for about 1 second and then freezes.
: 
: As I read this, the first thing that pops into my head is that you're not
: re-setting the struct timeval.  Try setting up strace in the kernel, and
: see what happens when you run it.
: 
: > It appears the serial driver queues stuff inbetween close/open's.  Have
: > you seen any of this activity?
: 
: My guess here (maybe) is a flow-control issue.  You might find the
: buffer/queue is actually in the mouse, and as soon as you open the device,
: the flow-control lets the data come in.
: 
: Davey
: 
	Yes, but Micro-Windows opens the mouse, and uses the same driver.
That's why I think there's still problems with select()

On Mon, 28 Jun 1999, Greg Haerr wrote:

> On Monday, June 28, 1999 1:37 PM, David Murn ####@####.#### wrote:
> : On Mon, 28 Jun 1999, Greg Haerr wrote:
> : 
> : > 	I can't get Micro-Windows to work with the serial mouse driver.
> : > It appears there's still possibly a bug in select().  Basically, the
> : > mouse works for about 1 second and then freezes.
> : 
> : As I read this, the first thing that pops into my head is that you're not
> : re-setting the struct timeval.  Try setting up strace in the kernel, and
> : see what happens when you run it.
> : 
> 	OK.  Al, does Micro-windows work for you?  How about anyone
> else?

It works fine for me, but the mouse doesn't work on a pentium. Still don't
know why.

Luke(Boo) Farrar.

Greg Haerr writes:
> 
> Al,
> 	I worked on getting up to date this weekend on the linux-86 devkit (v0.14.8)
> and ELKS (v0.77).  After quite a few compile time problems, I got everything
> compiling and working.  I'll detail the problems later, and submit a patch.
> 
> 	I can't get Micro-Windows to work with the serial mouse driver.  It appears
> there's still possibly a bug in select().  Basically, the mouse works for about 1 second
> and then freezes.  This is only the case if I move the mouse rapidly on startup.  I
> modified mou_ser.c to use /dev/ttys0 and mousetype "pc", as I use a logitech mouse,
> but this shouldn't affect anything.  If mou_ser.c is compiled -DTEST=1, it will
> run as a standalone app, and everything works fine. (it prints the x, y, buttonstatus
> to the console.)  If I run Microwindows and then move the mouse, nothing happens
> after 1 second, and then when I hit ESC to quite Microwindows, and run the mou_ser
> test program, it spits out 50 or so mouse moves that appeared to be queued up in
> the kernel??  It appears the serial driver queues stuff inbetween close/open's.
> Have you seen any of this activity?

I have seen something like this, but it has never frozen on me. In nano-X,
whenever I stop moving the mouse, the cursor stops moving (as expected),
but when I start moving it in the oposite direction, it moves a few pixels
in the same direction as before, then changes to move in the new direction
as if some bytes got stuck in the queue. I suspect this is a glitch in the
select code. I will dig into it.

> 
> 
> 	Also, when trying to set environment variables, what shell do you use?
> I can't get sash to set environment variables, and when I run sh, it chops the PATH
> and other env vars short, as though the env space is too small...

I have seen problems with env vars getting chopped, but was not able to
track down where it happened. Almost certainly a bug involving strlen()
somewhere in libc, or the shell. sash is far too primitive to be used as a
real shell. I include it still because it is so small and stable that it is
a good test platform for running binaries.

> 
> 	Finally, I don't yet understand how you build images.zip.  There are
> root, boot and  comb images, but the Makefiles in elkscmd and elks appear only to make
> the following:

Making the contents of images.zip is not fully automatic, but most of the
work is done by the Makefiles. I change the Makefile a bit between buidlign
comb and root (modify size etc.)

> 
> 	elks "make disk" makes boot, which seems to be called root in distribution.

Checking with my copy, boot in images.zip is definatly a boot kernel image,
and root is a root disk image.

> 	elkscmd "make floppy" makes root, which seems to be called comb...
> Isn't there a standard makefile function to make these images?

Make floppy attempts to build an image of a combined disk, but calls it
root.

Make min_rfs makes a small root disk without kernel boot loader on it.

> 
> 	On my system (Caldera 2.2) the dev86 package won't make.  It
> requires an old /usr/bin/ar (this core dumps), and a change in elksemu to include
> <sys/time.h>.  I will send a patch for this.
> 
> 	In addition, the elkscmd requires changes in byacc/Makefile
> and ash/mknode.c, and uses a redefinition of utmp stuff in utent.c.  I'll send
> a patch on this as well.
> 

From what you are saying I gather that you have not been able to get in
touch with Reb de Bath about the changes to Dev-86? I looks as though we
will have to put together an new patched version. The utent.c in elkscmd
fixes some major bugs in the dev86 version. I included it until the patch
was merged into dev86, but this has not happened.

Al

Greg Haerr writes:
> 
> On Monday, June 28, 1999 1:37 PM, David Murn ####@####.#### wrote:
> : On Mon, 28 Jun 1999, Greg Haerr wrote:
> : 
> : > 	I can't get Micro-Windows to work with the serial mouse driver.
> : > It appears there's still possibly a bug in select().  Basically, the
> : > mouse works for about 1 second and then freezes.
> : 
> : As I read this, the first thing that pops into my head is that you're not
> : re-setting the struct timeval.  Try setting up strace in the kernel, and
> : see what happens when you run it.
> : 
> 	OK.  Al, does Micro-windows work for you?  How about anyone
> else?
> 

It seems to work fine for me in terms of drawing, mouse movement etc. I was
not able to move the windows around however.

Al

: Making the contents of images.zip is not fully automatic, but most of the
: work is done by the Makefiles. I change the Makefile a bit between buidlign
: comb and root (modify size etc.)
: 
	I would *really* like a makefile that automatically makes your
distribution.  (its a quality thing...)



: > 
: > 	elks "make disk" makes boot, which seems to be called root in distribution.
: 
: Checking with my copy, boot in images.zip is definatly a boot kernel image,
: and root is a root disk image.

	This should be called "make root".  Then "make rootdisk" makes a root
floppy.  There should also be a "make boot"

: 
: > 	elkscmd "make floppy" makes root, which seems to be called comb...
: > Isn't there a standard makefile function to make these images?
: 
: Make floppy attempts to build an image of a combined disk, but calls it
: root.

	This should be changed to "make comb".  I also had to comment back
in the dd for the boot load 512 byte image or it wouldn't boot.

: 
: Make min_rfs makes a small root disk without kernel boot loader on it.
: 
: > 
: > 	On my system (Caldera 2.2) the dev86 package won't make.  It
: > requires an old /usr/bin/ar (this core dumps), and a change in elksemu to include
: > <sys/time.h>.  I will send a patch for this.

	Do you want the patch or shall I send it to Rob?


: > 
: > 	In addition, the elkscmd requires changes in byacc/Makefile
: > and ash/mknode.c, and uses a redefinition of utmp stuff in utent.c.  I'll send
: > a patch on this as well.
: > 
: 
: >From what you are saying I gather that you have not been able to get in
: touch with Reb de Bath about the changes to Dev-86? I looks as though we
: will have to put together an new patched version. The utent.c in elkscmd
: fixes some major bugs in the dev86 version. I included it until the patch
: was merged into dev86, but this has not happened.

	It wouldn't hurt for us to maintain working patches, since we're starting
to get some people trying to use/compile ELKS, nano-X, etc, and I don't
want people to have stupid problems...  We need all the help and input we can
get.  Everyone should be able to exactly duplicate the binary distribution
from scratch.

Greg

: It seems to work fine for me in terms of drawing, mouse movement etc. I was
: not able to move the windows around however.
: 

	Geez, that's half the fun of Micro-Windows!  Mouse down on the caption
of a window and drag the mouse, the window moves.  Mouse down on the non-caption
area, and the window jumps to top z-order.  

	What is your mouse-experience with it?

: I have seen something like this, but it has never frozen on me. In nano-X,
: whenever I stop moving the mouse, the cursor stops moving (as expected),
: but when I start moving it in the oposite direction, it moves a few pixels
: in the same direction as before, then changes to move in the new direction
: as if some bytes got stuck in the queue. I suspect this is a glitch in the
: select code. I will dig into it.
: 
	It seems that a read of the mouse fd is never attempted.
After this is totally debugged, the serial driver needs to be changed as well.
The driver open() should discard all chars in the queue.  Likewise, the system
shouldnt use memory to store chars after a driver close().