Question

So I have three installers for NVIDIA’s CUDA API — the first is a driver and comes with nice silent install flag options (but you have to be root and have to have run level 3).

The second two follow are shown manually installing below (cut out the long mess of install afterwards for brevity)

[root]# sh cudatoolkit_4.1.28_linux_64_rhel5.x.run Verifying archive
integrity… All good. Uncompressing NVIDIA
CUDA……………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………
Enter install path (default /usr/local/cuda, ‘/cuda’ will be
appended):
A previous version of CUDA was found in /usr/local/cuda/bin
Would you like to uninstall? (yes/no/abort): yes

In other words, I need to recognize:
“Enter install path” and output a ‘\n’

Now the tricky part is the uninstall may not be always be there. If it’s not I need to simply wait for the install to finish, but if I see “Would you like to uninstall?” I need to output “yes” to complete.

The third and final installer’s output is shown below….

[root]# sh gpucomputingsdk_4.1.28_linux.run
Verifying archive
integrity… All good.
Uncompressing NVIDIA GPU Computing

SDK………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………

Enter install path (default ~/NVIDIA_GPU_Computing_SDK):
/usr/local/CUDA_SDK

Located CUDA at /usr/local/cuda
If
this is correct, choose the default below.
If it is not correct,
enter the correct path to CUDA

Enter CUDA install path
(default /usr/local/cuda):

For this one, there is no uninstall action so it’s seemingly a bit simpler.

I just need to detect “Enter install path” and output “/usr/local/CUDA_SDK\n” and then detect “Enter CUDA install path” and output “\n”

My idea was to use a pair of expect scripts — one for each installer — but due to the nesting within the double quotes of the command to switch to root, I’m having some difficulties with this. What I currently have is:

#!/bin/bash
CR="\"\n\""
YES="\"Yes\""
INSTALL_PATH_REQUEST="\"Enter install path\""
CUDA_PATH_REQUEST="\"Enter CUDA install path\""
UNINSTALL_REQUEST="\"Would you like to uninstall?\""
TOOLKIT=`ls -t cudatoolkit* | head -n 1`
TOOLKIT_EXPECT="sh $TOOLKIT"
SDK=`ls -t gpucomputingsdk* | head -n 1`
SDK_INSTALL_PATH="\"/usr/local/CUDA_SDK\n\""
SDK_EXPECT="sh $SDK"
/bin/su root -c "yum -q -y install expect expectk;
/sbin/init 3; sh `ls -t NVIDIA*|head -n 1` -s --update -a -X;
/usr/bin/expect <<EOF;
spawn $TOOLKIT_EXPECT
expect $INSTALL_PATH_REQUEST
send $CR
expect $UNINSTALL_REQUEST
send $YES
EOF
/usr/bin/expect <<EOF;
spawn $SDK_EXPECT
expect $INSTALL_PATH_REQUEST
send $SDK_INSTALL_PATH
expect $CUDA_PATH_REQUEST
send $CR
EOF
/sbin/init 5"

This switches to root properly (once the password is entered) and installs the driver with the built in options correctly. It then appears to spawn the second install process and enter the first argument (a carriage return), but seems to exit the second installer prematurely (e.g. I don’t see the “yes” option.).

I feel like I’m pretty close, hopefully somebody can point me to where I’m going wrong and suggest the correct syntax.

NOTES:
I added the yum install command, as some of the machines I’m installing on didn’t have expect (stock CentOS 6), so that saves me the trouble there….

Answer 1

Might be an issue with timeout here… not sure how long the installer takes.
The default expect timeout is 10 seconds, if it doesn’t see the expected text in that time, it will proceed regardless, you could change the timeout values like so:

expect -timeout 100 $INSTALL_PATH_REQUEST

Also change your $YES from

YES="\"Yes\""

To:

YES="\"Yes\r\""

(Best to use \r instead of \n in $CR too)

It’s also a good idea to expect some ‘safety string’ at the end of the install, for example:

 expect -timeout 320 "Install Complete."

So the expect script doesn’t terminate before the spawned process is complete.