[ewg] install.pl feedback

[Jeff Squyres]

Here's my feedback in no particular order...

1. If you select option 3 (Show Installed Software), there seems to  
be some extra spaces between some package lines, and then when it's  
done, the screen is cleared and you can't see the list of packages  
that was displayed.

1a. It would be nice to be able to send the list of packages to a  
file (for later scripting purposes).

2. Option 4, "Configure IPoIB" doesn't seem to do anything -- it just  
goes back to the main menu.

3. Installing the software: hitting ctrl-C at any of the prompts  
seems to assume that you want the default value for that prompt.  It  
does not quit the installer nor give you the opportunity to ask "do  
you want to abort the current action?".

4. Hitting ctrl-Z at several (any?) prompt simply displays the prompt  
again, rather than suspending the installer.  I think that this  
violates the Law of Least Astonishment (LLA).

5. Hitting invalid answers at installer prompts just displays the  
prompt again -- it does not show you the valid options again.  What  
happens if the list of valid options scrolls off the screen?

6. I'm not sure if I like the "hit a single key without hitting  
enter" motif.  I think that many users expect to hit enter.  Indeed,  
before I realized that I didn't need to hit enter, I was hitting  
enter and then getting the default value of the *next* prompt and  
wondering why the installer [seemingly] blew by it without lettering  
me input an answer.

7. There is no progress indicator when you uninstall OFED.  Nor is  
there any indication of what it did, or even if it succeeded or  
failed.  When I did "uninstall", I eventually just got dumped back to  
a shell prompt.  Does that mean success?  In general, there is a lack  
of confirmation throughout the installer when an action was performed  
(e.g., whether it was successful or not).  For example:

OFED Distribution Software Installation Menu

    1) View OFED Installation Guide
    2) Install OFED Software
    3) Show Installed Software
    4) Configure IPoIB
    5) Uninstall OFED Software

    Q) Exit

Select Option [1-5]:5
Uninstalling the previous version of OFED
---> installer sat here for a while
[root at svbu-mpi010 OFED-1.3-20070823-1130]#

8. When I run the installer as non-root, a bunch of error messages  
flash by too quickly to read before the screen clears and the menu  
appears.  Upon further thought -- I don't know if these error  
messages are because I'm not root or because I [successfully?]  
uninstalled OFED before running the installer again.

8a. If we're going to be building/installing RPMs which requires root  
privlidges, and we're going to allow running the installer as non- 
root, then certain options that we *know* will require root access  
should be disallowed (whatever the ASCII-equivalent of "greyed out"  
is) rather than trying to run them and running into obscure  
failures.  E.g., the following is lame (note that I'm running as a  
non-root user):

OFED Distribution Software Installation Menu

    1) View OFED Installation Guide
    2) Install OFED Software
    3) Show Installed Software
    4) Configure IPoIB
    5) Uninstall OFED Software

    Q) Exit

Select Option [1-5]:2
Can't open /tmp/OFED-1.3-20070823-1130/ofed.conf: Permission denied  
at /tmp/OFED-1.3-20070823-1130/install.pl line 1574.
[9:41] svbu-mpi010:~ %

9. You are no longer given the option of which MPIs to install -- you  
have to install them all.  That seems bad.

10. The installer jumps right into the MVAPICH2 installation/ 
configuration questions but there is no specification that these  
questions *only* pertain to MVAPICH2 (not Open MPI and not MVAPICH).

11. Even when there is no OFED installed, the installer shows  
"Uninstalling the previous version of OFED" before building the OFA  
kernel code.

12. The installer failed for me on RHEL4U4:

Build ofa_kernel RPM
Running rpmbuild --rebuild --define '_topdir /var/tmp/OFED_topdir' -- 
define 'configure_options  --kernel-version=2.6.9-42.ELsmp --kernel- 
sources=/lib/modules/2.6.9-42.ELsmp/build  --with-core-mod --with- 
user_mad-mod --with-user_access-mod --with-addr_trans-mod --with- 
mthca-mod --with-mlx4-mod --with-cxgb3-mod --with-ipath_inf-mod -- 
with-ipoib-mod --with-sdp-mod --with-srp-mod --with-rds-mod --with- 
vnic-mod --with-iser-mod' --define 'build_kernel_ib 1' --define  
'build_kernel_ib_devel 1' /tmp/OFED-1.3-20070823-1130/SRPMS/ 
ofa_kernel-1.3-ofed20070823.src.rpm
Install kernel-ib RPM:
Running rpm -iv /tmp/OFED-1.3-20070823-1130/RPMS/redhat- 
release-4AS-5.5/kernel-ib-1.3-2.6.9_42.ELsmp.x86_64.rpm
Install kernel-ib-devel RPM:
Running rpm -iv /tmp/OFED-1.3-20070823-1130/RPMS/redhat- 
release-4AS-5.5/kernel-ib-devel-1.3-2.6.9_42.ELsmp.x86_64.rpm
Build ib-bonding RPM
Running rpmbuild --rebuild --define '_topdir /var/tmp/OFED_topdir' -- 
define 'KVERSION 2.6.9-42.ELsmp' --define '_release 2.6.9_42.ELsmp' -- 
define '_prefix /usr' /tmp/OFED-1.3-20070823-1130/SRPMS/ib- 
bonding-0.9.0-12.src.rpm
Install ib-bonding RPM:
Running rpm -iv /tmp/OFED-1.3-20070823-1130/RPMS/redhat- 
release-4AS-5.5/ib-bonding-0.9.0-2.6.9_42.ELsmp.x86_64.rpm
Build ib-bonding RPM
Running rpmbuild --rebuild --define '_topdir /var/tmp/OFED_topdir' -- 
define 'KVERSION 2.6.9-42.ELsmp' --define '_release 2.6.9_42.ELsmp' -- 
define '_prefix /usr' /tmp/OFED-1.3-20070823-1130/SRPMS/ib- 
bonding-0.9.0-12.src.rpm
ib-bonding-debuginfo was not created
[root at svbu-mpi010 OFED-1.3-20070823-1130]#

I don't know what the issue was about why it failed (note the lack of  
any explanation or pointers to logfiles where more information could  
be gleaned about why it failed / how to fix the problem).  So I  
didn't test any further than this.  :-)

-- 
Jeff Squyres
Cisco Systems