R devel - Mar 2006 - R started in terminal shell script or ESS steps on LD_LIBRARY_PATH?

Hi all,

Just noted this behavior in the past couple of days, where if R is
started in a shell script such as:

  gnome-terminal [-e][-x] R

or in ESS (version 5.2.12 with Emacs or XEmacs), the LD_LIBRARY_PATH
environment variable is not properly appended to, resulting in the loss
of the pre-start value.

This is using R Version 2.2.1 Patched (2006-02-28 r37448) on FC4.

I noted this when attempting to access an Oracle 10g server using RODBC
(version 1.1-5) and to the best of my recollection, this is a new
problem.

It took me a while to figure this out and was about to send an e-mail to
r-sig-db since I was having trouble connecting using odbcConnect(), when
I noted the problem with LD_LIBRARY_PATH. I noted this quite by accident
as I was writing the e-mail to r-sig-db.

I was on the verge of mental vapor lock with this, so if I missed a
documented change in behavior, my apologies. In looking at the R startup
script, it seems appropriate in terms of appending to the existing
value. So it seems that this is being altered elsewhere presumably.



In the two cases I note above, I get:
> Sys.getenv("LD_LIBRARY_PATH")
                                     LD_LIBRARY_PATH 
"/usr/local/lib/R/lib:/usr/local/lib:/usr/X11R6/lib"


whereas the pre-start value is:

  LD_LIBRARY_PATH=/usr/lib/oracle/10.2.0.1/client/lib



Importantly, if I start a gnome-terminal session and then start R from
the command line, I get:
> Sys.getenv("LD_LIBRARY_PATH")
LD_LIBRARY_PATH
"/usr/local/lib/R/lib:/usr/local/lib:/usr/X11R6/lib:/usr/lib/oracle/10.2.0.1/client/lib"


In this case, I can use odbcConnect() to the Oracle server and it works
the first time, every time.

This suggests to me some type of problem in the inheritance of the
pre-session environment in the two former cases, but again, I may be
missing something.


Even if I try:

LD_LIBRARY_PATH <- Sys.getenv("LD_LIBRARY_PATH")

Sys.putenv(LD_LIBRARY_PATH = paste(LD_LIBRARY_PATH,
           "/usr/lib/oracle/10.2.0.1/client/lib", sep =
":"))

and then:
> Sys.getenv("LD_LIBRARY_PATH")
LD_LIBRARY_PATH 
"/usr/local/lib/R/lib:/usr/local/lib:/usr/X11R6/lib:/usr/lib/oracle/10.2.0.1/client/lib"


I still cannot get odbcConnect() to work. I still get the error that I
was getting in the two initial R startup situations:

Warning messages:
1: [RODBC] ERROR: state 01000, code 0, message [unixODBC][Driver
Manager]Can't open lib
'/usr/lib/oracle/10.2.0.1/client/lib/libsqora.so.10.1' :
libclntsh.so.10.1: cannot open shared object file: No such file or
directory 
2: ODBC connection failed in: odbcDriverConnect(st, case = case,
believeNRows = believeNRows)


The above error is what has been driving me nuts for the past two days,
since clearly these files are present and it works in R from the
gnome-terminal command line and when using the sql*plus instant client
directly.  The LD_LIBRARY_PATH issue is the only one that I have noted
that appears to be different.


Any ideas on this?  Did I miss a change someplace?

Thanks for any insights.

Marc Schwartz

Hi all,

In follow up to my prior post on this issue, I have found a workaround,
but have not yet clearly identified the etiology of the problem.
Whatever it is, it is presumably unique to my system, though if anyone
can replicate this on another FC4 system...  :-)

The workaround involves booting to init 3 rather than init 5 and
starting X manually from the console. I found this after going through
some of the steps described below regarding my X configuration. In this
way, LD_LIBRARY_PATH is preserved and RODBC works without issue in both
ESS and the gnome-terminal shell start up script.

Dirk was kind enough to send me an offlist e-mail yesterday in reply,
which sparked some thoughts as I was away from the computer for a few
hours yesterday afternoon and evening.

Dirk's e-mail logically queried on any issues with gnome-terminal and/or
the bash shell itself.

Since this problem was new (this had all worked previously), I checked
to see if there had been any recent updates to either gnome-terminal or
bash in the FC repos. There were none, although there have been of
course updates to GNOME, GTK and other libs.

This got me to think about other updates since the last known time this
process worked properly. So I spent several hours last night and this
morning reviewing possibilities.

The last Xorg updates are from last September, so these are not new.

Other changes that I had made in the recent past include:

1. Modifying my xorg.conf to support nVidia TwinView hardware
acceleration functionality. TwinView is like xinerama mode, spanning
both displays to give me a virtual 3200x1200 screen, though supporting
HW acceleration on both displays. Previously I had been using two X
servers (also using the nVidia driver in non-xinerama mode) to support
my dual display configuration.  Reverting back to the old configuration
did not resolve the problem. 

2. The last nVidia driver update (8178) was in December and this process
had worked since then.

3. There was an updated Cisco VPN client for Linux (4.8) to support
recent kernels. The VPN client is installed from source. This normally
starts up on boot as a service. Disabling the service, thus removing the
kernel module, did not resolve the problem.

4. I had updated the encryption of several of my partitions on my laptop
to use dm-crypt/LUKS with 256 bit AES from regular dm-crypt to take
advantage of pending LUKS support updates in HAL and other system
functions. Disabling the encryption (so the relevant kernel modules did
not load), logging in as root and running ESS from root's home did not
resolve the problem.

5. Just in case, I also reinstalled kernel version 2.6.15-1.1830
(running 1833 now) to see if there was any change there. No joy. I
cannot locate any of the 2.6.14 FC4 kernel versions, as these have been
removed from the repos, so it leaves open the possibility of something
in the .14 to .15 rebase change.


Other than routine system updates via yum, these are the only
"self-inflicted" changes that I have made recently. If any of the
above
should spark some thoughts, let me know.

My plans are to live with this for now. FC5 is targeted for release next
Wednesday, presuming that it stays on schedule. I'll do a clean install
with that and see if anything is resolved, perhaps indicating some other
issue that is as yet unidentified.

Many thanks to Dirk for your assistance.

Best regards,

Marc Schwartz

On Wed, 2006-03-08 at 13:54 -0600, Marc Schwartz wrote:
> On Wed, 2006-03-08 at 18:39 +0000, Hin-Tak Leung wrote:  
> > This sounds like a shell init issue... and you probably want to hunt
> > down where LD_LIBRARY path is *set*, rather than how it is inherited.
> > 
> > When you log in in run-level three, you get a login shell rather than
> > a normal interactive shell, and your startx inherits your
login-shell's
> > environment, You get a normal interactive shell(?) inside 
> > gnome-terminal/xterm if you start at run-level 5, and finally, you get
> > a non-interactive shell if you run a script, most of the time.
> > The environments in the three cases are all different, and
> > sometimes security related environment variables are not inherited
> > by forked sub-shells, such as LD_LIBRARY_PATH; or more likely, 
> > LD_LIBRARY_PATH is set up for the login shell, and other shells
> > simply don't get it.
> > 
> > HTL
> > 
> >  From the INVOCATION part of bash's man page - assuming that's
your
> > login shell, otherwise, others.
> > ==========> > When  bash is invoked as an interactive login
shell, or as a non-inter-
> > active shell with the --login option, it first reads and executes 
com-
> > mands  from  the file /etc/profile, if that file exists.  After
reading
> > that file, it looks for ~/.bash_profile, ~/.bash_login, and
~/.profile,
> > in  that order, and reads and executes commands from the first one
that
> > exists and is readable.  The --noprofile option may be  used  when 
the
> > shell is started to inhibit this behavior.
> > 
> > When  a  login  shell  exits, bash reads and executes commands from
the
> > file ~/.bash_logout, if it exists.
> > 
> > When an interactive shell that is not a login shell  is  started, 
bash
> > reads  and executes commands from ~/.bashrc, if that file exists. 
This
> > may be inhibited by using the --norc option.  The --rcfile file 
option
> > will  force  bash  to  read  and  execute commands from file instead
of
> > ~/.bashrc.
> > 
> > When bash is started non-interactively, to  run  a  shell  script, 
for
> > example, it looks for the variable BASH_ENV in the environment,
expands
> > its value if it appears there, and uses the expanded value as the 
name
> > of  a  file to read and execute.  Bash behaves as if the following
com-
> > mand were executed:
> >     if [ -n "$BASH_ENV" ]; then . "$BASH_ENV"; fi
> > but the value of the PATH variable is not used to search for  the 
file
> > name.
> > ========> 
<SNIP>


Many thanks for the reply.  Given the subject matter feel free to
respond offlist with any further replies. I can post back should I
figure this out for the sake of closure on the thread.

LD_LIBRARY_PATH is set in ~/.bashrc and this has worked fine previously,
so I am still unclear as to what has changed. Though I am readily
willing to accept that something has been screwed up somehow. Presuming
that a system wide setting has been compromised in some fashion, the
pending clean install of FC 5 may be helpful.

If not, I may need to consider something in my own user profile
configuration.

I also logged into a KDE session from init 5, to see if perhaps whatever
was going on might have been GNOME specific. Unfortunately, the same
behavior is seen in KDE using ESS.

Two more data points under init 5 in GNOME however:

1. If I open a gnome-terminal console and start R from the CLI, things
work. If I exit R and use 'gnome-terminal -x R' within that same console
to mimic my startup script, it does not work, even though the variable
is clearly set in the console prior to entering the command. However, if
from the same initial gnome-terminal console session, I use 'xterm -e
R', it works.

2. If I use the "Run Application..." dialogue from the GNOME menu,
type
in 'R' and check "Run in terminal", it does not work.

There is something subtle going on here, that I am just not seeing.

Thanks again for taking the time to reply.

Best regards,

Marc