I think this is a result of a condor bug.  I am pretty sure that condor cannot
handle a '$' in the argument 
directive.  Jaime can you confirm?

If you use griodine to submit, you can source /opt/globus/set-globus.sh to get
access to some fairly recent GT4 clients. 

As for Stu's comment, I would expect the weird quoting to happen in the
jobmanager Perl scripts - aren't they part of Globus? 

Also, running Mike's arguments through plain Condor, no Globus, using a vanilla
universe (as the jobmanager does) yields correct results. 

Thus, I strongly suspect the Globus jobmanager-condor perl module. 

ok - thanks.  we'll take a look.

The submit routine in condor.pm is explicitly escaping dollar-sign, backslash,
double-quote, and back-
tick. Double-quote is the only character you need to escape.

In reference to comment #4 from Jaime:

But condor.pm must be doing handling the escape of $ differently from the other
characters you list (I would claim "incorrectly") as in the following examples:

$ globus-job-run evitable/jobmanager-condor /bin/echo x \' x
x ' x
$ globus-job-run evitable/jobmanager-condor /bin/echo x $ x 
x \$ x
$ 

The "$" should be reaching the echo command with no extra "\" prepended, as the
"'" does. I suspect its putting in the \ expecting some other layer of
evaluation to be taking it off, but that is not happening.

Mike accidentally used ' which was not in Jaime's list. But backslash is in his
list:

$ echo x \\ x
x \ x

$ globus-job-run griodine.uchicago.edu:2120/jobmanager-condor /bin/echo x \\ x
x \\ x

As Jaime said, there are some extra backslashes going around. I have this
nagging feeling as if I may have caused this during my NFS fixes two years ago.
I'd like to check the CVS. 

There was no activity on this bug for a month now. Will it be fixed in the next
release? Do you want me to elevate it to a blocker, since the quote/escaping
strategy produces the wrong thing? Do you want me to make a first pass to
suggest fixes? 

I've already said how to fix it. Has anyone made the changes?

If not, we should notify Alain, so VDT can ship a fixed version at least for
the
condor.pm file. The VDT release is immanent, and I am loath to see "broken"
stuff on OSG.

Created an attachment (id=712) [details]
replacement condor.in before gpt-postinstall or make install

Since nobody is going to do anything about this bug, here is the patched
version of $GLOBUS_LOCATION/setup/globus/condor.in. It should be installed
before running either gpt-postinstall or make install. I also have a patched
version of the installed
$GLOBUS_LOCATION/lib/perl/Globus/GRAM/JobManager/condor.pm, if anybody needs
it.

[Note that the first backslash is always the local shell escape]

$ gjr gk/jobmanager-fork -l /bin/echo x \\ x \\\"x\\\" \'\$x\' \`x\` 
x \ x \"x\" '$x' `x`

$ gjr gk/jobmanager-condor -l /bin/echo x \\ x \\\"x\\\" \'\$x\' \`x\` 
x \ x "x" '$x' \`x\`

This fixes most. However, I am still not 100% happy about the way the dquotes
are handled. Condor cannot handle dquotes without backslashes in front of them
in the submit file, and as we can see, the remote Condor then removes the
backslashes in front of the dquotes. It appears as if the Condor jobmanager
should escape dquotes, which is contradictory to Jaime's statement. Jaime? 

Your fix escapes back-tick, but not double-quote. It should the other way
around.

Created an attachment (id=713) [details]
replacement with proper escaping

Thanks, Jaime.

$ /bin/echo x \\ x \\\"x\\\" \'\$x\' \`x\` 
x \ x \"x\" '$x' `x`

$ gjr gk/jobmanager-fork -l /bin/echo x \\ x \\\"x\\\" \'\$x\' \`x\` 
x \ x \"x\" '$x' `x`

$ gjr gk/jobmanager-condor -l /bin/echo x \\ x \\\"x\\\" \'\$x\' \`x\` 
x \ x \"x\" '$x' `x`

This is finally as it should be. 

Stu, When will GT4 include this patch?
Alain, will this patch make it into VDT?

Jens,

I don't understand part of your condor.in. When I do a diff, I see:

@@ -520,9 +517,8 @@
     my $job_id = $description->jobid();
     my $count = 0;
 
-    if ($job_id =~ m/(\d+\.\d+)\.\d+/) {
-        $job_id = $1;
-    }
+    $job_id =~ s/,/ /g;
+    $job_id =~ s/(\d+\.\d+)\.\d+/$1/g;

Why did you make this change? Is it important, or coincidental to your testing?

By the way, this will make it into VDT 1.3.8, which we hope to release "soon".
We'll leave the defintion of "soon" to be purposefully vague, but hopefully less
than a month. 

I've checked out last night from Globus's CVS where the fix of the quoting is
still not in, and it does have the weird substitutions you point out. I do
suspect that you have at your hand yet another bug, and frankly, I don't know
what the "right" fix is for it. My posting of the module uses the vanilla from
GT-CVS condor.pm. Maybe this is from some of the VDT patches to Globus? 

Looking at the module, though, I'd say that your version provides a slightly
"safer" jobid, if the original jobid returned from $description->jobid() is for
some reason longer or scrambled. However, if it is completely scrambled for some
reason (unfathomable, but not impossible), your code version will continue with
an invalid job id, as does the no-if-subst version. The "right" fix is probably
something else completely. 

You could ask Jaime's opinion on it. 

Jaime didn't know why that bit was needed, though it's not clearly wrong. 

I'll patch the VDT, but I'll just do that patch to fix the quoting. It will be in VDT 1.3.8, which we hope 
will be released within a few weeks. 

Sut, the branch check-out from the 24th still provides the wrong quoting. Can
you please assign this fix to someone? 

I committed changes based on the attached condor.in to both the trunk and
globus_4_0_branch.