Hi Todd,
On 10/02/14 23:13, Heywood, Todd wrote:
It works for me with modules 3.2.10 (but not 3.2.9).
That's very weird, I didn't expect that...
In the mean time, I've figured out what the deal is with the single quotes.
After discussing this on the Lmod mailing list, things became clear to me.
Below is a copy-paste of my explanation on the Lmod mailing list.
The single quotes approach works but is actually wrong, and there's a
much cleaner way (i.e. not using is-loaded guards around dependency
"module load"s).
Here's what's going on. The two following expression in Tcl are equivalent:
is-loaded GCC/4.7.2
is-loaded "GCC/4.7.2"
In the second case, the quotes are stripped off. However, this does not
happen with single quotes:
is-loaded 'GCC/4.7.2'
The consequence of this is that this always evaluates to False, whether
the module is loaded or not (which is obviously wrong), since
"is-loaded" is checking whether a module with the name " 'GCC/4.7.2' "
exists (so the single quotes are effectively part of name of the module
being searched for).
This also explains the 'recursive unloading' behavior we're suddenly
getting.
Let me use the below as example, with line numbers:
1 if { ![is-loaded 'GCC/4.7.2'] } {
2 module load GCC/4.7.2
3 }
Since the condition used in line 1 will always evaluate to "not False"
a.k.a. "True", the body of the if clause will always be used, no matter
whether the GCC/4.7.2 module is loaded or not.
On "module load", this will trigger loading GCC/4.7.2 as a dependency;
no surprises there (except for always loading the module).
On "module unload", line 2 will be translated to "module unload
GCC/4.7.2", since on unload all statements in the module file are
basically translated to their complement.
WIthout single-quotes, the "is-loaded GCC/4.7.2" will be True on "module
unload" (because the GCC/4.7.2 will be loaded), so the condition in the
if clause will be "not True" a.k.a. "False", so the body of the if
clause is skipped.
Hence, it won't be translated to "module unload", hence no recursive
unloading.
So, basically using single-quote breaks things. Although the recursive
unloading is a nice side-effect, we're effectively making the
'is-loaded' check useless since it will always return False.
Simply dropping the "if is-loaded" guard around the "module load
<dependency>" will also have the effect of "recursive unloading", but
will also always load dependency modules, whether they were loaded
already or not.
I've looked into reloading modules that are already loaded, and
certainly with environment modules 3.2.10 this is harmless (the
statements in the module file are not re-executed). Lmod takes a
different approach (it unloads and reloads the module), so things may be
trickier there, but it should still work.
regards,
Kenneth
Thanks for the help!
Todd
From: Kenneth Hoste <[email protected]<mailto:[email protected]>>
Reply-To: "[email protected]<mailto:[email protected]>"
<[email protected]<mailto:[email protected]>>
Date: Monday, February 10, 2014 at 11:26 AM
To: "[email protected]<mailto:[email protected]>"
<[email protected]<mailto:[email protected]>>
Subject: Re: [easybuild] module unload for toolchains?
Hi Todd,
On 10/02/14 17:21, Heywood, Todd wrote:
Hi Kenneth,
Thanks. I am actually using an older release of modules (from our default/base
installation):
Modules Release 3.2.9 2011-11-24 (Copyright GNU GPL v2 1991):
That should do it, I think. I only have 3.2.10 around to play around
with though...
Can you take a look at the example I send right after my mail below, and
maybe try with v3.2.10 to see if that makes any difference for you?
regards,
Kenneth
Todd
From: Kenneth Hoste
<[email protected]<mailto:[email protected]><mailto:[email protected]>>
Reply-To:
"[email protected]<mailto:[email protected]><mailto:[email protected]>"
<[email protected]<mailto:[email protected]><mailto:[email protected]>>
Date: Monday, February 10, 2014 at 11:14 AM
To: "[email protected]<mailto:[email protected]><mailto:[email protected]>"
<[email protected]<mailto:[email protected]><mailto:[email protected]>>
Subject: Re: [easybuild] module unload for toolchains?
Hi Todd,
On 10/02/14 16:56, Heywood, Todd wrote:
Hi Pablo,
I only just got to this now. I tried your change (line 157 of
module_generator.py), and I also tried this:
"if { ![is-loaded %(mod_name)s] } {",
" module load %(mod_name)s",
“} else {",
" module unload %(mod_name)s",
“}
"",
I regenerated modulefiles via eb. Unfortunately, the first try had no effect
for unloading, and the second resulted in not being able to unload or purge any
modules.
I’m probably missing something fundamental.
I'm currently looking into fleshing out Pablo's PR a bit more to add a
configure option for the wrapping of module names in the is-loaded
conditional, which triggers 'recursive unloading'.
Are you saying that wrapping the module name in single quotes as shown
below does not trigger recursive unloading at your end?
if { ![is-loaded %(mod_name)s]} { # NO recursive unloading
vs
if { ![is-loaded '%(mod_name)s']} { # WITH recursive unloading
Which modules tool are you using (environment modules, Lmod, ...), and
which version?
Also: I agree with your amazement that simply wrapping the module name
in single quotes (note: double quotes don't work >_<) has such a 'major'
effect...
Has anyone been able to find a reference for that? Is this documented
somewhere?
regards,
Kenneth
