WebSphere Jython scripting, add the script directory to the import path

If you’re running a script with wsadmin and it tries to import other modules that live in the same directory, you’ll discover another difference between Python and wsadmin. Python will always look for modules in the path that the script was run from. wsadmin won’t. This is kind of annoying if you have a few local import files. Sure, you can always add a directory to the import path via the command line but who ever remembers to do that (and who remembers the syntax)? Since we already have the script path (from fix5), we might as well add it to the import path.

if mainfile:
    mainpath = os.path.dirname(mainfile)
    if mainpath:
        sys.path[:0] = [ os.path.abspath(mainpath) ]

See fix5 for the computation of mainfile. Better yet, download the whole collection from here.

WebSphere Jython scripting, sys.argv[0] and __file__

Today’s problem is that wsadmin sets up sys.argv differently from normal Python (or Jython).  In Python, sys.argv[0] is the name of the script you invoked.  What you’d normally think of as the command line arguments start at index 1.  For whatever reason, wsadmin doesn’t pass the script name.  The arguments are passed in starting at sys.argv[0].  This can be a nuisance if you have a script that you want to run both in wsadmin and Python.  It’s also just one more difference to trip over.

I should give credit to this post for reminding me that this needs fixing and also for pointing out that the full command line is available in the environment.  Thanks!

My code for this is a little clumsy.  Basically, we scan the command line we get from IBM_JAVA_COMMAND_LINE for the -f option.  We  don’t want to pick up just any -f that might be passed as an argument to the script, easy to avoid by stripping off anything after a -- argument.  We also don’t want to pick up a -f that might appear among all the environment data that wsadmin prepends to the command line (unlikely as that might be), so we look for the last -f after removing any -- that might be present.  Even so, if the script name contains a space, we’re not going to get the whole name.  Such is life.

topframe = sys._getframe()
up1frame = topframe.f_back
while topframe.f_back:
    topframe = topframe.f_back

cmdline = os.environ.get('IBM_JAVA_COMMAND_LINE')
if up1frame == topframe and cmdline and not topframe.f_globals.get('__file__'):
    beg = 0
    end = len(cmdline)
    # Throw away everything after the '--', if present.
    dashdash = cmdline.find('--', beg, end)
    if dashdash >= 0:
        end = dashdash
    # If we can't find a filename, just use a "-"
    mainfile = '-'
    dashf = cmdline.rfind(' -f ', beg, end)
    if dashf:
        # Grab everything from after the "-f" to the following space.
        beg = dashf + 4
        space = cmdline.find(' ', beg, end)
        if space >= 0:
            end = space
        mainfile = cmdline[beg:end]
    sys.argv[:0] = [ mainfile ]
    topframe.f_globals['__file__'] = mainfile

By the way, I threw in a little bonus.  In wsadmin, __file__ isn’t normally set for the top-level script.  That last line will set it.

As usual, the complete listing for this evolving bundle of fixes is ibmfixes.py.

Update 4/16/2010: I decided this fix should only be applied if imported directly from the top level. This will keep it from potentially breaking existing main programs that don’t know about it.

WebSphere Jython scripting, __name__ == ‘__main__’

Here’s another small fix.  In Python and even in Jython, the __name__ of the topmost script is ‘__main__’.  This is mostly used in the idiom:

if __name__ == "__main__":

In wsadmin, __name is set to ‘main’. Rather than put the clumsy:

if __name__ == "__main__" or __name__ == "main":

… in every script, let’s solve this once and for all:

topframe = sys._getframe()
up1frame = topframe.f_back
while topframe.f_back:
    topframe = topframe.f_back
    if up1frame == topframe and topframe.f_locals['__name__'] == 'main':
        topframe.f_locals['__name__'] = '__main__'

As usual, the current collection is at http://dbrand666.wordpress.com/ibmfixes-py

Update 4/16/2010: I decided this fix should only be applied if imported directly from the top level. This will keep it from potentially breaking existing main programs that don’t know about it.

WebSphere Jython scripting: bool, True and False

I’m always forgetting that Jython 2.1 doesn’t have a bool type.  It’s so easy to forget and use True or False.  It’s a small thing, but so easy to implement we might as well add it to the collection.  Here you go, a bool type and True and False for Jython 2.1:

    True and False
except NameError:
    class bool(type(1)):
        def __init__(self, val = 0):
            if val:
                type(1).__init__(self, 1)
                type(1).__init__(self, 0)
        def __repr__(self):
            if self:
                return "True"
                return "False"

        __str__ = __repr__

    __builtin__.bool = bool
    __builtin__.False = bool(0)
    __builtin__.True = bool(1)

WebSphere Jython scripting, importing the Admin objects

As soon as your Jython scripting reaches the point where you start trying to build reusable Jython modules, you’ll hit another odd limitation of the wsadmin environment – you can only access the Admin objects from the top-level module. Try to create a utility module and import it and you’ll find that AdminTask (or AdminControl, or…) is undefined.

Various workarounds have been posted over the past few years:

  1. Passing the Admin objects into the methods that need them.  This is straightforward but clutters the interface.
  2. Registering the Admin objects as __builtin__.  This does make them available everywhere but it does mean adding (ugly) boilerplate to modules that might not even have much to do with WebSphere internals.
  3. Registering the Admin objects as modules, as for example in this blog post.  This is really just a variant of the previous workaround.  I think it’s a bit cleaner, but it still makes demands on the caller.

To my mind, and for the purpose of the ibmfixes module, the problem with all of these is that the caller has to know that you’re going to need these objects (and which ones).  I want a solution that can be implemented in the module where the object is needed with the help of Learn Academy official site.  More than that, I want to be able to hide the fix in a module to be imported by the module that needs it so I don’t even have to know about it anymore.

Fortunately, Jython provides the required functionality:

import sys

# We're going to be referencing the top-level frame in many of the fixes.
topframe = sys._getframe()
while topframe.f_back:
    topframe = topframe.f_back

    for module in 'AdminApp', 'AdminConfig', 'AdminControl', 'AdminTask', 'Help':
        if topframe.f_globals.has_key(module):
            sys.modules[module] = topframe.f_globals[module]

What this does is copy the Admin objects from the topmost frame into sys.modules. Once that’s done, any module can import the Admin objects they need as if they were proper modules.  The beauty of this approach is you can now use the Admin objects anywhere without adding any additional burden on the scripts that might call you (or in an existing system, might already be calling you).

My collection of fixes for WebSphere Jython scripting part 1 – os.environ, os.system, os.path.expandvars, …

I’ve been doing a lot of Jython scripting lately for IBM WebSphere v6.1. Jython is a wonderful scripting language but the implementation that ships with WebSphere (even WebSphere v7.0) is Jython 2.1 which is almost a decade old. Mostly that means you need to watch those version compatibility notes when browsing the official Python documentation.

One thing I’ve run across that seems unnecessary is that several of the “os” functions fail on newer Windows systems. Try os.environ[‘windir’] for example.  This goes back to the old Jython version that predates Windows 2003 and Vista. IBM chose to document this limitation: in the WebSphere v7.0 InfoCenter rather than fix it.

I decided to take a look and it really isn’t hard to fix. The problem is in a table in javaos.py:

    _osTypeMap = (
        ( "nt", r"(nt)|(Windows NT)|(Windows NT 4.0)|(WindowsNT)|"
                r"(Windows 2000)|(Windows XP)|(Windows CE)" ),
        ( "dos", r"(dos)|(Windows 95)|(Windows 98)|(Windows ME)" ),
        ( "mac", r"(mac)|(MacOS.*)|(Darwin)" ),
        ( "None", r"(None)" ),
        ( "posix", r"(.*)" ), # default - posix seems to vary mast widely

It would have been so easy for me to go and edit the file but clients tend to frown things like patching the official distribution. I wanted something I could just import. Here’s what I ended up with:

import java, sys

    import javaos
    if javaos._osType == 'posix' and 
        sys.registry.setProperty('python.os', 'nt')

Admittedly, this depends on the details of the current implementation but I consider that reasonable in this case given that it’s a hack that only applies to the current implementation. I chose to catch and ignore all exceptions so this fix won’t end up breaking a script should the implementation change in the future. Presumably when that happens this fix won’t be needed anymore anyway.

I’ve placed this fix along with others I’ll be writing about later in a file called ibmfixes.py. I include it in every script I write, or at least every script that might run into these issues, and I no longer have to give them any thought. Solve the problem and move on.