Edward d'Auvergne wrote:
> Hi Gary,
>
>> wellcome back, I hope the search for jobs and fame was fruitful
>
>
> I'll definitely be around in the future. I'm not in a position to
> confirm anything though.
>
>
>> I have been trying to get the current unit tests to work from the
>> command line e.g. ' ./unit_test_runner.py maths_fns/test_chi2.py'.
>> However, most of them fail beacuase the global variable Debug is not
>> defined. Chris tells me that this is because debug has been added
to the
>> python __builtins__ namespace. This points to a couple of questions
>>
>> 1. why are we playing with __builtins__?
>
>
> A number of relax features are placed into __builtin__, most
> significantly the RelaxError and RelaxWarning systems. These mimic
> and simply extend the normal python error and warning exception
> systems. I did notice that this will be an important issue for stand
> alone operation of individual unit tests - the calls to RelaxErrors
> will necessarily cause exceptions. As for a solution I don't know the
> best way of handling the __builtin__ objects and flags.
The first thing to say is that I don't think we should be adding
anything to builtins that isn't there already in python(?)
>
>
>> 2. how do we get round this
>
>
> I'm not sure.
>
>
oops ;-)
>> as an aside __builtins__ is 'an implementation detail'
>> http://docs.python.org/lib/module-builtin.html and its use will
prevent
>> relax from working on iron python, jython and pypy. So is this use of
>> __builtins__ something we need to reconsider...
>
>
> Apart from __builtin__, how else would we have the equivalent of a
> 'global' variable (accessible to all parts of the program)? Are there
> alternative solutions to using __builtin__ or are there alternative
> solutions for getting the __builtin__ relax objects into the unit
> tests? I don't see the addition of objects to __builtin__ as a big
> implementation issue as __builtin__ exists in all python
> implementations
> (http://www.jython.org/docs/api/org/python/core/__builtin__.html and
> Google picks up the rest of the scattered docs).
well that maybe the case for jython now but the fact that the main
python documentation says it is implimentation detail (though Chris
pointed out to me that this was the per module __builtins__ variable).
However, good sources such as the python cookbook also point out that
__builtin__ shouldn't be used
http://www.oreilly.com/catalog/pythoncook/chapter/ch01.pdf
Importing __builtin__ and assigning to its attributes is a trick that
basically defines anew built-in object at runtime. All other modules
will automatically be able to access these new built-ins without having
to do an import. It's not good practice, though, since readers of those
modules should not need to know about the strange side effects of other
modules in the application. Nevertheless, it's a trick worth knowing
about in case you encounter it.
In general __builtin__ is just a way of avoiding an import. Now for the
relax interpreter that could be avoided just as well by doing an
implicit import of a module that defines all globals at the start of the
script if you wanted to. For the real hacky programmers dabbling in the
source code they would know to do the required imports anyway. e.g. from
relax import * would import all globals etc required for relax
porgramming...
> Wouldn't global
> variables have the same issue, being absent, with the stand alone unit
> tests?
yes, the globals was an aside to the point that if we really multithread
python at any time soon ;-) globals will become a no no for things that
are not global to all threads
> Could an alternative be the creation of a new module which
> sets up all the relax __builtin__ objects by importing 'error.py' and
> setting dummy values of 'Debug' to zero?
>
now that is one stratergy we could use. A module that you import and
just sets up all of the global/per thread variables required. However, I
would still urge against adding anything new to builtins as it is just a
method for changing the behaviour of types classes and objects that
alread exist and for avoiding pesky import statements for the main
python object types. Furthermore having one globals module is also not a
good design decisioon as variables start to loose obvious ownership
which is what the modules mechannism aims for... (see below)
Now to a more philosophical bent. Python global variables are in almost
all cases except for __builtin__ __per module__ globals variables. Now I
am sure that this is quite a careful design decision made to ensure that
the global name space is not poluted and that there is always a clear
owner for a variable. Thus in (positivley humble) opinion all globals
ought to be global to a module and imported via an import statement. Use
of the buitins dictionary to create global global variables is in almost
all cases trying to hack roun the philosophical under pinnings of the
language and its structures.
philosophical bent 2. I am not proposing that we outlaw global variables
persay. Just that we will just need to pay special attention. In general
it is easier to modify the behaviour of variables within classes (this
is what get and set mechanism is for) and overall the most robust long
term solution for mutable global and per thread variables is to access
them as variables from an instance of a singleton which is global to a
module
end of philosophical rant!
regards
gary
> Cheers,
>
> Edward
>
> .
>
--
-------------------------------------------------------------------
Dr Gary Thompson
Astbury Centre for Structural Molecular Biology,
University of Leeds, Astbury Building,
Leeds, LS2 9JT, West-Yorkshire, UK Tel. +44-113-3433024
email: garyt@xxxxxxxxxxxxxxx Fax +44-113-2331407
-------------------------------------------------------------------