From maintainers-request at octave dot org Wed Oct 26 09:37:55 2005
Subject: Re: Converting DejaGNU tests to Paul's test/assert infrastructure
From: David Bateman <David dot Bateman at motorola dot com>
To: Paul Kienzle <pkienzle at users dot sourceforge dot net>
CC: octave maintainers mailing list <maintainers at octave dot org>
Date: Wed, 26 Oct 2005 16:33:46 +0200

Paul Kienzle a écrit :

>
> On Oct 26, 2005, at 5:22 AM, David Bateman wrote:
>
>>
>> * The functions in the test scripts are converted to funtion handles 
>> and so things like
>>   fsolve("f", ....) won't work. Paul are the function handles really 
>> needed, or is there a way to
>>   get the same functionality in the test scripts without nested 
>> functions?
>
>
> The problem is that the test function might blow away a function of 
> the same name in the users environment.  Instead I'm generating a 
> unique name and returning a handle to that.  The alternative is to 
> find a way to create a function in a private namespace only, or to 
> modify the name of the function so that it is guaranteed not to collide.
>
> For now we can change the definition to use the function name 
> directly, but build a 'private namespace' in the usual way by changing 
> the function name to package_fn, or in this case, changing function a 
> to function __test_a in the definition and use.  Later when support is 
> added for private functions those tests will continue to work without 
> change.
>
> - Paul

Paul,

Is it necessary to allow arbitrary contexts or namespaces to allow this 
in test.m? Maybe if we could imagine a symbol table 
"global_fbi_sym_tab", then something like

global f;
function y = f(x)
  y = sin(x);
endfunction
function y = g(x);
  global f;
  y = 2 * f (x);
endfunction
g(1)

would be sufficient. Alternatively if "top_level_fbi_sym_tab" and 
"curr_caller_fbi_sym_tab" existed then we might do the same with evalin. 
A further means might be a syntax like

function y = f (context, x)
y = evalin(context, 'g(x)');
endfunction
function y = g(x)
sin(x);
endfunction
context = save_context();
y = f(context,1);

which would require an octave_value to save the current context and that 
evalin was capable of using it. One thing that I see that is odd 
behaviour o octave while looking at this is the following

global f;
f =  at sin;
function y = g(x)
global f;
y = f(x);
endfunction
g(1)
function y = f(x)
y = cos(x);
endfunction
f(1)
g(1)

which returns

ans = 1.6829
ans = 0.84147
ans = 1.6829

so the redefition of "f" as afunction works in fbi_sym_tab in the 
current context, but it doesn't alter the fact that its defined as a 
function handle in the global space. This is odd as in the above it is 
defined as global in the top-level context. So perhaps the first idea 
above needs to be done in any case to remove this little inconsistency...

Regards
David