[Date Index] [Thread Index] [Author Index]

Re: Compile. Not again

• To: mathgroup at smc.vnet.net
• Subject: [mg56482] Re: [mg56470] Compile. Not again
• From: "Wolf, Hartmut" <Hartmut.Wolf at t-systems.com>
• Date: Tue, 26 Apr 2005 21:52:37 -0400 (EDT)
• Sender: owner-wri-mathgroup at wolfram.com

>-----Original Message-----
>From: Maxim [mailto:ab_def at prontomail.com] 
To: mathgroup at smc.vnet.net
>Sent: Tuesday, April 26, 2005 7:34 AM
>Subject: [mg56482] [mg56470] Compile. Not again
>
>Here's what I get in version 5.1.0:
>
>In[1]:= f = Compile[{}, Module[{L = {0}}, L[[1]]++; L]]
>
>In[2]:= a = f[]
>Out[2]= {1}
>
>In[3]:= b = f[]
>Out[3]= {2}
>
>In[4]:= Table[f[], {3}]
>
>Out[4]= {{5}, {5}, {5}}
>
>In[5]:= {a, b}
>Out[5]= {{5}, {5}}
>
>Each subsequent call to f uses the modified value of L. 
>This also affects Table and all the objects involved 
>in assignments such as x = f[] (but not  
>x = f[][[1]]), including DownValues[Out]. The problem
>might be related to deciding whether a and b point to 
>the same shared object.
>
>Maxim Rytin
>m.r at inbox.ru
>
>

Maxim,

compare this to

In[1]:= L = {0};
In[2]:= f0[] := (L[[1]]++; L)

In[3]:= (f0[]; a0 = Unevaluated[L])
Out[3]= {1}

In[4]:= (f0[]; b0 = Unevaluated[L])
Out[4]= {2}

In[5]:=
Identity /@ Table[(f0[]; Unevaluated@Unevaluated[L]), {3}]
Out[5]= {{5}, {5}, {5}}

In[6]:= {a0, b0}
Out[6]= {{5}, {5}}

(There are yet other ways to simulate this.) So I might follow you that
the OwnValues of a and b "point to the same object" (which is thought to
be a constant, but isn't).  This reminds me of old days of Fortran,
where Compilers dealt constants as references, whith the effect that you
could turn 5 e.g. to 7, by a function call.

Table in your example is innocent, as this produces a list of the same
references.


We might have been warned off, however, by those meager words of Help:

	"Compiled code does not handle numerical 
	precision and local variables in the same
	way as ordinary Mathematica code." 

--
Hartmut Wolf