RE: [sv-bc] declaration vs reference order issue

Gord,

As I replied in my previous message, binding the same identifier in a
particular scope to more than one object is simply creating a debug
problem for users. This is what I meant by "the only consistent way" -
short of an error. Binding a name to multiple objects is (at least by my
definition) inconsistent. The problem exists in modules, but we have a
chance to avoid the problem in classes - just as C++ has managed to do.
I am arguing from a user's perspective, not in favor of a particular
implementation.

And I would definitely favor making your example an error - why would
any user want to subject themselves to the problems associated with
type/data aliases?

	Arturo

-----Original Message-----
From: Gordon Vreugdenhil [mailto:gordonv@model.com] 
Sent: Friday, September 05, 2008 8:06 AM
To: Arturo Salz
Cc: Rich, Dave; Daniel Mlynek; sv-bc@eda-stds.org; Mirek Forczek; Piotr
Winter; Sergei Zaychenko
Subject: Re: [sv-bc] declaration vs reference order issue

Arturo,

I don't agree that this is "the only consistent way".  What is not
consistent about claiming that the 1809 rules apply to classes
from "the point of reference" in the same way?  Clearly the
inherited names are "visible" prior to that point since the
point of reference must follow any "extends" clause so that is
a pretty easy interpretation.

I think that allowing the C++ style rules adds substantial
complexity -- you now have different rules for type and non-type
names since (I think) that everyone agrees that forward references
to type names in a class can't be permitted.

So then you have to add yet more rules to deal with "changes of
meaning" as you have in C++.  For example, the following is
not legal in C++:

   typedef int myint;
   class C {
     myint x;
     int myint;
   };


If you make this illegal in SV, you then have very different
rules for class lexical visibility than for modules, etc.
Why should forward ident references in classes be different than
in modules, etc?

The name resolution rules in SV are already horrendous -- the
interaction of dotted names, opaque types, hierarchical
referencing, compilation unit, and legacy "forward" tf naming,
make understanding things very, very difficult.

There is very little advantage to going to the C++ implicit
forward naming.  You can already always get a forward data
member reference by adding "this.", it is irregular with other
nested scopes to have special rules, type references
become even more irregular, and it is pretty easy to
avoid the whole issue by coding conventions (data first).

Although I understand the appeal of following C++ in terms of
what (some) users might expect, I just don't think that the
resulting language issues justify it.

Gord


Arturo Salz wrote:
> Dave,
> 
>  
> 
> This has been discussed in the past - in the context of randomize-with

> and package imports. The only consistent way to deal with the 2 cases 
> that Daniel shows is to resolve all class references to the class,
that 
> is, in the two examples, an undecorated 'i' always refers to the one 
> inside the class - both cases should display 2. This is similar to the

> way in which C++ handles class members: undecorated references to
class 
> members always resolve within the class, regardless of the declaration

> order. And, like SV, C++ also has a declaration-before use rule for 
> other identifiers. BTW, don't forget that resolving names in classes 
> must also consider the inheritance hierarchy, for example:
> 
>  
> 
> int v = 1;
> 
>  
> 
> class B;
> 
>   static int v = 2;
> 
> endclass
> 
>  
> 
> class C extends B;
> 
>   static int w = v;            // this 'v' resolves to the base class 
> B::v (2)
> 
> endclass
> 
>  
> 
> On another note, whether the compiler is a single-pass or
multiple-pass 
> is definitely an implementation issue that should not be mandated by
the 
> LRM. I recall that Mark Hartoog had proposed a scheme that would
support 
> all this, but I don't see it in the text.
> 
>  
> 
>             Arturo
> 
>  
> 
> *From:* owner-sv-bc@eda.org [mailto:owner-sv-bc@eda.org] *On Behalf Of

> *Rich, Dave
> *Sent:* Thursday, September 04, 2008 8:49 AM
> *To:* Daniel Mlynek; sv-bc@eda-stds.org
> *Cc:* Mirek Forczek; Piotr Winter; Sergei Zaychenko
> *Subject:* RE: [sv-bc] declaration vs reference order issue
> 
>  
> 
> Mantis 2106 added to 6.21
> 
>  
> 
> A variable declaration shall precede any simple reference 
> (non-hierarchical) to that variable.
> 
>  
> 
> So there can be no tool dependence on this issue. Customers certainly
do 
> not like when code compiles on one simulator and not on another. The 
> begrudgingly accept order of evaluation dependencies only if a single 
> implementation cannot guarantee ordering.
> 
>  
> 
> For case 2, I thought we had added text that defines the search rules 
> from the point of the source code, but I can't seem to find it. But 
> given the rule above, there's no way $display(i) can refer to this.i
and 
> display 2.
> 
>  
> 
> Dave
> 
>  
> 
>  
> 
>  
> 
>
------------------------------------------------------------------------
> 
> *From:* owner-sv-bc@server.eda.org [mailto:owner-sv-bc@server.eda.org]

> *On Behalf Of *Daniel Mlynek
> *Sent:* Thursday, September 04, 2008 2:53 AM
> *To:* sv-bc@server.eda-stds.org
> *Cc:* 'Mirek Forczek'; 'Piotr Winter'; 'Sergei Zaychenko'
> *Subject:* [sv-bc] declaration vs reference order issue
> 
>  
> 
> CASE1:
> 
> Does LRM defines rules on how tool should behave if reference to 
> an identifier which is unknown at this point of source code.
Idetnifier 
> is delcared later in code (case1). LRM specifies that both single pass

> parser and multi pass parser may be used for parsing SV code (LRM:" 
> Implementations may execute compilation in one or more passes") It is 
> obvious that single pass compiler will fail on case like below, while 
> mutli pass can handle with that. So this should be tool dependend if 
> CASE would pass compilation?
> 
>  
> 
> 2nd case is connected - what should happend if in current scope and in

> higher scope there is an identifier declared. But in current scope 
> declaration is place in code after reference - so which one should be 
> printed by case2. Is it defined or tool dependend?
> 
>  
> 
> LRM:22.9 says:"If an identifier is referenced directly (without a 
> hierarchical path) within a task, function, named block, or generate 
> block, it shall be declared either within the task, function, named 
> block, or generate block locally or within a module, interface,
program, 
> task, function, named block, or generate block that is higher in the 
> same branch of the name tree that contains the task, function, named 
> block, or generate block. If it is declared locally, then the local
item 
> shall be used; if not, the search shall continue upward until an item
by 
> that name is found or until a module,  interface, or program boundary
is 
> encountered."
> 
> So imho in both cases "2" should be displayed
> 
>  
> 
>  
> 
>  
> 
> CASE1: 
> 
> module top;
> 
>  
> 
>    class nested;
>        function new ();
>               $display(i);    //  displaying "i" - this should be a 
> failure - as i is unkcnown since here, or 2 shoudl be displayed?
>       endfunction
>       int i = 2;               // local member "i"
>    endclass
> 
>  
> 
>    nested n_inst = new();
> 
>  
> 
> endmodule
> 
>  
> 
> CASE2:
> 
> module top;
>     bit [2:0] i = 0;   // static member "i"
> 
>  
> 
>    class nested;
>        function new ();
>               $display(i);    //  displaying "i" - which i would be 
> displayed - ???????????????
>       endfunction
>       int i = 2;               // local member "i"
>    endclass
> 
>  
> 
>    nested n_inst = new();
> 
>  
> 
> endmodule
> 
>  
> 
>  
> 
>  
> 
> DANiel
> 
> 
> -- 
> This message has been scanned for viruses and
> dangerous content by *MailScanner* <http://www.mailscanner.info/>, and
is
> believed to be clean.
> -- 
> This message has been scanned for viruses and
> dangerous content by *MailScanner* <http://www.mailscanner.info/>, and
is
> believed to be clean.
> 
> 
> -- 
> This message has been scanned for viruses and
> dangerous content by *MailScanner* <http://www.mailscanner.info/>, and
is
> believed to be clean.

-- 
--------------------------------------------------------------------
Gordon Vreugdenhil                                503-685-0808
Model Technology (Mentor Graphics)                gordonv@model.com


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.