Re: [sv-bc] Ballot issue 216 (Mantis 693) generated identifiers

I am uncomfortable with this.

Some initial questions/issues:
   1) This completely changes how "genvar" works.  A genvar used
      to be "value-less" other than in the context of a generate
      loop where the iteration scheme completely determined the
      values.  Now a genvar is much more "stateful".  At very
      least there would need to be restrictions placed on
      assigning to a genvar within the context of a controlling
      loop.  You would also run into having to define default
      values for a genvar, etc. since that is not an issue at this
      point.
   2) A genvar is currently visible as a localparam in the context
      of the generated loop body.  What is the visibility for vpi, etc.
      of a genvar used in the manner that you suggest?
   3) With generated identifiers, we completely lose the ability to deal
      with implicit net creation at parse time; that now becomes an
      elaboration time activity.  This impacts things like 'default_nettype
      handling, etc. and delays any error reporting until elab time.
   4) I definitely don't like the syntax since the [] form can't be
      determined during parse as being either a packed dimension
      or a generated identifier:
         reg [a] b;   // is [a] a generated identifier making this an error?

Issues 1-3 are pretty big ones; the syntax can easily be fixed
in various ways.

There are likely also lurking issues in terms of hierarchical
references and other elaboration time interactions, but I would
need to think further about that.

At this point, I don't think I can support this concept at all.

Gord.



Maidment, Matthew R wrote:

> I have uploaded a proposal for Ballot Issue 216, Mantis 693:
> 
> http://www.eda.org/svdb/bug_view_page.php?bug_id=0000693
> 
> I believe the use of square brackets to be problematic, but
> I want to present the intent of the functionality.
> I hope someone is more creative with the limited special
> character set and can suggest an alternative syntax.
> 
> I'm not a fan of $identifier.
> 
> Matt
> --
> Matt Maidment
> mmaidmen@ichips.intel.com
>  
> 

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