(r2) EmptyRecord < P1076

P1076 Web>Vhdl2019CollectedRequirements>EmptyRecord (revision 2) (raw view)~~EditAttach~~
---+ Syntax Regularlization: Empty Records

%TOC%

---++ Proposal Information
   * Who Updates: Main.RyanHinton
   * Date Proposed: 2013-10-28
   * Date Last Updated: 2013-10-28
   * Priority:
   * Complexity:
   * Focus: General Language

---+++ Language Change Specification link

[[LCS2016_082][LCS-2016-082]]

---++ Requirement Summary

Allow record declarations to contain no elements.

---++ Rationale

Record with debug ports. Record has elements when debugging block. Record empty in final design.

The need is similar to that for null arrays. Often I want a port into a component, e.g. for status, and I don't know ahead of time what status I need to include. Assume I write my entity before my architecture and I want to wrap all the status signals in a single OUT-mode record port. Then I have several choices.

   * Be omniscient and know from the start exactly what I need on my interface. (This may work for you, but it doesn't for me.)
   * Don't define the record type or add the port until I have the first status item to report. This requires changing a the entity declaration, the package for the record type, any component declaraions, and any instantiations. And then later if all my status items go away for some reason, I have to remove the port and type in all of these contexts again.
   * Declare the type in the package and the port in the entity declaration. (Currently this can be done with a dummy element.) Then I can add and remove status items simply by adding them to the type declaration in the package, driving them in the architecture (which is where I usually discover what to drive in the first place), and then hook up a receiver somewhere if desired.

Obviously the third option is ideal.

---++ Related Issues: None = General
---++ Competing Issues: None at this time

---++ Current Syntax:

Currently, a "blank" record requires a dummy element or placeholder of some sort.

<verbatim>
  type confusing_record is record
    dummy : boolean;
    -- dummy placeholder to avoid compile errors: confuses designers, muddies 
    -- the interface, wastes processor cycles when ripped out by synthesis, 
    -- wastes memory in simulation; otherwise it works fine
  end record;</verbatim>

---++ Proposed syntax change

Allow records to have zero elements.

<verbatim>
  type accurate_interface is record
  end record;  -- not currently legal</verbatim>

---++ Possible Issues

---++ Arguments FOR

   * The LRM impact should be light.
   * No apparent syntax ambiguities or issues.
   * It shouldn't be difficult for tools to implement.
   * It's a nice productivity enhancement.

---++ Arguments AGAINST

Even C doesn't allow empty struct declarations. The only reason one needs a record type is to access fields of the type. If there are no fields, there's no need for a record type.

-- Main.ErnstChristen - 2015-01-21

Null ranges of vectors may occur in values, but objects declared with a null range are meaningless. IMO this proposal makes sense only if slices of records and corresponding concatenations are supported in the language.

Many C compilers do tolerate empty structs though (e.g. gcc), and C++ explicitly allows them. Different languages have different needs and this reference does not provide a helpful insight.
---++ General Comments

---++ Supporters

_Add your signature here to indicate support for this proposal._

-- Main.RyanHinton - 2013-10-29

-- [[Main.BrentHahoe][Brent Hayhoe]] - 2013-10-30

-- Main.LievenLemiengre - 2015-06-04

-- Main.ThomasPreusser - 2016-09-09
---++ In opposition

_Add your signature here to indicate opposition to this proposal._

--main.CliffordWalinsky - 2013-11-13

-- Main.ErnstChristen - 2015-01-21
Topic revision: r1 - 2020-02-17 - 15:34:29 - TWikiGuest
P1076