(r1) 2016_MeetingFebruary18 < P1076

P1076 Web>Vhdl2019MeetingMinutes>2016_MeetingFebruary18 (revision 1) (raw view)~~EditAttach~~
<!--
Set ALLOWTOPICCHANGE = P1076AdminGroup
-->
---+!! P1076 February 18, 2016 Meeting Minutes
---++!! Attendees:

   * Brent Hayhoe, Rob Gaddi, Patrick Lehmann, Jim Lewis, Peter Flake, Ernst Christen, Jing Pang

---++!! Agenda:

%TOC%

---++ Meeting Discussion
   * What's Next - see [[PrivateDocuments]] file: _summary_vhdl_requirements_priority.xlsx
   * Garbage Collection (item 31)
      * What stops a simulator from doing this now? Nothing
         * The language allows this (9.3.7 note 2), however, we cannot count on it happening
      * Cannot be handled at the user level
      * implied deallocate
      * Use model: to_string using write(buf, ...)
      * Allow user to suggest deallocation with deallocate
      * Good upside and little downside (simulator speed issue?)
      * Reference count (issues with a ring), Mark & sweep
   * Path_name, Shared Variables, and Subprograms
      * path_name and instance_name follow where the protected type is declared
      * We need an attribute that follows where the shared variable is declared
         * Exposes instance to implementation
      * within a protected type, use of this'instance_name makes sense (see Rob's notes)
      * within a regular subprogram, intent is to go back to the line that called the subprogram.
         * Intelligent call stack
         * Information model is already in VHPI
         * Maybe this type of information should be gathered via std.env
      * May require 2 separate solutions one for PT and one for regular subprograms
   * Composition of protected types
      * Deriving capability from existing objects and layering rather than recreating
      * Look at Suave tagged types for additional ideas (ADA has tagged types)
   * Generics for Protected Types
      * Look also at ADA discriminants as an alternative - discriminant is more like a subtype constraint
      * Coordinate with interfaces and bundles
      * Each generic instance of a type is a unique type (and likely needs to be for PT type applications)
   * Vector Literal Introspection
      * For single dimensional array objects, look at the object itself to help disambiguate subprogram calls
      * Compile time issue. 
      * Is it a valuable enough change to justify the effort in implementing it?
      * Need someone who is really interested in this to take it further
   * Conditional Expression or Operator
      * Addresses issues with initializing constants, signals, and variables
      * Straight forward add implementation 2 and justify that it makes the language symmetric with assignment capability
      * Implementation 1 would require some analysis when it is mixed with a conditional assignment. 
      * It seems to be a no-brainer to do it for initializers - leverage variable assignment capability
      * For a general expression it only makes sense if it generalizes many different things.
         * Complication is in signal assignments, as when allows waveform elements
   * Stopped at line 37

---++ Review and Approve Meeting Minutes:

Rob, Ernst
---++ Next Meeting: Thursday February 25, 2016 11 am Pacific
---+++ Previous Meeting: Thursday [[2016_MeetingFebruary11][February 11, 2016]]
Topic revision: r1 - 2020-02-17 - 15:36:16 - TWikiGuest
P1076