SV-EC Errata Meeting Minutes
Monday 8 November 2004   11:00 am. to 1:00pm 
[Distribued, for approval at next sv-ec errata meeting]

Attendees:
----------

(a--AA) Alex Wakefield
(AA-AA) Arturo Salz
(----A) Bill Paulsen 
(AAAAA) Brad Pierce
(-AAA-) Cliff Cummings 
(A-AAA) Dave Rich 
(A----) David Smith
(--a-a) Don Mills 
(A----) Dennis Brophy 
(AAA--) Eugene Zhang
(AAAAA) Francoise Martinolle
(A-A--) Karen Pieper
(AAAAA) Mehdi Mohtashemi 
(AAAAA) Neil Korpusik 
(AAAAA) Ray Ryan 
(--AAA) Steven Sharp
(A----) Stu Sutherland
(aAAAA) Surrendra Dudani
(-Aa--) Yogesh Pandey
 |||||
 |||||_   8 November  2004
 ||||__  25 October   2004
 |||
 |||___  27 September 2004
 ||____  13 September 2004 
 |_____  31 August    2004
 
 
 
			
A => Attended (Present for both roll calls: beginning and end of meeting)
a => Present for roll call at beginning, absent at end meeting roll call
- => Missed

Action items from November 8, 2004:
-----------------------------------
   1. Surrendra - errata 203, check for compatibility with sv-cc and
      compatibility with C++ stl. Also specify how strlen handles an 
      embedded null byte.
   2. Surrendra - errata 271, change the example to use a typedef plus a cast.
   3. Dave Rich - errata 275,  put together a proposal and post it on the web 
      site.     
   4. Surrendra and Ray - errata 4, discuss this off-line and resolve before 
      the next meeting.
   5. Ray - errata 175, send out a proposal.
   6. Ray - errata 211, respond to the comments from Cadence.
   7. Francoise - errata 260, come up with a proposed clarification.
   8. Steven - errata 261, make a proposal.
   9. Steven/Bill - errata 262 make a proposal. [261/262 may be combined]


Minutes 8 November 2004 taken by Neil Korpusik

1)  Review the IEEE patent policy
    ref:   http://standards.ieee.org/board/pat/pat-slideset.ppt 
    We agreed to assume that the policy was read since everyone is familiar 
    with it.

2)  Approve the minutes of the last meeting [October 25 2004 meeting]
		http://www.eda.org/sv-ec/Minutes.html

   Motion: to approve minutes of October 25, 2004: Ray
   Second: Surrendra
   Abstain: <none>
   Opposed: <none>
   Passed


3)  Next meeting schedule and p1800 proposed deadline guidance.
    The P1800 will be having their next meeting on Nov. 22, 2004. This is 
    the same day that the next SV-ec meeting is scheduled for. It was decided
    that there wouldn't be much of a conflict with people that attend this
    meeting. We will hold our next meeting as originally scheduled.

    Next Meeting - November 22, 2004. 11:00am - 1:00pm [PST]


4)  Review/discuss and vote on Errata in the database.
    
    a) String questions:
	Errata: 197, 203 (Per Dave Rich request to delay for next meeting)
        271   Incorrect example of assigning a string to a register in Section 3.7
	275   Several issues with section 3.7 String data type 

    Errata 197:  Is string an array? - section 3.7 first paragraph
    Dave: objected to the use of the word character contained in the re-wording 
	of this paragraph.
    Surrendra: character is defined in the table 3.1, byte or ASCII character.
    	Table 3-1 in section 3.3 contains a list of "Integer data types". For
	byte it states "2-state SystemVerilog data type, 8 bit signed integer 
	or ASCII character".
    Arturo: Section 2.6, it talks about characterfirst paragraph. "Non-printing
	 and other special characters are preceded with a backslash."

    Dave: Why was the dynamically allocated array of bytes not correct.
    Surrendra: string is not really an array.
    Dave: the fourth paragraph was deleted. If everyone is ok with 
	character. Other places we talk about dynamic array of bytes for
	the purposes of bit-stream.
    Ray: it does say if you read it you get a byte.
    Neil: Section 3.16 also talks about byte.
    Arturo: Character implies more than byte, it has ASCII notion.
    Dave: The reason character is not there because of unicode.
    Fransoice: What is ordered list. 
    Arturo: You can implement it as a list, for implementation purposes.
    Brad: Is it indexable, looks like unpacked-array.
    Ray: Is there anything that you can do to array that can not do string.
    Dave: Slice of string
    Surrendra: slice is not really defined.
    Neil: the methods associated with array is different from array.
    Ray: Is it any different from collection of bytes.
    Fransoice: We should remove character from the definition, since it is not	
	defined.  We need to provide the definition of character.
    Neil: The methods use the word character.
    Surrendra: it defines the operation, without going to detail
    Ray: Character is a concept but not a data-type.
    Francoise: should we include order collection of ascii characters. BNF
	says string literal is ascii character. 
    Arturo: String literal is not string data-type. It should not be limited
	to ASCII. 
    Neil: It returns a byte, in the table it is ascii,
    Arturo: the question is the 8th bit.
    Ray: All 8th bits are usable.
    NOTE: The 8th bit of an ASCII character is not defined. Leaving off the word 
      	ASCII was intentional.

 	Motion: to accept errata 197 proposal
	Move: Surrendra
	Second: Arturo
	Abstain: Dave (I like to see string behave more like array)
	Oppose
	197 passes


    Errata 203: String data type, Table 3-2: String operators, relational operators

	   "The comparison behaves like the ANSI C strcmp function (or the 
	   compare string method) (with regard to the lexical ordering) and 
	   embedded null bytes are included." 
 
    Ray: Strings list of bytes, it can have 0.
    Surrendra: If you allow null, then it is not consistent with 
	ANSI C. makes it incompatible with c-functions, also vpi and dpi
	taking c as main interfaces. It will not work properly.
    Ray: The null bytes have additional semantics than c.
    Dave: they will terminate if you use c function. you can still pass a string.
    Francoise: the string will be terminated.
    Dave: Some simulators will change null to space.
    Ray: This is $display command-- null to blank space.
    Brad: In c you can pass out register with null embedded in them, c-functions
	are not usable on them.
    Francoise: You have to know the size.
    Surrendra: Users do not allocate any bytes. In c you have full control 
	over the array. As soon as string, you can not have null.
    Dave: The original spec, has no restriction about null bytes. This is
	a change that is not needed.
    Ray: If we declare a string variable 8, put a null character in second byte,
	what would the size be.
    Surrendra: it will get 1,
    Ray: I think it should be 8.
    Surrendra: In c you have complete control. 
    Francoise: The simulator always know the size. 
    Surrendra: The string size can become smaller, 
    Ray: It would only get resized if only re-assigned.  Access an array of Lvalue,
 	can we access something beyond the size of array.
    Arturo: If you allow this.
    Francoise: I would like to have strings behave like c/c++. No null bytes
	in the middle of it.  The only way you know, then you need the size of
	the string.
    Surrendra: The user will not know the notion of size.
    Ray: the length function would give size.
    Francoise: The vpi does not work. they would need the size of string in
	addition.
    Arturo: would you not have the same problem with dynamic array. you have to pass
	the lenght.
    Francoise: String would be assumed to be mapped.
    Dave: The user has to be aware of this issue.
    Surrendra: What advantage does the user get out of null byte in the middle of
	string.
    Arturo: These are consistent with c++, in c you can put anything in it. there
	is only char in c no string.
    Surrendra: In c++ if you use stl library, problems in using string with null bytes.
    Dave: Two areas would be interesting.  Concatenation of two, with null byte. 
	Also string as verification construct to pass a null. Also this is 
	a feature of 3.1a. If we make the proposed change it would be removing existing
	3.1a functionality.
.
    Surrendra: It makes it incompatible with vpi
    Ray: Semantics of storing a null does that change the meaning of string.
	If you set it to null, then again to a character, 
    Neil: Does the implementation allow reclaiming of memory back.
    Francoise: If you allow null byte, the lenght does is not known.

   [AI] Surrendra: 1) check for compatibility with sv-cc committee.
		   2) how should a null byte in the middle of a string be 
		      handled? Is the last portion of the string reclaimed?
                   3) how does strlen handle an embedded null byte?


   
    Errata 271:  Incorrect example of assigning a string to a register in Section 3.7
    Surrendra: Is this doable.
    Brad: You need a typeof, $typeof, we do not know.
    Surrendra: we can define a typedeff a reg. Not a dynamic type assignment to
	a reg.
    Arturo: You can assign a dynamic assignment to static.
    Dave; You can not assign a dynamic to static types 
    Arturo: Section 4.7 shows the rules for dynamic-static assigment. Now that 
	strings are not arrays, we do not need to define it in terms of array.

	 reg [15:0] r;
	 string b="";
	 string a={"Hi",b};
	 r=a;    // should be r=reg'(a);  <---- proposal in the errata

    NOTES: A dynamically sized object can't be directly assigned to a fixed size 
      	data object without using a cast. 

   [AI] Surrendra: change the example to use a typedef, make the foraml proposal 
	in the data-base
	
   
    Errata 275:    Several issues with section 3.7 String data type
      This errata contains a list of all the other issues that were raised 
      during a series of recent emails. 

   [AI] Dave Rich: this errata has been assigned to Dave. He will put 
	           together a proposal and post it on the web site.
    
 
    b) Randomization/constraints/randcase
     Errata 4     Add rs_rule precedence and example in Section 12.16
    Surrendra: Item 4, may create some problems for constraints. The 
	same := is used for dist operators, the expression.  If we change
	the precedence that will not work.
    Ray: These are randsequences.
    Surrendra: It will change it over in the dist operators inside the constraint
	definition of the class.
    Ray: If the expression for the weight is expression not a value, if
	treated := first, 
    Arturo: The syntax will take care of it.
    Surrendra: you are treating them differently in two places.
    Ray: In the original example, has the ambiguity to it. can you write up an
	example that breaks it.  := is the operator.
    Surrendra: It is not an expression operator, it will not bind to it, the
	expression will get rid of | as an expression.
	The rule does not allow (), it would have been able to distinguish
	current BNF does not allow it.
    
   [AI] Surrendra and Ray: discuss this off-line and resolve before the 
			   next meeting.    

    
    Errata 175
	175   Ambiguous ordering of random variables
    Ray: We agreed on, does not have any effect on solution space, once
	that is defined, choosing the solution in the solution space. The statement
	in the LRM, states for groups. The previous paragraph has ordered set.
    Arturo: Are you saying that we are partitioning the set.
    Ray: Yes,
    Arturo: That is not correct.
    Ray: Solving for the solution space. You can have a partial ordering. 
    	Why not leave it completely up to the implementation. If you do not
	care about probability, then solve-before does not do anything.
	Distribution is implementation dependent?
    Arturo: Partitioning into sets, then you do not need to do it.
    Ray: If I have solve before statement, for example x,y and a,b. 
    	you can get a case where the distribution can be effected by 
	partial ordering. 
    Arturo: Why do you want to delay it. 
    Ray: If 4 variables, a,b,c,d.  a before b, b before c, d.  
    Arturo: Ordering without constraining, 
    Ray: if you put some constraints, a<b, b<c, c<d.  You choose a,
	then given a, choose b, then choose value pair for c, d. Gives you
	a different value. If i add constraint d before c. DOes that mean
	d gets solved before b, or a.
    Arturo: That gets into implementation.
    Ray: The LRM says, that variables that have no order, the requirement says
	they solve last. But it should say the variables that have no partial
	ordering, it should be as late as possible. 
    	Why then mandate it in such a way.
        I wold like to see a, b,c and then d.
	It says you can not solve a and d first, then b, then c.
     Surrendra: I would rather delete the sentence in the LRM. It should be
	that way.
     Arturo: You delay as late as possible. 
     Ray: I would like to say it specifically say, 'solve partial solution as 
	late as possible'.  
     
      [AI] Ray: Make the statement proposal and send it out.


     Errata: 211   Add clarifying example to builtin / scope randomize
      [AI] Ray: respond to the comments from Cadence.

     Errata 260   distributions and constraints solving
	       x dist { 0 := 90, 1 := 10 }; // x is more likely to be 0
	       y dist { 0 := 10, 1 := 90 }; // y is more likely to be 1
	       constraint c { x == y; } 
	 
	    The question is how to go about solving for conflicting 
	    distributions such as the one in this example.

     Francoise: Is this ok.
     Ray: I do not think it should be an error.
     Francoise: Not an error, some vendor may provide a warning.
     Bill: The solution is ok, but we need something in LRM.
     Arturo: It does say it. 
     Bill: the question is not so much failure, but what should be the
	behaviour.  Distributions are very wide.
     Ray: This is only the distribution rather than solve before.
     Arturo: This is implementation dependent. Very hard to comeup with
     Ray: The one you do not want to preclude. weighting only applies to when
	you have a solution

     NOTES: Whether an example should be added to the LRM. Clarifying text would 
	be preferable to an example.

	Section 12.4.9, has the following sentence (right above the 
	block for syntax 12-7), "However, a "solve...before..." constraint
	does not change the solution space, and so cannot cause the 
	solver to fail".

     [AI]: Francoise: create a proposed clarification statement(s).


     Errata 261   randcase width rules inconsistent with Verilog, section 12.15
	    What rules should govern the precision of weights within a randcase?

     Steve: text states values are self-determined, comparison, with 
	This would be consistent with ordinary case will be evaluated.
     Bill: If you have arithmetic overflow, it is not direct analogy with
	direct case statement. You are adding each element.
     Ray: That would be the way that normal verilog. 
     Steve: size shall be at least 32bits, integer value at least 32 bits.
     Bill: It does not convert them at the end.
     NOTES: Rules: 
	       - precision should be context-dependent - Verilog
	       - precision is self-determined - LRM


     [AI]: Steve to create an proposal.

     Errata 262   randcase unclear on negative weights, section 12.15
     Ray: They are treated as unsigned.  
     Steve: They will be treated as unsigned. 
  	What about x and z bits, conversion to 2-state,
     Ray: Unimplied conversion. 
     NOTES: It was mentioned that quite some time ago this was discussed in
	    the sv-ec committee. The response at that time was that a 
	    negative weight would show up as a very large positive value.

      [AI] Steven and Bill - make a proposal.

     Errata 263   wide randcase weights, section 12.15
      	How to compose a random value greater than 32 bits by combining multiple	
	calls to $urandom_range().

     Decision: turn this into an enhancement request. Right now it 
	    is up to the implementors to decide how to compose a random number
	    larger than 32 bits.


	(data base ref:  http://www.eda.org/svdb/ )  


5) Next Meeting Monday November 22, 2004.
	11:00 am-1:00pm.
	send the logistics, agenda.

6) Meeting adjourned at: 1:00 p.m.