Gordon Vreugdenhil wrote: > > Mark Hartoog wrote: > > I am confused by what the issue is here. > > It is clarity of the text for the non-implementor. > > > Compile time resolution is always in terms of the statically declared > > fields and methods. Some methods may be declared virtual in the static > > class. > > This means that at compile time you resolve to the virtual > > method. > > All type checking is done against the virtual method. The actual method > > that is called is selected at run time. Data field members are never > > virtual. > > > > In your example the 'x' in the randomize with block resolves to the int > > 'x' in > > class 'C'. > > > Sure, you and I and other implementors know that. I'm certainly > not suggesting any other binding. > > > What the text doesn't say is: > 1) that the inline constraint is effectively its own non-virtual > method rather than being "part of" randomize. Without > such a statement the only "this" that a reader could assume is > the "this" of the randomize method that is *actually called*. > That type is not known. > 2) that the class in which the "method" lives is the class > defined by the type of the reference > That language is not the clearest explanation (i.e., what does "lives" mean). Obviously, an inline constraint is unique and we run into problems when trying to define it in terms of other constructs. I think we should only attempt to state its semantics unambiguously. > The current approved 1858 text talks about "the randomize...with object" > and NOT the type of the reference. That could easily lead readers > to assume the highly dynamic behavior. > Actually, the text talks about "the randomize...with object class" and that language can certainly be made more explicit to mean the class of the object used in the call to randomize (the static type or the type of the reference). That has always been the intent of that verbiage. The LRM and the approved proposal do not include language that mention the randomize method for name binding - that issue was raised by Ray who pointed out that "this" is not part of the object scope but is available only in non-static methods. > By the way, you certainly cannot tell the type of the class reference > at compile time -- if the prefix is a general hierarchical reference > or similar then you can't possibly know until elaboration time. So > where the names bind is certainly an elaboration time decision > based on the static type of the reference; the non-virtual method > model is just a convenient way to talk about the effect. > True, but that is not unique to in-line constraints. The type of any hierarchical reference is not known at parse time and needs to be resolved after elaboration. I believe by compile time Mark meant to include elaboration, in much the same way as the LRM distinguishes only compile-time and run-time. > Gord. > Since at this point, we seem to agree on what the rules are (please correctly if that is not the case), shouldn't we concentrate on the verbiage that removes any potential ambiguity. I propose the following: The scope for resolution of variable names referenced in the in a constraint block begins with the randomize()...with object class, i.e., the class of the object used in the call to randomize. If a name fails to resolve within the randomize()...with object class, the name is resolved normally starting in the scope containing the inline constraint. If this is acceptable, I'll incorporate it into the proposal. Otherwise, would you please suggest a better alternative. Arturo > > > > > > -----Original Message----- > > From: owner-sv-ec@eda.org [mailto:owner-sv-ec@eda.org] On Behalf Of > > Gordon Vreugdenhil > > Sent: Tuesday, December 18, 2007 1:50 PM > > To: SV_EC List > > Subject: [sv-ec] "this" and member resolution in inline constraints > > > > > > In considering some of the wording for the 1858 (inline constraints name > > resolution) issues, I would like to raise an additional concern and/or > > need for clarification. > > > > "randomize" is definitely a virtual method. In the discussion, we were > > talking about "this" as following the normal method resolution rules for > > "this" references. That does pose a problem when everything is > > composed. > > > > Consider this case: > > > > class C; > > rand int x; > > endclass > > class D extends C; > > rand bit[1:0] x; > > endclass > > D d = new; > > C c = d; > > > > initial c.randomize with { x > 0 && x < 10; }; > > > > My working assumption has always been that the resolution of "x" is > > based on the static type of the object prefix, but I think that the > > wording and discussion related to "this" has at least muddied the waters > > if not compromised that interpretation completely. > > > > If I have: > > initial c.randomize with { this.x > 0 && this.x < 10; }; the > > meaning of "this" is now different. We would have to interpret "this" > > not as the formal of the virtual method "randomize" > > but as "a reference to the portion of the object addressable by the > > static type of c". > > > > There are many, many issues with a dynamic interpretation of "this" so I > > would not want to propose that we follow the dynamic resolution path. I > > believe that the statically typed interpretation I have above is > > consistent with Arturo's view. > > > > Since "this" would not follow the normal semantics of a virtual method, > > I think that this needs to be reexamined or clarified as part of 1858. > > > > If the more static view of name binding is followed, there are still > > potential questions related to handling of other virtual method > > references within a constraint and the overall coherency of object > > assumptions in such cases. > > > > Gord. > > -- > > -------------------------------------------------------------------- > > Gordon Vreugdenhil 503-685-0808 > > Model Technology (Mentor Graphics) gordonv@model.com > > > > > > -- > -------------------------------------------------------------------- > 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. -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.Received on Wed Dec 19 10:05:02 2007
This archive was generated by hypermail 2.1.8 : Wed Dec 19 2007 - 10:05:32 PST