Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!hc!lanl!opus!ted From: ted@nmsu.edu (Ted Dunning) Newsgroups: comp.sw.components Subject: Re: Ted Dunning's flamage (dousing) Message-ID: Date: 8 May 89 19:47:32 GMT References: <5421@hubcap.clemson.edu> Sender: news@nmsu.edu Organization: NMSU Computer Science Lines: 106 In-reply-to: billwolf%hazel.cs.clemson.edu@hubcap.clemson.edu's message of 7 May 89 23:06:26 GMT In article <5421@hubcap.clemson.edu> billwolf%hazel.cs.clemson.edu@hubcap.clemson.edu (William Thomas Wolfe,2847,) writes: ... Not some Madison Avenue hype, but the actual project objective: ... The objective of the Reuseability Library Framework project is to develop this essential knowledge-based foundations technology for building _intelligent_librarians_, and to demonstrate the use of this technology by building an example library for the domain of ... On the contrary: The initial librarian design will be a prototype that supports interactive search through a particular library taxonomy. Component ... note the sudden shift from the past to the future tense. ... [ with regard to the automatic test plan generator ] It's a mechanism: [The test plan assistant] is motivated by two factors, the need to support black box testing of Ada modules, and the parts engineering and qualification process for Ada parts to be inserted into reuseability libraries. [It] requires a network-based structural model of an Ada subprogram interface, and using actions provided by AdaTAU rule bases, queries the user for implicit assumptions about the unit under test (such as parameter interdependencies). [It] produces a structured set of test specifications where the set has been pruned by built-in heuristics and user-specified choices. In addition to a direct representation of the Ada type hierarchy as a semantic network via AdaKNET, [it] will utilize testing heuristics associated with various data types, as well as general rules drawn from the experiences of veteran Ada programmers. The knowledge base built to support [the test plan assistant] should be reuseable by other "smart" tools, and [the test plan assistant] will itself be absorbed in the Librarian systems where additional semantic information may be available from the library domain model. this is fascinating; a test plan generator that doesn't need to know what the unit under test is SUPPOSED TO DO. this fits in well with the experience with the software qualification of magellan and space telescope. ... > > the implication that the qualifying librarian and the test plan > assistant are automated. this is bullshit. No, this is fact. see above. on the other hand, if the qualifying librarian and the test plan generator can independently and automatically qualify and generate tests for arbitrary ada software, then they are automated. if, on the other hand, they can't, then they are streamlined manual procedures. ... Both semantic nets and rule-based systems have been around for years, and thousands (if not millions) of them exist. Your comments are indicative of the great extent to which you are unaware of the field. (is this what is meant by ad hominem attacks?) both semantic nets and rule-based systems have been around for years, and they still have the same severe limitations which prevent them from doing more than very limited, well specified tasks. the task of generating complete test suites is something that expert people still cannot do very well, much less something that expert systems can do. likewise with deriving useable taxonomies in novel domains. current ai approaches work well in limited domains, or in toy applications (credit apps, translation of weather reports). they work less well (but still useably) where the situation is basically somewhat open-ended, but enormous effort can be applied over a period of years (technical translation using systran is the canonical example). in situations requiring open-world reasoning of any sort, they still basically don't function. ACE may well be a significant advance in the state of the art in software development, but overselling ai techniques, and overdescribing systems that may in some degree utilize them is very counter-productive. (for one thing, it brings flamers out of the woodwork). ... Essentially because that's not true. I have used ACE; it's much more like an Ada-oriented operating system environment, running on top of something much less appropriate to the developer's needs (e.g., Unix). ahhhh..... now this sounds more like what the system really is. with this approach, we can talk less in terms of buzz-words, and more in terms of how dynamic the development cycle can be made and whether test-stubbing and pre-testing of components can be supported. let's hear from all the satisfied users of ACE.