#lisp - Mon 26 Mar 2007 between 00:00 and 00:16

NY Lost Funds



thomas001but i thought you wanted useful apis?
starseekerthomas001: Exactly. What I'm wondering is if that can be done in such a way as to respect "natural" conventions of a given language
thomas001starseeker: how whould you translate macros to java? ;)
starseekerthomas001: Right. So what constitutes a "useful" api in a given language, and given some specification of functionality can such an api be autogenerated?
thomas001or method combination
starseekerthomas001: No idea. It's what eructate said that I'm wondering about - a meta-semantic generalization of program logic
thomas001starseeker: imo writing really good APIs in any language requires much experience, i don't think it can be auto generated,but thats just my opinion
also people once believed you can't go faster than 30 km/h
starseekerthomas001: Heh :-)
thomas001: So I take it ORB and ILU are not resolutions of that problem?
thomas001i don't think so
RiastradhI think that, although generating interfaces completely automatically from C headers is bound to lose, it would be conceivable to have tools by which a programmer can direct the generation of interfaces.
I don't have anything more specific in mind, however, so I'm not prepared to defend that claim.
eructatestarseeker: the meta-semantic generalization is called the Plan Calculus. I'm sure there is information about it on the intarweb.
thomas001Plankalkl? ;)
starseekereructate: Is this it? http://www.yann-gael.gueheneuc.net/Work/Publications/Documents/020523+The+Programmers+Apprentice.ppt.pdf
actionstarseeker concedes that if it WERE possible someone way smarter than him would have already done it...
starseekerconcedes that if it WERE possible someone way smarter than him would have already done it...
eructatethat is a slideshow.
starseekerBut the Programmer's Apprentice - that's what you mentioned earlier?
eructateyes, the book that that is a report about, is the one that described the MIT research
thomas001starseeker: perhaps it whould be less work to just handcode the api wrappers,than to write such a generator
(which generates equal good code)
eructatehttp://ccs.mit.edu/papers/CCSWP193/main.html
that is a thesis about software components, which is also what you seem to be asking for
starseekerthomas001: In a specific case, that's true. But if it CAN be done in general, it means being able to quickly use all of the pre-existing code out there in new environments with minimal pain of re-implementation
eructate: Thanks! good stuff
thomas001starseeker: you still need to have all code export its interface in a useful way
generation from a plain c header is near to impossible
(to a lispy interface)
eructatehttp://ccs.mit.edu/1996wp.html&WP193
thomas001404
eructateI forgot how you spell "anchor" in urls.
it's # of course.
i just thought the abstract would be useful
http://ccs.mit.edu/1996wp.html#WP193
starseekerfound it :-)
thomas001: It's probably true that there are fundamental limits, but I'm curious as to what they are. The biggest problem with introducing new software to any environment nowadays is being able to support legacy systems and functionality flawlessly - arguably that's why OS/2 didn't succeed, for example - it couldn't run old DOS software well.
thomas001: If the code for old functionality can be abstracted as some kind of logic (at least in communicating with it, if nothing else) and mapped to the new system, it might allow new systems to quickly gain or retain old functionality that is critical
thomas001starseeker: "can be abstracted"...automatically?
starseekerHopefully

Page: 2 9 16 23 30 

IrcArchive

NY Lost Funds