What Tom suggests is very relevant and why I was concerned about querying individuals on a project. Each individuals opinion will be very different and I'm not sure that individual responses on a project as a whole will give you the data you can analyse.
I immediately threw out critics as unimportant not because that is my perspective (actually that is partly true but not really the reason). It was because if one major part of the analysis is on usage at runtime then the critics will get a major hit on processor time as they run continuously in a background thread. That would weight them more heavily than user interaction with the application. I think it is the code that is triggered as part of user interaction that would be most relevant to the application as a whole. In ArgoUML we refer to "subsystems" (I prefer the term components really) such as critics, diagrams, UML models, code generations, reverse engineering etc. Developers tend to arrive who want to improve one of these different subsystems. Maybe this tool should allow a single project to specify what packages make up the various subsystems and give separate analysis for each of these subsystems. If you're starting with a questionnaire then maybe the first question should be what subsystem do you believe exist and then for each one repeat the question set. >From there I still suspect that taking the results of the tool working without any human input (i.e. no specialist human knowledge of what are the most important class) would be the best first pass. The human factor would then be to remove the classes that cause some anomalous results, either because it runs frequently but does little (which would throw any usage analysis) or because it goes through many changes over time but does little (which would throw any subversion analysis). Asking a developer up-front what they think is most important and using that as input to the tool would most likely thrown the results entirely. If you're asking us so that you can use that to rate the tools results then that is just as difficult. I think at least the breakdown into subsystems would be required. Regards Bob On 26 May 2010 21:47, Tom Morris <[email protected]> wrote: > On Wed, May 26, 2010 at 3:57 PM, Bob Tarling <[email protected]> wrote: > >> I would think such a tool should preferably operate without any >> developer input but could then be tuned. > > My impression was that Alberto's questionaire was to establish a > ground truth of expert opinion against which to measure the results of > his tool(s). This is something that would be done during the > development/evaluation phase of the tool, not something that you'd do > as a matter of course if he were able to come up with suitable > automatic algorithms. > >> When viewing the results of the above the seasoned developwe may spot >> anomalies and then choose to remove items that are not relevant (for >> argouml that would almost certainly be the critics). > > Except for a developer who wanted to extend the critics to cover UML > 1.4 or 2.x or modify the critic infrastructure to operate differently. > That was my point about importance being context sensitive. I agree > that for developers focused on other areas, critics could be largely > irrelevant (although the general knowledge that ArgoUML has multiple > threads of execution is probably important context for all > developers). > > BTW Alberto, I'd avoid the term "hot spots" and stick to "important > classes" or something similar. For me, hotspots is something I > asssociate with performance profiling (which may also be why Bob was > referencing code coverage utilities). > > Tom > > ------------------------------------------------------ > http://argouml.tigris.org/ds/viewMessage.do?dsForumId=450&dsMessageId=2613763 > > To unsubscribe from this discussion, e-mail: > [[email protected]]. > To be allowed to post to the list contact the mailing list moderator, email: > [[email protected]] > ------------------------------------------------------ http://argouml.tigris.org/ds/viewMessage.do?dsForumId=450&dsMessageId=2613797 To unsubscribe from this discussion, e-mail: [[email protected]]. To be allowed to post to the list contact the mailing list moderator, email: [[email protected]]
