Hey Christoph,

I also have a few comments. I initially thought this could have potential.
But, I think it introduces too much complexity.

1) We already sort of have a BaseSensorDataType. The SensorDataEntry
provides absolute mandatory fields like timestamp, tool, etc.

2) I'm not sure what this Hierarchy would accomplish. Even a
LatexFileMetric SDT and JavaFileMetric SDT would have to have a file
attribute. And we could place all the Latex and Java specific things in the
PropertyList (see eSDT). The bottom line is that after getting the data
from the SensorDataEntry there is _no_ difference in how the semantics of
the SDT are organized.

3) The fancy thing that you've mentioned is pretty cool. That is why I
thought this has potential. Here's an example why this would be cool.
Currently, if we wanted to run an analysis on strictly FileMetric data from
Latex, then we would have to search through all FileMetric instances. In
other words, the indexing of Sensor Data is only limited to the SDT
(FileMetric, UnitTest, etc), User, and Day. There is no way to look up the
type of FileMetric data you want. This could be a problem in the future if
we ever generate DocBook FileMetric or JSP FileMetric. Or, we could do an
analysis over all types of FileMetric data, because they inherit from the
same parent.

However, I don't think we really need to introduce Hierarchical SDTs to
accomplish what you explained. Couldn't we just index the SDT entries by
Type (FileMetric, UnitTest, Coverage, etc), User, Day, and Tool? Maybe that
doesn't work. How about Type, User, Day, and some other attribute that
indicates the Subtype. You could get the specific entries for a subtype or
get all types.

To conclude. I like the benefits of your idea. But, I'm not sure I agree
with changing the SDTs to get them. Imagine the horror we would live in if
we had 50 SDTs! Imagine writing the DocBook for that.

thanks, aaron

At 11:59 PM 3/14/2005, you wrote:
Hi.

Following idea:
As we want to introduce these evolutionary sensor data type, which is a
lot of work and a major refactoring of the whole sensor data type thing,
how about including SensorDataType hierarchies in that update.

So my idea is defining inheritance mechanisms for data types.

So, for example, there could be a BaseSensorDataType, which has only the
most common attributes (like timestamp or origin).
Then there could be a FileMetricSensorDataType, which is a subtype of
BaseSensorDataType, containing very general filemetric attribute. Then
you could define things like a JavaFileMetricDataType or a
LatexFileMetricDataType which are subtypes of FileMetricDataType.

The fancy thing would be that you should be able to do analysis over all
FileMetricDatas in a general matter by casting them to the
FileMetricSensorDataType supertype, or do a fine grained, tool based
analysis by using the tool subtypes.

So in the end, we would have some kind of fancy sensor data type
inheritance tree (just like in Java).

I just thought about that; I don't know if you think if this might be
cool. Neither do I know if the rework effort needed for that is
justified by just the coolness factor.

Cheers,
Christoph

Reply via email to