> Another problem is, of course, that class diagrams
> (such as popularized
> by OMT, Booch, UML, you name it) are inherently unable to
> meaningfully deal with
> parameterized classes, at least with C++'s implementation of this concept.
>
>
I admit I don't have a clear view on this. However, I sort of think that for
leaf-classes, it is perfectly possible to have a clear inheritance diagram
(after all, if there isn't a clear inheritance tree, it won't compile...).
Indeed, at the end of a hierarchy tree, all 'intermediate' templates have to
be resolved (i.e. the only ones left are the template parameters of the
class itself).
The question really is what to do with 'intermediate' classes.
One option would be to do the same as for a 'leaf' class, i.e. you make the
inheritance tree only upto the current class, and do not show whatever is
below it (so, your intermediate class IS a leaf class). It looks like the
dot-generated hierarchies for leaves are already pretty close to this (my
previous mail was a counter-example, but I'd hope it's not too difficult to
take this into account as well).
Another option is to do show whatever comes lower but only heuristically,
i.e. without looking at template arguments at all. One could look at
doxygen's current attempt in this way. As long as this is stated clearly, I
could live with it.
I guess the latter approach could be improved upon by some kind of
postprocessing of the hierarchies: the hierarchy for any intermediate class
could be found by merging the hierarchies of all leaf classes. This doesn't
sound easy though.
What about the 'class hierarchy' list ? It attempts to start at the 'root'
of the hierarchy. This is difficult/impossible for exactly the same reasons
as before. Again, I could live with it if the doc states it somewhere
explicitly. I could imagine doxygen finding out it has troubles with the
templates, so inserting a warning on top of the class hierarchy page.
Summary: an inheritance diagram for anything upwards in the tree should be
doable. For anything downwards, it's getting muddy. The dot-generated
hierarchy seems pretty close, so maybe a few extra fixes make it usable
enough.
(Still beats me how the 'default',i.e. non-dot class hierarchy diagrams can
be different from the other ones. It seems to say that there are 2 pieces of
code to do the same thing in doxygen).
Kris Thielemans
Imaging Research Solutions Ltd
Cyclotron Building
Hammersmith Hospital
Du Cane Road
London W12 ONN, United Kingdom
Phone on : +44 (020)8383 3731
FAX on : +44 (020)8383 2029
web site address:
http://www.irsl.org/~kris
|
