[Qt5-feedback] QAIM::headerData()

[Charley Bay]

André spaketh:

> I think it is a mistake in the current QAIM design, that all cells have a
> separate QModelIndex, and that all cells can have multiple rows and columns
> of children, and that the number of columns does not need to match between
> them. It is extremely flexible, but it complicates the structure needlessly.
> The Qt model/view structure needs, in my opinion, not cater to the very
> exotic situation where people want very complex views with multiple levels
> of headers and the likes. Let that be the domain of 3rd party components;
> they are available if you need them. Let us not reinforce that design
> mistake by making headerData more complex than it needs to be.
>
> I vote against.
>

IMHO, this goes to the crux of the issue:  Use cases like "file system
models" include different types in the model (files, directories), and
different views are sometimes interesting (e.g., the current Windows
explorer lets you show different header views in different directories, such
as "thumbnails" or "details", etc.)

I think we'd agree that the QAIM design is complex.  I don't criticize that
-- it's a Hard Problem.  However, I think the issue is that QAIM *itself*
has not yet decided how it wants to support models of different *object
types* within it.

Only the QModelIndex *assumes* the *same type* for all items in the model.
 Every thing else is agnostic to what types-nest-in-types for a model.  That
design makes QAbstractItemModel (much) more complicated.  A possible
point-of-debate is the header data (for whether it should be constant across
all items in the model, which is this thread).

IMHO, if the goal is for people to merely create new models when they have
different item-type-demands (such as need for different header data for
different nodes), then the QAIM design could be *massively* simplified to
make that type of information constant, making the interface much more
trivial, and making it *far* easier to integrate into projects.

I understand in the Qt5 timeframe probably nothing can/will be done.
 However, I vote QAIM decide whether it supports models with heterogeneous
types, or not, and *picks* one.  I'm a general supporter of its use and
design, but IMHO it currently doesn't do either job very well (too
complicated for single-type-models, too restrictive for multi-type models).

--charley
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.qt.nokia.com/pipermail/qt5-feedback/attachments/20110726/e6d07800/attachment.html