Notation layout, staff types etc
Mail archive thread link (a long discussion about potential future changes)
See also Notation editor improvements.
See also Score layout, not just notation layout.
Note: this is not a description of code actually in Rosegarden, it's part of a discussion about future work. That particular discussion didn't really go anywhere, but it's possible material for later.
On Tuesday 21 Mar 2006 16:00, Michelle Donalies wrote:
A good summary, thanks.
So, how do these differ?
The first thing that strikes me (given the recent thread, and my comments in it) is that there are not many fundamental differences between how these different staffs handle horizontal layout.
I clipped some of your descriptions above, but I didn't see anything substantial in there that couldn't be explained as “one staff shows event type X but the other one doesn't”. Since a staff already gets to choose what events it selects from the underlying segment (via Staff::wrapEvent), this suggests that nothing outside of the staff itself should actually care whether a staff has (say) clefs and keys in it or not. Time signature and bar lines (which are not events) are a different matter, but it ought still to be a simple question of testing a flag to determine whether to lay them out or not, for any staff type.
However, there are certainly differences in implementation in Rosegarden, many of which are caused by the current implementation of NotationHLayout doing more than it should. These are examples of things that appear to me to be really different and potentially problematic:
1. Vertical layout. NotationVLayout could certainly be done by a staff type rather than a layout class. Different staffs have quite different vertical layout techniques and there's no call to reconcile between them. (btw I'm also not sure that the NotationView's current code to position the staffs on the page will do the right thing if the staffs are of unequal heights.)
2. Accidentals. The current HLayout manages which of these are shown and how they are spaced based on the current clef and key, which of course may differ or disappear in other staff types. This is confined to the HLayout::scanStaff area and would be easy to suppress, but it would not be so tidy a thing to do as just suppressing the appearance of clefs and keys for example.
3. The HLayout does several things which are not strictly its job at all, such as assigning stem lengths for beamed groups. (Well, it doesn't do these things itself, but it does cause them to happen.) It does so just because it's in the right place: after vertical layout but before horizontal positioning is finalised. A more correct implementation would probably have an intermediate pass between vertical and horizontal layout for these. This is again a fairly small amount of code in the HLayout class, but it's untidy.
4. Properties calculated in the layout phase are stored in the Event object, with names unique to the NotationView instance so that different views can do different layouts (e.g. in different sizes, or one in a page layout and another in a linear one). However, these are View-specific, not Staff-specific. If it became possible to have the same event on more than one staff in the same view at the same time, this would break.
I don't see any obvious disadvantages to refactoring 1, and I think 2 and 3 are both candidates for things that could usefully be refactored to improve the code even if we had no intention of adding any more staff types in the first place, although there may be some tricky details. Also I'm not especially happy with the way that beamed groups and tuplets are handled anyway. (Our reconciliation of simultaneous events on different staffs is quite good though, which is partly why I took badly to your observation that doing anything better than selecting the widest minimum spacing was “overkill” – it isn't at all, it makes a big difference and we do far better than that now.)
If you're thinking about that, then you might perhaps also think about multi-voice staffs, for which the general underlying thought has been that the user may choose optionally to show different segments (even on different tracks) as different voices on the same staff, via some sort of part arranger interface. Most of the basic layout code supports this already.
[ separate email, actually earlier ]
The original idea for the HLayout and VLayout classes was to use something like the “visitor” design pattern to separate out the layout logic from the staff container or the low-level drawing functions (both of which are either in, or invoked from, the staff class). This is not so much because you might want to use more than one type of layout for a given staff, as to keep the layout-related reasoning separate for code management purposes.
The fact that the things we want to do next aren't easily possible using the current classes implies that something is wrong, of course, but I'm not sure that making the staff class so much more complicated is a good idea on the road to fixing it.
Attached is a “whiteboard diagram” I sketched on paper (you can gather that I'm still not getting much time at the computer to think about these things at the moment). This shows the stages of the layout procedure (from memory, so doubtless some omissions) with the classes that currently carry them out. Start reading at top-left; the vertical division is broadly into single-staff layout at top, multi-staff reconcile and layout etc in the middle, drawing at the bottom.
So, for each of the steps on this diagram, which is the most appropriate class to do the work? Assuming for the moment that there's no obligation to use the existing class structure (there is an obligation effectively to use the existing _code_, but it can be rearranged at will). It seems that the existing classes mostly do multiple tasks that are not necessarily closely related but that share some of the same data or domain knowledge (e.g. the HLayout class doing things with beamed groups and accidentals), and there may well be a case for pulling these things out into more than one class rather than just changing which of the existing classes they go into.