[Adobe-source-devel] reverse fullorder for forest -

[Sean P. <sp...@ad...>]

A little more on the great forest cleanup... since nobody responded  
to my last note, depth is now moved to an adapter.

Next item, I don't like the interface around edges:

bool leading() const;
void set_leading(bool);
bool pivot();

My first thought was that there are really any invariants being  
protected here - and just expose leading as a data member:

bool leading;

That, however, doesn't work well with the adaptors -

Second thought:

bool leading() const;
bool& leading();

This fails with the reverse iterator though were toggling leading  
_is_ more complicated than toggling a bool (which gets us back to a  
set_leading() interface).

One option is to nuke the reverse fullorder iterator (Eric?) - you  
could still use a general reverse iterator adaptor, but you would  
lose the ability to toggle edges which in reverse is a questionable  
thing to do (it base iterator does a very odd flip around the pointed  
to node).

Another thing I somewhat dislike is the use of a bool (maybe it  
should be called edge and return an enum?) -

if (i.edge() == adobe::forest_leading_s)...

I like the idea but dealing with enum types is just plain cumbersome...

Another option that I'm somewhat fond of is to simply use leading_of  
and trailing_of as stand alone functions - then maybe have an  
is_leading and is_trailing for completeness-

bool is_leading(iterator);
bool is_trailing(iterator);
iterator leading_of(iterator);
iterator trailing_of(iterator);

if (adobe::is_leading(i)) i = adobe::trialing_of(i);

I can't find a case where it's very desirable to actually pivot -  
usually you want to just force to one side or another...

This could also be combined with the previous - something like:

is_edge<adobe::forest_leading_s>(i);
i = edge<adobe::forest_trailing_s>(i);

Anyone have suggestions?

Sean