This enables callers to opt out of automatic data wrapping in any case.

Also, use Reflection instead of instanceof, because that invokes autoload.

Say goodbye to the random background colors, and hello to different connector dots.
Hopefully we can add JS connection lines sometime as well.

Prevents constant slot disconnections by the pipeline updater.

Patch by sgilits: Remove unnecessary loop, and fix connection removal when only a single operation existed in a pipeline.

Plus a few minor rewordings in the file operation descriptions.

This should solve the problem of orphan connections, and therefore
enable me to finally roll a new release without data inconsistencies.

Further related stuff also present in this commit:
- Fix wrong order of array_diff() arguments.
- Call TfOperationSlotListener::observedInputChanged()
  from TfOperation::inputChanged(), so that propagation actually works.

Use with care, as here lie dragons.

They are already assigned by TfPipeline::createOperation(),
which the pipeline operation uses to instantiate its operations.

#502054 by sgilits, with modifications: Don't allow certain slot combinations to be connectable with JS.

acting as pseudo super-type for strings and numbers.

This input enables you to unnest the foreach operation's
output structure and only return a single output value per list item.

This is a major step for usability. Not only is it now possible to
drag and drop operation slots and the operation blocks themselves,
but also the layout of the non-JS version has been revamped and
rocks a lot more than it did before. (Bye, bye, submit buttons.)

This is an optional feature which only gets enabled if the
dependencies are satisfied. Currently depends on jQuery UI 1.7
and thus jQuery Update 6.x-2.x which is currently only available
as unstable -dev version, I'll look into lowering the requirements.

This is a fix for a bug that was introduced during the conversion
to variable input schemas.

That's because TfOperation's initilize() implementation will call
inputs() and outputs() where that state might already be required.

This is done by means of a considerable refactoring, and enables us to
completely remove type options in favor of a purely input-based solution.
Where type options were limited to a set of string or integer options,
inputs enjoy the full flexibility of data wrappers and data input widgets.

A large part of the refactoring concerns output schema listeners, which
have been expanded in scope and are now called operation slot listeners,
as they can react to input, input schema and output schema changes.
(This will also make it possible to properly update pipeline connections
when input/output slots change, but that's still to be implemented.)

Also, a few bugs have been fixed in the process (and a few others have
probably been introduced, let's see how it works out in the end).

Yay for improving flexibility while lowering complexity.

Inheritance is planned and prepared for, but not yet implemented.

Overall, this should make type features feasible to implement -
schemas were just too complex to get us any real-life benefit.

* use jQuery's $.each() instead of JS 1.6 for each()
* not only highlight the corresponding slots but also the currently hovered one

On the one hand, this makes it clear that it does not expose any UI to
the end user (in fact, not even to the admin) and on the other hand this
makes it easier to refer to the API module directly.

instead of throwing an exception. Same effect, nicer presentation.