Some of the most common operations used in the Scheme language are those
transforming lists: map, filter, take and so on. They work well, are well
understood, and are used daily by most Scheme programmers. They are however not
general because they only work on lists, and they do not compose very well
since combining N of them builds
(- N 1) intermediate lists.
Transducers are oblivious to what kind of process they are used in, and are composable without building intermediate collections. This means we can create a transducer that squares all odd numbers:
(compose (tfilter odd?) (tmap (lambda (x) (* x x))))
and reuse it with lists, vectors, or in just about any context where data flows in one direction. We could use it as a processing step for asynchronous channels, with an event framework as a pre-processing step, or even in lazy contexts where you pass a lazy collection and a transducer to a function and get a new lazy collection back.
The traditional Scheme approach of having collection-specific procedures is not changed. We instead specify a general form of transformations that complement these procedures. The benefits are obvious: a clear, well-understood way of describing common transformations in a way that is faster than just chaining the collection-specific counterparts. For guile in particular this means a lot better GC performance.
Notice however that
(compose …) composes transducers
left-to-right, due to how transducers are initiated.
|• SRFI-171 General Discussion||General information about transducers|
|• SRFI-171 Applying Transducers||Documentation of collection-specific forms|
|• SRFI-171 Reducers||Reducers specified by the SRFI|
|• SRFI-171 Transducers||Transducers specified by the SRFI|
|• SRFI-171 Helpers||Utilities for writing your own transducers|