Buffered transforms. A buffered transform represents a method for converting a stream of inputs to a stream of outputs. However, inputs can also be buffered, i.e. you can feed inputs to the transform and pull out outputs later. There is no requirement that 1 input produces exactly 1 output. It is common that multiple input values are needed to construct a single output, and vice versa.
Buffered transforms serve as a general method for working with streams of data and flexibly composing mappings from inputs to outputs. The buffering aspect supports asynchronous programming interfaces. Parsers and printers throughout Biocaml are implemented with this module whenever possible.
Often mappings need to account for errors, e.g. an input string cannot be converted to an integer. Several methods below explicitly support buffered transforms where the output type is a result.
val make :
?name:string ->
feed:('input -> unit) ->
next:(bool -> [ `output of 'output | `end_of_stream | `not_ready ]) ->
unit ->
('input, 'output) tmake ~feed ~next () creates a transform that can be fed with feed and read from with next.
feed input should store input in a buffer, which is presumably a shared state also available to next.next stopped should remove values from the buffer, convert it to an `output and return this output, or return `not_ready if there are not enough buffered inputs to create an output value, or return `end_of_stream if the buffer has been stopped, as determined by the supplied argument, and there is no more input.Depending on the specifics of the transform, it may be the case that the buffer has been stopped but there is not enough input to create an output value. It is the caller's choice how to handle this or any other kind of error, e.g. make the return type a result.
name an optional name for the transform that will be used in error messages.val feed : ('input, 'output) t -> 'input -> unitfeed t i stores i into the buffered transform.
val next :
('input, 'output) t ->
[ `output of 'output | `end_of_stream | `not_ready ]next t returns an output value if possible, `not_ready if t needs to be fed with more input before it can produce an output, or `end_of_stream if t has been stopped and has no more data.
val stop : ('input, 'output) t -> unitstop t declares t to be stopped, which means subsequent calls to:
feed t _ will raise Feeding_stopped_transform. Feeding a stopped transform is not allowed.next t will eventually return `end_of_stream, not necessarily the immediate next call as there may still be buffered values available for output.val name : ('input, 'output) t -> string optionname t returns the name of t.
val identity : ?name:string -> unit -> ('a, 'a) tidentity () returns a transform that simply returns its inputs as outputs without modification (it can be seen as a simple Queue.t).
val of_function : ?name:string -> ('a -> 'b) -> ('a, 'b) tof_function f is like identity () but the transform outputs are passed to the function f.
to_stream_fun t returns a function f that behaves like t but the inputs and outputs are on standard OCaml streams.
val in_channel_strings_to_stream :
?buffer_size:int ->
Core_kernel.In_channel.t ->
(string, 'output) t ->
'output Stream.tin_channel_strings_to_stream ic t returns a stream of 'outputs given a transform t that knows how to produce 'outputs from strings. The strings are read from the in_channel.
val stream_to_out_channel :
'input Stream.t ->
('input, string) t ->
Core_kernel.Out_channel.t ->
unitstream_to_out_channel xs t oc consumes a stream of 'inputs using t to transform them into strings, which are then written on the out_channel oc.
Buffered transforms are mutable and one should not expect nice mathematical properties from composing them. The intention here is to provide building blocks that allow the creation of more complex transforms from simpler ones. Only the final resultant transform should be used. Feeding/reading the transforms being composed is likely to lead to violations of the stated behavior of the above operations.
on_input f t returns a transform that converts its inputs with f and feeds the results to t.
on_output t f returns a transform that behaves like t except the outputs are first converted by f.
compose t u composes t and u. figure src/doc/figures/transform_compose.svg 50%
“Compose” two transforms
val mix :
('a1, 'b1) t ->
('a2, 'b2) t ->
('a1 * 'a2, [ `both of 'b1 * 'b2 | `left of 'b1 | `right of 'b2 ]) tmix t u returns a transform that takes as input a pair of the inputs expected by t and u, and outputs either both outputs, or, when one transform has reach the end of its stream, the output of the remaining one (as `left _ or `right _). figure src/doc/figures/transform_mix.svg 50%
“Mix” the inputs of two transforms
val filter_compose :
('il, 'ol) t ->
('ir, 'our) t ->
destruct:('ol -> [ `transform of 'ir | `bypass of 'filtered ]) ->
reconstruct:([ `bypassed of 'filtered | `transformed of 'our ] -> 'result) ->
('il, 'result) tfilter_compose t u ~destruct ~reconstruct produces a transform that feeds a filtered subset of ts outputs to u. Only those outputs ol of t for which destruct ol returns `transform are passed on to u. The filterd out values are combined with u's output using reconstruct. figure src/doc/figures/transform_filter_compose.svg 50%
“Compose” two transforms with a filtering function
val split_and_merge :
('il, 'ol) t ->
('ir, 'our) t ->
split:('input -> [ `left of 'il | `right of 'ir ]) ->
merge:([ `left of 'ol | `right of 'our ] -> 'output) ->
('input, 'output) tsplit_and_merge t u ~split ~merge returns a transform whose input is split using split, passing the result either to t or u, and then the outputs of t and u are combined using merge. There is no guarantee about the order in which the inputs are fed to t and u (it depends on the buffering done by the individual input transforms). figure src/doc/figures/transform_split_merge.svg 50%
“Split” the inputs of two transforms and “merge” the outputs
Operations analogous to those above, but for transforms whose output types are results.
val make_result :
?name:string ->
feed:('input -> unit) ->
next:
(bool ->
[ `output of ('a, 'b) Core_kernel.result | `end_of_stream | `not_ready ]) ->
unit ->
('input, ('a, 'b) Core_kernel.result) tval on_ok :
('input, ('ok, 'error) Core_kernel.result) t ->
f:('ok -> 'still_ok) ->
('input, ('still_ok, 'error) Core_kernel.result) tLike on_output but on the successful part of the output.
val on_error :
('input, ('ok, 'error) Core_kernel.result) t ->
f:('error -> 'another_errror) ->
('input, ('ok, 'another_errror) Core_kernel.result) tLike on_output but on the erroneous part of the output.
val compose_results :
on_error:([ `left of 'error_left | `right of 'error_right ] -> 'error) ->
('input_left, ('middle, 'error_left) Core_kernel.result) t ->
('middle, ('output_right, 'error_right) Core_kernel.result) t ->
('input_left, ('output_right, 'error) Core_kernel.result) tcompose_results t u is like compose but for transforms returning results. The on_error function specifies how errors in t or u should be converted into those in the resultant transform. figure src/doc/figures/transform_compose_results.svg 50%
“Compose” two transforms which may fail
val compose_results_merge_error :
('a, ('b, 'el) Core_kernel.result) t ->
('b, ('d, 'er) Core_kernel.result) t ->
('a, ('d, [ `left of 'el | `right of 'er ]) Core_kernel.result) tLike compose_results but with a pre-specified on_error function.
val compose_result_left :
('input_left, ('middle, 'error) Core_kernel.result) t ->
('middle, 'output_right) t ->
('input_left, ('output_right, 'error) Core_kernel.result) tLike compose_results but only the first transform returns results. figure src/doc/figures/transform_compose_result_left.svg 50%
“Compose” two transforms when only the first one may fail
class type ['input, 'output] object_t = object ... endGeneric transform type.
val make_general :
?name:string ->
next:(unit -> [ `output of 'output | `end_of_stream | `not_ready ]) ->
feed:('input -> unit) ->
stop:(unit -> unit) ->
unit ->
('input, 'output) tThe most general way to make a transform. All make functions above are implemented with this one.