Pipeline Toolkit

A small functional pipeline toolkit for Python.

pipeline-toolkit provides a simple way to build sequential pipelines from ordinary Python callables.

It also provides small utilities for composing functions, configuring callable steps, applying side effects, and managing pipeline state.

Features

• Sequential functional pipeline execution.
• Asynchronous execution using a worker thread.
• Positional and keyword arguments for pipeline steps.
• Configurable pipeline defaults.
• Stop, skip, wait, and rerun execution.
• Context manager support for automatic pipeline execution and cleanup.
• Manual synchronous step execution.
• Optional execution delays.
• Initial cancellation window before the first step when a delay is enabled.
• Configurable daemon worker threads.
• Optional stop-on-error behavior.
• Optional exception re-raising from worker threads.
• Result and error history using a stack.
• Pipeline modification with add(), insert(), pop(), and clear().
• Sequential callable composition with compose().
• Configurable callable steps with pipe and step.
• Step export and unpacking support.
• Side-effect operations with tap().

Installation

Install from PyPI:

pip install pipeline-toolkit

Or install directly from GitHub:

pip install git+https://github.com/Hoang-Long2012/pipeline-toolkit.git

Quick Start

A pipeline is created from an iterable of steps.

Each step is a tuple whose first item is a callable.

from pipeline import Pipeline

def add(value, amount):
	return value + amount

def multiply(value, factor):
	return value * factor

pipeline = Pipeline([
	(add, (5,)),
	(multiply, (2,)),
])

pipeline.run(10).wait()

print(pipeline.results.get())

The execution flow is:

10
 ↓
add(10, 5)
 ↓
15
 ↓
multiply(15, 2)
 ↓
30

The final result is 30.

Pipeline Steps

Each step can use one of four supported forms.

Callable only

(function,)

The callable receives the previous result:

pipeline = Pipeline([
	(str.upper,),
])

pipeline.run("hello").wait()

The step is executed as:

str.upper(previous_result)

Positional arguments

(function, args)

where args is a tuple:

pipeline = Pipeline([
	(add, (5,)),
	(multiply, (2,)),
])

A step such as:

(add, (5,))

is executed as:

add(previous_result, 5)

Keyword arguments

(function, kwargs)

where kwargs is a mapping:

pipeline = Pipeline([
	(pow, {"exp": 2}),
])

The step is executed as:

pow(previous_result, exp=2)

Positional and keyword arguments

(function, args, kwargs)

For example:

pipeline = Pipeline([
	(my_function, (1, 2), {"option": True}),
])

The callable receives the previous result followed by the supplied positional and keyword arguments.

Execution

run()

Start the pipeline asynchronously.

pipeline.run(default, delay=0, daemon=False, stop_on_error=True)

The initial result is determined by the supplied default value.

If default is omitted, the value configured when creating the pipeline is used.

None can be passed explicitly as the initial value:

pipeline = Pipeline([
	(add, (5,)),
], default=10)

pipeline.run(None).wait()

In this example, the first step receives None, not 10.

delay specifies the delay in seconds before the first step and between subsequent steps.

When delay is enabled, the initial delay provides an opportunity to cancel the pipeline before the first step begins.

pipeline.run(10, delay=1)

pipeline.stop()

daemon controls whether the worker thread is a daemon thread.

stop_on_error controls whether execution stops after the first exception.

When disabled, exceptions are stored in errors and execution continues with the previous result.

run() returns the pipeline instance, allowing calls such as:

pipeline.run(10).wait()

The pipeline is snapshotted when execution starts. Changes made to pipeline.pipeline after run() begins do not affect the current execution.

wait()

Wait for the current execution to finish.

pipeline.wait()

It returns the pipeline instance.

By default, exceptions raised by pipeline steps are stored in errors and are not raised by wait().

To re-raise the most recent worker exception in the calling thread, use reraise_exception=True:

try:
	pipeline.wait(reraise_exception=True)
except Exception as error:
	print(error)

This allows an exception raised by a worker step to be re-raised in the thread waiting for the pipeline.

stop()

Request the running pipeline to stop and wait for its worker thread to terminate.

step = pipeline.stop()

The return value is the current one-based step index when execution is stopped, or 0 if the pipeline was not running.

If the pipeline is stopped before the first step begins, the return value is 0.

skip()

Request the worker to skip the next step that reaches its skip check.

pipeline.skip()

The method returns the pipeline instance.

rerun()

Stop the current execution and start the pipeline again.

pipeline.rerun(10)

Arguments are passed directly to run().

Context Manager

Pipeline can be used as a context manager.

Entering the context automatically starts the pipeline if it is not already running. Exiting the context requests the pipeline to stop and waits for the worker thread to terminate.

with Pipeline([
	(add, (5,)),
	(multiply, (2,)),
]) as pipeline:
	print("Pipeline started")

pipeline.results.get()

This is useful when the lifetime of the pipeline should be tied to a with block.

The context manager does not start the pipeline again if it is already running:

pipeline.run(10)

with pipeline:
	# The existing execution continues.
	pass

When leaving the context, stop() is called regardless of whether the block exits normally or because of an exception.

Manual Step Execution

run_step() executes one configured step synchronously.

result = pipeline.run_step(2, 10)

Unlike run(), this method:

• Does not create a worker thread.
• Does not modify the pipeline's worker-thread state.
• Does not store the result in results.
• Does not store exceptions in errors.
• Allows exceptions to propagate to the caller.

This makes it useful when a single pipeline step needs to be executed manually.

Results and Errors

The pipeline provides two Stack instances:

pipeline.results
pipeline.errors

results contains the initial value and the results produced by successfully executed steps.

For example:

pipeline.run(10).wait()

print(pipeline.results.get())

errors contains exceptions raised by pipeline steps.

When stop_on_error=True, execution stops after the first exception.

When stop_on_error=False, the exception is stored in errors and execution continues with the previous result.

If multiple exceptions occur, they are stored in errors in execution order, with the most recent exception at the top of the stack.

Managing Pipeline Steps

Pipeline steps can be modified before or between executions.

add()

Append a step:

pipeline.add((str.upper,))

insert()

Insert a step at a one-based position:

pipeline.insert(2, (str.strip,))

Positions range from 1 to len(pipeline) + 1.

pop()

Remove and return a step:

step = pipeline.pop(1)

Pipeline indexes are one-based.

clear()

Remove all configured steps:

pipeline.clear()

Pipeline State

The running property indicates whether the worker thread is currently running:

if pipeline.running:
	print("Pipeline is running")

The step attribute contains the one-based index of the currently executing step. It is 0 when the pipeline is not running.

A Pipeline instance can also be used as a boolean:

if pipeline:
	print("Pipeline is running")

Calling a pipeline instance is equivalent to calling run():

pipeline(10)

is equivalent to:

pipeline.run(10)

The length of a pipeline is the number of configured steps:

len(pipeline)

A callable can be checked with the in operator:

if add in pipeline:
	print("add is part of the pipeline")

Callable membership uses identity comparison.

Functional Utilities

compose()

compose() applies callables sequentially to a value.

The result of each callable is passed as the first argument to the next callable.

from pipeline import compose

def add(value, amount):
	return value + amount

def multiply(value, factor):
	return value * factor

result = compose(
	lambda value: add(value, 5),
	lambda value: multiply(value, 2),
	default=10,
)

print(result)

The execution flow is:

10
 ↓
add(10, 5)
 ↓
15
 ↓
multiply(15, 2)
 ↓
30

compose() executes the callables immediately and returns the final result.

An empty composition returns the supplied default value.

pipe

pipe wraps a callable as a step factory.

It is useful when the same callable needs to be configured with different arguments.

from pipeline import pipe

@pipe
def add(value, amount):
	return value + amount

add_five = add(5)

print(add_five(10))

The pipe object itself is called to create a step.

add(5)

returns a step containing add and the argument 5.

A step can be used directly as a callable:

result = add_five(10)

It can also be placed inside a pipeline step tuple:

pipeline = Pipeline([
	(add_five,),
])

Because step objects are callable, they are compatible with the standard pipeline step format without requiring special handling by Pipeline.

step

step represents a callable with preconfigured arguments.

from pipeline import step

add_five = step(add, 5)

print(add_five(10))

A step passes its supplied value as the first argument to the wrapped callable, followed by its configured positional and keyword arguments.

It can also be used with the pipe operator:

result = 10 | add_five

This is equivalent to:

result = add_five(10)

A step can be repeated with the multiplication operator:

def add(value, amount):
	return (value or 0) + amount

result = step(add, 5) * 3

print(result)  # 15

The wrapped callable is executed once for each repetition, with each result passed to the next execution.

Repeated execution starts with None; each result is passed to the next execution.

Exporting a step

A step can be converted into the standard pipeline step format with export():

add_five = step(add, 5)

pipeline_step = add_five.export()

print(pipeline_step)

The exported value is one of the standard pipeline step forms:

(function,)
(function, args)
(function, kwargs)
(function, args, kwargs)

For example:

step(add, 5, amount=10).export()

produces:

(add, (5,), {"amount": 10})

Empty positional or keyword arguments are omitted from the exported tuple.

Unpacking a step

A step can also be unpacked directly with the * operator:

add_five = step(add, 5)

pipeline = Pipeline([
	(*add_five,),
])

This is equivalent to:

pipeline = Pipeline([
	add_five.export(),
])

Unpacking is therefore a convenient shorthand when constructing pipeline step tuples.

step itself remains a general callable object and is not a special pipeline step type. Pipeline continues to use its standard (function, args, kwargs) step format.

step also provides convenience support for file-like objects through < and >:

process_step = step(process)

result = process_step < file
process_step > file

These operations use the file-like object's read() and write() methods respectively.

tap

tap() applies a side effect to a deep copy of a value and returns the original value unchanged.

This makes it useful for logging, inspection, debugging, or other side effects that should not interrupt a functional chain.

from pipeline.tap import tap

value = {"count": 10}

def show(data):
	print(data)

result = tap(value, show)

print(result is value)  # True

The function receives a deep copy, so mutations made by the side-effect function do not modify the original value.

tap() can also be used as a pipeline step:

from pipeline import Pipeline
from pipeline.tap import tap

def add(value, amount):
	return value + amount

pipeline = Pipeline([
	(add, (5,)),
	(tap, (print,)),
	(add, (10,)),
])

pipeline.run(10).wait()

The value printed by tap() is still passed unchanged to the next step.

Stack

Stack is a simple LIFO stack container with optional capacity limits.

It supports common stack operations such as pushing, retrieving and removing items, with dedicated exceptions for overflow and underflow conditions.

Import it directly from its submodule:

from pipeline.stack import Stack

stack = Stack()

stack.push("first")
stack.push("second")

print(stack.get())

Stack is also used internally by Pipeline for storing results and errors.

Stack raises StackOverflowError when pushing to a full stack and StackUnderflowError when accessing or removing an item from an empty stack.

For detailed stack operations and behavior, see the pipeline.stack module.

API Overview

Pipeline

Member	Description
run()	Start asynchronous pipeline execution
run_step()	Execute one step synchronously
stop()	Stop the current execution
skip()	Request the next step to be skipped
wait()	Wait for the current execution
rerun()	Restart the pipeline
add()	Append a step
insert()	Insert a step
pop()	Remove and return a step
clear()	Remove all steps
running	Whether the worker is running
step	Current one-based step index
results	Stack of initial value and successful results
errors	Stack of raised exceptions

Functional Utilities

Member	Description
compose	Apply callables sequentially to a value
pipe	Wrap a callable as a step factory
tap	Apply a side effect to a deep copy of a value

step

Member	Description
export()	Convert the step to standard pipeline step format

Stack

Member	Description
Stack(maxsize=0)	Create an empty LIFO stack with optional maximum capacity.
push(value)	Push a value onto the top of the stack.
get()	Return the top value without removing it.
pop()	Remove and return the top value.
clear()	Remove all values from the stack.
empty()	Return whether the stack is empty.
full()	Return whether the stack has reached its maximum capacity.
len(stack)	Return the number of values currently in the stack.
bool(stack)	Return whether the stack contains at least one value.
value in stack	Check whether a value exists in the stack.
iter(stack)	Iterate over values from bottom to top.
repr(stack)	Return a developer-oriented representation of the stack.

Requirements

• Python 3.8 or newer

Changelog

See CHANGELOG.md.

License

This project is licensed under the MIT License. See LICENSE for details.

Contribution

If you'd like to contribute, feel free to submit a pull request.

If you'd like to report a bug or request a feature, please open an issue.

Copyright (C) 2026 Hoàng Long