Section 3 of 7 · Programming & Data Foundations

Python Execution Model

Later ML code is still Python code. Precise reasoning about objects, imports, cleanup and package boundaries prevents silent state errors that framework knowledge cannot repair.

9 units 19 topics 0/9

Units

9 in this section

3.1 Names, identity, mutability, copying and hashability
explain · measure
4 h
3.2 Argument binding, scope and closures
explain
3 h
3.3 Attribute lookup, object protocols and dataclasses
explain · implement
5 h
3.4 Iterators, generators and lazy pipelines
explain · implement
4 h
3.5 Decorators, wrappers and parameterized decorators
implement
3 h
3.6 Context managers, exceptions and deterministic cleanup
implement
4 h
3.7 Type hints, generics, protocols and narrowing
explain · implement
3 h
3.8 Imports, virtual environments and src-layout packaging
explain · implement
4 h
3.9 Reference counts, cycle collection and weak references
explain · measure
2 h

Before moving on

Checkpoint

Build and install a typed src-layout package from a blank directory, then trace and repair one seeded aliasing, cleanup or import defect without trial-and-error edits.

Deliberately out of scope

Leave for later

Metaclass frameworks and custom import hooks
CPython interpreter implementation details beyond a diagnosed need
Supporting multiple packaging backends

Resources for this section

Shared by every unit above

Stas' Python Cookbook - selected path
Stas Bekmancourse

primary

Run Part I on dictionaries, iteration, functions, closures, decorators, classes and context managers; Part II on modules, files, paths, environment and arguments; and Part IV on versions, dependencies, packaging and code quality. Finish by building one src-layout package from a blank directory.
Python language reference
Python Software Foundationreference

required

Check Data model, Execution model, Import system and Exceptions whenever a claim depends on Python semantics. These pages are the authority when the course and a runtime observation disagree.
Packaging Python Projects
Python Packaging Authorityimplementation

required

Follow the packaging tutorial once from a blank directory, using a src layout, pyproject.toml, an isolated build and a clean-environment installation test.
AlgoMaster Python - structured revision
AlgoMasterpractice

support

Use Functions, Modules and Packages, OOP, Advanced OOP, Exception Handling, Iterators and Generators, Type Hints and Annotations only after the cookbook pass. Open it for recall practice or a diagnosed weak mechanism, not as a second complete course.
dataclasses - data classes
Python Software Foundationreference

reference

Read the @dataclass parameters (eq, order, frozen, slots), field() with default_factory, frozen instances, post-init processing, replace, asdict and astuple. Use for the validated frozen dataclass in attributes-protocols-dataclasses.
typing - type hints and protocols
Python Software Foundationreference

reference

Read Protocols for structural typing and runtime_checkable limits, Generics for TypeVar generics, and assert_type for narrowing checks. Use for the Protocol and narrowing parts of type-hints-protocols; PEP 695 syntax itself is in the language reference.
gc - cycle garbage collector
Python Software Foundationreference

reference

Read the cycle collector interface: collect, get_count, garbage and disable/enable. Use for the cycle-collection and measurement parts of refcounting-gc-weakrefs.
weakref - weak references
Python Software Foundationreference

reference

Read weak-reference objects (ref callbacks, proxy objects) and the WeakValueDictionary example. Use for the weak-reference part of refcounting-gc-weakrefs.

Apply the section

Mini-project · Installable taskrunner package with a typed registry and two seeded defects (8 to 12 h)

Build the installable task runner package the later projects extend, with a typed task registry, a lazy run log and deterministic cleanup. Two seeded defects, one aliasing and one lifetime, train evidence-first diagnosis with a regression test for each.

Steps

Scaffold a src-layout package with uv init, build its wheel with uv build, install the wheel into a fresh environment and run its console entry point outside the repository, showing the installed copy is imported.
Model tasks as a frozen slotted dataclass validated in __post_init__, registered by a decorator that rejects duplicates, and checked against the AlgoMaster Python Data Classes and Decorators exercises, with mypy --strict clean.
Stream the JSON Lines run log through a generator and build the summary from generator expressions, then predict which peak grows with file size and compare peaks with tracemalloc against a list-building version.
Give each run a directory through a context manager that writes a marker and closes the log on every path, map failures to a small chained exception hierarchy, and test every path.
Seed an aliasing defect where a shallow copy shares nested run parameters across runs, write the regression test first, watch it fail, then fix it and explain why the chosen copy strategy removes the sharing.
Seed a cleanup defect where a stored bound method keeps a cycle alive, diagnose it with a weakref probe and the collector, then fix it so the regression test passes with the collector disabled.

Deliverables

The taskrunner repository: src-layout package, wheel, and a README with install and import evidence.
Tests for validation, registration, laziness, cleanup and causes, with mypy --strict clean.
Two defect write-ups with failing and passing regression output, diagnosis evidence and the laziness peak comparison.

Checks

rubric · Self-grade with the rubric below

Rubric

Criterion	Pass	Strong
Installable package	The wheel installs into a fresh environment and the entry point runs outside the repository.	A test asserts the imported package is the installed copy, and the `README` rebuilds it from a clean checkout.
Typed model	`TaskSpec` rejects bad values and round-trips, duplicates are rejected, and `mypy --strict` is clean.	Decorated tasks keep their own type, and the write-up shows the rejected `Protocol` typing and why.
Resources and laziness	Cleanup and markers hold on every path with causes chained, and the learner's tests cover each path.	The peak comparison shows the generator holding flat while the list version grows, as predicted.
Seeded defects	Both regression tests fail on the seeded code and pass after the fix.	Each write-up names the shared reference or cycle behind the bug and why the fix removes it.

Feeds into: The installable typed package, its entry point and the run log that the service and pipeline write; the lifetime write-up seeds the evidence-first debugging report.