pyflowchart claimed Python 3.6+ support but had never been tested against it, used a setup.py-only packaging setup with no pyproject.toml, carried several silent runtime bugs, and had no CI test matrix at all. This PR addresses all of it.

Version

Bumped to 0.5.0b1 (setup.py and pyproject.toml).

Packaging

• Add pyproject.toml (PEP 517/621): build system, classifiers for Python 3.7–3.14, [dev] extra, pyflowchart console-script entry point, pytest config, Bug Tracker URL
• setup.py: drop Python 3.6 classifier (not supported by setuptools>=61), add Python 3.7–3.14 classifiers, update python_requires to >=3.7, add entry_points for the console script
• .gitignore: remove pyproject.toml exclusion (a leftover "poetry experiment" comment was silently preventing it from ever being committed)
• README.md: updated to state "requires Python 3.7+ and is CI-tested on Python 3.7 through 3.14" — Python 3.6 is not supported because setuptools>=61.0 (required by the PEP 621 [project] table) itself requires Python ≥ 3.7

Bug fixes

ast_node.py

• import typing — unused bare import removed; List added explicitly to from typing import …
• Double parse() call in Loop.parse_loop_body() — original code called parse(self.ast_object.body, **kwargs) twice: the first result (progress) was used only for the head is not None guard then discarded, and a second identical call (process) was used for the actual graph wiring. This was not intentional design — a dirty implementation that inflated node IDs via itertools.count. Fixed by removing the redundant first call and using a single result throughout. The fix is safe: parse() is deterministic for a given input and the single result is used consistently.
• lstrip("return") in ReturnOutput — strips individual characters {r,e,t,u,n}, not the keyword; fixed to source[len("return"):].lstrip()
• _ast.Raise not handled — fell through to CommonOperation and connected forward; raise is terminal like break; added to __ctrl_stmts mapped to BreakContinueSubroutine
• _ast.YieldFrom not handled — fell through to CommonOperation instead of an I/O output node; added to __ctrl_stmts mapped to YieldOutput
• AsyncFunctionDef / AsyncFor not handled — fell through to CommonOperation; added to __func_stmts / __loop_stmts respectively, mapped to the same FunctionDef / Loop classes; assert guards in FunctionDefArgsInput.func_args_str() and FunctionDef.parse_func_body() updated to include _ast.AsyncFunctionDef
• except IndexError or AttributeError in Match.__init__ — Python evaluates this as except IndexError; AttributeError was never caught; fixed to except (IndexError, AttributeError)
• MatchCase.parse_body() — used bare _ast.match_case instead of _ast_match_case_t alias; called parse(body) without **kwargs, silently dropping simplify/conds_align inside every match-case body
• Match.parse_cases() — bare _ast.Match instead of _ast_Match_t alias
• Match.__init__() comments — restored the commented-out # self.head = TransparentNode(self) alternative with a note explaining why NodesGroup.__init__() is used instead (direct assignment hits multi-inheritance MRO issues); removed only the profane word
• parse() — type(x) == Y replaced with isinstance(x, Y) for both Match and Expr checks

flowchart.py

• Load-bearing assert statements in from_code() — silently skipped by python -O, producing a cryptic AttributeError; replaced with raise ValueError
• Control-flow assert in find_field_from_ast() — used inside try/except (AttributeError, AssertionError) to detect a missing field; disabled in -O mode, causing the wrong AST node to be silently returned and garbage flowcharts to be produced; replaced with if/raise ValueError and tightened except clause

__main__.py

• confidence < 0.9 in detect_decode() — raises TypeError when chardet returns confidence=None on unrecognised byte streams; fixed to if not confidence or confidence < 0.9:

__init__.py

• No __all__ — from pyflowchart import * leaked time, uuid, itertools, List, TypeVar, debug, and every private symbol; 26 public names now declared explicitly

CI / GitHub Actions

• codeql-analysis.yml: checkout@v2 → @v4, codeql-action/*@v1 → @v3, add permissions block
• TestPyPi-publish.yml / release-publish.yml: checkout@master (unpinned) → @v4, setup-python@v1 → @v5, pypi-publish@master (unpinned) → @release/v1, Python 3.7 (EOL) → 3.12, add permissions with id-token: write
• New tests.yml: CI test matrix covering the full supported range — Python 3.7 on ubuntu-22.04 (the only runner with Python 3.7 in its hosted tool cache) and Python 3.8–3.14 on ubuntu-latest, running on every push/PR

Tests

7 new tests added covering every fix with no prior test:

# Before: silently drops simplify=False inside match-case bodies
# After: kwargs forwarded correctly
fc = Flowchart.from_code(code, field='fn', simplify=False)
# → condition node for inner `if`, not a collapsed operation node

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.