Prepare 9.0.0: legacy defaults with opt-in modern slug generation

[un33k]

Summary

• Keep slugify(...) legacy-compatible by default; improved rules require algorithm="modern".
• Preserve upstream test.py byte-for-byte; add separate regression and differential compatibility checks.
• Add explicit backend and replacement-stage controls, CLI support, packaging modernization, and migration documentation.
• Set version to 9.0.0 without a development suffix. Include supporting issue-review evidence.

Verification

• All 82 unchanged upstream tests pass.
• Full local macOS matrix: CPython 3.10–3.14 and PyPy 3.11 across three backends (18 cells).
• 1,866 test passes and six intentional backend-isolation skips.
• Strict mypy, pycodestyle, flake8, packaging validation and diff checks passed; aggregate coverage 99% (includes tests).

Scope and limitations

• This PR does not publish a package or create a release tag.
• Windows/Linux execution, full installed-artifact regression matrices and real downstream persisted-slug datasets remain unverified.
• See docs/release-9/verification.md for evidence and historical verification notes.

🚀 Generated with Dojo ⛩️

[coveralls]

coverage: 99.864% (+2.4%) from 97.44% — release/9.0.0-legacy-default into master

[un33k]

Dojo — issue and PR disposition index

This is Dojo, posting the maintainer-authorized follow-up for #191. Legacy remains the default; modern behavior requires explicit opt-in. Individual contributor PRs are being closed as incorporated/superseded or declined, not merged separately. Deferred requests are not represented as implemented. A merged PR does not itself publish a package.

• \u2019 should be replaced by a separator #30: A global curly-apostrophe behavior change is not adopted. ASCII and curly apostrophes remain distinct; callers can choose explicit pre replacements.
• [WIP] Replicate #41 in Travis build #43: The installed-command testing intent is revived through isolated wheel/sdist installation and CLI smoke checks, not by adopting the old Travis patch.
• Use the actual separator in code #47: Partially addressed: modern mode budgets actual emitted separator width. Literal-hyphen preservation is NOT implemented; both algorithms retain global dash mapping.
• Shorten usage to one code and one shell example #48: The documentation now has a concise Python/CLI quickstart while retaining option references and recipes, reviving this documentation proposal's intent.
• MANIFEST.in: Add test.py to sdist #74: Explicit source-test inclusion and source archive checks are provided for newly built distributions. Historical published archives are not modified.
• Add UniqueSlugify for generating unique slugs #76: A stateful uniqueness API remains deferred. Collision handling belongs to the application; no database uniqueness guarantee is provided.
• Name conflict with other slugify package #81: Install-name versus import-name troubleshooting is documented. No package namespace rename is introduced.
• Issue with separator #91: Literal-hyphen preservation remains unsupported for compatibility. Modern separator length budgeting is a separate improvement, not a fix for literal dash preservation.
• Create command line tool that can set all parameters #92: CLI parameter parity is extended with regex forwarding and explicit algorithm/backend/replacement-stage controls.
• We cannot replace '-' before replacing ' ' by '-' #94: Explicit pre-only replacement stages address the replacement-order use case while retaining the historical two-pass default.
• publish pre-build wheels (manylinux) #96: Pure-Python wheel build and artifact checks are provided. This does not announce publication; platform-specific manylinux machinery is unnecessary for the pure-Python wheel.
• Question/Bug: how do I preserve dashes and not replace them with separator? #107: Global dash-to-separator mapping remains unchanged. No new literal-hyphen-preservation API is introduced; modern length budgeting does not resolve that separate request.
• Question: Why is save_order False by default? #118: save_order remains False by default to preserve existing outputs. The behavior is documented rather than silently changing the default.
• Question: Why are custom replacements ran twice? #119: The two-pass contract is documented and explicit pre/post stages are available. Replacement iterator snapshot/replay changes require modern mode.
• Document the use of what's in special.py #120: Exported substitution tables and an explicit German pre-replacement recipe are documented. Tables are optional, not automatic locale detection.
• Fuzzing Python Slugify on OSS-Fuzz #127: Bounded combinatorial regression coverage is added. OSS-Fuzz enrollment and a separate disclosure-response program are NOT implemented by Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191.
• Enable installation without GPL dependencies #128: Dependency-free normal installation remains deferred: text-unidecode is still a base dependency. Lazy loading permits Unicode-only runtime without a backend, but does not remove installation metadata requirements or provide a blanket licensing assurance.
• Replace flake8 and pycodestyle with ruff #131: Ruff migration remains deferred. Existing lint tools are retained; Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191 does not implement this tooling migration.
• Fix CI to run on master without duplicate push and pull_request runs #132: Workflow consolidation and duplicate-run cleanup remain deferred pending branch policy. Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191 does not claim to resolve this request.
• Option to convert a string to a "safe" filename and modify the original name as less as possible  #133: A broad safe-filename API remains deferred. Destination-specific validation is the application's responsibility; slug generation alone is not a universal filename safety guarantee.
• pip install python-slugify[unidecode] also installs text-unidecode #140: Documentation clarifies additive extras, installed dependencies and runtime backend selection. Selecting a backend does not subtract the base dependency.
• PyPi tarball is missing files #141: Source archive contents are explicitly included and checked for newly built artifacts. Existing published PyPI archives are not repaired or replaced.
• TypeError: decoding to str: need a bytes-like object, int found #144: Unsupported input types receive a clear TypeError; UTF-8 bytes and bytearray remain supported.
• Argument to make flugified strings valid XML id attributes. #150: An XML-specific slug option is not adopted. XML identifier validation and any required prefixing remain caller responsibilities.
• The problem of converting all Korean consonants to Korean initials in the slugfiy process when they are alone. #153: Normalization/jamo behavior is documented and regression-covered. Exact original code-point preservation is not implemented.
• rewrote slugify func with radon #155: The wholesale complexity-driven rewrite is not adopted. Focused helper changes in Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191 should not be read as integration of this historical patch.
• Adding Debian manpage to python-slugify #156: Debian manpage import remains deferred pending source and license review. Updated quickstart/CLI documentation is not a delivered manpage.
• [WIP] Rewriting in Rust PyO3 python lib - Student / Learning project #158: The native Rust/PyO3 rewrite remains deferred. No measured performance justification or Rust integration is claimed by Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191.
• Fix build process #159: Building on your packaging work, Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191 uses declarative metadata and isolated build checks and removes the executable publish shortcut. Thank you for the foundation.
• Reintroduce tox and thoroughly test the project #160: Your tox/build testing foundation is retained and expanded with backend/interpreter and artifact checks. This is an extension of that work, not a claim that testing was previously absent.
• Default to anyascii decoder #163: AnyASCII is available as an explicitly selected backend. Switching the automatic default to AnyASCII is declined to preserve existing output; its installation alone does not change auto selection.
• Fix CI by dropping EOL Python versions #164: Python >=3.10 support documentation and the modern interpreter matrix are synchronized, building on the existing supported-runtime policy.
• Fix the README badge to refer to the main.yml workflow results #165: The remaining README badge destination mismatch is corrected. Broader workflow consolidation is not completed by this change.
• Slugify licensing update and PRs #166: Addressed through backend, installation and license documentation. The base text-unidecode dependency remains; extras are additive. No blanket legal assurance is made.
• Inconsistent workflow badge reference in README #167: Fixed the README workflow badge link to match its main.yml image.
• docs: align workflow badge link with the main.yml #168: The badge correction is incorporated in Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191; this standalone PR is superseded, not merged separately.
• Add edge case test for whitespace-only input #169: Consolidated empty and whitespace-only regression coverage. This adds tests rather than claiming a new runtime fix.
• test: add edge case test for whitespace-only input #170: Whitespace regression coverage is consolidated with Add edge case test for whitespace-only input #174 and test: cover whitespace-only slugify input #188 in Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191; this standalone PR is superseded.
• Ordinal indicators are not slugified #172: Existing ordinal normalization is confirmed and regression-covered; no global transliteration-table change was needed.
• fix: align README badge link with badge image workflow #173: The badge-link correction is incorporated in Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191. Thank you for the duplicate investigation; no separate merge is needed.
• Add edge case test for whitespace-only input #174: The broader whitespace examples are consolidated into the new regression suite; this standalone PR is superseded.
• --regex-pattern option ignored by CLI #175: Fixed explicit CLI --regex-pattern forwarding in both legacy and modern modes. This intentionally corrects previously ignored flag behavior.
• Fix --regex-pattern being ignored by the CLI #176: Thank you, Jacobo de Vera. Your focused CLI regex-forwarding correction is incorporated, with additional CLI and installed-artifact checks. This standalone PR is superseded, not merged separately.
• Release time? #177: Release preparation is consolidated and tracked in Prepare 9.0.0: legacy defaults with opt-in modern slug generation #191. Closing this tracking issue does not mean version 9.0.0 has been published; package publication and tags are separate actions, with no date promised here.
• fix: recover from malformed numeric entities #178: Your earlier malformed numeric-reference recovery investigation is credited alongside Handle invalid numeric entity references individually #181. Recovery is incorporated ONLY in opt-in algorithm='modern'; legacy behavior is preserved.
• Don't apply self-referential replacements twice #179: The proposed automatic skipping of self-referential rules remains declined. Explicit pre/post replacement stages are available instead; reusable rules retain the intentional two-pass default. No universal idempotence guarantee is introduced.
• fix: wire --regex-pattern through the CLI (#175) #180: The duplicate CLI regex-forwarding fix is covered in both algorithms, retaining Fix --regex-pattern being ignored by the CLI #176's priority and acknowledging the additional regression investigation.
• Handle invalid numeric entity references individually #181: Per-reference invalid numeric entity recovery is incorporated ONLY in opt-in algorithm='modern', including invalid-neighbor and surrogate cases. Legacy retains historical substitution behavior. This standalone PR is superseded.
• fix: make slugify replacement passes idempotent (supersedes #179) #182: Declined as written: silently skipping replacement rules or sanitizing their final output changes caller intent. Explicit replacement_stage='pre'/'post' controls are supplied instead; the historical two-pass default remains. Universal idempotence for arbitrary replacements is not promised.
• Raise a clear TypeError for non-str/non-bytes input to slugify() #183: Clear unsupported-input validation is incorporated, while retaining UTF-8 bytes and bytearray support and decoding before replacements. Unrelated ignore-file changes are not included. This standalone PR is superseded.
• Pass --regex-pattern through to slugify() from the CLI #184: The duplicate CLI regex-forwarding correction is covered in both algorithms. Fix --regex-pattern being ignored by the CLI #176 remains the original focused contribution; no duplicate merge is needed.
• fix: treat non-positive smart_truncate max_length as no truncation #185: Not adopted for the public smart_truncate helper: changing its negative-limit slicing would break the legacy contract. Direct helper calls remain unchanged. Nonpositive slugify limits are unlimited; modern slugify uses a separate private helper. This is a compatibility decision, not a claim that the public-helper proposal was implemented.
• fix: hard-truncate when smart_truncate separator is empty #186: Not adopted for the public smart_truncate helper: direct calls with an empty separator still raise ValueError for compatibility. Modern slugify handles empty separators through a separate private helper. This is not a fix to the public helper.
• Transliterate HTML entities decoded after the unidecode pass #187: Encoded/literal transliteration parity is addressed ONLY in algorithm='modern', by decoding before the existing transliteration pass rather than adding an unconditional second pass. Legacy output remains unchanged. This standalone PR is superseded.
• test: cover whitespace-only slugify input #188: Whitespace regression examples are consolidated with test: add edge case test for whitespace-only input #170 and Add edge case test for whitespace-only input #174; this standalone PR is superseded.
• test: add regression coverage for ordinal indicators (º, ª) #189: Ordinal regression coverage is incorporated, confirming existing behavior rather than adding global transliteration rules. This standalone PR is superseded.
• Preserve case-sensitive stopword filtering for iterators #190: Case-sensitive stopword iterator membership is corrected ONLY in algorithm='modern' by snapshotting stopwords. Legacy deliberately retains historical iterator consumption. This standalone PR is superseded.

Thank you to all reporters and contributors. This index also links historical contributions and explicit decisions; already-closed items remain closed. No PR merge is performed by this follow-up.

🚀 Generated with Dojo ⛩️