repolish 0.7.0

Maintain consistency across repositories

repolish

Repolish is a hybrid of templating and diff/patch systems, useful for maintaining repo consistency while allowing local customizations. It uses templates with placeholders that can be filled from a context, and regex patterns to preserve existing local content in files.

Why this exists

Teams often need to enforce repository-level conventions (CI config, build tools, metadata, common docs) while letting individual projects keep local customizations. The naive approaches are painful:

• Copying templates into many repos means drift over time and manual syncs.
• Running destructive templating can overwrite local changes developers rely on.

Repolish solves this by combining templating (to generate canonical files) with a set of careful, reversible operations that preserve useful local content. Instead of blindly replacing files, Repolish can:

• Fill placeholders from provider-supplied context.
• Apply anchor-driven replacements to keep developer-customized sections.
• Track provider-specified deletions and record provenance so reviewers can see why a path was requested for deletion.

Design overview

Key concepts:

• Providers (templates): Each provider lives in a template directory and may include a repolish.py module that exports create_context(), create_anchors(), and/or create_delete_files() helpers. Providers supply cookiecutter context and may indicate files that should be removed from a project.
• Anchors: A small markup syntax placed in templates (and optionally in project files) that marks blocks or regex lines to preserve. Examples:
  • Block anchors: ## repolish-start[readme] ... repolish-end[readme]
  • Regex anchors: ## repolish-regex[keep]: ^important=.* The processors use these anchors to replace or merge the template content with the local project file while preserving the parts marked with anchors.
• Delete semantics: Providers can request deletions using POSIX-style paths. A ! prefix acts as a negation (keep). Config-level delete_files are applied last and recorded in provenance.
• Provenance: Repolish records a delete_history mapping that stores, for each candidate path, a list of decisions (which provider or config requested a delete or a keep). This helps reviewers and automation understand why a path was flagged.

How it works (high level)

1. Load providers configured in repolish.yaml (or the default config).
2. Merge provider contexts; config-level context overrides provider values.
3. Merge anchors from providers and config.
4. Stage all provider template directories into a single cookiecutter template (adjacent to the config under .repolish/setup-input).
5. Preprocess staged templates by applying anchor-driven replacements using local project files (looked up relative to the config location).
6. Render the merged cookiecutter template once into .repolish/setup-output.
7. In --check mode: compare generated files to project files and report either diffs, missing files, or paths that providers wanted deleted but which are still present.
8. In apply mode: copy generated files into the project and apply deletions as the final step.

Example usage

repolish.yaml (simple example):

directories:
  - ./templates/template_a
  - ./templates/template_b
context: {}
anchors: {}
delete_files: []

Run a dry-run check (useful for CI):

repolish --check --config repolish.yaml

Debugging preprocessors

For testing and understanding how preprocessors work, Repolish includes a debugger tool:

repolish-debugger debug.yaml

The debug YAML file contains template content and optional target/config. See the debugging guide for details.

Post-processing / formatters

Repolish supports an optional post_process list in repolish.yaml that runs commands after the template is rendered but before the --check diff or apply step. This is useful for running project formatters or other transformations so the checks operate on formatted output.

How it runs

• The commands are executed exactly once after rendering and before checking or applying.
• Commands are executed with the working directory set to the rendered project folder inside .repolish/setup-output/<project> (where _repolish_project is the generated project folder name; default repolish).

Command forms

You can provide post_process entries in two forms:

-- string (e.g. "ruff --fix .") — the string is tokenized with shlex.split and executed without a shell. This covers common tools and small python one-liners like python -c "open('file','w').write('x')".

• list/array of argv parts (e.g. ['prettier', '--write', '.']) — recommended when you want to avoid any ambiguity about argument parsing and quoting.

Platform note

• On Windows the tokenization rules differ; Repolish uses Python's shlex to split string commands into argv lists before executing them without a shell. Because quoting/escaping differs between POSIX shells and Windows, prefer the argv/list form in post_process on Windows to avoid surprises. If you must provide a single string entry, keep arguments simple (no shell metacharacters) or call a committed script from the argv form.

Security note

We intentionally avoid running commands with shell=True to reduce shell injection risks. If you need to run complex shell pipelines or use shell metacharacters, create a small script (bash or python) in the repository and invoke that script from post_process using the argv-list form. That keeps the command execution explicit and avoids embedding complex shells in the configuration.

Example repolish.yaml with formatters

directories:
  - ./templates/template_a
context: {}
anchors: {}
post_process:
  - ['python', '-m', 'pip', 'install', '-r', 'requirements-dev.txt']
  - 'ruff --fix .'
  - ['prettier', '--write', 'src/']
delete_files: []

If a post_process command exits with a non-zero status, Repolish will fail and return a non-zero exit code so CI can detect the problem.

This will produce structured logs that include:

• The merged provider context and delete_paths (so you can see what was requested).
• A check_result listing per-path diffs or deletion warnings like PRESENT_BUT_SHOULD_BE_DELETED.

Processor story (anchors)

We iterated on preserving local file semantics and landed on a simple, explicit anchor-based system. Anchors are easy for template authors to add and for maintainers to reason about:

• Block anchors allow entire sections of a file to be preserved or replaced while keeping the surrounding template-driven structure.
• Regex anchors can mark single lines or patterns to keep (useful for maintainer-inserted keys or comments that should survive templating).

Anchors are processed in staging before cookiecutter runs, so the generated output already reflects local overrides while still taking canonical values from templates when needed.

Conditional files (file mappings)

Template authors can provide multiple alternative files and conditionally choose which one to copy based on the context. This is useful when you want to offer different configurations (e.g., Poetry vs setuptools, GitHub Actions vs GitLab CI) without cluttering filenames with cookiecutter's {% if %} syntax.

How it works

Files in your template directory that start with _repolish. are treated as conditional/alternative files. They will only be copied to the project if explicitly referenced in the create_file_mappings() function (or file_mappings variable) in your repolish.py.

Conditional files can be placed anywhere in your template directory structure — at the root or nested in subdirectories. For example:

• _repolish.config.yml (root level)
• .github/workflows/_repolish.ci.yml (nested in subdirectories)
• configs/editors/_repolish.vscode-settings.json (deeply nested)

The file_mappings return value is a dictionary where:

• Keys are destination paths in the final project (must be unique)
• Values are source paths in the template, or None to skip

Since this is Python code, you have full control over the logic used to select source files (if/else, ternary expressions, function calls, etc.).

Example

Template directory structure:

templates/my-template/
├── repolish.py
└── repolish/
    ├── README.md                          # Always copied
    ├── _repolish.poetry-pyproject.toml    # Conditional (root level)
    ├── _repolish.setup-pyproject.toml     # Conditional (root level)
    └── .github/
        └── workflows/
            ├── _repolish.github-ci.yml    # Conditional (nested)
            └── _repolish.gitlab-ci.yml    # Conditional (nested)

In repolish.py:

def create_context():
    return {
        "use_github_actions": True,
        "use_poetry": False,
        "enable_precommit": True,
    }

def create_file_mappings():
    """Map destination paths to source files in the template.
    
    Returns dict[str, str | None]:
        - Key: destination path in final project
        - Value: source path in template, or None to skip
    
    Files starting with '_repolish.' are only copied when referenced here.
    They can be at any level in the template directory structure.
    """
    ctx = create_context()
    
    return {
        # Nested conditional file: rename and place in final location
        ".github/workflows/ci.yml": (
            ".github/workflows/_repolish.github-ci.yml" if ctx["use_github_actions"]
            else ".github/workflows/_repolish.gitlab-ci.yml"
        ),
        
        # Root-level conditional: pick between alternatives based on context
        "pyproject.toml": (
            "_repolish.poetry-pyproject.toml" if ctx["use_poetry"]
            else "_repolish.setup-pyproject.toml"
        ),
        
        # Conditional: only include if enabled (None means skip)
        ".pre-commit-config.yaml": (
            "_repolish.precommit-config.yaml" if ctx.get("enable_precommit")
            else None
        ),
    }

Key behaviors

• Conditional files (prefixed with _repolish.) are only copied when explicitly listed in file_mappings
• Regular files (no _repolish. prefix) are always copied normally
• Destinations are unique: you cannot map multiple sources to the same destination within a single provider
• None values are skipped: returning None as the source means "don't copy this destination"
• Multiple providers: file_mappings from multiple providers are merged; later providers override earlier ones for the same destination

Use cases

This feature is ideal for:

• Offering multiple CI configurations (GitHub Actions, GitLab CI, Jenkins)
• Supporting different build systems (Poetry, setuptools, Hatch)
• Providing optional configuration files that depend on context
• Keeping template filenames clean (no {% if use_poetry %}pyproject.toml{% endif %})

The key advantage over cookiecutter's conditional filename syntax is that you keep filenames clean and organize conditional logic in Python where you have full programmatic control.

Create-only files (initial scaffolding)

Template authors can specify files that should only be created when they don't exist in the project. This is useful for initial repository scaffolding — like setting up a package structure for a new project — where you want to create the basic layout once but allow repolish to continue updating other files like pyproject.toml, CI configs, etc.

How it works

Files listed in create_only_files will be:

• Created if they don't exist in the project (normal template behavior)
• Skipped if they already exist (preserving user modifications)

This is different from anchors, which merge template and user content. Create-only files are an all-or-nothing decision: create when missing, preserve when present.

Example: Scaffolding a new package

Imagine your template creates a Python package scaffold. When someone starts a new project called awesome-tool, the template should create the package structure once (directories, __init__.py files, example modules) but subsequent repolish runs should only update the tooling files (like pyproject.toml, .github/workflows/ci.yml, etc.) while leaving the package code alone.

Template directory structure:

templates/python-project/
├── repolish.py
└── repolish/
    ├── src/
    │   └── {{cookiecutter.package_name}}/
    │       ├── __init__.py      # Create-only: initial package structure
    │       ├── py.typed          # Create-only: type marker
    │       └── main.py           # Create-only: example starter module
    ├── tests/
    │   ├── __init__.py           # Create-only: test package marker
    │   └── test_main.py          # Create-only: example test
    ├── pyproject.toml            # Regular: always updated by repolish
    └── .github/
        └── workflows/
            └── ci.yml            # Regular: always updated by repolish

In repolish.py:

def create_context():
    return {
        "package_name": "awesome_tool",  # User's actual package name
        "project_name": "awesome-tool",
        "version": "0.1.0",
    }

def create_create_only_files():
    """Files that are created only if they don't exist.
    
    Returns list[str]:
        POSIX-style paths relative to project root
    
    Use this for initial scaffolding that creates the package structure
    but shouldn't be overwritten on subsequent repolish runs.
    """
    ctx = create_context()
    pkg = ctx["package_name"]
    
    return [
        # Package source structure (created once, then hands-off)
        f"src/{pkg}/__init__.py",
        f"src/{pkg}/py.typed",
        f"src/{pkg}/main.py",
        
        # Test structure (created once, then hands-off)
        "tests/__init__.py",
        "tests/test_main.py",
        
        # Project files users often customize
        ".gitignore",
        "README.md",  # If you want an initial README but let users own it
    ]

Alternatively, you can use a module-level variable:

create_only_files = [
    "src/awesome_tool/__init__.py",
    "src/awesome_tool/py.typed",
    "src/awesome_tool/main.py",
    "tests/__init__.py",
    "tests/test_main.py",
]

What happens

First run (new project):

$ repolish apply

• Creates entire package structure: src/awesome_tool/, tests/, etc.
• Creates pyproject.toml, .github/workflows/ci.yml
• User now has a complete, working package scaffold

Later runs (after user develops the package):

$ repolish apply

• Skips src/awesome_tool/__init__.py (user has added exports)
• Skips src/awesome_tool/main.py (user has implemented features)
• Skips tests/test_main.py (user has written tests)
• Updates pyproject.toml (latest dependencies, build config)
• Updates .github/workflows/ci.yml (latest CI improvements)

This gives you the best of both worlds: automated tooling/config updates via repolish while keeping your actual code untouched

Key behaviors

• First run: Files are created normally from the template
• Subsequent runs: Files are skipped if they exist, preserving user changes
• Check mode: Reports MISSING for create-only files that don't exist yet, but won't report diffs for files that exist (even if content differs)
• Multiple providers: create_only_files from multiple providers are merged (additive); later providers can add more create-only files
• Works with file_mappings: A file can be both in file_mappings (for conditional copying/renaming) and create_only_files (for preservation)
• Conflicts with delete_files: If a file is in both create_only_files and delete_files, the delete wins (file will be deleted)

Use cases

This feature is ideal for:

• Initial package scaffolding: Create the package directory structure, __init__.py files, and starter modules for a new project, but leave them alone once the user starts developing
• Example/template modules: Provide example code (main.py, test_example.py) that users can modify or replace without repolish overwriting their work
• Project-specific configs: Files like .gitignore, README.md, or .editorconfig that should be created initially but are meant to be customized per-project
• Type markers and metadata: Files like py.typed that are needed for the package structure but never need updating
• Separation of concerns: Keep repolish managing your tooling/CI/build configs (always updated) while preserving your actual source code (created once, then hands-off)

The key insight: use create-only files for anything that serves as a starting point rather than a continuously synced template. This lets you scaffold new projects quickly while still getting the benefits of repolish for keeping your tooling configuration up to date.

How do I add anchors?

Anchors are intentionally simple so template authors and maintainers can reason about them easily. There are two primary forms:

• Block anchors mark a named section to preserve or replace between repolish-start[...] and repolish-end[...] markers. Use them for multi-line sections such as README snippets, install blocks, or long descriptions.
• Regex anchors mark single-line patterns to keep using a regular expression. They are useful when you want to preserve a line that follows a predictable pattern (version lines, keys, simple single-line edits).

Below are two practical examples you can copy into templates and projects.

Dockerfile (block anchor)

Template (templates/template_a/Dockerfile):

# base image
FROM python:3.11-slim

## repolish-start[install]
# install system deps
RUN apt-get update && apt-get install -y build-essential libssl-dev
## repolish-end[install]

# copy + install python deps
COPY pyproject.toml .
RUN pip install --no-cache-dir .

Project Dockerfile (local override) — developer has custom install needs:

FROM python:3.11-slim

## repolish-start[install]
# custom build deps for project X
RUN apt-get update && apt-get install -y locales libpq-dev
## repolish-end[install]

# copy + install python deps
COPY pyproject.toml .
RUN pip install --no-cache-dir .

When Repolish runs its preprocessing, the install block from the local project will be preserved in the staged template (so the generated output keeps the local custom RUN command), while the rest of the Dockerfile comes from the template.

pyproject.toml (regex anchor + block anchor)

Template (templates/template_a/pyproject.toml):

[tool.poetry]
name = "{{ cookiecutter.package_name }}"
version = "0.1.0"
## repolish-regex[keep]: ^version\s*=\s*".*"

description = "A short description"

## repolish-start[extra-deps]
# optional extra deps (preserved when present)
## repolish-end[extra-deps]

Project pyproject.toml (developer bumped version and added extras):

[tool.poetry]
name = "myproj"
version = "0.2.0"

description = "Local project description"

## repolish-start[extra-deps]
requests = "^2.30"
## repolish-end[extra-deps]

In this example the ## repolish-regex[keep]: ^version\s*=\s*".*" anchor ensures the local version = "0.2.0" line is preserved instead of being replaced by the template's 0.1.0. The extra-deps block is preserved whole-cloth when present, letting projects keep local dependency additions.

Notes and tips

• Use meaningful anchor names (e.g., install, readme, extra-deps) so reviewers immediately understand the preserved section's intent.
• Regex anchors are applied line-by-line; prefer anchoring to a simple, easy to read pattern to avoid surprises.
• Anchors are processed before cookiecutter rendering, so template substitutions still work around preserved sections.

Regex anchors and capture groups

Regex anchors let you specify exactly what to preserve from a local file. Two important behaviors to know:

• Capture group preference: If your regex includes a capturing group (parentheses), Repolish will prefer the first capture group (group 1) as the block to insert into the template. If there are no capture groups, Repolish falls back to the entire match (group 0).
• Safeguard trimming: As a conservative safeguard Repolish trims captured blocks to a contiguous region based on indentation so that incidental following sections are not accidentally pulled into the template. However, the canonical way to express intent is an explicit capture group — authors should prefer to capture exactly what they mean.

Implementation note

• Repolish computes the absolute span of the selected capture (group 1 when present, otherwise the full match) inside the template and trims that slice using an indentation-aware heuristic before replacing it with the trimmed content extracted from the local file. Also note that the ## repolish-regex[...] declaration line is removed from the template prior to matching and replacement. This avoids accidentally removing unrelated template sections when patterns include surrounding context.

Example

Template (excerpt):

cat1:
  - line1
  - line2
  ## repolish-regex[cat1-filter]: (^\s*# cat1-filter-additional-paths.*\n(?:\s+.*\n)*)
  # cat1-filter-additional-paths

cat2:
  - from-template
  ## repolish-regex[cat2-filter]: (^\s*# cat2-filter-additional-paths.*\n(?:\s+.*\n)*)
  # cat2-filter-additional-paths

cat3:
  - cat3-line
  ## repolish-regex[cat3-filter]: (^\s*# cat3-filter-additional-paths.*\n(?:\s+.*\n)*)
  # cat3-filter-additional-paths

Local file (excerpt):

cat1:
  - line1
  - line2
  # cat1-filter-additional-paths
  - extra

cat3:
  - cat3-line
  # cat3-filter-additional-paths

Result after preprocessing:

cat1:
  - line1
  - line2
  # cat1-filter-additional-paths
  - extra

cat2:
  - from-template
  # cat2-filter-additional-paths

cat3:
  - cat3-line
  # cat3-filter-additional-paths

When your regex is too greedy, tighten it or add explicit parentheses around the intended capture so Repolish can reliably hydrate the template without importing unrelated content.

Where anchors are declared and uniqueness

Anchors can come from three places (and are merged in this order):

1. Provider templates: any ## repolish-start[...] / ## repolish-regex[...] markers present inside the provider's template files.
2. Provider code: a provider's create_anchors() callable can return an anchors mapping (key -> replacement text) used during preprocessing.
3. Config-level anchors: the anchors mapping in repolish.yaml applies last and can be used to override or add anchor values.

When anchors are merged, later sources override earlier ones (config wins). Anchor keys must be unique across the whole merged template set — keys are global identifiers used to find matching repolish-start[...] blocks or repolish-regex[...] declarations. If two different template files (or providers) use the same anchor key, the later provider's value will override the earlier one, which can produce surprising results.

Example conflict

Two provider templates accidentally use the same anchor key init:

• templates/a/Dockerfile contains ## repolish-start[init] … ## repolish-end[init]
• templates/b/README.md also contains ## repolish-start[init] … ## repolish-end[init]

Because anchor keys are merged globally, the init block from the provider that is processed later will replace (or be used in place of) the other one. That may not be what you want — for predictable behavior, choose anchor keys scoped to the file or the provider, e.g. docker-install or readme-intro.

Best practice: prefix anchor keys with the file or provider name when the content is file-scoped. This avoids accidental collisions when multiple providers contribute templates that contain similarly-named sections.

Why this is useful

• Safe consistency: teams get centralized templates without forcing destructive, manual rollouts.
• Clear explainability: the delete_history provenance makes it easy to review why a file was targeted for deletion or kept.
• CI-friendly: --check can be run in CI to detect drift; logs and diffs make it straightforward to require PRs to run repolish before merging.

Final notes

Repolish is intentionally small and composable. If you need per-file log artifacts, or slightly different merge rules, the processors and cookiecutter helpers are isolated so you can adapt them safely.

Contributions and issues are welcome — see the test-suite for practical examples of how the system behaves.