Merge pull request #1170 from RonnyPfannschmidt/fix-987-git-archival-docs-and-tools

expand documentation on git archival/gitattributes and add cli tools for creating them

Author: RonnyPfannschmidt

CHANGELOG.md

@@ -9,6 +9,8 @@
 - fix #1099 use file modification times for dirty working directory timestamps instead of current time
 - fix #1059: add `SETUPTOOLS_SCM_PRETEND_METADATA` environment variable to override individual ScmVersion fields
 - add `scm` parameter support to `get_version()` function for nested SCM configuration
+- fix #987: expand documentation on git archival files and add cli tools for good defaults
+
 ### Changed
 
 - add `pip` to test optional dependencies for improved uv venv compatibility

docs/usage.md

@@ -376,15 +376,41 @@ accordingly.
 
 ### Git archives
 
-Git archives are supported, but a few changes to your repository are required.
+Git archives are supported, but require specific setup and understanding of how they work with package building.
 
-Ensure the content of the following files:
+#### Overview
 
+When you create a `.git_archival.txt` file in your repository, it enables setuptools-scm to extract version information from git archives (e.g., GitHub's source downloads). However, this file contains template placeholders that must be expanded by `git archive` - they won't work when building directly from your working directory.
+
+#### Setting up git archival support
+
+You can generate a `.git_archival.txt` file using the setuptools-scm CLI:
+
+```commandline
+# Generate a stable archival file (recommended for releases)
+$ python -m setuptools_scm create-archival-file --stable
+
+# Generate a full archival file with all metadata (use with caution)
+$ python -m setuptools_scm create-archival-file --full
+```
+
+Alternatively, you can create the file manually:
+
+**Stable version (recommended):**
 ```{ .text file=".git_archival.txt"}
+node: $Format:%H$
+node-date: $Format:%cI$
+describe-name: $Format:%(describe:tags=true,match=*[0-9]*)$
+```
 
+**Full version (with branch information - can cause instability):**
+```{ .text file=".git_archival.txt"}
+# WARNING: Including ref-names can make archive checksums unstable
+# after commits are added post-release. Use only if describe-name is insufficient.
 node: $Format:%H$
 node-date: $Format:%cI$
 describe-name: $Format:%(describe:tags=true,match=*[0-9]*)$
+ref-names: $Format:%D$
 ```
 
 Feel free to alter the `match` field in `describe-name` to match your project's
@@ -402,14 +428,92 @@ tagging style.
 .git_archival.txt  export-subst
 ```
 
-Finally, don't forget to commit the two files:
+Finally, commit both files:
 ```commandline
-$ git add .git_archival.txt .gitattributes && git commit -m "add export config"
+$ git add .git_archival.txt .gitattributes && git commit -m "add git archive support"
+```
+
+#### Understanding the warnings
+
+If you see warnings like these when building your package:
+
 ```
+UserWarning: git archive did not support describe output
+UserWarning: unprocessed git archival found (no export subst applied)
+```
+
+This typically happens when:
+
+1. **Building from working directory**: You're running `python -m build` directly in your repository
+2. **Sdist extraction**: A build tool extracts your sdist to build wheels, but the extracted directory isn't a git repository
+
+#### Recommended build workflows
+
+**For development builds:**
+Exclude `.git_archival.txt` from your package to avoid warnings:
+
+```{ .text file="MANIFEST.in"}
+# Exclude archival file from development builds
+exclude .git_archival.txt
+```
+
+**For release builds from archives:**
+Build from an actual git archive to ensure proper template expansion:
+
+```commandline
+# Create archive from a specific tag/commit
+$ git archive --output=../source_archive.tar v1.2.3
+$ cd ..
+$ tar -xf source_archive.tar
+$ cd extracted_directory/
+$ python -m build .
+```
+
+**For automated releases:**
+Many CI systems and package repositories (like GitHub Actions) automatically handle this correctly when building from git archives.
+
+#### Integration with package managers
+
+**MANIFEST.in exclusions:**
+```{ .text file="MANIFEST.in"}
+# Exclude development files from packages
+exclude .git_archival.txt
+exclude .gitattributes
+```
+
+
+```{ .text file=".gitattributes"}
+# Archive configuration
+.git_archival.txt  export-subst
+.gitignore         export-ignore
+```
+
+#### Troubleshooting
+
+**Problem: "unprocessed git archival found" warnings**
+- ✅ **Solution**: Add `exclude .git_archival.txt` to `MANIFEST.in` for development builds
+- ✅ **Alternative**: Build from actual git archives for releases
+
+**Problem: "git archive did not support describe output" warnings**
+- ℹ️ **Information**: This is expected when `.git_archival.txt` contains unexpanded templates
+- ✅ **Solution**: Same as above - exclude file or build from git archives
+
+**Problem: Version detection fails in git archives**
+- ✅ **Check**: Is `.gitattributes` configured with `export-subst`?
+- ✅ **Check**: Are you building from a properly created git archive?
+- ✅ **Check**: Does your git hosting provider support archive template expansion?
+
+!!! warning "Branch Names and Archive Stability"
+
+    Including `ref-names: $Format:%D$` in your `.git_archival.txt` can make archive checksums change when new commits are added to branches referenced in the archive. This primarily affects GitHub's automatic source archives. Use the stable format (without `ref-names`) unless you specifically need branch information and understand the stability implications.
 
+!!! note "Version Files"
 
-Note that if you are creating a `_version.py` file, note that it should not
-be kept in version control. It's strongly recommended to be put into gitignore.
+    If you are creating a `_version.py` file, it should not be kept in version control. Add it to `.gitignore`:
+    ```
+    # Generated version file
+    src/mypackage/_version.py
+    ```
 
 [git-archive-issue]: https://github.com/pypa/setuptools-scm/issues/806
 

src/setuptools_scm/_cli.py

@@ -5,6 +5,7 @@
 import os
 import sys
 
+from pathlib import Path
 from typing import Any
 
 from setuptools_scm import Configuration
@@ -106,6 +107,28 @@ def _get_cli_opts(args: list[str] | None) -> argparse.Namespace:
     # We avoid `metavar` to prevent printing repetitive information
     desc = "List information about the package, e.g. included files"
     sub.add_parser("ls", help=desc[0].lower() + desc[1:], description=desc)
+
+    # Add create-archival-file subcommand
+    archival_desc = "Create .git_archival.txt file for git archive support"
+    archival_parser = sub.add_parser(
+        "create-archival-file",
+        help=archival_desc[0].lower() + archival_desc[1:],
+        description=archival_desc,
+    )
+    archival_group = archival_parser.add_mutually_exclusive_group(required=True)
+    archival_group.add_argument(
+        "--stable",
+        action="store_true",
+        help="create stable archival file (recommended, no branch names)",
+    )
+    archival_group.add_argument(
+        "--full",
+        action="store_true",
+        help="create full archival file with branch information (can cause instability)",
+    )
+    archival_parser.add_argument(
+        "--force", action="store_true", help="overwrite existing .git_archival.txt file"
+    )
     return parser.parse_args(args)
 
 
@@ -116,6 +139,9 @@ def command(opts: argparse.Namespace, version: str, config: Configuration) -> in
     if opts.command == "ls":
         opts.query = ["files"]
 
+    if opts.command == "create-archival-file":
+        return _create_archival_file(opts, config)
+
     if opts.query == []:
         opts.no_version = True
         sys.stderr.write("Available queries:\n\n")
@@ -187,3 +213,79 @@ def _find_pyproject(parent: str) -> str:
     return os.path.abspath(
         "pyproject.toml"
     )  # use default name to trigger the default errors
+
+
+def _create_archival_file(opts: argparse.Namespace, config: Configuration) -> int:
+    """Create .git_archival.txt file with appropriate content."""
+    archival_path = Path(config.root, ".git_archival.txt")
+
+    # Check if file exists and force flag
+    if archival_path.exists() and not opts.force:
+        print(
+            f"Error: {archival_path} already exists. Use --force to overwrite.",
+            file=sys.stderr,
+        )
+        return 1
+
+    if opts.stable:
+        content = _get_stable_archival_content()
+        print("Creating stable .git_archival.txt (recommended for releases)")
+    elif opts.full:
+        content = _get_full_archival_content()
+        print("Creating full .git_archival.txt with branch information")
+        print("WARNING: This can cause archive checksums to be unstable!")
+
+    try:
+        archival_path.write_text(content, encoding="utf-8")
+        print(f"Created: {archival_path}")
+
+        gitattributes_path = Path(config.root, ".gitattributes")
+        needs_gitattributes = True
+
+        if gitattributes_path.exists():
+            # TODO: more nuanced check later
+            gitattributes_content = gitattributes_path.read_text("utf-8")
+            if (
+                ".git_archival.txt" in gitattributes_content
+                and "export-subst" in gitattributes_content
+            ):
+                needs_gitattributes = False
+
+        if needs_gitattributes:
+            print("\nNext steps:")
+            print("1. Add this line to .gitattributes:")
+            print("   .git_archival.txt  export-subst")
+            print("2. Commit both files:")
+            print("   git add .git_archival.txt .gitattributes")
+            print("   git commit -m 'add git archive support'")
+        else:
+            print("\nNext step:")
+            print("Commit the archival file:")
+            print("   git add .git_archival.txt")
+            print("   git commit -m 'update git archival file'")
+
+        return 0
+    except OSError as e:
+        print(f"Error: Could not create {archival_path}: {e}", file=sys.stderr)
+        return 1
+
+
+def _get_stable_archival_content() -> str:
+    """Generate stable archival file content (no branch names)."""
+    return """\
+node: $Format:%H$
+node-date: $Format:%cI$
+describe-name: $Format:%(describe:tags=true,match=*[0-9]*)$
+"""
+
+
+def _get_full_archival_content() -> str:
+    """Generate full archival file content with branch information."""
+    return """\
+# WARNING: Including ref-names can make archive checksums unstable
+# after commits are added post-release. Use only if describe-name is insufficient.
+node: $Format:%H$
+node-date: $Format:%cI$
+describe-name: $Format:%(describe:tags=true,match=*[0-9]*)$
+ref-names: $Format:%D$
+"""