feat(updating)!: introduce new update algorithm based on git merge

BREAKING CHANGE: The `-o,--conflict` flag is removed, as `git merge`
only supports inline conflicts. Omit the flag and corresponding API
parameter as of now.
BREAKING CHANGE: The `-c,--context-lines` flag is removed, as
`git merge` does not support configurable context sizes. Omit the flag
and corresponding API parameter as of now.
BREAKING CHANGE: Inline conflict marker labels are different, as
`git merge` does not support their customization.
BREAKING CHANGE: `gitattributes` settings affect Copier's internal
`git merge` call.

Author: sisp

copier/_cli.py

@@ -395,23 +395,6 @@ class CopierUpdateSubApp(_Subcommand):
         """
     )
 
-    conflict = cli.SwitchAttr(
-        ["-o", "--conflict"],
-        cli.Set("rej", "inline"),
-        default="inline",
-        help=(
-            "Behavior on conflict: Create .rej files, or add inline conflict markers."
-        ),
-    )
-    context_lines = cli.SwitchAttr(
-        ["-c", "--context-lines"],
-        int,
-        default=3,
-        help=(
-            "Lines of context to use for detecting conflicts. Increase for "
-            "accuracy, decrease for resilience."
-        ),
-    )
     defaults = cli.Flag(
         ["-l", "-f", "--defaults"],
         help="Use default answers to questions, which might be null if not specified.",
@@ -456,8 +439,6 @@ def inner() -> None:
                 overwrite=True,
                 pretend=self.pretend,
                 quiet=self.quiet,
-                conflict=cast(Literal["rej", "inline"], self.conflict),
-                context_lines=self.context_lines,
                 unsafe=self.unsafe,
                 skip_answered=self.skip_answered,
                 skip_tasks=self.skip_tasks,

copier/_main.py

@@ -5,7 +5,6 @@
 import errno
 import os
 import platform
-import re
 import stat
 import sys
 from collections.abc import Callable, Iterator
@@ -19,7 +18,6 @@
 
 import colorama
 from packaging.version import Version
-from pathspec.patterns.gitwildmatch import GitWildMatchPattern
 from pydantic import StrictBool
 
 colorama.just_fix_windows_console()
@@ -178,92 +176,6 @@ def handle_remove_readonly(
         raise
 
 
-_re_whitespace = re.compile(r"^\s+|\s+$")
-
-
-def normalize_git_path(path: str) -> str:
-    r"""Convert weird characters returned by Git to normal UTF-8 path strings.
-
-    A filename like âñ will be reported by Git as "\\303\\242\\303\\261" (octal
-    notation).
-    Similarly, a filename like "<tab>foo\b<lf>ar" will be reported as "\tfoo\\b\nar".
-    This can be disabled with `git config core.quotepath off`.
-
-    Args:
-        path: The Git path to normalize.
-
-    Returns:
-        str: The normalized Git path.
-    """
-    # Remove surrounding quotes
-    if path[0] == path[-1] == '"':
-        path = path[1:-1]
-    # Repair double-quotes
-    path = path.replace('\\"', '"')
-    # Unescape escape characters
-    path = path.encode("latin-1", "backslashreplace").decode("unicode-escape")
-    # Convert octal to utf8
-    return path.encode("latin-1", "backslashreplace").decode("utf-8")
-
-
-def escape_git_path(path: str) -> str:
-    """Escape paths that will be used as literal gitwildmatch patterns.
-
-    If the path was returned by a Git command, it should be unescaped completely.
-    ``normalize_git_path`` can be used for this purpose.
-
-    Args:
-        path: The Git path to escape.
-
-    Returns:
-        str: The escaped Git path.
-    """
-    # Prior to PathSpec v1.1.0, `GitWildMatchPattern.escape` does not escape backslashes
-    # or trailing whitespace.
-    # TODO: Remove this workaround when support for PathSpec prior to v1.1.0 is dropped.
-    if GitWildMatchPattern.escape("\\") == "\\":
-        path = path.replace("\\", "\\\\")
-    path = GitWildMatchPattern.escape(path)
-    return _re_whitespace.sub(
-        lambda match: "".join(f"\\{whitespace}" for whitespace in match.group()),
-        path,
-    )
-
-
-def get_git_objects_dir(path: Path) -> Path:
-    """Get the absolute path of a Git repository's objects directory."""
-    # FIXME: A lazy import is currently necessary to avoid circular imports with
-    # `errors.py`.
-    from ._vcs import get_git
-
-    git = get_git()
-    return path.joinpath(
-        git(
-            "-C",
-            path,
-            "rev-parse",
-            "--git-path",
-            "objects",
-        ).strip()
-    ).absolute()
-
-
-def set_git_alternates(*repos: Path, path: Path = Path()) -> None:
-    """Set Git alternates to borrow Git objects from other repositories.
-
-    Alternates are paths of other repositories' object directories written to
-    `$GIT_DIR/objects/info/alternates` and delimited by the newline character.
-
-    Args:
-        *repos: The paths of repositories from which to borrow Git objects.
-        path: The path of the repository where to set Git alternates. Defaults
-            to the current working directory.
-    """
-    alternates_file = get_git_objects_dir(path) / "info" / "alternates"
-    alternates_file.parent.mkdir(parents=True, exist_ok=True)
-    alternates_file.write_bytes(b"\n".join(map(bytes, map(get_git_objects_dir, repos))))
-
-
 def scantree(path: str, follow_symlinks: bool) -> Iterator[os.DirEntry[str]]:
     """A recursive extension of `os.scandir`."""
     for entry in os.scandir(path):

copier/_tools.py

@@ -835,43 +835,6 @@ will delete that folder.
 Copier will never delete the folder if it didn't create it. For this reason, when
 running `copier update`, this setting has no effect.
 
-!!! info
-
-    Not supported in `copier.yml`.
-
-### `conflict`
-
-- Format: `Literal["rej", "inline"]`
-- CLI flags: `-o`, `--conflict` (only available in `copier update` subcommand)
-- Default value: `inline`
-
-When updating a project, sometimes Copier doesn't know what to do with a diff code hunk.
-This option controls the output format if this happens. Using `rej`, creates `*.rej`
-files that contain the unresolved diffs. The `inline` option (default) includes the diff
-code hunk in the file itself, similar to the behavior of `git merge`.
-
-!!! info
-
-    Not supported in `copier.yml`.
-
-### `context_lines`
-
-- Format: `Int`
-- CLI flags: `-c`, `--context-lines` (only available in `copier update` subcommand)
-- Default value: `1`
-
-During a project update, Copier needs to compare the template evolution with the
-subproject evolution. This way, it can detect what changed, where and how to merge those
-changes. [Refer here for more details on this process](updating.md).
-
-The more lines you use, the more accurate Copier will be when detecting conflicts. But
-you will also have more conflicts to solve by yourself. FWIW, Git uses 3 lines by
-default.
-
-The less lines you use, the less conflicts you will have. However, Copier will not be so
-accurate and could even move lines around if the file it's comparing has several similar
-code chunks.
-
 !!! info
 
     Not supported in `copier.yml`.

docs/configuring.md

@@ -108,8 +108,6 @@ Attributes:
 | ------------------ | ------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `answers_file`     | `PurePath`                                                    | The path for [the answers file](configuring.md#the-copier-answersyml-file) relative to `dst_path`.<br>See the [`answers_file`](configuring.md#answers_file) setting for related information.                                                        |
 | `cleanup_on_error` | `bool`                                                        | When `True`, delete `dst_path` if there's an error.<br>See the [`cleanup_on_error`](configuring.md#cleanup_on_error) setting for related information.                                                                                               |
-| `conflict`         | `Literal["inline", "rej"]`                                    | The output format of a diff code hunk when [updating][updating-a-project] a file yields conflicts.<br>See the [`conflict`](configuring.md#conflict) setting for related information.                                                                |
-| `context_lines`    | `PositiveInt`                                                 | Lines of context to consider when solving conflicts in updates.<br>See the [`context_lines`](configuring.md#context_lines) setting for related information.                                                                                         |
 | `data`             | `dict[str, Any]`                                              | Answers to the questionnaire, defined in the template, provided via CLI (`-d,--data`) or API (`data`).<br>See the [`data`](configuring.md#data) setting for related information.<br>⚠️ May contain secret answers.                                  |
 | `defaults`         | `bool`                                                        | When `True`, use default answers to questions.<br>See the [`defaults`](configuring.md#defaults) setting for related information.                                                                                                                    |
 | `dst_path`         | `PurePath`                                                    | Destination path where to render the subproject.<br>⚠️ When [updating a project](updating.md), it may be a temporary directory, as Copier's update algorithm generates fresh copies using the old and new template versions in temporary locations. |

docs/creating.md

@@ -38,15 +38,11 @@ other Git ref you want.
 
 When updating, Copier will do its best to respect your project evolution by using the
 answers you provided when copied last time. However, sometimes it's impossible for
-Copier to know what to do with a diff code hunk. In those cases, copier handles the
-conflict in one of two ways, controlled with the `--conflict` option:
-
-- `--conflict rej`: Creates a separate `.rej` file for each file with conflicts. These
-    files contain the unresolved diffs.
-- `--conflict inline` (default): Updates the file with conflict markers. This is quite
-    similar to the conflict markers created when a `git merge` command encounters a
-    conflict. For more information, see the "Checking Out Conflicts" section of the
-    [`git` documentation](https://git-scm.com/book/en/v2/Git-Tools-Advanced-Merging).
+Copier to know how to merge the changes from the evolved template into the evolved
+project. In those cases, Copier updates a conflicting file with conflict markers in the
+same ways as a `git merge` command encounters conflicts; in fact, Copier uses
+`git merge` internally. For more information, see the "Checking Out Conflicts" section
+of the [`git` documentation](https://git-scm.com/book/en/v2/Git-Tools-Advanced-Merging).
 
 If the update results in conflicts, _you should review those manually_ before
 committing.
@@ -56,13 +52,11 @@ Git history, but if you aren't careful, it's easy to make mistakes.
 
 That's why the recommended way to prevent these mistakes is to add a
 [pre-commit](https://pre-commit.com/) (or equivalent) hook that forbids committing
-conflict files or markers. The recommended hook configuration depends on the `conflict`
-setting you use.
+conflict markers.
 
 ## Preventing Commit of Merge Conflicts
 
-If you use `--conflict inline` (the default) then you need to check for conflicts
-markers in your files:
+You need to check for conflict markers in your files:
 
 ```yaml title=".pre-commit-config.yaml"
 repos:
@@ -74,29 +68,6 @@ repos:
             args: [--assume-in-merge]
 ```
 
-If you use `--conflict rej` then you need to review and remove all generated `.rej`
-files:
-
-```yaml title=".pre-commit-config.yaml"
-repos:
-    - repo: local
-      hooks:
-          # Prevent committing .rej files
-          - id: forbidden-files
-            name: forbidden files
-            entry:
-                found Copier update rejection files; review and remove them before
-                merging.
-            language: fail
-            files: "\\.rej$"
-```
-
-!!! note
-
-    For projects that use both `rej` and `inline` depending on each user's preference,
-    you can add both hooks to your `pre-commit-config.yaml` file, making sure that no
-    unresolved merge conflicts are committed.
-
 ## Never change the answers file manually
 
 !!! important
@@ -151,53 +122,55 @@ graph TD
 
 %% nodes ----------------------------------------------------------
 template_repo("template repository")
-template_current("/tmp/template<br>(current tag)")
-template_latest("/tmp/template<br>(latest tag)")
+template_current("/tmp/template-old<br>(current tag)")
+template_latest("/tmp/template-new<br>(latest tag)")
 
-project_regen("/tmp/project<br>(fresh, current version)")
+project_regen_current("/tmp/project-old<br>(fresh, current version)")
+project_regen_latest("/tmp/project-new<br>(fresh, latest version)")
 project_current("current project")
 project_half("half migrated<br>project")
 project_updated("updated project")
-project_applied("updated project<br>(diff applied)")
 project_full("fully updated<br>and migrated project")
 
-update["update current<br>project in-place<br>(prompting)<br>+ run tasks again"]
-compare["compare to get diff"]
-apply["apply diff"]
-
-diff("diff")
+update["3-way merge<br>& run tasks again"]
+regen_current["generate and run tasks"]
+regen_latest["generate and run tasks"]
 
 %% edges ----------------------------------------------------------
         template_repo --> |git clone| template_current
         template_repo --> |git clone| template_latest
 
-     template_current --> |generate and run tasks| project_regen
-      project_current --> compare
+     template_current --> regen_current
+      project_current .-> |use answers| regen_current
+        regen_current --> project_regen_current
+      template_latest --> regen_latest
+         regen_latest --> project_regen_latest
       project_current --> |apply pre-migrations| project_half
-        project_regen --> compare
+         project_half .-> |use answers| regen_latest
          project_half --> update
-      template_latest --> update
+project_regen_current --> update
+ project_regen_latest --> update
                update --> project_updated
-              compare --> diff
-                 diff --> apply
-      project_updated --> apply
-                apply --> project_applied
-      project_applied --> |apply post-migrations| project_full
+      project_updated --> |apply post-migrations| project_full
 
 %% style ----------------------------------------------------------
 classDef blackborder stroke:#000;
-class compare,update,apply blackborder;
+class regen_current,regen_latest,update blackborder;
 ```
 
 As you can see here, `copier` does several things:
 
-- It regenerates a fresh project from the current template version.
-- Then, it compares both version to get the diff from "fresh project" to "current
-    project".
-- Now, it applies pre-migrations to your project, and updates the current project with
-    the latest template changes (asking for confirmation).
-- Finally, it re-applies the previously obtained diff and then runs the
-    post-migrations.
+- Regenerate the project fresh from the **current** template version, using the
+    project's existing answers – this becomes the merge-base.
+- Regenerate the project fresh from the **latest** template version, using the same
+    answers (with pre-migrations applied to the project beforehand).
+- Build a synthetic Git commit graph from these three states: the current-version
+    regeneration (common ancestor), the latest-version regeneration, and the actual
+    current project.
+- Perform a Git 3-way merge (using `git merge`) of the latest-version regeneration
+    into the current project, using the current-version regeneration as their common
+    ancestor – conflicts are marked like any normal `git merge` conflict.
+- Run post-migrations on the merged result to produce the fully updated project.
 
 ### Handling of deleted paths
 
@@ -242,10 +215,15 @@ branch. The following strategies won't work:
 
 - `git checkout <branch>` – _error: you need to resolve your current index first_
 - `git checkout .` – _error: path '&lt;filename&gt;' is unmerged_
-- `git merge --abort` – _fatal: There is no merge to abort (MERGE_HEAD missing)_
 
 Here is what you can do using Git in the terminal to throw away all changes:
 
+```shell
+git merge --abort
+```
+
+or
+
 ```shell
 git reset           # throw away merge conflict information
 git checkout .      # restore modified files

docs/updating.md

@@ -193,3 +193,28 @@ def git_init(message: str = "hello world") -> None:
     git("init")
     git("add", ".")
     git("commit", "-m", message)
+
+
+def normalize_git_path(path: str) -> str:
+    r"""Convert weird characters returned by Git to normal UTF-8 path strings.
+
+    A filename like âñ will be reported by Git as "\\303\\242\\303\\261" (octal
+    notation).
+    Similarly, a filename like "<tab>foo\b<lf>ar" will be reported as "\tfoo\\b\nar".
+    This can be disabled with `git config core.quotepath off`.
+
+    Args:
+        path: The Git path to normalize.
+
+    Returns:
+        str: The normalized Git path.
+    """
+    # Remove surrounding quotes
+    if path[0] == path[-1] == '"':
+        path = path[1:-1]
+    # Repair double-quotes
+    path = path.replace('\\"', '"')
+    # Unescape escape characters
+    path = path.encode("latin-1", "backslashreplace").decode("unicode-escape")
+    # Convert octal to utf8
+    return path.encode("latin-1", "backslashreplace").decode("utf-8")