redeploy 0.2.16__tar.gz → 0.2.18__tar.gz

{redeploy-0.2.16 → redeploy-0.2.18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: redeploy
-Version: 0.2.16
+Version: 0.2.18
 Summary: Infrastructure migration toolkit: detect → plan → apply
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -27,17 +27,17 @@ Requires-Dist: pfix>=0.1.60; extra == "dev"
 
 ## AI Cost Tracking
 
-![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.16-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
-![AI Cost](https://img.shields.io/badge/AI%20Cost-$5.25-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-11.3h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
+![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.18-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![AI Cost](https://img.shields.io/badge/AI%20Cost-$5.55-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-12.6h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
 
-- 🤖 **LLM usage:** $5.2500 (35 commits)
-- 👤 **Human dev:** ~$1128 (11.3h @ $100/h, 30min dedup)
+- 🤖 **LLM usage:** $5.5500 (37 commits)
+- 👤 **Human dev:** ~$1255 (12.6h @ $100/h, 30min dedup)
 
-Generated on 2026-04-20 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
+Generated on 2026-04-21 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
 
 ---
 
-![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.16-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.18-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
 
 Infrastructure migration and device deploy toolkit — VPS, Raspberry Pi kiosk, Podman Quadlet, k3s.
 
@@ -401,6 +401,135 @@ def notify_slack(ctx: PluginContext) -> None:
 | `params` | `dict` | Shortcut for `step.plugin_params` |
 | `dry_run` | `bool` | Skip side-effects if True |
 
+## Inline Scripts
+
+Execute multiline bash scripts directly from YAML without external files:
+
+```yaml
+extra_steps:
+  - id: configure_kiosk
+    action: inline_script
+    description: "Deploy kiosk launch script"
+    command: |
+      #!/bin/bash
+      mkdir -p ~/c2004/config
+      cat > ~/c2004/config/kiosk-launch.sh << 'EOF'
+      #!/bin/bash
+      if command -v chromium-browser >/dev/null 2>&1; then
+        chromium-browser --kiosk http://localhost:8100
+      elif command -v firefox >/dev/null 2>&1; then
+        firefox --kiosk http://localhost:8100
+      fi
+      EOF
+      chmod +x ~/c2004/config/kiosk-launch.sh
+    risk: medium
+    timeout: 60
+```
+
+The script is base64-encoded and executed via SSH with automatic temp file cleanup. Use `command` field for multiline script content (YAML `|` preserves newlines).
+
+### Script References (`command_ref`)
+
+Instead of duplicating scripts in YAML, reference a script defined in a markdown codeblock:
+
+```yaml
+extra_steps:
+  - id: configure_kiosk
+    action: inline_script
+    description: "Execute kiosk script from markdown"
+    command_ref: "#kiosk-browser-configuration-script"
+    risk: medium
+```
+
+In your migration markdown file, define the script in a section:
+
+```markdown
+## Kiosk Browser Configuration Script
+
+```bash
+#!/bin/bash
+# Auto-detect browser...
+if command -v chromium-browser >/dev/null 2>&1; then
+  chromium-browser --kiosk http://localhost:8100
+fi
+```
+```
+
+**Benefits:**
+- Single source of truth — script lives in one place (markdown codeblock)
+- No duplication between markdown documentation and YAML
+- Easy to read and maintain
+- Changes to the codeblock automatically apply to the deployment
+
+**Reference formats:**
+- `"#section-id"` — script from section in current spec file
+- `"./file.md#section-id"` — script from section in specific file
+
+The section ID is derived from the heading: spaces become hyphens, lowercase.
+Example: `## Kiosk Browser Configuration Script` → `#kiosk-browser-configuration-script`
+
+### Execute Script by Reference (`redeploy exec`)
+
+Run a single script from markdown without running the full migration:
+
+```bash
+# Execute script from codeblock on remote host
+redeploy exec '#kiosk-browser-configuration-script' \
+    --host pi@192.168.188.108 \
+    --file migration.podman-rpi5-resume.md
+
+# With file in reference
+redeploy exec './migration.md#install-deps' --host root@server.com
+
+# Using markpact:ref (more explicit)
+redeploy exec 'kiosk-script-id' --host pi@192.168.188.108 --file migration.md
+
+# Dry-run to preview script
+redeploy exec '#backup-script' --host pi@192.168.188.108 --file ops.md --dry-run
+```
+
+This is useful for:
+- One-off operations defined in markdown docs
+- Testing individual scripts before full migration
+- Running maintenance tasks
+
+### Execute Multiple Scripts (`redeploy exec-multi`)
+
+Test multiple scripts at once:
+
+```bash
+# Execute multiple scripts by ref
+redeploy exec-multi 'kiosk-script,install-deps,cleanup' \
+    --host pi@192.168.188.108 \
+    --file migration.md
+
+# Mix of markpact:ref and section headings
+redeploy exec-multi 'script1,#section2,script3' \
+    --host root@server.com \
+    --file deploy.md \
+    --dry-run
+```
+
+### Marking Codeblocks with `markpact:ref`
+
+For more explicit script identification, use `markpact:ref <id>` in codeblock:
+
+```markdown
+```bash markpact:ref kiosk-browser-configuration-script
+#!/bin/bash
+# Auto-detect browser...
+if command -v chromium-browser >/dev/null 2>&1; then
+  chromium-browser --kiosk http://localhost:8100
+fi
+```
+```
+
+Benefits of `markpact:ref`:
+- Explicit ID assignment (not derived from heading)
+- Multiple scripts per section
+- Can reference by simple ID instead of full heading
+- Self-documenting in markdown
+
 
 
 Place in project root — `redeploy run` (no args) uses it automatically.

{redeploy-0.2.16 → redeploy-0.2.18}/README.md

@@ -3,17 +3,17 @@
 
 ## AI Cost Tracking
 
-![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.16-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
-![AI Cost](https://img.shields.io/badge/AI%20Cost-$5.25-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-11.3h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
+![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.18-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![AI Cost](https://img.shields.io/badge/AI%20Cost-$5.55-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-12.6h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
 
-- 🤖 **LLM usage:** $5.2500 (35 commits)
-- 👤 **Human dev:** ~$1128 (11.3h @ $100/h, 30min dedup)
+- 🤖 **LLM usage:** $5.5500 (37 commits)
+- 👤 **Human dev:** ~$1255 (12.6h @ $100/h, 30min dedup)
 
-Generated on 2026-04-20 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
+Generated on 2026-04-21 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
 
 ---
 
-![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.16-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.18-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
 
 Infrastructure migration and device deploy toolkit — VPS, Raspberry Pi kiosk, Podman Quadlet, k3s.
 
@@ -377,6 +377,135 @@ def notify_slack(ctx: PluginContext) -> None:
 | `params` | `dict` | Shortcut for `step.plugin_params` |
 | `dry_run` | `bool` | Skip side-effects if True |
 
+## Inline Scripts
+
+Execute multiline bash scripts directly from YAML without external files:
+
+```yaml
+extra_steps:
+  - id: configure_kiosk
+    action: inline_script
+    description: "Deploy kiosk launch script"
+    command: |
+      #!/bin/bash
+      mkdir -p ~/c2004/config
+      cat > ~/c2004/config/kiosk-launch.sh << 'EOF'
+      #!/bin/bash
+      if command -v chromium-browser >/dev/null 2>&1; then
+        chromium-browser --kiosk http://localhost:8100
+      elif command -v firefox >/dev/null 2>&1; then
+        firefox --kiosk http://localhost:8100
+      fi
+      EOF
+      chmod +x ~/c2004/config/kiosk-launch.sh
+    risk: medium
+    timeout: 60
+```
+
+The script is base64-encoded and executed via SSH with automatic temp file cleanup. Use `command` field for multiline script content (YAML `|` preserves newlines).
+
+### Script References (`command_ref`)
+
+Instead of duplicating scripts in YAML, reference a script defined in a markdown codeblock:
+
+```yaml
+extra_steps:
+  - id: configure_kiosk
+    action: inline_script
+    description: "Execute kiosk script from markdown"
+    command_ref: "#kiosk-browser-configuration-script"
+    risk: medium
+```
+
+In your migration markdown file, define the script in a section:
+
+```markdown
+## Kiosk Browser Configuration Script
+
+```bash
+#!/bin/bash
+# Auto-detect browser...
+if command -v chromium-browser >/dev/null 2>&1; then
+  chromium-browser --kiosk http://localhost:8100
+fi
+```
+```
+
+**Benefits:**
+- Single source of truth — script lives in one place (markdown codeblock)
+- No duplication between markdown documentation and YAML
+- Easy to read and maintain
+- Changes to the codeblock automatically apply to the deployment
+
+**Reference formats:**
+- `"#section-id"` — script from section in current spec file
+- `"./file.md#section-id"` — script from section in specific file
+
+The section ID is derived from the heading: spaces become hyphens, lowercase.
+Example: `## Kiosk Browser Configuration Script` → `#kiosk-browser-configuration-script`
+
+### Execute Script by Reference (`redeploy exec`)
+
+Run a single script from markdown without running the full migration:
+
+```bash
+# Execute script from codeblock on remote host
+redeploy exec '#kiosk-browser-configuration-script' \
+    --host pi@192.168.188.108 \
+    --file migration.podman-rpi5-resume.md
+
+# With file in reference
+redeploy exec './migration.md#install-deps' --host root@server.com
+
+# Using markpact:ref (more explicit)
+redeploy exec 'kiosk-script-id' --host pi@192.168.188.108 --file migration.md
+
+# Dry-run to preview script
+redeploy exec '#backup-script' --host pi@192.168.188.108 --file ops.md --dry-run
+```
+
+This is useful for:
+- One-off operations defined in markdown docs
+- Testing individual scripts before full migration
+- Running maintenance tasks
+
+### Execute Multiple Scripts (`redeploy exec-multi`)
+
+Test multiple scripts at once:
+
+```bash
+# Execute multiple scripts by ref
+redeploy exec-multi 'kiosk-script,install-deps,cleanup' \
+    --host pi@192.168.188.108 \
+    --file migration.md
+
+# Mix of markpact:ref and section headings
+redeploy exec-multi 'script1,#section2,script3' \
+    --host root@server.com \
+    --file deploy.md \
+    --dry-run
+```
+
+### Marking Codeblocks with `markpact:ref`
+
+For more explicit script identification, use `markpact:ref <id>` in codeblock:
+
+```markdown
+```bash markpact:ref kiosk-browser-configuration-script
+#!/bin/bash
+# Auto-detect browser...
+if command -v chromium-browser >/dev/null 2>&1; then
+  chromium-browser --kiosk http://localhost:8100
+fi
+```
+```
+
+Benefits of `markpact:ref`:
+- Explicit ID assignment (not derived from heading)
+- Multiple scripts per section
+- Can reference by simple ID instead of full heading
+- Self-documenting in markdown
+
 
 
 Place in project root — `redeploy run` (no args) uses it automatically.

{redeploy-0.2.16 → redeploy-0.2.18}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "redeploy"
-version = "0.2.16"
+version = "0.2.18"
 description = "Infrastructure migration toolkit: detect → plan → apply"
 readme = "README.md"
 requires-python = ">=3.11"

{redeploy-0.2.16 → redeploy-0.2.18}/redeploy/__init__.py

@@ -7,7 +7,7 @@ Public API (stable from 0.2.0, semver guaranteed)::
 
 Everything not listed in ``__all__`` is internal and may change without notice.
 """
-__version__ = "0.2.16"
+__version__ = "0.2.18"
 
 # ── Core models ───────────────────────────────────────────────────────────────
 from .models import (  # noqa: F401

redeploy-0.2.18/redeploy/apply/__init__.py

@@ -0,0 +1,16 @@
+"""apply — Execute a MigrationPlan step by step."""
+from .executor import Executor
+from .state import (
+    DEFAULT_STATE_DIR,
+    ResumeState,
+    default_state_path,
+    state_key,
+)
+
+__all__ = [
+    "Executor",
+    "ResumeState",
+    "default_state_path",
+    "state_key",
+    "DEFAULT_STATE_DIR",
+]

{redeploy-0.2.16 → redeploy-0.2.18}/redeploy/apply/executor.py

@@ -15,6 +15,7 @@ from loguru import logger
 from ..detect.remote import RemoteProbe
 from ..models import MigrationPlan, MigrationStep, StepAction, StepStatus
 from ..plugins import PluginContext, registry as _plugin_registry
+from .state import ResumeState, default_state_path
 
 
 class ProgressEmitter:
@@ -124,7 +125,11 @@ class Executor:
                  progress_yaml: bool = False,
                  progress_stream: IO[str] = None,
                  audit_log: bool = True,
-                 audit_path: Optional["Path"] = None):
+                 audit_path: Optional["Path"] = None,
+                 resume: bool = False,
+                 from_step: Optional[str] = None,
+                 state_path: Optional["Path"] = None,
+                 spec_path: Optional[str] = None):
         self.plan = plan
         self.dry_run = dry_run
         self.probe = RemoteProbe(plan.host)
@@ -138,10 +143,43 @@ class Executor:
         self._audit_path = audit_path
         self._t0: float = 0.0
 
+        # ── spec path for command_ref resolution ─────────────────────────────
+        if spec_path and not plan.spec_path:
+            plan.spec_path = spec_path
+
+        # ── resume / checkpoint ──────────────────────────────────────────────
+        self._resume = resume
+        self._from_step = from_step
+        spec_id = spec_path or plan.spec_path or plan.target_file or plan.infra_file or plan.app
+        if state_path is None and not dry_run:
+            state_path = default_state_path(spec_id, plan.host)
+        self._state_path: Optional[Path] = (
+            Path(state_path) if state_path is not None else None
+        )
+        self._state: Optional[ResumeState] = None
+        if self._state_path is not None and not dry_run:
+            self._state = ResumeState.load_or_new(
+                self._state_path,
+                spec_path=str(spec_id),
+                host=plan.host,
+                total_steps=len(plan.steps),
+            )
+            # Keep total_steps in sync if the spec changed between runs.
+            self._state.total_steps = len(plan.steps)
+
     @property
     def completed_steps(self) -> list[MigrationStep]:
         return list(self._completed)
 
+    @property
+    def state(self) -> Optional[ResumeState]:
+        """Current ResumeState (None when dry_run or disabled)."""
+        return self._state
+
+    @property
+    def state_path(self) -> Optional[Path]:
+        return self._state_path
+
     def run(self) -> bool:
         """Execute all steps. Returns True if all passed."""
         self._t0 = time.monotonic()
@@ -151,19 +189,34 @@ class Executor:
         if self._emitter:
             self._emitter.start(self.plan)
 
+        skip_ids = self._compute_skip_set()
+        if skip_ids:
+            logger.info(f"resume: skipping {len(skip_ids)} already-completed step(s): "
+                        f"{', '.join(sorted(skip_ids))}")
+
         ok = True
         for i, step in enumerate(self.plan.steps, 1):
+            if step.id in skip_ids:
+                step.status = StepStatus.SKIPPED
+                step.result = "resumed: previously completed"
+                if self._emitter:
+                    self._emitter.step_done(i, step)
+                continue
             try:
                 if self._emitter:
                     self._emitter.step_start(i, step)
                 self._execute_step(step)
                 self._completed.append(step)
+                if self._state is not None:
+                    self._state.mark_done(step.id)
                 if self._emitter:
                     self._emitter.step_done(i, step)
             except StepError as e:
                 logger.error(f"Step failed: {e}")
                 step.status = StepStatus.FAILED
                 step.error = str(e)
+                if self._state is not None:
+                    self._state.mark_failed(step.id, str(e))
                 if self._emitter:
                     self._emitter.step_fail(i, step, str(e))
                     self._emitter.failed(len(self._completed), len(self.plan.steps), str(e))
@@ -177,9 +230,40 @@ class Executor:
             logger.info(f"{'[DRY RUN] ' if self.dry_run else ''}All {len(self.plan.steps)} steps completed")
             if self._emitter:
                 self._emitter.done(len(self.plan.steps))
+            # Plan finished cleanly — drop the checkpoint.
+            if self._state is not None:
+                self._state.remove()
+        else:
+            if self._state is not None and self._state_path is not None:
+                logger.info(f"resume: checkpoint saved → {self._state_path} "
+                            f"({self._state.completed_count}/{self._state.total_steps} done)")
         self._write_audit(ok=ok, elapsed_s=elapsed)
         return ok
 
+    def _compute_skip_set(self) -> set[str]:
+        """Determine which step ids should be skipped this run.
+
+        Sources (in priority order):
+          1. ``--from-step <id>``: skip every step BEFORE that id.
+          2. ``--resume`` + persisted state: skip every previously completed id.
+        """
+        skip: set[str] = set()
+        ids = [s.id for s in self.plan.steps]
+
+        if self._from_step:
+            if self._from_step not in ids:
+                logger.warning(
+                    f"--from-step '{self._from_step}' not found in plan; running full plan"
+                )
+            else:
+                idx = ids.index(self._from_step)
+                skip.update(ids[:idx])
+
+        if self._resume and self._state is not None:
+            skip.update(self._state.completed_step_ids)
+
+        return skip
+
     def _write_audit(self, *, ok: bool, elapsed_s: float) -> None:
         if not self._audit_log:
             return
@@ -219,6 +303,7 @@ class Executor:
             StepAction.VERSION_CHECK:       self._run_version_check,
             StepAction.WAIT:                self._run_wait,
             StepAction.PLUGIN:              self._run_plugin,
+            StepAction.INLINE_SCRIPT:       self._run_inline_script,
         }
 
         handler = dispatch.get(step.action)
@@ -456,16 +541,104 @@ class Executor:
         step.status = StepStatus.DONE
         step.result = f"waited {total}s"
 
+    def _run_inline_script(self, step: MigrationStep) -> None:
+        """Execute multiline bash script via SSH using base64 encoding."""
+        import base64
+        from pathlib import Path
+
+        script = step.command
+        
+        # If command_ref is set, extract script from markdown file
+        if step.command_ref:
+            script = self._resolve_command_ref(step.command_ref, step)
+        
+        if not script:
+            raise StepError(step, "inline_script requires command field or command_ref with script content")
+
+        # Base64 encode the script to safely pass it through SSH
+        encoded = base64.b64encode(script.encode()).decode()
+        timeout = step.timeout or 300
+
+        # Create temp file, decode script, run it, then clean up
+        cmd = (
+            f"tmpfile=$(mktemp) && "
+            f"echo '{encoded}' | base64 -d > \"$tmpfile\" && "
+            f"chmod +x \"$tmpfile\" && "
+            f"\"$tmpfile\"; "
+            f"rc=$?; "
+            f"rm -f \"$tmpfile\"; "
+            f"exit $rc"
+        )
+
+        r = self.probe.run(cmd, timeout=timeout)
+        step.result = r.out[:500] if r.out else "script executed"
+        if not r.ok:
+            raise StepError(step, f"script failed with exit={r.exit_code}: {r.stderr[:200]}")
+        step.status = StepStatus.DONE
+
+    def _resolve_command_ref(self, command_ref: str, step: MigrationStep) -> str:
+        """Resolve command_ref to script content from markdown file.
+        
+        command_ref formats:
+        - "./file.md#section-id" - script from section in specific file
+        - "#section-id" - script from section in current spec file
+        - "#kiosk-browser-configuration-script" - markpact:ref block
+        """
+        from pathlib import Path
+        from ..markpact.parser import extract_script_from_markdown, extract_script_by_ref
+        
+        # Parse command_ref
+        if "#" in command_ref:
+            file_part, section_id = command_ref.split("#", 1)
+            file_path = file_part if file_part else getattr(self.plan, 'spec_path', None)
+        else:
+            section_id = command_ref
+            file_path = getattr(self.plan, 'spec_path', None)
+        
+        if not file_path:
+            raise StepError(step, f"Cannot resolve command_ref '{command_ref}': no file path available")
+        
+        file_path = Path(file_path)
+        if not file_path.exists():
+            raise StepError(step, f"Command ref file not found: {file_path}")
+        
+        # Extract script from markdown - try both methods:
+        # 1. First try markpact:ref codeblock (new format)
+        # 2. Then try section heading (old format)
+        markdown_content = file_path.read_text(encoding="utf-8")
+        
+        # Try markpact:ref format first
+        script = extract_script_by_ref(markdown_content, section_id, language="bash")
+        
+        # Fallback to section heading format
+        if script is None:
+            script = extract_script_from_markdown(markdown_content, section_id, language="bash")
+        
+        if script is None:
+            raise StepError(step, f"Could not find bash script for ref '{section_id}' in {file_path} (tried markpact:ref and section heading)")
+        
+        return script
+
     # ── rollback ──────────────────────────────────────────────────────────────
 
     def _rollback(self) -> None:
         logger.warning("Rolling back completed steps...")
+        rolled_back: list[str] = []
         for step in reversed(self._completed):
             if step.rollback_command:
                 logger.info(f"  ↩ rollback [{step.id}]: {step.rollback_command}")
                 r = self.probe.run(step.rollback_command, timeout=120)
                 if not r.ok:
                     logger.warning(f"    rollback failed: {r.stderr[:100]}")
+                else:
+                    rolled_back.append(step.id)
+        # Forget rolled-back steps so a subsequent --resume re-executes them.
+        if rolled_back and self._state is not None:
+            self._state.completed_step_ids = [
+                sid for sid in self._state.completed_step_ids
+                if sid not in set(rolled_back)
+            ]
+            self._state.save()
 
     # ── summary ───────────────────────────────────────────────────────────────