comprobar el esquema de CloudKit antes de publicar

Un campo que esta en el @Model pero no en el esquema desplegado no sincroniza, y no falla nada: los datos se quedan en el dispositivo que los escribio. Asi se colaron nueve campos fuera de Production entre la 2.0 y la 2.1.2 sin que nadie lo notara, incluidas las reglas de dia de las etiquetas y el desayuno/merienda activados. scripts/check_cloudkit_schema.py compara las propiedades almacenadas de cada @Model con los CD_<campo> del esquema real (exportado con cktool), saltando relaciones, computadas y el sufijo _ckAsset de los binarios. fastlane submit y release abortan si falta algo; beta solo avisa, porque en TestFlight es normal que el deploy a Production aun no se haya hecho. La lane check_schema lo ejecuta suelto. El deploy a Production sigue siendo manual: cktool no tiene subcomando y Apple bloquea el endpoint de esquema en ese entorno. Refs #38 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013su1ttRiMeMYxkZJ1Y3246
2026-09-14 17:23:22 +02:00
parent 8b75f496f8
commit 75c41b12ec
3 changed files with 194 additions and 0 deletions
@@ -52,6 +52,43 @@ pass show admob/mealmood/app-id
 pass show admob/mealmood/banner-home-unit-id
 ```

+### Antes de mandar una versión a la App Store: comprobar el esquema de CloudKit
+
+Un campo que está en el `@Model` pero no en el esquema de CloudKit **no
+sincroniza**, y no falla nada: los datos se quedan en el dispositivo que los
+escribió. Así se colaron nueve campos fuera de Production entre la 2.0 y la
+2.1.2 (issue #38) — desayuno/merienda activados, sus horas, el recordatorio,
+las fotos del planificador y las reglas de día de las etiquetas no viajaban
+entre dispositivos.
+
+```bash
+fastlane check_schema                          # o directamente:
+python3 scripts/check_cloudkit_schema.py --environment production
+```
+
+`fastlane submit` y `fastlane release` ya lo ejecutan y **abortan** si falta
+algo; `fastlane beta` solo avisa (en TestFlight aún puede faltar el deploy).
+
+Si faltan campos: el esquema de development solo se actualiza al ejecutar la
+app contra ese entorno con iCloud activo, o importándolo a mano:
+
+```bash
+TOKEN=$(pass show apple/mealmood/cloudkit-token)
+TEAM=$(pass show apple/mealmood/developer-team-id)
+xcrun cktool export-schema --token "$TOKEN" --team-id "$TEAM" \
+  --container-id iCloud.com.alexandrev.mealmood --environment development \
+  --output-file /tmp/schema.ckdb
+# añadir los CD_<campo> que falten (Bool/Int → INT64, String → STRING,
+# Date → TIMESTAMP, Data → CD_<campo>_ckAsset ASSET) y:
+xcrun cktool import-schema --token "$TOKEN" --team-id "$TEAM" \
+  --container-id iCloud.com.alexandrev.mealmood --environment development \
+  --file /tmp/schema.ckdb
+```
+
+El paso a Production **no se puede automatizar**: `cktool` no tiene deploy y
+Apple bloquea el endpoint de esquema en ese entorno. Lo hace Alexandre en
+CloudKit Console → Schema → *Deploy Schema Changes* → Deploy to Production.
+
 ### TestFlight upload
 ```bash
 export FASTLANE_APPLE_APPLICATION_SPECIFIC_PASSWORD=$(pass show apple/mealmood/app-specific-password-fastlane)
@@ -25,8 +25,29 @@ platform :ios do
    )
  end

+  # A field missing from the CloudKit schema doesn't crash anything — it just
+  # never syncs, silently. Nine of them drifted out of Production between 2.0
+  # and 2.1.2 before anyone noticed (issue #38), so shipping now checks first.
+  def check_cloudkit_schema(blocking:)
+    script = File.expand_path("../scripts/check_cloudkit_schema.py", __dir__)
+    ok = system("python3", script, "--environment", "production")
+    return if ok
+
+    if blocking
+      UI.user_error!("CloudKit Production schema is behind the models — deploy it before shipping (see the output above).")
+    else
+      UI.important("CloudKit Production schema is behind the models. Fine for TestFlight, but deploy it before submitting to the App Store.")
+    end
+  end
+
+  desc "Check the deployed CloudKit schema against the SwiftData models"
+  lane :check_schema do
+    check_cloudkit_schema(blocking: true)
+  end
+
  desc "Push a new beta build to TestFlight"
  lane :beta do
+    check_cloudkit_schema(blocking: false)
    increment_build_number(xcodeproj: "MealMood.xcodeproj")
    # No `sdk:` here: forcing one SDK globally breaks the watchOS targets,
    # which must build against watchos within the same archive.
@@ -166,6 +187,7 @@ platform :ios do
  desc "Submit an already-uploaded build for App Store review (automatic release)"
  lane :submit do |options|
    UI.user_error!("pass build:<number>") unless options[:build]
+    check_cloudkit_schema(blocking: true)
    version = options[:version] || shipped_app_version
    ensure_editable_version(version)
    upload_to_app_store(
@@ -239,6 +261,7 @@ platform :ios do

  desc "Push a new release build to the App Store"
  lane :release do
+    check_cloudkit_schema(blocking: true)
    increment_build_number(xcodeproj: "MealMood.xcodeproj")
    build_app(
      scheme: "MealMood",
@@ -0,0 +1,134 @@
+#!/usr/bin/env python3
+"""Compares the SwiftData models against the deployed CloudKit schema.
+
+Every stored property of a @Model becomes a CD_<name> field in CloudKit, and a
+field missing from an environment simply doesn't sync: the app keeps working
+and the data silently stays on the device that wrote it. That is how nine
+fields drifted out of Production between 2.0 and 2.1.2 (issue #38).
+
+The schema only picks up new fields when the app runs against the development
+environment, and promoting them to Production is a manual step in CloudKit
+Console. So this check has to run before shipping, not after.
+
+    python3 scripts/check_cloudkit_schema.py [--environment production]
+
+Exits non-zero when a stored property has no field in the schema.
+"""
+
+import argparse
+import re
+import subprocess
+import sys
+from pathlib import Path
+
+REPO = Path(__file__).resolve().parent.parent
+MODELS_DIR = REPO / "MealMood" / "Models"
+CONTAINER = "iCloud.com.alexandrev.mealmood"
+
+# Properties CloudKit does not mirror as a field of their own record type.
+IGNORED = {
+    # To-one relationships live on the child as CD_<name>; to-many ones have no
+    # field at all (the inverse carries the link).
+    "slots",
+    "weekPlan",
+}
+
+
+def pass_show(entry: str) -> str:
+    return subprocess.run(
+        ["pass", "show", entry], capture_output=True, text=True, check=True
+    ).stdout.splitlines()[0].strip()
+
+
+def export_schema(environment: str) -> str:
+    """Pulls the live schema. Needs the management token from `pass`."""
+    token = pass_show("apple/mealmood/cloudkit-token")
+    team = pass_show("apple/mealmood/developer-team-id")
+    result = subprocess.run(
+        [
+            "xcrun", "cktool", "export-schema",
+            "--token", token,
+            "--team-id", team,
+            "--container-id", CONTAINER,
+            "--environment", environment,
+        ],
+        capture_output=True, text=True,
+    )
+    if result.returncode != 0:
+        raise SystemExit(f"cktool export-schema failed:\n{result.stderr.strip()}")
+    return result.stdout
+
+
+def model_names() -> set:
+    names = set()
+    for path in MODELS_DIR.glob("*.swift"):
+        names.update(re.findall(r"@Model\s+final class (\w+)", path.read_text()))
+    return names
+
+
+def stored_properties(source: str, models: set) -> list:
+    """Stored `var`s of a @Model: computed ones end in `{`, and relationships to
+    other models are not fields of this record type."""
+    body = source.split("init(")[0]
+    properties = []
+    for match in re.finditer(r"^\s*(?:@Attribute\([^)]*\)\s*)?var (\w+)\s*:\s*([^\n=]+?)(\s*=|\s*\{|$)", body, re.M):
+        name, type_name, tail = match.group(1), match.group(2).strip(), match.group(3)
+        if tail.strip() == "{":
+            continue  # computed
+        if name in IGNORED:
+            continue
+        bare = type_name.strip("[]?").strip()
+        if bare in models:
+            continue  # relationship
+        properties.append(name)
+    return properties
+
+
+def schema_fields(schema: str, record_type: str) -> set:
+    block = re.search(rf"RECORD TYPE CD_{record_type} \((.*?)\);", schema, re.S)
+    if not block:
+        return set()
+    # Binary attributes land as CD_<name>_ckAsset.
+    return {
+        field.removesuffix("_ckAsset")
+        for field in re.findall(r"CD_(\w+)", block.group(1))
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--environment", default="production", choices=["production", "development"])
+    args = parser.parse_args()
+
+    schema = export_schema(args.environment)
+    models = model_names()
+    problems = []
+
+    for path in sorted(MODELS_DIR.glob("*.swift")):
+        source = path.read_text()
+        for model in re.findall(r"@Model\s+final class (\w+)", source):
+            fields = schema_fields(schema, model)
+            if not fields:
+                problems.append(f"{model}: record type CD_{model} missing from the schema")
+                continue
+            for prop in stored_properties(source, models):
+                if prop not in fields:
+                    problems.append(f"{model}.{prop} → CD_{prop} missing")
+
+    if problems:
+        print(f"CloudKit schema ({args.environment}) is behind the models:\n")
+        for problem in problems:
+            print(f"  ✗ {problem}")
+        print(
+            "\nThese fields will not sync. Run the app once against development so the\n"
+            "schema picks them up (or import it with `cktool import-schema`), then deploy\n"
+            "to Production from CloudKit Console → Schema → Deploy Schema Changes."
+        )
+        return 1
+
+    print(f"✅ CloudKit schema ({args.environment}) matches the models.")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())