snapshot of https://github.com/mattermost/mattermost.git (master, 2026-09-16; blobs > 131072B omitted)

server/scripts/check_backport_migrations.sh

@@ -0,0 +1,78 @@
+#!/usr/bin/env bash
+#
+# Backport safety check (Postgres only).
+#
+# A release branch is an older subset of master. When a PR backports a migration
+# it must keep the exact version number and name that migration already has on
+# master; renumbering or renaming a shipped migration breaks the upgrade path
+# (the MM-68848 class of bug).
+#
+# We therefore only look at the migrations this PR ADDS on top of its base branch
+# and require each of them to already exist on master. Files that are already on
+# the base branch are left untouched, so historical states such as the Postgres
+# migrations renamed before the pre-migration infra landed do not produce false
+# positives.
+#
+# MySQL is intentionally skipped: support was dropped after v11.0, so older
+# release branches still carry MySQL migrations that no longer exist on master.
+#
+# Portable across the bash 3.2 that ships with macOS and Linux CI: uses only
+# POSIX-ish tools plus process substitution, no bash 4 features.
+#
+set -euo pipefail
+export LC_ALL=C
+
+# Branch the PR targets. The migrations the PR adds are everything HEAD has that
+# this ref does not.
+base_ref="${MM_MIGRATION_CHECK_BASE_REF:-origin/master}"
+# Canonical source of shipped migrations the additions must match.
+canonical_ref="${MM_MIGRATION_CHECK_CANONICAL_REF:-origin/master}"
+
+repo_root="$(git rev-parse --show-toplevel)"
+# Postgres only; MySQL migrations linger on old release branches and are skipped.
+migrations_dir="server/channels/db/migrations/postgres"
+
+for ref in "$base_ref" "$canonical_ref"; do
+    if ! git -C "$repo_root" rev-parse --verify --quiet "${ref}^{commit}" >/dev/null; then
+        echo "Ref '$ref' not found. Fetch it first (e.g. 'git fetch origin master')" >&2
+        echo "or set MM_MIGRATION_CHECK_BASE_REF / MM_MIGRATION_CHECK_CANONICAL_REF." >&2
+        exit 2
+    fi
+done
+
+# path -> "version_name.kind"; the up/down kind is part of the identity so a
+# one-sided rename can't be masked by its surviving partner file.
+norm='s#^'"$migrations_dir"'/([0-9]+_.+)\.(up|down)\.sql$#\1.\2#p'
+
+ref_files() {
+    git -C "$repo_root" ls-tree -r --name-only "$1" -- "$migrations_dir" \
+        | sed -nE "$norm" | sort -u
+}
+
+head_files() {
+    # Tracked + untracked files actually present on disk, so a plain `mv`
+    # without `git add` is still caught.
+    git -C "$repo_root" ls-files --cached --others --exclude-standard -- "$migrations_dir" \
+        | while IFS= read -r path; do
+              [ -f "$repo_root/$path" ] && printf '%s\n' "$path"
+          done \
+        | sed -nE "$norm" | sort -u
+}
+
+# Migrations this PR adds on top of its base branch.
+added="$(comm -23 <(head_files) <(ref_files "$base_ref"))"
+
+# Of those, the ones that don't exist on the canonical branch are renames,
+# renumbers, or otherwise stray.
+stray="$(comm -23 <(printf '%s\n' "$added" | sed '/^$/d') <(ref_files "$canonical_ref"))"
+
+if [ -n "$stray" ]; then
+    count="$(printf '%s\n' "$stray" | wc -l | tr -d '[:space:]')"
+    echo "Found $count migration file(s) added by this branch that do not exist on $canonical_ref:" >&2
+    printf '%s\n' "$stray" | while IFS= read -r m; do
+        echo "  - ${m}.sql is added on this branch but not present on $canonical_ref; renaming or renumbering a shipped migration breaks upgrades. Add a new migration instead." >&2
+    done
+    exit 1
+fi
+
+echo "All migrations added by this branch exist on $canonical_ref."