android-test: stable-self failures are WARN only, not blockers

Published F-Droid remains the hard gate (FAIL_FAST). Self-build of the
release tag continues the matrix and does not alone fail overall exit.

scripts/taler-monitoring/android-test/GUI-AUTOMATION-NOTES.md

@@ -243,14 +243,14 @@ Scripts **today** implement Android only; they should honour:
 
 ### Build variants (anti–“weird local build”)
 
-| Variant | Source | Package | Purpose |
+| Variant | Source | Severity | Purpose |
-|---------|--------|---------|---------|
+|---------|--------|----------|---------|
-| **published** | F-Droid APK URL | `net.taler.wallet.fdroid` | What users install |
+| **stable-self** | tag `wallet-1.6.1` rebuilt here | **WARN only** | Same release tree, our Gradle — never hard-blocks the matrix |
-| **stable-self** | tag `wallet-1.6.1` (override `STABLE_TAG`) rebuilt here | `…fdroid.debug` | Same release tree as stable, **our** Gradle — excludes odd publisher-only or dirty-tree builds |
+| **published** | F-Droid APK URL | **BLOCKER** | What users install; FAIL_FAST stops here |
-| **master** | `origin/master` rebuilt here | `…fdroid.debug` | Tip of mainline |
+| **master** | `origin/master` | soft fail | Tip of mainline |
-| **fix** (optional) | inference fix branch | `…fdroid.debug` | GOA automation fixes |
+| **fix** (optional) | inference fix branch | soft fail | GOA automation fixes |
 
-Default matrix: `published,stable-self,master`.  
+Default matrix order: `stable-self,published,master`.  
 `./run-android-variant-matrix.sh` — see README.
 
 ---

scripts/taler-monitoring/android-test/README.md

@@ -36,18 +36,18 @@ Runs several smokes so a single odd developer build cannot be the only sample:
 
 | Variant | What |
 |---------|------|
-| Order | Variant | What |
+| Order | Variant | Severity | What |
-|------|---------|------|
+|------|---------|----------|------|
-| **1 (gate)** | **stable-self** | Rebuild tag `wallet-1.6.1` on this host |
+| **1** | **stable-self** | **WARN only** (not a blocker) | Rebuild tag `wallet-1.6.1` on this host |
-| **2 (gate)** | **published** | F-Droid stable APK (`net.taler.wallet.fdroid` 1.6.1 / 854) |
+| **2** | **published** | **BLOCKER** | F-Droid stable APK (`net.taler.wallet.fdroid` 1.6.1 / 854) |
-| **3** | **master** | Rebuild `origin/master` (only if gates pass when `FAIL_FAST=1`) |
+| **3** | **master** | fail counts / FAIL_FAST | Rebuild `origin/master` |
 
-**Default:** `FAIL_FAST=1` — first non-zero gate stops the matrix (no master if stable fails).
+**Default:** `FAIL_FAST=1` only stops on **blocker** failures (`published`).  
+`stable-self` failure → warning in `SUMMARY.txt`, matrix continues, exit code unaffected by that warn alone.
 
 ```bash
 ./run-android-variant-matrix.sh
 STACK=goa ./run-android-variant-matrix.sh
-# only the hard gates:
 VARIANTS=stable-self,published FAIL_FAST=1 ./run-android-variant-matrix.sh
 ```
 

scripts/taler-monitoring/android-test/run-android-variant-matrix.sh

@@ -2,22 +2,23 @@
 # Run Android wallet smoke against **several build variants** so a single
 # "weird" developer tree or a bad local build cannot be the only sample.
 #
-# Default matrix (gate order — fail → hard stop when FAIL_FAST=1):
+# Default matrix order:
 #   1) **stable-self** — rebuild release **tag** (same source as published)
+#        Failure → **WARN only** (not a hard blocker; continue matrix)
 #   2) **published**   — F-Droid stable APK (what users install)
-#   3) **master**      — rebuild origin/master (only if 1–2 pass with FAIL_FAST)
+#        Failure → **BLOCKER** (matrix exit ≠ 0; FAIL_FAST stops here)
+#   3) **master**      — rebuild origin/master
+#        Failure → counts toward overall fail; does not skip earlier gates
 #
 # Why stable-self in addition to published?
-#   Rare, but a published APK *or* a one-off developer build can be odd
+#   Isolates "our toolchain rebuild of the release tag" vs "F-Droid binary".
-#   (wrong flags, dirty tree, wrong ABI). Rebuilding the release tag on *this*
+#   Rebuild issues (signing flavor, dirty host, low-RAM ANR on debug) must not
-#   machine with the same script path as master isolates "our toolchain" vs
+#   block the published-stable gate — hence warn-only for stable-self.
-#   "F-Droid binary" vs "current master source".
 #
 # Usage:
 #   ./run-android-variant-matrix.sh
 #   STACK=goa ./run-android-variant-matrix.sh
 #   VARIANTS=master,stable-self ./run-android-variant-matrix.sh
-#   VARIANTS=published,stable-self,master,fix GUI=1 ./run-android-variant-matrix.sh
 #   STABLE_TAG=wallet-1.6.1 MASTER_BRANCH=master ./run-android-variant-matrix.sh
 #
 # Env:
@@ -27,7 +28,7 @@
 #   FIX_BRANCH       inference fix branch (variant "fix")
 #   VARIANTS         comma list: published,stable-self,master,fix
 #   GUI=1            use gui-smoke instead of pay-smoke after each build
-#   FAIL_FAST=1      stop on first non-zero smoke (default 0 = run all)
+#   FAIL_FAST=1      stop on first **blocker** failure (default 1)
 #   MATRIX_OUT       summary dir (default out-matrix-<timestamp>)
 #
 set -euo pipefail
@@ -40,7 +41,7 @@ ROOT=$(cd "$(dirname "$0")" && pwd)
 : "${STABLE_APK_NAME:=net.taler.wallet.fdroid_854.apk}"
 : "${MASTER_BRANCH:=master}"
 : "${FIX_BRANCH:=dev/hernani-inference/fix-bank-withdraw-auto-exchange}"
-# Gate: stable self-build → published F-Droid; then master. FAIL_FAST default on.
+# Order: stable-self (warn) → published (blocker) → master.
 : "${VARIANTS:=stable-self,published,master}"
 : "${FAIL_FAST:=1}"
 : "${GUI:=0}"
@@ -55,6 +56,7 @@ SUMMARY="$MATRIX_OUT/SUMMARY.txt"
 echo "=== Android variant matrix ===" | tee -a "$SUMMARY"
 echo "STACK=$STACK  GUI=$GUI  headless=${EMULATOR_HEADLESS}  variants=$VARIANTS" | tee -a "$SUMMARY"
 echo "STABLE_TAG=$STABLE_TAG  MASTER_BRANCH=$MASTER_BRANCH" | tee -a "$SUMMARY"
+echo "severity: stable-self=WARN  published=BLOCKER  master=fail-count" | tee -a "$SUMMARY"
 echo "out: $MATRIX_OUT" | tee -a "$SUMMARY"
 
 # Ensure device once (headless AVD by default)
@@ -67,13 +69,26 @@ echo "device: $SERIAL" | tee -a "$SUMMARY"
 
 IFS=',' read -r -a VLIST <<<"$VARIANTS"
 declare -a RESULTS=()
+declare -a WARNINGS=()
 overall=0
 
+# Severity: warn | blocker | soft
+#   warn     — log WARN, never FAIL_FAST, never overall≠0
+#   blocker  — FAIL_FAST eligible; overall≠0
+#   soft     — overall≠0 on fail, but FAIL_FAST only if FAIL_FAST=1 and not warn
+variant_severity() {
+  case "$1" in
+    stable-self|stable|self-stable|release-self) echo warn ;;
+    published|fdroid|stable-published) echo blocker ;;
+    *) echo soft ;;
+  esac
+}
+
 run_published() {
   local out="$MATRIX_OUT/published"
   mkdir -p "$out"
   echo "" | tee -a "$SUMMARY"
-  echo "-------- variant: published (F-Droid stable binary) --------" | tee -a "$SUMMARY"
+  echo "-------- variant: published (F-Droid stable binary) [BLOCKER] --------" | tee -a "$SUMMARY"
   local smoke="$ROOT/run-android-pay-smoke.sh"
   [ "$GUI" = "1" ] && smoke="$ROOT/run-android-gui-smoke.sh"
   set +e
@@ -90,17 +105,23 @@ run_published() {
   local rc=$?
   set -e
   echo "published exit=$rc" | tee -a "$SUMMARY"
-  echo "published $rc" >>"$MATRIX_OUT/results.tsv"
+  echo "published $rc blocker" >>"$MATRIX_OUT/results.tsv"
   return "$rc"
 }
 
 run_selfbuild() {
   local label="$1"
   local ref="$2"
+  local sev="${3:-soft}"
   local out="$MATRIX_OUT/$label"
   mkdir -p "$out"
   echo "" | tee -a "$SUMMARY"
-  echo "-------- variant: $label (self-build ref=$ref) --------" | tee -a "$SUMMARY"
+  echo "-------- variant: $label (self-build ref=$ref) [$(echo "$sev" | tr 'a-z' 'A-Z')] --------" | tee -a "$SUMMARY"
+  # Warn-only self-build: still collect evidence; SMOKE_STRICT optional (default off for warn)
+  local strict="${SMOKE_STRICT:-1}"
+  if [ "$sev" = "warn" ]; then
+    strict="${SMOKE_STRICT_SELF:-0}"
+  fi
   set +e
   env STACK="$STACK" \
     BRANCH="$ref" \
@@ -111,12 +132,12 @@ run_selfbuild() {
     ANDROID_SERIAL="$SERIAL" \
     AUTO_START_EMULATOR=0 \
     PKG=net.taler.wallet.fdroid.debug \
-    SMOKE_STRICT="${SMOKE_STRICT:-1}" \
+    SMOKE_STRICT="$strict" \
     "$ROOT/run-android-build-and-smoke.sh"
   local rc=$?
   set -e
-  echo "$label (ref=$ref) exit=$rc" | tee -a "$SUMMARY"
+  echo "$label (ref=$ref) exit=$rc severity=$sev" | tee -a "$SUMMARY"
-  echo "$label $rc $ref" >>"$MATRIX_OUT/results.tsv"
+  echo "$label $rc $ref $sev" >>"$MATRIX_OUT/results.tsv"
   if [ -f "$out/build-meta.txt" ]; then
     echo "build-meta:" | tee -a "$SUMMARY"
     cat "$out/build-meta.txt" | tee -a "$SUMMARY"
@@ -128,31 +149,54 @@ for raw in "${VLIST[@]}"; do
   v=$(echo "$raw" | tr -d '[:space:]')
   [ -n "$v" ] || continue
   rc=0
+  sev=$(variant_severity "$v")
   case "$v" in
     published|fdroid|stable-published)
       run_published || rc=$?
       ;;
     stable-self|stable|self-stable|release-self)
-      run_selfbuild "stable-self" "$STABLE_TAG" || rc=$?
+      run_selfbuild "stable-self" "$STABLE_TAG" warn || rc=$?
       ;;
     master|main)
-      run_selfbuild "master" "$MASTER_BRANCH" || rc=$?
+      run_selfbuild "master" "$MASTER_BRANCH" soft || rc=$?
       ;;
     fix|fix-branch|inference)
-      run_selfbuild "fix" "$FIX_BRANCH" || rc=$?
+      run_selfbuild "fix" "$FIX_BRANCH" soft || rc=$?
       ;;
     *)
-      # treat as raw git ref / label
+      run_selfbuild "ref-${v//\//-}" "$v" soft || rc=$?
-      run_selfbuild "ref-${v//\//-}" "$v" || rc=$?
       ;;
   esac
-  RESULTS+=("$v:$rc")
+
   if [ "$rc" -ne 0 ]; then
-    overall=1
+    case "$sev" in
-    if [ "$FAIL_FAST" = "1" ]; then
+      warn)
-      echo "FAIL_FAST: stopping after $v rc=$rc" | tee -a "$SUMMARY"
+        echo "WARN: variant $v failed rc=$rc — self-build only, NOT a blocker; continuing" | tee -a "$SUMMARY"
-      break
+        WARNINGS+=("$v:$rc")
-    fi
+        RESULTS+=("$v:WARN($rc)")
+        # do not set overall; do not FAIL_FAST
+        ;;
+      blocker)
+        echo "BLOCKER: variant $v failed rc=$rc" | tee -a "$SUMMARY"
+        RESULTS+=("$v:BLOCKER($rc)")
+        overall=1
+        if [ "$FAIL_FAST" = "1" ]; then
+          echo "FAIL_FAST: stopping after blocker $v" | tee -a "$SUMMARY"
+          break
+        fi
+        ;;
+      soft)
+        echo "FAIL: variant $v rc=$rc (counts toward overall; not stable-self)" | tee -a "$SUMMARY"
+        RESULTS+=("$v:FAIL($rc)")
+        overall=1
+        if [ "$FAIL_FAST" = "1" ]; then
+          echo "FAIL_FAST: stopping after $v rc=$rc" | tee -a "$SUMMARY"
+          break
+        fi
+        ;;
+    esac
+  else
+    RESULTS+=("$v:OK")
   fi
 done
 
@@ -161,10 +205,13 @@ echo "=== matrix done (overall=$overall) ===" | tee -a "$SUMMARY"
 for r in "${RESULTS[@]}"; do
   echo "  $r" | tee -a "$SUMMARY"
 done
+if [ "${#WARNINGS[@]}" -gt 0 ]; then
+  echo "warnings (non-blocking):" | tee -a "$SUMMARY"
+  for w in "${WARNINGS[@]}"; do
+    echo "  WARN $w" | tee -a "$SUMMARY"
+  done
+fi
 echo ""
 echo "Compare OUT_DIR trees under $MATRIX_OUT"
-echo "Intent: published vs stable-self should behave similarly;"
+echo "stable-self failure = WARN only; published failure = BLOCKER."
-echo "        master may diverge (new features / fixes)."
-echo "A odd stable-self vs published suggests toolchain or unsigned-debug differences;"
-echo "a odd master-only failure suggests tip breakage, not F-Droid stable."
 exit "$overall"