release-train: staging -> main

profile/README.md

@@ -36,7 +36,7 @@ Your data stays on your infrastructure. Fine-tuned weights stay on your infrastr
 | | |
 |---|---|
 | **[model-zoo](https://github.com/tracebloc/model-zoo)** | Pre-built models for vision, NLP, tabular, time series — ready to train |
-| **[start-training](https://github.com/tracebloc/start-training)** | Jupyter notebook to launch training in minutes |
+| **[quickstart](https://github.com/tracebloc/quickstart)** | Jupyter notebook to launch training in minutes |
 | **[data-ingestors](https://github.com/tracebloc/data-ingestors)** | Pipelines to validate, prepare, and ingest your datasets |
 | **[client](https://github.com/tracebloc/client)** | Deploy the tracebloc workspace on your Kubernetes cluster |
 

scripts/tests/bugbot-gate-mutations.py

@@ -81,6 +81,15 @@
     ("a repeating cursor is followed for ever instead of refused",
      '            if not cursor or cursor in seen_cursors or pages_left <= 0:',
      '            if not cursor:'),
+    # The regression the staging-hop Bugbot named on #464: the pageInfo self-check
+    # going blind for ONE connection. The selftest's stripper loop used to `break`
+    # after `checkSuites`, and its `reviewThreads` branch pasted a fixed indentation
+    # that matched the checkSuites line first -- so this row was UNCAUGHT until
+    # the loop was made to visit every member (measured before the fix landed).
+    ("the pageInfo self-check walks only the first connection, so reviewThreads' "
+     "cursor can be dropped from the query unnoticed",
+     '    for name in PAGED_TOPLEVEL:\n        match = re.search(name + r"\\(first:',
+     '    for name in list(PAGED_TOPLEVEL)[:1]:\n        match = re.search(name + r"\\(first:'),
 
     # --- (A) the load-bearing claim: a TERMINAL verdict on THIS head --------
     ("a missing Bugbot verdict reports PASS instead of UNCLAIMED",

scripts/tests/bugbot-gate-selftest.py

@@ -911,15 +911,54 @@ def _threads(n, start=0):
 # The query must keep asking for pageInfo, or none of the above ever runs live.
 check("the real QUERY asks both top-level connections for pageInfo",
       gate.connections_missing_pageinfo() == [], "missing=%r" % gate.connections_missing_pageinfo())
+
+# EVERY CONNECTION, AND THE RIGHT ONE. The first version of this loop ended in a
+# `break`, so only the first member of PAGED_TOPLEVEL (`checkSuites`) was ever
+# exercised, and its `reviewThreads` branch pasted a fixed 6-space indentation
+# into `str.replace` -- a needle that is a substring of the 14-space-indented
+# `checkSuites` line and so, had it ever run, would have stripped checkSuites'
+# pageInfo a second time and let the detector "pass" by naming the wrong
+# connection (measured: `connections_missing_pageinfo` answered
+# `['checkSuites']` for the reviewThreads branch). A self-check that went blind
+# for `reviewThreads` alone passed this suite. Same shape as the totalCount
+# stripper above, with three things pinned:
+#   1. the members are ALSO written down as literals, because a loop over the
+#      module's own dict cannot see a member being removed from it;
+#   2. the stripper is anchored on the connection's own name and is indentation-
+#      agnostic, and its substitution count is asserted, so it cannot hit the
+#      neighbour or silently strip nothing;
+#   3. the detector must name EXACTLY the connection stripped -- the other one,
+#      or both, is a wrong answer, not a pass;
+# and the loop's visit list is compared to the literals afterwards, so a `break`
+# (or a `continue` past the asserts) reddens the suite instead of shrinking it.
+PAGED_TOPLEVEL_LITERALS = ("checkSuites", "reviewThreads")
+for name in PAGED_TOPLEVEL_LITERALS:
+    check(
+        "%r is declared a paged top-level connection" % name,
+        name in gate.PAGED_TOPLEVEL,
+        "PAGED_TOPLEVEL = %r" % (list(gate.PAGED_TOPLEVEL),),
+    )
+pageinfo_visited = []
 for name in gate.PAGED_TOPLEVEL:
-    stripped = re.sub(r"pageInfo \{ hasNextPage endCursor \}\n", "", gate.QUERY, count=1) \
-        if name == "checkSuites" else gate.QUERY.replace("      pageInfo { hasNextPage endCursor }\n", "", 1)
-    # Whichever occurrence the stripper removed, the guard must name AT LEAST one
-    # connection -- the assertion is that the stripper applied and was seen.
-    check("the pageInfo stripper actually applied", stripped != gate.QUERY)
-    check("dropping pageInfo from the query is detected",
-          gate.connections_missing_pageinfo(stripped) != [], "guard stayed silent for %r" % name)
-    break
+    stripped, applied = re.subn(
+        r"(" + name + r"\(first:\s*\d+[^)]*\)\s*\{[^{]*?)pageInfo\s*\{[^}]*\}\s*", r"\1", gate.QUERY
+    )
+    check(
+        "the pageInfo stripper actually applied to %r" % name,
+        applied == 1,
+        "%d substitution(s) -- the anchor no longer matches the query" % applied,
+    )
+    check(
+        "dropping pageInfo from %r is detected, and %r alone is named" % (name, name),
+        gate.connections_missing_pageinfo(stripped) == [name],
+        "detector said %r" % (gate.connections_missing_pageinfo(stripped),),
+    )
+    pageinfo_visited.append(name)
+check(
+    "the pageInfo stripper loop visited every paged top-level connection",
+    sorted(pageinfo_visited) == sorted(PAGED_TOPLEVEL_LITERALS),
+    "visited %r, expected %r" % (pageinfo_visited, list(PAGED_TOPLEVEL_LITERALS)),
+)
 
 # 7. severity_of, directly.
 # --------------------------------------------------------------------------

scripts/tests/extract-advanced-prs-selftest.sh

@@ -26,6 +26,19 @@ no() { printf '  FAIL  %s\n     %s\n' "$1" "$2"; fail=$((fail + 1)); }
 assert_out()     { if grep -qF -- "$2" <<<"$3"; then ok "$1"; else no "$1" "expected: $2 -- got: $3"; fi; }
 assert_not_out() { if grep -qF -- "$2" <<<"$3"; then no "$1" "must NOT contain: $2 -- got: $3"; else ok "$1"; fi; }
 
+# result_line <out> -> ONLY the script's "Found PRs: ..." line, the attribution
+# itself. The "is NOT attributed" checks below must grep THIS, never the whole
+# output: the script also prints `::warning::commit <40-hex sha>: ...`, and the
+# fixture commits are minted at run time (random content + the wall clock), so
+# their shas are random hex. A negative grep for "901" over the whole output
+# therefore fails whenever the sha happens to contain "901" -- which is what
+# reddened `selftests` on main at eb89c1e5 (sha e57d5bafc191901afac...) with the
+# extractor itself correct. ~1% of runs per three-digit needle: a coin the
+# suite tossed on every push. An absent result line is returned as-is (empty),
+# so a negative assertion on it still passes only when the positive
+# "Found PRs" assertion next to it is what carries the check.
+result_line() { grep -- '^Found PRs:' <<<"$1" || true; }
+
 # A `gh` stub. `gh api repos/<repo>/commits/<sha>/pulls --jq <expr>`:
 #   * $STUB_DIR/<sha>.fail present -> exit 1 (a FAILED read, e.g. a 403)
 #   * else apply <expr> with real jq over $STUB_DIR/<sha>.json (empty array if none)
@@ -93,7 +106,7 @@ a="$(commit "$root" 'fix(tests): bound the k3d cleanup in all seven e2e EXIT tra
 printf '[{"number":985,"merged_at":"2026-09-07T08:22:52Z","base":{"ref":"develop"},"head":{"ref":"fix/979-k3d-cleanup"}}]\n' >"$STUB_DIR/${a}.json"
 out="$(run_extract "$root" "$base" "$a" "$bin")"
 assert_out     "client#985: the develop-based feature PR is attributed on a staging hop" "Found PRs: 985" "$out"
-assert_not_out "client#985: the subject issue is NOT used"                               "979"            "$out"
+assert_not_out "client#985: the subject issue is NOT used"                               "979"            "$(result_line "$out")"
 
 # ---------------------------------------------------------------------------
 # FIXTURE 2: tracebloc-engine#914 — subject names ticket backend#3013 in the
@@ -106,7 +119,7 @@ b="$(commit "$root" 'sec(deps): torch 2.13.0 + torchvision 0.28.0 on the cu129 i
 printf '[{"number":914,"merged_at":"2026-09-07T00:00:00Z","base":{"ref":"develop"},"head":{"ref":"sec/3013-torch"}}]\n' >"$STUB_DIR/${b}.json"
 out="$(run_extract "$root" "$base" "$b" "$bin")"
 assert_out     "engine#914: the develop-based feature PR is attributed on a staging hop" "Found PRs: 914" "$out"
-assert_not_out "engine#914: the subject ticket is NOT used"                              "3013"           "$out"
+assert_not_out "engine#914: the subject ticket is NOT used"                              "3013"           "$(result_line "$out")"
 
 # ---------------------------------------------------------------------------
 # PROMOTION PR EXCLUDED: a commit whose /pulls names a merged release-train/*
@@ -118,7 +131,7 @@ base="$(make_repo "$root")"
 p="$(commit "$root" 'chore(promote): develop -> staging (#900)')"
 printf '[{"number":901,"merged_at":"2026-09-07T00:00:00Z","base":{"ref":"staging"},"head":{"ref":"release-train/develop-to-staging"}}]\n' >"$STUB_DIR/${p}.json"
 out="$(run_extract "$root" "$base" "$p" "$bin")"
-assert_not_out "promotion: the release-train PR is not attributed"       "901"            "$out"
+assert_not_out "promotion: the release-train PR is not attributed"       "901"            "$(result_line "$out")"
 assert_out     "promotion: it falls through to the subject"              "Found PRs: 900" "$out"
 
 # ---------------------------------------------------------------------------