helly25 · Nov 1, 2026 · Nov 1, 2026 · Nov 1, 2026 · Nov 1, 2026 · Nov 1, 2026
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -23,6 +23,8 @@ repos:
     rev: v5.0.0
     hooks:
       - id: check-added-large-files
+        # Retain the exact lossless cross-platform audit evidence archive.
+        exclude: ^docs/benchmarks/executor-copy-audit-20261101\.json\.gz$
       - id: check-merge-conflict
       - id: check-yaml
       # BINARY_TYPES below: archive containers are binary but can START with text (a phar begins with

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -3,6 +3,8 @@
 
 # 0.11.0
 
+- Avoid temporary collection-name and comparison-category strings while retaining owned keys.
+
 # 0.10.0
 
 - Verify staged release executables generate Git-applicable comparison patches for binary and empty-file additions, removals, and updates, including reverse application.

diff --git a/TODO.md b/TODO.md
@@ -361,6 +361,12 @@ Measured follow-ups (see the P07/P08 sections of `docs/performance-analysis.md`)
 
 ## Performance optimization follow-up
 
+- [ ] Qualify executor admission after the clock correction with separate cold/warm controls and
+      16/32/63/64-entry matching boundaries; check Linux and macOS workload regressions before
+      promoting shared executors, caller-inclusive budgets or bounded directory frontiers.
+- [ ] Revisit recycling matcher batch buffers only when repeatable gains justify the additional API;
+      retain ownership and thread-safety tests plus benchmarks with identical bytes as negative controls.
+
 - [ ] Benchmark collection arena block policies: 4 KiB -> 64 KiB (current), 64 KiB -> 1 MiB,
       and 4 KiB -> 1 MiB. Compare allocation counts, CPU time, unused capacity and peak memory
       for small/large collections on Linux and macOS before choosing new defaults.

diff --git a/docs/benchmarks/executor-copy-audit-20261101.json.gz b/docs/benchmarks/executor-copy-audit-20261101.json.gz
diff --git a/docs/benchmarks/executor-copy-audit-20261101.md b/docs/benchmarks/executor-copy-audit-20261101.md
diff --git a/docs/performance-analysis.md b/docs/performance-analysis.md
@@ -2731,3 +2731,48 @@ results; all five macOS targets execute. Full CI remains required before merging
 The macOS samples, binary hashes, invocation settings, output oracles and probe
 source are retained in
 [`abseil-steady-clock-macos-20261031.json.gz`](benchmarks/abseil-steady-clock-macos-20261031.json.gz).
+
+## Executor and allocation audit after the clock correction (2026-11-01)
+
+Keep main's runtime scheduling. The [complete executor/copy audit](benchmarks/executor-copy-audit-20261101.md)
+compares all seven runtime/measurement PRs, their actual dependencies, manageable combinations and
+isolated mechanism controls on macOS and Linux. The initial native matrix has 53 cases and three
+sessions on each platform, with independent output oracles; supplemental native and in-memory
+measurements have separate contracts. The [lossless evidence bundle](benchmarks/executor-copy-audit-20261101.json.gz)
+retains all ten report populations, complete source patches, drivers, identities and provenance corrections.
+
+Reject #991's current four-millisecond directory-admission policy and #1001's current remaining-work
+cost model. A separate matching guard inherited by #992/#996 accumulates 8,192 entries and samples
+16 entries against a 500-microsecond threshold. This guard survives excluding #991's directory policy.
+Removing it restores near-baseline clustered-content and line-output performance on both platforms;
+it does not establish that executor reuse is a universal improvement or restore main's cold-batch policy.
+
+Command-wide execution and bounded read-ahead have useful Linux listing/metadata/comparison results,
+but retain material workload and platform regressions. At three workers, the 65,536-file long-name
+collection case regresses by 39.6% for #996 and 36.3% for #998 on macOS. The #992-only eager control
+also has rg-selection counterexamples. Keep these runtime integrations conditional rather than
+promoting a replacement policy from the favorable cells alone.
+
+Bounded read-ahead has a substantial measured memory benefit on both platforms. In the
+65,536-file metadata case at three workers, median whole-process peak RSS is:
+
+| Platform |      Main |      #996 |     #997 |     #998 |
+| :------- | --------: | --------: | -------: | -------: |
+| Linux    | 20.81 MiB | 20.06 MiB | 6.25 MiB | 6.39 MiB |
+| macOS    | 18.09 MiB | 17.34 MiB | 5.44 MiB | 5.47 MiB |
+
+The #997 frontier reduces peak RSS by about 69% relative to #996 on both platforms;
+#998 retains that improvement. This proves a benefit for this fixture, rather than a
+worst-case bound for arbitrary directory width, path size, archives or storage behavior.
+
+Retain C01/C03: avoid temporary collection-name and comparison-category strings while preserving
+owned keys and caller-storage lifetimes. This is a narrow construction/ownership improvement, without
+a universal wall-clock claim. Defer C02's matcher batch-buffer recycling API until repeatable gains
+justify its additional lifetime transition. Byte-identical collector binaries provide a negative control;
+small timing differences between those labels cannot be attributed to recycling.
+
+Infrastructure is isolated in [PR #1035](https://github.com/mboworks/xff/pull/1035): the query downloader
+repair and executor diagnostic use private test-only support. The shipping CLI is identical to baseline
+on Linux; all non-debug Mach-O sections match on macOS. Its pre-sync full CI passed after one targeted
+retry of a runner disk-capacity failure, with no source/head change. The subsequent main synchronization
+has a separate current-head CI context. Runtime experiments remain unmerged.
diff --git a/release-site.json b/release-site.json
@@ -4,6 +4,7 @@
     "CHANGELOG.md": "CHANGELOG.html",
     "CONTRIBUTING.md": "CONTRIBUTING.html",
     "docs/benchmarks/execution-resources.md": "docs/benchmarks/execution-resources.html",
+    "docs/benchmarks/executor-copy-audit-20261101.md": "docs/benchmarks/executor-copy-audit-20261101.html",
     "docs/benchmarks/executor-dispatch-probe.md": "docs/benchmarks/executor-dispatch-probe.html",
     "docs/execution-resources.md": "docs/execution-resources.html",
     "docs/ci-performance.md": "docs/ci-performance.html",

diff --git a/xff/cli/BUILD.bazel b/xff/cli/BUILD.bazel
@@ -1264,6 +1264,7 @@ bashtest(
 bashtest(
     name = "rg_test",
     size = "small",
+    timeout = "moderate",  # Repeated frontend subprocesses exceed 60 seconds under MSan.
     srcs = ["rg_test.sh"],
     data = ["//xff/cli/testing:xff"],
 )

diff --git a/xff/engine/BUILD.bazel b/xff/engine/BUILD.bazel
@@ -166,6 +166,7 @@ cc_test(
         "//xff/vfs:vfs_cc",
         "@abseil-cpp//absl/status:statusor",
         "@googletest//:gtest_main",
+        "@mboworks_mbo//mbo/testing:matchers_cc",
         "@mboworks_mbo//mbo/testing:status_cc",
     ],
 )

diff --git a/xff/engine/collect.cc b/xff/engine/collect.cc
@@ -67,7 +67,7 @@ bool Collections::Add(std::string_view name, const Visit& visit) {
   }
   rows_ += 1;
   bytes_ += entry_bytes;
-  by_name_[std::string(name)].push_back(
+  by_name_.try_emplace(name).first->second.push_back(
       Entry{
           .path = CopyText(visit.path),
           .name = CopyText(visit.name),

diff --git a/xff/engine/collect_test.cc b/xff/engine/collect_test.cc
@@ -22,6 +22,7 @@
 
 #include "gmock/gmock.h"
 #include "gtest/gtest.h"
+#include "mbo/testing/matchers.h"
 #include "mbo/testing/status.h"
 #include "xff/engine/walk.h"
 #include "xff/parser/parser.h"
@@ -30,6 +31,7 @@
 namespace xff::engine {
 namespace {
 
+using ::mbo::testing::EqualsText;
 using ::testing::AllOf;
 using ::testing::ElementsAre;
 using ::testing::Eq;
@@ -80,6 +82,21 @@ TEST_F(CollectTest, AddStoresUnderTheNamedCollection) {
   EXPECT_THAT(entry.metadata.size, Eq(42));
 }
 
+TEST_F(CollectTest, RepeatedLongNamesOwnTheirKeyBeyondCallerStorage) {
+  const vfs::Metadata metadata{.type = vfs::FileType::kRegular};
+  const std::string expected_name(100, 'n');
+  Collections collections;
+  {
+    std::string name = expected_name;
+    EXPECT_THAT(collections.Add(name, MakeVisit("./a", "a", ".", metadata)), IsTrue());
+    EXPECT_THAT(collections.Add(name, MakeVisit("./b", "b", ".", metadata)), IsTrue());
+    name.assign(100, 'x');
+  }
+  EXPECT_THAT(collections.Names(), ElementsAre(EqualsText(expected_name)));
+  EXPECT_THAT(collections.Entries(expected_name), SizeIs(2));
+  EXPECT_THAT(collections.Entries(expected_name).back().path, EqualsText("./b"));
+}
+
 TEST_F(CollectTest, AsVisitRebuildsTheEntryForASink) {
   const vfs::Metadata md{.type = vfs::FileType::kRegular, .size = 7};
   Collections collections;

diff --git a/xff/engine/run.cc b/xff/engine/run.cc
@@ -4574,7 +4574,7 @@ RunResult RunTreeCompare(
     return RunResult{.errors = std::max(left_run_result.errors, right_result.errors)};
   }
 
-  const auto record_category = [&](const std::string& path, std::string_view category) {
+  const auto record_category = [&](std::string_view path, std::string_view category) {
     if (side_summaries.at(0).per_entry || side_summaries.at(1).per_entry) {
       categories.emplace(path, category);
     }
@@ -4624,7 +4624,7 @@ RunResult RunTreeCompare(
       const std::size_t category = !pair.left ? 1 : !pair.right ? 0 : result->same ? 3 : 2;
       const auto category_name = kComparisonCategories.at(category);
       counts.Add(category, pair.left, pair.right);
-      record_category(std::string(pair.path), category_name);
+      record_category(pair.path, category_name);
       different |= category != 3;
       if (!selected.at(category)) {
         continue;