From 605bd10a051c2fc6bc3a02d336aa2bc23a2bf94a Mon Sep 17 00:00:00 2001
From: speak-agent <248744407+speak-agent@users.noreply.github.com>
Date: Fri, 4 Sep 2026 20:13:02 +0800
Subject: [PATCH] =?UTF-8?q?test(e2e):=20340=20=E5=8F=AA=E6=95=B0=E5=AE=83?=
 =?UTF-8?q?=E8=87=AA=E5=B7=B1=E9=82=A3=E4=B8=AA=E7=9B=AE=E5=BD=95=E9=87=8C?=
 =?UTF-8?q?=E7=9A=84=20ninja?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

THE PREDICATE WAS RIGHT AND THE OBJECT WAS WRONG.

`count_ninja` counted every ninja on the machine. A baseline taken before the
build covers one that was ALREADY running; it does not cover one that appears
during the window, and a build server, a second checkout or a parallel test can
start one at any moment.

Measured in an ecosystem sandbox, which shares the host's PID namespace: an
unrelated session's ninja, in a different project entirely, made this assertion
fail on a build that has the fix.

Counting is now scoped by the process's working directory to the test's own
`$work` tree. Verified that scoping did not weaken it: the test still FAILS
against the released 2026.9.4.2, naming the orphan and its directory.
---
 .../e2e/340_no_orphan_survives_a_killed_mcpp.sh | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/tests/e2e/340_no_orphan_survives_a_killed_mcpp.sh b/tests/e2e/340_no_orphan_survives_a_killed_mcpp.sh
index 9851ec57..e481d809 100755
--- a/tests/e2e/340_no_orphan_survives_a_killed_mcpp.sh
+++ b/tests/e2e/340_no_orphan_survives_a_killed_mcpp.sh
@@ -33,12 +33,23 @@ work="$(mktemp -d)"; trap 'rm -rf "$work"' EXIT
 cd "$work"
 mkdir -p src
 
-# Counting by executable name, not by a `pgrep -f` pattern: this script's own
-# command line contains the word, and a pattern match finds itself.
+# COUNTING BY EXECUTABLE NAME AND BY WORKING DIRECTORY, AND BOTH HALVES MATTER.
+#
+# Not by a `pgrep -f` pattern: this script's own command line contains the word,
+# and a pattern match finds itself.
+#
+# And not every ninja on the machine: a build server, a developer's second
+# checkout, or another test running in parallel can start one inside this
+# check's window, and a baseline taken before the build does not cover a process
+# that appears during it. Measured in an ecosystem sandbox — which shares the
+# host's PID namespace — where an unrelated session's ninja in a different
+# project failed this assertion. The predicate was right and the object was
+# wrong.
 count_ninja() {
   local n=0 p
   for p in /proc/[0-9]*; do
-    [[ "$(cat "$p/comm" 2>/dev/null)" == "ninja" ]] && n=$((n+1))
+    [[ "$(cat "$p/comm" 2>/dev/null)" == "ninja" ]] || continue
+    case "$(readlink "$p/cwd" 2>/dev/null)" in "$work"*) n=$((n+1));; esac
   done
   echo "$n"
 }