From e0574a5fa79d97efe2e262229a5377534d4b7acb Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Mon, 20 Jul 2026 22:44:14 +0000
Subject: [PATCH 1/2] Add -XX:+EnableDynamicAgentLoading to test argLine (all
 reactor modules)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Silences the JDK 21 'A Java agent has been loaded dynamically' warning
emitted when a test-time agent self-attaches at runtime with no static
-javaagent configured — in the core srcmorph module the Lincheck
instrumentation agent (AiGenerationKindLincheckTest runs in the default
surefire run), plus the jcstress/vmlens agents under their opt-in
profiles. That warning is written to the native stderr, outside
Surefire's wrapped System.err, and can intermittently corrupt the encoded
fork channel — surfacing as 'Corrupted channel by directly writing to
native stream' followed by a bogus 'There was a timeout in the fork' even
though every test passed.

Applied identically to all three module poms to keep the reactor's
surefire argLine uniform; mirrors the same guard added to the sibling
BitcoinAddressFinder repo (there triggered by Mockito's inline mock maker
self-attaching byte-buddy).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y84TKMNfhbFGi7EoCEvP99
---
 srcmorph-cli/pom.xml          | 13 ++++++++++++-
 srcmorph-maven-plugin/pom.xml | 14 +++++++++++++-
 srcmorph/pom.xml              | 17 ++++++++++++++++-
 3 files changed, 41 insertions(+), 3 deletions(-)

diff --git a/srcmorph-cli/pom.xml b/srcmorph-cli/pom.xml
index 467c39a..543dca0 100644
--- a/srcmorph-cli/pom.xml
+++ b/srcmorph-cli/pom.xml
@@ -210,7 +210,18 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log</argLine>
+                        <!--
+                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
+                            been loaded dynamically" warning that a test-time self-attaching agent
+                            (no static -javaagent) emits to the *native* stderr, which can
+                            intermittently corrupt Surefire's encoded fork channel ("Corrupted
+                            channel by directly writing to native stream" → bogus "There was a
+                            timeout in the fork"). This module runs no such agent today, but the
+                            flag is kept identical across the reactor's module poms and mirrors the
+                            guard in the core srcmorph module (Lincheck) and the sibling
+                            BitcoinAddressFinder repo (Mockito inline mock maker).
+                        -->
+                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             Force classpath-only test execution. Surefire's default module-path
                             auto-detection activates as soon as module-info.class exists in
diff --git a/srcmorph-maven-plugin/pom.xml b/srcmorph-maven-plugin/pom.xml
index 36ed9f8..c0213b0 100644
--- a/srcmorph-maven-plugin/pom.xml
+++ b/srcmorph-maven-plugin/pom.xml
@@ -285,7 +285,19 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log</argLine>
+                        <!--
+                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
+                            been loaded dynamically" warning that a test-time self-attaching agent
+                            (no static -javaagent) emits to the *native* stderr, which can
+                            intermittently corrupt Surefire's encoded fork channel ("Corrupted
+                            channel by directly writing to native stream" → bogus "There was a
+                            timeout in the fork"). This module's jcstress/vmlens agents run only
+                            under their opt-in profiles; the flag is kept identical across the
+                            reactor's module poms and mirrors the guard in the core srcmorph module
+                            (Lincheck) and the sibling BitcoinAddressFinder repo (Mockito inline
+                            mock maker).
+                        -->
+                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             The vmlens interleaving smoke test is meaningful only when run
                             under the vmlens agent (the `vmlens` profile / CI job). Without the
diff --git a/srcmorph/pom.xml b/srcmorph/pom.xml
index a712802..c8c634f 100644
--- a/srcmorph/pom.xml
+++ b/srcmorph/pom.xml
@@ -214,7 +214,22 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log</argLine>
+                        <!--
+                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
+                            been loaded dynamically" warning emitted when a test-time agent
+                            self-attaches at runtime with no static -javaagent configured — in this
+                            module the Lincheck instrumentation agent (AiGenerationKindLincheckTest
+                            runs in the default surefire run), plus the jcstress/vmlens agents under
+                            their opt-in profiles. That warning is written to the *native* stderr
+                            (outside Surefire's wrapped System.err) and can intermittently corrupt
+                            the encoded fork channel — surfacing as "Corrupted channel by directly
+                            writing to native stream" and then a bogus "There was a timeout in the
+                            fork", even though every test passed. Kept identical across the reactor's
+                            module poms and mirrors the same guard in the sibling
+                            BitcoinAddressFinder repo (there triggered by Mockito's inline mock
+                            maker self-attaching byte-buddy).
+                        -->
+                        <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             Force classpath-mode test execution. module-info.java here is release
                             metadata for downstream module-path consumers only (see its own Javadoc:

From 9bd5f1598a8534187443c2e5324f67751ed76747 Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Tue, 21 Jul 2026 08:52:01 +0000
Subject: [PATCH 2/2] Shorten EnableDynamicAgentLoading argLine comment to
 workspace policy pointer

Replace the inline rationale in all three reactor module poms with a one-line
comment pointing to the canonical workspace policy (ci-test-diagnostics.md
section 2.1) plus the root-cause upstream bug (byte-buddy#1639), keeping the
explanation in one synchronized place.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y84TKMNfhbFGi7EoCEvP99
---
 srcmorph-cli/pom.xml          | 12 +-----------
 srcmorph-maven-plugin/pom.xml | 13 +------------
 srcmorph/pom.xml              | 16 +---------------
 3 files changed, 3 insertions(+), 38 deletions(-)

diff --git a/srcmorph-cli/pom.xml b/srcmorph-cli/pom.xml
index 543dca0..1d442fe 100644
--- a/srcmorph-cli/pom.xml
+++ b/srcmorph-cli/pom.xml
@@ -210,17 +210,7 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <!--
-                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
-                            been loaded dynamically" warning that a test-time self-attaching agent
-                            (no static -javaagent) emits to the *native* stderr, which can
-                            intermittently corrupt Surefire's encoded fork channel ("Corrupted
-                            channel by directly writing to native stream" → bogus "There was a
-                            timeout in the fork"). This module runs no such agent today, but the
-                            flag is kept identical across the reactor's module poms and mirrors the
-                            guard in the core srcmorph module (Lincheck) and the sibling
-                            BitcoinAddressFinder repo (Mockito inline mock maker).
-                        -->
+                        <!-- -XX:+EnableDynamicAgentLoading: silences the JDK 21 byte-buddy self-attach agent warning that intermittently corrupts Surefire's fork channel ("Corrupted channel ..." / bogus "timeout in the fork"). See workspace policy ci-test-diagnostics.md section 2.1: https://github.com/bernardladenthin/workspace/blob/main/policies/ci-test-diagnostics.md - root cause: https://github.com/raphw/byte-buddy/issues/1639 -->
                         <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             Force classpath-only test execution. Surefire's default module-path
diff --git a/srcmorph-maven-plugin/pom.xml b/srcmorph-maven-plugin/pom.xml
index c0213b0..4c6e07e 100644
--- a/srcmorph-maven-plugin/pom.xml
+++ b/srcmorph-maven-plugin/pom.xml
@@ -285,18 +285,7 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <!--
-                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
-                            been loaded dynamically" warning that a test-time self-attaching agent
-                            (no static -javaagent) emits to the *native* stderr, which can
-                            intermittently corrupt Surefire's encoded fork channel ("Corrupted
-                            channel by directly writing to native stream" → bogus "There was a
-                            timeout in the fork"). This module's jcstress/vmlens agents run only
-                            under their opt-in profiles; the flag is kept identical across the
-                            reactor's module poms and mirrors the guard in the core srcmorph module
-                            (Lincheck) and the sibling BitcoinAddressFinder repo (Mockito inline
-                            mock maker).
-                        -->
+                        <!-- -XX:+EnableDynamicAgentLoading: silences the JDK 21 byte-buddy self-attach agent warning that intermittently corrupts Surefire's fork channel ("Corrupted channel ..." / bogus "timeout in the fork"). See workspace policy ci-test-diagnostics.md section 2.1: https://github.com/bernardladenthin/workspace/blob/main/policies/ci-test-diagnostics.md - root cause: https://github.com/raphw/byte-buddy/issues/1639 -->
                         <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             The vmlens interleaving smoke test is meaningful only when run
diff --git a/srcmorph/pom.xml b/srcmorph/pom.xml
index c8c634f..008c63f 100644
--- a/srcmorph/pom.xml
+++ b/srcmorph/pom.xml
@@ -214,21 +214,7 @@ SPDX-License-Identifier: Apache-2.0
                             a 2 GB heap cap, a heap dump on OutOfMemoryError into the working
                             directory, and a JVM fatal-error log named per crashing PID.
                         -->
-                        <!--
-                            -XX:+EnableDynamicAgentLoading silences the JDK 21 "A Java agent has
-                            been loaded dynamically" warning emitted when a test-time agent
-                            self-attaches at runtime with no static -javaagent configured — in this
-                            module the Lincheck instrumentation agent (AiGenerationKindLincheckTest
-                            runs in the default surefire run), plus the jcstress/vmlens agents under
-                            their opt-in profiles. That warning is written to the *native* stderr
-                            (outside Surefire's wrapped System.err) and can intermittently corrupt
-                            the encoded fork channel — surfacing as "Corrupted channel by directly
-                            writing to native stream" and then a bogus "There was a timeout in the
-                            fork", even though every test passed. Kept identical across the reactor's
-                            module poms and mirrors the same guard in the sibling
-                            BitcoinAddressFinder repo (there triggered by Mockito's inline mock
-                            maker self-attaching byte-buddy).
-                        -->
+                        <!-- -XX:+EnableDynamicAgentLoading: silences the JDK 21 byte-buddy self-attach agent warning that intermittently corrupts Surefire's fork channel ("Corrupted channel ..." / bogus "timeout in the fork"). See workspace policy ci-test-diagnostics.md section 2.1: https://github.com/bernardladenthin/workspace/blob/main/policies/ci-test-diagnostics.md - root cause: https://github.com/raphw/byte-buddy/issues/1639 -->
                         <argLine>@{argLine} -Xmx2g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=. -XX:ErrorFile=hs_err_pid%p.log -XX:+EnableDynamicAgentLoading</argLine>
                         <!--
                             Force classpath-mode test execution. module-info.java here is release