[bpamiri]

Fixes #25 (upstream: cybersonic#144)

What

ModuleArgBindingTest and McpCommandTest installed their fixture modules into LucliPaths.resolve().modulesDir(), which is the real ~/.lucli/modules by default. They then spawned dev-lucli.sh without passing a home, so the child resolved its own home from the account's user.home. That caused two problems:

• mvn test compiled the fixtures into the real ~/.lucli/lucee-server.
• With an isolated test home, the child couldn't find the fixtures: 17 tests failed.

Each class now has a static @TempDir LuCLI home. The fixture is installed there, and every child process gets that home as LUCLI_HOME through a small lucliProcess(...) helper. LUCLI_HOME also moves the child's Lucee context (configureLuceeDirectories), so nothing goes to ~/.lucli. Only test code changes.

Tests

• New ModuleArgBindingTest.childRunsInTheTemporaryLucliHome: asserts that the child created its lucee-server under the temporary home. RED: with the child pointed at a different home, it fails with "the child must create its Lucee context under LUCLI_HOME".
• mvn test -DargLine="-Duser.home=<empty dir>" with LUCLI_HOME unset: 788 tests, 0 failures (before: 17 failures). ~/.lucli had no files newer than the run's start. The child ran with the real user.home, so this checks the child, not just the test JVM.

🤖 Generated with Claude Code

[github-actions]

Unit Tests (ubuntu-latest, Java 21)

 65 files   65 suites   2m 29s ⏱️
783 tests 777 ✅ 6 💤 0 ❌
788 runs  782 ✅ 6 💤 0 ❌

Results for commit 3b79b9c.

[github-actions]

Integration Tests (ubuntu-latest, Java 21)

78 tests   78 ✅  2m 1s ⏱️
14 suites   0 💤
 1 files     0 ❌

Results for commit 3b79b9c.

[bpamiri]

rev1-r3 review @ 3b79b9c: APPROVE (final)

RED/GREEN in a clean home. I ran mvn test -Dtest='ModuleArgBindingTest,McpCommandTest' with a fresh sandbox (-DargLine="-Duser.home=<sandbox> -Dlucli.home=<sandbox>/.lucli"):

	ModuleArgBindingTest	McpCommandTest
base wheels	8 run, 8 failures	9 run, 4 failures + 5 errors
this branch	9/0/0 (including the new childRunsInTheTemporaryLucliHome)	9/0/0

The base failures are exactly the 17 known clean-home failures from #25, and this PR removes all of them.

No writes to the real home: find ~/.lucli ~/.wheels -newer <start-mark> after both runs shows nothing, even though the dev-lucli.sh children run with the real user.home, as the description says.

Code (test-only):

• A static @TempDir lucliHome per class, so JUnit cleans it up.
• The fixture is installed under lucliHome/modules, and every child ProcessBuilder gets LUCLI_HOME=lucliHome.
• The now-unused LucliPaths import is removed.
• The new test asserts the child creates its lucee-server context under the temp home.

Note, not blocking: each class now builds a fresh Lucee context on its first child run, so locally the two classes took about 49 s and 28 s instead of about 11 s each. That's a fair price for hermetic tests.

PR CI: 8 pass (Build ×3, Unit ×3, Integration ×2); APT smoke skipped. mergeStateStatus is CLEAN.

Ready to merge.