[isc-dchui]

Description

Resolves #1195
Resolves #1251

This cuts integration test time by about a third, based on a MONLBL profile of the full suite and per-class timings. Most of the time went to work the tests didn't need: scanning unrelated modules, downloading large Python packages, and querying the public community registry. It also makes every test clean up the files it creates in the repo.

CI timings

Each CI job runs the full test suite twice: in "Test and Verify", and again in "Clean install current version and test new one." The table compares the average of the two main runs just before this branch (a765da5, d3815fa) with the branch run (d5c57f2). The averages cover the four images of each flavor.

	main	this branch	saved
IRIS, Test and Verify	557s	392s	165s (30%)
IRIS, Clean install and test	611s	440s	171s (28%)
IRIS, whole job	24.9 min	19.4 min	5.5 min (22%)
IRIS for Health, Test and Verify	945s	576s	370s (39%)
IRIS for Health, Clean install and test	1002s	624s	378s (38%)
IRIS for Health, whole job	39.4 min	26.7 min	12.8 min (32%)

The slowest job, which bounds the workflow when nothing is queued, went from 42.8 min to 28.7 min. Runner timings vary by up to about 30% between runs of the same job on main, so read these as approximate.

Where the time came from, measured locally per class against an unprofiled full run:

Change	Class	Before	After
Smaller Python fixtures	ProcessPythonWheel	187s	44s
Narrower repo -fs root	FileCopy	93s	20s
Narrower repo -fs root	PublishModuleWithDeps	77s	4s
Clear repos before building graphs	BuildDependencyGraph	66s	13s
Stop re-enabling the community registry	InstallModule	36s	26s

What changed

Repo roots. A repo -fs repo scans every module.xml under its root when it is created. FileCopy and PublishModuleWithDeps rooted theirs at all of _data/, which holds 217 modules. Each is now rooted at its own fixture folder, and FileCopy no longer leaves its repo configured for later classes.

Python fixtures. The packages in requirements.txt and the committed wheels were much larger than the tests need:

Fixture	Before	After
package-with-python-deps	lune + ansible: 11 wheels, 53MB for ansible alone	lune + jinja2, which pulls in markupsafe
just-reqs	docx2md, which pulls in lxml and pillow	markdown-it-py, which pulls in mdurl
just-wheels	numpy wheel (16MB)	tomli wheel (15KB)

Each fixture still covers what its test checks: a package with a dependency of its own, a requirements.txt-only module, and a <PythonWheel> resource.

Community registry. Seven classes ended with repo -reset-defaults, which re-enables pm.community.intersystems.com. The verify namespace keeps its repos between classes and between runs, so every later dependency lookup also queried the internet:

any class ─ reset-defaults ─> community registry ON, into the next run too
   ├── BuildDependencyGraph   ~2.5s per graph build (0.02s without it),
   │                          paying for the registry left on by the previous run
   ├── InstallModule          +9s
   └── ... every class that resolves modules by name

The cleanups now only delete repos, and BuildDependencyGraph clears repos before creating its own. This also removes the suite's dependency on reaching the internet registry.

Files left in the repo. Run one class at a time, every integration test class and the unit suite now leave the working tree as they found it. The ones that didn't:

Class	Left behind
LockFile	rewrote 8 committed module-lock.json files, which are inputs for the "overwrite an existing lock file" cases, so it now saves and restores them
Module	_data/varresolver/module.xml whenever the test failed before its own cleanup. The module is now generated outside the repo
Resources	localize/HTTPStatuses.xml, which Activate exports into the module root
ProcessPythonWheel	the wheel build output in the fixture's package source, and the wheels -export-python-deps downloads into <module root>/wheels. The committed package-with-python-deps/wheels/ was leftover output from those downloads, so it is removed

The .gitignore files that hid the Resources and ProcessPythonWheel output are removed, so a future leak shows up in git status.

Product code.

• The shell welcome message lists modules in every namespace, and it was built on every one-shot zpm "..." call even though those never show it.
• LoadNewModule ran a full %XML.Reader pass over module.xml twice, once for <Defaults> and once for the version checks.

Testing

• Existing tests pass

Checklist

• This branch has the latest changes from the main branch rebased or merged.
• [N/A] Changelog entry added.
• Unit (zpm test -only) and integration tests (zpm verify -only) pass.
• Style matches the style guide in the contributing guide.
• Documentation has been/will be updated
  • Source controlled docs, e.g. README.md, should be included in this PR and Wiki changes should be made after this PR is merged (add an extra issue for this if needed)
• Pull request correctly renders in the "Preview" tab.

🤖 Generated with Claude Code