[26.1] Fix Workflow.copy() dropping readme, help, logo_url and doi

[mvdbeek]

readme, help, logo_url and doi were added to the workflow table in 25.0 (57cf7d38d53, e886353b90a, migration 71eeb8d91f92), but Workflow.copy() was never updated to copy them. Every operation that copies a workflow silently destroys them.

Export is guarded by if workflow.readme is not None, so the metadata doesn't come back as null — the key is absent from the exported .ga entirely, which is what makes this look like an export bug. It isn't: import, export, editor and API paths all handle readme correctly.

Affected paths

• Rename-only update (lib/galaxy/webapps/galaxy/api/workflows.py:466) — name_updated and not steps_updated copies the workflow into a new revision, so renaming a workflow destroys its readme.
• _import_shared_workflow (lib/galaxy/webapps/base/controller.py:626) — reached via /workflow/imp?id= and POST /api/workflows {shared_workflow_id}.
• Subworkflow copy during shared import.

Reported against an IWC workflow with a 7398-character readme on usegalaxy.eu (galaxyproject/iwc#1234). The uuid changing across the round-trip pins the cause to copy() rather than to export, and the survival pattern is a fingerprint: everything copy() handles survived, only the fields it omits vanished.

On source_metadata

Still deliberately not copied. It records where this exact content was fetched from and is already dropped whenever a workflow is modified (test_trs_import asserts this, and the comment at test_workflows.py:1484 calls it out as intentional). Copying it would also change get_workflow_by_trs_id_and_version, which matches on latest_workflow.source_metadata.

Worth noting separately: because source_metadata isn't copied, renaming a TRS-imported workflow still breaks its TRS dedup lookup. That's arguably its own bug, but it's a behavioral change I didn't fold into this fix.

Tests

copy() has regressed this way three times now — 470baa59216 added license + creator_metadata, bbbe7f2182c added reports_config — because the assignments are hand-maintained. The new unit test therefore drives its assertions off Workflow.__table__.columns minus an explicitly annotated not_copied set, so a column added by a later migration trips "<column> is not covered by this test" and has to be classified rather than silently dropped.

• test_workflow_copy_preserves_metadata (unit) — verified to fail on an unmodified tree (AssertionError: help / assert None == 'Some help text') and pass with the fix. I also deleted the doi assignment with the fix otherwise in place to confirm the guard catches exactly the recurrence mode.
• test_update_name (API) — extended; it already covered license for precisely this reason.
• test_import_published_preserves_metadata (API) — new, covers the shared-import path.

The existing suite is not evidence here: test_update_name passed the whole time the bug was live, which is how this regressed.

Bug dates to 25.0 — worth checking whether 25.0/26.0 are still supported and need this too.

[jmchilton]

Good comments and good tests - is this one also not Claude?

[mvdbeek]

Plan was fable, implementer was opus with a minor manual steer (wanted to copy all model attributes and maintain a string list of attributes to exclude, which i think was a little brittle)

[github-actions]

This PR was merged without a "kind/" label, please correct.