Repository navigation
feat(plugin-contract): prove exact unload after a reload #36
Description
Feature description
The plugin-contract reload scenario (#34, #35) proves load, unload, load: the second load restores the first load's surface and its callback still works. It never unloads again after that second load, and the ordinary scenario never unloads a reloaded plugin. A plugin whose unload is exact the first time but leaks after a reload therefore passes every case.
Related Code
tests/_support/plugin-contract/scenario.zsh in #35 (head 8799b23), lines 68-83: the reload scenario ends after the effect check, and its comment leaves exact unload to ordinary.
Reproduced on 8799b23 with a fixture whose unload removes its precmd hook only once, keyed on a _zunit_* flag the observer excludes (the device reload-stale uses):
reload noninteractive adv-second-unload: rc 0 (passes)
ordinary noninteractive adv-second-unload: rc 0 (passes)
The shipped fixture unloads exactly after a reload, so the check has a positive case today.
Additional Context
Found by the second independent review of #35 and left out of it by maintainer decision, so #35 merges as it stands with Refs #34. The organization testing standard asks that a separate clean-shell case load, unload and load again; it does not name the unload that follows, so this closes a gap the standard leaves open rather than one it requires.
Proposed change:
- In the
reloadscenario, snapshot before the first load, then after the effect check unload again and assertplugin_restoredagainst that first snapshot. - Add a negative fixture whose unload leaks only after a reload; it must fail
reloadand passordinary. - Reword the scenario comment that leaves exact unload to
ordinary.
Acceptance criteria:
- A fixture whose post-reload unload leaks fails the
reloadcase, and the shipped fixture passes it. ./zunit --tap testspasses on the native job.
Self-service
- I'd be willing to address this myself.
Have you read the Contributing Guidelines?
- I have read the Contributing Guidelines.
Are you familiar with the Contributor Covenant Code of Conduct?
- I have read the Contributor Covenant Code of Conduct.
Activity
ss-o commentedon Sep 30, 2026
Triage (Project 28 fields set from evidence):
Enhancement: extends the existingreloadscenario's coverage; no current behavior is wrong in shipped code.Low: a plugin would need an unload that leaks only after a reload to slip through; no known plugin does, and the shipped fixture already unloads exactly.S: one extra snapshot, unload and assertion inscenario.zsh, plus one negative fixture, as the reproduction in the body shows.P2 - Medium: important for the contract's completeness but not urgent; it came out of test(plugin-contract): prove reload after unload (#34) #35's review and does not block other work.Human delivery: ordinary maintainer-reviewed test work.Status stays
Triageuntil the scope is agreed.ss-o commentedon Oct 1, 2026
Project 28 triage pass (2026-10-02): Status
Triage->Todo.mainat6349b24: tests/_support/plugin-contract/scenario.zsh:68-83@6349b24 (main = merge of test(plugin-contract): prove reload after unload (#34) #35): the reload case ends after the effect check with no second unload and no pre-load snapshot. Probe: a fixture whose unload drops its precmd hook only once (flag under zunit*) passed bothscenario.zsh reload noninteractiveandscenario.zsh ordinary noninteractive(rc 0, rc 0); the shipped fixture also rc 0. A probe variant of the reload flow that snapshotsbefore, unloads again after the effect check and asserts_zunit_assert_plugin_restored before afterfails the leaky fixture (rc 1, 'hook:precmd') and passes the shipped fixture (rc 0), provided the scenario's own_contract_fixture_effectglobal is unset before the final snapshot.