lxbme opened a new pull request, #3438:
URL: https://github.com/apache/apisix-dashboard/pull/3438
**Why submit this pull request?**
- [x] Bugfix
- [ ] New feature provided
- [ ] Improve performance
- [ ] Backport patches
**What changes will this PR take into?**
Part of #3417 (Data-integrity section, last remaining item): "Empty-value
stripping inside surviving plugin configs: `produceRestoreEmptyPlugins` only
restores whole-plugin entries; nested meaningful empties (`{"headers": {}}`,
`[]`, `""`) are still silently removed. Inline route/service
`upstream.discovery_args: {}` also still dropped (the #3376 merge-back fix
covered the upstreams page only)."
The submit pipeline deep-cleans every empty value (`{}` / `[]` / `""` /
`null`) and then restored only WHOLE plugin entries that had vanished.
Consequences, both verified against a live gateway:
- empty members **inside** a partially-surviving plugin config were silently
removed on any edit-save — the Admin API accepts and stores these shapes for
loose-schema plugins (`{"key-auth": {"empty_obj": {}, "empty_arr": [],
"empty_str": ""}}` → 201, stored verbatim);
- `discovery_args: {}` on an **inline** upstream (routes/services) was
dropped — the #3376 fix only restored the root-level field on the upstreams
page.
**Fix.** The clean-then-restore pair is replaced by a single stage,
`produceCleanPreservingUserValues`: snapshot the `plugins` subtree, detach it,
deep-clean the rest, reattach verbatim, and re-add root/inline `discovery_args:
{}` when present. The principle: plugin configs are user-authored JSON and the
gateway is the only judge of their empties — loose schemas accept them, strict
ones reject with a descriptive 400 (`proxy-rewrite` `headers: {}` → "value
should match only one schema, but matches none", verified live).
**Behavior change, disclosed:** typing an empty member that a strict plugin
schema rejects now surfaces the gateway's 400 (toast via the existing error
path) instead of being silently stripped into a "successful" save that stored
something other than what the user wrote.
**Why a single-draft stage — two crashier designs died in testing, disclosed
for reviewers:** a restore stage that closes over the pipeline's input value
breaks under *nested* `pipeProduce` composition (the routes detail page
composes `produceRoute` — itself a `pipeProduce` — as a stage of another
`pipeProduce`), because inner stages receive the outer **live immer draft**
instead of a plain value: assigning the base reference back trips immer's set
trap, and `structuredClone` rejects proxies. Both variants passed unit tests
with plain inputs and were caught by the full e2e suite
(`routes.crud-all-fields`, `routes.empty-plugin-config` went red). The final
design does everything inside one draft with `current()` snapshots and
references nothing outside it; a dedicated unit test pins the nested
composition.
**Tests** (red on the unfixed build at the intended assertions, green after):
- Unit `producer.test.ts` additions: plugin configs pass through verbatim
(array-form nodes included), inline + root `discovery_args` restored,
nested-`pipeProduce` composition survives.
- E2E regression `form.plugin-config-nested-empties.spec.ts`: a no-op
edit-save keeps `{}` / `[]` / `""` members inside a plugin config, and keeps
`discovery_args: {}` on an inline upstream.
Blast radius: full local e2e suite — **177 passed** (the run that caught the
two intermediate designs' regressions is what forced the final shape);
remaining failures are documented environment items unrelated to this change
(one Monaco read-back race non-deterministic on the same build and reproducing
on pristine master, one bulk-page load flake green on isolated rerun, and
`stream_routes.show-disabled-error`, which cannot run outside the repo's own
compose project). Unit tests 37/37, lint and build clean.
**Related issues**
Part of #3417 (please do not auto-close the tracking issue)
**Checklist:**
- [x] Did you explain what problem does this PR solve? Or what new features
have been added?
- [x] Have you added corresponding test cases?
- [ ] Have you modified the corresponding document? (no user-facing document
covers the submit pipeline)
- [x] Is this PR backward compatible? If it is not backward compatible,
please discuss on the mailing list first
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
