feat(file-service): accept string|resource content for streamed writes

[rjzondervan]

Summary

• Widen the $content parameter on FileService::saveFile/addFile and the CreateFileHandler/UpdateFileHandler content paths from string to string|resource, so a stream can be handed straight to OCP\Files\File::putContent() (which already accepts one) instead of being buffered into a PHP string. This is the provider half of stream-file-content; the OpenConnector consumer half is fix(sync): stream file downloads to disk and record the contract before file rules openconnector#1087.
• Pure type widening — every existing string caller behaves exactly as before. PHP has no resource type keyword, so the runtime declaration is mixed and the enforced contract lives in the @param string|resource docblock that PHPStan/Psalm check at each call site.
• Preserve the security guarantees on the streamed path: the filename-extension check still runs, and the magic-byte check reads a bounded 512-byte prefix from the stream and rewinds before the write, rather than buffering the file.
• Preserve the change-detection optimisation: on a resource the md5 is computed with hash_update_stream (chunked, never buffered) and the stream rewound, so an unchanged file still skips both the write and the version bump.
• Declare the contract on the update path too. FileService::updateFile and UpdateFileHandler::updateFile said @param mixed, which silently widened the enforced contract there — an int, array or object would have passed static analysis. Both are nullable (null = update tags/metadata only), so the accurate narrowing is string|resource|null.
• Untrack the two generated js/openregister-push-*.js bundles. /js/ is gitignored but these stayed tracked from earlier web-push work — the ignore rule only prevents future adds.

Also in this PR

Two fixes that are not part of stream-file-content but were found while verifying it, and are separable if you'd prefer them split out:

• Mass delete was completely broken. objectStore.massDeleteObject() was called from three places but never defined, in neither this app's store nor @conduction/nextcloud-vue — it threw TypeError: not a function before any request, so the confirm button did nothing. Added, delegating to the package store's deleteObjects(type, ids). Two follow-on defects only became reachable once it existed: the store holds id strings while the dialog submitted map(obj => obj.id) (undefined for every entry), and _buildUrl(type, id) appends /${id} only if (id) — so an undefined id became a collection-scoped DELETE, i.e. HTTP 405 per selected row. And the post-delete refresh called refreshObjectList(), which refetches a different collection than the search view renders, leaving deleted rows on screen.
• Nextcloud 35 support declared in appinfo/info.xml.

Checks

• 313 tests, 706 assertions — tests/Unit/Service/File/ green
• Verified end to end against a live sync together with openconnector#1087: 5 objects created, 5 contracts persisted, 5 files written and attached, and a re-run reports updated: 5, created: 0
• Verified per-id DELETE → 204 and collection DELETE → 405, and that the object total moved 40 → 39 on a mass delete

Issues

Refs ConductionNL/openconnector#1079 — this is the OpenRegister provider half of that cross-repo change, so it does not auto-close the issue.

Test plan

• CI passes
• Tested locally against a live Nextcloud 35 instance
• Reviewed for regressions

[github-actions]

Quality Report — ConductionNL/openregister @ fa6da27

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
composer			✅	✅ 174/174
npm			✅	✅ 555/555
PHPUnit					✅
Newman					✅
Playwright					⏭️

---

Quality workflow — 2026-07-29 10:07 UTC

Download the full PDF report from the workflow artifacts.

[github-actions]

Quality Report — ConductionNL/openregister @ e96e6ba

Check	PHP	Vue	Security	License	Tests
lint	✅
phpcs	✅
phpmd	✅
psalm	✅
phpstan	✅
phpmetrics	✅
eslint		✅
stylelint		✅
build		✅
composer			✅	✅ 173/173
npm			✅	✅ 713/713
PHPUnit					✅
Newman					✅
Playwright					✅

---

Quality workflow — 2026-08-04 10:54 UTC

Download the full PDF report from the workflow artifacts.

[rubenvdlinde]

Brought this branch current with development (it was 166 commits behind, so its green run was measuring a base that no longer exists). The merge was clean and the rebased run is green except one job, which turns out to be a real defect rather than staleness.

quality / Frontend Tests (unit) fails: src/tests/openregister-push-sw.spec.js — "Test suite failed to run".

Cause is commit e8cd76a "chore(build): untrack generated js/ web-push bundles". Those two files are not generated. webpack.config.js:273-274 says so in as many words:

The Web Push Service Worker + opt-in client (openregister-push-client.js / openregister-push-sw.js) are hand-written static assets served as-is — NOT [built]

They are not webpack entries, so nothing regenerates them. Consequences:

1. The spec reads the file straight off disk (fs.readFileSync(path.resolve(__dirname, '../../js/openregister-push-sw.js'))), so it cannot load — that is the failing job.
2. More seriously, the service worker is served as-is at runtime. WebPushDispatchJob and AnnotationNotificationDispatcher both document their payload shape against js/openregister-push-sw.js. Deleting it means the SW 404s and web push stops working in production — and nothing else in CI would have said so, because the only thing that reads it is the test that is now failing.

development passes this job, so this failing set is not a subset of the base's and I have not merged it.

Suggested fix: revert e8cd76a (keep both files tracked) and, if the intent was to stop /js/ being ignored-but-tracked, add explicit negations — !/js/openregister-push-sw.js and !/js/openregister-push-client.js — so the ignore rule stops looking like it covers them.

Everything else on this branch (the string|resource streamed-write contract and its tests, the mass-delete store action, the NC 35 max-version bump) is green on the current base.