AppConfig and UserConfig unconditionally queried NC-only columns (type,
lazy, flags, indexed) that don't exist in ownCloud's database schema,
breaking ownCloud → Nextcloud upgrades entirely before the schema
migration steps could run.
Restore the fallback pattern in both classes: on first loadConfig() call,
if a DBException with REASON_INVALID_FIELD_NAME is thrown, set
$migrationCompleted = false and retry selecting only the columns present
in ownCloud's schema. INSERT and UPDATE statements also omit NC-only
columns when $migrationCompleted is false.
The catch block also guards against infinite recursion: if $migrationCompleted
is already false when the exception fires, the exception is re-thrown
instead of triggering another recursive call.
Fixes: https://github.com/nextcloud/server/issues/57340
Signed-off-by: Anna Larch <anna@nextcloud.com>
AI-Assisted-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Currently apps are broken if they have exports in the JS entry point,
because they then will import from the entry point but because they do
not know about the Nextcloud cache buster they will import without cache
buster.
This results in two problem:
1. The module might be outdated (old cached)
2. The module is duplicated, so the module will be loaded twice and will
have two different - out of sync - states. This also means it will
re-run sideeffects of the entry point.
To fix this we generate an import map which basically maps the plain
entry point script to the script with cache buster added.
(Some background: Bundler will try to minimize chunks (reduce page
loading time) so they can inline modules into entry points and thus
extend the entry point exports and then this issue would be caused).
For example:
```js
// entry.mjs
console.error('called')
async function onClick() {
await import('./chunk.mjs')
}
export const name = 'foo'
// chunk.mjs
import { name } from './entry.mjs'
console.error(name)
```
When calling `onClick` without this fix the output will be:
> called
> called
> foo
With this fix:
> called
> foo
Signed-off-by: Ferdinand Thiessen <opensource@fthiessen.de>
- Attempt delete before logging the warning, so the warning only fires
when we know recovery will succeed
- Log an error (not silently return) when delete itself fails
- Use catch (\Exception) without variable (PHP 8)
- Replace willReturnArgument(1) with explicit willReturn(true) in test
- Add blank lines between logical blocks in test for readability
Signed-off-by: Anna Larch <anna@nextcloud.com>
AI-Assisted-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When the appstore cache file exists but getContent() throws a
GenericFileException (I/O error or OS-level permission failure), explicitly
delete the file and recreate it before writing fresh data — mirroring the
NotFoundException recovery path. If deletion itself fails, return [] cleanly.
Previously, the unhandled exception caused the entire apps settings page to
crash. The new test covers both the recovery path and deletion failure.
Signed-off-by: Anna Larch <anna@nextcloud.com>
AI-Assisted-By: Claude Sonnet 4.6 <noreply@anthropic.com>
AWS SDK PHP v3.339.0+ introduced a breaking change requiring the Content-MD5
header for DeleteObjects operations. This causes 'MissingContentMD5' errors when
using S3-compatible services like MinIO.
Add middleware to automatically calculate and inject the Content-MD5 header on
all DeleteObjects requests. This is applied universally at the S3ConnectionTrait
level, fixing both external storage (AmazonS3) and core ObjectStore (S3) classes.
Fixes: https://github.com/aws/aws-sdk-php/issues/3068
Signed-off-by: John Molakvoæ (skjnldsv) <skjnldsv@protonmail.com>
- Add Electron to supported browsers regex so Electron-based environments
(including Cypress) are not redirected to the unsupported browser page.
The browserslist-useragent-regexp 4.1.4 upgrade stopped matching Electron
implicitly; this mirrors the existing AscDesktopEditor special-case.
- Fix TextProcessing mock: deleteOlderThan() returns int, not void.
PHP 8.4 strict mock type enforcement causes a TypeError when the
callback is declared void.
AI-Assisted-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Anna Larch <anna@nextcloud.com>
Signed-off-by: nextcloud-command <nextcloud-command@users.noreply.github.com>
First check which users have a shares and for which providers and then
only load these shares.
Avoid doing at most 5 SQL queries for each users a share was shared with if
there are no shares.
Signed-off-by: Carl Schwan <carlschwan@kde.org>
Stripping the underscore breaks support for all languages like de_AT,
de_DE and so on...
Signed-off-by: Salvatore Martire <4652631+salmart-dev@users.noreply.github.com>
Fix possible dead locks when running the propagator caused by two
requests updating the same amount rows in transactions.
- Lock rows always in the same deterministic order by sorting the
path_hash first
- On all database outside of sqlite, also do first a SELECT FOR UPDATE
to lock all the rows used in batch UPDATE calls, afterward to decrease
the risk of two requests trying to lock the same rows
Signed-off-by: Carl Schwan <carlschwan@kde.org>
(cherry picked from commit cae742d182)
Add InstallationCompletedEvent class in public API (OCP namespace) that
provides installation details: data directory, admin username, and admin
email. Event will be dispatched after successful installation.
Include comprehensive unit tests covering all event scenarios.
Signed-off-by: Misha M.-Kupriyanov <kupriyanov@strato.de>
If the current user belongs to both one or more groups excluded from
sharing and one or more groups not excluded from sharing the user is
allowed to share. However, in the contacts menu, as soon as the current
user belonged to a group excluded from sharing the user could not search
for local contacts. This has been unified now with the sharing
behaviour, so local contacts can still be searched if the user also
belongs to a group not excluded from sharing (or to no group at all,
which was also allowed before).
Signed-off-by: Daniel Calviño Sánchez <danxuliu@gmail.com>
