When installing an app from the appstore the `\OC_App::getAppVersion` code is triggered twice:
- First when the downloader tries to compare the current version to the new version on the appstore to check if there is a newer version. This protects against downgrade attacks and is implemented in `\OC\Installer::downloadApp`.
- Second, when the app is actually installed the current version is written to the database. (`\OC\Installer::installApp`)
This fails however when the version is actually cached. Because in step 1 the cached version will be set to "0" and then be reused in the second step.
While this is probably not the cleanest version I assume this is an approach that is least invasive. Feedback and suggestions welcome :)
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
Else the last-login-check fails hard because the session value is not
set and thus defaults to 0.
* Started with tests
Signed-off-by: Roeland Jago Douma <roeland@famdouma.nl>
On renew, a session token is duplicated. For some reason we did
not copy over the remember-me attribute value. Hence, the new token
was deleted too early in the background job and remember-me did
not work properly.
Signed-off-by: Christoph Wurst <christoph@winzerhof-wurst.at>
We have to respect the value of the remember-me checkbox. Due to an error
in the source code the default value for the session token was to remember
it.
Signed-off-by: Christoph Wurst <christoph@winzerhof-wurst.at>
As SemVer can be used apps could define a release like "10.0.0-alpha". This is something that we don't support at the moment in the server and we should filter all prereleases.
Ref https://github.com/nextcloud/server/pull/2307#issuecomment-262911588
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
The current implementation when fetching apps from the appstore is to assume that the first element is the newest version, this is now always applicable and leads to the fact that for some apps (e.g. nextant) the newest version is not delivered. This can be easily tested by comparing the version of the downloaded Nextant version.
This change will loop over all releases delivered by the appstore and chooses the newest compatible one. While not the cleanest solution, it does its job.
Most of the code are actually unit tests. Whereas I have copied the whole original response from the appstore and also have performed the transformation. So that's why the diff looks so huge.
Signed-off-by: Lukas Reschke <lukas@statuscode.ch>
* A repair step that inserts a background job for each user
* Each background job will delete for 15 seconds if it takes longer we
reschedule. This is done so instances that don't use the system cron
won't time out.
* Added tests
Signed-off-by: Roeland Jago Douma <roeland@famdouma.nl>
