Upgrade accounts saved as http:// to https on launch
Benchmarks / CPU and I/O (per commit) (push) Successful in 1m53s
Benchmarks / Frame budget (on demand) (push) Skipped
Build and test / Desktop (Linux) (push) Successful in 45m7s
Build and test / Layer separation (push) Successful in 41s
🐳 Android image / Build and push (push) Successful in 1s
Build and test / android-image (push) Successful in 1s
🐳 Windows image / Build and push (push) Successful in 1s
Build and test / windows-image (push) Successful in 2s
Traceability / Requirement traces (push) Successful in 40s
Build and test / Android (aarch64) (push) Successful in 28m49s
Build and test / Windows (x86_64, cross) (push) Successful in 17m9s
Build and test / Publish the release (push) Skipped
Benchmarks / CPU and I/O (per commit) (push) Successful in 1m53s
Benchmarks / Frame budget (on demand) (push) Skipped
Build and test / Desktop (Linux) (push) Successful in 45m7s
Build and test / Layer separation (push) Successful in 41s
🐳 Android image / Build and push (push) Successful in 1s
Build and test / android-image (push) Successful in 1s
🐳 Windows image / Build and push (push) Successful in 1s
Build and test / windows-image (push) Successful in 2s
Traceability / Requirement traces (push) Successful in 40s
Build and test / Android (aarch64) (push) Successful in 28m49s
Build and test / Windows (x86_64, cross) (push) Successful in 17m9s
Build and test / Publish the release (push) Skipped
Before the previous commit, browser sign-in could store an account as http://, and the client now refuses to send to one. Left alone, such a library would fail to open with a configuration error, so its stored endpoint is rewritten before anything reads it. The endpoint is half of two keys, and both are handled: - The keyring entry is filed under it. Rewriting only the record would strand the app password under the old key and sign the user out, so AccountStore::move_endpoint copies the secret across first, rewrites the record in place (the last record is the one resumed), and deletes the old entry only once nothing refers to it. - namespace() is built from it and names the catalog directory. For an http to https rewrite it does not change, because the namespace strips either scheme. A move that would change it is refused, not performed, so no later rewrite can abandon a catalog either. The rewrite is a new BackendProvider::upgrade_endpoint hook, which does nothing by default, and not a second call to normalise_endpoint. The folder connector's normalise_endpoint canonicalises the path and needs it to exist, so running it on every launch would fail a library on an unplugged disk, or rename one whose path now resolves differently. Only Nextcloud implements the hook. If the move fails (for example, a locked keyring), it is logged, the account is left as it was, and the move is tried again on the next launch. Closes #65.
This commit is contained in:
@@ -83,6 +83,20 @@ pub trait BackendProvider: Send + Sync {
|
||||
/// wrong rather than naming a type.
|
||||
fn normalise_endpoint(&self, input: &str) -> Result<String, String>;
|
||||
|
||||
/// The form a *stored* endpoint should take now, where an older build
|
||||
/// wrote one this build would not.
|
||||
///
|
||||
/// Not [`normalise_endpoint`](Self::normalise_endpoint) run again: that
|
||||
/// judges what a person typed, and may touch the world to do it — a folder
|
||||
/// is canonicalised and must exist — so rerunning it on every launch would
|
||||
/// fail a library whose disk is unplugged, or rename one whose path now
|
||||
/// resolves differently. This is a pure rewrite of the string, and `None`
|
||||
/// means leave it alone, which is the answer for almost every connector.
|
||||
fn upgrade_endpoint(&self, stored: &str) -> Option<String> {
|
||||
let _ = stored;
|
||||
None
|
||||
}
|
||||
|
||||
/// Build an account from a normalised endpoint alone.
|
||||
///
|
||||
/// Only meaningful for [`SignIn::EndpointOnly`]; a browser flow produces
|
||||
|
||||
Reference in New Issue
Block a user