Before Qt 5.10 QtNetworkAuth was in Tech Preview state and was missing
methods to set and retrieve refresh tokens.

This test only includes a very basic test case for now. More to come.

Prepare an advanced version of the unittest addition macro used for
test cases which require some non-standard set-up, like extra sources
or link libraries.

Instead of just popping the web browser authentication window display
a notification informing the user about the token expiration and let
him trigger the authentication by clicking the action button.

The implementation inside QOauthOobReplyHandler has one major problem
- in case something goes south it only prints an error to the log and
exits leaving the authentication process hanging and waiting for
something that will never come.

Instead use an additional signal to notify the OAuth2 authentication
controller that a problem occurred so that the state can be safely
reset.

In case this fails fall back to full authentication.

This involves traffic from all three possible sources:
  * Settings dialog
  * Storage (wallet)
  * EWS client

This avoids reauthentication when the existing tokens are valid. In
case there are no tokens or they're invalid the tokens retrieved from
the OAuth2 authentication process are taken back and stored.

This will allow use of already stored tokens in the configuration
dialog.

This most likely means that the OAuth2 access token is no longer
valid. Try to reset and refresh it.

Unline password authentication the OAuth2 authentication can at any
time change the tokens due to reauthentication or access token
refresh. These updated tokens need to be passed up to the main
resource in order for them to be stored for use after a restart.

The access token also needs to be reset in case it is found not to be
valid any more.

This merges the codepaths used to read and write the password into a
single wallet request that can be used for both. Existing
functionality, which satisfies a pending read password request with a
write password request is retained.

This prepares the code for introduction of storing the OAuth2 tokens
in the wallet.

This is needed whenever the browser window would need to be displayed
on top of an existing dialog window, which is the case when the
configuration dialog is open. Otherwise the modality order is wrong
and the browser window is unable to accept user input.

When a request is started in case of enabled OAuth2 the implementation
checks if a token is available and if not - launches the
authentication. The actual HTTP request is delayed until that process
is completed.

This process is also started in case a 401 Unauthorized HTTP response
is received with OAuth2 enabled.

The event loop executed by KJob::exec() explicitly blocks input
events. This will cause the Cancel button to be effectively block and
will also prevent interaction with the OAuth2-driven web browser.

This includes a class which will be responsible for driving the OAuth2
authentication using QtNetworkAuth framework. QWebEngine is used as
the browser for the interactive steps.

The class will be later used by the EWS request to retrieve the bearer
token.

This description explains the role of this checkbox and the difference
between "no domain" vs. "empty domain".

The field is disabled for now until support in code is added.

Split out the authentication and EWS-specific fields into dedicated
groups.

This is a preparatory change for introducing OAuth2 as an alternative
authentication option.

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

Using invoke method makes no sense as the QThread object always lives
in the parent (=main) thread, which means the message will get
dispatched to the current thread's event loop anyway. Since that event
loop will be blocked by the subsequent wait() call the message will
never reach the fake server thread.

Instead just call the method directly.

This ensures that the next test always sees a clean state of the
resource.

Whenever an option is set on a resource during the POP3 test wait for
the resource to acknowledge. This makes sure the resource has actually
consumed this setting before relying on it in subsequent tests.

The maildir resource should automatically create all the paths inside
its data dir. Unfortunately in case of the POP3 test this does not
happen for some reason. Work around by creating the paths manually.

The POP3 resource test simulates a series of POP3 messages to the
resource and waits for them to be processed. Unfortunately this is not
enough as even though the POP3 resource is done talking to the server
processing of items in Akonadi and in the associated maildir resource
can still take some time. Due to this fact the POP3 tests were failing
intermittently.

In order to prevent that run an additional wait loop with the help of
the monitor interface. Each time an item operation is notified a timer
is reset. Once no item operation happens within 200ms it is safe to
assume that the server is done and all the operations resulting from
the simulated POP3 traffic have been committed.

This is to prevent attempts to delete them again.

Fixes the testCountBasedLeaveRule, which excercises the following
scenario:

 * initial state: 3 messages known
 * two new messages are added on server
 * LIST returns 5 messages
 * UIDL returns 5 UIDs, 2 of them new
 * download of 2 messages starts
 * fetch of first message is completed
 * item create for first message is completed
 * item create completion triggers cleanup of unwanted server messages
 * messages 1 and 3 are deleted as they are the oldest
 * fetch of second message is completed
 * item create for second message is completed
 * item create completion triggers cleanup of unwanted server messages
 * messages 1 and 3 are deleted again - this causes the test to fail

Removing those identifiers causes the second search for items to
delete to return an empty set and no further delete attempts are made.

Reviewers: #kde_pim

Reviewed By: dvratil

Tags: #kde_pim

Differential Revision: https://phabricator.kde.org/D15975

When determining ids of mails to delete a special identifier -1 is
used to reference all e-mails. When the "leave on server" flag is not
set asking for deletion of all mails should really return a list of
all mail ids.

This fixes Pop3Test::testSimpleLeaveOnServer test case.

Reviewers: #kde_pim

Reviewed By: dvratil

Tags: #kde_pim

Differential Revision: https://phabricator.kde.org/D15962

Setting the maildir path is an asynchronous operation of which effects
can be delayed. Fetching the maildir root collection before that will
cause the fetched identifier to become stale and cause a cascade of
failures.

This makes use of the internal information that the remote identifier
of the maildir resource is the path to the storage directory.

GIT_SILENT: Cleanup: remove ECMOptionalAddSubdirectory which is never used\n ECMMarkNonGuiExecutable which is includes in ECMAddTests module\n CMakePackageConfigHelpers which is included by ECMSetupVersion

The test works by faking responses to some EWS requests and waiting
for the resource to reach an expected state based on mocked responses
to these requests.

The state expectation logic is driven by Akonadi collection change
monitor. Unfortunately every now and then some notifications get lost
causing the state not to be reached while the server is actually in a
correct one.

Work around this by introducing an optional recheck method that
fetches all collections and forces a recheck of the current state.

This should finally fix the isolated tests for good.

It looks like use of EntityDisplayAttribute was deprecated.

This fixes the isolated test not being able to reach expected state.

Note: the test still fails intermittently - to be investigated.

Commit a20701f5 added the "has domain" flag to fix ambiguity between
empty domain and no domain in the username. Unfortunately the default
behaviour was set to "domain present", which changed the default
username from "test" to "/test" causing an unexpected value of the
Authorization HTTP header and in turn a failure.

Fix this by explicitly setting the "has domain" flag to false.