When set programmatically, browsers don't always consistently show a focus indicator which is required for list navigation via keyboard to function properly. We already implemented a custom focus indicator for buttons which has now been expanded with slight tweaks to anchor tags as well. This is by no means a complete solution and will be revisited at a later date.
An accessibility requirement dictates that our custom shortcut system must be allowed to be disabled. A new accessibility settings page has been added to the user profile, similar to the theme settings. It is persisted in local storage.
Co-authored-by: Eduard Heimbuch <eduard.heimbuch@cloudogu.com>
We recently introduced shortcut-based list navigation that ran into problems when the list contained items that were loaded asynchronously but had stand at the beginning of the list. As the system is based on render order, we had to introduce a new layer in-between the main iterator and the asynchronously loaded items. This new layer is called a "sub-iterator", which functions literally like the main iterator with the exception that it can also be used as an item of a parent-iterator. This allows for more complicated use-cases and the support of keyboard iterators around extension points. This pull request also fixes an unreleased regression where usages of the deprecated confirmAlert function caused a nullpointer exception.
In `751343f` we forced words to wrap in a variety of places in the SCM-Manager to prevent unintentional overflow breaking the layout. This was also added to the markdown view which had implications for tables in these views including long headers that would now break in a not so pretty fashion. After investigating potential implications and checking other usages of the markdown view, we removed the problematic class again in this particular instance as it was seemingly not serving any positive purpose and removing it had no negative impact.
A new api is introduced to allow focus-based list iteration through keyboard shortcuts. The api is initially considered closed and only used in the repository overview.
Co-authored-by: Eduard Heimbuch <eduard.heimbuch@cloudogu.com>
The plugin center cache was not invalidated when the proxy configuration was changed in the global settings. This caused stale and inconsistent state to be displayed to the user while there was no feedback that something was wrong.
Global Modals as components, and as a result their registration hooks, have a lifecycle that spans the whole application. Because of this, they never get unmounted. The registration hook will simply be re-rendered with an updated "active" flag. Because the hook did not reset to its initial state when switching from "active" to "deactive", it decremented the modal count twice. Once in the cleanup of the previous hook render and once in the else block ("inactive").
Cloning repositories with LFS data via ssh requires the creation of access tokens using a subject authenticated with an already scoped token (the first scoped token is created by the ssh lfs authentication command; with this token the further requests are issued which will create further tokens for the individual lfs download requests). This failed, because such a creation was rejected without further checks. Now we test, whether the requested scope is permitted by the current scope.
This adds the possibility to load files managed by lfs to the repository import of git repositories.
Co-authored-by: Konstantin Schaper <konstantin.schaper@cloudogu.com>
Sometimes it happens that a git diff command request is performed with a non-existent target branch. This is usually fine but the underlying system might have already garbage-collected the revisions associated with that branch. In this case, the revision for that deleted branch might turn up null which currently causes a 500 error. We catch this specific corner-case and throw the correct NotFoundException instead.
In recent weeks we have created an api for declaring keyboard shortcuts and tested its usage in internal modules. After successfully verifying it, we are now exposing it for plugins to use. The api has also received some tweaks in the process to make it more flexible, such as allowing bound shortcuts not to appear in the documentation dialog or allowing shortcuts to explicitly allow event bubbling.
The pending queue is updated after all the plugins to be installed are collected, and then we already may have duplicate entries. Because of this we check right on the collecting step if the plugin was already added during this single action.
If a new branch is pushed in Git without new commits (the branch
is created on a commit that already exists in the SCM-Manager repository
and pushed without any further commit), all ancestors of the commit
the branch points to have been treated as if they were new by the
hook changeset provider. This led to severe errors like wrong
push logs (by the pushlog plugin) or re-evaluated commit messages
by the commit message checker plugin.
This fixes this wrong behaviour. If new commits are not provided by
the pack parser, no commit will be treated as a new one.
There is currently no documentation which shortcuts are available to the end user, neither within the application nor the documentation published on scm-manager.org. This PR adds the missing documentation in both places and introduces a new api for developers to add documentation when using `useShortcut`. It also improves the api for conditional shortcuts significantly.
If a POST request is submitted to the rest api for repostory permissions, the regex validator ignores the name field if it is null, which leads to an internal server error and breaks any further attempts to interact with that repository. An additional not-null constraint resolves this problem.
