Review Board 3.0.16 Release Notes¶

Release date: January 7, 2020

Packaging¶

Added an explicit dependency on requests-oauthlib>=0.8,<=1.0 and on django-braces==1.13, in order to work around conflicts caused by two of our dependencies.
rbintegrations 1.0.1 is now required.

Extensions¶

Custom SCMTools can now (and should) set their own registration IDs.

Subclasses should now set SCMTool.scmtool_id to a custom, unique ID. If not set, an ID will be computed based on the Python Entrypoint that registered it.
Custom hosting services can now (and should) set their own registration IDs.

In previous releases, any subclass of HostingService provided by an extension would use the service’s name as its unique ID, but this wasn’t very future-proof and led to various issues.

Subclasses should now set HostingService.hosting_service_id to a custom, unique ID. If not set, a new ID will be computed based on the name, but this should not be relied on going forward.
Custom hosting services can now normalize patches to a file before it’s applied.

This is an advanced feature for hosting services that may need to pre-process an uploaded patch to, for instance, normalize file encodings change some content to make it apply to a version of a file fetched from the repository. This is done by overriding the new HostingService.normalize_patch() method.
Added support for custom configuration forms for plain SCMTool-backed repositories.

The repository configuration page now allows SCMTools to provide their own forms for authentication and repository details, allowing them to augment or replace the Username, Password, Path, and Mirror Path fields. This can allow them to collect other types of data or provide options for interfacing with the repository. It mirrors the functionality that hosting services already have.

This is done by setting SCMTool.auth_form to a subclass of BaseSCMToolAuthForm and SCMTool.repository_form to a subclass of BaseSCMToolRepositoryForm.

Web API¶

Added API for updating a user’s name, e-mail address, and active flag.

Clients can now HTTP PUT to the User Resource to set the first_name, last_name, email, and is_active fields.

Users cannot change these values for other users unless they are administrators or have the special auth.change_user permission.

The name and e-mail address fields can only be updated if supported by the configured authentication backend. The backend will also be notified of the new values, providing the same behavior as if the user updated these fields on their My Account page.

Use the is_active flag to disable or re-enable user accounts.
Added an option to the Review Request Resource for filtering review requests by branch.

Callers can now pass ?branch=... to specify a branch to filter by. This is case-sensitive.

Note

At the moment, this does not make use of any database indexes, so for very large installs, this filter may be slow. We are not providing a database index for branch fields at this time, in order to avoid impacting the upgrade process for these servers.

If administrators need to make use of this API and are seeing slowdown, an index on the branch column on the reviews_reviewrequest table can be added manually, but please test this first on a copy of the database.

Bug Fixes¶

Dashboard¶

Fixed some problems calculating counters for the Dashboard sidebar.

We’ve found a few cases in which counters were sometimes calculated incorrectly. This can happen when changing ownership of a review request to a user whose account data wasn’t fully set up (such as a new user who hasn’t yet logged in), or when joining or leaving review groups.

Search¶

Updated user information is now included when performing an incremental search re-index.

This depends on the user having logged in since the information was updated. It will not index new information for users marked as inactive (disabled). This is due to the limited way in which we can currently detect information updates for users.

Repository Configuration¶

Saving the repository form no longer wipes out any custom content in the repository’s extra_data field.
Attempted to prevent browsers from auto-filling usernames and passwords when configuring repositories.

Browsers try pretty hard to auto-fill anything that looks like a username or password field with any stored login credentials, and would try to do this for a repository’s username and password fields. We had workarounds in the past for this, but browsers have since worked around those. We now employ some new, safer tricks that should prevent this from occurring.
Repository access control lists are now cleared when marking a repository as publicly-accessible.

This works around an obscure bug involving Local Sites that could occur when adding a user to a repository’s access list, marking the repository public, removing the user from the Local Site, and then attempting to re-save the repository.
Removed the Use hosting service’s bug tracker checkbox when the option isn’t available during repository configuration.
Improved the Publicly Accessible checkbox to appear more correct based on configuration.

The wording led people to misunderstand the purpose of the checkbox. We now label it differently depending on whether the repository is on a Local Site or whether the server allows anonymous access.

Beanstalk¶

Fixed Subversion diffs for files that contain SVN keywords.

Beanstalk previously didn’t include the API needed to let us query SVN keyword values for files, preventing us from processing diffs that made use of them. They were kind enough to add an API for us, so we’re now able to fully support all Subversion diffs.

Bitbucket¶

Fixed crashes when Bitbucket WebHook payloads are missing information.

We’ve completely bullet-proofed the payload parsing in order to ensure that we’re not susceptible to bugs or changes on Bitbucket’s side.
Limited the number of API requests we’ll make to Bitbucket when processing pushed commits in order to close review requests.

When Bitbucket sends information on pushed commits via a WebHook, it sometimes marks them as “truncated,” requiring that we perform our own API requests to fetch the information. This could be one page of data, or hundreds, which can swamp the server. We now limit our requests to 5 pages of commits.

Any review requests marked on commits on subsequent pages will not be closed automatically.

Subversion¶

Fixed parsing of diffs containing binary file entries that include property changes.

Contributors¶

Christian Hammond
David Trowbridge