Sentry error:
https://sentry.gitlab.net/gitlab/gitlabcom/issues/3174990

**Problem**

Incorrect sort value for branches query causes an ArgumentError

**Solution**

* Add a validation that sort has an accepted value.
* Notify user about the error

Changelog: fixed

Some specs need a clean Redis state to run successfully. We want to add
per-environment isolation using redis databases, but to do that, we
need to use flushdb instead of flushall.

Contributes to https://gitlab.com/gitlab-org/gitlab/-/issues/331297

**Problem**

 We set pagination headers during the request time(see
 Gitlab::Pagination:::OffsetHeaderBuilder).

But when the request is cached, then we skip the headers' setup.

**Solution**

Save headers in the cache and restore them if necessary

Changelog: fixed

Currently when `repository#add_branch` returns `false` in
`Branches::CreateService` we return error message that the source
reference was invalid. This is the most likely cause of this error, but
not the only one.

Another cause is creating a branch with a subcomponent that matches the
full name of an existing branch, e.g. `feature/widget` and `feature`. In
this scenario it is the new branch name that is the problem, not the
source ref, but our error message gives no clues to this.

This is a particular problem for pull mirroring, where the SHA1 of the
source branch is used rather than a human-readable name. This requires
the user to copy down the SHA and go to the source repo to attempt to
understand which branch is causing the failure.

Ideally we would have separate errors for invalid source refs and
invalid branch names, but `repository#add_branch` hides the information
we need to distinguish between them.

This commit expands the error message to mention which branch we failed
to create, making it easier for users to find out they were creating
invalid branch names.

This removes the `Links` header in favour of the standard `Link` header,
which was added in 13.1

See: https://gitlab.com/gitlab-org/gitlab/-/merge_requests/33714

Changelog: changed

Previously this was returning a 500 as the exception wasn't handled.

Changelog: fixed

Adds caching around entity generation for branches
on the API list endpoint.

Fixes Rails/SaveBang cop for spec files

- optional `pagination: 'keyset'` can be passed in to enable keyset
pagination
- adds next page link to Link and Links header

Pagination by Kaminari is slow because it requests all branches and
paginate after receiving results from Gitaly. It's a first iteration to
implement native pagination with Gitaly.

The plan for RSpec 4.0 is to disable monkey patching, reference:
https://rspec.info/blog/2013/07/the-plan-for-rspec-3/#zero-monkey-patching-mode
https://relishapp.com/rspec/rspec-core/docs/configuration/zero-monkey-patching-mode

This commit stops using RSpec monkey patching in several specs.

Issue: https://gitlab.com/gitlab-org/gitlab/-/issues/216545

Currently we display a misleading error message when a branch cannot
be created because of the incorrect reference.

I change the error message to return a reference name rather than a new
branch name.

Similar to gitlab-foss!13174 for files, Grape would try to return a
text result when the branch name ended in '.txt'. Fix this by forcing it
to always return JSON results for branches, no matter what the name.

Fixes: #217097

243 files inspected, 427 offenses detected, 427 offenses corrected

Changes `set` to `let_it_be`

This adds a first iteration of caching to `merged_branch_names` on
the Repository model. This is to address a performance issue as
raised in https://gitlab.com/gitlab-org/gitlab/issues/30536 and
it currently uses a fairly naive implementation using existing
cache stores defined on the Repository model.

Auto-correct some references in RSpec as:
- expect_any_instance_of -> expect_next_instance_of
- allow_any_instance_of -> allow_next_instance_of

Related to https://gitlab.com/gitlab-org/gitlab/issues/34997

Renamed CreateBranchService into Branches::CreateService and renamed
its usages.

Fixed related tests with new lines to make separation between the
4 stages of the test clear.

Some extra codestyle changes.

This commit handles N+1 calls being
submited at 'GET /projects/:id/repository/branches'
endpoint for protected branches.

It does so by using properly eagerloaded protected branches
relations, which also avoids N+1 ActiveRecord cached
calls.

Partially addresses:

https://gitlab.com/gitlab-org/gitlab/issues/27703

This would have made it easier to diagnose errors in
https://gitlab.com/gitlab-org/gitlab/issues/34310.

Files from which the specs are extracted:

- spec/requests/api/branches_spec.rb
- spec/requests/api/helpers_spec.rb
- spec/requests/api/protected_branches_spec.rb

The main point of this check is to ensure we aren't checking
all branches, not that we have an exact count.

https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/24034 introduced
a regression where only the first 20 branches were used to determine
whether a branch has been merged because the pagination was applied
incorrectly. Requesting the second page of branches via the API would
always have the wrong merge status. We fix this by properly paginating
the branches before requesting their merge status.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/56250

47d4890d changed the order of pagination so that the full list of
branches would be passed to Gitaly to determine which ones had been
merged, but this operation can timeout for large repositories with
many branches. We only need to determine whether the found branches have
been merged, so limit the scan to those.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/55724

Updates specs to use new rails5 format.

The old format:
`get :show, { some: params }, { some: headers }`

The new format:
`get :show, params: { some: params }, headers: { some: headers }`

Updates specs to use new rails5 format.

The old format:
`get :show, { some: params }, { some: headers }`

The new format:
`get :show, params: { some: params }, headers: { some: headers }`

In API v4 when requesting a branch with an invalid refname shows an
helpful error message: 'The branch refname is invalid'.

Signed-off-by: Rémy Coutable <remy@rymai.me>