3.4 KiB
On Pull Requests
-
Before you start a PR there needs to be a Github issue and a discussion about it on that issue with a core contributor, even if it's just a 'SGTM'.
-
A PR's description must reference the issue it closes with a
For <ISSUE NUMBER>
(e.g. For #293). -
A PR's description must contain link(s) to the line(s) in the OpenStack source code (on Github) that prove(s) the PR code to be valid. Links to documentation are not good enough. The link(s) should be to a non-
master
branch. For example, a pull request implementing the creation of a Neutron v2 subnet might put the following link in the description:https://github.com/openstack/neutron/blob/stable/mitaka/neutron/api/v2/attributes.py#L749
From that link, a reviewer (or user) can verify the fields in the request/response objects in the PR.
-
A PR that is in-progress should have
[wip]
in front of the PR's title. When ready for review, remove the[wip]
and ping a core contributor with an@
. -
Forcing PRs to be small can have the effect of users submitting PRs in a hierarchical chain, with one depending on the next. If a PR depends on another one, it should have a [Pending #PRNUM] prefix in the PR title. In addition, it will be the PR submitter's responsibility to remove the [Pending #PRNUM] tag once the PR has been updated with the merged, dependent PR. That will let reviewers know it is ready to review.
-
A PR should be small. Even if you intend on implementing an entire service, a PR should only be one route of that service (e.g. create server or get server, but not both).
-
Unless explicitly asked, do not squash commits in the middle of a review; only append. It makes it difficult for the reviewer to see what's changed from one review to the next.
On Code
-
In re design: follow as closely as is reasonable the code already in the library. Most operations (e.g. create, delete) admit the same design.
-
Unit tests and acceptance (integration) tests must be written to cover each PR. Tests for operations with several options (e.g. list, create) should include all the options in the tests. This will allow users to verify an operation on their own infrastructure and see an example of usage.
-
If in doubt, ask in-line on the PR.
File Structure
-
The following should be used in most cases:
requests.go
: contains all the functions that make HTTP requests and the types associated with the HTTP request (parameters for URL, body, etc)results.go
: contains all the response objects and their methodsurls.go
: contains the endpoints to which the requests are made
Naming
-
For methods on a type in
response.go
, the receiver should be namedr
and the variable into which it will be unmarshalleds
. -
Functions in
requests.go
, with the exception of functions that return apagination.Pager
, should be named returns of the namer
. -
Functions in
requests.go
that accept request bodies should accept as their last parameter aninterface
named<Action>OptsBuilder
(egCreateOptsBuilder
). Thisinterface
should have at the least a method namedTo<Resource><Action>Map
(egToPortCreateMap
). -
Functions in
requests.go
that accept query strings should accept as their last parameter aninterface
named<Action>OptsBuilder
(egListOptsBuilder
). Thisinterface
should have at the least a method namedTo<Resource><Action>Query
(egToServerListQuery
).