- 28.06.2021: Initial Draft
Draft
This ADR provides guidelines and recommended practices when updating Protobuf definitions. These guidelines are targeting module developers.
The SDK maintains a set of Protobuf definitions. It is important to correctly design Protobuf definitions to avoid any breaking changes within the same version. The reasons are to not break tooling (including indexers and explorers), wallets and other third-party integrations.
When making changes to these Protobuf definitions, the SDK currently only follows Buf's recommendations. We noticed however that Buf's recommendations might still result in breaking changes in the SDK in some cases. For example:
- Adding fields to
Msg
s. Adding fields is a not a Protobuf spec-breaking operation. However, when adding new fields toMsg
s, the unknown field rejection will throw an error when sending the newMsg
to an older node. - Marking fields as
reserved
. Protobuf proposes thereserved
keyword for removing fields without the need to bump the package version. However, by doing so, client backwards compatibility is broken as Protobuf doesn't generate anything forreserved
fields. See #9446 for more details on this issue.
Moreover, module developers often face other questions around Protobuf definitions such as "Can I rename a field?" or "Can I deprecate a field?" This ADR aims to answer all these questions by providing clear guidelines about allowed updates for Protobuf definitions.
We decide to keep Buf's recommendations with the following exceptions:
UNARY_RPC
: the SDK currently does not support streaming RPCs.COMMENT_FIELD
: the SDK allows fields with no comments.SERVICE_SUFFIX
: we use theQuery
andMsg
service naming convention, which doesn't use the-Service
suffix.PACKAGE_VERSION_SUFFIX
: some packages, such ascosmos.crypto.ed25519
, don't use a version suffix.RPC_REQUEST_STANDARD_NAME
: Requests for theMsg
service don't have the-Request
suffix to keep backwards compatibility.
On top of Buf's recommendations we add the following guidelines that are specific to the SDK.
When processing Msg
s, the SDK's antehandlers are strict and don't allow unknown fields in Msg
s. This is checked by the unknown field rejection in the codec/unknownproto
package.
Now imagine a v0.43 node accepting a MsgExample
transaction, and in v0.44 the chain developer decides to add a field to MsgExample
. A client developer, which only manipulates Protobuf definitions, would see that MsgExample
has a new field, and will populate it. However, sending the new MsgExample
to an old v0.43 node would cause the v0.43 node to reject the MsgExample
because of the unknown field. The expectation that the same Protobuf version can be used across multiple node versions MUST be guaranteed.
For this reason, module developers MUST NOT add new fields to existing Msg
s.
It is worth mentioning that this does not limit adding fields to a Msg
, but also to all nested structs and Any
s inside a Msg
.
On the other hand, module developers MAY add new fields to Protobuf definitions related to the Query
service or to objects which are saved in the store. This recommendation follows the Protobuf specification, but is added in this document for clarity.
3. Fields MAY be marked as deprecated
, and nodes MAY implement a protocol-breaking change for handling these fields.
Protobuf supports the deprecated
field option, and this option MAY be used on any field, including Msg
fields. If a node handles a Protobuf message with a non-empty deprecated field, the node MAY change its behavior upon processing it, even in a protocol-breaking way. When possible, the node MUST handle backwards compatibility without breaking the consensus (unless we increment the proto version).
As an example, the SDK v0.42 to v0.43 update contained two Protobuf-breaking changes, listed below. Instead of bumping the package versions from v1beta1
to v1
, the SDK team decided to follow this guideline, by reverting the breaking changes, marking those changes as deprecated, and modifying the node implementation when processing messages with deprecated fields. More specifically:
- The SDK recently removed support for time-based software upgrades. As such, the
time
field has been marked as deprecated incosmos.upgrade.v1beta1.Plan
. Moreover, the node will reject any proposal containing an upgrade Plan whosetime
field is non-empty. - The SDK now supports governance split votes. When querying for votes, the returned
cosmos.gov.v1beta1.Vote
message has itsoption
field (used for 1 vote option) deprecated in favor of itsoptions
field (allowing multiple vote options). Whenever possible, the SDK still populates the deprecatedoption
field, that is, if and only if thelen(options) == 1
andoptions[0].Weight == 1.0
.
Whereas the official Protobuf recommendations do not prohibit renaming fields, as it does not break the Protobuf binary representation, the SDK explicitly forbids renaming fields in Protobuf structs. The main reason for this choice is to avoid introducing breaking changes for clients, which often rely on hard-coded fields from generated types. Moreover, renaming fields will lead to client-breaking JSON representations of Protobuf definitions, used in REST endpoints and in the CLI.
TODO, needs architecture review. Some topics:
- Bumping versions frequency
- When bumping versions, should the SDK support both versions?
- i.e. v1beta1 -> v1, should we have two folders in the SDK, and handlers for both versions?
- mention ADR-023 Protobuf naming
This section describes the resulting context, after applying the decision. All consequences should be listed here, not just the "positive" ones. A particular decision may have positive, negative, and neutral consequences, but all of them affect the team and project in the future.
All ADRs that introduce backwards incompatibilities must include a section describing these incompatibilities and their severity. The ADR must explain how the author proposes to deal with these incompatibilities. ADR submissions without a sufficient backwards compatibility treatise may be rejected outright.
- less pain to tool developers
- more compatibility in the ecosystem
- ...
{negative consequences}
- more rigor in Protobuf review
This ADR is still in the DRAFT stage, and the "Incrementing Protobuf Package Version" will be filled in once we make a decision on how to correctly do it.
Test cases for an implementation are mandatory for ADRs that are affecting consensus changes. Other ADRs can choose to include links to test cases if applicable.