Breaking interface changes in a patch release

[tj-cisco]

Looks like you introduced breaking interface changes between 2.1.0 and 2.1.1. Please be mindful of downstream consumers of your library when making interface changes, the addition of secret_updated to the ConnectionCallback trait is a breaking interface change that should have been done in version 3.x.x not in a 2.x.x release. See https://semver.org/

[michaelklishin]

@tj-cisco not that I expect semver.org dogmatists to be very receptive to my opinion but I'll share it anyway, as a RabbitMQ core team member and a contributor to this client.

I'd argue that the lack of secretupdate (an AMQP 0-9-1 extension critical for JWT token adoption) support in 2025 was a bug, and bumping the major version over something like this would be ridiculous.

A minor version bump would be reasonable.

[tj-cisco]

I would disagree based on the way cargo treats version numbers in accordance with semver and this change breaks any downstream application that uses a custom ConnectionCallback implementation and not DefaultConnectionCallback. If this was done in 2.2.0 then it will still break down stream applications that use the default version requirement syntax for any minimum version before and including 2.1.0 (e.g. version="2.0.0" or version="2.1.0" both of these allow upgrading to any version < 3.0.0). The tilde and wildcard syntaxes would have protected downstream applications if this was done as a minor release but these are not the default behaviour. I'm not disagreeing with the importance of adding secret_updated, but please remember there are other consumers of libraries that should not have to make code changes to accommodate interface changes in a patch release or any version that is labeled interface compatible with their required minimum version under Rust's versioning semantics. If down stream applications needed secret update support then it is more than reasonable for them to have to change their version number requirement to get that feature.

[gftea]

@tj-cisco
I agree it does break the semantic of versioning.
Wonder how much time it takes for you to fix this in your application code?