Make bumping the wire format an API breaking change

[therealfrauholle]

Hey @Urhengulas after the latest release of the changes in https://github.com/knurling-rs/defmt/pull/747 I am still having problems aligning versions in my setup :sweat:

It appears to me that we missed bumping the dependency defmt-macros of defmt in https://github.com/knurling-rs/defmt/commit/37892413e3a64797bdc213d6a644754ac8bb2eef and also earlier in https://github.com/knurling-rs/defmt/commit/1cbbb334bf508d13c1d537623fe2a4165dfc5d65, it should now depend on defmt-macros-0.3.4.

Following this I tried to wrap my head around how the wire format plays together with the SemVer guarantees, and I think that bumping the wire format is an API breaking change.

This is related to the changes introduced as of https://github.com/knurling-rs/defmt/pull/540, which solved https://github.com/knurling-rs/defmt/issues/287.

[Dirbaio]

The original rationale for decoupling the wire format version from the defmt semver version was that doing a defmt semver-major bump is a giant pain. You can't mix defmt versions in a binary (#426), and this seems fundamentally unfixable with the defmt design. So upgrading defmt requires ALL libraries using it to update. And in turn, that update requires a major bump of the library, because it'll break binaries that are still using the older version.

This happened in the defmt 0.2 -> 0.3 upgrade. It was a giant pain, it took MONTHS.

Decoupling the versions allows defmt to evolve while avoiding regularly inflicting this pain into the ecosystem. I really don't think we should walk back on this decision.

However, I remember discussing making defmt-decoder support multiple wire format versions, which it seems it never happened. Currently old defmt-decoder/probe-run versions support only wire 3, newer ones support only 4, so you have to keep switching between them if some projects have been updated and some other haven't. I agree this is not great.

If the new versions added support for 4 instead of dropping support for 3 the experience would be much better: you upgrade defmt, you get an error from probe-run saying "please update probe-run", and then everything would be working again.

[jannic]

Would it be another option to do a semver bump for defmt-decoder only when the wire format changes? That way, binaries could support multiple wire format versions. Not as nice as having multiple versions supported by defmt-decoder, but perhaps a viable option if direct support is too difficult to add?

[Dirbaio]

It'd be nicer to have a single defmt-decoder version support multiple wire format versions.

If it's one-to-one, bins can hack it by depending on multiple defmt-decoder versions at once, yes, but:

• that work needs to be repeated for each project that uses defmt-decoder
• needs manual work to add new versions, vs just cargo updateing to a newer defmt-decoder.

[datdenkikniet]

Requiring an update to projects using defmt-decoder sounds far less problematic, IMO.

Crates actually using that are far fewer, and being able to support multiple wire versions so that actual users can safely cargo install <decoder thing> without having to worry that decoding logs from projects locked to older versions of defmt completely breaks (without cargo install-ing very specific versions of <decoder thing> locked to a specific version of defmt-decoder).

The fact that probe-run has defmt-decoder = "=0.3.5" (because if it didn't, how would I ever run a defmt 3 binary without patching a version of probe-run?) so it didn't update anyways isn't particularly supportive for the use case... Great that all my projects updated but the tools didn't :/

For defmt the rationale makes sense, but for defmt-decoder I don't think there's a good enough reason if backwards compatibility is not maintained, or backwards compatibility can't be maintained by 3rd party tools.