Release some stability plan

[rmannibucau]

Hi,

Currently Microprofile is a big sandbox where API are broken very often (too much for production applications, it is quite costly to migrate regularly and respect a company release policy). The worse is that it is often for no valid reason from an user point of view (object -> optional, dropping a method etc).
This issue is about providing a stability plan which would enable to use microprofile in production without having to pay a high price to stay up to date (or fork which seems current option).

A 5 years stability plan is something acceptable, at least for MP platform/cloud specs (metrics, health, tracing, openapi out of my head).

Romain

[Emily-Jiang]

This is something we should look at once MPWG is done. At the moment, the current policy is that we introduce breaking changes once a year. We need to revisit this. We have been tried hard not to introduce breaking changes unless it is absolutely necessary. Maybe we should review the breaking changes by the spec committer to ensure we have indeed taken this very seriously.

[rmannibucau]

We have been tried hard not to introduce breaking changes unless it is absolutely necessary.

This is not how it is perceived - and on a more personal side, it is clearly not the case. See metrics spec for instance, renaming, optional wrapping/unwrapping, and other changes are all unjustifiable except by "trying to be perfect". Same applies to health (2->3).. I'm not blaming anyone since today it is the contract of MP so it is aligned with that, but as users it is not justifiable and costs too much, therefore this issue. I also think the message i quoted shouldnt be promoted ("we try") since it is not the case - intentionnally or not) so, IMHO, it should be "we do" or we "don't" but nothing in between which is misleading and not guaranteed.

Hope it makes sense. Once again, no blaming at all since today there is no such policy, just highlighting the fact it does not work in time now API are actually no more (functionally) changing.