On Tue, Feb 28, 2017 at 04:11:50AM +0100, Cyril Brulebois wrote:
> > (If we really wanted, we could maybe avoid the ABI bump: no library
> > functions are being added/removed, only the name and meaning of a struct
> > member (a pointer, which remains the same length). The
> > dynamically-sized buffer it points to, would change from storing an MD5
> > to a SHA256 hash, and would only cause a regression where something is
> > still trying to validate MD5).
>
> Given the number of reverse dependencies, I doubt this is worth abusing
> md5 storage for sha256 things. Bumping the ABI seems reasonable to me,
> even if that's effectively starting a mini-transition from a release
> point of view.
On second thought, let's just do it without ABI name change. For d-i
breaks don't work well, but if we update them en block this will not
show any breakage. For the rest (exactl one user) breaks works fine.
Bastian
--
It would be illogical to assume that all conditions remain stable.
-- Spock, "The Enterprise Incident", stardate 5027.3
