On Wed, Jun 23, 2021 at 12:22:52PM -0000, Robie Basak wrote:
> > Regression potential: low, no-change rebuilds with the updated GCC
> 10.3.
> Won't this break any users who have built their own binaries against
> libgphobos1?
Yes, it will.
A proper fix for this must include upgrade handling. There are two ways to
accomplish this:
- new libgphobos1 declares versioned Breaks: against the complete list of
reverse-dependencies that use the old ABI; or
- package shipping libgphobos.so.1 gets a new name such as libgphobos1final
or libgphobos1stable or libgphobos1abi or something
I prefer the latter over the former for aesthetic reasons (including the
fact that it handles out-of-archive packages). But given the overall small
number of packages using this library (13 reverse-dependencies in focal, and
the problem went unreported by users, it was only picked up via
autopkgtests) I think the former would be acceptable in this case.
On Wed, Jun 23, 2021 at 12:22:52PM -0000, Robie Basak wrote:
> > Regression potential: low, no-change rebuilds with the updated GCC
> 10.3.
> Won't this break any users who have built their own binaries against
> libgphobos1?
Yes, it will.
A proper fix for this must include upgrade handling. There are two ways to
accomplish this:
- new libgphobos1 declares versioned Breaks: against the complete list of dependencies that use the old ABI; or
reverse-
- package shipping libgphobos.so.1 gets a new name such as libgphobos1final
or libgphobos1stable or libgphobos1abi or something
I prefer the latter over the former for aesthetic reasons (including the dependencies in focal, and
fact that it handles out-of-archive packages). But given the overall small
number of packages using this library (13 reverse-
the problem went unreported by users, it was only picked up via
autopkgtests) I think the former would be acceptable in this case.