Bug#1118939: rocr-runtime: Makes pymia to FTBFS due to unmet build-depends
control: reopen -1
Hi Cory,
there are two mistakes here, as I see it:
On 2025-10-25 20:32, Cordell Bloor wrote:
> The current version of libhsa-runtime64-1 legitimately breaks
> libamdhip64-5 at runtime [1], but there are many users that have
> libamdhip64-5 installed that don't actually use the features it
> provides. To unblock those users, I've removed the breaks annotation
> until the libamdhip64-6 transition is completed on unstable.
First: many users not being affected by this breakage is insufficient
grounds to accept this breakage.
The Breaks-relationship was objectively correct. Please re-add this.
rocr-runtime must not migrate to testing like this. Whether
libamdhip64-6 is available or not is irrelevant, as on a system with
libamdhip64-5 installed, a user may still elect to update only
libhsa-runtime64-1.
The Breaks enforces a correct upgrade path.
Second: even with the removed Breaks, this doesn't fix Santiago's issue
because Santiago's build is still trying to install libamdhip64-5, which
is no longer available in unstable.
The same applies to any other package that depends on libamdhip64-5.
You need to coordinate a transition with the Release Team. Here is the
transition tracker [1] that was created when the library bump was
detected, which should cover all packages that B-D on the problematic
versions.
As part of this transition, you'll need to file bugs with all affected
packages, so that they update their B-Ds. Normally, this is all done
while the breaking package is still in experimental, so the breakage in
unstable is kept as short as possible.
Don't worry, this happens to everyone at least once. But this still
needs to follow the correct path out.
Best,
Christian
[1]: https://release.debian.org/transitions/html/auto-rocm-hipamd.html
Reply to: