Re: Removal from testing / what's blocking our transition?
Hey Cory,
On 2024-04-19 22:17, Cordell Bloor wrote:
> On 2024-04-18 14:54, Christian Kastner wrote:
>> a number of our packages were removed from testing today [1]. It seems
>> to boil down to llvm-toolchain-17 blocking migration of
>> * rocr-runtime [2]
>> * rocm-compilersupport [3]
>> * rocm-device-libs [4],
>> but I'm not sure why. llvm-toolchain-17 is in testing [5].
>
> I think this is due to the time_t 64-bit transition. 1:17.0.6-5 is in
> testing and 1:17.0.6-9 is on unstable. The *t64 binary packages are only
> on unstable (e.g. libc++1-17t64). The ROCm packages likely depend on those.
Indeed, that is the case for at least bin:>libhsa-runtime64-1 from
src:rocr-runtime.
What threw me off a bit was that the excuse on the tracker lists
"Build-Depends(-Arch)", rather than "Depends". IIRC none of the packages
use versioned B-Ds, so 1:17.0.6-5 from testing should satisfy them.
However, that cannot be all, as rocm-device-libs has no binary
dependencies, only build dependencies. (As you can tell, it's been a
while since my last involvement in a transition).
> At the moment, the LLVM 17 transition is blocked by the migration of
> python3-defaults (3.11.6-1 to 3.11.8-1) [6]. I'm not an expert, but I
> could I imagine that migrating Python to the 64-bit time_t may be a
> difficult process given how large the Python ecosystem is and how much
> it depends on C extensions.
In addition to that, llvm-toolchain-17 also affected by its own bug [7].
In any case, I concur that this is probably because of the time_t
transition, so there is not much that we can do about it right now.
Best,
Christian
PS: As you can tell, it's been a while since my last involvement in a
major transition.
> [6]: https://tracker.debian.org/pkg/python3-defaults
[7]: https://lists.debian.org/debian-devel/2024/04/msg00358.html
Reply to: