Update to Ubuntu 26.04 and ROOT 6.38.04

[tvami]

I am adding a new package to the container, here are the details.

What new packages does this PR add to the development image?

• Update to Ubuntu 26.04 and
• ROOT 6.38.04

Check List

• I successfully built the container using docker

• I was able to build ldmx-sw using this new container build

Yes with LDMX-Software/ldmx-sw#2047

• I was able to test run a small simulation and reconstruction inside this container

• I was able to successfully use the new packages.

I can compile and run ldmx-sw

[tomeichlersmith]

Just as a heads up, when I was experimenting with bumping the ROOT version, the image was failing to push to DockerHub: https://github.com/LDMX-Software/dev-build-context/actions/runs/25692264293

This error happens at the end of the build and is fairly opaque:

#38 exporting cache to registry
... enumeration of cache layer digests being pushed
#38 ERROR: error writing layer blob: failed to copy: unexpected status from PUT request to https://registry-1.docker.io/v2/ldmx/dev/blobs/uploads/6dfe44a1-d8e5-4101-96b0-60ecb367ce2e?_state=YWmJR8ZplwmAgIsMakD5kyA9Din_n16W7g8aDlwfGqp7Ik5hbWUiOiJsZG14L2RldiIsIlVVSUQiOiI2ZGZlNDRhMS1kOGU1LTQxMDEtOTZiMC02MGVjYjM2N2NlMmUiLCJPZmZzZXQiOjAsIlN0YXJ0ZWRBdCI6IjIwMjYtMDUtMTJUMTk6NTI6MjkuMzU4NzQzMzg2WiIsIkV4cGlyZXNBdCI6IjIwMjYtMDUtMTJUMjA6MjI6MjkuNDM2NjE3MzY5WiIsIlVzZXJuYW1lIjoi***IiwiVXNlclVVSUQiOiJiY2I0ODZmOS04YmFhLTRiNTItOWQwMi1hZGJmZjg0MjFmYzIiLCJTZXNzaW9uSUQiOiJkY2tyX2p0aV96c3l5QTlmZkpHYjFZdkU0eDRSRnBVY3dvYWM9In0%3D&digest=sha256%3A0d116fb00beae894036892d63cedd3769966bc245456841abee862cc9d7fb8c7: 400 Bad request

I fear that it has to do with the caching since it fails in the build step of trying to push the cache. From a little searching, this looks like the layer being pushed is too large. Either we need to drop all caching (which works) or maybe we can get away with dropping the mode=max parameter? I am uncertain. Caching is not necessary, it just helps speed up re-builds when we merge a PR.

Another solution is to look back at customizing the CI workflow to cache locally on the runner. This makes sense for us since we just have the single VM acting as the runners doing the building. Right now, we treat it like two runners and have GitHub distribute the load, but we could treat it like one runner and do the parallel-building within the action itself.

[tomeichlersmith]

My concerns were confirmed. The amd64 run failed to push its cache to DockerHub like what happened on my branch. https://github.com/LDMX-Software/dev-build-context/actions/runs/25886123565/job/76078111961

I got around this issue by remove the cache-to line in the ci.yaml, but that means every run of the CI will build the image from scratch (taking two days ish).

The largest layer in the current cache on DockerHub (so the largest layer that was successfully pushed) is below 1GB.

tom@zuko:~/code/ldmx/dev-build-context$ docker buildx imagetools inspect ldmx/dev:build-cache-amd64 --raw | jq '.layers | map(.size) | max' | numfmt --to=si
726M

I think it makes sense that they would prevent individual layers from exceeding 1GB, so that is probably the issue. I don't know which layer is exceeding the size limit but I suspect its the ROOT one since ROOT is the one changing. I don't know how ROOT would increase the layer size by ~275MB though if that is the issue.

[tvami]

OK so I guess we have to remove the cache-to ?

[tomeichlersmith]

Yea :/ pending some other solution I'm unaware of

[tvami]

cache-from too, right?

[tvami]

So our old GEANT does not work out with the gcc15 within ldmx-sw (I think we have seen this issue with the lto build too, so it's not completely new), so I compile GEANT with gcc13 in 7e2efee

I have spent some time on trying to make the gcc15 based GEANT work, but at this point I gave up. Especially since we plan to update GEANT too anyway

[tvami]

So I should wait for https://github.com/LDMX-Software/dev-build-context/actions/runs/26109399080 to finish before merging, right?

[tomeichlersmith]

Yes, I would - then we will have a build at ldmx/dev:ubuntu_26.04 for easier testing across machines and at least see which versions of ldmx-sw are supported.

[tvami]

Yes, I would

Sounds good

least see which versions of ldmx-sw are supported.

I expect a lot to fail w/o
LDMX-Software/ldmx-sw#2047

[tomeichlersmith]

Yea, which motivates updating the legacy interop patches https://ldmx-software.github.io/dev-build-context/interop.html and maybe updating the oldest ldmx-sw version that we commit to compiling with the image.

[tvami]

Hmm so it's not failing?

https://github.com/LDMX-Software/dev-build-context/actions/runs/26109399080/job/76782195508?pr=188

[tomeichlersmith]

Nope, it takes ~6hr to build x86 and ~48hr to build the ARM (since its emulating ARM on an x86 machine). After both successfully build, the image is merged into one "image manifest" so ARM and x86 users can pull from the same tag. At this point, it runs the ldmx-sw tests.

The specific job your linked is just reading in the ci/ldmx-sw-to-test.json file and creating GitHub jobs from them. Those "Test X.Y.Z" Jobs will spawn after the build+merge completes.

Edit: You can see the final job tree as an example here: https://github.com/LDMX-Software/dev-build-context/actions/runs/23150843005