Tomviz Standalone Installer Packaging

This repository builds standalone installers for Tomviz from conda-forge packages.

Output Formats

Platform	Format	Notes
macOS x64	.dmg	Unsigned — sign/notarize internally before distribution
macOS ARM	.dmg	Unsigned — sign/notarize internally before distribution
Windows	.msi, .zip	MSI is unsigned — sign internally before distribution
Linux	.tar.gz	Relocatable bundle, extract anywhere
Linux (RHEL8/9)	.rpm	Installs to /opt/tomviz; relocatable via --prefix
Linux	.flatpak	Single-file bundle, org.freedesktop.Platform runtime

All Linux formats wrap the same relocatable conda-forge bundle (package.py

• conda-pack), so they stay in lockstep: the .tar.gz is the raw bundle, the .rpm drops it into /opt/tomviz with desktop integration, and the .flatpak copies it into the sandbox's /app.

How It Works

1. Check: A scheduled GitHub Action polls conda-forge every 6 hours for new tomviz releases
2. Build: When a new version is detected, it creates a conda environment with tomviz, uses conda-pack to make it relocatable, and wraps it in platform-specific bundles
3. Release: Uploads installers as GitHub Releases

Bundled ffmpeg

Every bundle includes the conda-forge gpl variant of ffmpeg, which tomviz's movie export runs (as a separate process) to encode H.264 MP4s; only the gpl variant links libx264, which the export's quality control requires. Because the ffmpeg executable is GPL, package.py copies its license texts and a source-availability notice into share/licenses/ffmpeg/ inside the bundle (tomviz itself remains 3-clause BSD; running a GPL executable as a subprocess does not change that). verify.py checks the binary is present, is the gpl variant, encodes a test MP4 with the exact arguments tomviz uses, and that the license files made it into the bundle.

TomoPy GPU support

The Linux and Windows bundles include the CUDA 12.9 build of libtomo, which provides libtomo-accel for the TomoPy operator's "Use GPU (CUDA)" option (GPU SIRT and MLEM). The build runners have no GPU, so package.py sets CONDA_OVERRIDE_CUDA=12.9 for the solve and pins ParaView's viskores to its CPU build. Only libtomo-accel links the CUDA libraries (NVIDIA NPP), so the bundles still run without an NVIDIA GPU; using the option needs an NVIDIA driver that supports CUDA 12. There is no CUDA build for macOS. verify.py checks that libtomo-accel and NPP made it into the bundle.

Manual Build

cd packaging
conda env create -f environment.yml
conda activate tomviz-package
# The newest tomviz on conda-forge for Python 3.14, as CI picks it
VERSION=$(python ../.github/scripts/latest_conda_forge_tomviz.py 3.14 | cut -d' ' -f1)
python package.py --tomviz-version "$VERSION" --python-version 3.14
TOMVIZ_VERSION="$VERSION" cpack --config CPackConfig.cmake  # .tar.gz / .dmg / .msi

Linux RPM and Flatpak

Both consume the bundle staged by package.py (_build/install/tomviz), with VERSION set as above:

cd packaging
python package.py --tomviz-version "$VERSION" --python-version 3.14

# RPM (relocatable, installs to /opt/tomviz)
bash linux/build_rpm.sh --staged _build/install --version "$VERSION" --out _build

# Flatpak (single-file bundle)
bash flatpak/build_flatpak.sh --staged _build/install --version "$VERSION" --out _build

The committed linux/tomviz.spec and flatpak/org.tomviz.Tomviz.yaml are the configuration files needed to rebuild each package.

Build + test locally via Docker (handles RHEL containers / flatpak sandbox; uses linux/amd64 emulation on Apple Silicon). Both take an optional version and Python version, defaulting to the newest tomviz on conda-forge and 3.14:

bash packaging/linux/build_and_test_rpm_local.sh          # RPM on Rocky 8 + 9
bash packaging/flatpak/build_and_test_flatpak_local.sh    # Flatpak smoke test

Both mirror the CI tests: structural verification, install, headless smoke test under software GL, and (for the RPM) a relocation test at a non-default prefix.

Publishing to OpenChemistry/tomviz

Releases in this repo are built automatically for every conda-forge update, but they are not auto-published to OpenChemistry/tomviz. The macOS DMGs and Windows MSI produced by CI are unsigned, and publishing unsigned binaries to the user-facing repo would trigger Gatekeeper/SmartScreen warnings for end users.

Instead, specific releases are hand-picked, signed, and published manually:

1. Pick a release from this repo's GitHub Releases page.
2. Download the macOS DMGs (x64 and ARM) and the Windows MSI.
3. Sign and notarize the DMGs, and sign the MSI, using the internal signing process.
4. Create a matching release on OpenChemistry/tomviz and upload the signed DMGs and MSI along with the Linux artifacts from this repo's release.