Skip to content

Datoviz v0.4.0rc1

First release candidate of Datoviz v0.4. Datoviz is an open-source GPU rendering engine for interactive scientific visualization, written in C with a Vulkan runtime and an experimental WebGPU/WASM subset, for 2D and 3D graphics with large datasets.

v0.4 is a major upgrade to the architecture, rendering engine, and API. This is a public testing milestone, not the final v0.4 — the public API may still change before then. See the announcement for what's new; this page covers install, scope, and known limitations.

Install

python -m pip install --pre datoviz==0.4.0rc1

The six native wheels published on PyPI were downloaded, matched byte-for-byte against the canonical release artifacts, and smoke-tested on all six target platforms. You can also download the same wheels from the release assets.

Release identity and validation

Field Value
Version and tag 0.4.0rc1, v0.4.0rc1
Tagged commit 12da2a6019c2e73bf29734eed482fffcd1e376b3
Tag date 2026-07-18
Artifacts Source archive and six native wheels attached; PyPI package verified
Source SHA-512 19f87ce8ba145e23bb15db55144f1575aabd34edb17a7da3b07a26de520de9745603afc7d4cc0e7d4d2517cab5759f8ad641b9167f6eb070d57d4ff3ce43f48e

The candidate campaign passed the native build and full test suite, specification checks, generated binding checks, documentation API checks, doctests, wheel inspection, and installed-package validation. The final macOS arm64 native suite passed 1,035 of 1,045 tests, with 10 capability skips and no failures. Canonical Wheels run 29644925786 passed all 29 jobs; hosted conformance run 29645577693 passed all six lanes; TestPyPI verification run 29652477816 proved the rehearsal index; and PyPI verification run 29666589331 proved indexed-wheel byte identity and clean installation on all six targets. Static validation covered compiler checks plus wheel architecture, linkage, dependency, and ABI inspection; no separate clang-tidy campaign is claimed for RC1.

Platform Validation
Linux x86_64/aarch64 Hosted build, native inspection, Python 3.10–3.14 installed-wheel smoke, software-Vulkan conformance, examples, shaderc, and CMake consumer passed.
macOS arm64/x86_64 Hosted build and Python 3.10–3.14 installed-wheel smoke passed; both original conformance lanes passed. Post-release testing found that those lanes covered offscreen rendering but not an installed native window. The seven attended labels covered checkout-built examples, not a separately recorded exact-wheel Quickstart.
Windows AMD64/ARM64 Hosted build, native architecture inspection, Python 3.10–3.14 installed-wheel smoke, shaderc, examples, CMake consumer, and conformance passed.

Release documentation is available in the tag-pinned Get Started, examples and gallery, and C API reference.

Scope

Status Surface
Primary testing surface Vulkan core, retained scene, native app and offscreen rendering, and the Python binding (datoviz, plus datoviz.raw for exact pointer/count calls).
Experimental WebGPU/WASM promoted example subset and compute-to-render path.
Source-build-only in RC1 Qt/PyQt hosting through datoviz_qtbridge; no packaged bridge provider is available in RC1.
External High-level plotting through VisPy 2 and GSP.
Not included v0.3 source/ABI compatibility, the old object-oriented Python plotting API, and vector PDF/SVG export.

The v0.3 parity page has the detailed migration disposition. The scope and limitations above supersede the tagged feature-status page where RC1 package availability is concerned.

Known limitations

  • macOS RC1 wheel native windows are broken. The packaged library and GLFW can select different Vulkan loaders, causing glfwCreateWindowSurface() to fail. Offscreen rendering is unaffected. Upgrade to the published RC2 package for native-window use. The RC1 release evidence originally overstated this path because the exact-wheel Quickstart was not represented separately and manual labels allowed empty observations.
  • Attended macOS arm64 testing covered checkout-built interaction examples, while the exact RC1 wheel had comprehensive unattended offscreen and installed-package evidence. The other five targets have unattended hosted evidence only.
  • WebGPU/WASM covers only the promoted example subset.
  • Compute+graphics is experimental, not a general compute framework.
  • Qt/PyQt hosting is implemented but source-build-only in RC1. The wheels include the datoviz.qt adapter, and the datoviz[qt] extra installs PyQt6, but neither supplies the native datoviz_qtbridge library required for hosted rendering. Build the bridge with DVZ_ENABLE_QT_BRIDGE=ON and, if necessary, set DATOVIZ_QTBRIDGE_LIBRARY. A packaged provider is planned for RC3, preferably through conda-forge first.

Feedback

Most useful: installation on clean machines, behavior across GPUs and drivers, clarity of the C and Python APIs, the examples, and the experimental WebGPU and compute paths.

Report issues on the GitHub tracker. Include your OS, GPU, driver version, Python version, install method, Datoviz version or commit, and a small reproducer — plus a screenshot or terminal output for rendering or install problems.