scaleway/compute-instance v6.0.0
Renames instance_config.enable_heic_transcoding to instance_config.enable_transcoding. This is a pure rename — the same ppa:savoury1/ffmpeg4 repository is added and the same ffmpeg package installed, at the same point in cloud-init, producing an identical instance.
The old name described one motivating use rather than the flag's actual effect. What it installs is a general-purpose ffmpeg build; libheif-based HEIC decode is the capability that made that PPA preferable to Ubuntu's own archive build, but it is one codec among everything else the build ships. A job running transform --input-file-type=png or --input-file-type=webp needs exactly the same flag, and reading enable_heic_transcoding = true on such a job would look like a configuration error rather than the correct setting. The instance_config description has been rewritten to match: it now leads with the general effect and keeps HEIC as the stated reason for the PPA choice.
This is a breaking change, and it breaks quietly — please read before upgrading. Terraform treats an unknown attribute on an object-typed variable as a warning rather than an error, so a caller that upgrades to v6.0.0 while still passing enable_heic_transcoding will not fail. tofu validate reports only the object type for input variable "instance_config" does not include an attribute named "enable_heic_transcoding", so this definition is unused, and plan and apply both succeed. The key is silently dropped, the instance comes up with no ffmpeg installed at all, and the problem surfaces much later as a transform job dying on its first file with ffmpeg's own "decoder not found" error.
So when bumping to v6.0.0, grep your configuration for enable_heic_transcoding and rename every occurrence — a clean plan is not evidence that you caught them all.
Recorded in ADR-0151, which also covers the matching pigeon-cluster passthrough rename and an unrelated object-bucket change landing separately.