Question. Is the renderer so tied to CLI strings that the extension substrate needs a new
internal diagnostic model before msbuild, github, sarif, a future Visual Studio host or a
future LSP can be served?
Answer: no. The model exists, it is the authoritative core's, and every surface is already one function of it. OX-01 adds no diagnostic layer. It consumes the existing one.
own_bridge::verdict::Finding (Rust, authoritative since the P-022 Stage 3 cutover; the Python
reference mirrors it field for field):
| field | the brief's concept |
|---|---|
code |
code (OWN001 …) |
severity (+ advisory) |
severity. The host's --severity changes only how a finding is shown, never the verdict |
message |
message |
file, line, column (optional) |
file, start line, start column |
related: Vec<Step>, flow: Vec<Step> |
related locations and the witness path |
kind ([resource: …]), component, event, handler |
the resource the finding is about |
ignore_reason |
suppression (// own:ignore …) |
Renderers, each a pure function of a Finding:
| function | output |
|---|---|
own_bridge::render_finding(f, "human" | "github" | "msbuild", severity) |
one line per surface |
own_bridge::build_sarif |
a SARIF 2.1.0 log |
own-cli ownir --format {human,github,msbuild,sarif} exposes all four. BR-V9 pins the
bytes, and the CLI ledgers pin both engines.
The build host (owen build-check) asks the core for --format msbuild and passes the lines
through. It changes exactly one thing: it makes the origin path absolute against the project
directory, so Visual Studio's Error List never has to guess a base. No message, code or
severity is re-rendered by the host.
The host's own conditions are the OWENB codes (spec/OwenExtension.md §4):
- a descriptor rejected;
- no engine for the platform;
- a refusal.
They are rendered by the host in the same canonical MSBuild shape. They are host
diagnostics, not analysis findings, and they never reuse an OWN code.
- No column on the
msbuildline.render_msbuildprintsfile(line)althoughFindingcarries an optionalcolumn.- Effect: the Error List navigates to the line, not the column.
- Why not here: changing the line is a change to the core's pinned bytes (BR-V9, the CLI ledgers), which is foundation work.
- The line and the wording of state-protocol findings. A stale state token is reported at the region entry, and in the generic resource wording ("IDisposable local … disposed"). This is issue #393, unchanged here.
- For a future IDE host or LSP: both consume the same
Finding.--format sarifalready carries everything they need: code, level, message, region, related locations, code flows. A futureOwen.VisualStudioor language server should read SARIF, or a structured JSON of the sameFinding, from the one core, and never re-derive a finding.