fix: normalize Additional Host Path Mounts textarea to a comma list

[ak684]

Problem

The KOTS Additional Host Path Mounts field (sandbox_additional_host_paths, help text "one per line") is rendered into a single-quoted inline YAML scalar:

ADDITIONAL_HOST_PATHS: 'repl{{ ConfigOption "sandbox_additional_host_paths" }}'

A multi-line value inside a single-quoted scalar gets YAML-folded — the newline between entries becomes a single space. runtime-api then split only on commas/newlines, so a two-line value arrived as one space-joined token, was read as a single malformed entry, and was silently skipped → zero hostPath mounts for the sandbox pods. A customer hit this verbatim (their support bundle showed ADDITIONAL_HOST_PATHS = "/proj/a:/proj/a /proj/b:/proj/b" and the runtime-api skip log).

Fix

Normalize the textarea to a clean single-line comma list at render time, using the same splitList "\n" | trim | join idiom already used for every model textarea in this file (e.g. anthropic_models at line 31/262, azure_deployments, custom_models, …):

ADDITIONAL_HOST_PATHS: 'repl{{ $hp := list }}repl{{ range $raw := splitList "\n" (ConfigOption "sandbox_additional_host_paths" | replace "\r" "\n") }}repl{{ $e := trim $raw }}repl{{ if $e }}repl{{ $hp = append $hp $e }}repl{{ end }}repl{{ end }}repl{{ $hp | join "," }}'

This feeds the shared charts/runtime-api/templates/_env.yaml helper, so the runtime-api Deployment and the warm-runtimes/cleanup CronJobs all get the clean value. Because it emits commas (not a folded space), it works with the currently shipped runtime-api image — a Cloud-only release fixes affected installs without a runtime-api rebuild.

Relationship to runtime-api#617

Belt-and-suspenders with OpenHands/runtime-api#617, which makes the parser also split on whitespace. Either alone fixes it; together the value is clean at the source and the parser is tolerant of any separator (so already-deployed space-folded configs auto-heal on upgrade).

Validation

replicated/ change only — no chart template touched, so no chart version bump. Template braces/quotes balanced; mirrors the proven model-textarea idiom. The end-to-end behavior (space → 0 mounts; comma list → mounts present + readable) was verified live on an embedded-cluster install via runtime-api#617's reproduction.

[all-hands-bot]

✅ Review complete.

This review was performed through OpenHands Cloud Automation. You can log in and view the conversation here.

[all-hands-bot]

🟢 Good taste — Minimal, correct fix following the exact idiom already established in this file for other textarea fields.

What the change does: Replaces a naive ConfigOption call with the same splitList/trim/join normalization pattern already used for anthropic_models, azure_deployments, custom_models, and other textarea fields in this file. This prevents KOTS's single-quoted YAML scalar folding from space-joining multi-line entries before runtime-api parses them — a real customer-impacting bug.

Edge case correctness:

• Empty field → "" (empty list join = empty string, identical to before) ✅
• Single path → "path" ✅
• Multiple paths → "path1,path2" ✅
• Windows \r\n line endings → handled via replace "\r" "\n" ✅
• Empty / whitespace-only lines → filtered by trim + if $e ✅

The inline comment on line 315 is warranted — the template is opaque and the YAML-folding root cause is non-obvious to anyone who hasn't encountered it. It's a "workaround for external bug" comment in the best sense.

[RISK ASSESSMENT]

• [Overall PR] ⚠️ Risk Assessment: 🟢 LOW
  replicated/-only change; no chart templates touched; no version bump needed. The comma-separated output is already accepted by the currently deployed runtime-api. Follows a proven pattern used for every other textarea field in the same file. Fully backward-compatible.

VERDICT:
✅ Worth merging: Correct fix, minimal scope, consistent with existing patterns in the file.

KEY INSIGHT:
The fix is exactly right because it reuses the same normalization idiom already proven for all other textarea fields in this config — consistency here is a feature, not laziness.

---

This review was generated by an AI agent (OpenHands) on behalf of the user through OpenHands Automation. View conversation