Repository navigation
feat: migrate mooncake-related variants with schema variants.inputs - #104
Conversation
huskcass
left a comment
There was a problem hiding this comment.
No blockers. CI was still pending when I looked.
Medium: the three variants keep their old placeholder master / te_meta_data_server / device in values (the deepseek ones also keep mooncake-master.example), while the new inputs carry the same strings as example. Since values is now the base an input overrides, a site that answers nothing silently gets the placeholder host (or an empty string for glm5.3) instead of the "no valid answer" refusal the schema description promises. Dropping those keys from values (they now live in inputs) would remove the duplication and keep the refusal; the relaxed inputProblems rule and the valueSetAt branch in catalogValues would then only be needed for whatever still sets a key on purpose.
Low: catalogValues now has three-way precedence (default, then values, then example) and the schema text says "a value values gives is the base"; if nothing in the catalog needs values-over-example after the cut above, the simpler rule (default, else example, values never set it) was enough and this relaxation can be skipped.
Verdict: fine to merge once CI is green; the Medium is worth a cleanup.
— Review Golden
New
inputsaddition only, nothing will break