mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-11 22:51:28 +00:00
7d15c5b6da
Dashboard tests located controls through antd's own class names (.ant-form-item, .ant-select-selector, .ant-select-item-option). Those break when a page moves to shadcn without any behaviour changing, and they miss regressions a user would notice. Replace them with role, label, title and accessible icon-name queries where antd exposes one, and add local/no-antd-class-selectors to keep them out. The rule is enabled as an error at zero violations, so there is no budget file and no suppressions baseline. It found eight more sites a '.ant-' search missed, written as bare class names. Eleven couplings remain and carry an inline suppression naming why: antd puts role="option" only on a hidden mirror list, so the visible options have no role, no aria-disabled and a tooltip in title; Skeleton and the modal mask expose nothing at all; and one assertion's whole purpose is that no antd modal renders. 7231 tests passed before, 7233 pass after: one conflated ModelSelector case became three focused ones. |
||
|---|---|---|
| .. | ||
| eslint-rules | ||
| check-lint-budgets.mjs | ||
| gen-api-types.mjs | ||
| generate_compliance_prompts.py | ||
| lint-budget-lib.mjs | ||