$ the-wire · showcase
Turborepo fixes multi-language cache and watch hashing
By RepoJournal · Filed · About Vercel
Three Turborepo fixes land today: cache keys no longer fragment on machine-specific paths, watch mode deduplicates hashes across Cargo and Python, and native tasks are discoverable in task listings.
Turborepo decouples contract environment projection from raw-value task hashing [1], so strict-mode execution still receives toolchain settings without machine-specific paths, like Cargo home and rustup home, fragmenting cache keys. Watch mode now registers repository-graph execution scopes with the hash watcher and reconciles them with JavaScript package discovery [2], suppressing no-op events while retaining real changes. Implicit native tasks are now discoverable anywhere package tasks are listed [3], including `turbo ls <package>`, with a new `command` field that displays resolved native commands. Chat absorbed a workspace diff view [4] with Chat and Diff tabs, rendering sandbox changes via `@pierre/diffs` and handling oversized diffs. vercel/chat added docs for the LINE community adapter [5], registered as a valid docs code-example import and passing 467 integration tests. The AI SDK shipped patch releases across providers; MiniMax 2.0.13 [6] now sends a default `16:9` ratio for MiniMax-H3 text-to-video.
Action items
- → Pull the latest turborepo to get the cache and watch hash fixes before your next build vercel/turborepo [plan]
- → Update @ai-sdk/minimax to 2.0.13 to get the default 16:9 ratio for MiniMax-H3 vercel/ai [monitor]
References
- [1] fix: Normalize multi-language cache environment inputs ↗ vercel/turborepo
- [2] fix: Deduplicate multi-language watch hashes ↗ vercel/turborepo
- [3] fix: Show native tasks in task listings ↗ vercel/turborepo
- [4] chore: Add workspace diff view ↗ vercel/turborepo
- [5] docs: add LINE community adapter ↗ vercel/chat
- [6] @ai-sdk/minimax@2.0.13 ↗ vercel/ai