FAQ

FAQ

Is Tycho production-ready?

No — and it doesn't claim to be. Tycho is an experimental proof-of-concept for one idea: implicit hierarchical arenas under value semantics. It is heavily tested for its age (every example built native and sanitized against a golden, a differential fuzzer, eight full programs run with zero compiler or runtime defects — see the repository README), but "experimental" is a scope statement, not a modesty tic: it has no security audit, no stability guarantees, and no package ecosystem. Use it to learn, to experiment, and to test the idea — not for anything where a compromise would matter. The security policy says the same, in blunter words.

No GC and no borrow checker — how is it memory-safe?

There is no reference type, so a dangling pointer is inexpressible — the bug that escape analysis exists to prevent can't be written. Memory frees per scope; values that outlive their scope are copied up. Option removes null, Result removes exceptions, indexing is bounds-checked, and copy-in/copy-out concurrency removes data races inside Tycho (concurrent FFI calls can still race). Every test runs under ASan + UBSan, plus LeakSanitizer and ThreadSanitizer.

What does value semantics cost?

No shared mutable references: you can't build a shared-mutable graph, doubly-linked list, or observer the pointer way — the idiom is a flat node pool (all nodes in one array, linked by integer index), which is also the cache-friendly layout data-oriented engines choose on purpose. Pointer-shaped data costs more, measured: a recursive trie ~1.55× C's memory, a fixed-capacity LRU ~2.8×; the flat-pool idiom brings the graph analog to ~1.3× C. Arenas reclaim at scope exit, not incrementally. The full loss column is in docs/internals/value-semantics-limits.md.

Deep-copying every value must be slow.

"Copied on assignment" is the semantic model, not the generated code. The transpiler drops the copy wherever a value is provably un-aliased — returns build in the caller's arena, acc = acc + x grows in place, b := a becomes a move when a is dead. A copy happens only when a value genuinely escapes to two live owners — exactly when a GC or refcount would also work. Measured, not asserted: see Performance in this wiki.

Where's the package manager?

There isn't one, on purpose. A package is a directory of .ty files you import by path; the corelib lives under core:. Adding third-party code is a deliberate manual act — vendor the source — never a one-line command that pulls a transitive graph you've never read.

How do I report a bug?

The Contributing page has the full answer; the short version: file an issue with a small .ty program that reproduces it, plus what you expected vs. what happened. A miscompile that a fixture in tests/ would have caught is gold — it shows where the fuzzer and suite have a blind spot.