# Security
## Untrusted Input
CBOR input must be treated as untrusted data.
The implementation must safely handle:
- malformed encodings
- truncated input
- invalid values
- adversarial input
- deeply nested structures
Never assume input is valid.
---
# Decoder Safety
Malformed CBOR must:
- return structured errors
- never panic
- never cause undefined behavior
- never silently recover
The decoder must validate:
- input boundaries
- lengths
- integer conversions
- UTF-8 text strings
- nesting limits
- tag representations
- floating-point encodings
---
# Error Handling
Errors must be:
- deterministic
- allocation-free
- explicit
- compact
Never:
- allocate while reporting malformed input
- store dynamic error messages
- hide validation failures
Error behavior should be predictable for both callers and fuzzing systems.
---
# Unsafe Code Policy
Unsafe code is permitted only when all requirements are satisfied:
1. Safe Rust cannot achieve equivalent performance.
2. The performance improvement is measurable.
3. Soundness is formally documented.
4. A safe API is exposed.
5. Every unsafe block includes a `Safety` section.
6. Benchmarks justify its existence.
Otherwise, use safe Rust.
---
# Unsafe Code Requirements
Unsafe code must be:
- isolated
- minimal
- easy to review
- surrounded by safety documentation
Every unsafe operation must explain:
- why unsafe is required
- what invariants must hold
- why the operation is sound
---
# Memory Safety
The implementation must preserve Rust safety guarantees.
Avoid:
- unnecessary unsafe code
- unchecked assumptions
- unchecked indexing without proof
- lifetime violations
- hidden aliasing requirements
Prefer safe abstractions unless unsafe provides a measurable and justified improvement.
---
# Robustness
The implementation should prioritize:
1. RFC correctness
2. memory safety
3. predictable failure behavior
4. performance
Never trade correctness or safety for speed.