Go: GoBytes+string double-copies FFI string payloads (large GET/MGET RSS) #6751

Describe the bug

In the Go client, string responses from the FFI layer are converted with:

byteSlice := C.GoBytes(..., len)
return string(byteSlice)

C.GoBytes allocates a Go []byte copy of the C buffer, and string(...) allocates a second copy. For large GET / MGET values this doubles peak Go heap traffic for every payload (binary-safe including interior NULs).

This is independent of the v2.5.0 ResponseArena payload leak (#6740 / #6741). Even with arenas releasing FFI buffers correctly, the Go binding still pays two heap copies per string response.

Expected Behavior

One copy from the FFI char* into a Go string (still preserving length and interior NULs).

Current Behavior

Two allocations / two full payload copies per string response (GoBytes then string).

Reproduction / evidence

Microbenchmark in the linked PR (1 MiB payload):

Proposed fix

Use C.GoStringN(ptr, len) in convertCharArrayToString and parseString. GoStringN is length-based and preserves interior NULs (same semantics as the current path).

Related

• (ffi): v2.5.0 leaks the full response payload of every command — ResponseArena pool parks arenas with payload buffers still populated (fixed on main in #6559 but unreleased) #6740 / fix(ffi): backport to release-2.5 — release response payload buffers when parking arenas in ARENA_POOL #6741 — FFI ResponseArena leak (orthogonal; both matter for large-value RSS)
• perf(core): zero-copy receive path for GET/MGET #6559 — Rust-side zero-copy receive (does not remove the Go binding double-copy)