Beagle SCM

JAB-017: Buf.grow's _grow hook is dead — growing a mapped Buf silently severs its file backing

Buf.grow (jab/buf.cpp:100-108) has an if (this._grow) mapped-backing branch that nothing ever installs (grep across jab/*.cpp, test/, jsrc/ — cont.cpp's P._grow is the unrelated PACK scratch grower), so growing an io.ram/io.mmap("rw") Buf falls into the heap-copy path: a fresh Uint8Array replaces the mapping and later msync() no longer flushes the file. API.md:118 promises "io.buf/ram: realloc-or-mremap", which the code cannot do. Found in the 2026-07-24 jab/ review; sibling of the JAB-006 contract gaps.

Input

Context

• Latent today (no caller grows a mapped Buf yet), but the failure is silent data loss the moment one does.

Goals

• Either wire _grow for mapped backings (mremap/ftruncate path) or make grow on a mapped Buf throw; API.md must match the ruling.

Constraints

• Repro-first: grow an mmap-backed Buf, write, msync, reopen — assert bytes hit the file (or assert the throw).
• Depends on the JAB-006 implement-vs-de-document ruling; land consistently with it.

WIP

Design decisions

• (pending — wire vs forbid)

TODOs

• Repro: grow-then-msync on a mapped Buf.
• Implement or forbid; align API.md:118.

Blockers and bummers

• Gated on the same API ruling as JAB-006.

Outcome

• (pending)