[mahaloz]

A stripped Go binary's interface dispatch referenced anonymous addresses — the itab (interface-table) records that bind a concrete type to an interface rendered as dat_<addr>:

go build -ldflags="-s -w" -o hello main.go
kuna decompile hello <fn-with-iface> --json | grep itab
#   ... == &dat_4a1c20          # before
#   ... == &go:itab.main.Point,fmt.Stringer    # after

The fix (rides gotypeglobals #1540 — no new option):

• Each static itab becomes a go:itab.<concrete>,<interface> named data symbol. The two type names come from the two descriptor addresses the record points at (offsets 0 and 8), resolved through the already-decoded type blob — never from a spelled name (the [AUTOMATED] Per-call-site output type lock + Go allocation result typing (sitetypelock, gonewtype) #1542 lesson); Inter is validated KIND_INTERFACE and both descriptors must land in the type blob or the record is rejected.
• Locates itabs via moduledata's itablinks slice (go1.18–1.25) or the contiguous [types+itaboffset, …) region go1.26+ walks record-by-record (the same restructuring as [AUTOMATED] Walk go1.26+'s descriptor region: moduledata moved and the typelinks slice is gone #1285's typelinks change), with the era-stable internal/abi.ITab layout.
• Name-only symbols (how Go's own symtab names itabs), not typed facts: an early typed design demoted the polymorphic itab-address pointer and lost 42 names on unstripped gofmt — the requirement-7 sweep caught it; the fix plus an occupied-address filter (never override a name the image already has) proves LOST=0 on unstripped binaries.

Measured (stripped, additive): hello 0→57 itab refs, gofmt 0→852 (358 distinct); gotypeglobals off emits zero (byte-identical to pre-change); unstripped off==on, zero names lost.

Tests: verify_gotypes::stripped_go_itab_renders_named green under go1.22/1.25/1.27; 5 hermetic unit tests on synthetic itab blobs (both locate paths, the garbage-stop, the method-tail size). Datatests 675/675 and stages unchanged.

🤖 Generated with Claude Code