`test -x` on the pinned stable passes on a foreign architecture

Found 2026-08-27 by the ianweb session on via while scoping whether that host could meet Track D's snippet-verification gate. Verified independently in /home/neo/frank2.

The measurement

via is aarch64. The repo ships only stable_linux_amd64:

$ uname -m
aarch64
$ file stable_linux_amd64/default/stable_pinned
ELF 64-bit LSB executable, x86-64, statically linked
$ ./stable_linux_amd64/default/stable_pinned --version
cannot execute binary file: Exec format error

The guard, Makefile:3485 in seed-from-stable:

@test -x $(PXX_STABLE) || \
  (echo "No pinned stable at $(PXX_STABLE). Run: make bootstrap (needs FPC) once."; exit 1)
cp $(PXX_STABLE) $(COMPILER)

test -x passes. The file exists and carries the executable bit — and that bit is a property of the file, not a statement about whether this CPU can run it. So the precondition check reports healthy, the recipe proceeds, and the failure surfaces later at the exec.

Why this is worth a ticket rather than a shrug

The diagnostic is not merely absent, it is actively misleading. The only message a reader gets says "No pinned stable at stable_linux_amd64/default/pinned" about a path where a pinned stable is demonstrably present and executable-bitted. The reasonable conclusion is "my checkout is broken", and the actual answer is "this binary is for another architecture" — a different problem with a different fix. On via the fallbacks are shut too: no qemu-x86_64, nothing in binfmt_misc, no FPC, so the suggested make bootstrap escape hatch cannot run either.

Five sites, not one — the same guard is repeated, so a fix wants to be one helper rather than five edits:

Makefile:3485    seed-from-stable
Makefile:12631   test-fpjson
Makefile:13695
Makefile:13796

(plus 13769, a test -f on the managed-stable dir, same family.)

Suggested shape

Keep test -x as the first check — it still catches the genuinely-missing case — and add an architecture check when it passes. Cheapest sufficient form is to actually run the thing: $(PXX_STABLE) --version >/dev/null 2>&1, and on failure emit a message that distinguishes the two causes and names uname -m versus the binary's target. A file-based check works too but adds a dependency the Makefile does not currently have.

Whatever the shape: the message must separate "absent" from "present but foreign". That is the entire value of the ticket.

Scope note — this is the diagnostic, not the port

Making pxx work on aarch64 hosts is a different and larger question (publish an aarch64 stable, or document qemu-user-static), and it is A/T work that would help more than one host. See the lane bound recorded in devdocs/dev/session-roster.md for via's situation. This ticket is only about the guard lying. Do not widen it into the port.

Why it was found at all — the day's recurring shape