vms-0d75: Ctrl/T reflexive status line (real handler over the SET CONTROL=T flag)

[baron-3dl]

What

Wires a real Ctrl/T handler onto the DCL interactive terminal. Previously SET CONTROL=T tracked a flag with no handler and Ctrl/T did nothing — a first-30-seconds fidelity tell (docs/design-vms-parity-map.md §9.1). Every VMS user hits Ctrl/T; now it prints the reflexive one-liner.

Format (clean-room, Rule 8)

NODE::USER   hh:mm:ss   [image   ]CPU=hh:mm:ss.cc PF=n [IO=n ]MEM=n

Taken from the OpenVMS User's Manual, "Interrupting Command Execution" / "Ctrl/T" — documented example GREEN::MCCARTHY 13:45:02 EVE CPU=00:00:03.33 PF=778 IO=295 MEM=315, with the fields defined as node/user, time of day, executing image, charged CPU time (hh:mm:ss.cc), page faults, I/O activity, and memory in CPU-specific pages. No VSI/HPE binary was disassembled.

Rendered example (test output): GREEN::MCCARTHY 13:45:02 CPU=00:00:03.33 PF=778 MEM=315

Data sourcing (INV-6 / Rule 9 / Rule 11)

Every field is read from the executive $GETJPI row (struct vms_procinfo) the handler fetches with vms_kif_getjpi_self() — the same source SHOW PROCESS reads. Nothing is fabricated:

Field	Source
node	ovmx_node_name() → SCSNODE (VMS node identity, not the Linux hostname)
user	info.username (executive-stamped)
time	wall clock, local hh:mm:ss
CPU	JPI$_CPUTIM (info.cputim, gated on VMS_PI_V_CPUTIM)
PF	JPI$_PAGEFLTS (gated on VMS_PI_V_PAGEFLTS)
MEM	JPI$_PPGCNT (info.pages, gated on VMS_PI_V_PAGES)

If /dev/vms is absent $GETJPI fails and the handler emits nothing — never a faked line.

Deferred honest omissions (filed, not faked)

• IO — JPI$_DIRIO/BUFIO are structurally unsourced in OVMX (their fields_valid bits are never set; src/kernel/vms_ioctl.h calls a syscall count wearing that heading a mislabel), so the token is omitted rather than fabricated.
• image name — OVMX does not yet source JPI$_IMAGNAME for a running image; empty at the DCL prompt (where no image is active anyway).

How it hooks

Readline binds Ctrl/T (0x14) to the handler, replacing the emacs default transpose-chars, gated on ctrl_t_enabled (SET CONTROL=T, default off). The status line is drawn between the current input line and a forced redraw, so it interrupts without disturbing typed input.

Test

tests/libvms/test_ctrl_t_status.c compiles the production renderer and drives dcl_format_ctrl_t_status() against a vms_procinfo shaped exactly as fill_proc_acct() fills one — the real $GETJPI boundary, no monkeypatch. Asserts: format, real-field consumption, field order (CPU→PF→MEM), image-present case, honest degrade when a valid bit is clear, and that IO= never appears (anti-fabrication, including that the struct's dirio/bufio values do not leak).

• test_libvms_ctrl_t_status PASS; libvms suite 33/33 green.
• dcl-integration, dcl-harness-selftest, terminal_identity_gate, runtime_target_gate/negctl all green.

Native-link

Edits existing DCL TUs only (dcl_terminal is already in both native-link enumerations — mk_dcl.sh, run_dcl_native.sh) and adds no cross-image symbol, so the native-link object graph is unchanged.

🤖 Generated with Claude Code