BenchForge Studio 1.1.0

BenchForge Studio — Release Notes

Version 1.0.0 · hardware profiles verified 2026-08-10

---

Read this first if you are testing against TestController

Three behaviours will look like defects and are not. They are the emulator
being faithful to hardware we measured.

1. TestController cannot connect to the E5810A persona

It will fail. So will a real E5810A.

TestController's LXIInterface.open() creates its VXI-11 link with the device
string "inst0" (hardcoded at LXIInterface.java:62). A physical Agilent
E5810A rejects inst0 with error 3, "device not accessible" — it addresses
instruments as gpib0,<primary address>, and exposes no logical device by that
name. We searched TestController's source for gpib0 and found no occurrences.

BenchForge refuses inst0 exactly as the hardware does, and the Debug Log
explains why rather than returning a bare error code. There is deliberately no
lenient mode: a client that passes here and hangs on the bench would be worse
than useless.

Full write-up, with the caveats, in
TESTCONTROLLER_OBSERVATIONS.md.

The E5810A's SICL interface name is configurable and defaults to gpib0.
Renaming it to inst0 may well make the existing client code work. We have
not tested that.

2. device_abort does nothing

The E5810A advertises an abort port that accepts TCP connections and never
speaks — not even PROG_UNAVAIL to a bogus program. Program 395184 is answered
on the core port instead, as a stub returning error 4 for every link
identifier, valid or garbage. Aborting a genuinely blocked read has no effect;
the read still runs its full io_timeout.

Measured three ways with a control proving the client sound. BenchForge
reproduces all of it.

3. WARN lines in the Debug Log are not emulator faults

They are SCPI errors a real instrument would have queued in response to your
client's traffic:

Code	Raised when
-420 Query UNTERMINATED	a read found nothing queued
-410 Query INTERRUPTED	a query arrived over an unread reply
-113 Undefined header	a query named a node the instrument lacks

All three look like plain silence on the wire, which is what makes them
expensive to diagnose. That is why the log exists.

---

What is verified, and how

Area	Status
Prologix ++ command set, framing, escaping, terminators	byte-identical to a physical GPIB-ETHERNET, firmware 01.06.06.00
Prologix instrument replies	86/86 checks against 7 physical instruments
E5810A VXI-11 protocol	link fields, read reason bits, error paths, timeouts measured and reproduced
E5810A instrument replies	86/86 checks over real VXI-11
Serial-poll status bytes	agree across both gateways
AR488	SOURCE-DERIVED ONLY — no hardware has been on the bench

Reproduce any of it:

python -m unittest discover -s tests      # 26 tests, no physical hardware
python tools/verify_offline.py            # 7 fidelity checks, no hardware
python tools/verify_hardware.py --host <prologix>
python tools/ab_instruments.py --host <prologix>
python tools/ab_instruments.py --gateway e5810 --host <e5810a>

---

What is in this release

Two gateway emulators verified against hardware. Prologix Ethernet, and a
real VXI-11/ONC-RPC E5810A — portmapper, core channel, interrupt channel and
the abort stub, not a raw-socket stand-in.

Instrument personalities: DMM, counter, function generator and power
supply, with vendor-correct reply formats. Keithley and Agilent disagree on how
FUNC? spells DC volts, on whether integers carry a sign, and on whether a
reading is a bare number or carries its elements — all reproduced.

SCPI error queue per instrument, with *ESR? latching, feeding a live
Debug Log split from the raw traffic feed. Both panes export.

A verification toolchain in tools/, documented in its own README,
including the two hardware A/B harnesses and an error-queue drainer.

---

Known limitations

• AR488 is unverified. The profile carries a warning at the top. Treat that
  persona as a best-effort reading of the firmware source until an adapter is
  on the bench.
• Port 111 usually needs elevation. The E5810A persona binds the RPC
  portmapper on 111. Without privileges it logs an ERROR line and falls back to
  the raw SCPI socket. Run elevated for full VXI-11.
• device_docmd returns "operation not supported" — measured; this gateway
  does not expose bus-level operations that way.
• The abort port is allocated once per run. The hardware allocates it per
  core-channel connection (975, 1005, 1002, 999 and 984 observed). A deliberate
  simplification.
• The HP E3631A has no TestController driver. AgilentHP E363xA.TXT covers
  the E3632A/E3633A/E3634A only. It is emulated for bus fidelity but kept out of
  the startup bench, and the library warns before assigning it.
• Unrecognised set commands do not raise -113. They are
  indistinguishable from valid ones the emulator accepts silently, so raising
  there would invent failures.

---

Reporting a problem

Export both panes from the Traffic page — Export traffic… and
Export log… — and send them together. The traffic file is what went over
the wire; the log is what the emulator thought about it. A defect is usually
obvious from the pair and very hard to find from either alone.

If the claim is that BenchForge differs from real hardware, the deciding
evidence is a run of tools/ab_instruments.py against that hardware. It
compares byte for byte and does not take anyone's word for it.