Skip to content

README.md

Latest commit

 

History

History
64 lines (47 loc) · 2.52 KB

README.md

File metadata and controls

64 lines (47 loc) · 2.52 KB
Raw

Enhancement-492 — a node named only in a control position

python3 verify_ctrlnode.py

37 checks, a few seconds. 19/37 against the pre-fix binary — 18 checks discriminate.

What it is

E, G and S take a controlling node pair, and those two names were bound exactly the way the output pair is — INPtermInsert, which creates the node. So a typo simply invented a node and the run continued against it.

The one that answers anyway

For E and G the invented node has no path to ground, the matrix goes singular, and the run fails — with the wrong explanation ("check node nosuch", a node the user never wrote). A switch does not fail. It only reads its control voltage to decide open or closed and stamps nothing for it, so the matrix stays non-singular and the solve succeeds:

v(b) state
S1 a b ctl 0 sw 0.999001 CLOSED
S1 a b nosuch 0 sw 9.99999e-07 OPEN

rc=0, no diagnostic. A factor of a million from one mistyped character.

A phantom reference is dangerous exactly where it is read but not stamped.

Everything else already checked

  • .ic / .nodeset → "IC on non-existent node - nosuch, ignored"
  • F, H, W, B-source i() → "unknown controlling source vnope"
  • all thirteen output constructs name a missing vector
  • warn_physics validates the switch's own ron, roff, vh

Only the controlling-node pair skipped it.

The controls

Half the suite pins what must not move, because the fix sits in the node table every device shares:

  • a control node that is connected — including the source's own output (E1 out 0 out 0 2), one defined later in the deck, ground, and one passed through a subcircuit port.
  • F, H and W, which already named their missing controlling source.
  • .ic on a non-existent node — E-429's own path, whose devRef mechanism this reuses.
  • a real Verilog-A $fatal, still named as such and still pointing at its OSDI(fatal) line.
  • both solvers on an ordinary divider, and a mixed-signal KLU run, which must get no empty-matrix note.

The model

ctrlnode.va holds one module, vfatal, which raises $fatal above a trip voltage. It is the control for the other half of the round: CKTop's abort message names Verilog-A, and it used to name Verilog-A for any E_PANIC that reached it — including decks containing no Verilog-A device at all. When a $fatal really is raised, the message must still say so.