Enhancement-676: the string an unknown $simparam$str hands on after its $fatal is empty — it was the U+FFFD replacement character

Enhancement-676: the string an unknown $simparam$str hands on after its $fatal is empty — it was the U+FFFD replacement character

Scope: F4 of the
bug hunt of 2026-09-19.
openvaf: osdi/stdlib.c (simparam_str, the value returned after the
fatal). examples/simparamstrdef_examples/ (one check added, 14 per solver).
Handbook §2.11 row. openvaf
only — the stdlib is compiled into every model, so a model picks this up
when it is recompiled.

Suites: simparamstrdef_examples
14 of 14 per solver, both solvers (the new check fails on the E-670
binaries); full sweep 531 of 531.

What was wrong

string sinst;
analog begin
  sinst = $simparam$str("instance");
  @(initial_step) $strobe("instance=%s", sinst);
  ...

stdout:  OSDI n1: instance=�
stderr:  OSDI(fatal) n1: unknown $simparam$str "instance" (at the operating point)
         Error: a Verilog-A device raised $fatal during the operating point; aborting.

The compile-time L025 and the run-time $fatal are the LRM's (9.15: no
default, unknown name — an error). But an OSDI evaluation runs on to its end
after a fatal is raised: the simulator abandons the results (E_PANIC, the
operating-point variables left invalid) and not the code path, so whatever
the failed lookup returns reaches every later use in that evaluation — a
$strobe, a string compare, a $fwrite. The stdlib's simparam_str
returned the literal "�" after raising the flag: not a stray pointer or
uninitialised bytes, as the hunt guessed, but a deliberate sentinel — EF BF
BD, U+FFFD — which a reader cannot tell from a corrupt string.

What changed

The value is the empty string. Well defined, prints as nothing, equal to
nothing meaningful — the string twin of the 0.0 the numeric simparam
returns on the same path. The strobe above prints instance=, and
sinst == "" is true. The warning, the fatal and the abort are as before.

Verification

check	result
@(initial_step) $strobe("instance=[%s] empty=%d", s, (s == "") ? 1 : 0) after the failed lookup	instance=[] empty=1 (was instance=[�] empty=0)
the fatal and the abort	unchanged: unknown $simparam$str "instance" (at the operating point), "aborting"
an unconditional $strobe in the same evaluation	not printed, on either binary: it belongs to the iteration that was never accepted and is dropped (LRM 9.4.6, 9.7.3)
$simparam$str("instance", "d")	the default, no fatal (E-598, unchanged)
the E-670 binaries on the suite	the new check fails
full sweep	531 of 531

What this does not do

• The evaluation is not cut short at the $fatal: OSDI code cannot unwind
  out of a model, so the statements after it still run, now with a
  well-defined value. An initial-step $strobe after it prints at once, as
  every initial-step output does; an iteration-deferred one does not.
• The numeric $simparam("nosuch") without a default returns 0.0 after its
  fatal, as before.

Write-up: enhancements_doc/Enhancement-676.md