Script IDE 1.16.1

Signed module, md5-verified identical to the file ignition-module-testing is running (b6ecf18a…). Ignition 8.3, signed with the Gaskony keystore.

feat: search that covers what the IDE actually edits, guards on the way out, and most of the product review's own backlog.

Nigel, 06/09/2026, after five more ideas: "do all 5 they are good" — and then
"also fix up everything from the past review. that should have all been done
already."

Added — the five

• Search covers everything the IDE edits. ProjectIndex filtered to
  library scripts while the module had grown to edit gateway event scripts, Web
  Dev handlers and pages, and named-query SQL. So search, references and (since
  1.15.0) replace covered about a quarter of what a user can open, and looking
  for a string in your own timer script returned nothing — which reads as "it
  isn't there". The NAME index stays library-only, deliberately: definition and
  quick-open symbols are about IMPORTABLE modules, and a timer script is not one.
• A save that does not parse asks once. Nothing stopped writing broken
  Python into a running gateway; on a timer script that starts failing on the
  next tick. Only an ERROR stops — a warning never does.
• Rename a script, with the call sites offered as a separate, confirmed
  project-wide replace. One resource, never a folder: the server refuses a
  folder rather than half-moving it, and the dialog says so before you type.
• Compare an override with its parent. Discard Overrides has always been
  one click and there was no way to see what it would discard — in this IDE or
  the Designer. Read-only, and it walks the parent CHAIN rather than stopping at
  the immediate parent.
• An unused report. Top-level functions and classes nothing else names.
  Name-based, and the panel says so at length: anything called from a
  Perspective binding, a Vision window or outside the project appears here and
  is not unused.

Added — the product review's backlog

• R1, impact before save. A save that removes or re-declares a top-level
  function names its call sites first. A line scan, not the AST: the question is
  about two versions of a buffer and the server only knows one of them.
• R3, run history that survives a restart. The console kept nothing;
  ExecAudit stores a SHA-256 of the source by design, so it could say a run
  happened and never what was run. Source and output now, per user, bounded.
  "Load into console" does not re-run it.
• R4, tag event scripts — the blocker was never the Designer. This type has
  been body-only since 1.1.0 waiting for someone to measure the tag-path list in
  the Designer's workspace. A real tag-change resource on the rig had the shape
  written on it all along: paths and changeTypes as JSON ARRAYS and
  enabled. Both arrays are now editable, with changeTypes as an allowlist —
  the platform ignores an unrecognised one silently, so the script would sit
  there configured and never fire.
• F4, the Administrator literal is a policy key now
  (com.gaskony.scriptide.admin.role), resolved file > -D > default. A
  gateway whose admin role is called anything else fell back to a check that
  could only say no.
• R2, half of it, and the half that is real. The review asked for last fire,
  duration and next fire per event script. The 8.3.0 SDK exposes none of
  them — there is no timer-task registry to ask, and a log line appears only
  on failure, so inventing "last run" from one would make a healthy script look
  like one that never runs. What ships is which gateway event scripts are
  FAILING, how often and when, in the Problems panel.

Not done, and why

R5 (two gateways side by side) and R6 (a Jython test runner) are
untouched. R5 is a cross-gateway authentication surface rather than a feature
flag; R6 is a product in its own right. F1 (internal or public) and F2
(prove it on a second gateway) are decisions and environment, not code.

Fixed

• Two over-broad test assertions that counted every button in the Search panel
  and broke the moment it grew a control — the same shape as 1.15.0's class-name
  collisions. They assert the result rows now, which is what they meant.

Verified

deploy_gate.py PASS (6 checks, 14 routes, none unmounted). v13 27/27,
v15_tree 18/18, v16_nav 25/25, v17_nq 45/45, v23_insight 27/27.
Java 450, Vitest 619.

The Tag Change labels came from driving the real Designer, and they had to:
the resource stores ValueChange and the Designer's Change Triggers row says
Value. Reading the JSON alone — which is how the shape was found — would have
shipped a vocabulary this IDE invented for a control that already has names
people know. Screenshot at docs/images/designer-tag-change.png.