[jeany55]

The ledger the audit (#40) asked for, built before step 4: one script per feature of TrigScript, each saying which step of the way off eudplib makes it work in a map of our own, and whether a map of our own takes it today. It turned up three bugs that both builds had, now fixed.

The ledger. ledger/<area>/ holds 831 scripts. Each one's first line is // ledger: step 4, refused: a product of two values or // ledger: step 1, takes.

• tests/ledger.test.ts compiles each script as the editor does and asks emitOwn, the checks build-own makes, now shared. It fails when a first line no longer says what happens.
• It also fails when something is in no script: a kind of the IR (129, less six the front end always takes out), a name a script has from the library, a member of on, ui, memory, local, a unit, a set of units or a task, or an export of a standard-library module (340 names).
• npx tsx scripts/ledger.mts:
  • with no option, prints the table by step;
  • --write rewrites the first lines after a step lands, so the diff is the step's account;
  • --check <files> checks a few and shows the IR kinds they compiled to;
  • --missing lists what nothing reaches.

Step	Taken
0 The start	157 of 158
1 Calls and recursion	33 of 33
2 Texts with numbers	6 of 6
3 Arrays of a fixed length	53 of 54
4 Numbers	0 of 59
5 The game's conditions and actions	0 of 193
6 The game's memory	0 of 49
7 Programs, owners and players	1 of 22
8 Texts as values	0 of 43
9 Arrays that grow, the heap	0 of 27
10 Units	0 of 99
11 Tasks, async and errors	0 of 44
12 Input	0 of 31
13 The map's own triggers	0 of 13

Two steps counted as done each still have one gap: a ?: of two booleans (step 0), and a cell read at a shifted index, xs[a >> 1] (step 3). Every script a map of our own takes was also run from its laid-out bytes against the simulator, with about forty edge cases besides. All printed the same.

Refusals say what they refuse. Most read "a assign it does not lower (no reason given)", which is how the missing arithmetic stayed hidden. Now unlowered names the first part of the statement the emitter never takes, with its line: "a product of two values, line 51", "the action Set Resources", "a write of unit.sight", "a unit's hp", "an event handled (on.killed)", "an error thrown or caught (try, throw)". Each refused statement keeps its own reason. A print to the owner named by its slot is now taken. A program owned by P9–P12 is now refused, since the game runs triggers for P1–P8 only (still to confirm in the game). The maps build-own makes of the five own probes are byte for byte as before.

Fixed, in both builds:

• A field read straight off a pick was worked out when the script was built, as undefined. print(`${first(…)?.hp}`) printed "undefined", and if (first(…)?.cloaked) compiled to nothing. It is now read in the game, as u.hp is.
• The action order(…) did not compile inside a program. on.order's name was taken for it.
• music.loop(song) never compiled, not even the module's own example.

Each has a test that fails without its fix.

Open, recorded in the plan's new section "The ledger":

• A function written outside the program that keeps a state, called inside it, is worked out once when the script is built: a counter prints 1, 1, 1. Its script is kept out of the ledger until it is refused or right.
• order(…, orders.Move) makes an Order action with modifier 6.
• disabled() and preserve() inside a program pass without a word.
• Some false compile errors.
• What a program lacks next to JavaScript.
• The README's language reference predates TrigScript 4; the ledger is the list to write it again from at 4.0.

The suite: 1239 pass, the ledger included (the whole run 222 s against 215).

🤖 Generated with Claude Code

https://claude.ai/code/session_01T1CM9pLuTuAn3mgTEkjgVg