@genroc/eval-node 0.0.0-edge.d773aa8 → 0.0.0-edge.dadd892

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -96,15 +96,18 @@ The `input` of the external task IS the evaluation request:
96
96
 
97
97
  ```jsonc
98
98
  {
99
- "code": "return { fee: input.amount * 0.1 };", // required — an async function body
100
- "input": { "amount": 250 }, // optional — bound as `input`
101
- "timeout_ms": 5000 // optional — default 5000
99
+ "code": "export default (input) => ({ fee: input.amount * 0.1 });", // required — a module
100
+ "input": { "amount": 250 }, // optional — its argument
101
+ "timeout_ms": 5000 // optional — default 5000
102
102
  }
103
103
  ```
104
104
 
105
- `code` is the **body of an async function**, so `await` works and the value reaches genroc
106
- through `return`. It is compiled with `input` and `require` as parameters, under
107
- `"use strict"` `require` is what a bundled `import` of a node builtin lands on.
105
+ `code` is an **ES module**, and the evaluator imports it and calls its **default export** with
106
+ `input`. The value it returns reaches genroc; the function may be async, and so may the module's
107
+ top level. A module exporting anything else or nothing — is a `compile_error`: there is
108
+ nothing to call, and so is one that will not parse; a throw from its top level is a `threw`
109
+ like any other. `import` of a node builtin resolves as it does anywhere else, and that is what
110
+ a bundled builtin lands on.
108
111
 
109
112
  ## The answer — the failure kind IS the error code
110
113
 
@@ -140,12 +143,14 @@ no retry policy to write.
140
143
  type: external
141
144
  input:
142
145
  code: |
143
- if (input.amount > 100) {
144
- const e = new Error('amount over the limit');
145
- e.name = 'LimitExceeded';
146
- throw e;
146
+ export default function (input) {
147
+ if (input.amount > 100) {
148
+ const e = new Error('amount over the limit');
149
+ e.name = 'LimitExceeded';
150
+ throw e;
151
+ }
152
+ return { fee: input.amount * 0.1 };
147
153
  }
148
- return { fee: input.amount * 0.1 };
149
154
  input: "$: input"
150
155
  result_schema: { type: object, properties: { fee: { type: number } }, required: [fee] }
151
156
  raises:
@@ -164,7 +169,7 @@ no retry policy to write.
164
169
 
165
170
  - id: script_failed
166
171
  switch:
167
- - case: 'error.data.name == "LimitExceeded"'
172
+ - case: 'last_error.data.name == "LimitExceeded"'
168
173
  raise: { code: limit_exceeded, message: "the script rejected the amount" }
169
174
  - raise: { code: script_failed, message: "the script failed" }
170
175
  ```
@@ -227,12 +232,14 @@ the inferred type of what the definition passes; `Output` is what it declares
227
232
  `tsc --noEmit`, and bundles — so **a type error is a failed apply**, and a stored definition
228
233
  cannot hold code that failed to typecheck.
229
234
 
230
- The bundle is emitted as CJS and wrapped as a function body, so the evaluator needs to know
231
- nothing about modules. Imports resolve through TypeScript under the same config the check
232
- ran with, so a `paths` alias that typechecks also bundles. They are inlined at build time, so the string a
233
- definition version stores is self-contained forever with one exception: **node builtins
234
- stay as `require` calls**, which the realm satisfies. A package is frozen into the
235
- definition; `node:fs` is resolved by whatever runner executes it.
235
+ The bundle is one self-contained ES module whose default export is the author's own nothing
236
+ wraps or rewrites it, so what the editor checks and what the realm runs are the same module, and
237
+ **a missing default export is a failed apply** rather than a fault at run time. Imports resolve
238
+ through TypeScript under the same config the check ran with, so a `paths` alias that typechecks
239
+ also bundles. They are inlined at build time, so the string a definition version stores is
240
+ self-contained forever — with one exception: **node builtins stay as imports**, which the realm
241
+ resolves. A package is frozen into the definition; `node:fs` is resolved by whatever runner
242
+ executes it.
236
243
 
237
244
  ### Your tsconfig, your types
238
245
 
@@ -258,8 +265,9 @@ never read by genroc, because genctl doubles every `$` on splice.
258
265
 
259
266
  ## The realm — one Worker per execution
260
267
 
261
- `evaluate()` starts a Worker (`realm.ts`), posts the code into it, and races the reply against the budget;
262
- `terminate()` runs on every path. That thread is what the contract rests on, and it buys
268
+ `evaluate()` starts a Worker (`realm.ts`), posts the code into it where a module loader hook
269
+ serves it to `import()`, so the engine numbers the script's frames from the author's own source
270
+ and races the reply against the budget; `terminate()` runs on every path. That thread is what the contract rests on, and it buys
263
271
  exactly three things the previous in-process evaluator could not:
264
272
 
265
273
  - **The budget is enforced, not merely reported.** A synchronous `while(true){}` never
@@ -271,16 +279,15 @@ exactly three things the previous in-process evaluator could not:
271
279
  - **The script's mistakes stay the script's.** An uncaught throw, and `process.exit()`, end
272
280
  the realm and come back as a `422` — neither reaches the runner.
273
281
 
274
- It costs about **50ms per execution** end to end for a trivial script (Node 24, M-series
275
- laptop), a 200 KiB body about 63ms. Roughly 19ms of that is Node re-stripping `worker.ts`'s
276
- types on every realm — precompiling it to JavaScript would buy that back, and is deliberately
277
- not done: a build artefact that goes stale against its source fails silently, and this file
278
- is the one where a wrong line number is invisible.
282
+ It costs about **27ms per execution** end to end for a trivial script (Node 24, M-series
283
+ laptop), a 200 KiB body about 30ms. Roughly 11ms of that is Node re-stripping `realm.ts`'s
284
+ types on every realm — precompiling it to JavaScript measures 17ms and would buy that back, and
285
+ is deliberately not done: a build artefact that goes stale against its source fails silently,
286
+ and this file is the one where a wrong line number is invisible.
279
287
 
280
- That also changes an old trade-off. A subprocess per execution measures ~48ms here within
281
- noise of the thread where on the previous runtime it was ten times the thread's cost. The
282
- thread no longer wins on price, and a subprocess contains the two things a thread cannot
283
- (below), so it is the live upgrade path rather than a theoretical one.
288
+ A subprocess per execution measures ~55ms here, twice the thread rather than the ten times it
289
+ cost on the previous runtime. It contains the two things a thread cannot (below), so the upgrade
290
+ path is a live one at a price worth naming rather than a theoretical one.
284
291
 
285
292
  ## What this is not
286
293
 
@@ -291,7 +298,7 @@ thread no longer wins on price, and a subprocess contains the two things a threa
291
298
  it needed was surface with nothing behind it. A value that must survive a retry belongs in
292
299
  the definition, passed through `input`.
293
300
  - **Not a sandbox.** The realm isolates *execution*, not *authority*: a script gets the
294
- worker's filesystem, network and environment, and `require` of any node builtin. That is
301
+ worker's filesystem, network and environment, and any node builtin it imports. That is
295
302
  deliberate — a script task is meant to do real work — but the trust boundary stays the
296
303
  same-trust-domain one (your genroc, your worker host). It is not the multi-tenant story, and
297
304
  nothing here should be mistaken for one. Pulling does move the boundary in one useful way:
@@ -305,9 +312,9 @@ thread no longer wins on price, and a subprocess contains the two things a threa
305
312
  - **Concurrency is capped by this worker, not by genroc.** Each evaluation is a thread, and
306
313
  `CONCURRENCY` is how many it will claim at once. Raising it past what the host can run turns
307
314
  a queue back into threads fighting over a core.
308
- - **Not where imports and type checking happen.** The evaluator still takes one
309
- self-contained function body and knows nothing about TypeScript; `import.ts` is what turns
310
- a module into that body, at author time. See above.
315
+ - **Not where imports and type checking happen.** The evaluator takes one self-contained
316
+ module and knows nothing about TypeScript; `import.ts` is what bundles an author's script and
317
+ its dependencies into that module, at author time. See above.
311
318
  - **`eval.ts` and `realm.ts` know nothing about genroc.** `worker.ts` is the entire
312
319
  queue-facing half, which is what keeps the containment strategy swappable — and what lets
313
320
  the realm's own properties be tested by calling `evaluate()` directly.
package/dist/eval.js CHANGED
@@ -1,6 +1,7 @@
1
- // Evaluation core: run a code string in its OWN realm and classify every outcome into one of
2
- // the failure kinds in README.md. Nothing here knows about genroc — worker.ts is the only
3
- // thing that talks to the queue — so this stays testable and the containment stays swappable.
1
+ // Evaluation core: import a script module in its OWN realm, call its default export, and
2
+ // classify every outcome into one of the failure kinds in README.md. Nothing here knows about
3
+ // genroc — worker.ts is the only thing that talks to the queue — so this stays testable and
4
+ // the containment stays swappable.
4
5
  //
5
6
  // The containment is a Worker per execution (realm.ts). It is what makes the budget real:
6
7
  // a synchronous busy loop never yields, so no in-process timer can interrupt it, and only a
package/dist/import.js CHANGED
@@ -7,9 +7,10 @@
7
7
  // code; "build" typechecks and bundles. A separate types hook would mean a second `tsc`
8
8
  // over the same project.
9
9
  import { spawn } from "node:child_process";
10
+ import { existsSync } from "node:fs";
10
11
  import { access, mkdir, writeFile } from "node:fs/promises";
11
12
  import { builtinModules } from "node:module";
12
- import { dirname, join, relative } from "node:path";
13
+ import { dirname, join, relative, resolve } from "node:path";
13
14
  import { fileURLToPath } from "node:url";
14
15
  import commonjs from "@rollup/plugin-commonjs";
15
16
  import json from "@rollup/plugin-json";
@@ -86,7 +87,9 @@ function objectType(s, used) {
86
87
  const required = new Set(s.required ?? []);
87
88
  const lines = [];
88
89
  for (const [key, sub] of Object.entries(props)) {
89
- const doc = typeof sub.description === "string" ? ` /** ${sub.description} */\n` : "";
90
+ const doc = typeof sub.description === "string"
91
+ ? ` /** ${sub.description} */\n`
92
+ : "";
90
93
  lines.push(`${doc} ${propKey(key)}${required.has(key) ? "" : "?"}: ${tsType(sub, used)};`);
91
94
  }
92
95
  if (s.additionalProperties && typeof s.additionalProperties === "object") {
@@ -102,7 +105,7 @@ function union(parts) {
102
105
  }
103
106
  const IDENT = /^[A-Za-z_$][A-Za-z0-9_$]*$/;
104
107
  const propKey = (k) => (IDENT.test(k) ? k : JSON.stringify(k));
105
- const identifier = (n) => (IDENT.test(n) ? n : `Def_${n.replace(/[^A-Za-z0-9_$]/g, "_")}`);
108
+ const identifier = (n) => IDENT.test(n) ? n : `Def_${n.replace(/[^A-Za-z0-9_$]/g, "_")}`;
106
109
  function deref(s, defs) {
107
110
  let cur = s;
108
111
  for (let i = 0; cur && typeof cur.$ref === "string" && i < 16; i++) {
@@ -110,21 +113,14 @@ function deref(s, defs) {
110
113
  }
111
114
  return cur;
112
115
  }
113
- /** The manifest's `input` is the type of the whole ACTION input — for /eval that is
114
- * `{code, input, timeout_ms, …}`, and only its `input` field is bound as the script's
115
- * parameter. genroc cannot know that; this file owns the evaluator's wire contract, so
116
- * the navigation belongs here rather than in genctl. */
117
- function scriptInput(site, defs) {
118
- const action = deref(site.input, defs);
119
- const bound = action?.properties?.input;
120
- return bound ?? site.input;
121
- }
122
116
  /** Emits one named type per reachable $def rather than inlining: a task output may
123
117
  * reference itself (specs/recursive-type-inference.md) and inlining would not terminate. */
124
- function declarations(site, defs) {
118
+ function declarations(at) {
119
+ const { site, where } = at;
120
+ const defs = where.$defs ?? {};
125
121
  const used = new Set();
126
- const input = tsType(scriptInput(site, defs), used);
127
- const output = tsType(site.output, used);
122
+ const input = tsType(site.types?.Input, used);
123
+ const output = tsType(site.types?.Output, used);
128
124
  const emitted = [];
129
125
  const done = new Set();
130
126
  while (true) {
@@ -137,7 +133,7 @@ function declarations(site, defs) {
137
133
  }
138
134
  return [
139
135
  "// Generated by genroc. Do not edit - regenerate with `genctl types`.",
140
- `// ${site.process}${site.task ? ` / ${site.task}` : ""} (${site.pointer})`,
136
+ `// ${where.name} (${address(site.pointer)})`,
141
137
  "",
142
138
  ...emitted,
143
139
  emitted.length ? "" : "",
@@ -165,7 +161,7 @@ async function nearestTsconfig(from, root) {
165
161
  return null;
166
162
  }
167
163
  }
168
- async function typecheck(root, sites) {
164
+ async function typecheck(sites) {
169
165
  // NOT `.genroc`: that is the project config FILE, and a directory of the same name cannot
170
166
  // coexist with it. The suffix is what keeps the scratch area out of its way.
171
167
  const dir = join(root, ".genroc-cache");
@@ -173,13 +169,13 @@ async function typecheck(root, sites) {
173
169
  // One tsc per distinct base config: `extends` takes a single base, so merging two would
174
170
  // check each script under the other author's options.
175
171
  const groups = new Map();
176
- for (const site of sites) {
177
- const base = (await nearestTsconfig(dirname(site.path), root)) ?? "";
172
+ for (const at of sites) {
173
+ const base = (await nearestTsconfig(dirname(at.file), root)) ?? "";
178
174
  const group = groups.get(base);
179
175
  if (group)
180
- group.push(site);
176
+ group.push(at);
181
177
  else
182
- groups.set(base, [site]);
178
+ groups.set(base, [at]);
183
179
  }
184
180
  let n = 0;
185
181
  for (const [base, group] of groups) {
@@ -200,7 +196,10 @@ async function typecheck(root, sites) {
200
196
  // base config there is nothing to opt in with, so the default stays none.
201
197
  ...(base ? {} : { types: [] }),
202
198
  },
203
- files: group.flatMap((s) => [relative(dir, s.path), relative(dir, typesPathFor(s.path))]),
199
+ files: group.flatMap((s) => [
200
+ relative(dir, s.file),
201
+ relative(dir, typesPathFor(s.file)),
202
+ ]),
204
203
  // `files` overrides the base's, but a base `include` survives beside it and would
205
204
  // drag the author's whole tree in, to be checked under the worker lib.
206
205
  include: [],
@@ -250,7 +249,10 @@ const transpile = {
250
249
  return { code: out.outputText, map: out.sourceMapText ?? null };
251
250
  },
252
251
  };
253
- const BUILTIN = new Set([...builtinModules, ...builtinModules.map((m) => `node:${m}`)]);
252
+ const BUILTIN = new Set([
253
+ ...builtinModules,
254
+ ...builtinModules.map((m) => `node:${m}`),
255
+ ]);
254
256
  /** Resolves imports through TYPESCRIPT, using the same config the typecheck ran under, so a
255
257
  * `paths` alias that compiles also bundles. Reimplementing `paths` here would be a second
256
258
  * resolver to keep in agreement with tsc; this one cannot disagree.
@@ -270,22 +272,25 @@ function tsResolve(configPath) {
270
272
  const { resolvedModule } = ts.resolveModuleName(source, importer, options, ts.sys);
271
273
  if (!resolvedModule || resolvedModule.isExternalLibraryImport)
272
274
  return null;
273
- return resolvedModule.resolvedFileName.endsWith(".d.ts") ? null : resolvedModule.resolvedFileName;
275
+ return resolvedModule.resolvedFileName.endsWith(".d.ts")
276
+ ? null
277
+ : resolvedModule.resolvedFileName;
274
278
  },
275
279
  };
276
280
  }
277
- /** Bundles to CJS and wraps it as an async function BODY, which is what /eval compiles.
278
- * The runtime stays unchanged: bundling is entirely the importer's job, and the string it
279
- * produces is self-contained, so a definition version pins its code forever. */
280
- async function bundle(site, root) {
281
- // Builtins are EXTERNALISED as `require` calls that worker.ts satisfies. Anything else
282
- // unresolved is a REFUSAL, not an external: rollup's default is to leave it as a require
281
+ /** Bundles to a self-contained ES module, which is what the evaluator imports: the default
282
+ * export it calls is the author's own, so nothing wraps or rewrites the code between the two.
283
+ * Bundling is entirely the importer's job, so a definition version pins its code forever. */
284
+ async function bundle(at) {
285
+ const site = at.site;
286
+ // Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
287
+ // unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
283
288
  // of a module that will not be there, which bundles clean and fails at runtime.
284
289
  const built = await rollup({
285
- input: site.path,
290
+ input: at.file,
286
291
  external: (id) => BUILTIN.has(id),
287
292
  plugins: [
288
- tsResolve(await nearestTsconfig(dirname(site.path), root)),
293
+ tsResolve(await nearestTsconfig(dirname(at.file), root)),
289
294
  nodeResolve({ extensions: [".ts", ".tsx", ".mjs", ".js", ".json"] }),
290
295
  commonjs(),
291
296
  // A `.json` import is a data file inlined at build time, which the previous bundler
@@ -295,22 +300,21 @@ async function bundle(site, root) {
295
300
  ],
296
301
  onwarn(warning) {
297
302
  if (warning.code === "UNRESOLVED_IMPORT") {
298
- die(`${site.path}: cannot resolve ${warning.exporter ?? "an import"} — is it installed?`);
303
+ die(`${at.file}: cannot resolve ${warning.exporter ?? "an import"} — is it installed?`);
299
304
  }
300
305
  },
301
- }).catch((e) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
302
- const { output } = await built.generate({ format: "cjs", exports: "auto", inlineDynamicImports: true });
306
+ }).catch((e) => die(`${at.file}: ${e instanceof Error ? e.message : String(e)}`));
307
+ const { output } = await built.generate({
308
+ format: "es",
309
+ inlineDynamicImports: true,
310
+ });
303
311
  await built.close();
304
- const cjs = output[0].code;
305
- return [
306
- "var module = { exports: {} }, exports = module.exports;",
307
- cjs,
308
- "var __genroc_main = module.exports.default ?? module.exports;",
309
- 'if (typeof __genroc_main !== "function") {',
310
- ` throw new Error(${JSON.stringify(`${site.path} has no default export function`)});`,
311
- "}",
312
- "return await __genroc_main(input);",
313
- ].join("\n");
312
+ // Refused here rather than in the realm: the evaluator can only report it against a running
313
+ // instance, and the file it names is on this machine.
314
+ if (!output[0].exports.includes("default")) {
315
+ die(`${at.file}: a script must \`export default\` the function to run`);
316
+ }
317
+ return output[0].code;
314
318
  }
315
319
  // ── main ───────────────────────────────────────────────────────────────────────
316
320
  const stdin = await new Promise((resolve, reject) => {
@@ -321,32 +325,73 @@ const stdin = await new Promise((resolve, reject) => {
321
325
  process.stdin.on("error", reject);
322
326
  });
323
327
  const manifest = JSON.parse(stdin);
324
- if (!manifest || !Array.isArray(manifest.sites))
328
+ // genctl runs a resolver with the project root as its cwd, so the manifest need not say so.
329
+ const root = process.cwd();
330
+ if (!manifest || !Array.isArray(manifest.processes))
325
331
  die("stdin is not a genroc resolver manifest");
332
+ /** A pointer as the address it is, so a generated comment or an error can be pasted into
333
+ * `genctl schema`. A key no identifier can spell is bracketed, as the grammar spells it. */
334
+ function address(pointer) {
335
+ return pointer
336
+ .map((seg) => typeof seg === "string" && /^[A-Za-z_][A-Za-z0-9_]*$/.test(seg)
337
+ ? `.${seg}`
338
+ : `[${JSON.stringify(seg)}]`)
339
+ .join("")
340
+ .replace(/^\./, "");
341
+ }
342
+ const located = manifest.processes.flatMap((where) => where.sites.map((site) => ({
343
+ site,
344
+ where,
345
+ file: resolve(where.dir, site.argument),
346
+ })));
347
+ // genctl is agnostic about what a script is for; the contract that an evaluation request carries
348
+ // its module in `code` is THIS resolver's, so it is the one that checks a directive landed there.
349
+ // `child` is the shape the scaffold uses — a call to a process that forwards to the evaluator —
350
+ // and `external` is the same request made directly.
351
+ for (const at of located) {
352
+ // Two halves, from the two places that carry them: the action's KIND is a field, and WHERE in
353
+ // it the directive sits is the pointer. A child call is the shape the scaffold generates — to a
354
+ // process that forwards to the evaluator — and an external task is the same request made
355
+ // directly.
356
+ const kind = at.site.level === "action" ? (at.site.action ?? "") : "";
357
+ const slot = at.site.pointer.slice(-2).join(".");
358
+ if ((kind !== "child" && kind !== "external") || slot !== "input.code") {
359
+ die(`${join(at.where.dir, at.where.file)}: ${address(at.site.pointer)}: an evaluated script ` +
360
+ "belongs in the `code` field of a child or external task's input, and this is " +
361
+ `${kind ? `a ${kind} task's ` : ""}\`${slot}\`.`);
362
+ }
363
+ }
364
+ // genctl no longer stats the argument — it does not know it is a file — so the resolver that
365
+ // does is the one that must say when it is not there.
366
+ for (const at of located) {
367
+ if (!existsSync(at.file)) {
368
+ die(`${join(at.where.dir, at.where.file)}: ${address(at.site.pointer)}: ` +
369
+ `"$import: ${at.site.argument}" names no file (looked at ${at.file})`);
370
+ }
371
+ }
326
372
  // One script at two sites with different input types is a refusal, not a union: the union
327
373
  // is sound and would typecheck a body that is wrong at one of the sites.
328
374
  const byPath = new Map();
329
- for (const site of manifest.sites) {
330
- const seen = byPath.get(site.path);
331
- const defsOf = (x) => (manifest.schemas[x.process]?.$defs ?? {});
375
+ for (const at of located) {
376
+ const seen = byPath.get(at.file);
332
377
  if (seen &&
333
- JSON.stringify(scriptInput(seen, defsOf(seen))) !==
334
- JSON.stringify(scriptInput(site, defsOf(site)))) {
335
- die(`${site.path} is imported at ${seen.pointer} and ${site.pointer} with different input types.\n` +
378
+ JSON.stringify(seen.site.types) !== JSON.stringify(at.site.types)) {
379
+ die(`${at.file} is imported at ${address(seen.site.pointer)} and ` +
380
+ `${address(at.site.pointer)} with different input types.\n` +
336
381
  "Split it into two scripts, or make the two call sites pass the same shape.");
337
382
  }
338
- byPath.set(site.path, site);
383
+ byPath.set(at.file, at);
339
384
  }
340
- for (const site of byPath.values()) {
341
- const defs = (manifest.schemas[site.process]?.$defs ?? {});
342
- await write(typesPathFor(site.path), declarations(site, defs));
385
+ for (const at of byPath.values()) {
386
+ await write(typesPathFor(at.file), declarations(at));
343
387
  }
344
388
  if (manifest.mode === "types") {
389
+ console.log(JSON.stringify(manifest, null, 2));
345
390
  process.exit(0);
346
391
  }
347
- await typecheck(manifest.root, [...byPath.values()]);
392
+ await typecheck([...byPath.values()]);
348
393
  const code = [];
349
- for (const site of manifest.sites) {
350
- code.push(await bundle(site, manifest.root));
394
+ for (const at of located) {
395
+ code.push(await bundle(at));
351
396
  }
352
397
  process.stdout.write(JSON.stringify({ code }));
package/dist/realm.js CHANGED
@@ -2,80 +2,46 @@
2
2
  // thread the host can kill mid-loop — the only thing that bounds a synchronous busy loop.
3
3
  // eval.ts owns the budget and does the killing; nothing here knows about time.
4
4
  //
5
- // Everything that touches the script's VALUE lives on this side of the boundary — compiling,
5
+ // Everything that touches the script's VALUE lives on this side of the boundary — loading,
6
6
  // classifying, serialising — because this is the only realm the value exists in.
7
- import { createRequire } from "node:module";
7
+ var __rewriteRelativeImportExtension = (this && this.__rewriteRelativeImportExtension) || function (path, preserveJsx) {
8
+ if (typeof path === "string" && /^\.\.?\//.test(path)) {
9
+ return path.replace(/\.(tsx)$|((?:\.d)?)((?:\.[^./]+?)?)\.([cm]?)ts$/i, function (m, tsx, d, ext, cm) {
10
+ return tsx ? preserveJsx ? ".jsx" : ".js" : d && (!ext || !cm) ? m : (d + ext + "." + cm.toLowerCase() + "js");
11
+ });
12
+ }
13
+ return path;
14
+ };
15
+ import { registerHooks } from "node:module";
8
16
  import { parentPort } from "node:worker_threads";
9
- const AsyncFunction = async function () { }.constructor;
10
- const STRICT = '"use strict";\n';
11
- // Bundled `node:*` imports survive as `require` calls — the importer externalises builtins and
12
- // inlines everything else — and a function built by the AsyncFunction constructor has no
13
- // `require` in scope. Passing one in is what makes an import of a builtin work at runtime rather than at
14
- // typecheck only. Resolution is anchored here, which is right: only builtins reach it.
15
- const scriptRequire = createRequire(import.meta.url);
17
+ // The script is IMPORTED as a module under a URL of our own, not compiled from a string: its
18
+ // frames then carry the author's own line numbers, and an `import` of a node builtin resolves
19
+ // the way it does everywhere else.
20
+ const SCRIPT_URL = "script:main";
16
21
  const STACK_BYTES = 2_048;
17
- /**
18
- * Line offset the AsyncFunction preamble adds, measured rather than assumed: the generated
19
- * wrapper's shape is engine-specific, and a hardcoded number silently misreports every
20
- * script's error location the day it changes.
21
- */
22
- const lineOffset = (async () => {
23
- // Same parameter list as a real compile: the preamble is what is being measured.
24
- const probe = new AsyncFunction("input", "require", STRICT + "throw new Error('probe');");
25
- try {
26
- await probe();
27
- return 0;
28
- }
29
- catch (err) {
30
- return reportedLine(err) - 1;
31
- }
32
- })();
33
- // V8 marks a frame compiled by the AsyncFunction constructor with the site that CALLED the
34
- // constructor, then the script's OWN position:
35
- // at inner (eval at run (file:///…/worker.ts:107:10), <anonymous>:6:9)
36
- // `eval at` is therefore what separates script frames from runner plumbing — matched without
37
- // the function name, which is whatever encloses the `new AsyncFunction` below. The LAST such
38
- // frame is the body's top level: frames interleave, since a script can throw inside a native
39
- // callback.
40
- const SCRIPT_FRAME = /\(eval at /;
41
- // The trailing `<anonymous>:LINE:COL` — the script's position, after the host file's own.
42
- const POSITION = /<anonymous>:(\d+):(\d+)\)?\s*$/;
43
- // ` at name (` — absent on the top-level frame, which V8 names `eval`.
44
- const FRAME_NAME = /^\s*at\s+(?:async\s+)?([^\s(]+)\s*\(/;
45
- /** Line number of the throw as the engine reported it, or 1 if the stack is unreadable. */
46
- function reportedLine(err) {
47
- const stack = err instanceof Error && typeof err.stack === "string" ? err.stack : "";
48
- const frame = stack.split("\n").find((l) => SCRIPT_FRAME.test(l)) ?? "";
49
- const m = frame.match(POSITION);
50
- return m ? Number(m[1]) : 1;
51
- }
52
- /** Renumbers each script frame to the line the AUTHOR wrote and drops the runner's own.
53
- * Rewriting the whole location is also what keeps the runner's path out of a script's
54
- * stack — V8 puts it inside every compiled frame. */
55
- function scriptStack(err, offset) {
22
+ // The only channel a load hook has to the source. Written per request and read once — a realm
23
+ // evaluates one script and is then discarded, so no second execution can observe it.
24
+ let source = "";
25
+ registerHooks({
26
+ resolve: (specifier, context, next) => specifier === SCRIPT_URL ? { url: SCRIPT_URL, shortCircuit: true } : next(specifier, context),
27
+ load: (url, context, next) => url === SCRIPT_URL ? { format: "module", source, shortCircuit: true } : next(url, context),
28
+ });
29
+ /** Keeps the frames that are the script's own. Everything below the last of them is runner
30
+ * plumbing the author cannot act on, and V8 puts this file's path in it. */
31
+ function scriptStack(err) {
56
32
  if (!(err instanceof Error) || typeof err.stack !== "string")
57
33
  return undefined;
58
34
  const lines = err.stack.split("\n");
59
- let boundary = -1;
35
+ let last = 0;
60
36
  for (let i = 0; i < lines.length; i++)
61
- if (SCRIPT_FRAME.test(lines[i]))
62
- boundary = i;
63
- const frames = (boundary >= 0 ? lines.slice(0, boundary + 1) : lines.slice(0, 1))
64
- .map((line) => {
65
- const pos = line.match(POSITION);
66
- if (!pos)
67
- return line; // the `Error: message` header and native frames, kept as-is
68
- const name = line.match(FRAME_NAME)?.[1];
69
- const at = name && name !== "eval" && name !== "anonymous" ? `at ${name} ` : "at ";
70
- const indent = line.match(/^\s*/)[0];
71
- return `${indent}${at}(script:${Math.max(1, Number(pos[1]) - offset)}:${pos[2]})`;
72
- })
73
- .join("\n");
74
- return frames.length > STACK_BYTES ? frames.slice(0, STACK_BYTES) : frames;
37
+ if (lines[i].includes(SCRIPT_URL))
38
+ last = i;
39
+ const stack = lines.slice(0, last + 1).join("\n");
40
+ return stack.length > STACK_BYTES ? stack.slice(0, STACK_BYTES) : stack;
75
41
  }
76
- function describe(err, kind, offset) {
42
+ function describe(err, kind) {
77
43
  if (err instanceof Error) {
78
- return { kind, name: err.name, message: err.message, stack: scriptStack(err, offset) };
44
+ return { kind, name: err.name, message: err.message, stack: scriptStack(err) };
79
45
  }
80
46
  // A script may throw a non-Error (`throw {code: "x"}`), so name/message must not assume one.
81
47
  return { kind, name: "Thrown", message: safeText(err) };
@@ -88,23 +54,39 @@ function safeText(v) {
88
54
  return String(v);
89
55
  }
90
56
  }
57
+ /** A module that will not parse, or names an import nothing resolves, is broken code — only
58
+ * editing it helps. Anything else thrown by the import is the module's top level running,
59
+ * which is the script throwing. */
60
+ function unloadable(err) {
61
+ const code = err?.code;
62
+ return err instanceof SyntaxError || (typeof code === "string" && code.startsWith("ERR_MODULE"));
63
+ }
91
64
  async function run(req) {
92
- const offset = await lineOffset;
93
- let fn;
65
+ source = req.code;
66
+ let main;
94
67
  try {
95
- // No compile cache: the realm is discarded after this execution, so a cache in it could
96
- // never be hit. Repeated compilation is the price of the fresh global object.
97
- fn = new AsyncFunction("input", "require", STRICT + req.code);
68
+ main = (await import(__rewriteRelativeImportExtension(SCRIPT_URL))).default;
98
69
  }
99
70
  catch (err) {
100
- return { ok: false, failure: describe(err, "compile_error", offset) };
71
+ return { ok: false, failure: describe(err, unloadable(err) ? "compile_error" : "threw") };
72
+ }
73
+ if (typeof main !== "function") {
74
+ const got = main === undefined ? "no default export" : `a ${typeof main}`;
75
+ return {
76
+ ok: false,
77
+ failure: {
78
+ kind: "compile_error",
79
+ name: "NoDefaultExport",
80
+ message: `a script must export default a function; this one has ${got}`,
81
+ },
82
+ };
101
83
  }
102
84
  let value;
103
85
  try {
104
- value = await fn(req.input, scriptRequire);
86
+ value = await main(req.input);
105
87
  }
106
88
  catch (err) {
107
- return { ok: false, failure: describe(err, "threw", offset) };
89
+ return { ok: false, failure: describe(err, "threw") };
108
90
  }
109
91
  try {
110
92
  // undefined stringifies to undefined, not "undefined"; an empty body is how genroc
@@ -112,11 +94,20 @@ async function run(req) {
112
94
  return { ok: true, body: value === undefined ? "" : JSON.stringify(value) ?? "" };
113
95
  }
114
96
  catch (err) {
115
- return { ok: false, failure: describe(err, "nonserializable", offset) };
97
+ return { ok: false, failure: describe(err, "nonserializable") };
116
98
  }
117
99
  }
100
+ // The realm's stdio is a pipe to the host thread, and eval.ts terminates this thread the moment
101
+ // the reply lands — so whatever a script wrote last is still in the pipe when it dies. An empty
102
+ // write's callback fires once the queue ahead of it has drained, which makes the reply a barrier
103
+ // for `console` and a direct process.stdout.write alike: both are this stream.
104
+ function flush(stream) {
105
+ return new Promise((resolve) => stream.write("", () => resolve()));
106
+ }
118
107
  // Non-null because this module only ever runs as a worker entry point; a null port here
119
108
  // would mean eval.ts loaded it as a plain module, which nothing does.
120
109
  parentPort.on("message", async (req) => {
121
- parentPort.postMessage(await run(req));
110
+ const reply = await run(req);
111
+ await Promise.all([flush(process.stdout), flush(process.stderr)]);
112
+ parentPort.postMessage(reply);
122
113
  });
package/eval.ts CHANGED
@@ -1,6 +1,7 @@
1
- // Evaluation core: run a code string in its OWN realm and classify every outcome into one of
2
- // the failure kinds in README.md. Nothing here knows about genroc — worker.ts is the only
3
- // thing that talks to the queue — so this stays testable and the containment stays swappable.
1
+ // Evaluation core: import a script module in its OWN realm, call its default export, and
2
+ // classify every outcome into one of the failure kinds in README.md. Nothing here knows about
3
+ // genroc — worker.ts is the only thing that talks to the queue — so this stays testable and
4
+ // the containment stays swappable.
4
5
  //
5
6
  // The containment is a Worker per execution (realm.ts). It is what makes the budget real:
6
7
  // a synchronous busy loop never yields, so no in-process timer can interrupt it, and only a
@@ -9,6 +10,7 @@
9
10
  import { Worker } from "node:worker_threads";
10
11
 
11
12
  export type EvalRequest = {
13
+ /** An ES module whose default export is the function to run; `input` is its argument. */
12
14
  code: string;
13
15
  input?: unknown;
14
16
  timeout_ms?: number;
package/import.ts CHANGED
@@ -8,9 +8,10 @@
8
8
  // over the same project.
9
9
 
10
10
  import { spawn } from "node:child_process";
11
+ import { existsSync } from "node:fs";
11
12
  import { access, mkdir, writeFile } from "node:fs/promises";
12
13
  import { builtinModules } from "node:module";
13
- import { dirname, join, relative } from "node:path";
14
+ import { dirname, join, relative, resolve } from "node:path";
14
15
  import { fileURLToPath } from "node:url";
15
16
 
16
17
  import commonjs from "@rollup/plugin-commonjs";
@@ -22,20 +23,37 @@ import ts from "typescript";
22
23
  type Schema = Record<string, any>;
23
24
 
24
25
  type Site = {
25
- resolver: string;
26
- process: string;
26
+ // Which namespace the directive sits in: `process`, `task` or `action`.
27
+ level: "process" | "task" | "action";
27
28
  task?: string;
28
- pointer: string;
29
- path: string;
30
- input?: Schema;
31
- output?: Schema;
29
+ // What the site IS, at action level: the action's type, and the process a child call names.
30
+ action?: string;
31
+ child?: string;
32
+ // The slot as a path of keys and indices, so nothing has to unescape a JSON Pointer.
33
+ pointer: (string | number)[];
34
+ // Everything after `$<resolver>:`, verbatim. genctl does not read it as a path — that is
35
+ // this resolver's reading of it.
36
+ argument: string;
37
+ // What .genroc asked genctl to type, keyed by the name it chose. This resolver reads Input
38
+ // and Output; which ADDRESS each came from is the config's business, not ours.
39
+ types?: Record<string, Schema>;
40
+ };
41
+
42
+ type ManifestProcess = {
43
+ name: string;
44
+ // The definition's own location, split because a relative argument is relative to the
45
+ // DIRECTORY — joining is ours to do, and this is the base.
46
+ dir: string;
47
+ file: string;
48
+ sites: Site[];
49
+ // Only what this process's fragments reach — a `$ref` in `types` points here.
50
+ $defs?: Record<string, Schema>;
32
51
  };
33
52
 
34
53
  type Manifest = {
35
54
  mode: "types" | "build";
36
55
  root: string;
37
- schemas: Record<string, Schema>;
38
- sites: Site[];
56
+ processes: ManifestProcess[];
39
57
  };
40
58
 
41
59
  function die(message: string): never {
@@ -72,13 +90,16 @@ function tsType(s: Schema | undefined, used: Set<string>): string {
72
90
  if (Array.isArray(s.enum)) {
73
91
  return s.enum.map((v: unknown) => JSON.stringify(v)).join(" | ") || "never";
74
92
  }
75
- if (Array.isArray(s.anyOf)) return union(s.anyOf.map((a: Schema) => tsType(a, used)));
76
- if (Array.isArray(s.oneOf)) return union(s.oneOf.map((a: Schema) => tsType(a, used)));
93
+ if (Array.isArray(s.anyOf))
94
+ return union(s.anyOf.map((a: Schema) => tsType(a, used)));
95
+ if (Array.isArray(s.oneOf))
96
+ return union(s.oneOf.map((a: Schema) => tsType(a, used)));
77
97
  if (Array.isArray(s.allOf)) {
78
98
  return s.allOf.map((a: Schema) => tsType(a, used)).join(" & ") || "unknown";
79
99
  }
80
100
 
81
- const types: string[] = s.type === undefined ? [] : Array.isArray(s.type) ? s.type : [s.type];
101
+ const types: string[] =
102
+ s.type === undefined ? [] : Array.isArray(s.type) ? s.type : [s.type];
82
103
  if (types.length === 0) {
83
104
  // The top type: `{}` means unknown, not "an empty object". specs/unknown-type.md.
84
105
  return s.properties ? objectType(s, used) : "unknown";
@@ -111,8 +132,13 @@ function objectType(s: Schema, used: Set<string>): string {
111
132
  const required = new Set<string>(s.required ?? []);
112
133
  const lines: string[] = [];
113
134
  for (const [key, sub] of Object.entries(props)) {
114
- const doc = typeof sub.description === "string" ? ` /** ${sub.description} */\n` : "";
115
- lines.push(`${doc} ${propKey(key)}${required.has(key) ? "" : "?"}: ${tsType(sub, used)};`);
135
+ const doc =
136
+ typeof sub.description === "string"
137
+ ? ` /** ${sub.description} */\n`
138
+ : "";
139
+ lines.push(
140
+ `${doc} ${propKey(key)}${required.has(key) ? "" : "?"}: ${tsType(sub, used)};`,
141
+ );
116
142
  }
117
143
  if (s.additionalProperties && typeof s.additionalProperties === "object") {
118
144
  lines.push(` [key: string]: ${tsType(s.additionalProperties, used)};`);
@@ -128,9 +154,13 @@ function union(parts: string[]): string {
128
154
 
129
155
  const IDENT = /^[A-Za-z_$][A-Za-z0-9_$]*$/;
130
156
  const propKey = (k: string) => (IDENT.test(k) ? k : JSON.stringify(k));
131
- const identifier = (n: string) => (IDENT.test(n) ? n : `Def_${n.replace(/[^A-Za-z0-9_$]/g, "_")}`);
157
+ const identifier = (n: string) =>
158
+ IDENT.test(n) ? n : `Def_${n.replace(/[^A-Za-z0-9_$]/g, "_")}`;
132
159
 
133
- function deref(s: Schema | undefined, defs: Record<string, Schema>): Schema | undefined {
160
+ function deref(
161
+ s: Schema | undefined,
162
+ defs: Record<string, Schema>,
163
+ ): Schema | undefined {
134
164
  let cur = s;
135
165
  for (let i = 0; cur && typeof cur.$ref === "string" && i < 16; i++) {
136
166
  cur = defs[cur.$ref.replace(/^#\/\$defs\//, "")];
@@ -138,22 +168,14 @@ function deref(s: Schema | undefined, defs: Record<string, Schema>): Schema | un
138
168
  return cur;
139
169
  }
140
170
 
141
- /** The manifest's `input` is the type of the whole ACTION input — for /eval that is
142
- * `{code, input, timeout_ms, …}`, and only its `input` field is bound as the script's
143
- * parameter. genroc cannot know that; this file owns the evaluator's wire contract, so
144
- * the navigation belongs here rather than in genctl. */
145
- function scriptInput(site: Site, defs: Record<string, Schema>): Schema | undefined {
146
- const action = deref(site.input, defs);
147
- const bound = action?.properties?.input;
148
- return bound ?? site.input;
149
- }
150
-
151
171
  /** Emits one named type per reachable $def rather than inlining: a task output may
152
172
  * reference itself (specs/recursive-type-inference.md) and inlining would not terminate. */
153
- function declarations(site: Site, defs: Record<string, Schema>): string {
173
+ function declarations(at: Located): string {
174
+ const { site, where } = at;
175
+ const defs = where.$defs ?? {};
154
176
  const used = new Set<string>();
155
- const input = tsType(scriptInput(site, defs), used);
156
- const output = tsType(site.output, used);
177
+ const input = tsType(site.types?.Input, used);
178
+ const output = tsType(site.types?.Output, used);
157
179
 
158
180
  const emitted: string[] = [];
159
181
  const done = new Set<string>();
@@ -167,7 +189,7 @@ function declarations(site: Site, defs: Record<string, Schema>): string {
167
189
 
168
190
  return [
169
191
  "// Generated by genroc. Do not edit - regenerate with `genctl types`.",
170
- `// ${site.process}${site.task ? ` / ${site.task}` : ""} (${site.pointer})`,
192
+ `// ${where.name} (${address(site.pointer)})`,
171
193
  "",
172
194
  ...emitted,
173
195
  emitted.length ? "" : "",
@@ -189,7 +211,10 @@ function typesPathFor(scriptPath: string): string {
189
211
  /** The nearest tsconfig above the script — the one the author's editor already reads. Two
190
212
  * different configs mean a red editor over a clean apply, or the reverse. The walk stops at
191
213
  * the project root: above it is not this project. */
192
- async function nearestTsconfig(from: string, root: string): Promise<string | null> {
214
+ async function nearestTsconfig(
215
+ from: string,
216
+ root: string,
217
+ ): Promise<string | null> {
193
218
  for (let dir = from; ; dir = dirname(dir)) {
194
219
  const candidate = join(dir, "tsconfig.json");
195
220
  if (await exists(candidate)) return candidate;
@@ -197,7 +222,7 @@ async function nearestTsconfig(from: string, root: string): Promise<string | nul
197
222
  }
198
223
  }
199
224
 
200
- async function typecheck(root: string, sites: Site[]): Promise<void> {
225
+ async function typecheck(sites: Located[]): Promise<void> {
201
226
  // NOT `.genroc`: that is the project config FILE, and a directory of the same name cannot
202
227
  // coexist with it. The suffix is what keeps the scratch area out of its way.
203
228
  const dir = join(root, ".genroc-cache");
@@ -205,12 +230,12 @@ async function typecheck(root: string, sites: Site[]): Promise<void> {
205
230
 
206
231
  // One tsc per distinct base config: `extends` takes a single base, so merging two would
207
232
  // check each script under the other author's options.
208
- const groups = new Map<string, Site[]>();
209
- for (const site of sites) {
210
- const base = (await nearestTsconfig(dirname(site.path), root)) ?? "";
233
+ const groups = new Map<string, Located[]>();
234
+ for (const at of sites) {
235
+ const base = (await nearestTsconfig(dirname(at.file), root)) ?? "";
211
236
  const group = groups.get(base);
212
- if (group) group.push(site);
213
- else groups.set(base, [site]);
237
+ if (group) group.push(at);
238
+ else groups.set(base, [at]);
214
239
  }
215
240
 
216
241
  let n = 0;
@@ -232,12 +257,18 @@ async function typecheck(root: string, sites: Site[]): Promise<void> {
232
257
  // base config there is nothing to opt in with, so the default stays none.
233
258
  ...(base ? {} : { types: [] }),
234
259
  },
235
- files: group.flatMap((s) => [relative(dir, s.path), relative(dir, typesPathFor(s.path))]),
260
+ files: group.flatMap((s) => [
261
+ relative(dir, s.file),
262
+ relative(dir, typesPathFor(s.file)),
263
+ ]),
236
264
  // `files` overrides the base's, but a base `include` survives beside it and would
237
265
  // drag the author's whole tree in, to be checked under the worker lib.
238
266
  include: [],
239
267
  };
240
- const configPath = join(dir, groups.size === 1 ? "tsconfig.json" : `tsconfig.${n++}.json`);
268
+ const configPath = join(
269
+ dir,
270
+ groups.size === 1 ? "tsconfig.json" : `tsconfig.${n++}.json`,
271
+ );
241
272
  await write(configPath, JSON.stringify(config, null, 2));
242
273
  await runTsc(root, configPath);
243
274
  }
@@ -285,7 +316,10 @@ const transpile: Plugin = {
285
316
  },
286
317
  };
287
318
 
288
- const BUILTIN = new Set([...builtinModules, ...builtinModules.map((m) => `node:${m}`)]);
319
+ const BUILTIN = new Set([
320
+ ...builtinModules,
321
+ ...builtinModules.map((m) => `node:${m}`),
322
+ ]);
289
323
 
290
324
  /** Resolves imports through TYPESCRIPT, using the same config the typecheck ran under, so a
291
325
  * `paths` alias that compiles also bundles. Reimplementing `paths` here would be a second
@@ -296,31 +330,44 @@ function tsResolve(configPath: string | null): Plugin {
296
330
  let options: ts.CompilerOptions = {};
297
331
  if (configPath) {
298
332
  const read = ts.readConfigFile(configPath, ts.sys.readFile);
299
- options = ts.parseJsonConfigFileContent(read.config ?? {}, ts.sys, dirname(configPath)).options;
333
+ options = ts.parseJsonConfigFileContent(
334
+ read.config ?? {},
335
+ ts.sys,
336
+ dirname(configPath),
337
+ ).options;
300
338
  }
301
339
  return {
302
340
  name: "genroc-ts-resolve",
303
341
  resolveId(source, importer) {
304
342
  if (!importer || BUILTIN.has(source)) return null;
305
- const { resolvedModule } = ts.resolveModuleName(source, importer, options, ts.sys);
306
- if (!resolvedModule || resolvedModule.isExternalLibraryImport) return null;
307
- return resolvedModule.resolvedFileName.endsWith(".d.ts") ? null : resolvedModule.resolvedFileName;
343
+ const { resolvedModule } = ts.resolveModuleName(
344
+ source,
345
+ importer,
346
+ options,
347
+ ts.sys,
348
+ );
349
+ if (!resolvedModule || resolvedModule.isExternalLibraryImport)
350
+ return null;
351
+ return resolvedModule.resolvedFileName.endsWith(".d.ts")
352
+ ? null
353
+ : resolvedModule.resolvedFileName;
308
354
  },
309
355
  };
310
356
  }
311
357
 
312
- /** Bundles to CJS and wraps it as an async function BODY, which is what /eval compiles.
313
- * The runtime stays unchanged: bundling is entirely the importer's job, and the string it
314
- * produces is self-contained, so a definition version pins its code forever. */
315
- async function bundle(site: Site, root: string): Promise<string> {
316
- // Builtins are EXTERNALISED as `require` calls that worker.ts satisfies. Anything else
317
- // unresolved is a REFUSAL, not an external: rollup's default is to leave it as a require
358
+ /** Bundles to a self-contained ES module, which is what the evaluator imports: the default
359
+ * export it calls is the author's own, so nothing wraps or rewrites the code between the two.
360
+ * Bundling is entirely the importer's job, so a definition version pins its code forever. */
361
+ async function bundle(at: Located): Promise<string> {
362
+ const site = at.site;
363
+ // Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
364
+ // unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
318
365
  // of a module that will not be there, which bundles clean and fails at runtime.
319
366
  const built = await rollup({
320
- input: site.path,
367
+ input: at.file,
321
368
  external: (id) => BUILTIN.has(id),
322
369
  plugins: [
323
- tsResolve(await nearestTsconfig(dirname(site.path), root)),
370
+ tsResolve(await nearestTsconfig(dirname(at.file), root)),
324
371
  nodeResolve({ extensions: [".ts", ".tsx", ".mjs", ".js", ".json"] }),
325
372
  commonjs(),
326
373
  // A `.json` import is a data file inlined at build time, which the previous bundler
@@ -330,23 +377,26 @@ async function bundle(site: Site, root: string): Promise<string> {
330
377
  ],
331
378
  onwarn(warning) {
332
379
  if (warning.code === "UNRESOLVED_IMPORT") {
333
- die(`${site.path}: cannot resolve ${warning.exporter ?? "an import"} — is it installed?`);
380
+ die(
381
+ `${at.file}: cannot resolve ${warning.exporter ?? "an import"} — is it installed?`,
382
+ );
334
383
  }
335
384
  },
336
- }).catch((e: unknown) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
385
+ }).catch((e: unknown) =>
386
+ die(`${at.file}: ${e instanceof Error ? e.message : String(e)}`),
387
+ );
337
388
 
338
- const { output } = await built.generate({ format: "cjs", exports: "auto", inlineDynamicImports: true });
389
+ const { output } = await built.generate({
390
+ format: "es",
391
+ inlineDynamicImports: true,
392
+ });
339
393
  await built.close();
340
- const cjs = output[0].code;
341
- return [
342
- "var module = { exports: {} }, exports = module.exports;",
343
- cjs,
344
- "var __genroc_main = module.exports.default ?? module.exports;",
345
- 'if (typeof __genroc_main !== "function") {',
346
- ` throw new Error(${JSON.stringify(`${site.path} has no default export function`)});`,
347
- "}",
348
- "return await __genroc_main(input);",
349
- ].join("\n");
394
+ // Refused here rather than in the realm: the evaluator can only report it against a running
395
+ // instance, and the file it names is on this machine.
396
+ if (!output[0].exports.includes("default")) {
397
+ die(`${at.file}: a script must \`export default\` the function to run`);
398
+ }
399
+ return output[0].code;
350
400
  }
351
401
 
352
402
  // ── main ───────────────────────────────────────────────────────────────────────
@@ -359,40 +409,99 @@ const stdin: string = await new Promise((resolve, reject) => {
359
409
  process.stdin.on("error", reject);
360
410
  });
361
411
  const manifest = JSON.parse(stdin) as Manifest;
362
- if (!manifest || !Array.isArray(manifest.sites)) die("stdin is not a genroc resolver manifest");
412
+ // genctl runs a resolver with the project root as its cwd, so the manifest need not say so.
413
+ const root = process.cwd();
414
+ if (!manifest || !Array.isArray(manifest.processes))
415
+ die("stdin is not a genroc resolver manifest");
416
+
417
+ /** Every site with the process it sits in and the file its argument names, in the manifest's own
418
+ * order — which is the order `code` must answer in. genctl passes the argument verbatim, so
419
+ * joining it to the definition's directory is this resolver's business: only it knows the
420
+ * argument is a path at all. */
421
+ type Located = { site: Site; where: ManifestProcess; file: string };
422
+
423
+ /** A pointer as the address it is, so a generated comment or an error can be pasted into
424
+ * `genctl schema`. A key no identifier can spell is bracketed, as the grammar spells it. */
425
+ function address(pointer: (string | number)[]): string {
426
+ return pointer
427
+ .map((seg) =>
428
+ typeof seg === "string" && /^[A-Za-z_][A-Za-z0-9_]*$/.test(seg)
429
+ ? `.${seg}`
430
+ : `[${JSON.stringify(seg)}]`,
431
+ )
432
+ .join("")
433
+ .replace(/^\./, "");
434
+ }
435
+ const located: Located[] = manifest.processes.flatMap((where) =>
436
+ where.sites.map((site) => ({
437
+ site,
438
+ where,
439
+ file: resolve(where.dir, site.argument),
440
+ })),
441
+ );
442
+
443
+ // genctl is agnostic about what a script is for; the contract that an evaluation request carries
444
+ // its module in `code` is THIS resolver's, so it is the one that checks a directive landed there.
445
+ // `child` is the shape the scaffold uses — a call to a process that forwards to the evaluator —
446
+ // and `external` is the same request made directly.
447
+ for (const at of located) {
448
+ // Two halves, from the two places that carry them: the action's KIND is a field, and WHERE in
449
+ // it the directive sits is the pointer. A child call is the shape the scaffold generates — to a
450
+ // process that forwards to the evaluator — and an external task is the same request made
451
+ // directly.
452
+ const kind = at.site.level === "action" ? (at.site.action ?? "") : "";
453
+ const slot = at.site.pointer.slice(-2).join(".");
454
+ if ((kind !== "child" && kind !== "external") || slot !== "input.code") {
455
+ die(
456
+ `${join(at.where.dir, at.where.file)}: ${address(at.site.pointer)}: an evaluated script ` +
457
+ "belongs in the `code` field of a child or external task's input, and this is " +
458
+ `${kind ? `a ${kind} task's ` : ""}\`${slot}\`.`,
459
+ );
460
+ }
461
+ }
462
+
463
+ // genctl no longer stats the argument — it does not know it is a file — so the resolver that
464
+ // does is the one that must say when it is not there.
465
+ for (const at of located) {
466
+ if (!existsSync(at.file)) {
467
+ die(
468
+ `${join(at.where.dir, at.where.file)}: ${address(at.site.pointer)}: ` +
469
+ `"$import: ${at.site.argument}" names no file (looked at ${at.file})`,
470
+ );
471
+ }
472
+ }
363
473
 
364
474
  // One script at two sites with different input types is a refusal, not a union: the union
365
475
  // is sound and would typecheck a body that is wrong at one of the sites.
366
- const byPath = new Map<string, Site>();
367
- for (const site of manifest.sites) {
368
- const seen = byPath.get(site.path);
369
- const defsOf = (x: Site) => (manifest.schemas[x.process]?.$defs ?? {}) as Record<string, Schema>;
476
+ const byPath = new Map<string, Located>();
477
+ for (const at of located) {
478
+ const seen = byPath.get(at.file);
370
479
  if (
371
480
  seen &&
372
- JSON.stringify(scriptInput(seen, defsOf(seen))) !==
373
- JSON.stringify(scriptInput(site, defsOf(site)))
481
+ JSON.stringify(seen.site.types) !== JSON.stringify(at.site.types)
374
482
  ) {
375
483
  die(
376
- `${site.path} is imported at ${seen.pointer} and ${site.pointer} with different input types.\n` +
484
+ `${at.file} is imported at ${address(seen.site.pointer)} and ` +
485
+ `${address(at.site.pointer)} with different input types.\n` +
377
486
  "Split it into two scripts, or make the two call sites pass the same shape.",
378
487
  );
379
488
  }
380
- byPath.set(site.path, site);
489
+ byPath.set(at.file, at);
381
490
  }
382
491
 
383
- for (const site of byPath.values()) {
384
- const defs = (manifest.schemas[site.process]?.$defs ?? {}) as Record<string, Schema>;
385
- await write(typesPathFor(site.path), declarations(site, defs));
492
+ for (const at of byPath.values()) {
493
+ await write(typesPathFor(at.file), declarations(at));
386
494
  }
387
495
 
388
496
  if (manifest.mode === "types") {
497
+ console.log(JSON.stringify(manifest, null, 2));
389
498
  process.exit(0);
390
499
  }
391
500
 
392
- await typecheck(manifest.root, [...byPath.values()]);
501
+ await typecheck([...byPath.values()]);
393
502
 
394
503
  const code: string[] = [];
395
- for (const site of manifest.sites) {
396
- code.push(await bundle(site, manifest.root));
504
+ for (const at of located) {
505
+ code.push(await bundle(at));
397
506
  }
398
507
  process.stdout.write(JSON.stringify({ code }));
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@genroc/eval-node",
3
- "version": "0.0.0-edge.d773aa8",
3
+ "version": "0.0.0-edge.dadd892",
4
4
  "type": "module",
5
5
  "engines": {
6
6
  "node": ">=24"
package/realm.ts CHANGED
@@ -2,88 +2,45 @@
2
2
  // thread the host can kill mid-loop — the only thing that bounds a synchronous busy loop.
3
3
  // eval.ts owns the budget and does the killing; nothing here knows about time.
4
4
  //
5
- // Everything that touches the script's VALUE lives on this side of the boundary — compiling,
5
+ // Everything that touches the script's VALUE lives on this side of the boundary — loading,
6
6
  // classifying, serialising — because this is the only realm the value exists in.
7
7
 
8
- import { createRequire } from "node:module";
8
+ import { registerHooks } from "node:module";
9
9
  import { parentPort } from "node:worker_threads";
10
10
 
11
11
  import type { EvalFailure, FailureKind, WorkerReply, WorkerRequest } from "./eval.ts";
12
12
 
13
- const AsyncFunction = async function () {}.constructor as new (
14
- ...args: string[]
15
- ) => (...args: unknown[]) => Promise<unknown>;
16
-
17
- const STRICT = '"use strict";\n';
18
-
19
- // Bundled `node:*` imports survive as `require` calls — the importer externalises builtins and
20
- // inlines everything else — and a function built by the AsyncFunction constructor has no
21
- // `require` in scope. Passing one in is what makes an import of a builtin work at runtime rather than at
22
- // typecheck only. Resolution is anchored here, which is right: only builtins reach it.
23
- const scriptRequire = createRequire(import.meta.url);
13
+ // The script is IMPORTED as a module under a URL of our own, not compiled from a string: its
14
+ // frames then carry the author's own line numbers, and an `import` of a node builtin resolves
15
+ // the way it does everywhere else.
16
+ const SCRIPT_URL = "script:main";
24
17
  const STACK_BYTES = 2_048;
25
18
 
26
- /**
27
- * Line offset the AsyncFunction preamble adds, measured rather than assumed: the generated
28
- * wrapper's shape is engine-specific, and a hardcoded number silently misreports every
29
- * script's error location the day it changes.
30
- */
31
- const lineOffset: Promise<number> = (async () => {
32
- // Same parameter list as a real compile: the preamble is what is being measured.
33
- const probe = new AsyncFunction("input", "require", STRICT + "throw new Error('probe');");
34
- try {
35
- await probe();
36
- return 0;
37
- } catch (err) {
38
- return reportedLine(err) - 1;
39
- }
40
- })();
41
-
42
- // V8 marks a frame compiled by the AsyncFunction constructor with the site that CALLED the
43
- // constructor, then the script's OWN position:
44
- // at inner (eval at run (file:///…/worker.ts:107:10), <anonymous>:6:9)
45
- // `eval at` is therefore what separates script frames from runner plumbing — matched without
46
- // the function name, which is whatever encloses the `new AsyncFunction` below. The LAST such
47
- // frame is the body's top level: frames interleave, since a script can throw inside a native
48
- // callback.
49
- const SCRIPT_FRAME = /\(eval at /;
50
- // The trailing `<anonymous>:LINE:COL` — the script's position, after the host file's own.
51
- const POSITION = /<anonymous>:(\d+):(\d+)\)?\s*$/;
52
- // ` at name (` — absent on the top-level frame, which V8 names `eval`.
53
- const FRAME_NAME = /^\s*at\s+(?:async\s+)?([^\s(]+)\s*\(/;
19
+ // The only channel a load hook has to the source. Written per request and read once — a realm
20
+ // evaluates one script and is then discarded, so no second execution can observe it.
21
+ let source = "";
54
22
 
55
- /** Line number of the throw as the engine reported it, or 1 if the stack is unreadable. */
56
- function reportedLine(err: unknown): number {
57
- const stack = err instanceof Error && typeof err.stack === "string" ? err.stack : "";
58
- const frame = stack.split("\n").find((l) => SCRIPT_FRAME.test(l)) ?? "";
59
- const m = frame.match(POSITION);
60
- return m ? Number(m[1]) : 1;
61
- }
23
+ registerHooks({
24
+ resolve: (specifier, context, next) =>
25
+ specifier === SCRIPT_URL ? { url: SCRIPT_URL, shortCircuit: true } : next(specifier, context),
26
+ load: (url, context, next) =>
27
+ url === SCRIPT_URL ? { format: "module", source, shortCircuit: true } : next(url, context),
28
+ });
62
29
 
63
- /** Renumbers each script frame to the line the AUTHOR wrote and drops the runner's own.
64
- * Rewriting the whole location is also what keeps the runner's path out of a script's
65
- * stack V8 puts it inside every compiled frame. */
66
- function scriptStack(err: unknown, offset: number): string | undefined {
30
+ /** Keeps the frames that are the script's own. Everything below the last of them is runner
31
+ * plumbing the author cannot act on, and V8 puts this file's path in it. */
32
+ function scriptStack(err: unknown): string | undefined {
67
33
  if (!(err instanceof Error) || typeof err.stack !== "string") return undefined;
68
34
  const lines = err.stack.split("\n");
69
- let boundary = -1;
70
- for (let i = 0; i < lines.length; i++) if (SCRIPT_FRAME.test(lines[i]!)) boundary = i;
71
- const frames = (boundary >= 0 ? lines.slice(0, boundary + 1) : lines.slice(0, 1))
72
- .map((line) => {
73
- const pos = line.match(POSITION);
74
- if (!pos) return line; // the `Error: message` header and native frames, kept as-is
75
- const name = line.match(FRAME_NAME)?.[1];
76
- const at = name && name !== "eval" && name !== "anonymous" ? `at ${name} ` : "at ";
77
- const indent = line.match(/^\s*/)![0];
78
- return `${indent}${at}(script:${Math.max(1, Number(pos[1]) - offset)}:${pos[2]})`;
79
- })
80
- .join("\n");
81
- return frames.length > STACK_BYTES ? frames.slice(0, STACK_BYTES) : frames;
35
+ let last = 0;
36
+ for (let i = 0; i < lines.length; i++) if (lines[i]!.includes(SCRIPT_URL)) last = i;
37
+ const stack = lines.slice(0, last + 1).join("\n");
38
+ return stack.length > STACK_BYTES ? stack.slice(0, STACK_BYTES) : stack;
82
39
  }
83
40
 
84
- function describe(err: unknown, kind: FailureKind, offset: number): EvalFailure {
41
+ function describe(err: unknown, kind: FailureKind): EvalFailure {
85
42
  if (err instanceof Error) {
86
- return { kind, name: err.name, message: err.message, stack: scriptStack(err, offset) };
43
+ return { kind, name: err.name, message: err.message, stack: scriptStack(err) };
87
44
  }
88
45
  // A script may throw a non-Error (`throw {code: "x"}`), so name/message must not assume one.
89
46
  return { kind, name: "Thrown", message: safeText(err) };
@@ -97,23 +54,40 @@ function safeText(v: unknown): string {
97
54
  }
98
55
  }
99
56
 
57
+ /** A module that will not parse, or names an import nothing resolves, is broken code — only
58
+ * editing it helps. Anything else thrown by the import is the module's top level running,
59
+ * which is the script throwing. */
60
+ function unloadable(err: unknown): boolean {
61
+ const code = (err as { code?: unknown } | null)?.code;
62
+ return err instanceof SyntaxError || (typeof code === "string" && code.startsWith("ERR_MODULE"));
63
+ }
64
+
100
65
  async function run(req: WorkerRequest): Promise<WorkerReply> {
101
- const offset = await lineOffset;
66
+ source = req.code;
102
67
 
103
- let fn: (...args: unknown[]) => Promise<unknown>;
68
+ let main: unknown;
104
69
  try {
105
- // No compile cache: the realm is discarded after this execution, so a cache in it could
106
- // never be hit. Repeated compilation is the price of the fresh global object.
107
- fn = new AsyncFunction("input", "require", STRICT + req.code);
70
+ main = (await import(SCRIPT_URL)).default;
108
71
  } catch (err) {
109
- return { ok: false, failure: describe(err, "compile_error", offset) };
72
+ return { ok: false, failure: describe(err, unloadable(err) ? "compile_error" : "threw") };
73
+ }
74
+ if (typeof main !== "function") {
75
+ const got = main === undefined ? "no default export" : `a ${typeof main}`;
76
+ return {
77
+ ok: false,
78
+ failure: {
79
+ kind: "compile_error",
80
+ name: "NoDefaultExport",
81
+ message: `a script must export default a function; this one has ${got}`,
82
+ },
83
+ };
110
84
  }
111
85
 
112
86
  let value: unknown;
113
87
  try {
114
- value = await fn(req.input, scriptRequire);
88
+ value = await (main as (input: unknown) => unknown)(req.input);
115
89
  } catch (err) {
116
- return { ok: false, failure: describe(err, "threw", offset) };
90
+ return { ok: false, failure: describe(err, "threw") };
117
91
  }
118
92
 
119
93
  try {
@@ -121,12 +95,22 @@ async function run(req: WorkerRequest): Promise<WorkerReply> {
121
95
  // spells null, which is the right reading of a script that returned nothing.
122
96
  return { ok: true, body: value === undefined ? "" : JSON.stringify(value) ?? "" };
123
97
  } catch (err) {
124
- return { ok: false, failure: describe(err, "nonserializable", offset) };
98
+ return { ok: false, failure: describe(err, "nonserializable") };
125
99
  }
126
100
  }
127
101
 
102
+ // The realm's stdio is a pipe to the host thread, and eval.ts terminates this thread the moment
103
+ // the reply lands — so whatever a script wrote last is still in the pipe when it dies. An empty
104
+ // write's callback fires once the queue ahead of it has drained, which makes the reply a barrier
105
+ // for `console` and a direct process.stdout.write alike: both are this stream.
106
+ function flush(stream: NodeJS.WriteStream): Promise<void> {
107
+ return new Promise((resolve) => stream.write("", () => resolve()));
108
+ }
109
+
128
110
  // Non-null because this module only ever runs as a worker entry point; a null port here
129
111
  // would mean eval.ts loaded it as a plain module, which nothing does.
130
112
  parentPort!.on("message", async (req: WorkerRequest) => {
131
- parentPort!.postMessage(await run(req));
113
+ const reply = await run(req);
114
+ await Promise.all([flush(process.stdout), flush(process.stderr)]);
115
+ parentPort!.postMessage(reply);
132
116
  });