@genroc/eval-node 0.0.0-edge.d773aa8 → 0.0.0-edge.ee19083

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:
@@ -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
@@ -274,12 +274,12 @@ function tsResolve(configPath) {
274
274
  },
275
275
  };
276
276
  }
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. */
277
+ /** Bundles to a self-contained ES module, which is what the evaluator imports: the default
278
+ * export it calls is the author's own, so nothing wraps or rewrites the code between the two.
279
+ * Bundling is entirely the importer's job, so a definition version pins its code forever. */
280
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
+ // Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
282
+ // unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
283
283
  // of a module that will not be there, which bundles clean and fails at runtime.
284
284
  const built = await rollup({
285
285
  input: site.path,
@@ -299,18 +299,14 @@ async function bundle(site, root) {
299
299
  }
300
300
  },
301
301
  }).catch((e) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
302
- const { output } = await built.generate({ format: "cjs", exports: "auto", inlineDynamicImports: true });
302
+ const { output } = await built.generate({ format: "es", inlineDynamicImports: true });
303
303
  await built.close();
304
- const cjs = output[0].code;
305
- return [
306
- "var module = { exports: {} }, exports = module.exports;",
307
- cjs,
308
- "var __genroc_main = module.exports.default ?? module.exports;",
309
- 'if (typeof __genroc_main !== "function") {',
310
- ` throw new Error(${JSON.stringify(`${site.path} has no default export function`)});`,
311
- "}",
312
- "return await __genroc_main(input);",
313
- ].join("\n");
304
+ // Refused here rather than in the realm: the evaluator can only report it against a running
305
+ // instance, and the file it names is on this machine.
306
+ if (!output[0].exports.includes("default")) {
307
+ die(`${site.path}: a script must \`export default\` the function to run`);
308
+ }
309
+ return output[0].code;
314
310
  }
315
311
  // ── main ───────────────────────────────────────────────────────────────────────
316
312
  const stdin = await new Promise((resolve, reject) => {
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
@@ -309,12 +309,12 @@ function tsResolve(configPath: string | null): Plugin {
309
309
  };
310
310
  }
311
311
 
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. */
312
+ /** Bundles to a self-contained ES module, which is what the evaluator imports: the default
313
+ * export it calls is the author's own, so nothing wraps or rewrites the code between the two.
314
+ * Bundling is entirely the importer's job, so a definition version pins its code forever. */
315
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
316
+ // Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
317
+ // unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
318
318
  // of a module that will not be there, which bundles clean and fails at runtime.
319
319
  const built = await rollup({
320
320
  input: site.path,
@@ -335,18 +335,14 @@ async function bundle(site: Site, root: string): Promise<string> {
335
335
  },
336
336
  }).catch((e: unknown) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
337
337
 
338
- const { output } = await built.generate({ format: "cjs", exports: "auto", inlineDynamicImports: true });
338
+ const { output } = await built.generate({ format: "es", inlineDynamicImports: true });
339
339
  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");
340
+ // Refused here rather than in the realm: the evaluator can only report it against a running
341
+ // instance, and the file it names is on this machine.
342
+ if (!output[0].exports.includes("default")) {
343
+ die(`${site.path}: a script must \`export default\` the function to run`);
344
+ }
345
+ return output[0].code;
350
346
  }
351
347
 
352
348
  // ── main ───────────────────────────────────────────────────────────────────────
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.ee19083",
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
  });