@genroc/eval-node 0.0.0-edge.86dae06 → 0.0.0-edge.939198f
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 +40 -33
- package/dist/eval.js +4 -3
- package/dist/import.js +104 -59
- package/dist/realm.js +66 -75
- package/dist/worker.js +6 -1
- package/eval.ts +5 -3
- package/import.ts +189 -80
- package/package.json +1 -1
- package/realm.ts +61 -77
- package/worker.ts +7 -1
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": "
|
|
100
|
-
"input": { "amount": 250 },
|
|
101
|
-
"timeout_ms": 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
|
|
106
|
-
|
|
107
|
-
|
|
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
|
-
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
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: "$: input"
|
|
150
155
|
result_schema: { type: object, properties: { fee: { type: number } }, required: [fee] }
|
|
151
156
|
raises:
|
|
@@ -164,7 +169,7 @@ no retry policy to write.
|
|
|
164
169
|
|
|
165
170
|
- id: script_failed
|
|
166
171
|
switch:
|
|
167
|
-
- case: '
|
|
172
|
+
- case: 'last_error.data.name == "LimitExceeded"'
|
|
168
173
|
raise: { code: limit_exceeded, message: "the script rejected the amount" }
|
|
169
174
|
- raise: { code: script_failed, message: "the script failed" }
|
|
170
175
|
```
|
|
@@ -227,12 +232,14 @@ the inferred type of what the definition passes; `Output` is what it declares
|
|
|
227
232
|
`tsc --noEmit`, and bundles — so **a type error is a failed apply**, and a stored definition
|
|
228
233
|
cannot hold code that failed to typecheck.
|
|
229
234
|
|
|
230
|
-
The bundle is
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
|
|
235
|
-
|
|
235
|
+
The bundle is one self-contained ES module whose default export is the author's own — nothing
|
|
236
|
+
wraps or rewrites it, so what the editor checks and what the realm runs are the same module, and
|
|
237
|
+
**a missing default export is a failed apply** rather than a fault at run time. Imports resolve
|
|
238
|
+
through TypeScript under the same config the check ran with, so a `paths` alias that typechecks
|
|
239
|
+
also bundles. They are inlined at build time, so the string a definition version stores is
|
|
240
|
+
self-contained forever — with one exception: **node builtins stay as imports**, which the realm
|
|
241
|
+
resolves. A package is frozen into the definition; `node:fs` is resolved by whatever runner
|
|
242
|
+
executes it.
|
|
236
243
|
|
|
237
244
|
### Your tsconfig, your types
|
|
238
245
|
|
|
@@ -258,8 +265,9 @@ never read by genroc, because genctl doubles every `$` 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
|
|
262
|
-
`
|
|
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 **
|
|
275
|
-
laptop), a 200 KiB body about
|
|
276
|
-
types on every realm — precompiling it to JavaScript would buy that back, and
|
|
277
|
-
not done: a build artefact that goes stale against its source fails silently,
|
|
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
|
-
|
|
281
|
-
|
|
282
|
-
|
|
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
|
|
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
|
|
309
|
-
|
|
310
|
-
|
|
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:
|
|
2
|
-
// the failure kinds in README.md. Nothing here knows about
|
|
3
|
-
// thing that talks to the queue — so this stays testable and
|
|
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"
|
|
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) =>
|
|
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 $def rather than inlining: a task output may
|
|
123
117
|
* reference itself (specs/recursive-type-inference.md) and inlining would not terminate. */
|
|
124
|
-
function declarations(
|
|
118
|
+
function declarations(at) {
|
|
119
|
+
const { site, where } = at;
|
|
120
|
+
const defs = where.$defs ?? {};
|
|
125
121
|
const used = new Set();
|
|
126
|
-
const input = tsType(
|
|
127
|
-
const output = tsType(site.
|
|
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
|
-
`// ${
|
|
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(
|
|
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
|
|
177
|
-
const base = (await nearestTsconfig(dirname(
|
|
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(
|
|
176
|
+
group.push(at);
|
|
181
177
|
else
|
|
182
|
-
groups.set(base, [
|
|
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) => [
|
|
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([
|
|
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")
|
|
275
|
+
return resolvedModule.resolvedFileName.endsWith(".d.ts")
|
|
276
|
+
? null
|
|
277
|
+
: resolvedModule.resolvedFileName;
|
|
274
278
|
},
|
|
275
279
|
};
|
|
276
280
|
}
|
|
277
|
-
/** Bundles to
|
|
278
|
-
*
|
|
279
|
-
*
|
|
280
|
-
async function bundle(
|
|
281
|
-
|
|
282
|
-
//
|
|
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:
|
|
290
|
+
input: at.file,
|
|
286
291
|
external: (id) => BUILTIN.has(id),
|
|
287
292
|
plugins: [
|
|
288
|
-
tsResolve(await nearestTsconfig(dirname(
|
|
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(`${
|
|
303
|
+
die(`${at.file}: cannot resolve ${warning.exporter ?? "an import"} — is it installed?`);
|
|
299
304
|
}
|
|
300
305
|
},
|
|
301
|
-
}).catch((e) => die(`${
|
|
302
|
-
const { output } = await built.generate({
|
|
306
|
+
}).catch((e) => die(`${at.file}: ${e instanceof Error ? e.message : String(e)}`));
|
|
307
|
+
const { output } = await built.generate({
|
|
308
|
+
format: "es",
|
|
309
|
+
inlineDynamicImports: true,
|
|
310
|
+
});
|
|
303
311
|
await built.close();
|
|
304
|
-
|
|
305
|
-
|
|
306
|
-
|
|
307
|
-
|
|
308
|
-
|
|
309
|
-
|
|
310
|
-
` throw new Error(${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
|
-
|
|
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 ${at.file})`);
|
|
370
|
+
}
|
|
371
|
+
}
|
|
326
372
|
// One script at two sites with different input types is a refusal, not a union: the union
|
|
327
373
|
// is sound and would typecheck a body that is wrong at one of the sites.
|
|
328
374
|
const byPath = new Map();
|
|
329
|
-
for (const
|
|
330
|
-
const seen = byPath.get(
|
|
331
|
-
const defsOf = (x) => (manifest.schemas[x.process]?.$defs ?? {});
|
|
375
|
+
for (const at of located) {
|
|
376
|
+
const seen = byPath.get(at.file);
|
|
332
377
|
if (seen &&
|
|
333
|
-
JSON.stringify(
|
|
334
|
-
|
|
335
|
-
|
|
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(
|
|
383
|
+
byPath.set(at.file, at);
|
|
339
384
|
}
|
|
340
|
-
for (const
|
|
341
|
-
|
|
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(
|
|
392
|
+
await typecheck([...byPath.values()]);
|
|
348
393
|
const code = [];
|
|
349
|
-
for (const
|
|
350
|
-
code.push(await bundle(
|
|
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 —
|
|
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
|
+
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
|
-
|
|
10
|
-
|
|
11
|
-
//
|
|
12
|
-
|
|
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
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
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
|
|
35
|
+
let last = 0;
|
|
60
36
|
for (let i = 0; i < lines.length; i++)
|
|
61
|
-
if (
|
|
62
|
-
|
|
63
|
-
const
|
|
64
|
-
|
|
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
|
|
42
|
+
function describe(err, kind) {
|
|
77
43
|
if (err instanceof Error) {
|
|
78
|
-
return { kind, name: err.name, message: err.message, stack: scriptStack(err
|
|
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
|
-
|
|
93
|
-
let
|
|
65
|
+
source = req.code;
|
|
66
|
+
let main;
|
|
94
67
|
try {
|
|
95
|
-
|
|
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"
|
|
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
|
|
86
|
+
value = await main(req.input);
|
|
105
87
|
}
|
|
106
88
|
catch (err) {
|
|
107
|
-
return { ok: false, failure: describe(err, "threw"
|
|
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"
|
|
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
|
-
|
|
110
|
+
const reply = await run(req);
|
|
111
|
+
await Promise.all([flush(process.stdout), flush(process.stderr)]);
|
|
112
|
+
parentPort.postMessage(reply);
|
|
122
113
|
});
|
package/dist/worker.js
CHANGED
|
@@ -4,6 +4,7 @@
|
|
|
4
4
|
// nothing about the queue, which is what keeps the containment strategy swappable.
|
|
5
5
|
//
|
|
6
6
|
// See README.md for the contract, and specs/external-task-queue.md for the queue itself.
|
|
7
|
+
import { readFileSync } from "node:fs";
|
|
7
8
|
import { evaluate } from "./eval.js";
|
|
8
9
|
const SERVER = (process.env.GENROC_SERVER ?? "http://localhost:8448").replace(/\/$/, "");
|
|
9
10
|
const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
|
|
@@ -15,7 +16,11 @@ const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
|
|
|
15
16
|
// Sent as a header rather than in the URL because Node's fetch REFUSES a URL carrying
|
|
16
17
|
// credentials ("Request cannot be constructed from a URL that includes credentials"), so the
|
|
17
18
|
// basic-auth-in-the-URL trick that works for genctl is not available here.
|
|
18
|
-
|
|
19
|
+
// GENROC_TOKEN_FILE is the mounted-secret shape: a credential in a file rather than an
|
|
20
|
+
// environment variable, so it stays out of `docker inspect` and out of the process environment
|
|
21
|
+
// any child inherits. The inline variable wins when both are set.
|
|
22
|
+
const TOKEN = process.env.GENROC_TOKEN ??
|
|
23
|
+
(process.env.GENROC_TOKEN_FILE ? readFileSync(process.env.GENROC_TOKEN_FILE, "utf8").trim() : "");
|
|
19
24
|
const authHeaders = TOKEN ? { authorization: `Bearer ${TOKEN}` } : {};
|
|
20
25
|
// Concurrency is the worker's to set, and that is the point of pulling: under the old fetch
|
|
21
26
|
// shape genroc decided how many scripts ran at once (--max-concurrent, default 200) and the
|