@ai-sdk-tool/parser 1.0.2 → 2.0.0
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/dist/index.d.mts +11 -12
- package/dist/index.d.ts +11 -12
- package/dist/index.js +652 -65
- package/dist/index.js.map +1 -1
- package/dist/index.mjs +650 -54
- package/dist/index.mjs.map +1 -1
- package/package.json +4 -4
package/dist/index.mjs
CHANGED
|
@@ -1,10 +1,7 @@
|
|
|
1
1
|
// src/tool-call-middleware.ts
|
|
2
|
-
import {
|
|
3
|
-
generateId
|
|
4
|
-
} from "ai";
|
|
5
|
-
import * as RJSON from "relaxed-json";
|
|
2
|
+
import { generateId } from "@ai-sdk/provider-utils";
|
|
6
3
|
|
|
7
|
-
// src/utils.ts
|
|
4
|
+
// src/utils/get-potential-start-index.ts
|
|
8
5
|
function getPotentialStartIndex(text, searchedText) {
|
|
9
6
|
if (searchedText.length === 0) {
|
|
10
7
|
return null;
|
|
@@ -22,26 +19,606 @@ function getPotentialStartIndex(text, searchedText) {
|
|
|
22
19
|
return null;
|
|
23
20
|
}
|
|
24
21
|
|
|
22
|
+
// src/utils/relaxed-json.ts
|
|
23
|
+
function some(array, f) {
|
|
24
|
+
let acc = false;
|
|
25
|
+
for (let i = 0; i < array.length; i++) {
|
|
26
|
+
const result = f(array[i], i, array);
|
|
27
|
+
acc = result === void 0 ? false : result;
|
|
28
|
+
if (acc) {
|
|
29
|
+
return acc;
|
|
30
|
+
}
|
|
31
|
+
}
|
|
32
|
+
return acc;
|
|
33
|
+
}
|
|
34
|
+
function makeLexer(tokenSpecs) {
|
|
35
|
+
return function(contents) {
|
|
36
|
+
const tokens = [];
|
|
37
|
+
let line = 1;
|
|
38
|
+
function findToken() {
|
|
39
|
+
const result = some(tokenSpecs, (tokenSpec) => {
|
|
40
|
+
const m = tokenSpec.re.exec(contents);
|
|
41
|
+
if (m) {
|
|
42
|
+
const raw = m[0];
|
|
43
|
+
contents = contents.slice(raw.length);
|
|
44
|
+
return {
|
|
45
|
+
raw,
|
|
46
|
+
matched: tokenSpec.f(m)
|
|
47
|
+
// Process the match using the spec's function
|
|
48
|
+
};
|
|
49
|
+
} else {
|
|
50
|
+
return void 0;
|
|
51
|
+
}
|
|
52
|
+
});
|
|
53
|
+
return result === false ? void 0 : result;
|
|
54
|
+
}
|
|
55
|
+
while (contents !== "") {
|
|
56
|
+
const matched = findToken();
|
|
57
|
+
if (!matched) {
|
|
58
|
+
const err = new SyntaxError(
|
|
59
|
+
`Unexpected character: ${contents[0]}; input: ${contents.substr(
|
|
60
|
+
0,
|
|
61
|
+
100
|
|
62
|
+
)}`
|
|
63
|
+
);
|
|
64
|
+
err.line = line;
|
|
65
|
+
throw err;
|
|
66
|
+
}
|
|
67
|
+
const tokenWithLine = matched.matched;
|
|
68
|
+
tokenWithLine.line = line;
|
|
69
|
+
line += matched.raw.replace(/[^\n]/g, "").length;
|
|
70
|
+
tokens.push(tokenWithLine);
|
|
71
|
+
}
|
|
72
|
+
return tokens;
|
|
73
|
+
};
|
|
74
|
+
}
|
|
75
|
+
function fStringSingle(m) {
|
|
76
|
+
const content = m[1].replace(
|
|
77
|
+
/([^'\\]|\\['bnrtf\\]|\\u[0-9a-fA-F]{4})/g,
|
|
78
|
+
(mm) => {
|
|
79
|
+
if (mm === '"') {
|
|
80
|
+
return '\\"';
|
|
81
|
+
} else if (mm === "\\'") {
|
|
82
|
+
return "'";
|
|
83
|
+
} else {
|
|
84
|
+
return mm;
|
|
85
|
+
}
|
|
86
|
+
}
|
|
87
|
+
);
|
|
88
|
+
const match = `"${content}"`;
|
|
89
|
+
return {
|
|
90
|
+
type: "string",
|
|
91
|
+
match,
|
|
92
|
+
// The transformed, double-quoted string representation
|
|
93
|
+
// Use JSON.parse on the transformed string to handle escape sequences correctly
|
|
94
|
+
value: JSON.parse(match)
|
|
95
|
+
};
|
|
96
|
+
}
|
|
97
|
+
function fStringDouble(m) {
|
|
98
|
+
return {
|
|
99
|
+
type: "string",
|
|
100
|
+
match: m[0],
|
|
101
|
+
// The raw matched string (including quotes)
|
|
102
|
+
value: JSON.parse(m[0])
|
|
103
|
+
// Use JSON.parse to handle escapes and get the value
|
|
104
|
+
};
|
|
105
|
+
}
|
|
106
|
+
function fIdentifier(m) {
|
|
107
|
+
const value = m[0];
|
|
108
|
+
const match = '"' + value.replace(/\\/g, "\\\\").replace(/"/g, '\\"') + // Escape backslashes and quotes
|
|
109
|
+
'"';
|
|
110
|
+
return {
|
|
111
|
+
type: "string",
|
|
112
|
+
// Treat identifiers as strings
|
|
113
|
+
value,
|
|
114
|
+
// The original identifier name
|
|
115
|
+
match
|
|
116
|
+
// The double-quoted string representation
|
|
117
|
+
};
|
|
118
|
+
}
|
|
119
|
+
function fComment(m) {
|
|
120
|
+
const match = m[0].replace(/./g, (c) => /\s/.test(c) ? c : " ");
|
|
121
|
+
return {
|
|
122
|
+
type: " ",
|
|
123
|
+
// Represent comments as whitespace tokens
|
|
124
|
+
match,
|
|
125
|
+
// String containing original newlines and spaces for other chars
|
|
126
|
+
value: void 0
|
|
127
|
+
// Comments don't have a semantic value
|
|
128
|
+
};
|
|
129
|
+
}
|
|
130
|
+
function fNumber(m) {
|
|
131
|
+
return {
|
|
132
|
+
type: "number",
|
|
133
|
+
match: m[0],
|
|
134
|
+
// The raw matched number string
|
|
135
|
+
value: parseFloat(m[0])
|
|
136
|
+
// Convert string to number
|
|
137
|
+
};
|
|
138
|
+
}
|
|
139
|
+
function fKeyword(m) {
|
|
140
|
+
let value;
|
|
141
|
+
switch (m[0]) {
|
|
142
|
+
case "null":
|
|
143
|
+
value = null;
|
|
144
|
+
break;
|
|
145
|
+
case "true":
|
|
146
|
+
value = true;
|
|
147
|
+
break;
|
|
148
|
+
case "false":
|
|
149
|
+
value = false;
|
|
150
|
+
break;
|
|
151
|
+
default:
|
|
152
|
+
throw new Error(`Unexpected keyword: ${m[0]}`);
|
|
153
|
+
}
|
|
154
|
+
return {
|
|
155
|
+
type: "atom",
|
|
156
|
+
// Use 'atom' type for these literals
|
|
157
|
+
match: m[0],
|
|
158
|
+
// The raw matched keyword
|
|
159
|
+
value
|
|
160
|
+
// The corresponding JavaScript value
|
|
161
|
+
};
|
|
162
|
+
}
|
|
163
|
+
function makeTokenSpecs(relaxed) {
|
|
164
|
+
function f(type) {
|
|
165
|
+
return function(m) {
|
|
166
|
+
return { type, match: m[0], value: void 0 };
|
|
167
|
+
};
|
|
168
|
+
}
|
|
169
|
+
let tokenSpecs = [
|
|
170
|
+
{ re: /^\s+/, f: f(" ") },
|
|
171
|
+
// Whitespace
|
|
172
|
+
{ re: /^\{/, f: f("{") },
|
|
173
|
+
// Object start
|
|
174
|
+
{ re: /^\}/, f: f("}") },
|
|
175
|
+
// Object end
|
|
176
|
+
{ re: /^\[/, f: f("[") },
|
|
177
|
+
// Array start
|
|
178
|
+
{ re: /^\]/, f: f("]") },
|
|
179
|
+
// Array end
|
|
180
|
+
{ re: /^,/, f: f(",") },
|
|
181
|
+
// Comma separator
|
|
182
|
+
{ re: /^:/, f: f(":") },
|
|
183
|
+
// Key-value separator
|
|
184
|
+
{ re: /^(?:true|false|null)/, f: fKeyword },
|
|
185
|
+
// Keywords
|
|
186
|
+
// Number: optional sign, digits, optional decimal part, optional exponent
|
|
187
|
+
{ re: /^\-?\d+(?:\.\d+)?(?:[eE][+-]?\d+)?/, f: fNumber },
|
|
188
|
+
// String: double-quoted, handles escapes
|
|
189
|
+
{ re: /^"(?:[^"\\]|\\["bnrtf\\\/]|\\u[0-9a-fA-F]{4})*"/, f: fStringDouble }
|
|
190
|
+
];
|
|
191
|
+
if (relaxed) {
|
|
192
|
+
tokenSpecs = tokenSpecs.concat([
|
|
193
|
+
// Single-quoted strings
|
|
194
|
+
{
|
|
195
|
+
re: /^'((?:[^'\\]|\\['bnrtf\\\/]|\\u[0-9a-fA-F]{4})*)'/,
|
|
196
|
+
f: fStringSingle
|
|
197
|
+
},
|
|
198
|
+
// Single-line comments (// ...)
|
|
199
|
+
{ re: /^\/\/.*?(?:\r\n|\r|\n)/, f: fComment },
|
|
200
|
+
// Multi-line comments (/* ... */)
|
|
201
|
+
{ re: /^\/\*[\s\S]*?\*\//, f: fComment },
|
|
202
|
+
// Unquoted identifiers (treated as strings)
|
|
203
|
+
// Allows letters, numbers, _, -, +, ., *, ?, !, |, &, %, ^, /, #, \
|
|
204
|
+
{ re: /^[$a-zA-Z0-9_\-+\.\*\?!\|&%\^\/#\\]+/, f: fIdentifier }
|
|
205
|
+
// Note: The order matters here. Identifiers are checked after keywords/numbers.
|
|
206
|
+
]);
|
|
207
|
+
}
|
|
208
|
+
return tokenSpecs;
|
|
209
|
+
}
|
|
210
|
+
var lexer = makeLexer(makeTokenSpecs(true));
|
|
211
|
+
var strictLexer = makeLexer(makeTokenSpecs(false));
|
|
212
|
+
function previousNWSToken(tokens, index) {
|
|
213
|
+
for (; index >= 0; index--) {
|
|
214
|
+
if (tokens[index].type !== " ") {
|
|
215
|
+
return index;
|
|
216
|
+
}
|
|
217
|
+
}
|
|
218
|
+
return void 0;
|
|
219
|
+
}
|
|
220
|
+
function stripTrailingComma(tokens) {
|
|
221
|
+
const res = [];
|
|
222
|
+
tokens.forEach((token, index) => {
|
|
223
|
+
if (index > 0 && (token.type === "]" || token.type === "}")) {
|
|
224
|
+
const prevNWSTokenIndex = previousNWSToken(res, res.length - 1);
|
|
225
|
+
if (prevNWSTokenIndex !== void 0 && res[prevNWSTokenIndex].type === ",") {
|
|
226
|
+
const preCommaIndex = previousNWSToken(res, prevNWSTokenIndex - 1);
|
|
227
|
+
if (preCommaIndex !== void 0 && res[preCommaIndex].type !== "[" && res[preCommaIndex].type !== "{") {
|
|
228
|
+
res[prevNWSTokenIndex] = {
|
|
229
|
+
type: " ",
|
|
230
|
+
match: " ",
|
|
231
|
+
// Represent as a single space
|
|
232
|
+
value: void 0,
|
|
233
|
+
// Whitespace has no value
|
|
234
|
+
line: res[prevNWSTokenIndex].line
|
|
235
|
+
// Preserve original line number
|
|
236
|
+
};
|
|
237
|
+
}
|
|
238
|
+
}
|
|
239
|
+
}
|
|
240
|
+
res.push(token);
|
|
241
|
+
});
|
|
242
|
+
return res;
|
|
243
|
+
}
|
|
244
|
+
function popToken(tokens, state) {
|
|
245
|
+
const token = tokens[state.pos];
|
|
246
|
+
state.pos += 1;
|
|
247
|
+
if (!token) {
|
|
248
|
+
const lastLine = tokens.length !== 0 ? tokens[tokens.length - 1].line : 1;
|
|
249
|
+
return { type: "eof", match: "", value: void 0, line: lastLine };
|
|
250
|
+
}
|
|
251
|
+
return token;
|
|
252
|
+
}
|
|
253
|
+
function strToken(token) {
|
|
254
|
+
switch (token.type) {
|
|
255
|
+
case "atom":
|
|
256
|
+
case "string":
|
|
257
|
+
case "number":
|
|
258
|
+
return `${token.type} ${token.match}`;
|
|
259
|
+
case "eof":
|
|
260
|
+
return "end-of-file";
|
|
261
|
+
default:
|
|
262
|
+
return `'${token.type}'`;
|
|
263
|
+
}
|
|
264
|
+
}
|
|
265
|
+
function skipColon(tokens, state) {
|
|
266
|
+
const colon = popToken(tokens, state);
|
|
267
|
+
if (colon.type !== ":") {
|
|
268
|
+
const message = `Unexpected token: ${strToken(colon)}, expected ':'`;
|
|
269
|
+
if (state.tolerant) {
|
|
270
|
+
state.warnings.push({
|
|
271
|
+
message,
|
|
272
|
+
line: colon.line
|
|
273
|
+
});
|
|
274
|
+
state.pos -= 1;
|
|
275
|
+
} else {
|
|
276
|
+
const err = new SyntaxError(message);
|
|
277
|
+
err.line = colon.line;
|
|
278
|
+
throw err;
|
|
279
|
+
}
|
|
280
|
+
}
|
|
281
|
+
}
|
|
282
|
+
function skipPunctuation(tokens, state, valid) {
|
|
283
|
+
const punctuation = [",", ":", "]", "}"];
|
|
284
|
+
let token = popToken(tokens, state);
|
|
285
|
+
while (true) {
|
|
286
|
+
if (valid && valid.includes(token.type)) {
|
|
287
|
+
return token;
|
|
288
|
+
} else if (token.type === "eof") {
|
|
289
|
+
return token;
|
|
290
|
+
} else if (punctuation.includes(token.type)) {
|
|
291
|
+
const message = `Unexpected token: ${strToken(
|
|
292
|
+
token
|
|
293
|
+
)}, expected '[', '{', number, string or atom`;
|
|
294
|
+
if (state.tolerant) {
|
|
295
|
+
state.warnings.push({
|
|
296
|
+
message,
|
|
297
|
+
line: token.line
|
|
298
|
+
});
|
|
299
|
+
token = popToken(tokens, state);
|
|
300
|
+
} else {
|
|
301
|
+
const err = new SyntaxError(message);
|
|
302
|
+
err.line = token.line;
|
|
303
|
+
throw err;
|
|
304
|
+
}
|
|
305
|
+
} else {
|
|
306
|
+
return token;
|
|
307
|
+
}
|
|
308
|
+
}
|
|
309
|
+
}
|
|
310
|
+
function raiseError(state, token, message) {
|
|
311
|
+
if (state.tolerant) {
|
|
312
|
+
state.warnings.push({
|
|
313
|
+
message,
|
|
314
|
+
line: token.line
|
|
315
|
+
});
|
|
316
|
+
} else {
|
|
317
|
+
const err = new SyntaxError(message);
|
|
318
|
+
err.line = token.line;
|
|
319
|
+
throw err;
|
|
320
|
+
}
|
|
321
|
+
}
|
|
322
|
+
function raiseUnexpected(state, token, expected) {
|
|
323
|
+
raiseError(
|
|
324
|
+
state,
|
|
325
|
+
token,
|
|
326
|
+
`Unexpected token: ${strToken(token)}, expected ${expected}`
|
|
327
|
+
);
|
|
328
|
+
}
|
|
329
|
+
function checkDuplicates(state, obj, token) {
|
|
330
|
+
const key = String(token.value);
|
|
331
|
+
if (!state.duplicate && Object.prototype.hasOwnProperty.call(obj, key)) {
|
|
332
|
+
raiseError(state, token, `Duplicate key: ${key}`);
|
|
333
|
+
}
|
|
334
|
+
}
|
|
335
|
+
function appendPair(state, obj, key, value) {
|
|
336
|
+
const finalValue = state.reviver ? state.reviver(key, value) : value;
|
|
337
|
+
if (finalValue !== void 0) {
|
|
338
|
+
obj[key] = finalValue;
|
|
339
|
+
}
|
|
340
|
+
}
|
|
341
|
+
function parsePair(tokens, state, obj) {
|
|
342
|
+
let token = skipPunctuation(tokens, state, [":", "string", "number", "atom"]);
|
|
343
|
+
let key;
|
|
344
|
+
let value;
|
|
345
|
+
if (token.type !== "string") {
|
|
346
|
+
raiseUnexpected(state, token, "string key");
|
|
347
|
+
if (state.tolerant) {
|
|
348
|
+
switch (token.type) {
|
|
349
|
+
case ":":
|
|
350
|
+
token = {
|
|
351
|
+
type: "string",
|
|
352
|
+
value: "null",
|
|
353
|
+
match: '"null"',
|
|
354
|
+
line: token.line
|
|
355
|
+
};
|
|
356
|
+
state.pos -= 1;
|
|
357
|
+
break;
|
|
358
|
+
case "number":
|
|
359
|
+
// Use number as string key
|
|
360
|
+
case "atom":
|
|
361
|
+
token = {
|
|
362
|
+
type: "string",
|
|
363
|
+
value: String(token.value),
|
|
364
|
+
match: `"${token.value}"`,
|
|
365
|
+
line: token.line
|
|
366
|
+
};
|
|
367
|
+
break;
|
|
368
|
+
case "[":
|
|
369
|
+
// Assume missing key before an array
|
|
370
|
+
case "{":
|
|
371
|
+
state.pos -= 1;
|
|
372
|
+
value = parseAny(tokens, state);
|
|
373
|
+
checkDuplicates(state, obj, {
|
|
374
|
+
type: "string",
|
|
375
|
+
value: "null",
|
|
376
|
+
match: '"null"',
|
|
377
|
+
line: token.line
|
|
378
|
+
});
|
|
379
|
+
appendPair(state, obj, "null", value);
|
|
380
|
+
return;
|
|
381
|
+
// Finished parsing this "pair"
|
|
382
|
+
case "eof":
|
|
383
|
+
return;
|
|
384
|
+
// Cannot recover
|
|
385
|
+
default:
|
|
386
|
+
return;
|
|
387
|
+
}
|
|
388
|
+
} else {
|
|
389
|
+
return;
|
|
390
|
+
}
|
|
391
|
+
}
|
|
392
|
+
checkDuplicates(state, obj, token);
|
|
393
|
+
key = String(token.value);
|
|
394
|
+
skipColon(tokens, state);
|
|
395
|
+
value = parseAny(tokens, state);
|
|
396
|
+
appendPair(state, obj, key, value);
|
|
397
|
+
}
|
|
398
|
+
function parseElement(tokens, state, arr) {
|
|
399
|
+
const key = arr.length;
|
|
400
|
+
const value = parseAny(tokens, state);
|
|
401
|
+
arr[key] = state.reviver ? state.reviver(String(key), value) : value;
|
|
402
|
+
}
|
|
403
|
+
function parseObject(tokens, state) {
|
|
404
|
+
const obj = {};
|
|
405
|
+
return parseMany(tokens, state, obj, {
|
|
406
|
+
skip: [":", "}"],
|
|
407
|
+
// Initially skip over colon or closing brace (for empty/tolerant cases)
|
|
408
|
+
elementParser: parsePair,
|
|
409
|
+
// Use parsePair to parse each key-value element
|
|
410
|
+
elementName: "string key",
|
|
411
|
+
// Expected element type for errors
|
|
412
|
+
endSymbol: "}"
|
|
413
|
+
// The closing token for an object
|
|
414
|
+
});
|
|
415
|
+
}
|
|
416
|
+
function parseArray(tokens, state) {
|
|
417
|
+
const arr = [];
|
|
418
|
+
return parseMany(tokens, state, arr, {
|
|
419
|
+
skip: ["]"],
|
|
420
|
+
// Initially skip over closing bracket (for empty/tolerant cases)
|
|
421
|
+
elementParser: parseElement,
|
|
422
|
+
// Use parseElement to parse each array item
|
|
423
|
+
elementName: "json value",
|
|
424
|
+
// Expected element type for errors
|
|
425
|
+
endSymbol: "]"
|
|
426
|
+
// The closing token for an array
|
|
427
|
+
});
|
|
428
|
+
}
|
|
429
|
+
function parseMany(tokens, state, result, opts) {
|
|
430
|
+
let token = skipPunctuation(tokens, state, opts.skip);
|
|
431
|
+
if (token.type === "eof") {
|
|
432
|
+
raiseUnexpected(state, token, `'${opts.endSymbol}' or ${opts.elementName}`);
|
|
433
|
+
if (state.tolerant) {
|
|
434
|
+
return result;
|
|
435
|
+
} else {
|
|
436
|
+
return result;
|
|
437
|
+
}
|
|
438
|
+
}
|
|
439
|
+
if (token.type === opts.endSymbol) {
|
|
440
|
+
return result;
|
|
441
|
+
}
|
|
442
|
+
state.pos -= 1;
|
|
443
|
+
opts.elementParser(tokens, state, result);
|
|
444
|
+
while (true) {
|
|
445
|
+
token = popToken(tokens, state);
|
|
446
|
+
if (token.type !== opts.endSymbol && token.type !== ",") {
|
|
447
|
+
raiseUnexpected(state, token, `',' or '${opts.endSymbol}'`);
|
|
448
|
+
if (state.tolerant) {
|
|
449
|
+
if (token.type === "eof") {
|
|
450
|
+
return result;
|
|
451
|
+
}
|
|
452
|
+
state.pos -= 1;
|
|
453
|
+
} else {
|
|
454
|
+
return result;
|
|
455
|
+
}
|
|
456
|
+
}
|
|
457
|
+
switch (token.type) {
|
|
458
|
+
case opts.endSymbol:
|
|
459
|
+
return result;
|
|
460
|
+
case ",":
|
|
461
|
+
const nextToken = tokens[state.pos];
|
|
462
|
+
if (state.tolerant && nextToken && nextToken.type === opts.endSymbol) {
|
|
463
|
+
raiseError(state, token, `Trailing comma before '${opts.endSymbol}'`);
|
|
464
|
+
popToken(tokens, state);
|
|
465
|
+
return result;
|
|
466
|
+
}
|
|
467
|
+
opts.elementParser(tokens, state, result);
|
|
468
|
+
break;
|
|
469
|
+
// Default case is only reachable in tolerant mode recovery above
|
|
470
|
+
default:
|
|
471
|
+
opts.elementParser(tokens, state, result);
|
|
472
|
+
break;
|
|
473
|
+
}
|
|
474
|
+
}
|
|
475
|
+
}
|
|
476
|
+
function endChecks(tokens, state, ret) {
|
|
477
|
+
if (state.pos < tokens.length) {
|
|
478
|
+
if (state.tolerant) {
|
|
479
|
+
skipPunctuation(tokens, state);
|
|
480
|
+
}
|
|
481
|
+
if (state.pos < tokens.length) {
|
|
482
|
+
raiseError(
|
|
483
|
+
state,
|
|
484
|
+
tokens[state.pos],
|
|
485
|
+
`Unexpected token: ${strToken(tokens[state.pos])}, expected end-of-input`
|
|
486
|
+
);
|
|
487
|
+
}
|
|
488
|
+
}
|
|
489
|
+
if (state.tolerant && state.warnings.length > 0) {
|
|
490
|
+
const message = state.warnings.length === 1 ? state.warnings[0].message : `${state.warnings.length} parse warnings`;
|
|
491
|
+
const err = new SyntaxError(message);
|
|
492
|
+
err.line = state.warnings[0].line;
|
|
493
|
+
err.warnings = state.warnings;
|
|
494
|
+
err.obj = ret;
|
|
495
|
+
throw err;
|
|
496
|
+
}
|
|
497
|
+
}
|
|
498
|
+
function parseAny(tokens, state, end = false) {
|
|
499
|
+
const token = skipPunctuation(tokens, state);
|
|
500
|
+
let ret;
|
|
501
|
+
if (token.type === "eof") {
|
|
502
|
+
if (end) {
|
|
503
|
+
raiseUnexpected(state, token, "json value");
|
|
504
|
+
}
|
|
505
|
+
raiseUnexpected(state, token, "json value");
|
|
506
|
+
return void 0;
|
|
507
|
+
}
|
|
508
|
+
switch (token.type) {
|
|
509
|
+
case "{":
|
|
510
|
+
ret = parseObject(tokens, state);
|
|
511
|
+
break;
|
|
512
|
+
case "[":
|
|
513
|
+
ret = parseArray(tokens, state);
|
|
514
|
+
break;
|
|
515
|
+
case "string":
|
|
516
|
+
// String literal
|
|
517
|
+
case "number":
|
|
518
|
+
// Number literal
|
|
519
|
+
case "atom":
|
|
520
|
+
ret = token.value;
|
|
521
|
+
break;
|
|
522
|
+
default:
|
|
523
|
+
raiseUnexpected(state, token, "json value");
|
|
524
|
+
if (state.tolerant) {
|
|
525
|
+
ret = null;
|
|
526
|
+
} else {
|
|
527
|
+
return void 0;
|
|
528
|
+
}
|
|
529
|
+
}
|
|
530
|
+
if (end) {
|
|
531
|
+
ret = state.reviver ? state.reviver("", ret) : ret;
|
|
532
|
+
endChecks(tokens, state, ret);
|
|
533
|
+
}
|
|
534
|
+
return ret;
|
|
535
|
+
}
|
|
536
|
+
function parse(text, optsOrReviver) {
|
|
537
|
+
let options = {};
|
|
538
|
+
if (typeof optsOrReviver === "function") {
|
|
539
|
+
options.reviver = optsOrReviver;
|
|
540
|
+
} else if (optsOrReviver !== null && typeof optsOrReviver === "object") {
|
|
541
|
+
options = { ...optsOrReviver };
|
|
542
|
+
} else if (optsOrReviver !== void 0) {
|
|
543
|
+
throw new TypeError(
|
|
544
|
+
"Second argument must be a reviver function or an options object."
|
|
545
|
+
);
|
|
546
|
+
}
|
|
547
|
+
if (options.relaxed === void 0) {
|
|
548
|
+
if (options.warnings === true || options.tolerant === true) {
|
|
549
|
+
options.relaxed = true;
|
|
550
|
+
} else if (options.warnings === false && options.tolerant === false) {
|
|
551
|
+
options.relaxed = false;
|
|
552
|
+
} else {
|
|
553
|
+
options.relaxed = true;
|
|
554
|
+
}
|
|
555
|
+
}
|
|
556
|
+
options.tolerant = options.tolerant || options.warnings || false;
|
|
557
|
+
options.warnings = options.warnings || false;
|
|
558
|
+
options.duplicate = options.duplicate || false;
|
|
559
|
+
if (!options.relaxed && !options.warnings && !options.tolerant) {
|
|
560
|
+
if (!options.duplicate) {
|
|
561
|
+
} else {
|
|
562
|
+
return JSON.parse(text, options.reviver);
|
|
563
|
+
}
|
|
564
|
+
}
|
|
565
|
+
const lexerToUse = options.relaxed ? lexer : strictLexer;
|
|
566
|
+
let tokens = lexerToUse(text);
|
|
567
|
+
if (options.relaxed) {
|
|
568
|
+
tokens = stripTrailingComma(tokens);
|
|
569
|
+
}
|
|
570
|
+
if (options.warnings || options.tolerant) {
|
|
571
|
+
tokens = tokens.filter((token) => token.type !== " ");
|
|
572
|
+
const state = {
|
|
573
|
+
pos: 0,
|
|
574
|
+
reviver: options.reviver,
|
|
575
|
+
tolerant: options.tolerant,
|
|
576
|
+
duplicate: !options.duplicate,
|
|
577
|
+
// Internal state: true means *check* for duplicates
|
|
578
|
+
warnings: []
|
|
579
|
+
};
|
|
580
|
+
return parseAny(tokens, state, true);
|
|
581
|
+
} else {
|
|
582
|
+
const newtext = tokens.reduce((str, token) => {
|
|
583
|
+
return str + token.match;
|
|
584
|
+
}, "");
|
|
585
|
+
if (!options.relaxed && !options.warnings && !options.tolerant && options.duplicate) {
|
|
586
|
+
return JSON.parse(text, options.reviver);
|
|
587
|
+
} else if (options.warnings || options.tolerant || !options.duplicate) {
|
|
588
|
+
tokens = lexerToUse(text);
|
|
589
|
+
if (options.relaxed) {
|
|
590
|
+
tokens = stripTrailingComma(tokens);
|
|
591
|
+
}
|
|
592
|
+
tokens = tokens.filter((token) => token.type !== " ");
|
|
593
|
+
const state = {
|
|
594
|
+
pos: 0,
|
|
595
|
+
reviver: options.reviver,
|
|
596
|
+
tolerant: options.tolerant || false,
|
|
597
|
+
// Ensure boolean
|
|
598
|
+
duplicate: !options.duplicate,
|
|
599
|
+
// true = check duplicates
|
|
600
|
+
warnings: []
|
|
601
|
+
};
|
|
602
|
+
return parseAny(tokens, state, true);
|
|
603
|
+
} else {
|
|
604
|
+
tokens = lexer(text);
|
|
605
|
+
tokens = stripTrailingComma(tokens);
|
|
606
|
+
const newtext2 = tokens.reduce((str, token) => str + token.match, "");
|
|
607
|
+
return JSON.parse(newtext2, options.reviver);
|
|
608
|
+
}
|
|
609
|
+
}
|
|
610
|
+
}
|
|
611
|
+
|
|
25
612
|
// src/tool-call-middleware.ts
|
|
26
|
-
var defaultTemplate = (tools) => `You are a function calling AI model.
|
|
27
|
-
You are provided with function signatures within <tools></tools> XML tags.
|
|
28
|
-
You may call one or more functions to assist with the user query.
|
|
29
|
-
Don't make assumptions about what values to plug into functions.
|
|
30
|
-
Here are the available tools: <tools>${tools}</tools>
|
|
31
|
-
Use the following pydantic model json schema for each tool call you will make: {'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
|
|
32
|
-
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
|
|
33
|
-
<tool_call>
|
|
34
|
-
{'arguments': <args-dict>, 'name': <function-name>}
|
|
35
|
-
</tool_call>`;
|
|
36
613
|
function createToolMiddleware({
|
|
37
|
-
toolCallTag
|
|
38
|
-
toolCallEndTag
|
|
39
|
-
toolResponseTag
|
|
40
|
-
toolResponseEndTag
|
|
41
|
-
toolSystemPromptTemplate
|
|
614
|
+
toolCallTag,
|
|
615
|
+
toolCallEndTag,
|
|
616
|
+
toolResponseTag,
|
|
617
|
+
toolResponseEndTag,
|
|
618
|
+
toolSystemPromptTemplate
|
|
42
619
|
}) {
|
|
43
620
|
return {
|
|
44
|
-
middlewareVersion: "
|
|
621
|
+
middlewareVersion: "v2",
|
|
45
622
|
wrapStream: async ({ doStream }) => {
|
|
46
623
|
const { stream, ...rest } = await doStream();
|
|
47
624
|
let isFirstToolCall = true;
|
|
@@ -57,7 +634,7 @@ function createToolMiddleware({
|
|
|
57
634
|
if (toolCallBuffer.length > 0) {
|
|
58
635
|
toolCallBuffer.forEach((toolCall) => {
|
|
59
636
|
try {
|
|
60
|
-
const parsedToolCall =
|
|
637
|
+
const parsedToolCall = parse(toolCall);
|
|
61
638
|
controller.enqueue({
|
|
62
639
|
type: "tool-call",
|
|
63
640
|
toolCallType: "function",
|
|
@@ -68,19 +645,19 @@ function createToolMiddleware({
|
|
|
68
645
|
} catch (e) {
|
|
69
646
|
console.error(`Error parsing tool call: ${toolCall}`, e);
|
|
70
647
|
controller.enqueue({
|
|
71
|
-
type: "text
|
|
72
|
-
|
|
648
|
+
type: "text",
|
|
649
|
+
text: `Failed to parse tool call: ${e}`
|
|
73
650
|
});
|
|
74
651
|
}
|
|
75
652
|
});
|
|
76
653
|
}
|
|
77
654
|
controller.enqueue(chunk);
|
|
78
655
|
return;
|
|
79
|
-
} else if (chunk.type !== "text
|
|
656
|
+
} else if (chunk.type !== "text") {
|
|
80
657
|
controller.enqueue(chunk);
|
|
81
658
|
return;
|
|
82
659
|
}
|
|
83
|
-
buffer += chunk.
|
|
660
|
+
buffer += chunk.text;
|
|
84
661
|
function publish(text) {
|
|
85
662
|
if (text.length > 0) {
|
|
86
663
|
const prefix = afterSwitch && (isToolCall ? !isFirstToolCall : !isFirstText) ? "\n" : "";
|
|
@@ -91,8 +668,8 @@ function createToolMiddleware({
|
|
|
91
668
|
toolCallBuffer[toolCallIndex] += text;
|
|
92
669
|
} else {
|
|
93
670
|
controller.enqueue({
|
|
94
|
-
type: "text
|
|
95
|
-
|
|
671
|
+
type: "text",
|
|
672
|
+
text: prefix + text
|
|
96
673
|
});
|
|
97
674
|
}
|
|
98
675
|
afterSwitch = false;
|
|
@@ -131,32 +708,37 @@ function createToolMiddleware({
|
|
|
131
708
|
};
|
|
132
709
|
},
|
|
133
710
|
wrapGenerate: async ({ doGenerate }) => {
|
|
134
|
-
var _a;
|
|
135
711
|
const result = await doGenerate();
|
|
136
|
-
if (
|
|
712
|
+
if (result.content.length !== 1) {
|
|
713
|
+
return result;
|
|
714
|
+
}
|
|
715
|
+
if (result.content[0].type === "text" && !result.content[0].text.includes(toolCallTag)) {
|
|
137
716
|
return result;
|
|
138
717
|
}
|
|
139
718
|
const toolCallRegex = new RegExp(
|
|
140
719
|
`${toolCallTag}(.*?)(?:${toolCallEndTag}|$)`,
|
|
141
720
|
"gs"
|
|
142
721
|
);
|
|
143
|
-
const matches = [
|
|
722
|
+
const matches = result.content[0].type === "text" ? Array.from(result.content[0].text.matchAll(toolCallRegex)) : [];
|
|
144
723
|
const function_call_tuples = matches.map((match) => match[1] || match[2]);
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
|
|
148
|
-
text: "",
|
|
149
|
-
toolCalls: function_call_tuples.map((toolCall) => {
|
|
150
|
-
const parsedToolCall = RJSON.parse(toolCall);
|
|
724
|
+
const tool_calls = function_call_tuples.map(
|
|
725
|
+
(toolCall) => {
|
|
726
|
+
const parsedToolCall = parse(toolCall);
|
|
151
727
|
const toolName = parsedToolCall.name;
|
|
152
728
|
const args = parsedToolCall.arguments;
|
|
153
729
|
return {
|
|
730
|
+
type: "tool-call",
|
|
154
731
|
toolCallType: "function",
|
|
155
732
|
toolCallId: generateId(),
|
|
156
733
|
toolName,
|
|
157
|
-
args:
|
|
734
|
+
args: JSON.stringify(args)
|
|
158
735
|
};
|
|
159
|
-
}
|
|
736
|
+
}
|
|
737
|
+
);
|
|
738
|
+
return {
|
|
739
|
+
...result,
|
|
740
|
+
// TODO: Return the remaining value after extracting the tool call tag.
|
|
741
|
+
content: [...tool_calls]
|
|
160
742
|
};
|
|
161
743
|
},
|
|
162
744
|
transformParams: async ({ params }) => {
|
|
@@ -195,9 +777,8 @@ function createToolMiddleware({
|
|
|
195
777
|
}
|
|
196
778
|
return message;
|
|
197
779
|
});
|
|
198
|
-
const originalToolDefinitions = params.mode.type === "regular" && params.mode.tools ? params.mode.tools : {};
|
|
199
780
|
const HermesPrompt = toolSystemPromptTemplate(
|
|
200
|
-
JSON.stringify(Object.entries(
|
|
781
|
+
JSON.stringify(Object.entries(params.tools || {}))
|
|
201
782
|
);
|
|
202
783
|
const toolSystemPrompt = processedPrompt[0].role === "system" ? [
|
|
203
784
|
{
|
|
@@ -214,11 +795,10 @@ function createToolMiddleware({
|
|
|
214
795
|
];
|
|
215
796
|
return {
|
|
216
797
|
...params,
|
|
217
|
-
|
|
218
|
-
|
|
219
|
-
|
|
220
|
-
|
|
221
|
-
prompt: toolSystemPrompt
|
|
798
|
+
prompt: toolSystemPrompt,
|
|
799
|
+
// set the mode back to regular and remove the default tools.
|
|
800
|
+
tools: [],
|
|
801
|
+
toolChoice: void 0
|
|
222
802
|
};
|
|
223
803
|
}
|
|
224
804
|
};
|
|
@@ -228,22 +808,38 @@ function createToolMiddleware({
|
|
|
228
808
|
var gemmaToolMiddleware = createToolMiddleware({
|
|
229
809
|
toolSystemPromptTemplate(tools) {
|
|
230
810
|
return `You have access to functions. If you decide to invoke any of the function(s),
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
|
|
235
|
-
|
|
236
|
-
|
|
811
|
+
you MUST put it in the format of
|
|
812
|
+
\`\`\`tool_call
|
|
813
|
+
{'name': <function-name>, 'arguments': <args-dict>}
|
|
814
|
+
\`\`\`
|
|
815
|
+
You SHOULD NOT include any other text in the response if you call a function
|
|
816
|
+
${tools}`;
|
|
237
817
|
},
|
|
238
818
|
toolCallTag: "```tool_call\n",
|
|
239
819
|
toolCallEndTag: "```",
|
|
240
820
|
toolResponseTag: "```tool_response\n",
|
|
241
821
|
toolResponseEndTag: "\n```"
|
|
242
822
|
});
|
|
243
|
-
var hermesToolMiddleware = createToolMiddleware({
|
|
823
|
+
var hermesToolMiddleware = createToolMiddleware({
|
|
824
|
+
toolSystemPromptTemplate(tools) {
|
|
825
|
+
return `You are a function calling AI model.
|
|
826
|
+
You are provided with function signatures within <tools></tools> XML tags.
|
|
827
|
+
You may call one or more functions to assist with the user query.
|
|
828
|
+
Don't make assumptions about what values to plug into functions.
|
|
829
|
+
Here are the available tools: <tools>${tools}</tools>
|
|
830
|
+
Use the following pydantic model json schema for each tool call you will make: {'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}
|
|
831
|
+
For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
|
|
832
|
+
<tool_call>
|
|
833
|
+
{'arguments': <args-dict>, 'name': <function-name>}
|
|
834
|
+
</tool_call>`;
|
|
835
|
+
},
|
|
836
|
+
toolCallTag: "<tool_call>",
|
|
837
|
+
toolCallEndTag: "</tool_call>",
|
|
838
|
+
toolResponseTag: "<tool_response>",
|
|
839
|
+
toolResponseEndTag: "</tool_response>"
|
|
840
|
+
});
|
|
244
841
|
export {
|
|
245
842
|
createToolMiddleware,
|
|
246
|
-
defaultTemplate,
|
|
247
843
|
gemmaToolMiddleware,
|
|
248
844
|
hermesToolMiddleware
|
|
249
845
|
};
|