feat: add transliteration support for internationalization in tool names

This commit is contained in:
suyao 2025-12-03 14:13:07 +08:00
parent b361d29940
commit b6efe41b86
No known key found for this signature in database
4 changed files with 138 additions and 40 deletions

View File

@ -363,6 +363,7 @@
"tar": "^7.4.3",
"tiny-pinyin": "^1.3.2",
"tokenx": "^1.1.0",
"transliteration": "^2.3.5",
"tsx": "^4.20.3",
"turndown-plugin-gfm": "^1.0.2",
"tw-animate-css": "^1.3.8",

View File

@ -61,12 +61,13 @@ describe('buildFunctionCallToolName', () => {
it('should replace invalid characters with underscores', () => {
const result = buildFunctionCallToolName('test@server', 'tool#name')
expect(result).not.toMatch(/[@#]/)
expect(result).toMatch(/^[\p{L}\p{N}_-]+$/u)
// Should only contain ASCII alphanumeric, underscore, dash, dot, colon
expect(result).toMatch(/^[a-zA-Z0-9_.\-:]+$/)
})
it('should ensure name starts with a letter or underscore', () => {
const result = buildFunctionCallToolName('123server', '456tool')
expect(result).toMatch(/^[\p{L}_]/u)
expect(result).toMatch(/^[a-zA-Z_]/)
})
it('should handle consecutive underscores/dashes', () => {
@ -130,7 +131,7 @@ describe('buildFunctionCallToolName', () => {
// Should still produce a valid unique suffix via fallback hash
expect(result).toBeTruthy()
expect(result.length).toBeLessThanOrEqual(63)
expect(result).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
expect(result).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
// Should have a suffix (underscore followed by something)
expect(result).toMatch(/_[a-z0-9]+$/)
})
@ -177,9 +178,9 @@ describe('buildFunctionCallToolName', () => {
// Should be different
expect(tool1).not.toBe(tool2)
// Both should be valid identifiers
expect(tool1).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
expect(tool2).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
// Both should be valid AI model tool names (ASCII only)
expect(tool1).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
expect(tool2).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
// Both should be <= 63 chars
expect(tool1.length).toBeLessThanOrEqual(63)
@ -194,12 +195,14 @@ describe('buildFunctionCallToolName', () => {
})
})
describe('internationalization support', () => {
it('should preserve Chinese characters in tool names', () => {
describe('internationalization support (CJK to ASCII transliteration)', () => {
it('should convert Chinese characters to pinyin', () => {
const result = buildFunctionCallToolName('ocr', '行驶证OCR_轻盈版')
expect(result).toContain('行驶证')
expect(result).toContain('OCR')
expect(result).toContain('轻盈版')
// Chinese characters should be transliterated to pinyin
expect(result).not.toMatch(/[\u4e00-\u9fff]/) // No Chinese characters
expect(result).toContain('ocr') // OCR is lowercased
// Should only contain ASCII characters (lowercase)
expect(result).toMatch(/^[a-z_][a-z0-9_.\-:]*$/)
})
it('should distinguish between different Chinese OCR tools', () => {
@ -210,47 +213,84 @@ describe('buildFunctionCallToolName', () => {
buildFunctionCallToolName('ocr', '身份证OCR')
]
// All tools should be unique
// All tools should be unique (pinyin transliterations are different)
const uniqueTools = new Set(tools)
expect(uniqueTools.size).toBe(4)
// Verify each tool contains its distinctive Chinese characters
expect(tools[0]).toContain('行驶证')
expect(tools[1]).toContain('营业执照')
expect(tools[2]).toContain('车牌')
expect(tools[3]).toContain('身份证')
// All should be ASCII-only valid tool names
tools.forEach((tool) => {
expect(tool).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
expect(tool).not.toMatch(/[\u4e00-\u9fff]/) // No Chinese characters
})
// Verify they contain transliterated pinyin (with underscores between characters)
// 行驶证 = xing_shi_zheng, 营业执照 = ying_ye_zhi_zhao, 车牌 = che_pai, 身份证 = shen_fen_zheng
expect(tools[0]).toContain('xing_shi_zheng')
expect(tools[1]).toContain('ying_ye_zhi_zhao')
expect(tools[2]).toContain('che_pai')
expect(tools[3]).toContain('shen_fen_zheng')
})
it('should handle Japanese characters', () => {
it('should handle Japanese characters with base36 encoding', () => {
const result = buildFunctionCallToolName('server', 'ユーザー検索')
expect(result).toContain('ユーザー検索')
expect(result).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
// Should be ASCII-only
expect(result).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
// Should not contain original Japanese characters
expect(result).not.toMatch(/[\u3040-\u309f\u30a0-\u30ff]/)
})
it('should handle Korean characters', () => {
it('should handle Korean characters with base36 encoding', () => {
const result = buildFunctionCallToolName('server', '사용자검색')
expect(result).toContain('사용자검색')
expect(result).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
// Should be ASCII-only
expect(result).toMatch(/^[a-zA-Z_][a-zA-Z0-9_.\-:]*$/)
// Should not contain original Korean characters
expect(result).not.toMatch(/[\uac00-\ud7af]/)
})
it('should handle mixed language tool names', () => {
const result = buildFunctionCallToolName('api', 'search用户by名称')
// ASCII parts should be preserved (lowercased)
expect(result).toContain('search')
expect(result).toContain('用户')
expect(result).toContain('by')
expect(result).toContain('名称')
expect(result).toMatch(/^[\p{L}_][\p{L}\p{N}_-]*$/u)
// Chinese parts should be transliterated (用户 = yong_hu, 名称 = ming_cheng)
expect(result).toContain('yong_hu')
expect(result).toContain('ming_cheng')
// Final result should be ASCII-only (lowercase)
expect(result).toMatch(/^[a-z_][a-z0-9_.\-:]*$/)
})
it('should replace only control characters and special symbols, not Unicode letters', () => {
it('should transliterate Chinese and replace special symbols', () => {
const result = buildFunctionCallToolName('test', '文件@上传#工具')
// @ and # should be replaced with underscores
expect(result).not.toContain('@')
expect(result).not.toContain('#')
// Chinese characters should be preserved
expect(result).toContain('文件')
expect(result).toContain('上传')
expect(result).toContain('工具')
// Chinese characters should be transliterated
// 文件 = wen_jian, 上传 = shang_chuan, 工具 = gong_ju
expect(result).toContain('wen_jian')
expect(result).toContain('shang_chuan')
expect(result).toContain('gong_ju')
// Should be ASCII-only (lowercase)
expect(result).toMatch(/^[a-z_][a-z0-9_.\-:]*$/)
})
it('should produce AI model compatible tool names', () => {
const testCases = [
'行驶证OCR',
'营业执照识别',
'get用户info',
'文件@处理',
'数据分析_v2'
]
testCases.forEach((testCase) => {
const result = buildFunctionCallToolName('server', testCase)
// Must start with letter or underscore
expect(result).toMatch(/^[a-zA-Z_]/)
// Must only contain a-z, A-Z, 0-9, _, -, ., :
expect(result).toMatch(/^[a-zA-Z0-9_.\-:]+$/)
// Must be <= 64 characters
expect(result.length).toBeLessThanOrEqual(64)
})
})
})
})

View File

@ -1,6 +1,51 @@
import { transliterate } from 'transliteration'
/**
* Transliterate non-ASCII characters to ASCII equivalents
* - Chinese Pinyin (e.g., xingshizheng)
* - Japanese Romaji (e.g., yūzā)
* - Korean Romanization (e.g., sayongja)
* - Other special characters underscores
*/
/**
* Transliterates non-ASCII text (including CJK characters) to ASCII-compatible format.
*
* Converts input text to lowercase ASCII representation, replacing spaces with underscores
* and removing special characters. Unknown or special characters are replaced with underscores.
*
* @param text - The input string to transliterate, may contain Unicode characters including CJK
* @returns A lowercase ASCII string with spaces converted to underscores and special characters removed,
* preserving only alphanumeric characters, underscores, dots, hyphens, and colons
*
* @example
* ```typescript
* transliterateToAscii("Hello World") // returns "hello_world"
* transliterateToAscii("你好世界") // returns transliterated version with underscores
* transliterateToAscii("Café-123") // returns "cafe_123"
* ```
*/
function transliterateToAscii(text: string): string {
// Use transliteration library which supports CJK (Chinese, Japanese, Korean)
const result = transliterate(text, {
// Unknown/special characters become underscores
unknown: '_',
ignore: []
})
// Convert to lowercase, remove spaces, and clean up special chars
return result
.toLowerCase()
.replace(/\s+/g, '_')
.replace(/[^a-z0-9_.\-:]/g, '_')
}
export function buildFunctionCallToolName(serverName: string, toolName: string, serverId?: string) {
const sanitizedServer = serverName.trim().replace(/-/g, '_')
const sanitizedTool = toolName.trim().replace(/-/g, '_')
// First, transliterate non-ASCII characters to ASCII
const transliteratedServer = transliterateToAscii(serverName.trim())
const transliteratedTool = transliterateToAscii(toolName.trim())
const sanitizedServer = transliteratedServer.replace(/-/g, '_')
const sanitizedTool = transliteratedTool.replace(/-/g, '_')
// Calculate suffix first to reserve space for it
// Suffix format: "_" + 6 alphanumeric chars = 7 chars total
@ -26,14 +71,13 @@ export function buildFunctionCallToolName(serverName: string, toolName: string,
name = `${sanitizedServer.slice(0, 7) || ''}-${sanitizedTool || ''}`
}
// Replace invalid characters with underscores or dashes
// Keep Unicode letters (\p{L}), Unicode numbers (\p{N}), underscores and dashes
// This supports international characters (Chinese, Japanese, Korean, etc.)
name = name.replace(/[^\p{L}\p{N}_-]/gu, '_')
// Replace invalid characters with underscores
// Keep only a-z, A-Z, 0-9, underscores, dashes, dots, colons (AI model compatible)
name = name.replace(/[^a-zA-Z0-9_.\-:]/g, '_')
// Ensure name starts with a letter or underscore (supports Unicode letters)
if (!/^[\p{L}_]/u.test(name)) {
name = `tool-${name}`
// Ensure name starts with a letter or underscore (AI model requirement)
if (!/^[a-zA-Z_]/.test(name)) {
name = `tool_${name}`
}
// Remove consecutive underscores/dashes (optional improvement)

View File

@ -10285,6 +10285,7 @@ __metadata:
tesseract.js: "patch:tesseract.js@npm%3A6.0.1#~/.yarn/patches/tesseract.js-npm-6.0.1-2562a7e46d.patch"
tiny-pinyin: "npm:^1.3.2"
tokenx: "npm:^1.1.0"
transliteration: "npm:^2.3.5"
tsx: "npm:^4.20.3"
turndown: "npm:7.2.0"
turndown-plugin-gfm: "npm:^1.0.2"
@ -24607,6 +24608,18 @@ __metadata:
languageName: node
linkType: hard
"transliteration@npm:^2.3.5":
version: 2.3.5
resolution: "transliteration@npm:2.3.5"
dependencies:
yargs: "npm:^17.5.1"
bin:
slugify: dist/bin/slugify
transliterate: dist/bin/transliterate
checksum: 10c0/68397225c2ca59b8e33206c65f905724e86b64460cbf90576d352dc2366e763ded97e2c7b8b1f140fb36a565d61a97c51080df9fa638e6b1769f6cb24f383756
languageName: node
linkType: hard
"tree-kill@npm:1.2.2, tree-kill@npm:^1.2.2":
version: 1.2.2
resolution: "tree-kill@npm:1.2.2"