@comfanion/usethis_search 3.0.0-dev.4 → 3.0.0-dev.6
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/index.ts +0 -2
- package/package.json +1 -2
- package/vectorizer/index.js +22 -3
- package/tools/read-interceptor.ts +0 -127
package/index.ts
CHANGED
|
@@ -2,7 +2,6 @@ import type { Plugin } from "@opencode-ai/plugin"
|
|
|
2
2
|
|
|
3
3
|
import search from "./tools/search"
|
|
4
4
|
import codeindex from "./tools/codeindex"
|
|
5
|
-
import readInterceptor from "./tools/read-interceptor"
|
|
6
5
|
import FileIndexerPlugin from "./file-indexer"
|
|
7
6
|
|
|
8
7
|
const UsethisSearchPlugin: Plugin = async ({ directory, client }) => {
|
|
@@ -19,7 +18,6 @@ const UsethisSearchPlugin: Plugin = async ({ directory, client }) => {
|
|
|
19
18
|
tool: {
|
|
20
19
|
search,
|
|
21
20
|
codeindex,
|
|
22
|
-
read: readInterceptor,
|
|
23
21
|
},
|
|
24
22
|
|
|
25
23
|
event: async (args: any) => {
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@comfanion/usethis_search",
|
|
3
|
-
"version": "3.0.0-dev.
|
|
3
|
+
"version": "3.0.0-dev.6",
|
|
4
4
|
"description": "OpenCode plugin: semantic search with graph-based context (v3: graph relations, 1-hop context, LSP + regex analyzers)",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./index.ts",
|
|
@@ -16,7 +16,6 @@
|
|
|
16
16
|
"file-indexer.ts",
|
|
17
17
|
"tools/search.ts",
|
|
18
18
|
"tools/codeindex.ts",
|
|
19
|
-
"tools/read-interceptor.ts",
|
|
20
19
|
"vectorizer/index.js",
|
|
21
20
|
"vectorizer/content-cleaner.ts",
|
|
22
21
|
"vectorizer/metadata-extractor.ts",
|
package/vectorizer/index.js
CHANGED
|
@@ -603,7 +603,13 @@ class CodebaseIndexer {
|
|
|
603
603
|
if (!tables.includes(tableName)) return null;
|
|
604
604
|
|
|
605
605
|
const table = await this.db.openTable(tableName);
|
|
606
|
-
|
|
606
|
+
let allRows;
|
|
607
|
+
try {
|
|
608
|
+
allRows = await table.search([0]).limit(100000).execute();
|
|
609
|
+
} catch (e) {
|
|
610
|
+
if (DEBUG) console.log("[vectorizer] BM25 index build failed (corrupted table?):", e.message);
|
|
611
|
+
return null;
|
|
612
|
+
}
|
|
607
613
|
|
|
608
614
|
if (allRows.length === 0) return null;
|
|
609
615
|
|
|
@@ -643,7 +649,14 @@ class CodebaseIndexer {
|
|
|
643
649
|
(options.tags && options.tags.length > 0);
|
|
644
650
|
const isHybrid = HYBRID_CONFIG.enabled || options.hybrid;
|
|
645
651
|
const fetchLimit = (hasFilters || isHybrid) ? Math.max(limit * 3, 50) : limit;
|
|
646
|
-
let results
|
|
652
|
+
let results;
|
|
653
|
+
try {
|
|
654
|
+
results = await table.search(queryEmbedding).limit(fetchLimit).execute();
|
|
655
|
+
} catch (e) {
|
|
656
|
+
// LanceDB schema error (e.g. missing vector column) — index is corrupted
|
|
657
|
+
if (DEBUG) console.log("[vectorizer] Vector search failed (corrupted index?):", e.message);
|
|
658
|
+
return [];
|
|
659
|
+
}
|
|
647
660
|
|
|
648
661
|
// ── Hybrid search ───────────────────────────────────────────────────────
|
|
649
662
|
if (HYBRID_CONFIG.enabled || options.hybrid) {
|
|
@@ -833,7 +846,13 @@ class CodebaseIndexer {
|
|
|
833
846
|
if (!tables.includes(tableName)) return null;
|
|
834
847
|
|
|
835
848
|
const table = await this.db.openTable(tableName);
|
|
836
|
-
|
|
849
|
+
let rows;
|
|
850
|
+
try {
|
|
851
|
+
rows = await table.search([0]).limit(100000).execute();
|
|
852
|
+
} catch (e) {
|
|
853
|
+
if (DEBUG) console.log("[vectorizer] Chunk cache build failed (corrupted table?):", e.message);
|
|
854
|
+
return null;
|
|
855
|
+
}
|
|
837
856
|
this._chunkCache = new Map();
|
|
838
857
|
for (const row of rows) {
|
|
839
858
|
if (row.chunk_id) {
|
|
@@ -1,127 +0,0 @@
|
|
|
1
|
-
import { tool } from "@opencode-ai/plugin"
|
|
2
|
-
import path from "path"
|
|
3
|
-
import fs from "fs/promises"
|
|
4
|
-
|
|
5
|
-
import { CodebaseIndexer } from "../vectorizer/index.js"
|
|
6
|
-
|
|
7
|
-
// FR-043: Logging for intercepted Read() calls
|
|
8
|
-
const DEBUG = process.env.DEBUG?.includes("vectorizer") || process.env.DEBUG === "*"
|
|
9
|
-
|
|
10
|
-
interface ReadLogEntry {
|
|
11
|
-
timestamp: number
|
|
12
|
-
filePath: string
|
|
13
|
-
relPath: string
|
|
14
|
-
chunksFound: number
|
|
15
|
-
relatedContextCount: number
|
|
16
|
-
durationMs: number
|
|
17
|
-
fallback: boolean
|
|
18
|
-
}
|
|
19
|
-
|
|
20
|
-
const LOG_MAX_ENTRIES = 500
|
|
21
|
-
|
|
22
|
-
/**
|
|
23
|
-
* Append a log entry to the Read() interception log file.
|
|
24
|
-
* Non-blocking, non-fatal — errors are silently ignored.
|
|
25
|
-
*/
|
|
26
|
-
async function logReadInterception(projectRoot: string, entry: ReadLogEntry): Promise<void> {
|
|
27
|
-
try {
|
|
28
|
-
const logPath = path.join(projectRoot, ".opencode", "vectors", "read-intercept.log.json")
|
|
29
|
-
await fs.mkdir(path.dirname(logPath), { recursive: true })
|
|
30
|
-
|
|
31
|
-
let entries: ReadLogEntry[] = []
|
|
32
|
-
try {
|
|
33
|
-
const raw = await fs.readFile(logPath, "utf-8")
|
|
34
|
-
entries = JSON.parse(raw)
|
|
35
|
-
} catch {
|
|
36
|
-
// file doesn't exist or is invalid — start fresh
|
|
37
|
-
}
|
|
38
|
-
|
|
39
|
-
entries.push(entry)
|
|
40
|
-
// Cap log size to avoid unbounded growth
|
|
41
|
-
if (entries.length > LOG_MAX_ENTRIES) {
|
|
42
|
-
entries = entries.slice(-LOG_MAX_ENTRIES)
|
|
43
|
-
}
|
|
44
|
-
|
|
45
|
-
await fs.writeFile(logPath, JSON.stringify(entries, null, 2), "utf-8")
|
|
46
|
-
} catch {
|
|
47
|
-
// non-fatal — logging must never break Read
|
|
48
|
-
}
|
|
49
|
-
}
|
|
50
|
-
|
|
51
|
-
export default tool({
|
|
52
|
-
description: `Read file with graph-aware context attachment. When available, this tool searches the file in the index and returns content + related context from the graph (imports, links, etc.).
|
|
53
|
-
|
|
54
|
-
Use this instead of the standard Read tool for better context awareness.`,
|
|
55
|
-
|
|
56
|
-
args: {
|
|
57
|
-
filePath: tool.schema.string().describe("Path to the file to read"),
|
|
58
|
-
},
|
|
59
|
-
|
|
60
|
-
async execute(args) {
|
|
61
|
-
const startTime = Date.now()
|
|
62
|
-
const projectRoot = process.cwd()
|
|
63
|
-
const filePath = path.isAbsolute(args.filePath) ? args.filePath : path.join(projectRoot, args.filePath)
|
|
64
|
-
|
|
65
|
-
const relPath = path.relative(projectRoot, filePath)
|
|
66
|
-
|
|
67
|
-
if (DEBUG) {
|
|
68
|
-
console.log(`[read-interceptor] Intercepted Read("${relPath}")`)
|
|
69
|
-
}
|
|
70
|
-
|
|
71
|
-
const indexer = await new CodebaseIndexer(projectRoot, "code").init()
|
|
72
|
-
const results = await indexer.search(relPath, 20, false, {})
|
|
73
|
-
const fileChunks = results.filter(r => r.file === relPath)
|
|
74
|
-
await indexer.unloadModel()
|
|
75
|
-
|
|
76
|
-
const allRelated = fileChunks
|
|
77
|
-
.flatMap(c => c.relatedContext || [])
|
|
78
|
-
.filter((r, i, arr) => arr.findIndex(x => x.chunk_id === r.chunk_id) === i)
|
|
79
|
-
|
|
80
|
-
const durationMs = Date.now() - startTime
|
|
81
|
-
const fallback = fileChunks.length === 0
|
|
82
|
-
|
|
83
|
-
// FR-043: Log the interception asynchronously (non-blocking)
|
|
84
|
-
logReadInterception(projectRoot, {
|
|
85
|
-
timestamp: startTime,
|
|
86
|
-
filePath: args.filePath,
|
|
87
|
-
relPath,
|
|
88
|
-
chunksFound: fileChunks.length,
|
|
89
|
-
relatedContextCount: allRelated.length,
|
|
90
|
-
durationMs,
|
|
91
|
-
fallback,
|
|
92
|
-
}).catch(() => {})
|
|
93
|
-
|
|
94
|
-
if (DEBUG) {
|
|
95
|
-
console.log(
|
|
96
|
-
`[read-interceptor] ${relPath}: ${fileChunks.length} chunks, ${allRelated.length} related, ${durationMs}ms${fallback ? " (fallback)" : ""}`
|
|
97
|
-
)
|
|
98
|
-
}
|
|
99
|
-
|
|
100
|
-
if (fallback) {
|
|
101
|
-
return `File "${relPath}" not indexed. Use original Read tool or run codeindex({ action: "reindex", index: "code" })`
|
|
102
|
-
}
|
|
103
|
-
|
|
104
|
-
let output = `## ${relPath}\n\n`
|
|
105
|
-
|
|
106
|
-
output += `### Content\n\n`
|
|
107
|
-
for (const chunk of fileChunks) {
|
|
108
|
-
output += chunk.content + "\n\n"
|
|
109
|
-
}
|
|
110
|
-
|
|
111
|
-
if (allRelated.length > 0) {
|
|
112
|
-
output += `### Related Context\n\n`
|
|
113
|
-
for (const rel of allRelated) {
|
|
114
|
-
const snippet = rel.content.length > 300
|
|
115
|
-
? rel.content.substring(0, 300) + "..."
|
|
116
|
-
: rel.content
|
|
117
|
-
output += `**${rel.file}** (${rel.relation})\n`
|
|
118
|
-
output += `\`\`\`\n${snippet}\n\`\`\`\n\n`
|
|
119
|
-
}
|
|
120
|
-
}
|
|
121
|
-
|
|
122
|
-
return output
|
|
123
|
-
},
|
|
124
|
-
})
|
|
125
|
-
|
|
126
|
-
// Export for testing
|
|
127
|
-
export { logReadInterception, ReadLogEntry }
|