gitnexus 1.6.10-rc.5 → 1.6.10-rc.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/dist/cli/doctor.js +10 -0
- package/dist/core/lbug/extension-load-error.d.ts +67 -0
- package/dist/core/lbug/extension-load-error.js +320 -0
- package/dist/core/lbug/extension-loader.d.ts +7 -0
- package/dist/core/lbug/extension-loader.js +9 -1
- package/dist/core/run-analyze.js +27 -7
- package/dist/core/search/fts-indexes.js +11 -1
- package/package.json +1 -1
- package/scripts/cross-platform-tests.ts +5 -0
- package/scripts/install-duckdb-extension.mjs +3 -1
package/dist/cli/doctor.js
CHANGED
|
@@ -5,6 +5,7 @@ import { getLocalEmbeddingRuntimeBlocker, localEmbeddingPrefixUnloadableMessage,
|
|
|
5
5
|
import { isPrefixRuntimeLoadable, resolveEmbeddingRuntime, } from '../core/embeddings/runtime-install.js';
|
|
6
6
|
import { cudaRedirectDoctorStatus } from '../core/embeddings/onnxruntime-node-resolver.js';
|
|
7
7
|
import { checkLbugNative, probeFtsExtensionLoad } from '../core/lbug/native-check.js';
|
|
8
|
+
import { diagnoseExtensionLoad } from '../core/lbug/extension-load-error.js';
|
|
8
9
|
import { getExtensionInstallPolicy } from '../core/lbug/extension-loader.js';
|
|
9
10
|
import { t } from './i18n/index.js';
|
|
10
11
|
function isCombiningMark(codePoint) {
|
|
@@ -116,6 +117,15 @@ export const doctorCommand = async () => {
|
|
|
116
117
|
console.log(` ${label('doctor.labels.fullTextSearch', 18)}${ftsProbe.loaded ? 'available' : 'unavailable'}`);
|
|
117
118
|
if (!ftsProbe.loaded && ftsProbe.reason) {
|
|
118
119
|
console.log(` ${padDisplayEnd('', 18)}${ftsProbe.reason}`);
|
|
120
|
+
// Add an actionable remedy for recognized failure classes (#2374). The
|
|
121
|
+
// Windows missing-dependency case is the point of this: the raw error 126
|
|
122
|
+
// ("specified module could not be found") is opaque, so name the fix (VC++
|
|
123
|
+
// redist, then OpenSSL) instead of leaving the user to reinstall in vain.
|
|
124
|
+
// `unknown`'s remedy is "run doctor", which would be circular here.
|
|
125
|
+
const { kind, remedy } = diagnoseExtensionLoad(ftsProbe.reason);
|
|
126
|
+
if (kind !== 'unknown') {
|
|
127
|
+
console.log(` ${padDisplayEnd('', 18)}${remedy}`);
|
|
128
|
+
}
|
|
119
129
|
}
|
|
120
130
|
console.log(` ${label('doctor.labels.vectorIndex', 18)}${capabilities.vector}`);
|
|
121
131
|
console.log(` ${label('doctor.labels.semanticMode', 18)}${capabilities.semanticMode}`);
|
|
@@ -0,0 +1,67 @@
|
|
|
1
|
+
export type ExtensionLoadErrorKind = 'missing_file' | 'corrupt_file' | 'missing_dependency' | 'unknown';
|
|
2
|
+
export interface ExtensionLoadDiagnosis {
|
|
3
|
+
readonly kind: ExtensionLoadErrorKind;
|
|
4
|
+
/** Actionable, literal-English remedy suited to the class. */
|
|
5
|
+
readonly remedy: string;
|
|
6
|
+
}
|
|
7
|
+
/**
|
|
8
|
+
* On-disk file corruption / wrong-platform. FORCE INSTALL re-downloads.
|
|
9
|
+
* Kept byte-identical to `FILE_CORRUPTION_SIGNATURES` in
|
|
10
|
+
* scripts/install-duckdb-extension.mjs (that `.mjs` cannot import this `.ts`;
|
|
11
|
+
* the duplication is deliberate — the two serve different call sites). Note
|
|
12
|
+
* `/not a valid/i` already covers Windows error 193 ("is not a valid Win32
|
|
13
|
+
* application"), so a truncated Windows download is caught here, before the
|
|
14
|
+
* missing-dependency branch.
|
|
15
|
+
*/
|
|
16
|
+
export declare const FILE_CORRUPTION_SIGNATURES: readonly RegExp[];
|
|
17
|
+
/**
|
|
18
|
+
* Classify a collapsed LadybugDB LOAD error. Order is most-specific-first and is
|
|
19
|
+
* load-bearing: corrupt-file is tested before missing-dependency so a truncated
|
|
20
|
+
* Windows download (error 193, matched by `/not a valid/i`) routes to
|
|
21
|
+
* FORCE-reinstall rather than to the runtime-install remedy.
|
|
22
|
+
*/
|
|
23
|
+
export declare function classifyExtensionLoadError(reason: string | undefined | null): ExtensionLoadDiagnosis;
|
|
24
|
+
/** Well-formedness of the extension binary for the host platform + arch. */
|
|
25
|
+
export type ExtensionBinaryState = 'absent' | 'corrupt' | 'valid' | 'indeterminate';
|
|
26
|
+
/**
|
|
27
|
+
* Pull the extension file path out of lbug's load error. lbug's wrapper is
|
|
28
|
+
* English regardless of OS language — `Failed to load library: {path} which is
|
|
29
|
+
* needed by extension: {name}` (real lbug), or the quoted `Failed to load
|
|
30
|
+
* library '{path}': {reason}` variant — so the path is recoverable in any locale.
|
|
31
|
+
* Only paths ending in `.lbug_extension` are accepted, so a regex misfire can
|
|
32
|
+
* never point the inspector at an arbitrary file.
|
|
33
|
+
*/
|
|
34
|
+
export declare function extractExtensionPath(reason: string | undefined | null): string | null;
|
|
35
|
+
/**
|
|
36
|
+
* A structural verdict on a binary header. `indeterminate` means the probe could
|
|
37
|
+
* not prove validity OR corruption from what it read (e.g. the PE header sits past
|
|
38
|
+
* the BINARY_HEADER_BYTES window) — the caller defers to the string classifier
|
|
39
|
+
* rather than assert a false verdict.
|
|
40
|
+
*/
|
|
41
|
+
type HeaderVerdict = 'valid' | 'corrupt' | 'indeterminate';
|
|
42
|
+
/**
|
|
43
|
+
* Decide whether a binary header is a well-formed shared library for the given
|
|
44
|
+
* platform + architecture — using only the file's structure, no localized text.
|
|
45
|
+
* Pure and injectable (platform/arch as params) so every format+arch combination
|
|
46
|
+
* is unit-testable regardless of the host it runs on.
|
|
47
|
+
*/
|
|
48
|
+
export declare function classifyBinaryHeader(buf: Buffer, bytesRead: number, platform: NodeJS.Platform, arch: string): HeaderVerdict;
|
|
49
|
+
/**
|
|
50
|
+
* Best-effort language-independent inspection of the extension file. Reads the
|
|
51
|
+
* header and classifies it; never throws — a missing file is `absent`, an
|
|
52
|
+
* unreadable one is `indeterminate`.
|
|
53
|
+
*/
|
|
54
|
+
export declare function inspectExtensionBinary(extensionPath: string | null | undefined): ExtensionBinaryState;
|
|
55
|
+
/**
|
|
56
|
+
* Diagnose a LadybugDB load failure, preferring a LANGUAGE-INDEPENDENT structural
|
|
57
|
+
* check of the extension binary over the localized error text:
|
|
58
|
+
* - file absent → missing_file
|
|
59
|
+
* - present but malformed → corrupt_file (bad magic / wrong architecture)
|
|
60
|
+
* - present and well-formed → missing_dependency (a valid binary the loader rejected)
|
|
61
|
+
* The path comes from lbug's own English wrapper, so this holds in any OS display
|
|
62
|
+
* language. When the file cannot be located or read, it falls back to the string
|
|
63
|
+
* classifier (which still carries the language-independent hedged fallback). This
|
|
64
|
+
* is the entry point every surface should call.
|
|
65
|
+
*/
|
|
66
|
+
export declare function diagnoseExtensionLoad(reason: string | undefined | null): ExtensionLoadDiagnosis;
|
|
67
|
+
export {};
|
|
@@ -0,0 +1,320 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Classify a LadybugDB `LOAD EXTENSION` failure into one of four actionable
|
|
3
|
+
* classes and produce an accurate, literal-English remedy.
|
|
4
|
+
*
|
|
5
|
+
* Background (#2374): PR #2375 made the real LadybugDB LOAD error visible
|
|
6
|
+
* (instead of a false "not pre-installed" message). The rc.4 reproduction then
|
|
7
|
+
* showed the remaining defect — on Windows the extension file downloads and
|
|
8
|
+
* INSTALLs fine, but `LoadLibrary` fails with error 126 ("the specified module
|
|
9
|
+
* could not be found" / `找不到指定的模块`) because the extension dynamically
|
|
10
|
+
* imports OpenSSL 3 / MSVC 14 DLLs that ship nowhere. For that class, telling
|
|
11
|
+
* the user to reinstall/redownload is wrong — the file is fine; a *runtime
|
|
12
|
+
* dependency* is missing. This module decides which class an error is so each
|
|
13
|
+
* surface (doctor, --repair-fts, the analyze degrade warning, and
|
|
14
|
+
* ftsDegradedWarning) can emit the right remedy instead of a one-size-fits-all
|
|
15
|
+
* "reinstall over the network".
|
|
16
|
+
*
|
|
17
|
+
* `classifyExtensionLoadError` is pure string logic — no `@ladybugdb/core`
|
|
18
|
+
* import, no filesystem — which keeps `native-check.ts` free of a static lbug
|
|
19
|
+
* dependency. `diagnoseExtensionLoad` layers a LANGUAGE-INDEPENDENT structural
|
|
20
|
+
* check on top: it pulls the extension's file path out of lbug's own (English)
|
|
21
|
+
* wrapper and inspects the binary header directly (PE/ELF/Mach-O magic +
|
|
22
|
+
* architecture), so corrupt-vs-valid is decided by the file itself, not by the
|
|
23
|
+
* localized OS error tail. It reads the file (node:fs core module only, still no
|
|
24
|
+
* lbug) and never throws — any read failure degrades to the string classifier.
|
|
25
|
+
*/
|
|
26
|
+
import { closeSync, openSync, readSync } from 'node:fs';
|
|
27
|
+
/** LadybugDB says the extension file was never installed. INSTALL can heal it. */
|
|
28
|
+
const MISSING_FILE_SIGNATURES = [
|
|
29
|
+
/has not been installed/i,
|
|
30
|
+
/not been installed/i,
|
|
31
|
+
];
|
|
32
|
+
/**
|
|
33
|
+
* On-disk file corruption / wrong-platform. FORCE INSTALL re-downloads.
|
|
34
|
+
* Kept byte-identical to `FILE_CORRUPTION_SIGNATURES` in
|
|
35
|
+
* scripts/install-duckdb-extension.mjs (that `.mjs` cannot import this `.ts`;
|
|
36
|
+
* the duplication is deliberate — the two serve different call sites). Note
|
|
37
|
+
* `/not a valid/i` already covers Windows error 193 ("is not a valid Win32
|
|
38
|
+
* application"), so a truncated Windows download is caught here, before the
|
|
39
|
+
* missing-dependency branch.
|
|
40
|
+
*/
|
|
41
|
+
// Exported so a parity test can assert this stays byte-identical to the copy in
|
|
42
|
+
// scripts/install-duckdb-extension.mjs (that `.mjs` cannot import this `.ts`), #2383 F5b.
|
|
43
|
+
export const FILE_CORRUPTION_SIGNATURES = [
|
|
44
|
+
/invalid elf/i,
|
|
45
|
+
/file too short/i,
|
|
46
|
+
/not a valid/i,
|
|
47
|
+
/bad magic/i,
|
|
48
|
+
/wrong architecture/i,
|
|
49
|
+
/mach-o/i,
|
|
50
|
+
/truncat/i,
|
|
51
|
+
];
|
|
52
|
+
/**
|
|
53
|
+
* A *transitive dependency* of the extension is missing — the file loaded far
|
|
54
|
+
* enough to be found, but a library it needs is absent. Reinstalling the
|
|
55
|
+
* extension is a no-op for this class.
|
|
56
|
+
*
|
|
57
|
+
* WINDOWS CATCH-ALL GUARD (adversarial review): LadybugDB wraps *every* Windows
|
|
58
|
+
* load failure in `Failed to load library … which is needed by extension`, so
|
|
59
|
+
* that generic wrapper must NOT be sufficient — otherwise error 127 (wrong
|
|
60
|
+
* OpenSSL minor / unresolved procedure), 5 (AV/permission lock), and 1114
|
|
61
|
+
* (dependency DllMain failure) would all be mislabeled `missing_dependency` and
|
|
62
|
+
* told to install a runtime, the opposite of their real fix. We key strictly on
|
|
63
|
+
* the specific error-126 tail. Linux/macOS loaders name the missing library
|
|
64
|
+
* directly, so their signals are unambiguous.
|
|
65
|
+
*
|
|
66
|
+
* Localized Windows tails we do not enumerate (French, German, Japanese, …) and
|
|
67
|
+
* mojibake renderings of the Chinese text won't match here — but they still
|
|
68
|
+
* carry lbug's language-independent `Failed to load library` wrapper, so they
|
|
69
|
+
* are caught by the hedged fallback (LOAD_FAILURE_WRAPPER) with a non-committal
|
|
70
|
+
* remedy, never a wrong confident "reinstall" instruction.
|
|
71
|
+
*/
|
|
72
|
+
const WINDOWS_MISSING_DEPENDENCY_SIGNATURES = [
|
|
73
|
+
/找不到指定的模块/,
|
|
74
|
+
/specified module could not be found/i,
|
|
75
|
+
];
|
|
76
|
+
const POSIX_MISSING_DEPENDENCY_SIGNATURES = [
|
|
77
|
+
/cannot open shared object file/i, // Linux ld.so
|
|
78
|
+
/image not found/i, // macOS dyld
|
|
79
|
+
/library not loaded/i, // macOS dyld
|
|
80
|
+
];
|
|
81
|
+
/**
|
|
82
|
+
* LadybugDB's own English wrapper for a dlopen/LoadLibrary failure
|
|
83
|
+
* (extension.cpp: `Failed to load library: {path} which is needed by extension:
|
|
84
|
+
* {name}`). It is emitted for EVERY extension load failure regardless of the OS
|
|
85
|
+
* display language — the only localized part is the OS-error tail after it. So
|
|
86
|
+
* it is the language-independent fallback signal once the specific tails miss: a
|
|
87
|
+
* French/German/Japanese Windows 126 has a localized tail we cannot enumerate,
|
|
88
|
+
* but it still carries this wrapper. See HEDGED_LOAD_FAILURE_REMEDY.
|
|
89
|
+
*/
|
|
90
|
+
const LOAD_FAILURE_WRAPPER = /failed to load library/i;
|
|
91
|
+
const MISSING_FILE_REMEDY = 'The FTS extension is not installed. Re-run with network access and ' +
|
|
92
|
+
'GITNEXUS_LBUG_EXTENSION_INSTALL=auto (or `gitnexus analyze --repair-fts`) to download it.';
|
|
93
|
+
const CORRUPT_FILE_REMEDY = 'The FTS extension file is present but unreadable (corrupt, truncated, or built for another ' +
|
|
94
|
+
'platform). Re-download it with network access and GITNEXUS_LBUG_EXTENSION_INSTALL=auto ' +
|
|
95
|
+
'(`gitnexus analyze --repair-fts`).';
|
|
96
|
+
// Single source of truth for the VC++ runtime-install pointer, shared by the
|
|
97
|
+
// Windows-126 and structural missing-dependency remedies so the name/URL cannot
|
|
98
|
+
// drift between them (#2383 F5).
|
|
99
|
+
const VC_REDIST_INSTALL_HINT = 'the Microsoft Visual C++ 2015-2022 Redistributable (x64) from ' +
|
|
100
|
+
'https://aka.ms/vs/17/release/vc_redist.x64.exe';
|
|
101
|
+
// MSVC-first per DuckDB's canonical answer for this exact error; OpenSSL second.
|
|
102
|
+
const WINDOWS_MISSING_DEPENDENCY_REMEDY = 'The FTS extension is present but a required runtime library is missing (Windows error 126). ' +
|
|
103
|
+
'Reinstalling the extension will NOT help. Install ' +
|
|
104
|
+
VC_REDIST_INSTALL_HINT +
|
|
105
|
+
'; if the error persists, the extension also needs OpenSSL 3 ' +
|
|
106
|
+
'(libcrypto-3-x64.dll / libssl-3-x64.dll) on the DLL search path.';
|
|
107
|
+
const POSIX_MISSING_DEPENDENCY_REMEDY = 'The FTS extension is present but a shared library it depends on could not be loaded (named in ' +
|
|
108
|
+
'the error above). Reinstalling the extension will NOT help — install that library or add it to ' +
|
|
109
|
+
'your loader search path.';
|
|
110
|
+
// Language-independent fallback: we know the extension failed to load, but the
|
|
111
|
+
// OS-error tail is in a locale we did not enumerate, so we cannot say which class
|
|
112
|
+
// it is. Hedge honestly — point at the user's own localized error and give both
|
|
113
|
+
// branches — rather than confidently prescribing the wrong single fix. The clean
|
|
114
|
+
// long-term fix is upstream: have LadybugDB include the numeric GetLastError/errno
|
|
115
|
+
// in the message (as it already does elsewhere), so this becomes a code match.
|
|
116
|
+
const HEDGED_LOAD_FAILURE_REMEDY = 'The FTS extension file was found but could not be loaded — see the "Error:" text above (shown ' +
|
|
117
|
+
"in your system's language). Reinstalling usually will not help. If it names a missing module or " +
|
|
118
|
+
'library, install the required runtime (on Windows: the Microsoft Visual C++ 2015-2022 ' +
|
|
119
|
+
'Redistributable x64 and OpenSSL 3); if it names a corrupt or invalid file, run ' +
|
|
120
|
+
'`gitnexus analyze --repair-fts` to re-download.';
|
|
121
|
+
const UNKNOWN_REMEDY = 'The FTS extension failed to load for an unrecognized reason. Run `gitnexus doctor` for live ' +
|
|
122
|
+
'FTS status and verify the extension file and platform.';
|
|
123
|
+
const matchesAny = (reason, signatures) => signatures.some((re) => re.test(reason));
|
|
124
|
+
/**
|
|
125
|
+
* Classify a collapsed LadybugDB LOAD error. Order is most-specific-first and is
|
|
126
|
+
* load-bearing: corrupt-file is tested before missing-dependency so a truncated
|
|
127
|
+
* Windows download (error 193, matched by `/not a valid/i`) routes to
|
|
128
|
+
* FORCE-reinstall rather than to the runtime-install remedy.
|
|
129
|
+
*/
|
|
130
|
+
export function classifyExtensionLoadError(reason) {
|
|
131
|
+
const text = reason ?? '';
|
|
132
|
+
if (matchesAny(text, MISSING_FILE_SIGNATURES)) {
|
|
133
|
+
return { kind: 'missing_file', remedy: MISSING_FILE_REMEDY };
|
|
134
|
+
}
|
|
135
|
+
if (matchesAny(text, FILE_CORRUPTION_SIGNATURES)) {
|
|
136
|
+
return { kind: 'corrupt_file', remedy: CORRUPT_FILE_REMEDY };
|
|
137
|
+
}
|
|
138
|
+
if (matchesAny(text, WINDOWS_MISSING_DEPENDENCY_SIGNATURES)) {
|
|
139
|
+
return { kind: 'missing_dependency', remedy: WINDOWS_MISSING_DEPENDENCY_REMEDY };
|
|
140
|
+
}
|
|
141
|
+
if (matchesAny(text, POSIX_MISSING_DEPENDENCY_SIGNATURES)) {
|
|
142
|
+
return { kind: 'missing_dependency', remedy: POSIX_MISSING_DEPENDENCY_REMEDY };
|
|
143
|
+
}
|
|
144
|
+
// Language-independent fallback: the extension demonstrably failed to load
|
|
145
|
+
// (lbug's English wrapper is present) but the localized OS tail matched no
|
|
146
|
+
// specific class. Treat as a dependency/runtime load failure with a hedged
|
|
147
|
+
// remedy — strictly better than the generic `unknown` for non-English hosts,
|
|
148
|
+
// and it never prescribes the wrong fix.
|
|
149
|
+
if (LOAD_FAILURE_WRAPPER.test(text)) {
|
|
150
|
+
return { kind: 'missing_dependency', remedy: HEDGED_LOAD_FAILURE_REMEDY };
|
|
151
|
+
}
|
|
152
|
+
return { kind: 'unknown', remedy: UNKNOWN_REMEDY };
|
|
153
|
+
}
|
|
154
|
+
const STRUCTURAL_MISSING_DEPENDENCY_REMEDY = 'The FTS extension file is valid, so the failure is a missing or incompatible runtime dependency, ' +
|
|
155
|
+
'not the extension itself — reinstalling will NOT help. On Windows, install ' +
|
|
156
|
+
VC_REDIST_INSTALL_HINT +
|
|
157
|
+
' and ensure OpenSSL 3 is available; on Linux/macOS install the shared library named in the error above.';
|
|
158
|
+
/**
|
|
159
|
+
* Pull the extension file path out of lbug's load error. lbug's wrapper is
|
|
160
|
+
* English regardless of OS language — `Failed to load library: {path} which is
|
|
161
|
+
* needed by extension: {name}` (real lbug), or the quoted `Failed to load
|
|
162
|
+
* library '{path}': {reason}` variant — so the path is recoverable in any locale.
|
|
163
|
+
* Only paths ending in `.lbug_extension` are accepted, so a regex misfire can
|
|
164
|
+
* never point the inspector at an arbitrary file.
|
|
165
|
+
*/
|
|
166
|
+
export function extractExtensionPath(reason) {
|
|
167
|
+
const text = reason ?? '';
|
|
168
|
+
const m = /failed to load library:?\s*['"]?(.+?\.lbug_extension)/i.exec(text);
|
|
169
|
+
const path = m?.[1]?.trim();
|
|
170
|
+
return path && path.length > 0 ? path : null;
|
|
171
|
+
}
|
|
172
|
+
/** Node `process.arch` → PE `Machine`. Undefined for arches we don't map. */
|
|
173
|
+
const PE_MACHINE = { x64: 0x8664, arm64: 0xaa64 };
|
|
174
|
+
/** Node `process.arch` → ELF `e_machine`. */
|
|
175
|
+
const ELF_MACHINE = { x64: 0x3e, arm64: 0xb7 };
|
|
176
|
+
/** Node `process.arch` → Mach-O `cputype`. */
|
|
177
|
+
const MACHO_CPUTYPE = { x64: 0x01000007, arm64: 0x0100000c };
|
|
178
|
+
function classifyPE(buf, bytesRead, arch) {
|
|
179
|
+
if (bytesRead < 0x40 || buf[0] !== 0x4d || buf[1] !== 0x5a)
|
|
180
|
+
return 'corrupt'; // 'MZ'
|
|
181
|
+
const peOffset = buf.readUInt32LE(0x3c);
|
|
182
|
+
// The PE header (e_lfanew) points beyond what we read. A large-DOS-stub VALID PE
|
|
183
|
+
// and a garbage e_lfanew are indistinguishable from here, so don't claim 'corrupt'
|
|
184
|
+
// — defer to the loader's own report (#2383 F1-secondary).
|
|
185
|
+
if (peOffset + 6 > bytesRead)
|
|
186
|
+
return 'indeterminate';
|
|
187
|
+
const isPE = buf[peOffset] === 0x50 &&
|
|
188
|
+
buf[peOffset + 1] === 0x45 &&
|
|
189
|
+
buf[peOffset + 2] === 0 &&
|
|
190
|
+
buf[peOffset + 3] === 0;
|
|
191
|
+
if (!isPE)
|
|
192
|
+
return 'corrupt';
|
|
193
|
+
const expected = PE_MACHINE[arch];
|
|
194
|
+
if (expected === undefined)
|
|
195
|
+
return 'valid'; // arch we don't map: don't claim corrupt
|
|
196
|
+
return buf.readUInt16LE(peOffset + 4) === expected ? 'valid' : 'corrupt';
|
|
197
|
+
}
|
|
198
|
+
function classifyELF(buf, bytesRead, arch) {
|
|
199
|
+
if (bytesRead < 20)
|
|
200
|
+
return 'corrupt';
|
|
201
|
+
if (buf[0] !== 0x7f || buf[1] !== 0x45 || buf[2] !== 0x4c || buf[3] !== 0x46)
|
|
202
|
+
return 'corrupt'; // 0x7F ELF
|
|
203
|
+
const littleEndian = buf[5] === 1; // EI_DATA
|
|
204
|
+
const eMachine = littleEndian ? buf.readUInt16LE(18) : buf.readUInt16BE(18);
|
|
205
|
+
const expected = ELF_MACHINE[arch];
|
|
206
|
+
if (expected === undefined)
|
|
207
|
+
return 'valid';
|
|
208
|
+
return eMachine === expected ? 'valid' : 'corrupt';
|
|
209
|
+
}
|
|
210
|
+
function classifyMachO(buf, bytesRead, arch) {
|
|
211
|
+
if (bytesRead < 8)
|
|
212
|
+
return 'corrupt';
|
|
213
|
+
const magicLE = buf.readUInt32LE(0);
|
|
214
|
+
const magicBE = buf.readUInt32BE(0);
|
|
215
|
+
// Universal ("fat") binary — assume it carries the host slice.
|
|
216
|
+
if (magicBE === 0xcafebabe || magicLE === 0xcafebabe)
|
|
217
|
+
return 'valid';
|
|
218
|
+
const thin = magicLE === 0xfeedfacf || magicLE === 0xfeedface;
|
|
219
|
+
const thinSwapped = magicBE === 0xfeedfacf || magicBE === 0xfeedface;
|
|
220
|
+
if (!thin && !thinSwapped)
|
|
221
|
+
return 'corrupt';
|
|
222
|
+
const cpuType = thin ? buf.readUInt32LE(4) : buf.readUInt32BE(4);
|
|
223
|
+
const expected = MACHO_CPUTYPE[arch];
|
|
224
|
+
if (expected === undefined)
|
|
225
|
+
return 'valid';
|
|
226
|
+
return cpuType === expected ? 'valid' : 'corrupt';
|
|
227
|
+
}
|
|
228
|
+
/**
|
|
229
|
+
* Decide whether a binary header is a well-formed shared library for the given
|
|
230
|
+
* platform + architecture — using only the file's structure, no localized text.
|
|
231
|
+
* Pure and injectable (platform/arch as params) so every format+arch combination
|
|
232
|
+
* is unit-testable regardless of the host it runs on.
|
|
233
|
+
*/
|
|
234
|
+
export function classifyBinaryHeader(buf, bytesRead, platform, arch) {
|
|
235
|
+
if (platform === 'win32')
|
|
236
|
+
return classifyPE(buf, bytesRead, arch);
|
|
237
|
+
if (platform === 'linux')
|
|
238
|
+
return classifyELF(buf, bytesRead, arch);
|
|
239
|
+
if (platform === 'darwin')
|
|
240
|
+
return classifyMachO(buf, bytesRead, arch);
|
|
241
|
+
return 'valid'; // unknown host: never claim corrupt
|
|
242
|
+
}
|
|
243
|
+
const BINARY_HEADER_BYTES = 4096;
|
|
244
|
+
/**
|
|
245
|
+
* Best-effort language-independent inspection of the extension file. Reads the
|
|
246
|
+
* header and classifies it; never throws — a missing file is `absent`, an
|
|
247
|
+
* unreadable one is `indeterminate`.
|
|
248
|
+
*/
|
|
249
|
+
export function inspectExtensionBinary(extensionPath) {
|
|
250
|
+
if (!extensionPath)
|
|
251
|
+
return 'indeterminate';
|
|
252
|
+
let fd;
|
|
253
|
+
try {
|
|
254
|
+
fd = openSync(extensionPath, 'r');
|
|
255
|
+
}
|
|
256
|
+
catch (err) {
|
|
257
|
+
return err?.code === 'ENOENT' ? 'absent' : 'indeterminate';
|
|
258
|
+
}
|
|
259
|
+
try {
|
|
260
|
+
const buf = Buffer.alloc(BINARY_HEADER_BYTES);
|
|
261
|
+
const bytesRead = readSync(fd, buf, 0, BINARY_HEADER_BYTES, 0);
|
|
262
|
+
return classifyBinaryHeader(buf, bytesRead, process.platform, process.arch);
|
|
263
|
+
}
|
|
264
|
+
catch {
|
|
265
|
+
return 'indeterminate';
|
|
266
|
+
}
|
|
267
|
+
finally {
|
|
268
|
+
try {
|
|
269
|
+
closeSync(fd);
|
|
270
|
+
}
|
|
271
|
+
catch {
|
|
272
|
+
/* closing the probe fd must never surface */
|
|
273
|
+
}
|
|
274
|
+
}
|
|
275
|
+
}
|
|
276
|
+
/**
|
|
277
|
+
* Diagnose a LadybugDB load failure, preferring a LANGUAGE-INDEPENDENT structural
|
|
278
|
+
* check of the extension binary over the localized error text:
|
|
279
|
+
* - file absent → missing_file
|
|
280
|
+
* - present but malformed → corrupt_file (bad magic / wrong architecture)
|
|
281
|
+
* - present and well-formed → missing_dependency (a valid binary the loader rejected)
|
|
282
|
+
* The path comes from lbug's own English wrapper, so this holds in any OS display
|
|
283
|
+
* language. When the file cannot be located or read, it falls back to the string
|
|
284
|
+
* classifier (which still carries the language-independent hedged fallback). This
|
|
285
|
+
* is the entry point every surface should call.
|
|
286
|
+
*/
|
|
287
|
+
export function diagnoseExtensionLoad(reason) {
|
|
288
|
+
const text = reason ?? '';
|
|
289
|
+
const stringResult = classifyExtensionLoadError(text);
|
|
290
|
+
const fileState = inspectExtensionBinary(extractExtensionPath(text));
|
|
291
|
+
if (fileState === 'corrupt') {
|
|
292
|
+
return { kind: 'corrupt_file', remedy: CORRUPT_FILE_REMEDY };
|
|
293
|
+
}
|
|
294
|
+
if (fileState === 'valid') {
|
|
295
|
+
// The structural probe only inspects the first BINARY_HEADER_BYTES, so a file
|
|
296
|
+
// truncated AFTER its header still reads 'valid'. When the loader itself reported
|
|
297
|
+
// corruption (e.g. "file too short" / Windows error 193 "not a valid Win32
|
|
298
|
+
// application"), that whole-file verdict is stronger evidence than an intact-looking
|
|
299
|
+
// header — honor it and route to re-download, not a runtime-dependency install (#2383
|
|
300
|
+
// F1). Localized corrupt tails classify as hedged missing_dependency (not
|
|
301
|
+
// corrupt_file), so they still fall through to the dependency remedy below.
|
|
302
|
+
if (stringResult.kind === 'corrupt_file') {
|
|
303
|
+
return stringResult;
|
|
304
|
+
}
|
|
305
|
+
// A structurally sound binary that still failed to load ⇒ a dependency/runtime
|
|
306
|
+
// problem, decided WITHOUT the localized tail. Keep the string classifier's
|
|
307
|
+
// sharper remedy when it recognized the specific case (e.g. English 126).
|
|
308
|
+
const remedy = stringResult.kind === 'missing_dependency'
|
|
309
|
+
? stringResult.remedy
|
|
310
|
+
: STRUCTURAL_MISSING_DEPENDENCY_REMEDY;
|
|
311
|
+
return { kind: 'missing_dependency', remedy };
|
|
312
|
+
}
|
|
313
|
+
// 'absent' or 'indeterminate' → no positive structural evidence, so defer to the
|
|
314
|
+
// string classifier. Note a real never-installed extension has NO path in its
|
|
315
|
+
// reason (lbug says "has not been installed"), so it lands here via
|
|
316
|
+
// 'indeterminate' and the string classifier reports missing_file correctly; a
|
|
317
|
+
// path that lbug named but that is now gone (stale/racy) is better judged by
|
|
318
|
+
// what lbug actually reported than by re-deriving from disk.
|
|
319
|
+
return stringResult;
|
|
320
|
+
}
|
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { type ExtensionLoadDiagnosis } from './extension-load-error.js';
|
|
1
2
|
/**
|
|
2
3
|
* Lifecycle policy for an optional DuckDB extension.
|
|
3
4
|
*
|
|
@@ -20,6 +21,12 @@ export interface ExtensionCapability {
|
|
|
20
21
|
loaded: boolean;
|
|
21
22
|
/** Human-readable reason when `loaded` is false. */
|
|
22
23
|
reason?: string;
|
|
24
|
+
/**
|
|
25
|
+
* Classified diagnosis of `reason`, computed ONCE at mark-unavailable time so
|
|
26
|
+
* per-request surfaces (ftsDegradedWarning on /api/search + MCP query) read the
|
|
27
|
+
* cached remedy instead of re-inspecting the extension file on every call (#2383 F3).
|
|
28
|
+
*/
|
|
29
|
+
diagnosis?: ExtensionLoadDiagnosis;
|
|
23
30
|
}
|
|
24
31
|
/** Per-call overrides applied on top of `ExtensionManager` defaults. */
|
|
25
32
|
export interface ExtensionEnsureOptions {
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import { spawn } from 'child_process';
|
|
2
2
|
import { fileURLToPath } from 'node:url';
|
|
3
3
|
import { LBUG_MAX_DB_SIZE } from './lbug-config.js';
|
|
4
|
+
import { diagnoseExtensionLoad } from './extension-load-error.js';
|
|
4
5
|
import { logger } from '../logger.js';
|
|
5
6
|
const DEFAULT_EXTENSION_INSTALL_TIMEOUT_MS = 15_000;
|
|
6
7
|
const EXTENSION_NAME_PATTERN = /^[A-Za-z][A-Za-z0-9_]*$/;
|
|
@@ -208,7 +209,14 @@ export class ExtensionManager {
|
|
|
208
209
|
this.capabilities.set(name, { name, loaded: true });
|
|
209
210
|
}
|
|
210
211
|
markUnavailable(name, label, reason, warn) {
|
|
211
|
-
|
|
212
|
+
// Classify once here (the single load-failure sink, run per Database not per
|
|
213
|
+
// request) so the hot per-request warning path does no file I/O (#2383 F3).
|
|
214
|
+
this.capabilities.set(name, {
|
|
215
|
+
name,
|
|
216
|
+
loaded: false,
|
|
217
|
+
reason,
|
|
218
|
+
diagnosis: diagnoseExtensionLoad(reason),
|
|
219
|
+
});
|
|
212
220
|
const key = `${name}:${reason}`;
|
|
213
221
|
if (this.warnedKeys.has(key))
|
|
214
222
|
return;
|
package/dist/core/run-analyze.js
CHANGED
|
@@ -17,6 +17,7 @@ import { initLbug, loadGraphToLbug, getLbugStats, executeQuery, executeWithReuse
|
|
|
17
17
|
import { createSearchFTSIndexes, initialiseSearchFTSStemmer, verifySearchFTSIndexes, } from './search/fts-indexes.js';
|
|
18
18
|
import { cjkSegmentationModeMismatch, getSearchFTSCjkSegmentation, initialiseSearchFTSCjkSegmentation, } from './search/cjk-segmentation.js';
|
|
19
19
|
import { getExtensionCapabilities, resolveAnalyzeInstallPolicy } from './lbug/extension-loader.js';
|
|
20
|
+
import { diagnoseExtensionLoad } from './lbug/extension-load-error.js';
|
|
20
21
|
import { startWalCheckpointDriver, } from './lbug/wal-checkpoint-driver.js';
|
|
21
22
|
import { getStoragePaths, resolveBranchPlacement, saveMeta, loadMeta, ensureGitNexusIgnored, registerRepo, adoptFlatBranchLabel, isReadOnlyFilesystemError, isRepoRegistered, cleanupOldKuzuFiles, reconcileMetadataFiles, isMissingFilesystemError, INDEX_METADATA_FILE, INCREMENTAL_SCHEMA_VERSION, } from '../storage/repo-manager.js';
|
|
22
23
|
import { DEFAULT_PDG_MAX_FUNCTION_LINES } from './ingestion/cfg/collect.js';
|
|
@@ -41,7 +42,11 @@ import { STALE_HASH_SENTINEL } from './lbug/schema.js';
|
|
|
41
42
|
* a full analyze. Kept as a named constant so the env-var/command guidance
|
|
42
43
|
* stays in one place (mirrors the VECTOR message in embedding-pipeline.ts).
|
|
43
44
|
*/
|
|
44
|
-
|
|
45
|
+
// Class-neutral lead, reused for the missing-dependency degrade path (#2383 F2):
|
|
46
|
+
// its remedy already explains that reinstalling will NOT help, so appending the
|
|
47
|
+
// generic "install with network access" tail below would contradict it.
|
|
48
|
+
const FTS_UNAVAILABLE_LEAD = 'FTS extension unavailable; skipping search-index creation.';
|
|
49
|
+
const FTS_UNAVAILABLE_MESSAGE = `${FTS_UNAVAILABLE_LEAD} ` +
|
|
45
50
|
'Full-text/BM25 search will be disabled until the LadybugDB FTS extension is ' +
|
|
46
51
|
'installed once with network access (GITNEXUS_LBUG_EXTENSION_INSTALL=auto) or ' +
|
|
47
52
|
'pre-installed for offline use. Run `gitnexus doctor` for details.';
|
|
@@ -368,13 +373,20 @@ export async function runFullAnalysis(repoPath, options, callbacks) {
|
|
|
368
373
|
// Surface the load-side reason (#2374): "not pre-installed" was wrong
|
|
369
374
|
// and doctor never installed anything, so the old message trapped
|
|
370
375
|
// users in a query → repair-fts → doctor loop with no way out.
|
|
371
|
-
const
|
|
372
|
-
|
|
373
|
-
|
|
376
|
+
const rawFtsReason = getExtensionCapabilities().find((c) => c.name === 'fts')?.reason;
|
|
377
|
+
const ftsReason = rawFtsReason?.replace(/\.$/, '');
|
|
378
|
+
// A missing runtime dependency (Windows error 126, #2374) is not healed
|
|
379
|
+
// by re-installing — the file is already present. Route that class to the
|
|
380
|
+
// classified remedy (install VC++ redist / OpenSSL) instead of the old
|
|
381
|
+
// "retry the network install" text that trapped the user in a loop.
|
|
382
|
+
const { kind, remedy } = diagnoseExtensionLoad(rawFtsReason);
|
|
383
|
+
const remedyTail = kind === 'missing_dependency'
|
|
384
|
+
? ` ${remedy}`
|
|
385
|
+
: '. Retry with network access and GITNEXUS_LBUG_EXTENSION_INSTALL=auto to install it, ' +
|
|
386
|
+
'or pre-install the extension file; run `gitnexus doctor` for live FTS status.';
|
|
374
387
|
throw new Error('Cannot repair FTS indexes: the LadybugDB FTS extension failed to load' +
|
|
375
388
|
(ftsReason ? ` — ${ftsReason}` : '') +
|
|
376
|
-
|
|
377
|
-
'or pre-install the extension file; run `gitnexus doctor` for live FTS status.');
|
|
389
|
+
remedyTail);
|
|
378
390
|
}
|
|
379
391
|
progress('fts', 85, 'Repairing search indexes...');
|
|
380
392
|
await createSearchFTSIndexes({
|
|
@@ -963,7 +975,15 @@ export async function runFullAnalysis(repoPath, options, callbacks) {
|
|
|
963
975
|
progress('fts', 90, 'Search indexes ready');
|
|
964
976
|
}
|
|
965
977
|
else {
|
|
966
|
-
|
|
978
|
+
// For a missing runtime dependency (#2374) the file is present, so the
|
|
979
|
+
// generic "install it with network access" tail in FTS_UNAVAILABLE_MESSAGE
|
|
980
|
+
// contradicts the remedy's own "reinstalling will NOT help" (#2383 F2). Lead
|
|
981
|
+
// with the class-neutral sentence and append only the classified remedy.
|
|
982
|
+
const ftsReason = getExtensionCapabilities().find((c) => c.name === 'fts')?.reason;
|
|
983
|
+
const { kind, remedy } = diagnoseExtensionLoad(ftsReason);
|
|
984
|
+
log(kind === 'missing_dependency'
|
|
985
|
+
? `${FTS_UNAVAILABLE_LEAD} ${remedy}`
|
|
986
|
+
: FTS_UNAVAILABLE_MESSAGE);
|
|
967
987
|
progress('fts', 90, 'Search indexes skipped (FTS unavailable)');
|
|
968
988
|
}
|
|
969
989
|
// ── Phase 3.5: Re-insert cached embeddings ────────────────────────
|
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { createFTSIndex, dropFTSIndex, DEFAULT_FTS_STEMMER } from '../lbug/lbug-adapter.js';
|
|
2
2
|
import { getExtensionCapabilities } from '../lbug/extension-loader.js';
|
|
3
|
+
import { classifyExtensionLoadError } from '../lbug/extension-load-error.js';
|
|
3
4
|
import { FTS_INDEXES } from './fts-schema.js';
|
|
4
5
|
/**
|
|
5
6
|
* Strip filesystem paths from a LadybugDB error before it reaches the HTTP
|
|
@@ -21,9 +22,18 @@ export const ftsDegradedWarning = () => {
|
|
|
21
22
|
const fts = getExtensionCapabilities().find((c) => c.name === 'fts');
|
|
22
23
|
if (fts && !fts.loaded) {
|
|
23
24
|
const reason = fts.reason ? redactPaths(fts.reason).replace(/\.$/, '') : undefined;
|
|
25
|
+
// A missing *runtime dependency* (Windows error 126, etc.) is not healed by
|
|
26
|
+
// reinstalling (#2374) — surface the classified remedy instead of the generic
|
|
27
|
+
// reinstall tail. Read the diagnosis cached at mark-unavailable time so this
|
|
28
|
+
// per-request path (HTTP /api/search + MCP query) does NO file I/O (#2383 F3);
|
|
29
|
+
// fall back to the pure, no-I/O string classifier if it is somehow absent.
|
|
30
|
+
const { kind, remedy } = fts.diagnosis ?? classifyExtensionLoadError(fts.reason);
|
|
31
|
+
const tail = kind === 'missing_dependency'
|
|
32
|
+
? ` ${remedy}`
|
|
33
|
+
: '. Run `gitnexus doctor` for details, then `gitnexus analyze --repair-fts` with network access to reinstall.';
|
|
24
34
|
return ('FTS extension failed to load — keyword search degraded' +
|
|
25
35
|
(reason ? ` (${reason})` : '') +
|
|
26
|
-
|
|
36
|
+
tail);
|
|
27
37
|
}
|
|
28
38
|
return 'FTS indexes missing — keyword search degraded. Run: gitnexus analyze --repair-fts (or gitnexus analyze --force) to rebuild indexes.';
|
|
29
39
|
};
|
package/package.json
CHANGED
|
@@ -61,6 +61,11 @@ const PLATFORM_LOGIC = [
|
|
|
61
61
|
// array form. Runs on every platform (the ubuntu suite covers Linux; this
|
|
62
62
|
// registration adds windows + macos).
|
|
63
63
|
'test/unit/embedding-install-arg-delivery.test.ts',
|
|
64
|
+
// Structural FTS-extension classifier against REAL binaries (#2374): on this
|
|
65
|
+
// matrix `process.execPath` / `lbugjs.node` are a real PE (windows) and Mach-O
|
|
66
|
+
// (macos), so the header parsing is proven on genuine binaries, not synthetic
|
|
67
|
+
// buffers (the ubuntu suite covers the ELF path).
|
|
68
|
+
'test/integration/extension-binary-real.test.ts',
|
|
64
69
|
];
|
|
65
70
|
|
|
66
71
|
// Native LadybugDB integration tests — exercise the @ladybugdb/core
|
|
@@ -13,7 +13,9 @@ const EXTENSION_NAME_PATTERN = /^[A-Za-z][A-Za-z0-9_]*$/;
|
|
|
13
13
|
// a missing file (plain INSTALL downloads it), or a permanent non-file failure a
|
|
14
14
|
// re-download can never fix (missing runtime dep: "cannot open shared object") —
|
|
15
15
|
// plain INSTALL avoids re-downloading ~2 MB on every analyze run forever.
|
|
16
|
-
|
|
16
|
+
// Exported so a parity test keeps this byte-identical to the copy in
|
|
17
|
+
// src/core/lbug/extension-load-error.ts (this `.mjs` cannot import that `.ts`), #2383 F5b.
|
|
18
|
+
export const FILE_CORRUPTION_SIGNATURES = [
|
|
17
19
|
/invalid elf/i,
|
|
18
20
|
/file too short/i,
|
|
19
21
|
/not a valid/i,
|